xref: /openbmc/qemu/hw/nvme/ctrl.c (revision 764a6ee9)
1 /*
2  * QEMU NVM Express Controller
3  *
4  * Copyright (c) 2012, Intel Corporation
5  *
6  * Written by Keith Busch <keith.busch@intel.com>
7  *
8  * This code is licensed under the GNU GPL v2 or later.
9  */
10 
11 /**
12  * Reference Specs: http://www.nvmexpress.org, 1.4, 1.3, 1.2, 1.1, 1.0e
13  *
14  *  https://nvmexpress.org/developers/nvme-specification/
15  *
16  *
17  * Notes on coding style
18  * ---------------------
19  * While QEMU coding style prefers lowercase hexadecimals in constants, the
20  * NVMe subsystem use this format from the NVMe specifications in the comments
21  * (i.e. 'h' suffix instead of '0x' prefix).
22  *
23  * Usage
24  * -----
25  * See docs/system/nvme.rst for extensive documentation.
26  *
27  * Add options:
28  *      -drive file=<file>,if=none,id=<drive_id>
29  *      -device nvme-subsys,id=<subsys_id>,nqn=<nqn_id>
30  *      -device nvme,serial=<serial>,id=<bus_name>, \
31  *              cmb_size_mb=<cmb_size_mb[optional]>, \
32  *              [pmrdev=<mem_backend_file_id>,] \
33  *              max_ioqpairs=<N[optional]>, \
34  *              aerl=<N[optional]>,aer_max_queued=<N[optional]>, \
35  *              mdts=<N[optional]>,vsl=<N[optional]>, \
36  *              zoned.zasl=<N[optional]>, \
37  *              zoned.auto_transition=<on|off[optional]>, \
38  *              sriov_max_vfs=<N[optional]> \
39  *              sriov_vq_flexible=<N[optional]> \
40  *              sriov_vi_flexible=<N[optional]> \
41  *              sriov_max_vi_per_vf=<N[optional]> \
42  *              sriov_max_vq_per_vf=<N[optional]> \
43  *              atomic.dn=<on|off[optional]>, \
44  *              atomic.awun<N[optional]>, \
45  *              atomic.awupf<N[optional]>, \
46  *              subsys=<subsys_id>
47  *      -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
48  *              zoned=<true|false[optional]>, \
49  *              subsys=<subsys_id>,shared=<true|false[optional]>, \
50  *              detached=<true|false[optional]>, \
51  *              zoned.zone_size=<N[optional]>, \
52  *              zoned.zone_capacity=<N[optional]>, \
53  *              zoned.descr_ext_size=<N[optional]>, \
54  *              zoned.max_active=<N[optional]>, \
55  *              zoned.max_open=<N[optional]>, \
56  *              zoned.cross_read=<true|false[optional]>
57  *
58  * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
59  * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. By default, the
60  * device will use the "v1.4 CMB scheme" - use the `legacy-cmb` parameter to
61  * always enable the CMBLOC and CMBSZ registers (v1.3 behavior).
62  *
63  * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
64  * For example:
65  * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
66  *  size=<size> .... -device nvme,...,pmrdev=<mem_id>
67  *
68  * The PMR will use BAR 4/5 exclusively.
69  *
70  * To place controller(s) and namespace(s) to a subsystem, then provide
71  * nvme-subsys device as above.
72  *
73  * nvme subsystem device parameters
74  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
75  * - `nqn`
76  *   This parameter provides the `<nqn_id>` part of the string
77  *   `nqn.2019-08.org.qemu:<nqn_id>` which will be reported in the SUBNQN field
78  *   of subsystem controllers. Note that `<nqn_id>` should be unique per
79  *   subsystem, but this is not enforced by QEMU. If not specified, it will
80  *   default to the value of the `id` parameter (`<subsys_id>`).
81  *
82  * nvme device parameters
83  * ~~~~~~~~~~~~~~~~~~~~~~
84  * - `subsys`
85  *   Specifying this parameter attaches the controller to the subsystem and
86  *   the SUBNQN field in the controller will report the NQN of the subsystem
87  *   device. This also enables multi controller capability represented in
88  *   Identify Controller data structure in CMIC (Controller Multi-path I/O and
89  *   Namespace Sharing Capabilities).
90  *
91  * - `aerl`
92  *   The Asynchronous Event Request Limit (AERL). Indicates the maximum number
93  *   of concurrently outstanding Asynchronous Event Request commands support
94  *   by the controller. This is a 0's based value.
95  *
96  * - `aer_max_queued`
97  *   This is the maximum number of events that the device will enqueue for
98  *   completion when there are no outstanding AERs. When the maximum number of
99  *   enqueued events are reached, subsequent events will be dropped.
100  *
101  * - `mdts`
102  *   Indicates the maximum data transfer size for a command that transfers data
103  *   between host-accessible memory and the controller. The value is specified
104  *   as a power of two (2^n) and is in units of the minimum memory page size
105  *   (CAP.MPSMIN). The default value is 7 (i.e. 512 KiB).
106  *
107  * - `vsl`
108  *   Indicates the maximum data size limit for the Verify command. Like `mdts`,
109  *   this value is specified as a power of two (2^n) and is in units of the
110  *   minimum memory page size (CAP.MPSMIN). The default value is 7 (i.e. 512
111  *   KiB).
112  *
113  * - `zoned.zasl`
114  *   Indicates the maximum data transfer size for the Zone Append command. Like
115  *   `mdts`, the value is specified as a power of two (2^n) and is in units of
116  *   the minimum memory page size (CAP.MPSMIN). The default value is 0 (i.e.
117  *   defaulting to the value of `mdts`).
118  *
119  * - `zoned.auto_transition`
120  *   Indicates if zones in zone state implicitly opened can be automatically
121  *   transitioned to zone state closed for resource management purposes.
122  *   Defaults to 'on'.
123  *
124  * - `sriov_max_vfs`
125  *   Indicates the maximum number of PCIe virtual functions supported
126  *   by the controller. The default value is 0. Specifying a non-zero value
127  *   enables reporting of both SR-IOV and ARI capabilities by the NVMe device.
128  *   Virtual function controllers will not report SR-IOV capability.
129  *
130  *   NOTE: Single Root I/O Virtualization support is experimental.
131  *   All the related parameters may be subject to change.
132  *
133  * - `sriov_vq_flexible`
134  *   Indicates the total number of flexible queue resources assignable to all
135  *   the secondary controllers. Implicitly sets the number of primary
136  *   controller's private resources to `(max_ioqpairs - sriov_vq_flexible)`.
137  *
138  * - `sriov_vi_flexible`
139  *   Indicates the total number of flexible interrupt resources assignable to
140  *   all the secondary controllers. Implicitly sets the number of primary
141  *   controller's private resources to `(msix_qsize - sriov_vi_flexible)`.
142  *
143  * - `sriov_max_vi_per_vf`
144  *   Indicates the maximum number of virtual interrupt resources assignable
145  *   to a secondary controller. The default 0 resolves to
146  *   `(sriov_vi_flexible / sriov_max_vfs)`.
147  *
148  * - `sriov_max_vq_per_vf`
149  *   Indicates the maximum number of virtual queue resources assignable to
150  *   a secondary controller. The default 0 resolves to
151  *   `(sriov_vq_flexible / sriov_max_vfs)`.
152  *
153  * nvme namespace device parameters
154  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
155  * - `shared`
156  *   When the parent nvme device (as defined explicitly by the 'bus' parameter
157  *   or implicitly by the most recently defined NvmeBus) is linked to an
158  *   nvme-subsys device, the namespace will be attached to all controllers in
159  *   the subsystem. If set to 'off' (the default), the namespace will remain a
160  *   private namespace and may only be attached to a single controller at a
161  *   time.
162  *
163  * - `detached`
164  *   This parameter is only valid together with the `subsys` parameter. If left
165  *   at the default value (`false/off`), the namespace will be attached to all
166  *   controllers in the NVMe subsystem at boot-up. If set to `true/on`, the
167  *   namespace will be available in the subsystem but not attached to any
168  *   controllers.
169  *
170  * Setting `zoned` to true selects Zoned Command Set at the namespace.
171  * In this case, the following namespace properties are available to configure
172  * zoned operation:
173  *     zoned.zone_size=<zone size in bytes, default: 128MiB>
174  *         The number may be followed by K, M, G as in kilo-, mega- or giga-.
175  *
176  *     zoned.zone_capacity=<zone capacity in bytes, default: zone size>
177  *         The value 0 (default) forces zone capacity to be the same as zone
178  *         size. The value of this property may not exceed zone size.
179  *
180  *     zoned.descr_ext_size=<zone descriptor extension size, default 0>
181  *         This value needs to be specified in 64B units. If it is zero,
182  *         namespace(s) will not support zone descriptor extensions.
183  *
184  *     zoned.max_active=<Maximum Active Resources (zones), default: 0>
185  *         The default value means there is no limit to the number of
186  *         concurrently active zones.
187  *
188  *     zoned.max_open=<Maximum Open Resources (zones), default: 0>
189  *         The default value means there is no limit to the number of
190  *         concurrently open zones.
191  *
192  *     zoned.cross_read=<enable RAZB, default: false>
193  *         Setting this property to true enables Read Across Zone Boundaries.
194  */
195 
196 #include "qemu/osdep.h"
197 #include "qemu/cutils.h"
198 #include "qemu/error-report.h"
199 #include "qemu/log.h"
200 #include "qemu/units.h"
201 #include "qemu/range.h"
202 #include "qapi/error.h"
203 #include "qapi/visitor.h"
204 #include "sysemu/sysemu.h"
205 #include "sysemu/block-backend.h"
206 #include "sysemu/hostmem.h"
207 #include "hw/pci/msix.h"
208 #include "hw/pci/pcie_sriov.h"
209 #include "sysemu/spdm-socket.h"
210 #include "migration/vmstate.h"
211 
212 #include "nvme.h"
213 #include "dif.h"
214 #include "trace.h"
215 
216 #define NVME_MAX_IOQPAIRS 0xffff
217 #define NVME_DB_SIZE  4
218 #define NVME_SPEC_VER 0x00010400
219 #define NVME_CMB_BIR 2
220 #define NVME_PMR_BIR 4
221 #define NVME_TEMPERATURE 0x143
222 #define NVME_TEMPERATURE_WARNING 0x157
223 #define NVME_TEMPERATURE_CRITICAL 0x175
224 #define NVME_NUM_FW_SLOTS 1
225 #define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
226 #define NVME_VF_RES_GRANULARITY 1
227 #define NVME_VF_OFFSET 0x1
228 #define NVME_VF_STRIDE 1
229 
230 #define NVME_GUEST_ERR(trace, fmt, ...) \
231     do { \
232         (trace_##trace)(__VA_ARGS__); \
233         qemu_log_mask(LOG_GUEST_ERROR, #trace \
234             " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
235     } while (0)
236 
237 static const bool nvme_feature_support[NVME_FID_MAX] = {
238     [NVME_ARBITRATION]              = true,
239     [NVME_POWER_MANAGEMENT]         = true,
240     [NVME_TEMPERATURE_THRESHOLD]    = true,
241     [NVME_ERROR_RECOVERY]           = true,
242     [NVME_VOLATILE_WRITE_CACHE]     = true,
243     [NVME_NUMBER_OF_QUEUES]         = true,
244     [NVME_INTERRUPT_COALESCING]     = true,
245     [NVME_INTERRUPT_VECTOR_CONF]    = true,
246     [NVME_WRITE_ATOMICITY]          = true,
247     [NVME_ASYNCHRONOUS_EVENT_CONF]  = true,
248     [NVME_TIMESTAMP]                = true,
249     [NVME_HOST_BEHAVIOR_SUPPORT]    = true,
250     [NVME_COMMAND_SET_PROFILE]      = true,
251     [NVME_FDP_MODE]                 = true,
252     [NVME_FDP_EVENTS]               = true,
253 };
254 
255 static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
256     [NVME_TEMPERATURE_THRESHOLD]    = NVME_FEAT_CAP_CHANGE,
257     [NVME_ERROR_RECOVERY]           = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
258     [NVME_VOLATILE_WRITE_CACHE]     = NVME_FEAT_CAP_CHANGE,
259     [NVME_NUMBER_OF_QUEUES]         = NVME_FEAT_CAP_CHANGE,
260     [NVME_WRITE_ATOMICITY]          = NVME_FEAT_CAP_CHANGE,
261     [NVME_ASYNCHRONOUS_EVENT_CONF]  = NVME_FEAT_CAP_CHANGE,
262     [NVME_TIMESTAMP]                = NVME_FEAT_CAP_CHANGE,
263     [NVME_HOST_BEHAVIOR_SUPPORT]    = NVME_FEAT_CAP_CHANGE,
264     [NVME_COMMAND_SET_PROFILE]      = NVME_FEAT_CAP_CHANGE,
265     [NVME_FDP_MODE]                 = NVME_FEAT_CAP_CHANGE,
266     [NVME_FDP_EVENTS]               = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
267 };
268 
269 static const uint32_t nvme_cse_acs[256] = {
270     [NVME_ADM_CMD_DELETE_SQ]        = NVME_CMD_EFF_CSUPP,
271     [NVME_ADM_CMD_CREATE_SQ]        = NVME_CMD_EFF_CSUPP,
272     [NVME_ADM_CMD_GET_LOG_PAGE]     = NVME_CMD_EFF_CSUPP,
273     [NVME_ADM_CMD_DELETE_CQ]        = NVME_CMD_EFF_CSUPP,
274     [NVME_ADM_CMD_CREATE_CQ]        = NVME_CMD_EFF_CSUPP,
275     [NVME_ADM_CMD_IDENTIFY]         = NVME_CMD_EFF_CSUPP,
276     [NVME_ADM_CMD_ABORT]            = NVME_CMD_EFF_CSUPP,
277     [NVME_ADM_CMD_SET_FEATURES]     = NVME_CMD_EFF_CSUPP,
278     [NVME_ADM_CMD_GET_FEATURES]     = NVME_CMD_EFF_CSUPP,
279     [NVME_ADM_CMD_ASYNC_EV_REQ]     = NVME_CMD_EFF_CSUPP,
280     [NVME_ADM_CMD_NS_ATTACHMENT]    = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_NIC,
281     [NVME_ADM_CMD_VIRT_MNGMT]       = NVME_CMD_EFF_CSUPP,
282     [NVME_ADM_CMD_DBBUF_CONFIG]     = NVME_CMD_EFF_CSUPP,
283     [NVME_ADM_CMD_FORMAT_NVM]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
284     [NVME_ADM_CMD_DIRECTIVE_RECV]   = NVME_CMD_EFF_CSUPP,
285     [NVME_ADM_CMD_DIRECTIVE_SEND]   = NVME_CMD_EFF_CSUPP,
286 };
287 
288 static const uint32_t nvme_cse_iocs_none[256];
289 
290 static const uint32_t nvme_cse_iocs_nvm[256] = {
291     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
292     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
293     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
294     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
295     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
296     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
297     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
298     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
299     [NVME_CMD_IO_MGMT_RECV]         = NVME_CMD_EFF_CSUPP,
300     [NVME_CMD_IO_MGMT_SEND]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
301 };
302 
303 static const uint32_t nvme_cse_iocs_zoned[256] = {
304     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
305     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
306     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
307     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
308     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
309     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
310     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
311     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
312     [NVME_CMD_ZONE_APPEND]          = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
313     [NVME_CMD_ZONE_MGMT_SEND]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
314     [NVME_CMD_ZONE_MGMT_RECV]       = NVME_CMD_EFF_CSUPP,
315 };
316 
317 static void nvme_process_sq(void *opaque);
318 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst);
319 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n);
320 
321 static uint16_t nvme_sqid(NvmeRequest *req)
322 {
323     return le16_to_cpu(req->sq->sqid);
324 }
325 
326 static inline uint16_t nvme_make_pid(NvmeNamespace *ns, uint16_t rg,
327                                      uint16_t ph)
328 {
329     uint16_t rgif = ns->endgrp->fdp.rgif;
330 
331     if (!rgif) {
332         return ph;
333     }
334 
335     return (rg << (16 - rgif)) | ph;
336 }
337 
338 static inline bool nvme_ph_valid(NvmeNamespace *ns, uint16_t ph)
339 {
340     return ph < ns->fdp.nphs;
341 }
342 
343 static inline bool nvme_rg_valid(NvmeEnduranceGroup *endgrp, uint16_t rg)
344 {
345     return rg < endgrp->fdp.nrg;
346 }
347 
348 static inline uint16_t nvme_pid2ph(NvmeNamespace *ns, uint16_t pid)
349 {
350     uint16_t rgif = ns->endgrp->fdp.rgif;
351 
352     if (!rgif) {
353         return pid;
354     }
355 
356     return pid & ((1 << (15 - rgif)) - 1);
357 }
358 
359 static inline uint16_t nvme_pid2rg(NvmeNamespace *ns, uint16_t pid)
360 {
361     uint16_t rgif = ns->endgrp->fdp.rgif;
362 
363     if (!rgif) {
364         return 0;
365     }
366 
367     return pid >> (16 - rgif);
368 }
369 
370 static inline bool nvme_parse_pid(NvmeNamespace *ns, uint16_t pid,
371                                   uint16_t *ph, uint16_t *rg)
372 {
373     *rg = nvme_pid2rg(ns, pid);
374     *ph = nvme_pid2ph(ns, pid);
375 
376     return nvme_ph_valid(ns, *ph) && nvme_rg_valid(ns->endgrp, *rg);
377 }
378 
379 static void nvme_assign_zone_state(NvmeNamespace *ns, NvmeZone *zone,
380                                    NvmeZoneState state)
381 {
382     if (QTAILQ_IN_USE(zone, entry)) {
383         switch (nvme_get_zone_state(zone)) {
384         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
385             QTAILQ_REMOVE(&ns->exp_open_zones, zone, entry);
386             break;
387         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
388             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
389             break;
390         case NVME_ZONE_STATE_CLOSED:
391             QTAILQ_REMOVE(&ns->closed_zones, zone, entry);
392             break;
393         case NVME_ZONE_STATE_FULL:
394             QTAILQ_REMOVE(&ns->full_zones, zone, entry);
395         default:
396             ;
397         }
398     }
399 
400     nvme_set_zone_state(zone, state);
401 
402     switch (state) {
403     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
404         QTAILQ_INSERT_TAIL(&ns->exp_open_zones, zone, entry);
405         break;
406     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
407         QTAILQ_INSERT_TAIL(&ns->imp_open_zones, zone, entry);
408         break;
409     case NVME_ZONE_STATE_CLOSED:
410         QTAILQ_INSERT_TAIL(&ns->closed_zones, zone, entry);
411         break;
412     case NVME_ZONE_STATE_FULL:
413         QTAILQ_INSERT_TAIL(&ns->full_zones, zone, entry);
414     case NVME_ZONE_STATE_READ_ONLY:
415         break;
416     default:
417         zone->d.za = 0;
418     }
419 }
420 
421 static uint16_t nvme_zns_check_resources(NvmeNamespace *ns, uint32_t act,
422                                          uint32_t opn, uint32_t zrwa)
423 {
424     if (ns->params.max_active_zones != 0 &&
425         ns->nr_active_zones + act > ns->params.max_active_zones) {
426         trace_pci_nvme_err_insuff_active_res(ns->params.max_active_zones);
427         return NVME_ZONE_TOO_MANY_ACTIVE | NVME_DNR;
428     }
429 
430     if (ns->params.max_open_zones != 0 &&
431         ns->nr_open_zones + opn > ns->params.max_open_zones) {
432         trace_pci_nvme_err_insuff_open_res(ns->params.max_open_zones);
433         return NVME_ZONE_TOO_MANY_OPEN | NVME_DNR;
434     }
435 
436     if (zrwa > ns->zns.numzrwa) {
437         return NVME_NOZRWA | NVME_DNR;
438     }
439 
440     return NVME_SUCCESS;
441 }
442 
443 /*
444  * Check if we can open a zone without exceeding open/active limits.
445  * AOR stands for "Active and Open Resources" (see TP 4053 section 2.5).
446  */
447 static uint16_t nvme_aor_check(NvmeNamespace *ns, uint32_t act, uint32_t opn)
448 {
449     return nvme_zns_check_resources(ns, act, opn, 0);
450 }
451 
452 static NvmeFdpEvent *nvme_fdp_alloc_event(NvmeCtrl *n, NvmeFdpEventBuffer *ebuf)
453 {
454     NvmeFdpEvent *ret = NULL;
455     bool is_full = ebuf->next == ebuf->start && ebuf->nelems;
456 
457     ret = &ebuf->events[ebuf->next++];
458     if (unlikely(ebuf->next == NVME_FDP_MAX_EVENTS)) {
459         ebuf->next = 0;
460     }
461     if (is_full) {
462         ebuf->start = ebuf->next;
463     } else {
464         ebuf->nelems++;
465     }
466 
467     memset(ret, 0, sizeof(NvmeFdpEvent));
468     ret->timestamp = nvme_get_timestamp(n);
469 
470     return ret;
471 }
472 
473 static inline int log_event(NvmeRuHandle *ruh, uint8_t event_type)
474 {
475     return (ruh->event_filter >> nvme_fdp_evf_shifts[event_type]) & 0x1;
476 }
477 
478 static bool nvme_update_ruh(NvmeCtrl *n, NvmeNamespace *ns, uint16_t pid)
479 {
480     NvmeEnduranceGroup *endgrp = ns->endgrp;
481     NvmeRuHandle *ruh;
482     NvmeReclaimUnit *ru;
483     NvmeFdpEvent *e = NULL;
484     uint16_t ph, rg, ruhid;
485 
486     if (!nvme_parse_pid(ns, pid, &ph, &rg)) {
487         return false;
488     }
489 
490     ruhid = ns->fdp.phs[ph];
491 
492     ruh = &endgrp->fdp.ruhs[ruhid];
493     ru = &ruh->rus[rg];
494 
495     if (ru->ruamw) {
496         if (log_event(ruh, FDP_EVT_RU_NOT_FULLY_WRITTEN)) {
497             e = nvme_fdp_alloc_event(n, &endgrp->fdp.host_events);
498             e->type = FDP_EVT_RU_NOT_FULLY_WRITTEN;
499             e->flags = FDPEF_PIV | FDPEF_NSIDV | FDPEF_LV;
500             e->pid = cpu_to_le16(pid);
501             e->nsid = cpu_to_le32(ns->params.nsid);
502             e->rgid = cpu_to_le16(rg);
503             e->ruhid = cpu_to_le16(ruhid);
504         }
505 
506         /* log (eventual) GC overhead of prematurely swapping the RU */
507         nvme_fdp_stat_inc(&endgrp->fdp.mbmw, nvme_l2b(ns, ru->ruamw));
508     }
509 
510     ru->ruamw = ruh->ruamw;
511 
512     return true;
513 }
514 
515 static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
516 {
517     hwaddr hi, lo;
518 
519     if (!n->cmb.cmse) {
520         return false;
521     }
522 
523     lo = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
524     hi = lo + int128_get64(n->cmb.mem.size);
525 
526     return addr >= lo && addr < hi;
527 }
528 
529 static inline void *nvme_addr_to_cmb(NvmeCtrl *n, hwaddr addr)
530 {
531     hwaddr base = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
532     return &n->cmb.buf[addr - base];
533 }
534 
535 static bool nvme_addr_is_pmr(NvmeCtrl *n, hwaddr addr)
536 {
537     hwaddr hi;
538 
539     if (!n->pmr.cmse) {
540         return false;
541     }
542 
543     hi = n->pmr.cba + int128_get64(n->pmr.dev->mr.size);
544 
545     return addr >= n->pmr.cba && addr < hi;
546 }
547 
548 static inline void *nvme_addr_to_pmr(NvmeCtrl *n, hwaddr addr)
549 {
550     return memory_region_get_ram_ptr(&n->pmr.dev->mr) + (addr - n->pmr.cba);
551 }
552 
553 static inline bool nvme_addr_is_iomem(NvmeCtrl *n, hwaddr addr)
554 {
555     hwaddr hi, lo;
556 
557     /*
558      * The purpose of this check is to guard against invalid "local" access to
559      * the iomem (i.e. controller registers). Thus, we check against the range
560      * covered by the 'bar0' MemoryRegion since that is currently composed of
561      * two subregions (the NVMe "MBAR" and the MSI-X table/pba). Note, however,
562      * that if the device model is ever changed to allow the CMB to be located
563      * in BAR0 as well, then this must be changed.
564      */
565     lo = n->bar0.addr;
566     hi = lo + int128_get64(n->bar0.size);
567 
568     return addr >= lo && addr < hi;
569 }
570 
571 static int nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
572 {
573     hwaddr hi = addr + size - 1;
574     if (hi < addr) {
575         return 1;
576     }
577 
578     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
579         memcpy(buf, nvme_addr_to_cmb(n, addr), size);
580         return 0;
581     }
582 
583     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
584         memcpy(buf, nvme_addr_to_pmr(n, addr), size);
585         return 0;
586     }
587 
588     return pci_dma_read(PCI_DEVICE(n), addr, buf, size);
589 }
590 
591 static int nvme_addr_write(NvmeCtrl *n, hwaddr addr, const void *buf, int size)
592 {
593     hwaddr hi = addr + size - 1;
594     if (hi < addr) {
595         return 1;
596     }
597 
598     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
599         memcpy(nvme_addr_to_cmb(n, addr), buf, size);
600         return 0;
601     }
602 
603     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
604         memcpy(nvme_addr_to_pmr(n, addr), buf, size);
605         return 0;
606     }
607 
608     return pci_dma_write(PCI_DEVICE(n), addr, buf, size);
609 }
610 
611 static bool nvme_nsid_valid(NvmeCtrl *n, uint32_t nsid)
612 {
613     return nsid &&
614         (nsid == NVME_NSID_BROADCAST || nsid <= NVME_MAX_NAMESPACES);
615 }
616 
617 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
618 {
619     return sqid < n->conf_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
620 }
621 
622 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
623 {
624     return cqid < n->conf_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
625 }
626 
627 static void nvme_inc_cq_tail(NvmeCQueue *cq)
628 {
629     cq->tail++;
630     if (cq->tail >= cq->size) {
631         cq->tail = 0;
632         cq->phase = !cq->phase;
633     }
634 }
635 
636 static void nvme_inc_sq_head(NvmeSQueue *sq)
637 {
638     sq->head = (sq->head + 1) % sq->size;
639 }
640 
641 static uint8_t nvme_cq_full(NvmeCQueue *cq)
642 {
643     return (cq->tail + 1) % cq->size == cq->head;
644 }
645 
646 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
647 {
648     return sq->head == sq->tail;
649 }
650 
651 static void nvme_irq_check(NvmeCtrl *n)
652 {
653     PCIDevice *pci = PCI_DEVICE(n);
654     uint32_t intms = ldl_le_p(&n->bar.intms);
655 
656     if (msix_enabled(pci)) {
657         return;
658     }
659     if (~intms & n->irq_status) {
660         pci_irq_assert(pci);
661     } else {
662         pci_irq_deassert(pci);
663     }
664 }
665 
666 static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
667 {
668     PCIDevice *pci = PCI_DEVICE(n);
669 
670     if (cq->irq_enabled) {
671         if (msix_enabled(pci)) {
672             trace_pci_nvme_irq_msix(cq->vector);
673             msix_notify(pci, cq->vector);
674         } else {
675             trace_pci_nvme_irq_pin();
676             assert(cq->vector < 32);
677             n->irq_status |= 1 << cq->vector;
678             nvme_irq_check(n);
679         }
680     } else {
681         trace_pci_nvme_irq_masked();
682     }
683 }
684 
685 static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
686 {
687     if (cq->irq_enabled) {
688         if (msix_enabled(PCI_DEVICE(n))) {
689             return;
690         } else {
691             assert(cq->vector < 32);
692             if (!n->cq_pending) {
693                 n->irq_status &= ~(1 << cq->vector);
694             }
695             nvme_irq_check(n);
696         }
697     }
698 }
699 
700 static void nvme_req_clear(NvmeRequest *req)
701 {
702     req->ns = NULL;
703     req->opaque = NULL;
704     req->aiocb = NULL;
705     memset(&req->cqe, 0x0, sizeof(req->cqe));
706     req->status = NVME_SUCCESS;
707 }
708 
709 static inline void nvme_sg_init(NvmeCtrl *n, NvmeSg *sg, bool dma)
710 {
711     if (dma) {
712         pci_dma_sglist_init(&sg->qsg, PCI_DEVICE(n), 0);
713         sg->flags = NVME_SG_DMA;
714     } else {
715         qemu_iovec_init(&sg->iov, 0);
716     }
717 
718     sg->flags |= NVME_SG_ALLOC;
719 }
720 
721 static inline void nvme_sg_unmap(NvmeSg *sg)
722 {
723     if (!(sg->flags & NVME_SG_ALLOC)) {
724         return;
725     }
726 
727     if (sg->flags & NVME_SG_DMA) {
728         qemu_sglist_destroy(&sg->qsg);
729     } else {
730         qemu_iovec_destroy(&sg->iov);
731     }
732 
733     memset(sg, 0x0, sizeof(*sg));
734 }
735 
736 /*
737  * When metadata is transferred as extended LBAs, the DPTR mapped into `sg`
738  * holds both data and metadata. This function splits the data and metadata
739  * into two separate QSG/IOVs.
740  */
741 static void nvme_sg_split(NvmeSg *sg, NvmeNamespace *ns, NvmeSg *data,
742                           NvmeSg *mdata)
743 {
744     NvmeSg *dst = data;
745     uint32_t trans_len, count = ns->lbasz;
746     uint64_t offset = 0;
747     bool dma = sg->flags & NVME_SG_DMA;
748     size_t sge_len;
749     size_t sg_len = dma ? sg->qsg.size : sg->iov.size;
750     int sg_idx = 0;
751 
752     assert(sg->flags & NVME_SG_ALLOC);
753 
754     while (sg_len) {
755         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
756 
757         trans_len = MIN(sg_len, count);
758         trans_len = MIN(trans_len, sge_len - offset);
759 
760         if (dst) {
761             if (dma) {
762                 qemu_sglist_add(&dst->qsg, sg->qsg.sg[sg_idx].base + offset,
763                                 trans_len);
764             } else {
765                 qemu_iovec_add(&dst->iov,
766                                sg->iov.iov[sg_idx].iov_base + offset,
767                                trans_len);
768             }
769         }
770 
771         sg_len -= trans_len;
772         count -= trans_len;
773         offset += trans_len;
774 
775         if (count == 0) {
776             dst = (dst == data) ? mdata : data;
777             count = (dst == data) ? ns->lbasz : ns->lbaf.ms;
778         }
779 
780         if (sge_len == offset) {
781             offset = 0;
782             sg_idx++;
783         }
784     }
785 }
786 
787 static uint16_t nvme_map_addr_cmb(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
788                                   size_t len)
789 {
790     if (!len) {
791         return NVME_SUCCESS;
792     }
793 
794     trace_pci_nvme_map_addr_cmb(addr, len);
795 
796     if (!nvme_addr_is_cmb(n, addr) || !nvme_addr_is_cmb(n, addr + len - 1)) {
797         return NVME_DATA_TRAS_ERROR;
798     }
799 
800     qemu_iovec_add(iov, nvme_addr_to_cmb(n, addr), len);
801 
802     return NVME_SUCCESS;
803 }
804 
805 static uint16_t nvme_map_addr_pmr(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
806                                   size_t len)
807 {
808     if (!len) {
809         return NVME_SUCCESS;
810     }
811 
812     if (!nvme_addr_is_pmr(n, addr) || !nvme_addr_is_pmr(n, addr + len - 1)) {
813         return NVME_DATA_TRAS_ERROR;
814     }
815 
816     qemu_iovec_add(iov, nvme_addr_to_pmr(n, addr), len);
817 
818     return NVME_SUCCESS;
819 }
820 
821 static uint16_t nvme_map_addr(NvmeCtrl *n, NvmeSg *sg, hwaddr addr, size_t len)
822 {
823     bool cmb = false, pmr = false;
824 
825     if (!len) {
826         return NVME_SUCCESS;
827     }
828 
829     trace_pci_nvme_map_addr(addr, len);
830 
831     if (nvme_addr_is_iomem(n, addr)) {
832         return NVME_DATA_TRAS_ERROR;
833     }
834 
835     if (nvme_addr_is_cmb(n, addr)) {
836         cmb = true;
837     } else if (nvme_addr_is_pmr(n, addr)) {
838         pmr = true;
839     }
840 
841     if (cmb || pmr) {
842         if (sg->flags & NVME_SG_DMA) {
843             return NVME_INVALID_USE_OF_CMB | NVME_DNR;
844         }
845 
846         if (sg->iov.niov + 1 > IOV_MAX) {
847             goto max_mappings_exceeded;
848         }
849 
850         if (cmb) {
851             return nvme_map_addr_cmb(n, &sg->iov, addr, len);
852         } else {
853             return nvme_map_addr_pmr(n, &sg->iov, addr, len);
854         }
855     }
856 
857     if (!(sg->flags & NVME_SG_DMA)) {
858         return NVME_INVALID_USE_OF_CMB | NVME_DNR;
859     }
860 
861     if (sg->qsg.nsg + 1 > IOV_MAX) {
862         goto max_mappings_exceeded;
863     }
864 
865     qemu_sglist_add(&sg->qsg, addr, len);
866 
867     return NVME_SUCCESS;
868 
869 max_mappings_exceeded:
870     NVME_GUEST_ERR(pci_nvme_ub_too_many_mappings,
871                    "number of mappings exceed 1024");
872     return NVME_INTERNAL_DEV_ERROR | NVME_DNR;
873 }
874 
875 static inline bool nvme_addr_is_dma(NvmeCtrl *n, hwaddr addr)
876 {
877     return !(nvme_addr_is_cmb(n, addr) || nvme_addr_is_pmr(n, addr));
878 }
879 
880 static uint16_t nvme_map_prp(NvmeCtrl *n, NvmeSg *sg, uint64_t prp1,
881                              uint64_t prp2, uint32_t len)
882 {
883     hwaddr trans_len = n->page_size - (prp1 % n->page_size);
884     trans_len = MIN(len, trans_len);
885     int num_prps = (len >> n->page_bits) + 1;
886     uint16_t status;
887     int ret;
888 
889     trace_pci_nvme_map_prp(trans_len, len, prp1, prp2, num_prps);
890 
891     nvme_sg_init(n, sg, nvme_addr_is_dma(n, prp1));
892 
893     status = nvme_map_addr(n, sg, prp1, trans_len);
894     if (status) {
895         goto unmap;
896     }
897 
898     len -= trans_len;
899     if (len) {
900         if (len > n->page_size) {
901             g_autofree uint64_t *prp_list = g_new(uint64_t, n->max_prp_ents);
902             uint32_t nents, prp_trans;
903             int i = 0;
904 
905             /*
906              * The first PRP list entry, pointed to by PRP2 may contain offset.
907              * Hence, we need to calculate the number of entries in based on
908              * that offset.
909              */
910             nents = (n->page_size - (prp2 & (n->page_size - 1))) >> 3;
911             prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
912             ret = nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
913             if (ret) {
914                 trace_pci_nvme_err_addr_read(prp2);
915                 status = NVME_DATA_TRAS_ERROR;
916                 goto unmap;
917             }
918             while (len != 0) {
919                 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
920 
921                 if (i == nents - 1 && len > n->page_size) {
922                     if (unlikely(prp_ent & (n->page_size - 1))) {
923                         trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
924                         status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
925                         goto unmap;
926                     }
927 
928                     i = 0;
929                     nents = (len + n->page_size - 1) >> n->page_bits;
930                     nents = MIN(nents, n->max_prp_ents);
931                     prp_trans = nents * sizeof(uint64_t);
932                     ret = nvme_addr_read(n, prp_ent, (void *)prp_list,
933                                          prp_trans);
934                     if (ret) {
935                         trace_pci_nvme_err_addr_read(prp_ent);
936                         status = NVME_DATA_TRAS_ERROR;
937                         goto unmap;
938                     }
939                     prp_ent = le64_to_cpu(prp_list[i]);
940                 }
941 
942                 if (unlikely(prp_ent & (n->page_size - 1))) {
943                     trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
944                     status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
945                     goto unmap;
946                 }
947 
948                 trans_len = MIN(len, n->page_size);
949                 status = nvme_map_addr(n, sg, prp_ent, trans_len);
950                 if (status) {
951                     goto unmap;
952                 }
953 
954                 len -= trans_len;
955                 i++;
956             }
957         } else {
958             if (unlikely(prp2 & (n->page_size - 1))) {
959                 trace_pci_nvme_err_invalid_prp2_align(prp2);
960                 status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
961                 goto unmap;
962             }
963             status = nvme_map_addr(n, sg, prp2, len);
964             if (status) {
965                 goto unmap;
966             }
967         }
968     }
969 
970     return NVME_SUCCESS;
971 
972 unmap:
973     nvme_sg_unmap(sg);
974     return status;
975 }
976 
977 /*
978  * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
979  * number of bytes mapped in len.
980  */
981 static uint16_t nvme_map_sgl_data(NvmeCtrl *n, NvmeSg *sg,
982                                   NvmeSglDescriptor *segment, uint64_t nsgld,
983                                   size_t *len, NvmeCmd *cmd)
984 {
985     dma_addr_t addr, trans_len;
986     uint32_t dlen;
987     uint16_t status;
988 
989     for (int i = 0; i < nsgld; i++) {
990         uint8_t type = NVME_SGL_TYPE(segment[i].type);
991 
992         switch (type) {
993         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
994             break;
995         case NVME_SGL_DESCR_TYPE_SEGMENT:
996         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
997             return NVME_INVALID_NUM_SGL_DESCRS | NVME_DNR;
998         default:
999             return NVME_SGL_DESCR_TYPE_INVALID | NVME_DNR;
1000         }
1001 
1002         dlen = le32_to_cpu(segment[i].len);
1003 
1004         if (!dlen) {
1005             continue;
1006         }
1007 
1008         if (*len == 0) {
1009             /*
1010              * All data has been mapped, but the SGL contains additional
1011              * segments and/or descriptors. The controller might accept
1012              * ignoring the rest of the SGL.
1013              */
1014             uint32_t sgls = le32_to_cpu(n->id_ctrl.sgls);
1015             if (sgls & NVME_CTRL_SGLS_EXCESS_LENGTH) {
1016                 break;
1017             }
1018 
1019             trace_pci_nvme_err_invalid_sgl_excess_length(dlen);
1020             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1021         }
1022 
1023         trans_len = MIN(*len, dlen);
1024 
1025         addr = le64_to_cpu(segment[i].addr);
1026 
1027         if (UINT64_MAX - addr < dlen) {
1028             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1029         }
1030 
1031         status = nvme_map_addr(n, sg, addr, trans_len);
1032         if (status) {
1033             return status;
1034         }
1035 
1036         *len -= trans_len;
1037     }
1038 
1039     return NVME_SUCCESS;
1040 }
1041 
1042 static uint16_t nvme_map_sgl(NvmeCtrl *n, NvmeSg *sg, NvmeSglDescriptor sgl,
1043                              size_t len, NvmeCmd *cmd)
1044 {
1045     /*
1046      * Read the segment in chunks of 256 descriptors (one 4k page) to avoid
1047      * dynamically allocating a potentially huge SGL. The spec allows the SGL
1048      * to be larger (as in number of bytes required to describe the SGL
1049      * descriptors and segment chain) than the command transfer size, so it is
1050      * not bounded by MDTS.
1051      */
1052 #define SEG_CHUNK_SIZE 256
1053 
1054     NvmeSglDescriptor segment[SEG_CHUNK_SIZE], *sgld, *last_sgld;
1055     uint64_t nsgld;
1056     uint32_t seg_len;
1057     uint16_t status;
1058     hwaddr addr;
1059     int ret;
1060 
1061     sgld = &sgl;
1062     addr = le64_to_cpu(sgl.addr);
1063 
1064     trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl.type), len);
1065 
1066     nvme_sg_init(n, sg, nvme_addr_is_dma(n, addr));
1067 
1068     /*
1069      * If the entire transfer can be described with a single data block it can
1070      * be mapped directly.
1071      */
1072     if (NVME_SGL_TYPE(sgl.type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1073         status = nvme_map_sgl_data(n, sg, sgld, 1, &len, cmd);
1074         if (status) {
1075             goto unmap;
1076         }
1077 
1078         goto out;
1079     }
1080 
1081     for (;;) {
1082         switch (NVME_SGL_TYPE(sgld->type)) {
1083         case NVME_SGL_DESCR_TYPE_SEGMENT:
1084         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
1085             break;
1086         default:
1087             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1088         }
1089 
1090         seg_len = le32_to_cpu(sgld->len);
1091 
1092         /* check the length of the (Last) Segment descriptor */
1093         if (!seg_len || seg_len & 0xf) {
1094             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1095         }
1096 
1097         if (UINT64_MAX - addr < seg_len) {
1098             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1099         }
1100 
1101         nsgld = seg_len / sizeof(NvmeSglDescriptor);
1102 
1103         while (nsgld > SEG_CHUNK_SIZE) {
1104             if (nvme_addr_read(n, addr, segment, sizeof(segment))) {
1105                 trace_pci_nvme_err_addr_read(addr);
1106                 status = NVME_DATA_TRAS_ERROR;
1107                 goto unmap;
1108             }
1109 
1110             status = nvme_map_sgl_data(n, sg, segment, SEG_CHUNK_SIZE,
1111                                        &len, cmd);
1112             if (status) {
1113                 goto unmap;
1114             }
1115 
1116             nsgld -= SEG_CHUNK_SIZE;
1117             addr += SEG_CHUNK_SIZE * sizeof(NvmeSglDescriptor);
1118         }
1119 
1120         ret = nvme_addr_read(n, addr, segment, nsgld *
1121                              sizeof(NvmeSglDescriptor));
1122         if (ret) {
1123             trace_pci_nvme_err_addr_read(addr);
1124             status = NVME_DATA_TRAS_ERROR;
1125             goto unmap;
1126         }
1127 
1128         last_sgld = &segment[nsgld - 1];
1129 
1130         /*
1131          * If the segment ends with a Data Block, then we are done.
1132          */
1133         if (NVME_SGL_TYPE(last_sgld->type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1134             status = nvme_map_sgl_data(n, sg, segment, nsgld, &len, cmd);
1135             if (status) {
1136                 goto unmap;
1137             }
1138 
1139             goto out;
1140         }
1141 
1142         /*
1143          * If the last descriptor was not a Data Block, then the current
1144          * segment must not be a Last Segment.
1145          */
1146         if (NVME_SGL_TYPE(sgld->type) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT) {
1147             status = NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1148             goto unmap;
1149         }
1150 
1151         sgld = last_sgld;
1152         addr = le64_to_cpu(sgld->addr);
1153 
1154         /*
1155          * Do not map the last descriptor; it will be a Segment or Last Segment
1156          * descriptor and is handled by the next iteration.
1157          */
1158         status = nvme_map_sgl_data(n, sg, segment, nsgld - 1, &len, cmd);
1159         if (status) {
1160             goto unmap;
1161         }
1162     }
1163 
1164 out:
1165     /* if there is any residual left in len, the SGL was too short */
1166     if (len) {
1167         status = NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1168         goto unmap;
1169     }
1170 
1171     return NVME_SUCCESS;
1172 
1173 unmap:
1174     nvme_sg_unmap(sg);
1175     return status;
1176 }
1177 
1178 uint16_t nvme_map_dptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1179                        NvmeCmd *cmd)
1180 {
1181     uint64_t prp1, prp2;
1182 
1183     switch (NVME_CMD_FLAGS_PSDT(cmd->flags)) {
1184     case NVME_PSDT_PRP:
1185         prp1 = le64_to_cpu(cmd->dptr.prp1);
1186         prp2 = le64_to_cpu(cmd->dptr.prp2);
1187 
1188         return nvme_map_prp(n, sg, prp1, prp2, len);
1189     case NVME_PSDT_SGL_MPTR_CONTIGUOUS:
1190     case NVME_PSDT_SGL_MPTR_SGL:
1191         return nvme_map_sgl(n, sg, cmd->dptr.sgl, len, cmd);
1192     default:
1193         return NVME_INVALID_FIELD;
1194     }
1195 }
1196 
1197 static uint16_t nvme_map_mptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1198                               NvmeCmd *cmd)
1199 {
1200     int psdt = NVME_CMD_FLAGS_PSDT(cmd->flags);
1201     hwaddr mptr = le64_to_cpu(cmd->mptr);
1202     uint16_t status;
1203 
1204     if (psdt == NVME_PSDT_SGL_MPTR_SGL) {
1205         NvmeSglDescriptor sgl;
1206 
1207         if (nvme_addr_read(n, mptr, &sgl, sizeof(sgl))) {
1208             return NVME_DATA_TRAS_ERROR;
1209         }
1210 
1211         status = nvme_map_sgl(n, sg, sgl, len, cmd);
1212         if (status && (status & 0x7ff) == NVME_DATA_SGL_LEN_INVALID) {
1213             status = NVME_MD_SGL_LEN_INVALID | NVME_DNR;
1214         }
1215 
1216         return status;
1217     }
1218 
1219     nvme_sg_init(n, sg, nvme_addr_is_dma(n, mptr));
1220     status = nvme_map_addr(n, sg, mptr, len);
1221     if (status) {
1222         nvme_sg_unmap(sg);
1223     }
1224 
1225     return status;
1226 }
1227 
1228 static uint16_t nvme_map_data(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1229 {
1230     NvmeNamespace *ns = req->ns;
1231     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1232     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1233     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1234     size_t len = nvme_l2b(ns, nlb);
1235     uint16_t status;
1236 
1237     if (nvme_ns_ext(ns) &&
1238         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1239         NvmeSg sg;
1240 
1241         len += nvme_m2b(ns, nlb);
1242 
1243         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1244         if (status) {
1245             return status;
1246         }
1247 
1248         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1249         nvme_sg_split(&sg, ns, &req->sg, NULL);
1250         nvme_sg_unmap(&sg);
1251 
1252         return NVME_SUCCESS;
1253     }
1254 
1255     return nvme_map_dptr(n, &req->sg, len, &req->cmd);
1256 }
1257 
1258 static uint16_t nvme_map_mdata(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1259 {
1260     NvmeNamespace *ns = req->ns;
1261     size_t len = nvme_m2b(ns, nlb);
1262     uint16_t status;
1263 
1264     if (nvme_ns_ext(ns)) {
1265         NvmeSg sg;
1266 
1267         len += nvme_l2b(ns, nlb);
1268 
1269         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1270         if (status) {
1271             return status;
1272         }
1273 
1274         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1275         nvme_sg_split(&sg, ns, NULL, &req->sg);
1276         nvme_sg_unmap(&sg);
1277 
1278         return NVME_SUCCESS;
1279     }
1280 
1281     return nvme_map_mptr(n, &req->sg, len, &req->cmd);
1282 }
1283 
1284 static uint16_t nvme_tx_interleaved(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr,
1285                                     uint32_t len, uint32_t bytes,
1286                                     int32_t skip_bytes, int64_t offset,
1287                                     NvmeTxDirection dir)
1288 {
1289     hwaddr addr;
1290     uint32_t trans_len, count = bytes;
1291     bool dma = sg->flags & NVME_SG_DMA;
1292     int64_t sge_len;
1293     int sg_idx = 0;
1294     int ret;
1295 
1296     assert(sg->flags & NVME_SG_ALLOC);
1297 
1298     while (len) {
1299         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
1300 
1301         if (sge_len - offset < 0) {
1302             offset -= sge_len;
1303             sg_idx++;
1304             continue;
1305         }
1306 
1307         if (sge_len == offset) {
1308             offset = 0;
1309             sg_idx++;
1310             continue;
1311         }
1312 
1313         trans_len = MIN(len, count);
1314         trans_len = MIN(trans_len, sge_len - offset);
1315 
1316         if (dma) {
1317             addr = sg->qsg.sg[sg_idx].base + offset;
1318         } else {
1319             addr = (hwaddr)(uintptr_t)sg->iov.iov[sg_idx].iov_base + offset;
1320         }
1321 
1322         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1323             ret = nvme_addr_read(n, addr, ptr, trans_len);
1324         } else {
1325             ret = nvme_addr_write(n, addr, ptr, trans_len);
1326         }
1327 
1328         if (ret) {
1329             return NVME_DATA_TRAS_ERROR;
1330         }
1331 
1332         ptr += trans_len;
1333         len -= trans_len;
1334         count -= trans_len;
1335         offset += trans_len;
1336 
1337         if (count == 0) {
1338             count = bytes;
1339             offset += skip_bytes;
1340         }
1341     }
1342 
1343     return NVME_SUCCESS;
1344 }
1345 
1346 static uint16_t nvme_tx(NvmeCtrl *n, NvmeSg *sg, void *ptr, uint32_t len,
1347                         NvmeTxDirection dir)
1348 {
1349     assert(sg->flags & NVME_SG_ALLOC);
1350 
1351     if (sg->flags & NVME_SG_DMA) {
1352         const MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
1353         dma_addr_t residual;
1354 
1355         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1356             dma_buf_write(ptr, len, &residual, &sg->qsg, attrs);
1357         } else {
1358             dma_buf_read(ptr, len, &residual, &sg->qsg, attrs);
1359         }
1360 
1361         if (unlikely(residual)) {
1362             trace_pci_nvme_err_invalid_dma();
1363             return NVME_INVALID_FIELD | NVME_DNR;
1364         }
1365     } else {
1366         size_t bytes;
1367 
1368         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1369             bytes = qemu_iovec_to_buf(&sg->iov, 0, ptr, len);
1370         } else {
1371             bytes = qemu_iovec_from_buf(&sg->iov, 0, ptr, len);
1372         }
1373 
1374         if (unlikely(bytes != len)) {
1375             trace_pci_nvme_err_invalid_dma();
1376             return NVME_INVALID_FIELD | NVME_DNR;
1377         }
1378     }
1379 
1380     return NVME_SUCCESS;
1381 }
1382 
1383 static inline uint16_t nvme_c2h(NvmeCtrl *n, void *ptr, uint32_t len,
1384                                 NvmeRequest *req)
1385 {
1386     uint16_t status;
1387 
1388     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1389     if (status) {
1390         return status;
1391     }
1392 
1393     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_FROM_DEVICE);
1394 }
1395 
1396 static inline uint16_t nvme_h2c(NvmeCtrl *n, void *ptr, uint32_t len,
1397                                 NvmeRequest *req)
1398 {
1399     uint16_t status;
1400 
1401     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1402     if (status) {
1403         return status;
1404     }
1405 
1406     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_TO_DEVICE);
1407 }
1408 
1409 uint16_t nvme_bounce_data(NvmeCtrl *n, void *ptr, uint32_t len,
1410                           NvmeTxDirection dir, NvmeRequest *req)
1411 {
1412     NvmeNamespace *ns = req->ns;
1413     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1414     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1415     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1416 
1417     if (nvme_ns_ext(ns) &&
1418         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1419         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbasz,
1420                                    ns->lbaf.ms, 0, dir);
1421     }
1422 
1423     return nvme_tx(n, &req->sg, ptr, len, dir);
1424 }
1425 
1426 uint16_t nvme_bounce_mdata(NvmeCtrl *n, void *ptr, uint32_t len,
1427                            NvmeTxDirection dir, NvmeRequest *req)
1428 {
1429     NvmeNamespace *ns = req->ns;
1430     uint16_t status;
1431 
1432     if (nvme_ns_ext(ns)) {
1433         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbaf.ms,
1434                                    ns->lbasz, ns->lbasz, dir);
1435     }
1436 
1437     nvme_sg_unmap(&req->sg);
1438 
1439     status = nvme_map_mptr(n, &req->sg, len, &req->cmd);
1440     if (status) {
1441         return status;
1442     }
1443 
1444     return nvme_tx(n, &req->sg, ptr, len, dir);
1445 }
1446 
1447 static inline void nvme_blk_read(BlockBackend *blk, int64_t offset,
1448                                  uint32_t align, BlockCompletionFunc *cb,
1449                                  NvmeRequest *req)
1450 {
1451     assert(req->sg.flags & NVME_SG_ALLOC);
1452 
1453     if (req->sg.flags & NVME_SG_DMA) {
1454         req->aiocb = dma_blk_read(blk, &req->sg.qsg, offset, align, cb, req);
1455     } else {
1456         req->aiocb = blk_aio_preadv(blk, offset, &req->sg.iov, 0, cb, req);
1457     }
1458 }
1459 
1460 static inline void nvme_blk_write(BlockBackend *blk, int64_t offset,
1461                                   uint32_t align, BlockCompletionFunc *cb,
1462                                   NvmeRequest *req)
1463 {
1464     assert(req->sg.flags & NVME_SG_ALLOC);
1465 
1466     if (req->sg.flags & NVME_SG_DMA) {
1467         req->aiocb = dma_blk_write(blk, &req->sg.qsg, offset, align, cb, req);
1468     } else {
1469         req->aiocb = blk_aio_pwritev(blk, offset, &req->sg.iov, 0, cb, req);
1470     }
1471 }
1472 
1473 static void nvme_update_cq_eventidx(const NvmeCQueue *cq)
1474 {
1475     trace_pci_nvme_update_cq_eventidx(cq->cqid, cq->head);
1476 
1477     stl_le_pci_dma(PCI_DEVICE(cq->ctrl), cq->ei_addr, cq->head,
1478                    MEMTXATTRS_UNSPECIFIED);
1479 }
1480 
1481 static void nvme_update_cq_head(NvmeCQueue *cq)
1482 {
1483     ldl_le_pci_dma(PCI_DEVICE(cq->ctrl), cq->db_addr, &cq->head,
1484                    MEMTXATTRS_UNSPECIFIED);
1485 
1486     trace_pci_nvme_update_cq_head(cq->cqid, cq->head);
1487 }
1488 
1489 static void nvme_post_cqes(void *opaque)
1490 {
1491     NvmeCQueue *cq = opaque;
1492     NvmeCtrl *n = cq->ctrl;
1493     NvmeRequest *req, *next;
1494     bool pending = cq->head != cq->tail;
1495     int ret;
1496 
1497     QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
1498         NvmeSQueue *sq;
1499         hwaddr addr;
1500 
1501         if (n->dbbuf_enabled) {
1502             nvme_update_cq_eventidx(cq);
1503             nvme_update_cq_head(cq);
1504         }
1505 
1506         if (nvme_cq_full(cq)) {
1507             break;
1508         }
1509 
1510         sq = req->sq;
1511         req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
1512         req->cqe.sq_id = cpu_to_le16(sq->sqid);
1513         req->cqe.sq_head = cpu_to_le16(sq->head);
1514         addr = cq->dma_addr + (cq->tail << NVME_CQES);
1515         ret = pci_dma_write(PCI_DEVICE(n), addr, (void *)&req->cqe,
1516                             sizeof(req->cqe));
1517         if (ret) {
1518             trace_pci_nvme_err_addr_write(addr);
1519             trace_pci_nvme_err_cfs();
1520             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
1521             break;
1522         }
1523         QTAILQ_REMOVE(&cq->req_list, req, entry);
1524         nvme_inc_cq_tail(cq);
1525         nvme_sg_unmap(&req->sg);
1526         QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
1527     }
1528     if (cq->tail != cq->head) {
1529         if (cq->irq_enabled && !pending) {
1530             n->cq_pending++;
1531         }
1532 
1533         nvme_irq_assert(n, cq);
1534     }
1535 }
1536 
1537 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
1538 {
1539     assert(cq->cqid == req->sq->cqid);
1540     trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
1541                                           le32_to_cpu(req->cqe.result),
1542                                           le32_to_cpu(req->cqe.dw1),
1543                                           req->status);
1544 
1545     if (req->status) {
1546         trace_pci_nvme_err_req_status(nvme_cid(req), nvme_nsid(req->ns),
1547                                       req->status, req->cmd.opcode);
1548     }
1549 
1550     QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
1551     QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
1552 
1553     qemu_bh_schedule(cq->bh);
1554 }
1555 
1556 static void nvme_process_aers(void *opaque)
1557 {
1558     NvmeCtrl *n = opaque;
1559     NvmeAsyncEvent *event, *next;
1560 
1561     trace_pci_nvme_process_aers(n->aer_queued);
1562 
1563     QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
1564         NvmeRequest *req;
1565         NvmeAerResult *result;
1566 
1567         /* can't post cqe if there is nothing to complete */
1568         if (!n->outstanding_aers) {
1569             trace_pci_nvme_no_outstanding_aers();
1570             break;
1571         }
1572 
1573         /* ignore if masked (cqe posted, but event not cleared) */
1574         if (n->aer_mask & (1 << event->result.event_type)) {
1575             trace_pci_nvme_aer_masked(event->result.event_type, n->aer_mask);
1576             continue;
1577         }
1578 
1579         QTAILQ_REMOVE(&n->aer_queue, event, entry);
1580         n->aer_queued--;
1581 
1582         n->aer_mask |= 1 << event->result.event_type;
1583         n->outstanding_aers--;
1584 
1585         req = n->aer_reqs[n->outstanding_aers];
1586 
1587         result = (NvmeAerResult *) &req->cqe.result;
1588         result->event_type = event->result.event_type;
1589         result->event_info = event->result.event_info;
1590         result->log_page = event->result.log_page;
1591         g_free(event);
1592 
1593         trace_pci_nvme_aer_post_cqe(result->event_type, result->event_info,
1594                                     result->log_page);
1595 
1596         nvme_enqueue_req_completion(&n->admin_cq, req);
1597     }
1598 }
1599 
1600 static void nvme_enqueue_event(NvmeCtrl *n, uint8_t event_type,
1601                                uint8_t event_info, uint8_t log_page)
1602 {
1603     NvmeAsyncEvent *event;
1604 
1605     trace_pci_nvme_enqueue_event(event_type, event_info, log_page);
1606 
1607     if (n->aer_queued == n->params.aer_max_queued) {
1608         trace_pci_nvme_enqueue_event_noqueue(n->aer_queued);
1609         return;
1610     }
1611 
1612     event = g_new(NvmeAsyncEvent, 1);
1613     event->result = (NvmeAerResult) {
1614         .event_type = event_type,
1615         .event_info = event_info,
1616         .log_page   = log_page,
1617     };
1618 
1619     QTAILQ_INSERT_TAIL(&n->aer_queue, event, entry);
1620     n->aer_queued++;
1621 
1622     nvme_process_aers(n);
1623 }
1624 
1625 static void nvme_smart_event(NvmeCtrl *n, uint8_t event)
1626 {
1627     uint8_t aer_info;
1628 
1629     /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
1630     if (!(NVME_AEC_SMART(n->features.async_config) & event)) {
1631         return;
1632     }
1633 
1634     switch (event) {
1635     case NVME_SMART_SPARE:
1636         aer_info = NVME_AER_INFO_SMART_SPARE_THRESH;
1637         break;
1638     case NVME_SMART_TEMPERATURE:
1639         aer_info = NVME_AER_INFO_SMART_TEMP_THRESH;
1640         break;
1641     case NVME_SMART_RELIABILITY:
1642     case NVME_SMART_MEDIA_READ_ONLY:
1643     case NVME_SMART_FAILED_VOLATILE_MEDIA:
1644     case NVME_SMART_PMR_UNRELIABLE:
1645         aer_info = NVME_AER_INFO_SMART_RELIABILITY;
1646         break;
1647     default:
1648         return;
1649     }
1650 
1651     nvme_enqueue_event(n, NVME_AER_TYPE_SMART, aer_info, NVME_LOG_SMART_INFO);
1652 }
1653 
1654 static void nvme_clear_events(NvmeCtrl *n, uint8_t event_type)
1655 {
1656     NvmeAsyncEvent *event, *next;
1657 
1658     n->aer_mask &= ~(1 << event_type);
1659 
1660     QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
1661         if (event->result.event_type == event_type) {
1662             QTAILQ_REMOVE(&n->aer_queue, event, entry);
1663             n->aer_queued--;
1664             g_free(event);
1665         }
1666     }
1667 }
1668 
1669 static inline uint16_t nvme_check_mdts(NvmeCtrl *n, size_t len)
1670 {
1671     uint8_t mdts = n->params.mdts;
1672 
1673     if (mdts && len > n->page_size << mdts) {
1674         trace_pci_nvme_err_mdts(len);
1675         return NVME_INVALID_FIELD | NVME_DNR;
1676     }
1677 
1678     return NVME_SUCCESS;
1679 }
1680 
1681 static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
1682                                          uint32_t nlb)
1683 {
1684     uint64_t nsze = le64_to_cpu(ns->id_ns.nsze);
1685 
1686     if (unlikely(UINT64_MAX - slba < nlb || slba + nlb > nsze)) {
1687         trace_pci_nvme_err_invalid_lba_range(slba, nlb, nsze);
1688         return NVME_LBA_RANGE | NVME_DNR;
1689     }
1690 
1691     return NVME_SUCCESS;
1692 }
1693 
1694 static int nvme_block_status_all(NvmeNamespace *ns, uint64_t slba,
1695                                  uint32_t nlb, int flags)
1696 {
1697     BlockDriverState *bs = blk_bs(ns->blkconf.blk);
1698 
1699     int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
1700     int64_t offset = nvme_l2b(ns, slba);
1701     int ret;
1702 
1703     /*
1704      * `pnum` holds the number of bytes after offset that shares the same
1705      * allocation status as the byte at offset. If `pnum` is different from
1706      * `bytes`, we should check the allocation status of the next range and
1707      * continue this until all bytes have been checked.
1708      */
1709     do {
1710         bytes -= pnum;
1711 
1712         ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
1713         if (ret < 0) {
1714             return ret;
1715         }
1716 
1717 
1718         trace_pci_nvme_block_status(offset, bytes, pnum, ret,
1719                                     !!(ret & BDRV_BLOCK_ZERO));
1720 
1721         if (!(ret & flags)) {
1722             return 1;
1723         }
1724 
1725         offset += pnum;
1726     } while (pnum != bytes);
1727 
1728     return 0;
1729 }
1730 
1731 static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
1732                                  uint32_t nlb)
1733 {
1734     int ret;
1735     Error *err = NULL;
1736 
1737     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_DATA);
1738     if (ret) {
1739         if (ret < 0) {
1740             error_setg_errno(&err, -ret, "unable to get block status");
1741             error_report_err(err);
1742 
1743             return NVME_INTERNAL_DEV_ERROR;
1744         }
1745 
1746         return NVME_DULB;
1747     }
1748 
1749     return NVME_SUCCESS;
1750 }
1751 
1752 static void nvme_aio_err(NvmeRequest *req, int ret)
1753 {
1754     uint16_t status = NVME_SUCCESS;
1755     Error *local_err = NULL;
1756 
1757     switch (req->cmd.opcode) {
1758     case NVME_CMD_READ:
1759         status = NVME_UNRECOVERED_READ;
1760         break;
1761     case NVME_CMD_FLUSH:
1762     case NVME_CMD_WRITE:
1763     case NVME_CMD_WRITE_ZEROES:
1764     case NVME_CMD_ZONE_APPEND:
1765     case NVME_CMD_COPY:
1766         status = NVME_WRITE_FAULT;
1767         break;
1768     default:
1769         status = NVME_INTERNAL_DEV_ERROR;
1770         break;
1771     }
1772 
1773     if (ret == -ECANCELED) {
1774         status = NVME_CMD_ABORT_REQ;
1775     }
1776 
1777     trace_pci_nvme_err_aio(nvme_cid(req), strerror(-ret), status);
1778 
1779     error_setg_errno(&local_err, -ret, "aio failed");
1780     error_report_err(local_err);
1781 
1782     /*
1783      * Set the command status code to the first encountered error but allow a
1784      * subsequent Internal Device Error to trump it.
1785      */
1786     if (req->status && status != NVME_INTERNAL_DEV_ERROR) {
1787         return;
1788     }
1789 
1790     req->status = status;
1791 }
1792 
1793 static inline uint32_t nvme_zone_idx(NvmeNamespace *ns, uint64_t slba)
1794 {
1795     return ns->zone_size_log2 > 0 ? slba >> ns->zone_size_log2 :
1796                                     slba / ns->zone_size;
1797 }
1798 
1799 static inline NvmeZone *nvme_get_zone_by_slba(NvmeNamespace *ns, uint64_t slba)
1800 {
1801     uint32_t zone_idx = nvme_zone_idx(ns, slba);
1802 
1803     if (zone_idx >= ns->num_zones) {
1804         return NULL;
1805     }
1806 
1807     return &ns->zone_array[zone_idx];
1808 }
1809 
1810 static uint16_t nvme_check_zone_state_for_write(NvmeZone *zone)
1811 {
1812     uint64_t zslba = zone->d.zslba;
1813 
1814     switch (nvme_get_zone_state(zone)) {
1815     case NVME_ZONE_STATE_EMPTY:
1816     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1817     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1818     case NVME_ZONE_STATE_CLOSED:
1819         return NVME_SUCCESS;
1820     case NVME_ZONE_STATE_FULL:
1821         trace_pci_nvme_err_zone_is_full(zslba);
1822         return NVME_ZONE_FULL;
1823     case NVME_ZONE_STATE_OFFLINE:
1824         trace_pci_nvme_err_zone_is_offline(zslba);
1825         return NVME_ZONE_OFFLINE;
1826     case NVME_ZONE_STATE_READ_ONLY:
1827         trace_pci_nvme_err_zone_is_read_only(zslba);
1828         return NVME_ZONE_READ_ONLY;
1829     default:
1830         g_assert_not_reached();
1831     }
1832 
1833     return NVME_INTERNAL_DEV_ERROR;
1834 }
1835 
1836 static uint16_t nvme_check_zone_write(NvmeNamespace *ns, NvmeZone *zone,
1837                                       uint64_t slba, uint32_t nlb)
1838 {
1839     uint64_t zcap = nvme_zone_wr_boundary(zone);
1840     uint16_t status;
1841 
1842     status = nvme_check_zone_state_for_write(zone);
1843     if (status) {
1844         return status;
1845     }
1846 
1847     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1848         uint64_t ezrwa = zone->w_ptr + 2 * ns->zns.zrwas;
1849 
1850         if (slba < zone->w_ptr || slba + nlb > ezrwa) {
1851             trace_pci_nvme_err_zone_invalid_write(slba, zone->w_ptr);
1852             return NVME_ZONE_INVALID_WRITE;
1853         }
1854     } else {
1855         if (unlikely(slba != zone->w_ptr)) {
1856             trace_pci_nvme_err_write_not_at_wp(slba, zone->d.zslba,
1857                                                zone->w_ptr);
1858             return NVME_ZONE_INVALID_WRITE;
1859         }
1860     }
1861 
1862     if (unlikely((slba + nlb) > zcap)) {
1863         trace_pci_nvme_err_zone_boundary(slba, nlb, zcap);
1864         return NVME_ZONE_BOUNDARY_ERROR;
1865     }
1866 
1867     return NVME_SUCCESS;
1868 }
1869 
1870 static uint16_t nvme_check_zone_state_for_read(NvmeZone *zone)
1871 {
1872     switch (nvme_get_zone_state(zone)) {
1873     case NVME_ZONE_STATE_EMPTY:
1874     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1875     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1876     case NVME_ZONE_STATE_FULL:
1877     case NVME_ZONE_STATE_CLOSED:
1878     case NVME_ZONE_STATE_READ_ONLY:
1879         return NVME_SUCCESS;
1880     case NVME_ZONE_STATE_OFFLINE:
1881         trace_pci_nvme_err_zone_is_offline(zone->d.zslba);
1882         return NVME_ZONE_OFFLINE;
1883     default:
1884         g_assert_not_reached();
1885     }
1886 
1887     return NVME_INTERNAL_DEV_ERROR;
1888 }
1889 
1890 static uint16_t nvme_check_zone_read(NvmeNamespace *ns, uint64_t slba,
1891                                      uint32_t nlb)
1892 {
1893     NvmeZone *zone;
1894     uint64_t bndry, end;
1895     uint16_t status;
1896 
1897     zone = nvme_get_zone_by_slba(ns, slba);
1898     assert(zone);
1899 
1900     bndry = nvme_zone_rd_boundary(ns, zone);
1901     end = slba + nlb;
1902 
1903     status = nvme_check_zone_state_for_read(zone);
1904     if (status) {
1905         ;
1906     } else if (unlikely(end > bndry)) {
1907         if (!ns->params.cross_zone_read) {
1908             status = NVME_ZONE_BOUNDARY_ERROR;
1909         } else {
1910             /*
1911              * Read across zone boundary - check that all subsequent
1912              * zones that are being read have an appropriate state.
1913              */
1914             do {
1915                 zone++;
1916                 status = nvme_check_zone_state_for_read(zone);
1917                 if (status) {
1918                     break;
1919                 }
1920             } while (end > nvme_zone_rd_boundary(ns, zone));
1921         }
1922     }
1923 
1924     return status;
1925 }
1926 
1927 static uint16_t nvme_zrm_finish(NvmeNamespace *ns, NvmeZone *zone)
1928 {
1929     switch (nvme_get_zone_state(zone)) {
1930     case NVME_ZONE_STATE_FULL:
1931         return NVME_SUCCESS;
1932 
1933     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1934     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1935         nvme_aor_dec_open(ns);
1936         /* fallthrough */
1937     case NVME_ZONE_STATE_CLOSED:
1938         nvme_aor_dec_active(ns);
1939 
1940         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1941             zone->d.za &= ~NVME_ZA_ZRWA_VALID;
1942             if (ns->params.numzrwa) {
1943                 ns->zns.numzrwa++;
1944             }
1945         }
1946 
1947         /* fallthrough */
1948     case NVME_ZONE_STATE_EMPTY:
1949         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_FULL);
1950         return NVME_SUCCESS;
1951 
1952     default:
1953         return NVME_ZONE_INVAL_TRANSITION;
1954     }
1955 }
1956 
1957 static uint16_t nvme_zrm_close(NvmeNamespace *ns, NvmeZone *zone)
1958 {
1959     switch (nvme_get_zone_state(zone)) {
1960     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1961     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1962         nvme_aor_dec_open(ns);
1963         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
1964         /* fall through */
1965     case NVME_ZONE_STATE_CLOSED:
1966         return NVME_SUCCESS;
1967 
1968     default:
1969         return NVME_ZONE_INVAL_TRANSITION;
1970     }
1971 }
1972 
1973 static uint16_t nvme_zrm_reset(NvmeNamespace *ns, NvmeZone *zone)
1974 {
1975     switch (nvme_get_zone_state(zone)) {
1976     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1977     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1978         nvme_aor_dec_open(ns);
1979         /* fallthrough */
1980     case NVME_ZONE_STATE_CLOSED:
1981         nvme_aor_dec_active(ns);
1982 
1983         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1984             if (ns->params.numzrwa) {
1985                 ns->zns.numzrwa++;
1986             }
1987         }
1988 
1989         /* fallthrough */
1990     case NVME_ZONE_STATE_FULL:
1991         zone->w_ptr = zone->d.zslba;
1992         zone->d.wp = zone->w_ptr;
1993         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
1994         /* fallthrough */
1995     case NVME_ZONE_STATE_EMPTY:
1996         return NVME_SUCCESS;
1997 
1998     default:
1999         return NVME_ZONE_INVAL_TRANSITION;
2000     }
2001 }
2002 
2003 static void nvme_zrm_auto_transition_zone(NvmeNamespace *ns)
2004 {
2005     NvmeZone *zone;
2006 
2007     if (ns->params.max_open_zones &&
2008         ns->nr_open_zones == ns->params.max_open_zones) {
2009         zone = QTAILQ_FIRST(&ns->imp_open_zones);
2010         if (zone) {
2011             /*
2012              * Automatically close this implicitly open zone.
2013              */
2014             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
2015             nvme_zrm_close(ns, zone);
2016         }
2017     }
2018 }
2019 
2020 enum {
2021     NVME_ZRM_AUTO = 1 << 0,
2022     NVME_ZRM_ZRWA = 1 << 1,
2023 };
2024 
2025 static uint16_t nvme_zrm_open_flags(NvmeCtrl *n, NvmeNamespace *ns,
2026                                     NvmeZone *zone, int flags)
2027 {
2028     int act = 0;
2029     uint16_t status;
2030 
2031     switch (nvme_get_zone_state(zone)) {
2032     case NVME_ZONE_STATE_EMPTY:
2033         act = 1;
2034 
2035         /* fallthrough */
2036 
2037     case NVME_ZONE_STATE_CLOSED:
2038         if (n->params.auto_transition_zones) {
2039             nvme_zrm_auto_transition_zone(ns);
2040         }
2041         status = nvme_zns_check_resources(ns, act, 1,
2042                                           (flags & NVME_ZRM_ZRWA) ? 1 : 0);
2043         if (status) {
2044             return status;
2045         }
2046 
2047         if (act) {
2048             nvme_aor_inc_active(ns);
2049         }
2050 
2051         nvme_aor_inc_open(ns);
2052 
2053         if (flags & NVME_ZRM_AUTO) {
2054             nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_IMPLICITLY_OPEN);
2055             return NVME_SUCCESS;
2056         }
2057 
2058         /* fallthrough */
2059 
2060     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
2061         if (flags & NVME_ZRM_AUTO) {
2062             return NVME_SUCCESS;
2063         }
2064 
2065         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EXPLICITLY_OPEN);
2066 
2067         /* fallthrough */
2068 
2069     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
2070         if (flags & NVME_ZRM_ZRWA) {
2071             ns->zns.numzrwa--;
2072 
2073             zone->d.za |= NVME_ZA_ZRWA_VALID;
2074         }
2075 
2076         return NVME_SUCCESS;
2077 
2078     default:
2079         return NVME_ZONE_INVAL_TRANSITION;
2080     }
2081 }
2082 
2083 static inline uint16_t nvme_zrm_auto(NvmeCtrl *n, NvmeNamespace *ns,
2084                                      NvmeZone *zone)
2085 {
2086     return nvme_zrm_open_flags(n, ns, zone, NVME_ZRM_AUTO);
2087 }
2088 
2089 static void nvme_advance_zone_wp(NvmeNamespace *ns, NvmeZone *zone,
2090                                  uint32_t nlb)
2091 {
2092     zone->d.wp += nlb;
2093 
2094     if (zone->d.wp == nvme_zone_wr_boundary(zone)) {
2095         nvme_zrm_finish(ns, zone);
2096     }
2097 }
2098 
2099 static void nvme_zoned_zrwa_implicit_flush(NvmeNamespace *ns, NvmeZone *zone,
2100                                            uint32_t nlbc)
2101 {
2102     uint16_t nzrwafgs = DIV_ROUND_UP(nlbc, ns->zns.zrwafg);
2103 
2104     nlbc = nzrwafgs * ns->zns.zrwafg;
2105 
2106     trace_pci_nvme_zoned_zrwa_implicit_flush(zone->d.zslba, nlbc);
2107 
2108     zone->w_ptr += nlbc;
2109 
2110     nvme_advance_zone_wp(ns, zone, nlbc);
2111 }
2112 
2113 static void nvme_finalize_zoned_write(NvmeNamespace *ns, NvmeRequest *req)
2114 {
2115     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2116     NvmeZone *zone;
2117     uint64_t slba;
2118     uint32_t nlb;
2119 
2120     slba = le64_to_cpu(rw->slba);
2121     nlb = le16_to_cpu(rw->nlb) + 1;
2122     zone = nvme_get_zone_by_slba(ns, slba);
2123     assert(zone);
2124 
2125     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
2126         uint64_t ezrwa = zone->w_ptr + ns->zns.zrwas - 1;
2127         uint64_t elba = slba + nlb - 1;
2128 
2129         if (elba > ezrwa) {
2130             nvme_zoned_zrwa_implicit_flush(ns, zone, elba - ezrwa);
2131         }
2132 
2133         return;
2134     }
2135 
2136     nvme_advance_zone_wp(ns, zone, nlb);
2137 }
2138 
2139 static inline bool nvme_is_write(NvmeRequest *req)
2140 {
2141     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2142 
2143     return rw->opcode == NVME_CMD_WRITE ||
2144            rw->opcode == NVME_CMD_ZONE_APPEND ||
2145            rw->opcode == NVME_CMD_WRITE_ZEROES;
2146 }
2147 
2148 static void nvme_misc_cb(void *opaque, int ret)
2149 {
2150     NvmeRequest *req = opaque;
2151 
2152     trace_pci_nvme_misc_cb(nvme_cid(req));
2153 
2154     if (ret) {
2155         nvme_aio_err(req, ret);
2156     }
2157 
2158     nvme_enqueue_req_completion(nvme_cq(req), req);
2159 }
2160 
2161 void nvme_rw_complete_cb(void *opaque, int ret)
2162 {
2163     NvmeRequest *req = opaque;
2164     NvmeNamespace *ns = req->ns;
2165     BlockBackend *blk = ns->blkconf.blk;
2166     BlockAcctCookie *acct = &req->acct;
2167     BlockAcctStats *stats = blk_get_stats(blk);
2168 
2169     trace_pci_nvme_rw_complete_cb(nvme_cid(req), blk_name(blk));
2170 
2171     if (ret) {
2172         block_acct_failed(stats, acct);
2173         nvme_aio_err(req, ret);
2174     } else {
2175         block_acct_done(stats, acct);
2176     }
2177 
2178     if (ns->params.zoned && nvme_is_write(req)) {
2179         nvme_finalize_zoned_write(ns, req);
2180     }
2181 
2182     nvme_enqueue_req_completion(nvme_cq(req), req);
2183 }
2184 
2185 static void nvme_rw_cb(void *opaque, int ret)
2186 {
2187     NvmeRequest *req = opaque;
2188     NvmeNamespace *ns = req->ns;
2189 
2190     BlockBackend *blk = ns->blkconf.blk;
2191 
2192     trace_pci_nvme_rw_cb(nvme_cid(req), blk_name(blk));
2193 
2194     if (ret) {
2195         goto out;
2196     }
2197 
2198     if (ns->lbaf.ms) {
2199         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2200         uint64_t slba = le64_to_cpu(rw->slba);
2201         uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
2202         uint64_t offset = nvme_moff(ns, slba);
2203 
2204         if (req->cmd.opcode == NVME_CMD_WRITE_ZEROES) {
2205             size_t mlen = nvme_m2b(ns, nlb);
2206 
2207             req->aiocb = blk_aio_pwrite_zeroes(blk, offset, mlen,
2208                                                BDRV_REQ_MAY_UNMAP,
2209                                                nvme_rw_complete_cb, req);
2210             return;
2211         }
2212 
2213         if (nvme_ns_ext(ns) || req->cmd.mptr) {
2214             uint16_t status;
2215 
2216             nvme_sg_unmap(&req->sg);
2217             status = nvme_map_mdata(nvme_ctrl(req), nlb, req);
2218             if (status) {
2219                 ret = -EFAULT;
2220                 goto out;
2221             }
2222 
2223             if (req->cmd.opcode == NVME_CMD_READ) {
2224                 return nvme_blk_read(blk, offset, 1, nvme_rw_complete_cb, req);
2225             }
2226 
2227             return nvme_blk_write(blk, offset, 1, nvme_rw_complete_cb, req);
2228         }
2229     }
2230 
2231 out:
2232     nvme_rw_complete_cb(req, ret);
2233 }
2234 
2235 static void nvme_verify_cb(void *opaque, int ret)
2236 {
2237     NvmeBounceContext *ctx = opaque;
2238     NvmeRequest *req = ctx->req;
2239     NvmeNamespace *ns = req->ns;
2240     BlockBackend *blk = ns->blkconf.blk;
2241     BlockAcctCookie *acct = &req->acct;
2242     BlockAcctStats *stats = blk_get_stats(blk);
2243     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2244     uint64_t slba = le64_to_cpu(rw->slba);
2245     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2246     uint16_t apptag = le16_to_cpu(rw->apptag);
2247     uint16_t appmask = le16_to_cpu(rw->appmask);
2248     uint64_t reftag = le32_to_cpu(rw->reftag);
2249     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2250     uint16_t status;
2251 
2252     reftag |= cdw3 << 32;
2253 
2254     trace_pci_nvme_verify_cb(nvme_cid(req), prinfo, apptag, appmask, reftag);
2255 
2256     if (ret) {
2257         block_acct_failed(stats, acct);
2258         nvme_aio_err(req, ret);
2259         goto out;
2260     }
2261 
2262     block_acct_done(stats, acct);
2263 
2264     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2265         status = nvme_dif_mangle_mdata(ns, ctx->mdata.bounce,
2266                                        ctx->mdata.iov.size, slba);
2267         if (status) {
2268             req->status = status;
2269             goto out;
2270         }
2271 
2272         req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2273                                      ctx->mdata.bounce, ctx->mdata.iov.size,
2274                                      prinfo, slba, apptag, appmask, &reftag);
2275     }
2276 
2277 out:
2278     qemu_iovec_destroy(&ctx->data.iov);
2279     g_free(ctx->data.bounce);
2280 
2281     qemu_iovec_destroy(&ctx->mdata.iov);
2282     g_free(ctx->mdata.bounce);
2283 
2284     g_free(ctx);
2285 
2286     nvme_enqueue_req_completion(nvme_cq(req), req);
2287 }
2288 
2289 
2290 static void nvme_verify_mdata_in_cb(void *opaque, int ret)
2291 {
2292     NvmeBounceContext *ctx = opaque;
2293     NvmeRequest *req = ctx->req;
2294     NvmeNamespace *ns = req->ns;
2295     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2296     uint64_t slba = le64_to_cpu(rw->slba);
2297     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2298     size_t mlen = nvme_m2b(ns, nlb);
2299     uint64_t offset = nvme_moff(ns, slba);
2300     BlockBackend *blk = ns->blkconf.blk;
2301 
2302     trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req), blk_name(blk));
2303 
2304     if (ret) {
2305         goto out;
2306     }
2307 
2308     ctx->mdata.bounce = g_malloc(mlen);
2309 
2310     qemu_iovec_reset(&ctx->mdata.iov);
2311     qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2312 
2313     req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2314                                 nvme_verify_cb, ctx);
2315     return;
2316 
2317 out:
2318     nvme_verify_cb(ctx, ret);
2319 }
2320 
2321 struct nvme_compare_ctx {
2322     struct {
2323         QEMUIOVector iov;
2324         uint8_t *bounce;
2325     } data;
2326 
2327     struct {
2328         QEMUIOVector iov;
2329         uint8_t *bounce;
2330     } mdata;
2331 };
2332 
2333 static void nvme_compare_mdata_cb(void *opaque, int ret)
2334 {
2335     NvmeRequest *req = opaque;
2336     NvmeNamespace *ns = req->ns;
2337     NvmeCtrl *n = nvme_ctrl(req);
2338     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2339     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2340     uint16_t apptag = le16_to_cpu(rw->apptag);
2341     uint16_t appmask = le16_to_cpu(rw->appmask);
2342     uint64_t reftag = le32_to_cpu(rw->reftag);
2343     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2344     struct nvme_compare_ctx *ctx = req->opaque;
2345     g_autofree uint8_t *buf = NULL;
2346     BlockBackend *blk = ns->blkconf.blk;
2347     BlockAcctCookie *acct = &req->acct;
2348     BlockAcctStats *stats = blk_get_stats(blk);
2349     uint16_t status = NVME_SUCCESS;
2350 
2351     reftag |= cdw3 << 32;
2352 
2353     trace_pci_nvme_compare_mdata_cb(nvme_cid(req));
2354 
2355     if (ret) {
2356         block_acct_failed(stats, acct);
2357         nvme_aio_err(req, ret);
2358         goto out;
2359     }
2360 
2361     buf = g_malloc(ctx->mdata.iov.size);
2362 
2363     status = nvme_bounce_mdata(n, buf, ctx->mdata.iov.size,
2364                                NVME_TX_DIRECTION_TO_DEVICE, req);
2365     if (status) {
2366         req->status = status;
2367         goto out;
2368     }
2369 
2370     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2371         uint64_t slba = le64_to_cpu(rw->slba);
2372         uint8_t *bufp;
2373         uint8_t *mbufp = ctx->mdata.bounce;
2374         uint8_t *end = mbufp + ctx->mdata.iov.size;
2375         int16_t pil = 0;
2376 
2377         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2378                                 ctx->mdata.bounce, ctx->mdata.iov.size, prinfo,
2379                                 slba, apptag, appmask, &reftag);
2380         if (status) {
2381             req->status = status;
2382             goto out;
2383         }
2384 
2385         /*
2386          * When formatted with protection information, do not compare the DIF
2387          * tuple.
2388          */
2389         if (!(ns->id_ns.dps & NVME_ID_NS_DPS_FIRST_EIGHT)) {
2390             pil = ns->lbaf.ms - nvme_pi_tuple_size(ns);
2391         }
2392 
2393         for (bufp = buf; mbufp < end; bufp += ns->lbaf.ms, mbufp += ns->lbaf.ms) {
2394             if (memcmp(bufp + pil, mbufp + pil, ns->lbaf.ms - pil)) {
2395                 req->status = NVME_CMP_FAILURE | NVME_DNR;
2396                 goto out;
2397             }
2398         }
2399 
2400         goto out;
2401     }
2402 
2403     if (memcmp(buf, ctx->mdata.bounce, ctx->mdata.iov.size)) {
2404         req->status = NVME_CMP_FAILURE | NVME_DNR;
2405         goto out;
2406     }
2407 
2408     block_acct_done(stats, acct);
2409 
2410 out:
2411     qemu_iovec_destroy(&ctx->data.iov);
2412     g_free(ctx->data.bounce);
2413 
2414     qemu_iovec_destroy(&ctx->mdata.iov);
2415     g_free(ctx->mdata.bounce);
2416 
2417     g_free(ctx);
2418 
2419     nvme_enqueue_req_completion(nvme_cq(req), req);
2420 }
2421 
2422 static void nvme_compare_data_cb(void *opaque, int ret)
2423 {
2424     NvmeRequest *req = opaque;
2425     NvmeCtrl *n = nvme_ctrl(req);
2426     NvmeNamespace *ns = req->ns;
2427     BlockBackend *blk = ns->blkconf.blk;
2428     BlockAcctCookie *acct = &req->acct;
2429     BlockAcctStats *stats = blk_get_stats(blk);
2430 
2431     struct nvme_compare_ctx *ctx = req->opaque;
2432     g_autofree uint8_t *buf = NULL;
2433     uint16_t status;
2434 
2435     trace_pci_nvme_compare_data_cb(nvme_cid(req));
2436 
2437     if (ret) {
2438         block_acct_failed(stats, acct);
2439         nvme_aio_err(req, ret);
2440         goto out;
2441     }
2442 
2443     buf = g_malloc(ctx->data.iov.size);
2444 
2445     status = nvme_bounce_data(n, buf, ctx->data.iov.size,
2446                               NVME_TX_DIRECTION_TO_DEVICE, req);
2447     if (status) {
2448         req->status = status;
2449         goto out;
2450     }
2451 
2452     if (memcmp(buf, ctx->data.bounce, ctx->data.iov.size)) {
2453         req->status = NVME_CMP_FAILURE | NVME_DNR;
2454         goto out;
2455     }
2456 
2457     if (ns->lbaf.ms) {
2458         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2459         uint64_t slba = le64_to_cpu(rw->slba);
2460         uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2461         size_t mlen = nvme_m2b(ns, nlb);
2462         uint64_t offset = nvme_moff(ns, slba);
2463 
2464         ctx->mdata.bounce = g_malloc(mlen);
2465 
2466         qemu_iovec_init(&ctx->mdata.iov, 1);
2467         qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2468 
2469         req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2470                                     nvme_compare_mdata_cb, req);
2471         return;
2472     }
2473 
2474     block_acct_done(stats, acct);
2475 
2476 out:
2477     qemu_iovec_destroy(&ctx->data.iov);
2478     g_free(ctx->data.bounce);
2479     g_free(ctx);
2480 
2481     nvme_enqueue_req_completion(nvme_cq(req), req);
2482 }
2483 
2484 typedef struct NvmeDSMAIOCB {
2485     BlockAIOCB common;
2486     BlockAIOCB *aiocb;
2487     NvmeRequest *req;
2488     int ret;
2489 
2490     NvmeDsmRange *range;
2491     unsigned int nr;
2492     unsigned int idx;
2493 } NvmeDSMAIOCB;
2494 
2495 static void nvme_dsm_cancel(BlockAIOCB *aiocb)
2496 {
2497     NvmeDSMAIOCB *iocb = container_of(aiocb, NvmeDSMAIOCB, common);
2498 
2499     /* break nvme_dsm_cb loop */
2500     iocb->idx = iocb->nr;
2501     iocb->ret = -ECANCELED;
2502 
2503     if (iocb->aiocb) {
2504         blk_aio_cancel_async(iocb->aiocb);
2505         iocb->aiocb = NULL;
2506     } else {
2507         /*
2508          * We only reach this if nvme_dsm_cancel() has already been called or
2509          * the command ran to completion.
2510          */
2511         assert(iocb->idx == iocb->nr);
2512     }
2513 }
2514 
2515 static const AIOCBInfo nvme_dsm_aiocb_info = {
2516     .aiocb_size   = sizeof(NvmeDSMAIOCB),
2517     .cancel_async = nvme_dsm_cancel,
2518 };
2519 
2520 static void nvme_dsm_cb(void *opaque, int ret);
2521 
2522 static void nvme_dsm_md_cb(void *opaque, int ret)
2523 {
2524     NvmeDSMAIOCB *iocb = opaque;
2525     NvmeRequest *req = iocb->req;
2526     NvmeNamespace *ns = req->ns;
2527     NvmeDsmRange *range;
2528     uint64_t slba;
2529     uint32_t nlb;
2530 
2531     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
2532         goto done;
2533     }
2534 
2535     range = &iocb->range[iocb->idx - 1];
2536     slba = le64_to_cpu(range->slba);
2537     nlb = le32_to_cpu(range->nlb);
2538 
2539     /*
2540      * Check that all block were discarded (zeroed); otherwise we do not zero
2541      * the metadata.
2542      */
2543 
2544     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_ZERO);
2545     if (ret) {
2546         if (ret < 0) {
2547             goto done;
2548         }
2549 
2550         nvme_dsm_cb(iocb, 0);
2551         return;
2552     }
2553 
2554     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_moff(ns, slba),
2555                                         nvme_m2b(ns, nlb), BDRV_REQ_MAY_UNMAP,
2556                                         nvme_dsm_cb, iocb);
2557     return;
2558 
2559 done:
2560     nvme_dsm_cb(iocb, ret);
2561 }
2562 
2563 static void nvme_dsm_cb(void *opaque, int ret)
2564 {
2565     NvmeDSMAIOCB *iocb = opaque;
2566     NvmeRequest *req = iocb->req;
2567     NvmeCtrl *n = nvme_ctrl(req);
2568     NvmeNamespace *ns = req->ns;
2569     NvmeDsmRange *range;
2570     uint64_t slba;
2571     uint32_t nlb;
2572 
2573     if (iocb->ret < 0) {
2574         goto done;
2575     } else if (ret < 0) {
2576         iocb->ret = ret;
2577         goto done;
2578     }
2579 
2580 next:
2581     if (iocb->idx == iocb->nr) {
2582         goto done;
2583     }
2584 
2585     range = &iocb->range[iocb->idx++];
2586     slba = le64_to_cpu(range->slba);
2587     nlb = le32_to_cpu(range->nlb);
2588 
2589     trace_pci_nvme_dsm_deallocate(slba, nlb);
2590 
2591     if (nlb > n->dmrsl) {
2592         trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
2593         goto next;
2594     }
2595 
2596     if (nvme_check_bounds(ns, slba, nlb)) {
2597         trace_pci_nvme_err_invalid_lba_range(slba, nlb,
2598                                              ns->id_ns.nsze);
2599         goto next;
2600     }
2601 
2602     iocb->aiocb = blk_aio_pdiscard(ns->blkconf.blk, nvme_l2b(ns, slba),
2603                                    nvme_l2b(ns, nlb),
2604                                    nvme_dsm_md_cb, iocb);
2605     return;
2606 
2607 done:
2608     iocb->aiocb = NULL;
2609     iocb->common.cb(iocb->common.opaque, iocb->ret);
2610     g_free(iocb->range);
2611     qemu_aio_unref(iocb);
2612 }
2613 
2614 static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
2615 {
2616     NvmeNamespace *ns = req->ns;
2617     NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
2618     uint32_t attr = le32_to_cpu(dsm->attributes);
2619     uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
2620     uint16_t status = NVME_SUCCESS;
2621 
2622     trace_pci_nvme_dsm(nr, attr);
2623 
2624     if (attr & NVME_DSMGMT_AD) {
2625         NvmeDSMAIOCB *iocb = blk_aio_get(&nvme_dsm_aiocb_info, ns->blkconf.blk,
2626                                          nvme_misc_cb, req);
2627 
2628         iocb->req = req;
2629         iocb->ret = 0;
2630         iocb->range = g_new(NvmeDsmRange, nr);
2631         iocb->nr = nr;
2632         iocb->idx = 0;
2633 
2634         status = nvme_h2c(n, (uint8_t *)iocb->range, sizeof(NvmeDsmRange) * nr,
2635                           req);
2636         if (status) {
2637             g_free(iocb->range);
2638             qemu_aio_unref(iocb);
2639 
2640             return status;
2641         }
2642 
2643         req->aiocb = &iocb->common;
2644         nvme_dsm_cb(iocb, 0);
2645 
2646         return NVME_NO_COMPLETE;
2647     }
2648 
2649     return status;
2650 }
2651 
2652 static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
2653 {
2654     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2655     NvmeNamespace *ns = req->ns;
2656     BlockBackend *blk = ns->blkconf.blk;
2657     uint64_t slba = le64_to_cpu(rw->slba);
2658     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2659     size_t len = nvme_l2b(ns, nlb);
2660     size_t data_len = len;
2661     int64_t offset = nvme_l2b(ns, slba);
2662     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2663     uint32_t reftag = le32_to_cpu(rw->reftag);
2664     NvmeBounceContext *ctx = NULL;
2665     uint16_t status;
2666 
2667     trace_pci_nvme_verify(nvme_cid(req), nvme_nsid(ns), slba, nlb);
2668 
2669     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2670         status = nvme_check_prinfo(ns, prinfo, slba, reftag);
2671         if (status) {
2672             return status;
2673         }
2674 
2675         if (prinfo & NVME_PRINFO_PRACT) {
2676             return NVME_INVALID_PROT_INFO | NVME_DNR;
2677         }
2678     }
2679 
2680     if (nvme_ns_ext(ns) && !(NVME_ID_CTRL_CTRATT_MEM(n->id_ctrl.ctratt))) {
2681         data_len += nvme_m2b(ns, nlb);
2682     }
2683 
2684     if (data_len > (n->page_size << n->params.vsl)) {
2685         return NVME_INVALID_FIELD | NVME_DNR;
2686     }
2687 
2688     status = nvme_check_bounds(ns, slba, nlb);
2689     if (status) {
2690         return status;
2691     }
2692 
2693     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2694         status = nvme_check_dulbe(ns, slba, nlb);
2695         if (status) {
2696             return status;
2697         }
2698     }
2699 
2700     ctx = g_new0(NvmeBounceContext, 1);
2701     ctx->req = req;
2702 
2703     ctx->data.bounce = g_malloc(len);
2704 
2705     qemu_iovec_init(&ctx->data.iov, 1);
2706     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, len);
2707 
2708     block_acct_start(blk_get_stats(blk), &req->acct, ctx->data.iov.size,
2709                      BLOCK_ACCT_READ);
2710 
2711     req->aiocb = blk_aio_preadv(ns->blkconf.blk, offset, &ctx->data.iov, 0,
2712                                 nvme_verify_mdata_in_cb, ctx);
2713     return NVME_NO_COMPLETE;
2714 }
2715 
2716 typedef struct NvmeCopyAIOCB {
2717     BlockAIOCB common;
2718     BlockAIOCB *aiocb;
2719     NvmeRequest *req;
2720     NvmeCtrl *n;
2721     int ret;
2722 
2723     void *ranges;
2724     unsigned int format;
2725     int nr;
2726     int idx;
2727 
2728     uint8_t *bounce;
2729     QEMUIOVector iov;
2730     struct {
2731         BlockAcctCookie read;
2732         BlockAcctCookie write;
2733     } acct;
2734 
2735     uint64_t reftag;
2736     uint64_t slba;
2737 
2738     NvmeZone *zone;
2739     NvmeNamespace *sns;
2740     uint32_t tcl;
2741 } NvmeCopyAIOCB;
2742 
2743 static void nvme_copy_cancel(BlockAIOCB *aiocb)
2744 {
2745     NvmeCopyAIOCB *iocb = container_of(aiocb, NvmeCopyAIOCB, common);
2746 
2747     iocb->ret = -ECANCELED;
2748 
2749     if (iocb->aiocb) {
2750         blk_aio_cancel_async(iocb->aiocb);
2751         iocb->aiocb = NULL;
2752     }
2753 }
2754 
2755 static const AIOCBInfo nvme_copy_aiocb_info = {
2756     .aiocb_size   = sizeof(NvmeCopyAIOCB),
2757     .cancel_async = nvme_copy_cancel,
2758 };
2759 
2760 static void nvme_copy_done(NvmeCopyAIOCB *iocb)
2761 {
2762     NvmeRequest *req = iocb->req;
2763     NvmeNamespace *ns = req->ns;
2764     BlockAcctStats *stats = blk_get_stats(ns->blkconf.blk);
2765 
2766     if (iocb->idx != iocb->nr) {
2767         req->cqe.result = cpu_to_le32(iocb->idx);
2768     }
2769 
2770     qemu_iovec_destroy(&iocb->iov);
2771     g_free(iocb->bounce);
2772 
2773     if (iocb->ret < 0) {
2774         block_acct_failed(stats, &iocb->acct.read);
2775         block_acct_failed(stats, &iocb->acct.write);
2776     } else {
2777         block_acct_done(stats, &iocb->acct.read);
2778         block_acct_done(stats, &iocb->acct.write);
2779     }
2780 
2781     iocb->common.cb(iocb->common.opaque, iocb->ret);
2782     qemu_aio_unref(iocb);
2783 }
2784 
2785 static void nvme_do_copy(NvmeCopyAIOCB *iocb);
2786 
2787 static void nvme_copy_source_range_parse_format0_2(void *ranges,
2788                                                    int idx, uint64_t *slba,
2789                                                    uint32_t *nlb,
2790                                                    uint32_t *snsid,
2791                                                    uint16_t *apptag,
2792                                                    uint16_t *appmask,
2793                                                    uint64_t *reftag)
2794 {
2795     NvmeCopySourceRangeFormat0_2 *_ranges = ranges;
2796 
2797     if (snsid) {
2798         *snsid = le32_to_cpu(_ranges[idx].sparams);
2799     }
2800 
2801     if (slba) {
2802         *slba = le64_to_cpu(_ranges[idx].slba);
2803     }
2804 
2805     if (nlb) {
2806         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2807     }
2808 
2809     if (apptag) {
2810         *apptag = le16_to_cpu(_ranges[idx].apptag);
2811     }
2812 
2813     if (appmask) {
2814         *appmask = le16_to_cpu(_ranges[idx].appmask);
2815     }
2816 
2817     if (reftag) {
2818         *reftag = le32_to_cpu(_ranges[idx].reftag);
2819     }
2820 }
2821 
2822 static void nvme_copy_source_range_parse_format1_3(void *ranges, int idx,
2823                                                    uint64_t *slba,
2824                                                    uint32_t *nlb,
2825                                                    uint32_t *snsid,
2826                                                    uint16_t *apptag,
2827                                                    uint16_t *appmask,
2828                                                    uint64_t *reftag)
2829 {
2830     NvmeCopySourceRangeFormat1_3 *_ranges = ranges;
2831 
2832     if (snsid) {
2833         *snsid = le32_to_cpu(_ranges[idx].sparams);
2834     }
2835 
2836     if (slba) {
2837         *slba = le64_to_cpu(_ranges[idx].slba);
2838     }
2839 
2840     if (nlb) {
2841         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2842     }
2843 
2844     if (apptag) {
2845         *apptag = le16_to_cpu(_ranges[idx].apptag);
2846     }
2847 
2848     if (appmask) {
2849         *appmask = le16_to_cpu(_ranges[idx].appmask);
2850     }
2851 
2852     if (reftag) {
2853         *reftag = 0;
2854 
2855         *reftag |= (uint64_t)_ranges[idx].sr[4] << 40;
2856         *reftag |= (uint64_t)_ranges[idx].sr[5] << 32;
2857         *reftag |= (uint64_t)_ranges[idx].sr[6] << 24;
2858         *reftag |= (uint64_t)_ranges[idx].sr[7] << 16;
2859         *reftag |= (uint64_t)_ranges[idx].sr[8] << 8;
2860         *reftag |= (uint64_t)_ranges[idx].sr[9];
2861     }
2862 }
2863 
2864 static void nvme_copy_source_range_parse(void *ranges, int idx, uint8_t format,
2865                                          uint64_t *slba, uint32_t *nlb,
2866                                          uint32_t *snsid, uint16_t *apptag,
2867                                          uint16_t *appmask, uint64_t *reftag)
2868 {
2869     switch (format) {
2870     case NVME_COPY_FORMAT_0:
2871     case NVME_COPY_FORMAT_2:
2872         nvme_copy_source_range_parse_format0_2(ranges, idx, slba, nlb, snsid,
2873                                                apptag, appmask, reftag);
2874         break;
2875 
2876     case NVME_COPY_FORMAT_1:
2877     case NVME_COPY_FORMAT_3:
2878         nvme_copy_source_range_parse_format1_3(ranges, idx, slba, nlb, snsid,
2879                                                apptag, appmask, reftag);
2880         break;
2881 
2882     default:
2883         abort();
2884     }
2885 }
2886 
2887 static inline uint16_t nvme_check_copy_mcl(NvmeNamespace *ns,
2888                                            NvmeCopyAIOCB *iocb, uint16_t nr)
2889 {
2890     uint32_t copy_len = 0;
2891 
2892     for (int idx = 0; idx < nr; idx++) {
2893         uint32_t nlb;
2894         nvme_copy_source_range_parse(iocb->ranges, idx, iocb->format, NULL,
2895                                      &nlb, NULL, NULL, NULL, NULL);
2896         copy_len += nlb;
2897     }
2898     iocb->tcl = copy_len;
2899     if (copy_len > ns->id_ns.mcl) {
2900         return NVME_CMD_SIZE_LIMIT | NVME_DNR;
2901     }
2902 
2903     return NVME_SUCCESS;
2904 }
2905 
2906 static void nvme_copy_out_completed_cb(void *opaque, int ret)
2907 {
2908     NvmeCopyAIOCB *iocb = opaque;
2909     NvmeRequest *req = iocb->req;
2910     NvmeNamespace *dns = req->ns;
2911     uint32_t nlb;
2912 
2913     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2914                                  &nlb, NULL, NULL, NULL, NULL);
2915 
2916     if (ret < 0) {
2917         iocb->ret = ret;
2918         goto out;
2919     } else if (iocb->ret < 0) {
2920         goto out;
2921     }
2922 
2923     if (dns->params.zoned) {
2924         nvme_advance_zone_wp(dns, iocb->zone, nlb);
2925     }
2926 
2927     iocb->idx++;
2928     iocb->slba += nlb;
2929 out:
2930     nvme_do_copy(iocb);
2931 }
2932 
2933 static void nvme_copy_out_cb(void *opaque, int ret)
2934 {
2935     NvmeCopyAIOCB *iocb = opaque;
2936     NvmeRequest *req = iocb->req;
2937     NvmeNamespace *dns = req->ns;
2938     uint32_t nlb;
2939     size_t mlen;
2940     uint8_t *mbounce;
2941 
2942     if (ret < 0 || iocb->ret < 0 || !dns->lbaf.ms) {
2943         goto out;
2944     }
2945 
2946     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2947                                  &nlb, NULL, NULL, NULL, NULL);
2948 
2949     mlen = nvme_m2b(dns, nlb);
2950     mbounce = iocb->bounce + nvme_l2b(dns, nlb);
2951 
2952     qemu_iovec_reset(&iocb->iov);
2953     qemu_iovec_add(&iocb->iov, mbounce, mlen);
2954 
2955     iocb->aiocb = blk_aio_pwritev(dns->blkconf.blk, nvme_moff(dns, iocb->slba),
2956                                   &iocb->iov, 0, nvme_copy_out_completed_cb,
2957                                   iocb);
2958 
2959     return;
2960 
2961 out:
2962     nvme_copy_out_completed_cb(iocb, ret);
2963 }
2964 
2965 static void nvme_copy_in_completed_cb(void *opaque, int ret)
2966 {
2967     NvmeCopyAIOCB *iocb = opaque;
2968     NvmeRequest *req = iocb->req;
2969     NvmeNamespace *sns = iocb->sns;
2970     NvmeNamespace *dns = req->ns;
2971     NvmeCopyCmd *copy = NULL;
2972     uint8_t *mbounce = NULL;
2973     uint32_t nlb;
2974     uint64_t slba;
2975     uint16_t apptag, appmask;
2976     uint64_t reftag;
2977     size_t len, mlen;
2978     uint16_t status;
2979 
2980     if (ret < 0) {
2981         iocb->ret = ret;
2982         goto out;
2983     } else if (iocb->ret < 0) {
2984         goto out;
2985     }
2986 
2987     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
2988                                  &nlb, NULL, &apptag, &appmask, &reftag);
2989 
2990     trace_pci_nvme_copy_out(iocb->slba, nlb);
2991 
2992     len = nvme_l2b(sns, nlb);
2993 
2994     if (NVME_ID_NS_DPS_TYPE(sns->id_ns.dps)) {
2995         copy = (NvmeCopyCmd *)&req->cmd;
2996 
2997         uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
2998 
2999         mlen = nvme_m2b(sns, nlb);
3000         mbounce = iocb->bounce + nvme_l2b(sns, nlb);
3001 
3002         status = nvme_dif_mangle_mdata(sns, mbounce, mlen, slba);
3003         if (status) {
3004             goto invalid;
3005         }
3006         status = nvme_dif_check(sns, iocb->bounce, len, mbounce, mlen, prinfor,
3007                                 slba, apptag, appmask, &reftag);
3008         if (status) {
3009             goto invalid;
3010         }
3011     }
3012 
3013     if (NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3014         copy = (NvmeCopyCmd *)&req->cmd;
3015         uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
3016 
3017         mlen = nvme_m2b(dns, nlb);
3018         mbounce = iocb->bounce + nvme_l2b(dns, nlb);
3019 
3020         apptag = le16_to_cpu(copy->apptag);
3021         appmask = le16_to_cpu(copy->appmask);
3022 
3023         if (prinfow & NVME_PRINFO_PRACT) {
3024             status = nvme_check_prinfo(dns, prinfow, iocb->slba, iocb->reftag);
3025             if (status) {
3026                 goto invalid;
3027             }
3028 
3029             nvme_dif_pract_generate_dif(dns, iocb->bounce, len, mbounce, mlen,
3030                                         apptag, &iocb->reftag);
3031         } else {
3032             status = nvme_dif_check(dns, iocb->bounce, len, mbounce, mlen,
3033                                     prinfow, iocb->slba, apptag, appmask,
3034                                     &iocb->reftag);
3035             if (status) {
3036                 goto invalid;
3037             }
3038         }
3039     }
3040 
3041     status = nvme_check_bounds(dns, iocb->slba, nlb);
3042     if (status) {
3043         goto invalid;
3044     }
3045 
3046     if (dns->params.zoned) {
3047         status = nvme_check_zone_write(dns, iocb->zone, iocb->slba, nlb);
3048         if (status) {
3049             goto invalid;
3050         }
3051 
3052         if (!(iocb->zone->d.za & NVME_ZA_ZRWA_VALID)) {
3053             iocb->zone->w_ptr += nlb;
3054         }
3055     }
3056 
3057     qemu_iovec_reset(&iocb->iov);
3058     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3059 
3060     block_acct_start(blk_get_stats(dns->blkconf.blk), &iocb->acct.write, 0,
3061                      BLOCK_ACCT_WRITE);
3062 
3063     iocb->aiocb = blk_aio_pwritev(dns->blkconf.blk, nvme_l2b(dns, iocb->slba),
3064                                   &iocb->iov, 0, nvme_copy_out_cb, iocb);
3065 
3066     return;
3067 
3068 invalid:
3069     req->status = status;
3070     iocb->ret = -1;
3071 out:
3072     nvme_do_copy(iocb);
3073 }
3074 
3075 static void nvme_copy_in_cb(void *opaque, int ret)
3076 {
3077     NvmeCopyAIOCB *iocb = opaque;
3078     NvmeNamespace *sns = iocb->sns;
3079     uint64_t slba;
3080     uint32_t nlb;
3081 
3082     if (ret < 0 || iocb->ret < 0 || !sns->lbaf.ms) {
3083         goto out;
3084     }
3085 
3086     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
3087                                  &nlb, NULL, NULL, NULL, NULL);
3088 
3089     qemu_iovec_reset(&iocb->iov);
3090     qemu_iovec_add(&iocb->iov, iocb->bounce + nvme_l2b(sns, nlb),
3091                    nvme_m2b(sns, nlb));
3092 
3093     iocb->aiocb = blk_aio_preadv(sns->blkconf.blk, nvme_moff(sns, slba),
3094                                  &iocb->iov, 0, nvme_copy_in_completed_cb,
3095                                  iocb);
3096     return;
3097 
3098 out:
3099     nvme_copy_in_completed_cb(iocb, ret);
3100 }
3101 
3102 static inline bool nvme_csi_supports_copy(uint8_t csi)
3103 {
3104     return csi == NVME_CSI_NVM || csi == NVME_CSI_ZONED;
3105 }
3106 
3107 static inline bool nvme_copy_ns_format_match(NvmeNamespace *sns,
3108                                              NvmeNamespace *dns)
3109 {
3110     return sns->lbaf.ds == dns->lbaf.ds && sns->lbaf.ms == dns->lbaf.ms;
3111 }
3112 
3113 static bool nvme_copy_matching_ns_format(NvmeNamespace *sns, NvmeNamespace *dns,
3114                                          bool pi_enable)
3115 {
3116     if (!nvme_csi_supports_copy(sns->csi) ||
3117         !nvme_csi_supports_copy(dns->csi)) {
3118         return false;
3119     }
3120 
3121     if (!pi_enable && !nvme_copy_ns_format_match(sns, dns)) {
3122             return false;
3123     }
3124 
3125     if (pi_enable && (!nvme_copy_ns_format_match(sns, dns) ||
3126         sns->id_ns.dps != dns->id_ns.dps)) {
3127             return false;
3128     }
3129 
3130     return true;
3131 }
3132 
3133 static inline bool nvme_copy_corresp_pi_match(NvmeNamespace *sns,
3134                                               NvmeNamespace *dns)
3135 {
3136     return sns->lbaf.ms == 0 &&
3137            ((dns->lbaf.ms == 8 && dns->pif == 0) ||
3138            (dns->lbaf.ms == 16 && dns->pif == 1));
3139 }
3140 
3141 static bool nvme_copy_corresp_pi_format(NvmeNamespace *sns, NvmeNamespace *dns,
3142                                         bool sns_pi_en)
3143 {
3144     if (!nvme_csi_supports_copy(sns->csi) ||
3145         !nvme_csi_supports_copy(dns->csi)) {
3146         return false;
3147     }
3148 
3149     if (!sns_pi_en && !nvme_copy_corresp_pi_match(sns, dns)) {
3150         return false;
3151     }
3152 
3153     if (sns_pi_en && !nvme_copy_corresp_pi_match(dns, sns)) {
3154         return false;
3155     }
3156 
3157     return true;
3158 }
3159 
3160 static void nvme_do_copy(NvmeCopyAIOCB *iocb)
3161 {
3162     NvmeRequest *req = iocb->req;
3163     NvmeNamespace *sns;
3164     NvmeNamespace *dns = req->ns;
3165     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
3166     uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
3167     uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
3168     uint64_t slba;
3169     uint32_t nlb;
3170     size_t len;
3171     uint16_t status;
3172     uint32_t dnsid = le32_to_cpu(req->cmd.nsid);
3173     uint32_t snsid = dnsid;
3174 
3175     if (iocb->ret < 0) {
3176         goto done;
3177     }
3178 
3179     if (iocb->idx == iocb->nr) {
3180         goto done;
3181     }
3182 
3183     if (iocb->format == 2 || iocb->format == 3) {
3184         nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format,
3185                                      &slba, &nlb, &snsid, NULL, NULL, NULL);
3186         if (snsid != dnsid) {
3187             if (snsid == NVME_NSID_BROADCAST ||
3188                 !nvme_nsid_valid(iocb->n, snsid)) {
3189                 status = NVME_INVALID_NSID | NVME_DNR;
3190                 goto invalid;
3191             }
3192             iocb->sns = nvme_ns(iocb->n, snsid);
3193             if (unlikely(!iocb->sns)) {
3194                 status = NVME_INVALID_FIELD | NVME_DNR;
3195                 goto invalid;
3196             }
3197         } else {
3198             if (((slba + nlb) > iocb->slba) &&
3199                 ((slba + nlb) < (iocb->slba + iocb->tcl))) {
3200                 status = NVME_CMD_OVERLAP_IO_RANGE | NVME_DNR;
3201                 goto invalid;
3202             }
3203         }
3204     } else {
3205         nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format,
3206                                      &slba, &nlb, NULL, NULL, NULL, NULL);
3207     }
3208 
3209     sns = iocb->sns;
3210     if ((snsid == dnsid) && NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3211         ((prinfor & NVME_PRINFO_PRACT) != (prinfow & NVME_PRINFO_PRACT))) {
3212         status = NVME_INVALID_FIELD | NVME_DNR;
3213         goto invalid;
3214     } else if (snsid != dnsid) {
3215         if (!NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3216             !NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3217             if (!nvme_copy_matching_ns_format(sns, dns, false)) {
3218                 status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3219                 goto invalid;
3220             }
3221         }
3222         if (NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3223             NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3224             if ((prinfor & NVME_PRINFO_PRACT) !=
3225                 (prinfow & NVME_PRINFO_PRACT)) {
3226                 status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3227                 goto invalid;
3228             } else {
3229                 if (!nvme_copy_matching_ns_format(sns, dns, true)) {
3230                     status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3231                     goto invalid;
3232                 }
3233             }
3234         }
3235 
3236         if (!NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3237             NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3238             if (!(prinfow & NVME_PRINFO_PRACT)) {
3239                 status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3240                 goto invalid;
3241             } else {
3242                 if (!nvme_copy_corresp_pi_format(sns, dns, false)) {
3243                     status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3244                     goto invalid;
3245                 }
3246             }
3247         }
3248 
3249         if (NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3250             !NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3251             if (!(prinfor & NVME_PRINFO_PRACT)) {
3252                 status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3253                 goto invalid;
3254             } else {
3255                 if (!nvme_copy_corresp_pi_format(sns, dns, true)) {
3256                     status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3257                     goto invalid;
3258                 }
3259             }
3260         }
3261     }
3262     len = nvme_l2b(sns, nlb);
3263 
3264     trace_pci_nvme_copy_source_range(slba, nlb);
3265 
3266     if (nlb > le16_to_cpu(sns->id_ns.mssrl)) {
3267         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3268         goto invalid;
3269     }
3270 
3271     status = nvme_check_bounds(sns, slba, nlb);
3272     if (status) {
3273         goto invalid;
3274     }
3275 
3276     if (NVME_ERR_REC_DULBE(sns->features.err_rec)) {
3277         status = nvme_check_dulbe(sns, slba, nlb);
3278         if (status) {
3279             goto invalid;
3280         }
3281     }
3282 
3283     if (sns->params.zoned) {
3284         status = nvme_check_zone_read(sns, slba, nlb);
3285         if (status) {
3286             goto invalid;
3287         }
3288     }
3289 
3290     g_free(iocb->bounce);
3291     iocb->bounce = g_malloc_n(le16_to_cpu(sns->id_ns.mssrl),
3292                               sns->lbasz + sns->lbaf.ms);
3293 
3294     qemu_iovec_reset(&iocb->iov);
3295     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3296 
3297     block_acct_start(blk_get_stats(sns->blkconf.blk), &iocb->acct.read, 0,
3298                      BLOCK_ACCT_READ);
3299 
3300     iocb->aiocb = blk_aio_preadv(sns->blkconf.blk, nvme_l2b(sns, slba),
3301                                  &iocb->iov, 0, nvme_copy_in_cb, iocb);
3302     return;
3303 
3304 invalid:
3305     req->status = status;
3306     iocb->ret = -1;
3307 done:
3308     nvme_copy_done(iocb);
3309 }
3310 
3311 static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
3312 {
3313     NvmeNamespace *ns = req->ns;
3314     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
3315     NvmeCopyAIOCB *iocb = blk_aio_get(&nvme_copy_aiocb_info, ns->blkconf.blk,
3316                                       nvme_misc_cb, req);
3317     uint16_t nr = copy->nr + 1;
3318     uint8_t format = copy->control[0] & 0xf;
3319     size_t len = sizeof(NvmeCopySourceRangeFormat0_2);
3320 
3321     uint16_t status;
3322 
3323     trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);
3324 
3325     iocb->ranges = NULL;
3326     iocb->zone = NULL;
3327 
3328     if (!(n->id_ctrl.ocfs & (1 << format)) ||
3329         ((format == 2 || format == 3) &&
3330          !(n->features.hbs.cdfe & (1 << format)))) {
3331         trace_pci_nvme_err_copy_invalid_format(format);
3332         status = NVME_INVALID_FIELD | NVME_DNR;
3333         goto invalid;
3334     }
3335 
3336     if (nr > ns->id_ns.msrc + 1) {
3337         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3338         goto invalid;
3339     }
3340 
3341     if ((ns->pif == 0x0 && (format != 0x0 && format != 0x2)) ||
3342         (ns->pif != 0x0 && (format != 0x1 && format != 0x3))) {
3343         status = NVME_INVALID_FORMAT | NVME_DNR;
3344         goto invalid;
3345     }
3346 
3347     if (ns->pif) {
3348         len = sizeof(NvmeCopySourceRangeFormat1_3);
3349     }
3350 
3351     iocb->format = format;
3352     iocb->ranges = g_malloc_n(nr, len);
3353     status = nvme_h2c(n, (uint8_t *)iocb->ranges, len * nr, req);
3354     if (status) {
3355         goto invalid;
3356     }
3357 
3358     iocb->slba = le64_to_cpu(copy->sdlba);
3359 
3360     if (ns->params.zoned) {
3361         iocb->zone = nvme_get_zone_by_slba(ns, iocb->slba);
3362         if (!iocb->zone) {
3363             status = NVME_LBA_RANGE | NVME_DNR;
3364             goto invalid;
3365         }
3366 
3367         status = nvme_zrm_auto(n, ns, iocb->zone);
3368         if (status) {
3369             goto invalid;
3370         }
3371     }
3372 
3373     status = nvme_check_copy_mcl(ns, iocb, nr);
3374     if (status) {
3375         goto invalid;
3376     }
3377 
3378     iocb->req = req;
3379     iocb->ret = 0;
3380     iocb->nr = nr;
3381     iocb->idx = 0;
3382     iocb->reftag = le32_to_cpu(copy->reftag);
3383     iocb->reftag |= (uint64_t)le32_to_cpu(copy->cdw3) << 32;
3384 
3385     qemu_iovec_init(&iocb->iov, 1);
3386 
3387     req->aiocb = &iocb->common;
3388     iocb->sns = req->ns;
3389     iocb->n = n;
3390     iocb->bounce = NULL;
3391     nvme_do_copy(iocb);
3392 
3393     return NVME_NO_COMPLETE;
3394 
3395 invalid:
3396     g_free(iocb->ranges);
3397     qemu_aio_unref(iocb);
3398     return status;
3399 }
3400 
3401 static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req)
3402 {
3403     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3404     NvmeNamespace *ns = req->ns;
3405     BlockBackend *blk = ns->blkconf.blk;
3406     uint64_t slba = le64_to_cpu(rw->slba);
3407     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
3408     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3409     size_t data_len = nvme_l2b(ns, nlb);
3410     size_t len = data_len;
3411     int64_t offset = nvme_l2b(ns, slba);
3412     struct nvme_compare_ctx *ctx = NULL;
3413     uint16_t status;
3414 
3415     trace_pci_nvme_compare(nvme_cid(req), nvme_nsid(ns), slba, nlb);
3416 
3417     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) && (prinfo & NVME_PRINFO_PRACT)) {
3418         return NVME_INVALID_PROT_INFO | NVME_DNR;
3419     }
3420 
3421     if (nvme_ns_ext(ns)) {
3422         len += nvme_m2b(ns, nlb);
3423     }
3424 
3425     if (NVME_ID_CTRL_CTRATT_MEM(n->id_ctrl.ctratt)) {
3426         status = nvme_check_mdts(n, data_len);
3427     } else {
3428         status = nvme_check_mdts(n, len);
3429     }
3430     if (status) {
3431         return status;
3432     }
3433 
3434     status = nvme_check_bounds(ns, slba, nlb);
3435     if (status) {
3436         return status;
3437     }
3438 
3439     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3440         status = nvme_check_dulbe(ns, slba, nlb);
3441         if (status) {
3442             return status;
3443         }
3444     }
3445 
3446     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
3447     if (status) {
3448         return status;
3449     }
3450 
3451     ctx = g_new(struct nvme_compare_ctx, 1);
3452     ctx->data.bounce = g_malloc(data_len);
3453 
3454     req->opaque = ctx;
3455 
3456     qemu_iovec_init(&ctx->data.iov, 1);
3457     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, data_len);
3458 
3459     block_acct_start(blk_get_stats(blk), &req->acct, data_len,
3460                      BLOCK_ACCT_READ);
3461     req->aiocb = blk_aio_preadv(blk, offset, &ctx->data.iov, 0,
3462                                 nvme_compare_data_cb, req);
3463 
3464     return NVME_NO_COMPLETE;
3465 }
3466 
3467 typedef struct NvmeFlushAIOCB {
3468     BlockAIOCB common;
3469     BlockAIOCB *aiocb;
3470     NvmeRequest *req;
3471     int ret;
3472 
3473     NvmeNamespace *ns;
3474     uint32_t nsid;
3475     bool broadcast;
3476 } NvmeFlushAIOCB;
3477 
3478 static void nvme_flush_cancel(BlockAIOCB *acb)
3479 {
3480     NvmeFlushAIOCB *iocb = container_of(acb, NvmeFlushAIOCB, common);
3481 
3482     iocb->ret = -ECANCELED;
3483 
3484     if (iocb->aiocb) {
3485         blk_aio_cancel_async(iocb->aiocb);
3486         iocb->aiocb = NULL;
3487     }
3488 }
3489 
3490 static const AIOCBInfo nvme_flush_aiocb_info = {
3491     .aiocb_size = sizeof(NvmeFlushAIOCB),
3492     .cancel_async = nvme_flush_cancel,
3493 };
3494 
3495 static void nvme_do_flush(NvmeFlushAIOCB *iocb);
3496 
3497 static void nvme_flush_ns_cb(void *opaque, int ret)
3498 {
3499     NvmeFlushAIOCB *iocb = opaque;
3500     NvmeNamespace *ns = iocb->ns;
3501 
3502     if (ret < 0) {
3503         iocb->ret = ret;
3504         goto out;
3505     } else if (iocb->ret < 0) {
3506         goto out;
3507     }
3508 
3509     if (ns) {
3510         trace_pci_nvme_flush_ns(iocb->nsid);
3511 
3512         iocb->ns = NULL;
3513         iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);
3514         return;
3515     }
3516 
3517 out:
3518     nvme_do_flush(iocb);
3519 }
3520 
3521 static void nvme_do_flush(NvmeFlushAIOCB *iocb)
3522 {
3523     NvmeRequest *req = iocb->req;
3524     NvmeCtrl *n = nvme_ctrl(req);
3525     int i;
3526 
3527     if (iocb->ret < 0) {
3528         goto done;
3529     }
3530 
3531     if (iocb->broadcast) {
3532         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
3533             iocb->ns = nvme_ns(n, i);
3534             if (iocb->ns) {
3535                 iocb->nsid = i;
3536                 break;
3537             }
3538         }
3539     }
3540 
3541     if (!iocb->ns) {
3542         goto done;
3543     }
3544 
3545     nvme_flush_ns_cb(iocb, 0);
3546     return;
3547 
3548 done:
3549     iocb->common.cb(iocb->common.opaque, iocb->ret);
3550     qemu_aio_unref(iocb);
3551 }
3552 
3553 static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req)
3554 {
3555     NvmeFlushAIOCB *iocb;
3556     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
3557     uint16_t status;
3558 
3559     iocb = qemu_aio_get(&nvme_flush_aiocb_info, NULL, nvme_misc_cb, req);
3560 
3561     iocb->req = req;
3562     iocb->ret = 0;
3563     iocb->ns = NULL;
3564     iocb->nsid = 0;
3565     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
3566 
3567     if (!iocb->broadcast) {
3568         if (!nvme_nsid_valid(n, nsid)) {
3569             status = NVME_INVALID_NSID | NVME_DNR;
3570             goto out;
3571         }
3572 
3573         iocb->ns = nvme_ns(n, nsid);
3574         if (!iocb->ns) {
3575             status = NVME_INVALID_FIELD | NVME_DNR;
3576             goto out;
3577         }
3578 
3579         iocb->nsid = nsid;
3580     }
3581 
3582     req->aiocb = &iocb->common;
3583     nvme_do_flush(iocb);
3584 
3585     return NVME_NO_COMPLETE;
3586 
3587 out:
3588     qemu_aio_unref(iocb);
3589 
3590     return status;
3591 }
3592 
3593 static uint16_t nvme_read(NvmeCtrl *n, NvmeRequest *req)
3594 {
3595     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3596     NvmeNamespace *ns = req->ns;
3597     uint64_t slba = le64_to_cpu(rw->slba);
3598     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3599     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3600     uint64_t data_size = nvme_l2b(ns, nlb);
3601     uint64_t mapped_size = data_size;
3602     uint64_t data_offset;
3603     BlockBackend *blk = ns->blkconf.blk;
3604     uint16_t status;
3605 
3606     if (nvme_ns_ext(ns) && !(NVME_ID_CTRL_CTRATT_MEM(n->id_ctrl.ctratt))) {
3607         mapped_size += nvme_m2b(ns, nlb);
3608 
3609         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3610             bool pract = prinfo & NVME_PRINFO_PRACT;
3611 
3612             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3613                 mapped_size = data_size;
3614             }
3615         }
3616     }
3617 
3618     trace_pci_nvme_read(nvme_cid(req), nvme_nsid(ns), nlb, mapped_size, slba);
3619 
3620     status = nvme_check_mdts(n, mapped_size);
3621     if (status) {
3622         goto invalid;
3623     }
3624 
3625     status = nvme_check_bounds(ns, slba, nlb);
3626     if (status) {
3627         goto invalid;
3628     }
3629 
3630     if (ns->params.zoned) {
3631         status = nvme_check_zone_read(ns, slba, nlb);
3632         if (status) {
3633             trace_pci_nvme_err_zone_read_not_ok(slba, nlb, status);
3634             goto invalid;
3635         }
3636     }
3637 
3638     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3639         status = nvme_check_dulbe(ns, slba, nlb);
3640         if (status) {
3641             goto invalid;
3642         }
3643     }
3644 
3645     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3646         return nvme_dif_rw(n, req);
3647     }
3648 
3649     status = nvme_map_data(n, nlb, req);
3650     if (status) {
3651         goto invalid;
3652     }
3653 
3654     data_offset = nvme_l2b(ns, slba);
3655 
3656     block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3657                      BLOCK_ACCT_READ);
3658     nvme_blk_read(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3659     return NVME_NO_COMPLETE;
3660 
3661 invalid:
3662     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_READ);
3663     return status | NVME_DNR;
3664 }
3665 
3666 static void nvme_do_write_fdp(NvmeCtrl *n, NvmeRequest *req, uint64_t slba,
3667                               uint32_t nlb)
3668 {
3669     NvmeNamespace *ns = req->ns;
3670     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3671     uint64_t data_size = nvme_l2b(ns, nlb);
3672     uint32_t dw12 = le32_to_cpu(req->cmd.cdw12);
3673     uint8_t dtype = (dw12 >> 20) & 0xf;
3674     uint16_t pid = le16_to_cpu(rw->dspec);
3675     uint16_t ph, rg, ruhid;
3676     NvmeReclaimUnit *ru;
3677 
3678     if (dtype != NVME_DIRECTIVE_DATA_PLACEMENT ||
3679         !nvme_parse_pid(ns, pid, &ph, &rg)) {
3680         ph = 0;
3681         rg = 0;
3682     }
3683 
3684     ruhid = ns->fdp.phs[ph];
3685     ru = &ns->endgrp->fdp.ruhs[ruhid].rus[rg];
3686 
3687     nvme_fdp_stat_inc(&ns->endgrp->fdp.hbmw, data_size);
3688     nvme_fdp_stat_inc(&ns->endgrp->fdp.mbmw, data_size);
3689 
3690     while (nlb) {
3691         if (nlb < ru->ruamw) {
3692             ru->ruamw -= nlb;
3693             break;
3694         }
3695 
3696         nlb -= ru->ruamw;
3697         nvme_update_ruh(n, ns, pid);
3698     }
3699 }
3700 
3701 static uint16_t nvme_do_write(NvmeCtrl *n, NvmeRequest *req, bool append,
3702                               bool wrz)
3703 {
3704     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3705     NvmeNamespace *ns = req->ns;
3706     uint64_t slba = le64_to_cpu(rw->slba);
3707     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3708     uint16_t ctrl = le16_to_cpu(rw->control);
3709     uint8_t prinfo = NVME_RW_PRINFO(ctrl);
3710     uint64_t data_size = nvme_l2b(ns, nlb);
3711     uint64_t mapped_size = data_size;
3712     uint64_t data_offset;
3713     NvmeZone *zone;
3714     NvmeZonedResult *res = (NvmeZonedResult *)&req->cqe;
3715     BlockBackend *blk = ns->blkconf.blk;
3716     uint16_t status;
3717 
3718     if (nvme_ns_ext(ns) && !(NVME_ID_CTRL_CTRATT_MEM(n->id_ctrl.ctratt))) {
3719         mapped_size += nvme_m2b(ns, nlb);
3720 
3721         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3722             bool pract = prinfo & NVME_PRINFO_PRACT;
3723 
3724             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3725                 mapped_size -= nvme_m2b(ns, nlb);
3726             }
3727         }
3728     }
3729 
3730     trace_pci_nvme_write(nvme_cid(req), nvme_io_opc_str(rw->opcode),
3731                          nvme_nsid(ns), nlb, mapped_size, slba);
3732 
3733     if (!wrz) {
3734         status = nvme_check_mdts(n, mapped_size);
3735         if (status) {
3736             goto invalid;
3737         }
3738     }
3739 
3740     status = nvme_check_bounds(ns, slba, nlb);
3741     if (status) {
3742         goto invalid;
3743     }
3744 
3745     if (ns->params.zoned) {
3746         zone = nvme_get_zone_by_slba(ns, slba);
3747         assert(zone);
3748 
3749         if (append) {
3750             bool piremap = !!(ctrl & NVME_RW_PIREMAP);
3751 
3752             if (unlikely(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3753                 return NVME_INVALID_ZONE_OP | NVME_DNR;
3754             }
3755 
3756             if (unlikely(slba != zone->d.zslba)) {
3757                 trace_pci_nvme_err_append_not_at_start(slba, zone->d.zslba);
3758                 status = NVME_INVALID_FIELD;
3759                 goto invalid;
3760             }
3761 
3762             if (n->params.zasl &&
3763                 data_size > (uint64_t)n->page_size << n->params.zasl) {
3764                 trace_pci_nvme_err_zasl(data_size);
3765                 return NVME_INVALID_FIELD | NVME_DNR;
3766             }
3767 
3768             slba = zone->w_ptr;
3769             rw->slba = cpu_to_le64(slba);
3770             res->slba = cpu_to_le64(slba);
3771 
3772             switch (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3773             case NVME_ID_NS_DPS_TYPE_1:
3774                 if (!piremap) {
3775                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3776                 }
3777 
3778                 /* fallthrough */
3779 
3780             case NVME_ID_NS_DPS_TYPE_2:
3781                 if (piremap) {
3782                     uint32_t reftag = le32_to_cpu(rw->reftag);
3783                     rw->reftag = cpu_to_le32(reftag + (slba - zone->d.zslba));
3784                 }
3785 
3786                 break;
3787 
3788             case NVME_ID_NS_DPS_TYPE_3:
3789                 if (piremap) {
3790                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3791                 }
3792 
3793                 break;
3794             }
3795         }
3796 
3797         status = nvme_check_zone_write(ns, zone, slba, nlb);
3798         if (status) {
3799             goto invalid;
3800         }
3801 
3802         status = nvme_zrm_auto(n, ns, zone);
3803         if (status) {
3804             goto invalid;
3805         }
3806 
3807         if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3808             zone->w_ptr += nlb;
3809         }
3810     } else if (ns->endgrp && ns->endgrp->fdp.enabled) {
3811         nvme_do_write_fdp(n, req, slba, nlb);
3812     }
3813 
3814     data_offset = nvme_l2b(ns, slba);
3815 
3816     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3817         return nvme_dif_rw(n, req);
3818     }
3819 
3820     if (!wrz) {
3821         status = nvme_map_data(n, nlb, req);
3822         if (status) {
3823             goto invalid;
3824         }
3825 
3826         block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3827                          BLOCK_ACCT_WRITE);
3828         nvme_blk_write(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3829     } else {
3830         req->aiocb = blk_aio_pwrite_zeroes(blk, data_offset, data_size,
3831                                            BDRV_REQ_MAY_UNMAP, nvme_rw_cb,
3832                                            req);
3833     }
3834 
3835     return NVME_NO_COMPLETE;
3836 
3837 invalid:
3838     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_WRITE);
3839     return status | NVME_DNR;
3840 }
3841 
3842 static inline uint16_t nvme_write(NvmeCtrl *n, NvmeRequest *req)
3843 {
3844     return nvme_do_write(n, req, false, false);
3845 }
3846 
3847 static inline uint16_t nvme_write_zeroes(NvmeCtrl *n, NvmeRequest *req)
3848 {
3849     return nvme_do_write(n, req, false, true);
3850 }
3851 
3852 static inline uint16_t nvme_zone_append(NvmeCtrl *n, NvmeRequest *req)
3853 {
3854     return nvme_do_write(n, req, true, false);
3855 }
3856 
3857 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace *ns, NvmeCmd *c,
3858                                             uint64_t *slba, uint32_t *zone_idx)
3859 {
3860     uint32_t dw10 = le32_to_cpu(c->cdw10);
3861     uint32_t dw11 = le32_to_cpu(c->cdw11);
3862 
3863     if (!ns->params.zoned) {
3864         trace_pci_nvme_err_invalid_opc(c->opcode);
3865         return NVME_INVALID_OPCODE | NVME_DNR;
3866     }
3867 
3868     *slba = ((uint64_t)dw11) << 32 | dw10;
3869     if (unlikely(*slba >= ns->id_ns.nsze)) {
3870         trace_pci_nvme_err_invalid_lba_range(*slba, 0, ns->id_ns.nsze);
3871         *slba = 0;
3872         return NVME_LBA_RANGE | NVME_DNR;
3873     }
3874 
3875     *zone_idx = nvme_zone_idx(ns, *slba);
3876     assert(*zone_idx < ns->num_zones);
3877 
3878     return NVME_SUCCESS;
3879 }
3880 
3881 typedef uint16_t (*op_handler_t)(NvmeNamespace *, NvmeZone *, NvmeZoneState,
3882                                  NvmeRequest *);
3883 
3884 enum NvmeZoneProcessingMask {
3885     NVME_PROC_CURRENT_ZONE    = 0,
3886     NVME_PROC_OPENED_ZONES    = 1 << 0,
3887     NVME_PROC_CLOSED_ZONES    = 1 << 1,
3888     NVME_PROC_READ_ONLY_ZONES = 1 << 2,
3889     NVME_PROC_FULL_ZONES      = 1 << 3,
3890 };
3891 
3892 static uint16_t nvme_open_zone(NvmeNamespace *ns, NvmeZone *zone,
3893                                NvmeZoneState state, NvmeRequest *req)
3894 {
3895     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
3896     int flags = 0;
3897 
3898     if (cmd->zsflags & NVME_ZSFLAG_ZRWA_ALLOC) {
3899         uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3900 
3901         if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3902             return NVME_INVALID_ZONE_OP | NVME_DNR;
3903         }
3904 
3905         if (zone->w_ptr % ns->zns.zrwafg) {
3906             return NVME_NOZRWA | NVME_DNR;
3907         }
3908 
3909         flags = NVME_ZRM_ZRWA;
3910     }
3911 
3912     return nvme_zrm_open_flags(nvme_ctrl(req), ns, zone, flags);
3913 }
3914 
3915 static uint16_t nvme_close_zone(NvmeNamespace *ns, NvmeZone *zone,
3916                                 NvmeZoneState state, NvmeRequest *req)
3917 {
3918     return nvme_zrm_close(ns, zone);
3919 }
3920 
3921 static uint16_t nvme_finish_zone(NvmeNamespace *ns, NvmeZone *zone,
3922                                  NvmeZoneState state, NvmeRequest *req)
3923 {
3924     return nvme_zrm_finish(ns, zone);
3925 }
3926 
3927 static uint16_t nvme_offline_zone(NvmeNamespace *ns, NvmeZone *zone,
3928                                   NvmeZoneState state, NvmeRequest *req)
3929 {
3930     switch (state) {
3931     case NVME_ZONE_STATE_READ_ONLY:
3932         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_OFFLINE);
3933         /* fall through */
3934     case NVME_ZONE_STATE_OFFLINE:
3935         return NVME_SUCCESS;
3936     default:
3937         return NVME_ZONE_INVAL_TRANSITION;
3938     }
3939 }
3940 
3941 static uint16_t nvme_set_zd_ext(NvmeNamespace *ns, NvmeZone *zone)
3942 {
3943     uint16_t status;
3944     uint8_t state = nvme_get_zone_state(zone);
3945 
3946     if (state == NVME_ZONE_STATE_EMPTY) {
3947         status = nvme_aor_check(ns, 1, 0);
3948         if (status) {
3949             return status;
3950         }
3951         nvme_aor_inc_active(ns);
3952         zone->d.za |= NVME_ZA_ZD_EXT_VALID;
3953         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
3954         return NVME_SUCCESS;
3955     }
3956 
3957     return NVME_ZONE_INVAL_TRANSITION;
3958 }
3959 
3960 static uint16_t nvme_bulk_proc_zone(NvmeNamespace *ns, NvmeZone *zone,
3961                                     enum NvmeZoneProcessingMask proc_mask,
3962                                     op_handler_t op_hndlr, NvmeRequest *req)
3963 {
3964     uint16_t status = NVME_SUCCESS;
3965     NvmeZoneState zs = nvme_get_zone_state(zone);
3966     bool proc_zone;
3967 
3968     switch (zs) {
3969     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3970     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3971         proc_zone = proc_mask & NVME_PROC_OPENED_ZONES;
3972         break;
3973     case NVME_ZONE_STATE_CLOSED:
3974         proc_zone = proc_mask & NVME_PROC_CLOSED_ZONES;
3975         break;
3976     case NVME_ZONE_STATE_READ_ONLY:
3977         proc_zone = proc_mask & NVME_PROC_READ_ONLY_ZONES;
3978         break;
3979     case NVME_ZONE_STATE_FULL:
3980         proc_zone = proc_mask & NVME_PROC_FULL_ZONES;
3981         break;
3982     default:
3983         proc_zone = false;
3984     }
3985 
3986     if (proc_zone) {
3987         status = op_hndlr(ns, zone, zs, req);
3988     }
3989 
3990     return status;
3991 }
3992 
3993 static uint16_t nvme_do_zone_op(NvmeNamespace *ns, NvmeZone *zone,
3994                                 enum NvmeZoneProcessingMask proc_mask,
3995                                 op_handler_t op_hndlr, NvmeRequest *req)
3996 {
3997     NvmeZone *next;
3998     uint16_t status = NVME_SUCCESS;
3999     int i;
4000 
4001     if (!proc_mask) {
4002         status = op_hndlr(ns, zone, nvme_get_zone_state(zone), req);
4003     } else {
4004         if (proc_mask & NVME_PROC_CLOSED_ZONES) {
4005             QTAILQ_FOREACH_SAFE(zone, &ns->closed_zones, entry, next) {
4006                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
4007                                              req);
4008                 if (status && status != NVME_NO_COMPLETE) {
4009                     goto out;
4010                 }
4011             }
4012         }
4013         if (proc_mask & NVME_PROC_OPENED_ZONES) {
4014             QTAILQ_FOREACH_SAFE(zone, &ns->imp_open_zones, entry, next) {
4015                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
4016                                              req);
4017                 if (status && status != NVME_NO_COMPLETE) {
4018                     goto out;
4019                 }
4020             }
4021 
4022             QTAILQ_FOREACH_SAFE(zone, &ns->exp_open_zones, entry, next) {
4023                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
4024                                              req);
4025                 if (status && status != NVME_NO_COMPLETE) {
4026                     goto out;
4027                 }
4028             }
4029         }
4030         if (proc_mask & NVME_PROC_FULL_ZONES) {
4031             QTAILQ_FOREACH_SAFE(zone, &ns->full_zones, entry, next) {
4032                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
4033                                              req);
4034                 if (status && status != NVME_NO_COMPLETE) {
4035                     goto out;
4036                 }
4037             }
4038         }
4039 
4040         if (proc_mask & NVME_PROC_READ_ONLY_ZONES) {
4041             for (i = 0; i < ns->num_zones; i++, zone++) {
4042                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
4043                                              req);
4044                 if (status && status != NVME_NO_COMPLETE) {
4045                     goto out;
4046                 }
4047             }
4048         }
4049     }
4050 
4051 out:
4052     return status;
4053 }
4054 
4055 typedef struct NvmeZoneResetAIOCB {
4056     BlockAIOCB common;
4057     BlockAIOCB *aiocb;
4058     NvmeRequest *req;
4059     int ret;
4060 
4061     bool all;
4062     int idx;
4063     NvmeZone *zone;
4064 } NvmeZoneResetAIOCB;
4065 
4066 static void nvme_zone_reset_cancel(BlockAIOCB *aiocb)
4067 {
4068     NvmeZoneResetAIOCB *iocb = container_of(aiocb, NvmeZoneResetAIOCB, common);
4069     NvmeRequest *req = iocb->req;
4070     NvmeNamespace *ns = req->ns;
4071 
4072     iocb->idx = ns->num_zones;
4073 
4074     iocb->ret = -ECANCELED;
4075 
4076     if (iocb->aiocb) {
4077         blk_aio_cancel_async(iocb->aiocb);
4078         iocb->aiocb = NULL;
4079     }
4080 }
4081 
4082 static const AIOCBInfo nvme_zone_reset_aiocb_info = {
4083     .aiocb_size = sizeof(NvmeZoneResetAIOCB),
4084     .cancel_async = nvme_zone_reset_cancel,
4085 };
4086 
4087 static void nvme_zone_reset_cb(void *opaque, int ret);
4088 
4089 static void nvme_zone_reset_epilogue_cb(void *opaque, int ret)
4090 {
4091     NvmeZoneResetAIOCB *iocb = opaque;
4092     NvmeRequest *req = iocb->req;
4093     NvmeNamespace *ns = req->ns;
4094     int64_t moff;
4095     int count;
4096 
4097     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
4098         goto out;
4099     }
4100 
4101     moff = nvme_moff(ns, iocb->zone->d.zslba);
4102     count = nvme_m2b(ns, ns->zone_size);
4103 
4104     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, moff, count,
4105                                         BDRV_REQ_MAY_UNMAP,
4106                                         nvme_zone_reset_cb, iocb);
4107     return;
4108 
4109 out:
4110     nvme_zone_reset_cb(iocb, ret);
4111 }
4112 
4113 static void nvme_zone_reset_cb(void *opaque, int ret)
4114 {
4115     NvmeZoneResetAIOCB *iocb = opaque;
4116     NvmeRequest *req = iocb->req;
4117     NvmeNamespace *ns = req->ns;
4118 
4119     if (iocb->ret < 0) {
4120         goto done;
4121     } else if (ret < 0) {
4122         iocb->ret = ret;
4123         goto done;
4124     }
4125 
4126     if (iocb->zone) {
4127         nvme_zrm_reset(ns, iocb->zone);
4128 
4129         if (!iocb->all) {
4130             goto done;
4131         }
4132     }
4133 
4134     while (iocb->idx < ns->num_zones) {
4135         NvmeZone *zone = &ns->zone_array[iocb->idx++];
4136 
4137         switch (nvme_get_zone_state(zone)) {
4138         case NVME_ZONE_STATE_EMPTY:
4139             if (!iocb->all) {
4140                 goto done;
4141             }
4142 
4143             continue;
4144 
4145         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
4146         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
4147         case NVME_ZONE_STATE_CLOSED:
4148         case NVME_ZONE_STATE_FULL:
4149             iocb->zone = zone;
4150             break;
4151 
4152         default:
4153             continue;
4154         }
4155 
4156         trace_pci_nvme_zns_zone_reset(zone->d.zslba);
4157 
4158         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk,
4159                                             nvme_l2b(ns, zone->d.zslba),
4160                                             nvme_l2b(ns, ns->zone_size),
4161                                             BDRV_REQ_MAY_UNMAP,
4162                                             nvme_zone_reset_epilogue_cb,
4163                                             iocb);
4164         return;
4165     }
4166 
4167 done:
4168     iocb->aiocb = NULL;
4169 
4170     iocb->common.cb(iocb->common.opaque, iocb->ret);
4171     qemu_aio_unref(iocb);
4172 }
4173 
4174 static uint16_t nvme_zone_mgmt_send_zrwa_flush(NvmeCtrl *n, NvmeZone *zone,
4175                                                uint64_t elba, NvmeRequest *req)
4176 {
4177     NvmeNamespace *ns = req->ns;
4178     uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
4179     uint64_t wp = zone->d.wp;
4180     uint32_t nlb = elba - wp + 1;
4181     uint16_t status;
4182 
4183 
4184     if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
4185         return NVME_INVALID_ZONE_OP | NVME_DNR;
4186     }
4187 
4188     if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
4189         return NVME_INVALID_FIELD | NVME_DNR;
4190     }
4191 
4192     if (elba < wp || elba > wp + ns->zns.zrwas) {
4193         return NVME_ZONE_BOUNDARY_ERROR | NVME_DNR;
4194     }
4195 
4196     if (nlb % ns->zns.zrwafg) {
4197         return NVME_INVALID_FIELD | NVME_DNR;
4198     }
4199 
4200     status = nvme_zrm_auto(n, ns, zone);
4201     if (status) {
4202         return status;
4203     }
4204 
4205     zone->w_ptr += nlb;
4206 
4207     nvme_advance_zone_wp(ns, zone, nlb);
4208 
4209     return NVME_SUCCESS;
4210 }
4211 
4212 static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4213 {
4214     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
4215     NvmeNamespace *ns = req->ns;
4216     NvmeZone *zone;
4217     NvmeZoneResetAIOCB *iocb;
4218     uint8_t *zd_ext;
4219     uint64_t slba = 0;
4220     uint32_t zone_idx = 0;
4221     uint16_t status;
4222     uint8_t action = cmd->zsa;
4223     bool all;
4224     enum NvmeZoneProcessingMask proc_mask = NVME_PROC_CURRENT_ZONE;
4225 
4226     all = cmd->zsflags & NVME_ZSFLAG_SELECT_ALL;
4227 
4228     req->status = NVME_SUCCESS;
4229 
4230     if (!all) {
4231         status = nvme_get_mgmt_zone_slba_idx(ns, &req->cmd, &slba, &zone_idx);
4232         if (status) {
4233             return status;
4234         }
4235     }
4236 
4237     zone = &ns->zone_array[zone_idx];
4238     if (slba != zone->d.zslba && action != NVME_ZONE_ACTION_ZRWA_FLUSH) {
4239         trace_pci_nvme_err_unaligned_zone_cmd(action, slba, zone->d.zslba);
4240         return NVME_INVALID_FIELD | NVME_DNR;
4241     }
4242 
4243     switch (action) {
4244 
4245     case NVME_ZONE_ACTION_OPEN:
4246         if (all) {
4247             proc_mask = NVME_PROC_CLOSED_ZONES;
4248         }
4249         trace_pci_nvme_open_zone(slba, zone_idx, all);
4250         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_open_zone, req);
4251         break;
4252 
4253     case NVME_ZONE_ACTION_CLOSE:
4254         if (all) {
4255             proc_mask = NVME_PROC_OPENED_ZONES;
4256         }
4257         trace_pci_nvme_close_zone(slba, zone_idx, all);
4258         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_close_zone, req);
4259         break;
4260 
4261     case NVME_ZONE_ACTION_FINISH:
4262         if (all) {
4263             proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES;
4264         }
4265         trace_pci_nvme_finish_zone(slba, zone_idx, all);
4266         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_finish_zone, req);
4267         break;
4268 
4269     case NVME_ZONE_ACTION_RESET:
4270         trace_pci_nvme_reset_zone(slba, zone_idx, all);
4271 
4272         iocb = blk_aio_get(&nvme_zone_reset_aiocb_info, ns->blkconf.blk,
4273                            nvme_misc_cb, req);
4274 
4275         iocb->req = req;
4276         iocb->ret = 0;
4277         iocb->all = all;
4278         iocb->idx = zone_idx;
4279         iocb->zone = NULL;
4280 
4281         req->aiocb = &iocb->common;
4282         nvme_zone_reset_cb(iocb, 0);
4283 
4284         return NVME_NO_COMPLETE;
4285 
4286     case NVME_ZONE_ACTION_OFFLINE:
4287         if (all) {
4288             proc_mask = NVME_PROC_READ_ONLY_ZONES;
4289         }
4290         trace_pci_nvme_offline_zone(slba, zone_idx, all);
4291         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_offline_zone, req);
4292         break;
4293 
4294     case NVME_ZONE_ACTION_SET_ZD_EXT:
4295         trace_pci_nvme_set_descriptor_extension(slba, zone_idx);
4296         if (all || !ns->params.zd_extension_size) {
4297             return NVME_INVALID_FIELD | NVME_DNR;
4298         }
4299         zd_ext = nvme_get_zd_extension(ns, zone_idx);
4300         status = nvme_h2c(n, zd_ext, ns->params.zd_extension_size, req);
4301         if (status) {
4302             trace_pci_nvme_err_zd_extension_map_error(zone_idx);
4303             return status;
4304         }
4305 
4306         status = nvme_set_zd_ext(ns, zone);
4307         if (status == NVME_SUCCESS) {
4308             trace_pci_nvme_zd_extension_set(zone_idx);
4309             return status;
4310         }
4311         break;
4312 
4313     case NVME_ZONE_ACTION_ZRWA_FLUSH:
4314         if (all) {
4315             return NVME_INVALID_FIELD | NVME_DNR;
4316         }
4317 
4318         return nvme_zone_mgmt_send_zrwa_flush(n, zone, slba, req);
4319 
4320     default:
4321         trace_pci_nvme_err_invalid_mgmt_action(action);
4322         status = NVME_INVALID_FIELD;
4323     }
4324 
4325     if (status == NVME_ZONE_INVAL_TRANSITION) {
4326         trace_pci_nvme_err_invalid_zone_state_transition(action, slba,
4327                                                          zone->d.za);
4328     }
4329     if (status) {
4330         status |= NVME_DNR;
4331     }
4332 
4333     return status;
4334 }
4335 
4336 static bool nvme_zone_matches_filter(uint32_t zafs, NvmeZone *zl)
4337 {
4338     NvmeZoneState zs = nvme_get_zone_state(zl);
4339 
4340     switch (zafs) {
4341     case NVME_ZONE_REPORT_ALL:
4342         return true;
4343     case NVME_ZONE_REPORT_EMPTY:
4344         return zs == NVME_ZONE_STATE_EMPTY;
4345     case NVME_ZONE_REPORT_IMPLICITLY_OPEN:
4346         return zs == NVME_ZONE_STATE_IMPLICITLY_OPEN;
4347     case NVME_ZONE_REPORT_EXPLICITLY_OPEN:
4348         return zs == NVME_ZONE_STATE_EXPLICITLY_OPEN;
4349     case NVME_ZONE_REPORT_CLOSED:
4350         return zs == NVME_ZONE_STATE_CLOSED;
4351     case NVME_ZONE_REPORT_FULL:
4352         return zs == NVME_ZONE_STATE_FULL;
4353     case NVME_ZONE_REPORT_READ_ONLY:
4354         return zs == NVME_ZONE_STATE_READ_ONLY;
4355     case NVME_ZONE_REPORT_OFFLINE:
4356         return zs == NVME_ZONE_STATE_OFFLINE;
4357     default:
4358         return false;
4359     }
4360 }
4361 
4362 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4363 {
4364     NvmeCmd *cmd = &req->cmd;
4365     NvmeNamespace *ns = req->ns;
4366     /* cdw12 is zero-based number of dwords to return. Convert to bytes */
4367     uint32_t data_size = (le32_to_cpu(cmd->cdw12) + 1) << 2;
4368     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
4369     uint32_t zone_idx, zra, zrasf, partial;
4370     uint64_t max_zones, nr_zones = 0;
4371     uint16_t status;
4372     uint64_t slba;
4373     NvmeZoneDescr *z;
4374     NvmeZone *zone;
4375     NvmeZoneReportHeader *header;
4376     void *buf, *buf_p;
4377     size_t zone_entry_sz;
4378     int i;
4379 
4380     req->status = NVME_SUCCESS;
4381 
4382     status = nvme_get_mgmt_zone_slba_idx(ns, cmd, &slba, &zone_idx);
4383     if (status) {
4384         return status;
4385     }
4386 
4387     zra = dw13 & 0xff;
4388     if (zra != NVME_ZONE_REPORT && zra != NVME_ZONE_REPORT_EXTENDED) {
4389         return NVME_INVALID_FIELD | NVME_DNR;
4390     }
4391     if (zra == NVME_ZONE_REPORT_EXTENDED && !ns->params.zd_extension_size) {
4392         return NVME_INVALID_FIELD | NVME_DNR;
4393     }
4394 
4395     zrasf = (dw13 >> 8) & 0xff;
4396     if (zrasf > NVME_ZONE_REPORT_OFFLINE) {
4397         return NVME_INVALID_FIELD | NVME_DNR;
4398     }
4399 
4400     if (data_size < sizeof(NvmeZoneReportHeader)) {
4401         return NVME_INVALID_FIELD | NVME_DNR;
4402     }
4403 
4404     status = nvme_check_mdts(n, data_size);
4405     if (status) {
4406         return status;
4407     }
4408 
4409     partial = (dw13 >> 16) & 0x01;
4410 
4411     zone_entry_sz = sizeof(NvmeZoneDescr);
4412     if (zra == NVME_ZONE_REPORT_EXTENDED) {
4413         zone_entry_sz += ns->params.zd_extension_size;
4414     }
4415 
4416     max_zones = (data_size - sizeof(NvmeZoneReportHeader)) / zone_entry_sz;
4417     buf = g_malloc0(data_size);
4418 
4419     zone = &ns->zone_array[zone_idx];
4420     for (i = zone_idx; i < ns->num_zones; i++) {
4421         if (partial && nr_zones >= max_zones) {
4422             break;
4423         }
4424         if (nvme_zone_matches_filter(zrasf, zone++)) {
4425             nr_zones++;
4426         }
4427     }
4428     header = buf;
4429     header->nr_zones = cpu_to_le64(nr_zones);
4430 
4431     buf_p = buf + sizeof(NvmeZoneReportHeader);
4432     for (; zone_idx < ns->num_zones && max_zones > 0; zone_idx++) {
4433         zone = &ns->zone_array[zone_idx];
4434         if (nvme_zone_matches_filter(zrasf, zone)) {
4435             z = buf_p;
4436             buf_p += sizeof(NvmeZoneDescr);
4437 
4438             z->zt = zone->d.zt;
4439             z->zs = zone->d.zs;
4440             z->zcap = cpu_to_le64(zone->d.zcap);
4441             z->zslba = cpu_to_le64(zone->d.zslba);
4442             z->za = zone->d.za;
4443 
4444             if (nvme_wp_is_valid(zone)) {
4445                 z->wp = cpu_to_le64(zone->d.wp);
4446             } else {
4447                 z->wp = cpu_to_le64(~0ULL);
4448             }
4449 
4450             if (zra == NVME_ZONE_REPORT_EXTENDED) {
4451                 if (zone->d.za & NVME_ZA_ZD_EXT_VALID) {
4452                     memcpy(buf_p, nvme_get_zd_extension(ns, zone_idx),
4453                            ns->params.zd_extension_size);
4454                 }
4455                 buf_p += ns->params.zd_extension_size;
4456             }
4457 
4458             max_zones--;
4459         }
4460     }
4461 
4462     status = nvme_c2h(n, (uint8_t *)buf, data_size, req);
4463 
4464     g_free(buf);
4465 
4466     return status;
4467 }
4468 
4469 static uint16_t nvme_io_mgmt_recv_ruhs(NvmeCtrl *n, NvmeRequest *req,
4470                                        size_t len)
4471 {
4472     NvmeNamespace *ns = req->ns;
4473     NvmeEnduranceGroup *endgrp;
4474     NvmeRuhStatus *hdr;
4475     NvmeRuhStatusDescr *ruhsd;
4476     unsigned int nruhsd;
4477     uint16_t rg, ph, *ruhid;
4478     size_t trans_len;
4479     g_autofree uint8_t *buf = NULL;
4480 
4481     if (!n->subsys) {
4482         return NVME_INVALID_FIELD | NVME_DNR;
4483     }
4484 
4485     if (ns->params.nsid == 0 || ns->params.nsid == 0xffffffff) {
4486         return NVME_INVALID_NSID | NVME_DNR;
4487     }
4488 
4489     if (!n->subsys->endgrp.fdp.enabled) {
4490         return NVME_FDP_DISABLED | NVME_DNR;
4491     }
4492 
4493     endgrp = ns->endgrp;
4494 
4495     nruhsd = ns->fdp.nphs * endgrp->fdp.nrg;
4496     trans_len = sizeof(NvmeRuhStatus) + nruhsd * sizeof(NvmeRuhStatusDescr);
4497     buf = g_malloc0(trans_len);
4498 
4499     trans_len = MIN(trans_len, len);
4500 
4501     hdr = (NvmeRuhStatus *)buf;
4502     ruhsd = (NvmeRuhStatusDescr *)(buf + sizeof(NvmeRuhStatus));
4503 
4504     hdr->nruhsd = cpu_to_le16(nruhsd);
4505 
4506     ruhid = ns->fdp.phs;
4507 
4508     for (ph = 0; ph < ns->fdp.nphs; ph++, ruhid++) {
4509         NvmeRuHandle *ruh = &endgrp->fdp.ruhs[*ruhid];
4510 
4511         for (rg = 0; rg < endgrp->fdp.nrg; rg++, ruhsd++) {
4512             uint16_t pid = nvme_make_pid(ns, rg, ph);
4513 
4514             ruhsd->pid = cpu_to_le16(pid);
4515             ruhsd->ruhid = *ruhid;
4516             ruhsd->earutr = 0;
4517             ruhsd->ruamw = cpu_to_le64(ruh->rus[rg].ruamw);
4518         }
4519     }
4520 
4521     return nvme_c2h(n, buf, trans_len, req);
4522 }
4523 
4524 static uint16_t nvme_io_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4525 {
4526     NvmeCmd *cmd = &req->cmd;
4527     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4528     uint32_t numd = le32_to_cpu(cmd->cdw11);
4529     uint8_t mo = (cdw10 & 0xff);
4530     size_t len = (numd + 1) << 2;
4531 
4532     switch (mo) {
4533     case NVME_IOMR_MO_NOP:
4534         return 0;
4535     case NVME_IOMR_MO_RUH_STATUS:
4536         return nvme_io_mgmt_recv_ruhs(n, req, len);
4537     default:
4538         return NVME_INVALID_FIELD | NVME_DNR;
4539     };
4540 }
4541 
4542 static uint16_t nvme_io_mgmt_send_ruh_update(NvmeCtrl *n, NvmeRequest *req)
4543 {
4544     NvmeCmd *cmd = &req->cmd;
4545     NvmeNamespace *ns = req->ns;
4546     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4547     uint16_t ret = NVME_SUCCESS;
4548     uint32_t npid = (cdw10 >> 16) + 1;
4549     unsigned int i = 0;
4550     g_autofree uint16_t *pids = NULL;
4551     uint32_t maxnpid;
4552 
4553     if (!ns->endgrp || !ns->endgrp->fdp.enabled) {
4554         return NVME_FDP_DISABLED | NVME_DNR;
4555     }
4556 
4557     maxnpid = n->subsys->endgrp.fdp.nrg * n->subsys->endgrp.fdp.nruh;
4558 
4559     if (unlikely(npid >= MIN(NVME_FDP_MAXPIDS, maxnpid))) {
4560         return NVME_INVALID_FIELD | NVME_DNR;
4561     }
4562 
4563     pids = g_new(uint16_t, npid);
4564 
4565     ret = nvme_h2c(n, pids, npid * sizeof(uint16_t), req);
4566     if (ret) {
4567         return ret;
4568     }
4569 
4570     for (; i < npid; i++) {
4571         if (!nvme_update_ruh(n, ns, pids[i])) {
4572             return NVME_INVALID_FIELD | NVME_DNR;
4573         }
4574     }
4575 
4576     return ret;
4577 }
4578 
4579 static uint16_t nvme_io_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4580 {
4581     NvmeCmd *cmd = &req->cmd;
4582     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4583     uint8_t mo = (cdw10 & 0xff);
4584 
4585     switch (mo) {
4586     case NVME_IOMS_MO_NOP:
4587         return 0;
4588     case NVME_IOMS_MO_RUH_UPDATE:
4589         return nvme_io_mgmt_send_ruh_update(n, req);
4590     default:
4591         return NVME_INVALID_FIELD | NVME_DNR;
4592     };
4593 }
4594 
4595 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
4596 {
4597     NvmeNamespace *ns;
4598     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4599 
4600     trace_pci_nvme_io_cmd(nvme_cid(req), nsid, nvme_sqid(req),
4601                           req->cmd.opcode, nvme_io_opc_str(req->cmd.opcode));
4602 
4603     /*
4604      * In the base NVM command set, Flush may apply to all namespaces
4605      * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4606      * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4607      *
4608      * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4609      * opcode with a specific command since we cannot determine a unique I/O
4610      * command set. Opcode 0h could have any other meaning than something
4611      * equivalent to flushing and say it DOES have completely different
4612      * semantics in some other command set - does an NSID of FFFFFFFFh then
4613      * mean "for all namespaces, apply whatever command set specific command
4614      * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4615      * whatever command that uses the 0h opcode if, and only if, it allows NSID
4616      * to be FFFFFFFFh"?
4617      *
4618      * Anyway (and luckily), for now, we do not care about this since the
4619      * device only supports namespace types that includes the NVM Flush command
4620      * (NVM and Zoned), so always do an NVM Flush.
4621      */
4622 
4623     if (req->cmd.opcode == NVME_CMD_FLUSH) {
4624         return nvme_flush(n, req);
4625     }
4626 
4627     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4628         return NVME_INVALID_NSID | NVME_DNR;
4629     }
4630 
4631     ns = nvme_ns(n, nsid);
4632     if (unlikely(!ns)) {
4633         return NVME_INVALID_FIELD | NVME_DNR;
4634     }
4635 
4636     if (!(ns->iocs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
4637         trace_pci_nvme_err_invalid_opc(req->cmd.opcode);
4638         return NVME_INVALID_OPCODE | NVME_DNR;
4639     }
4640 
4641     if (ns->status) {
4642         return ns->status;
4643     }
4644 
4645     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
4646         return NVME_INVALID_FIELD;
4647     }
4648 
4649     req->ns = ns;
4650 
4651     switch (req->cmd.opcode) {
4652     case NVME_CMD_WRITE_ZEROES:
4653         return nvme_write_zeroes(n, req);
4654     case NVME_CMD_ZONE_APPEND:
4655         return nvme_zone_append(n, req);
4656     case NVME_CMD_WRITE:
4657         return nvme_write(n, req);
4658     case NVME_CMD_READ:
4659         return nvme_read(n, req);
4660     case NVME_CMD_COMPARE:
4661         return nvme_compare(n, req);
4662     case NVME_CMD_DSM:
4663         return nvme_dsm(n, req);
4664     case NVME_CMD_VERIFY:
4665         return nvme_verify(n, req);
4666     case NVME_CMD_COPY:
4667         return nvme_copy(n, req);
4668     case NVME_CMD_ZONE_MGMT_SEND:
4669         return nvme_zone_mgmt_send(n, req);
4670     case NVME_CMD_ZONE_MGMT_RECV:
4671         return nvme_zone_mgmt_recv(n, req);
4672     case NVME_CMD_IO_MGMT_RECV:
4673         return nvme_io_mgmt_recv(n, req);
4674     case NVME_CMD_IO_MGMT_SEND:
4675         return nvme_io_mgmt_send(n, req);
4676     default:
4677         g_assert_not_reached();
4678     }
4679 
4680     return NVME_INVALID_OPCODE | NVME_DNR;
4681 }
4682 
4683 static void nvme_cq_notifier(EventNotifier *e)
4684 {
4685     NvmeCQueue *cq = container_of(e, NvmeCQueue, notifier);
4686     NvmeCtrl *n = cq->ctrl;
4687 
4688     if (!event_notifier_test_and_clear(e)) {
4689         return;
4690     }
4691 
4692     nvme_update_cq_head(cq);
4693 
4694     if (cq->tail == cq->head) {
4695         if (cq->irq_enabled) {
4696             n->cq_pending--;
4697         }
4698 
4699         nvme_irq_deassert(n, cq);
4700     }
4701 
4702     qemu_bh_schedule(cq->bh);
4703 }
4704 
4705 static int nvme_init_cq_ioeventfd(NvmeCQueue *cq)
4706 {
4707     NvmeCtrl *n = cq->ctrl;
4708     uint16_t offset = (cq->cqid << 3) + (1 << 2);
4709     int ret;
4710 
4711     ret = event_notifier_init(&cq->notifier, 0);
4712     if (ret < 0) {
4713         return ret;
4714     }
4715 
4716     event_notifier_set_handler(&cq->notifier, nvme_cq_notifier);
4717     memory_region_add_eventfd(&n->iomem,
4718                               0x1000 + offset, 4, false, 0, &cq->notifier);
4719 
4720     return 0;
4721 }
4722 
4723 static void nvme_sq_notifier(EventNotifier *e)
4724 {
4725     NvmeSQueue *sq = container_of(e, NvmeSQueue, notifier);
4726 
4727     if (!event_notifier_test_and_clear(e)) {
4728         return;
4729     }
4730 
4731     nvme_process_sq(sq);
4732 }
4733 
4734 static int nvme_init_sq_ioeventfd(NvmeSQueue *sq)
4735 {
4736     NvmeCtrl *n = sq->ctrl;
4737     uint16_t offset = sq->sqid << 3;
4738     int ret;
4739 
4740     ret = event_notifier_init(&sq->notifier, 0);
4741     if (ret < 0) {
4742         return ret;
4743     }
4744 
4745     event_notifier_set_handler(&sq->notifier, nvme_sq_notifier);
4746     memory_region_add_eventfd(&n->iomem,
4747                               0x1000 + offset, 4, false, 0, &sq->notifier);
4748 
4749     return 0;
4750 }
4751 
4752 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
4753 {
4754     uint16_t offset = sq->sqid << 3;
4755 
4756     n->sq[sq->sqid] = NULL;
4757     qemu_bh_delete(sq->bh);
4758     if (sq->ioeventfd_enabled) {
4759         memory_region_del_eventfd(&n->iomem,
4760                                   0x1000 + offset, 4, false, 0, &sq->notifier);
4761         event_notifier_set_handler(&sq->notifier, NULL);
4762         event_notifier_cleanup(&sq->notifier);
4763     }
4764     g_free(sq->io_req);
4765     if (sq->sqid) {
4766         g_free(sq);
4767     }
4768 }
4769 
4770 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
4771 {
4772     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4773     NvmeRequest *r, *next;
4774     NvmeSQueue *sq;
4775     NvmeCQueue *cq;
4776     uint16_t qid = le16_to_cpu(c->qid);
4777 
4778     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
4779         trace_pci_nvme_err_invalid_del_sq(qid);
4780         return NVME_INVALID_QID | NVME_DNR;
4781     }
4782 
4783     trace_pci_nvme_del_sq(qid);
4784 
4785     sq = n->sq[qid];
4786     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
4787         r = QTAILQ_FIRST(&sq->out_req_list);
4788         assert(r->aiocb);
4789         blk_aio_cancel(r->aiocb);
4790     }
4791 
4792     assert(QTAILQ_EMPTY(&sq->out_req_list));
4793 
4794     if (!nvme_check_cqid(n, sq->cqid)) {
4795         cq = n->cq[sq->cqid];
4796         QTAILQ_REMOVE(&cq->sq_list, sq, entry);
4797 
4798         nvme_post_cqes(cq);
4799         QTAILQ_FOREACH_SAFE(r, &cq->req_list, entry, next) {
4800             if (r->sq == sq) {
4801                 QTAILQ_REMOVE(&cq->req_list, r, entry);
4802                 QTAILQ_INSERT_TAIL(&sq->req_list, r, entry);
4803             }
4804         }
4805     }
4806 
4807     nvme_free_sq(sq, n);
4808     return NVME_SUCCESS;
4809 }
4810 
4811 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
4812                          uint16_t sqid, uint16_t cqid, uint16_t size)
4813 {
4814     int i;
4815     NvmeCQueue *cq;
4816 
4817     sq->ctrl = n;
4818     sq->dma_addr = dma_addr;
4819     sq->sqid = sqid;
4820     sq->size = size;
4821     sq->cqid = cqid;
4822     sq->head = sq->tail = 0;
4823     sq->io_req = g_new0(NvmeRequest, sq->size);
4824 
4825     QTAILQ_INIT(&sq->req_list);
4826     QTAILQ_INIT(&sq->out_req_list);
4827     for (i = 0; i < sq->size; i++) {
4828         sq->io_req[i].sq = sq;
4829         QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
4830     }
4831 
4832     sq->bh = qemu_bh_new_guarded(nvme_process_sq, sq,
4833                                  &DEVICE(sq->ctrl)->mem_reentrancy_guard);
4834 
4835     if (n->dbbuf_enabled) {
4836         sq->db_addr = n->dbbuf_dbs + (sqid << 3);
4837         sq->ei_addr = n->dbbuf_eis + (sqid << 3);
4838 
4839         if (n->params.ioeventfd && sq->sqid != 0) {
4840             if (!nvme_init_sq_ioeventfd(sq)) {
4841                 sq->ioeventfd_enabled = true;
4842             }
4843         }
4844     }
4845 
4846     assert(n->cq[cqid]);
4847     cq = n->cq[cqid];
4848     QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
4849     n->sq[sqid] = sq;
4850 }
4851 
4852 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeRequest *req)
4853 {
4854     NvmeSQueue *sq;
4855     NvmeCreateSq *c = (NvmeCreateSq *)&req->cmd;
4856 
4857     uint16_t cqid = le16_to_cpu(c->cqid);
4858     uint16_t sqid = le16_to_cpu(c->sqid);
4859     uint16_t qsize = le16_to_cpu(c->qsize);
4860     uint16_t qflags = le16_to_cpu(c->sq_flags);
4861     uint64_t prp1 = le64_to_cpu(c->prp1);
4862 
4863     trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
4864 
4865     if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
4866         trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
4867         return NVME_INVALID_CQID | NVME_DNR;
4868     }
4869     if (unlikely(!sqid || sqid > n->conf_ioqpairs || n->sq[sqid] != NULL)) {
4870         trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
4871         return NVME_INVALID_QID | NVME_DNR;
4872     }
4873     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
4874         trace_pci_nvme_err_invalid_create_sq_size(qsize);
4875         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4876     }
4877     if (unlikely(prp1 & (n->page_size - 1))) {
4878         trace_pci_nvme_err_invalid_create_sq_addr(prp1);
4879         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4880     }
4881     if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
4882         trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
4883         return NVME_INVALID_FIELD | NVME_DNR;
4884     }
4885     sq = g_malloc0(sizeof(*sq));
4886     nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
4887     return NVME_SUCCESS;
4888 }
4889 
4890 struct nvme_stats {
4891     uint64_t units_read;
4892     uint64_t units_written;
4893     uint64_t read_commands;
4894     uint64_t write_commands;
4895 };
4896 
4897 static void nvme_set_blk_stats(NvmeNamespace *ns, struct nvme_stats *stats)
4898 {
4899     BlockAcctStats *s = blk_get_stats(ns->blkconf.blk);
4900 
4901     stats->units_read += s->nr_bytes[BLOCK_ACCT_READ];
4902     stats->units_written += s->nr_bytes[BLOCK_ACCT_WRITE];
4903     stats->read_commands += s->nr_ops[BLOCK_ACCT_READ];
4904     stats->write_commands += s->nr_ops[BLOCK_ACCT_WRITE];
4905 }
4906 
4907 static uint16_t nvme_smart_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4908                                 uint64_t off, NvmeRequest *req)
4909 {
4910     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4911     struct nvme_stats stats = { 0 };
4912     NvmeSmartLog smart = { 0 };
4913     uint32_t trans_len;
4914     NvmeNamespace *ns;
4915     time_t current_ms;
4916     uint64_t u_read, u_written;
4917 
4918     if (off >= sizeof(smart)) {
4919         return NVME_INVALID_FIELD | NVME_DNR;
4920     }
4921 
4922     if (nsid != 0xffffffff) {
4923         ns = nvme_ns(n, nsid);
4924         if (!ns) {
4925             return NVME_INVALID_NSID | NVME_DNR;
4926         }
4927         nvme_set_blk_stats(ns, &stats);
4928     } else {
4929         int i;
4930 
4931         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4932             ns = nvme_ns(n, i);
4933             if (!ns) {
4934                 continue;
4935             }
4936             nvme_set_blk_stats(ns, &stats);
4937         }
4938     }
4939 
4940     trans_len = MIN(sizeof(smart) - off, buf_len);
4941     smart.critical_warning = n->smart_critical_warning;
4942 
4943     u_read = DIV_ROUND_UP(stats.units_read >> BDRV_SECTOR_BITS, 1000);
4944     u_written = DIV_ROUND_UP(stats.units_written >> BDRV_SECTOR_BITS, 1000);
4945 
4946     smart.data_units_read[0] = cpu_to_le64(u_read);
4947     smart.data_units_written[0] = cpu_to_le64(u_written);
4948     smart.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4949     smart.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4950 
4951     smart.temperature = cpu_to_le16(n->temperature);
4952 
4953     if ((n->temperature >= n->features.temp_thresh_hi) ||
4954         (n->temperature <= n->features.temp_thresh_low)) {
4955         smart.critical_warning |= NVME_SMART_TEMPERATURE;
4956     }
4957 
4958     current_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4959     smart.power_on_hours[0] =
4960         cpu_to_le64((((current_ms - n->starttime_ms) / 1000) / 60) / 60);
4961 
4962     if (!rae) {
4963         nvme_clear_events(n, NVME_AER_TYPE_SMART);
4964     }
4965 
4966     return nvme_c2h(n, (uint8_t *) &smart + off, trans_len, req);
4967 }
4968 
4969 static uint16_t nvme_endgrp_info(NvmeCtrl *n,  uint8_t rae, uint32_t buf_len,
4970                                  uint64_t off, NvmeRequest *req)
4971 {
4972     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
4973     uint16_t endgrpid = (dw11 >> 16) & 0xffff;
4974     struct nvme_stats stats = {};
4975     NvmeEndGrpLog info = {};
4976     int i;
4977 
4978     if (!n->subsys || endgrpid != 0x1) {
4979         return NVME_INVALID_FIELD | NVME_DNR;
4980     }
4981 
4982     if (off >= sizeof(info)) {
4983         return NVME_INVALID_FIELD | NVME_DNR;
4984     }
4985 
4986     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4987         NvmeNamespace *ns = nvme_subsys_ns(n->subsys, i);
4988         if (!ns) {
4989             continue;
4990         }
4991 
4992         nvme_set_blk_stats(ns, &stats);
4993     }
4994 
4995     info.data_units_read[0] =
4996         cpu_to_le64(DIV_ROUND_UP(stats.units_read / 1000000000, 1000000000));
4997     info.data_units_written[0] =
4998         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4999     info.media_units_written[0] =
5000         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
5001 
5002     info.host_read_commands[0] = cpu_to_le64(stats.read_commands);
5003     info.host_write_commands[0] = cpu_to_le64(stats.write_commands);
5004 
5005     buf_len = MIN(sizeof(info) - off, buf_len);
5006 
5007     return nvme_c2h(n, (uint8_t *)&info + off, buf_len, req);
5008 }
5009 
5010 
5011 static uint16_t nvme_fw_log_info(NvmeCtrl *n, uint32_t buf_len, uint64_t off,
5012                                  NvmeRequest *req)
5013 {
5014     uint32_t trans_len;
5015     NvmeFwSlotInfoLog fw_log = {
5016         .afi = 0x1,
5017     };
5018 
5019     if (off >= sizeof(fw_log)) {
5020         return NVME_INVALID_FIELD | NVME_DNR;
5021     }
5022 
5023     strpadcpy((char *)&fw_log.frs1, sizeof(fw_log.frs1), "1.0", ' ');
5024     trans_len = MIN(sizeof(fw_log) - off, buf_len);
5025 
5026     return nvme_c2h(n, (uint8_t *) &fw_log + off, trans_len, req);
5027 }
5028 
5029 static uint16_t nvme_error_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
5030                                 uint64_t off, NvmeRequest *req)
5031 {
5032     uint32_t trans_len;
5033     NvmeErrorLog errlog;
5034 
5035     if (off >= sizeof(errlog)) {
5036         return NVME_INVALID_FIELD | NVME_DNR;
5037     }
5038 
5039     if (!rae) {
5040         nvme_clear_events(n, NVME_AER_TYPE_ERROR);
5041     }
5042 
5043     memset(&errlog, 0x0, sizeof(errlog));
5044     trans_len = MIN(sizeof(errlog) - off, buf_len);
5045 
5046     return nvme_c2h(n, (uint8_t *)&errlog, trans_len, req);
5047 }
5048 
5049 static uint16_t nvme_changed_nslist(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
5050                                     uint64_t off, NvmeRequest *req)
5051 {
5052     uint32_t nslist[1024];
5053     uint32_t trans_len;
5054     int i = 0;
5055     uint32_t nsid;
5056 
5057     if (off >= sizeof(nslist)) {
5058         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(nslist));
5059         return NVME_INVALID_FIELD | NVME_DNR;
5060     }
5061 
5062     memset(nslist, 0x0, sizeof(nslist));
5063     trans_len = MIN(sizeof(nslist) - off, buf_len);
5064 
5065     while ((nsid = find_first_bit(n->changed_nsids, NVME_CHANGED_NSID_SIZE)) !=
5066             NVME_CHANGED_NSID_SIZE) {
5067         /*
5068          * If more than 1024 namespaces, the first entry in the log page should
5069          * be set to FFFFFFFFh and the others to 0 as spec.
5070          */
5071         if (i == ARRAY_SIZE(nslist)) {
5072             memset(nslist, 0x0, sizeof(nslist));
5073             nslist[0] = 0xffffffff;
5074             break;
5075         }
5076 
5077         nslist[i++] = nsid;
5078         clear_bit(nsid, n->changed_nsids);
5079     }
5080 
5081     /*
5082      * Remove all the remaining list entries in case returns directly due to
5083      * more than 1024 namespaces.
5084      */
5085     if (nslist[0] == 0xffffffff) {
5086         bitmap_zero(n->changed_nsids, NVME_CHANGED_NSID_SIZE);
5087     }
5088 
5089     if (!rae) {
5090         nvme_clear_events(n, NVME_AER_TYPE_NOTICE);
5091     }
5092 
5093     return nvme_c2h(n, ((uint8_t *)nslist) + off, trans_len, req);
5094 }
5095 
5096 static uint16_t nvme_cmd_effects(NvmeCtrl *n, uint8_t csi, uint32_t buf_len,
5097                                  uint64_t off, NvmeRequest *req)
5098 {
5099     NvmeEffectsLog log = {};
5100     const uint32_t *src_iocs = NULL;
5101     uint32_t trans_len;
5102 
5103     if (off >= sizeof(log)) {
5104         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(log));
5105         return NVME_INVALID_FIELD | NVME_DNR;
5106     }
5107 
5108     switch (NVME_CC_CSS(ldl_le_p(&n->bar.cc))) {
5109     case NVME_CC_CSS_NVM:
5110         src_iocs = nvme_cse_iocs_nvm;
5111         /* fall through */
5112     case NVME_CC_CSS_ADMIN_ONLY:
5113         break;
5114     case NVME_CC_CSS_CSI:
5115         switch (csi) {
5116         case NVME_CSI_NVM:
5117             src_iocs = nvme_cse_iocs_nvm;
5118             break;
5119         case NVME_CSI_ZONED:
5120             src_iocs = nvme_cse_iocs_zoned;
5121             break;
5122         }
5123     }
5124 
5125     memcpy(log.acs, nvme_cse_acs, sizeof(nvme_cse_acs));
5126 
5127     if (src_iocs) {
5128         memcpy(log.iocs, src_iocs, sizeof(log.iocs));
5129     }
5130 
5131     trans_len = MIN(sizeof(log) - off, buf_len);
5132 
5133     return nvme_c2h(n, ((uint8_t *)&log) + off, trans_len, req);
5134 }
5135 
5136 static size_t sizeof_fdp_conf_descr(size_t nruh, size_t vss)
5137 {
5138     size_t entry_siz = sizeof(NvmeFdpDescrHdr) + nruh * sizeof(NvmeRuhDescr)
5139                        + vss;
5140     return ROUND_UP(entry_siz, 8);
5141 }
5142 
5143 static uint16_t nvme_fdp_confs(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
5144                                uint64_t off, NvmeRequest *req)
5145 {
5146     uint32_t log_size, trans_len;
5147     g_autofree uint8_t *buf = NULL;
5148     NvmeFdpDescrHdr *hdr;
5149     NvmeRuhDescr *ruhd;
5150     NvmeEnduranceGroup *endgrp;
5151     NvmeFdpConfsHdr *log;
5152     size_t nruh, fdp_descr_size;
5153     int i;
5154 
5155     if (endgrpid != 1 || !n->subsys) {
5156         return NVME_INVALID_FIELD | NVME_DNR;
5157     }
5158 
5159     endgrp = &n->subsys->endgrp;
5160 
5161     if (endgrp->fdp.enabled) {
5162         nruh = endgrp->fdp.nruh;
5163     } else {
5164         nruh = 1;
5165     }
5166 
5167     fdp_descr_size = sizeof_fdp_conf_descr(nruh, FDPVSS);
5168     log_size = sizeof(NvmeFdpConfsHdr) + fdp_descr_size;
5169 
5170     if (off >= log_size) {
5171         return NVME_INVALID_FIELD | NVME_DNR;
5172     }
5173 
5174     trans_len = MIN(log_size - off, buf_len);
5175 
5176     buf = g_malloc0(log_size);
5177     log = (NvmeFdpConfsHdr *)buf;
5178     hdr = (NvmeFdpDescrHdr *)(log + 1);
5179     ruhd = (NvmeRuhDescr *)(buf + sizeof(*log) + sizeof(*hdr));
5180 
5181     log->num_confs = cpu_to_le16(0);
5182     log->size = cpu_to_le32(log_size);
5183 
5184     hdr->descr_size = cpu_to_le16(fdp_descr_size);
5185     if (endgrp->fdp.enabled) {
5186         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, VALID, 1);
5187         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, RGIF, endgrp->fdp.rgif);
5188         hdr->nrg = cpu_to_le16(endgrp->fdp.nrg);
5189         hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5190         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5191         hdr->nnss = cpu_to_le32(NVME_MAX_NAMESPACES);
5192         hdr->runs = cpu_to_le64(endgrp->fdp.runs);
5193 
5194         for (i = 0; i < nruh; i++) {
5195             ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5196             ruhd++;
5197         }
5198     } else {
5199         /* 1 bit for RUH in PIF -> 2 RUHs max. */
5200         hdr->nrg = cpu_to_le16(1);
5201         hdr->nruh = cpu_to_le16(1);
5202         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5203         hdr->nnss = cpu_to_le32(1);
5204         hdr->runs = cpu_to_le64(96 * MiB);
5205 
5206         ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5207     }
5208 
5209     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5210 }
5211 
5212 static uint16_t nvme_fdp_ruh_usage(NvmeCtrl *n, uint32_t endgrpid,
5213                                    uint32_t dw10, uint32_t dw12,
5214                                    uint32_t buf_len, uint64_t off,
5215                                    NvmeRequest *req)
5216 {
5217     NvmeRuHandle *ruh;
5218     NvmeRuhuLog *hdr;
5219     NvmeRuhuDescr *ruhud;
5220     NvmeEnduranceGroup *endgrp;
5221     g_autofree uint8_t *buf = NULL;
5222     uint32_t log_size, trans_len;
5223     uint16_t i;
5224 
5225     if (endgrpid != 1 || !n->subsys) {
5226         return NVME_INVALID_FIELD | NVME_DNR;
5227     }
5228 
5229     endgrp = &n->subsys->endgrp;
5230 
5231     if (!endgrp->fdp.enabled) {
5232         return NVME_FDP_DISABLED | NVME_DNR;
5233     }
5234 
5235     log_size = sizeof(NvmeRuhuLog) + endgrp->fdp.nruh * sizeof(NvmeRuhuDescr);
5236 
5237     if (off >= log_size) {
5238         return NVME_INVALID_FIELD | NVME_DNR;
5239     }
5240 
5241     trans_len = MIN(log_size - off, buf_len);
5242 
5243     buf = g_malloc0(log_size);
5244     hdr = (NvmeRuhuLog *)buf;
5245     ruhud = (NvmeRuhuDescr *)(hdr + 1);
5246 
5247     ruh = endgrp->fdp.ruhs;
5248     hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5249 
5250     for (i = 0; i < endgrp->fdp.nruh; i++, ruhud++, ruh++) {
5251         ruhud->ruha = ruh->ruha;
5252     }
5253 
5254     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5255 }
5256 
5257 static uint16_t nvme_fdp_stats(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
5258                                uint64_t off, NvmeRequest *req)
5259 {
5260     NvmeEnduranceGroup *endgrp;
5261     NvmeFdpStatsLog log = {};
5262     uint32_t trans_len;
5263 
5264     if (off >= sizeof(NvmeFdpStatsLog)) {
5265         return NVME_INVALID_FIELD | NVME_DNR;
5266     }
5267 
5268     if (endgrpid != 1 || !n->subsys) {
5269         return NVME_INVALID_FIELD | NVME_DNR;
5270     }
5271 
5272     if (!n->subsys->endgrp.fdp.enabled) {
5273         return NVME_FDP_DISABLED | NVME_DNR;
5274     }
5275 
5276     endgrp = &n->subsys->endgrp;
5277 
5278     trans_len = MIN(sizeof(log) - off, buf_len);
5279 
5280     /* spec value is 128 bit, we only use 64 bit */
5281     log.hbmw[0] = cpu_to_le64(endgrp->fdp.hbmw);
5282     log.mbmw[0] = cpu_to_le64(endgrp->fdp.mbmw);
5283     log.mbe[0] = cpu_to_le64(endgrp->fdp.mbe);
5284 
5285     return nvme_c2h(n, (uint8_t *)&log + off, trans_len, req);
5286 }
5287 
5288 static uint16_t nvme_fdp_events(NvmeCtrl *n, uint32_t endgrpid,
5289                                 uint32_t buf_len, uint64_t off,
5290                                 NvmeRequest *req)
5291 {
5292     NvmeEnduranceGroup *endgrp;
5293     NvmeCmd *cmd = &req->cmd;
5294     bool host_events = (cmd->cdw10 >> 8) & 0x1;
5295     uint32_t log_size, trans_len;
5296     NvmeFdpEventBuffer *ebuf;
5297     g_autofree NvmeFdpEventsLog *elog = NULL;
5298     NvmeFdpEvent *event;
5299 
5300     if (endgrpid != 1 || !n->subsys) {
5301         return NVME_INVALID_FIELD | NVME_DNR;
5302     }
5303 
5304     endgrp = &n->subsys->endgrp;
5305 
5306     if (!endgrp->fdp.enabled) {
5307         return NVME_FDP_DISABLED | NVME_DNR;
5308     }
5309 
5310     if (host_events) {
5311         ebuf = &endgrp->fdp.host_events;
5312     } else {
5313         ebuf = &endgrp->fdp.ctrl_events;
5314     }
5315 
5316     log_size = sizeof(NvmeFdpEventsLog) + ebuf->nelems * sizeof(NvmeFdpEvent);
5317 
5318     if (off >= log_size) {
5319         return NVME_INVALID_FIELD | NVME_DNR;
5320     }
5321 
5322     trans_len = MIN(log_size - off, buf_len);
5323     elog = g_malloc0(log_size);
5324     elog->num_events = cpu_to_le32(ebuf->nelems);
5325     event = (NvmeFdpEvent *)(elog + 1);
5326 
5327     if (ebuf->nelems && ebuf->start == ebuf->next) {
5328         unsigned int nelems = (NVME_FDP_MAX_EVENTS - ebuf->start);
5329         /* wrap over, copy [start;NVME_FDP_MAX_EVENTS[ and [0; next[ */
5330         memcpy(event, &ebuf->events[ebuf->start],
5331                sizeof(NvmeFdpEvent) * nelems);
5332         memcpy(event + nelems, ebuf->events,
5333                sizeof(NvmeFdpEvent) * ebuf->next);
5334     } else if (ebuf->start < ebuf->next) {
5335         memcpy(event, &ebuf->events[ebuf->start],
5336                sizeof(NvmeFdpEvent) * (ebuf->next - ebuf->start));
5337     }
5338 
5339     return nvme_c2h(n, (uint8_t *)elog + off, trans_len, req);
5340 }
5341 
5342 static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
5343 {
5344     NvmeCmd *cmd = &req->cmd;
5345 
5346     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5347     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5348     uint32_t dw12 = le32_to_cpu(cmd->cdw12);
5349     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
5350     uint8_t  lid = dw10 & 0xff;
5351     uint8_t  lsp = (dw10 >> 8) & 0xf;
5352     uint8_t  rae = (dw10 >> 15) & 0x1;
5353     uint8_t  csi = le32_to_cpu(cmd->cdw14) >> 24;
5354     uint32_t numdl, numdu, lspi;
5355     uint64_t off, lpol, lpou;
5356     size_t   len;
5357     uint16_t status;
5358 
5359     numdl = (dw10 >> 16);
5360     numdu = (dw11 & 0xffff);
5361     lspi = (dw11 >> 16);
5362     lpol = dw12;
5363     lpou = dw13;
5364 
5365     len = (((numdu << 16) | numdl) + 1) << 2;
5366     off = (lpou << 32ULL) | lpol;
5367 
5368     if (off & 0x3) {
5369         return NVME_INVALID_FIELD | NVME_DNR;
5370     }
5371 
5372     trace_pci_nvme_get_log(nvme_cid(req), lid, lsp, rae, len, off);
5373 
5374     status = nvme_check_mdts(n, len);
5375     if (status) {
5376         return status;
5377     }
5378 
5379     switch (lid) {
5380     case NVME_LOG_ERROR_INFO:
5381         return nvme_error_info(n, rae, len, off, req);
5382     case NVME_LOG_SMART_INFO:
5383         return nvme_smart_info(n, rae, len, off, req);
5384     case NVME_LOG_FW_SLOT_INFO:
5385         return nvme_fw_log_info(n, len, off, req);
5386     case NVME_LOG_CHANGED_NSLIST:
5387         return nvme_changed_nslist(n, rae, len, off, req);
5388     case NVME_LOG_CMD_EFFECTS:
5389         return nvme_cmd_effects(n, csi, len, off, req);
5390     case NVME_LOG_ENDGRP:
5391         return nvme_endgrp_info(n, rae, len, off, req);
5392     case NVME_LOG_FDP_CONFS:
5393         return nvme_fdp_confs(n, lspi, len, off, req);
5394     case NVME_LOG_FDP_RUH_USAGE:
5395         return nvme_fdp_ruh_usage(n, lspi, dw10, dw12, len, off, req);
5396     case NVME_LOG_FDP_STATS:
5397         return nvme_fdp_stats(n, lspi, len, off, req);
5398     case NVME_LOG_FDP_EVENTS:
5399         return nvme_fdp_events(n, lspi, len, off, req);
5400     default:
5401         trace_pci_nvme_err_invalid_log_page(nvme_cid(req), lid);
5402         return NVME_INVALID_FIELD | NVME_DNR;
5403     }
5404 }
5405 
5406 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
5407 {
5408     PCIDevice *pci = PCI_DEVICE(n);
5409     uint16_t offset = (cq->cqid << 3) + (1 << 2);
5410 
5411     n->cq[cq->cqid] = NULL;
5412     qemu_bh_delete(cq->bh);
5413     if (cq->ioeventfd_enabled) {
5414         memory_region_del_eventfd(&n->iomem,
5415                                   0x1000 + offset, 4, false, 0, &cq->notifier);
5416         event_notifier_set_handler(&cq->notifier, NULL);
5417         event_notifier_cleanup(&cq->notifier);
5418     }
5419     if (msix_enabled(pci)) {
5420         msix_vector_unuse(pci, cq->vector);
5421     }
5422     if (cq->cqid) {
5423         g_free(cq);
5424     }
5425 }
5426 
5427 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)
5428 {
5429     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
5430     NvmeCQueue *cq;
5431     uint16_t qid = le16_to_cpu(c->qid);
5432 
5433     if (unlikely(!qid || nvme_check_cqid(n, qid))) {
5434         trace_pci_nvme_err_invalid_del_cq_cqid(qid);
5435         return NVME_INVALID_CQID | NVME_DNR;
5436     }
5437 
5438     cq = n->cq[qid];
5439     if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
5440         trace_pci_nvme_err_invalid_del_cq_notempty(qid);
5441         return NVME_INVALID_QUEUE_DEL;
5442     }
5443 
5444     if (cq->irq_enabled && cq->tail != cq->head) {
5445         n->cq_pending--;
5446     }
5447 
5448     nvme_irq_deassert(n, cq);
5449     trace_pci_nvme_del_cq(qid);
5450     nvme_free_cq(cq, n);
5451     return NVME_SUCCESS;
5452 }
5453 
5454 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
5455                          uint16_t cqid, uint16_t vector, uint16_t size,
5456                          uint16_t irq_enabled)
5457 {
5458     PCIDevice *pci = PCI_DEVICE(n);
5459 
5460     if (msix_enabled(pci)) {
5461         msix_vector_use(pci, vector);
5462     }
5463     cq->ctrl = n;
5464     cq->cqid = cqid;
5465     cq->size = size;
5466     cq->dma_addr = dma_addr;
5467     cq->phase = 1;
5468     cq->irq_enabled = irq_enabled;
5469     cq->vector = vector;
5470     cq->head = cq->tail = 0;
5471     QTAILQ_INIT(&cq->req_list);
5472     QTAILQ_INIT(&cq->sq_list);
5473     if (n->dbbuf_enabled) {
5474         cq->db_addr = n->dbbuf_dbs + (cqid << 3) + (1 << 2);
5475         cq->ei_addr = n->dbbuf_eis + (cqid << 3) + (1 << 2);
5476 
5477         if (n->params.ioeventfd && cqid != 0) {
5478             if (!nvme_init_cq_ioeventfd(cq)) {
5479                 cq->ioeventfd_enabled = true;
5480             }
5481         }
5482     }
5483     n->cq[cqid] = cq;
5484     cq->bh = qemu_bh_new_guarded(nvme_post_cqes, cq,
5485                                  &DEVICE(cq->ctrl)->mem_reentrancy_guard);
5486 }
5487 
5488 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeRequest *req)
5489 {
5490     NvmeCQueue *cq;
5491     NvmeCreateCq *c = (NvmeCreateCq *)&req->cmd;
5492     uint16_t cqid = le16_to_cpu(c->cqid);
5493     uint16_t vector = le16_to_cpu(c->irq_vector);
5494     uint16_t qsize = le16_to_cpu(c->qsize);
5495     uint16_t qflags = le16_to_cpu(c->cq_flags);
5496     uint64_t prp1 = le64_to_cpu(c->prp1);
5497     uint32_t cc = ldq_le_p(&n->bar.cc);
5498     uint8_t iocqes = NVME_CC_IOCQES(cc);
5499     uint8_t iosqes = NVME_CC_IOSQES(cc);
5500 
5501     trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
5502                              NVME_CQ_FLAGS_IEN(qflags) != 0);
5503 
5504     if (iosqes != NVME_SQES || iocqes != NVME_CQES) {
5505         trace_pci_nvme_err_invalid_create_cq_entry_size(iosqes, iocqes);
5506         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5507     }
5508 
5509     if (unlikely(!cqid || cqid > n->conf_ioqpairs || n->cq[cqid] != NULL)) {
5510         trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
5511         return NVME_INVALID_QID | NVME_DNR;
5512     }
5513     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
5514         trace_pci_nvme_err_invalid_create_cq_size(qsize);
5515         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5516     }
5517     if (unlikely(prp1 & (n->page_size - 1))) {
5518         trace_pci_nvme_err_invalid_create_cq_addr(prp1);
5519         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
5520     }
5521     if (unlikely(!msix_enabled(PCI_DEVICE(n)) && vector)) {
5522         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5523         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5524     }
5525     if (unlikely(vector >= n->conf_msix_qsize)) {
5526         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5527         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5528     }
5529     if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
5530         trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
5531         return NVME_INVALID_FIELD | NVME_DNR;
5532     }
5533 
5534     cq = g_malloc0(sizeof(*cq));
5535     nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
5536                  NVME_CQ_FLAGS_IEN(qflags));
5537 
5538     /*
5539      * It is only required to set qs_created when creating a completion queue;
5540      * creating a submission queue without a matching completion queue will
5541      * fail.
5542      */
5543     n->qs_created = true;
5544     return NVME_SUCCESS;
5545 }
5546 
5547 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl *n, NvmeRequest *req)
5548 {
5549     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5550 
5551     return nvme_c2h(n, id, sizeof(id), req);
5552 }
5553 
5554 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeRequest *req)
5555 {
5556     trace_pci_nvme_identify_ctrl();
5557 
5558     return nvme_c2h(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl), req);
5559 }
5560 
5561 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl *n, NvmeRequest *req)
5562 {
5563     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5564     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5565     NvmeIdCtrlNvm *id_nvm = (NvmeIdCtrlNvm *)&id;
5566 
5567     trace_pci_nvme_identify_ctrl_csi(c->csi);
5568 
5569     switch (c->csi) {
5570     case NVME_CSI_NVM:
5571         id_nvm->vsl = n->params.vsl;
5572         id_nvm->dmrsl = cpu_to_le32(n->dmrsl);
5573         break;
5574 
5575     case NVME_CSI_ZONED:
5576         ((NvmeIdCtrlZoned *)&id)->zasl = n->params.zasl;
5577         break;
5578 
5579     default:
5580         return NVME_INVALID_FIELD | NVME_DNR;
5581     }
5582 
5583     return nvme_c2h(n, id, sizeof(id), req);
5584 }
5585 
5586 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeRequest *req, bool active)
5587 {
5588     NvmeNamespace *ns;
5589     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5590     uint32_t nsid = le32_to_cpu(c->nsid);
5591 
5592     trace_pci_nvme_identify_ns(nsid);
5593 
5594     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5595         return NVME_INVALID_NSID | NVME_DNR;
5596     }
5597 
5598     ns = nvme_ns(n, nsid);
5599     if (unlikely(!ns)) {
5600         if (!active) {
5601             ns = nvme_subsys_ns(n->subsys, nsid);
5602             if (!ns) {
5603                 return nvme_rpt_empty_id_struct(n, req);
5604             }
5605         } else {
5606             return nvme_rpt_empty_id_struct(n, req);
5607         }
5608     }
5609 
5610     if (active || ns->csi == NVME_CSI_NVM) {
5611         return nvme_c2h(n, (uint8_t *)&ns->id_ns, sizeof(NvmeIdNs), req);
5612     }
5613 
5614     return NVME_INVALID_CMD_SET | NVME_DNR;
5615 }
5616 
5617 static uint16_t nvme_identify_ctrl_list(NvmeCtrl *n, NvmeRequest *req,
5618                                         bool attached)
5619 {
5620     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5621     uint32_t nsid = le32_to_cpu(c->nsid);
5622     uint16_t min_id = le16_to_cpu(c->ctrlid);
5623     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5624     uint16_t *ids = &list[1];
5625     NvmeNamespace *ns;
5626     NvmeCtrl *ctrl;
5627     int cntlid, nr_ids = 0;
5628 
5629     trace_pci_nvme_identify_ctrl_list(c->cns, min_id);
5630 
5631     if (!n->subsys) {
5632         return NVME_INVALID_FIELD | NVME_DNR;
5633     }
5634 
5635     if (attached) {
5636         if (nsid == NVME_NSID_BROADCAST) {
5637             return NVME_INVALID_FIELD | NVME_DNR;
5638         }
5639 
5640         ns = nvme_subsys_ns(n->subsys, nsid);
5641         if (!ns) {
5642             return NVME_INVALID_FIELD | NVME_DNR;
5643         }
5644     }
5645 
5646     for (cntlid = min_id; cntlid < ARRAY_SIZE(n->subsys->ctrls); cntlid++) {
5647         ctrl = nvme_subsys_ctrl(n->subsys, cntlid);
5648         if (!ctrl) {
5649             continue;
5650         }
5651 
5652         if (attached && !nvme_ns(ctrl, nsid)) {
5653             continue;
5654         }
5655 
5656         ids[nr_ids++] = cntlid;
5657     }
5658 
5659     list[0] = nr_ids;
5660 
5661     return nvme_c2h(n, (uint8_t *)list, sizeof(list), req);
5662 }
5663 
5664 static uint16_t nvme_identify_pri_ctrl_cap(NvmeCtrl *n, NvmeRequest *req)
5665 {
5666     trace_pci_nvme_identify_pri_ctrl_cap(le16_to_cpu(n->pri_ctrl_cap.cntlid));
5667 
5668     return nvme_c2h(n, (uint8_t *)&n->pri_ctrl_cap,
5669                     sizeof(NvmePriCtrlCap), req);
5670 }
5671 
5672 static uint16_t nvme_identify_sec_ctrl_list(NvmeCtrl *n, NvmeRequest *req)
5673 {
5674     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5675     uint16_t pri_ctrl_id = le16_to_cpu(n->pri_ctrl_cap.cntlid);
5676     uint16_t min_id = le16_to_cpu(c->ctrlid);
5677     uint8_t num_sec_ctrl = n->nr_sec_ctrls;
5678     NvmeSecCtrlList list = {0};
5679     uint8_t i;
5680 
5681     for (i = 0; i < num_sec_ctrl; i++) {
5682         if (n->sec_ctrl_list[i].scid >= min_id) {
5683             list.numcntl = MIN(num_sec_ctrl - i, 127);
5684             memcpy(&list.sec, n->sec_ctrl_list + i,
5685                    list.numcntl * sizeof(NvmeSecCtrlEntry));
5686             break;
5687         }
5688     }
5689 
5690     trace_pci_nvme_identify_sec_ctrl_list(pri_ctrl_id, list.numcntl);
5691 
5692     return nvme_c2h(n, (uint8_t *)&list, sizeof(list), req);
5693 }
5694 
5695 static uint16_t nvme_identify_ns_csi(NvmeCtrl *n, NvmeRequest *req,
5696                                      bool active)
5697 {
5698     NvmeNamespace *ns;
5699     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5700     uint32_t nsid = le32_to_cpu(c->nsid);
5701 
5702     trace_pci_nvme_identify_ns_csi(nsid, c->csi);
5703 
5704     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5705         return NVME_INVALID_NSID | NVME_DNR;
5706     }
5707 
5708     ns = nvme_ns(n, nsid);
5709     if (unlikely(!ns)) {
5710         if (!active) {
5711             ns = nvme_subsys_ns(n->subsys, nsid);
5712             if (!ns) {
5713                 return nvme_rpt_empty_id_struct(n, req);
5714             }
5715         } else {
5716             return nvme_rpt_empty_id_struct(n, req);
5717         }
5718     }
5719 
5720     if (c->csi == NVME_CSI_NVM) {
5721         return nvme_c2h(n, (uint8_t *)&ns->id_ns_nvm, sizeof(NvmeIdNsNvm),
5722                         req);
5723     } else if (c->csi == NVME_CSI_ZONED && ns->csi == NVME_CSI_ZONED) {
5724         return nvme_c2h(n, (uint8_t *)ns->id_ns_zoned, sizeof(NvmeIdNsZoned),
5725                         req);
5726     }
5727 
5728     return NVME_INVALID_FIELD | NVME_DNR;
5729 }
5730 
5731 static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeRequest *req,
5732                                      bool active)
5733 {
5734     NvmeNamespace *ns;
5735     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5736     uint32_t min_nsid = le32_to_cpu(c->nsid);
5737     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5738     static const int data_len = sizeof(list);
5739     uint32_t *list_ptr = (uint32_t *)list;
5740     int i, j = 0;
5741 
5742     trace_pci_nvme_identify_nslist(min_nsid);
5743 
5744     /*
5745      * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
5746      * since the Active Namespace ID List should return namespaces with ids
5747      * *higher* than the NSID specified in the command. This is also specified
5748      * in the spec (NVM Express v1.3d, Section 5.15.4).
5749      */
5750     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5751         return NVME_INVALID_NSID | NVME_DNR;
5752     }
5753 
5754     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5755         ns = nvme_ns(n, i);
5756         if (!ns) {
5757             if (!active) {
5758                 ns = nvme_subsys_ns(n->subsys, i);
5759                 if (!ns) {
5760                     continue;
5761                 }
5762             } else {
5763                 continue;
5764             }
5765         }
5766         if (ns->params.nsid <= min_nsid) {
5767             continue;
5768         }
5769         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5770         if (j == data_len / sizeof(uint32_t)) {
5771             break;
5772         }
5773     }
5774 
5775     return nvme_c2h(n, list, data_len, req);
5776 }
5777 
5778 static uint16_t nvme_identify_nslist_csi(NvmeCtrl *n, NvmeRequest *req,
5779                                          bool active)
5780 {
5781     NvmeNamespace *ns;
5782     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5783     uint32_t min_nsid = le32_to_cpu(c->nsid);
5784     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5785     static const int data_len = sizeof(list);
5786     uint32_t *list_ptr = (uint32_t *)list;
5787     int i, j = 0;
5788 
5789     trace_pci_nvme_identify_nslist_csi(min_nsid, c->csi);
5790 
5791     /*
5792      * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
5793      */
5794     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5795         return NVME_INVALID_NSID | NVME_DNR;
5796     }
5797 
5798     if (c->csi != NVME_CSI_NVM && c->csi != NVME_CSI_ZONED) {
5799         return NVME_INVALID_FIELD | NVME_DNR;
5800     }
5801 
5802     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5803         ns = nvme_ns(n, i);
5804         if (!ns) {
5805             if (!active) {
5806                 ns = nvme_subsys_ns(n->subsys, i);
5807                 if (!ns) {
5808                     continue;
5809                 }
5810             } else {
5811                 continue;
5812             }
5813         }
5814         if (ns->params.nsid <= min_nsid || c->csi != ns->csi) {
5815             continue;
5816         }
5817         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5818         if (j == data_len / sizeof(uint32_t)) {
5819             break;
5820         }
5821     }
5822 
5823     return nvme_c2h(n, list, data_len, req);
5824 }
5825 
5826 static uint16_t nvme_endurance_group_list(NvmeCtrl *n, NvmeRequest *req)
5827 {
5828     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5829     uint16_t *nr_ids = &list[0];
5830     uint16_t *ids = &list[1];
5831     uint16_t endgid = le32_to_cpu(req->cmd.cdw11) & 0xffff;
5832 
5833     /*
5834      * The current nvme-subsys only supports Endurance Group #1.
5835      */
5836     if (!endgid) {
5837         *nr_ids = 1;
5838         ids[0] = 1;
5839     } else {
5840         *nr_ids = 0;
5841     }
5842 
5843     return nvme_c2h(n, list, sizeof(list), req);
5844 }
5845 
5846 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl *n, NvmeRequest *req)
5847 {
5848     NvmeNamespace *ns;
5849     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5850     uint32_t nsid = le32_to_cpu(c->nsid);
5851     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5852     uint8_t *pos = list;
5853     struct {
5854         NvmeIdNsDescr hdr;
5855         uint8_t v[NVME_NIDL_UUID];
5856     } QEMU_PACKED uuid = {};
5857     struct {
5858         NvmeIdNsDescr hdr;
5859         uint8_t v[NVME_NIDL_NGUID];
5860     } QEMU_PACKED nguid = {};
5861     struct {
5862         NvmeIdNsDescr hdr;
5863         uint64_t v;
5864     } QEMU_PACKED eui64 = {};
5865     struct {
5866         NvmeIdNsDescr hdr;
5867         uint8_t v;
5868     } QEMU_PACKED csi = {};
5869 
5870     trace_pci_nvme_identify_ns_descr_list(nsid);
5871 
5872     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5873         return NVME_INVALID_NSID | NVME_DNR;
5874     }
5875 
5876     ns = nvme_ns(n, nsid);
5877     if (unlikely(!ns)) {
5878         return NVME_INVALID_FIELD | NVME_DNR;
5879     }
5880 
5881     if (!qemu_uuid_is_null(&ns->params.uuid)) {
5882         uuid.hdr.nidt = NVME_NIDT_UUID;
5883         uuid.hdr.nidl = NVME_NIDL_UUID;
5884         memcpy(uuid.v, ns->params.uuid.data, NVME_NIDL_UUID);
5885         memcpy(pos, &uuid, sizeof(uuid));
5886         pos += sizeof(uuid);
5887     }
5888 
5889     if (!nvme_nguid_is_null(&ns->params.nguid)) {
5890         nguid.hdr.nidt = NVME_NIDT_NGUID;
5891         nguid.hdr.nidl = NVME_NIDL_NGUID;
5892         memcpy(nguid.v, ns->params.nguid.data, NVME_NIDL_NGUID);
5893         memcpy(pos, &nguid, sizeof(nguid));
5894         pos += sizeof(nguid);
5895     }
5896 
5897     if (ns->params.eui64) {
5898         eui64.hdr.nidt = NVME_NIDT_EUI64;
5899         eui64.hdr.nidl = NVME_NIDL_EUI64;
5900         eui64.v = cpu_to_be64(ns->params.eui64);
5901         memcpy(pos, &eui64, sizeof(eui64));
5902         pos += sizeof(eui64);
5903     }
5904 
5905     csi.hdr.nidt = NVME_NIDT_CSI;
5906     csi.hdr.nidl = NVME_NIDL_CSI;
5907     csi.v = ns->csi;
5908     memcpy(pos, &csi, sizeof(csi));
5909     pos += sizeof(csi);
5910 
5911     return nvme_c2h(n, list, sizeof(list), req);
5912 }
5913 
5914 static uint16_t nvme_identify_cmd_set(NvmeCtrl *n, NvmeRequest *req)
5915 {
5916     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5917     static const int data_len = sizeof(list);
5918 
5919     trace_pci_nvme_identify_cmd_set();
5920 
5921     NVME_SET_CSI(*list, NVME_CSI_NVM);
5922     NVME_SET_CSI(*list, NVME_CSI_ZONED);
5923 
5924     return nvme_c2h(n, list, data_len, req);
5925 }
5926 
5927 static uint16_t nvme_identify(NvmeCtrl *n, NvmeRequest *req)
5928 {
5929     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5930 
5931     trace_pci_nvme_identify(nvme_cid(req), c->cns, le16_to_cpu(c->ctrlid),
5932                             c->csi);
5933 
5934     switch (c->cns) {
5935     case NVME_ID_CNS_NS:
5936         return nvme_identify_ns(n, req, true);
5937     case NVME_ID_CNS_NS_PRESENT:
5938         return nvme_identify_ns(n, req, false);
5939     case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST:
5940         return nvme_identify_ctrl_list(n, req, true);
5941     case NVME_ID_CNS_CTRL_LIST:
5942         return nvme_identify_ctrl_list(n, req, false);
5943     case NVME_ID_CNS_PRIMARY_CTRL_CAP:
5944         return nvme_identify_pri_ctrl_cap(n, req);
5945     case NVME_ID_CNS_SECONDARY_CTRL_LIST:
5946         return nvme_identify_sec_ctrl_list(n, req);
5947     case NVME_ID_CNS_CS_NS:
5948         return nvme_identify_ns_csi(n, req, true);
5949     case NVME_ID_CNS_CS_NS_PRESENT:
5950         return nvme_identify_ns_csi(n, req, false);
5951     case NVME_ID_CNS_CTRL:
5952         return nvme_identify_ctrl(n, req);
5953     case NVME_ID_CNS_CS_CTRL:
5954         return nvme_identify_ctrl_csi(n, req);
5955     case NVME_ID_CNS_NS_ACTIVE_LIST:
5956         return nvme_identify_nslist(n, req, true);
5957     case NVME_ID_CNS_NS_PRESENT_LIST:
5958         return nvme_identify_nslist(n, req, false);
5959     case NVME_ID_CNS_CS_NS_ACTIVE_LIST:
5960         return nvme_identify_nslist_csi(n, req, true);
5961     case NVME_ID_CNS_ENDURANCE_GROUP_LIST:
5962         return nvme_endurance_group_list(n, req);
5963     case NVME_ID_CNS_CS_NS_PRESENT_LIST:
5964         return nvme_identify_nslist_csi(n, req, false);
5965     case NVME_ID_CNS_NS_DESCR_LIST:
5966         return nvme_identify_ns_descr_list(n, req);
5967     case NVME_ID_CNS_IO_COMMAND_SET:
5968         return nvme_identify_cmd_set(n, req);
5969     default:
5970         trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
5971         return NVME_INVALID_FIELD | NVME_DNR;
5972     }
5973 }
5974 
5975 static uint16_t nvme_abort(NvmeCtrl *n, NvmeRequest *req)
5976 {
5977     uint16_t sqid = le32_to_cpu(req->cmd.cdw10) & 0xffff;
5978     uint16_t cid  = (le32_to_cpu(req->cmd.cdw10) >> 16) & 0xffff;
5979     NvmeSQueue *sq = n->sq[sqid];
5980     NvmeRequest *r, *next;
5981     int i;
5982 
5983     req->cqe.result = 1;
5984     if (nvme_check_sqid(n, sqid)) {
5985         return NVME_INVALID_FIELD | NVME_DNR;
5986     }
5987 
5988     if (sqid == 0) {
5989         for (i = 0; i < n->outstanding_aers; i++) {
5990             NvmeRequest *re = n->aer_reqs[i];
5991             if (re->cqe.cid == cid) {
5992                 memmove(n->aer_reqs + i, n->aer_reqs + i + 1,
5993                          (n->outstanding_aers - i - 1) * sizeof(NvmeRequest *));
5994                 n->outstanding_aers--;
5995                 re->status = NVME_CMD_ABORT_REQ;
5996                 req->cqe.result = 0;
5997                 nvme_enqueue_req_completion(&n->admin_cq, re);
5998                 return NVME_SUCCESS;
5999             }
6000         }
6001     }
6002 
6003     QTAILQ_FOREACH_SAFE(r, &sq->out_req_list, entry, next) {
6004         if (r->cqe.cid == cid) {
6005             if (r->aiocb) {
6006                 blk_aio_cancel_async(r->aiocb);
6007             }
6008             break;
6009         }
6010     }
6011 
6012     return NVME_SUCCESS;
6013 }
6014 
6015 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
6016 {
6017     trace_pci_nvme_setfeat_timestamp(ts);
6018 
6019     n->host_timestamp = le64_to_cpu(ts);
6020     n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
6021 }
6022 
6023 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
6024 {
6025     uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
6026     uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
6027 
6028     union nvme_timestamp {
6029         struct {
6030             uint64_t timestamp:48;
6031             uint64_t sync:1;
6032             uint64_t origin:3;
6033             uint64_t rsvd1:12;
6034         };
6035         uint64_t all;
6036     };
6037 
6038     union nvme_timestamp ts;
6039     ts.all = 0;
6040     ts.timestamp = n->host_timestamp + elapsed_time;
6041 
6042     /* If the host timestamp is non-zero, set the timestamp origin */
6043     ts.origin = n->host_timestamp ? 0x01 : 0x00;
6044 
6045     trace_pci_nvme_getfeat_timestamp(ts.all);
6046 
6047     return cpu_to_le64(ts.all);
6048 }
6049 
6050 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
6051 {
6052     uint64_t timestamp = nvme_get_timestamp(n);
6053 
6054     return nvme_c2h(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
6055 }
6056 
6057 static int nvme_get_feature_fdp(NvmeCtrl *n, uint32_t endgrpid,
6058                                 uint32_t *result)
6059 {
6060     *result = 0;
6061 
6062     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6063         return NVME_INVALID_FIELD | NVME_DNR;
6064     }
6065 
6066     *result = FIELD_DP16(0, FEAT_FDP, FDPE, 1);
6067     *result = FIELD_DP16(*result, FEAT_FDP, CONF_NDX, 0);
6068 
6069     return NVME_SUCCESS;
6070 }
6071 
6072 static uint16_t nvme_get_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
6073                                             NvmeRequest *req, uint32_t *result)
6074 {
6075     NvmeCmd *cmd = &req->cmd;
6076     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
6077     uint16_t ph = cdw11 & 0xffff;
6078     uint8_t noet = (cdw11 >> 16) & 0xff;
6079     uint16_t ruhid, ret;
6080     uint32_t nentries = 0;
6081     uint8_t s_events_ndx = 0;
6082     size_t s_events_siz = sizeof(NvmeFdpEventDescr) * noet;
6083     g_autofree NvmeFdpEventDescr *s_events = g_malloc0(s_events_siz);
6084     NvmeRuHandle *ruh;
6085     NvmeFdpEventDescr *s_event;
6086 
6087     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6088         return NVME_FDP_DISABLED | NVME_DNR;
6089     }
6090 
6091     if (!nvme_ph_valid(ns, ph)) {
6092         return NVME_INVALID_FIELD | NVME_DNR;
6093     }
6094 
6095     ruhid = ns->fdp.phs[ph];
6096     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
6097 
6098     assert(ruh);
6099 
6100     if (unlikely(noet == 0)) {
6101         return NVME_INVALID_FIELD | NVME_DNR;
6102     }
6103 
6104     for (uint8_t event_type = 0; event_type < FDP_EVT_MAX; event_type++) {
6105         uint8_t shift = nvme_fdp_evf_shifts[event_type];
6106         if (!shift && event_type) {
6107             /*
6108              * only first entry (event_type == 0) has a shift value of 0
6109              * other entries are simply unpopulated.
6110              */
6111             continue;
6112         }
6113 
6114         nentries++;
6115 
6116         s_event = &s_events[s_events_ndx];
6117         s_event->evt = event_type;
6118         s_event->evta = (ruh->event_filter >> shift) & 0x1;
6119 
6120         /* break if all `noet` entries are filled */
6121         if ((++s_events_ndx) == noet) {
6122             break;
6123         }
6124     }
6125 
6126     ret = nvme_c2h(n, s_events, s_events_siz, req);
6127     if (ret) {
6128         return ret;
6129     }
6130 
6131     *result = nentries;
6132     return NVME_SUCCESS;
6133 }
6134 
6135 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeRequest *req)
6136 {
6137     NvmeCmd *cmd = &req->cmd;
6138     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
6139     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
6140     uint32_t nsid = le32_to_cpu(cmd->nsid);
6141     uint32_t result = 0;
6142     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
6143     NvmeGetFeatureSelect sel = NVME_GETFEAT_SELECT(dw10);
6144     uint16_t iv;
6145     NvmeNamespace *ns;
6146     int i;
6147     uint16_t endgrpid = 0, ret = NVME_SUCCESS;
6148 
6149     static const uint32_t nvme_feature_default[NVME_FID_MAX] = {
6150         [NVME_ARBITRATION] = NVME_ARB_AB_NOLIMIT,
6151     };
6152 
6153     trace_pci_nvme_getfeat(nvme_cid(req), nsid, fid, sel, dw11);
6154 
6155     if (!nvme_feature_support[fid]) {
6156         return NVME_INVALID_FIELD | NVME_DNR;
6157     }
6158 
6159     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
6160         if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
6161             /*
6162              * The Reservation Notification Mask and Reservation Persistence
6163              * features require a status code of Invalid Field in Command when
6164              * NSID is FFFFFFFFh. Since the device does not support those
6165              * features we can always return Invalid Namespace or Format as we
6166              * should do for all other features.
6167              */
6168             return NVME_INVALID_NSID | NVME_DNR;
6169         }
6170 
6171         if (!nvme_ns(n, nsid)) {
6172             return NVME_INVALID_FIELD | NVME_DNR;
6173         }
6174     }
6175 
6176     switch (sel) {
6177     case NVME_GETFEAT_SELECT_CURRENT:
6178         break;
6179     case NVME_GETFEAT_SELECT_SAVED:
6180         /* no features are saveable by the controller; fallthrough */
6181     case NVME_GETFEAT_SELECT_DEFAULT:
6182         goto defaults;
6183     case NVME_GETFEAT_SELECT_CAP:
6184         result = nvme_feature_cap[fid];
6185         goto out;
6186     }
6187 
6188     switch (fid) {
6189     case NVME_TEMPERATURE_THRESHOLD:
6190         result = 0;
6191 
6192         /*
6193          * The controller only implements the Composite Temperature sensor, so
6194          * return 0 for all other sensors.
6195          */
6196         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6197             goto out;
6198         }
6199 
6200         switch (NVME_TEMP_THSEL(dw11)) {
6201         case NVME_TEMP_THSEL_OVER:
6202             result = n->features.temp_thresh_hi;
6203             goto out;
6204         case NVME_TEMP_THSEL_UNDER:
6205             result = n->features.temp_thresh_low;
6206             goto out;
6207         }
6208 
6209         return NVME_INVALID_FIELD | NVME_DNR;
6210     case NVME_ERROR_RECOVERY:
6211         if (!nvme_nsid_valid(n, nsid)) {
6212             return NVME_INVALID_NSID | NVME_DNR;
6213         }
6214 
6215         ns = nvme_ns(n, nsid);
6216         if (unlikely(!ns)) {
6217             return NVME_INVALID_FIELD | NVME_DNR;
6218         }
6219 
6220         result = ns->features.err_rec;
6221         goto out;
6222     case NVME_VOLATILE_WRITE_CACHE:
6223         result = 0;
6224         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6225             ns = nvme_ns(n, i);
6226             if (!ns) {
6227                 continue;
6228             }
6229 
6230             result = blk_enable_write_cache(ns->blkconf.blk);
6231             if (result) {
6232                 break;
6233             }
6234         }
6235         trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
6236         goto out;
6237     case NVME_ASYNCHRONOUS_EVENT_CONF:
6238         result = n->features.async_config;
6239         goto out;
6240     case NVME_TIMESTAMP:
6241         return nvme_get_feature_timestamp(n, req);
6242     case NVME_HOST_BEHAVIOR_SUPPORT:
6243         return nvme_c2h(n, (uint8_t *)&n->features.hbs,
6244                         sizeof(n->features.hbs), req);
6245     case NVME_FDP_MODE:
6246         endgrpid = dw11 & 0xff;
6247 
6248         if (endgrpid != 0x1) {
6249             return NVME_INVALID_FIELD | NVME_DNR;
6250         }
6251 
6252         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6253         if (ret) {
6254             return ret;
6255         }
6256         goto out;
6257     case NVME_FDP_EVENTS:
6258         if (!nvme_nsid_valid(n, nsid)) {
6259             return NVME_INVALID_NSID | NVME_DNR;
6260         }
6261 
6262         ns = nvme_ns(n, nsid);
6263         if (unlikely(!ns)) {
6264             return NVME_INVALID_FIELD | NVME_DNR;
6265         }
6266 
6267         ret = nvme_get_feature_fdp_events(n, ns, req, &result);
6268         if (ret) {
6269             return ret;
6270         }
6271         goto out;
6272     default:
6273         break;
6274     }
6275 
6276 defaults:
6277     switch (fid) {
6278     case NVME_TEMPERATURE_THRESHOLD:
6279         result = 0;
6280 
6281         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6282             break;
6283         }
6284 
6285         if (NVME_TEMP_THSEL(dw11) == NVME_TEMP_THSEL_OVER) {
6286             result = NVME_TEMPERATURE_WARNING;
6287         }
6288 
6289         break;
6290     case NVME_NUMBER_OF_QUEUES:
6291         result = (n->conf_ioqpairs - 1) | ((n->conf_ioqpairs - 1) << 16);
6292         trace_pci_nvme_getfeat_numq(result);
6293         break;
6294     case NVME_INTERRUPT_VECTOR_CONF:
6295         iv = dw11 & 0xffff;
6296         if (iv >= n->conf_ioqpairs + 1) {
6297             return NVME_INVALID_FIELD | NVME_DNR;
6298         }
6299 
6300         result = iv;
6301         if (iv == n->admin_cq.vector) {
6302             result |= NVME_INTVC_NOCOALESCING;
6303         }
6304         break;
6305     case NVME_FDP_MODE:
6306         endgrpid = dw11 & 0xff;
6307 
6308         if (endgrpid != 0x1) {
6309             return NVME_INVALID_FIELD | NVME_DNR;
6310         }
6311 
6312         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6313         if (ret) {
6314             return ret;
6315         }
6316         break;
6317 
6318     case NVME_WRITE_ATOMICITY:
6319         result = n->dn;
6320         break;
6321     default:
6322         result = nvme_feature_default[fid];
6323         break;
6324     }
6325 
6326 out:
6327     req->cqe.result = cpu_to_le32(result);
6328     return ret;
6329 }
6330 
6331 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
6332 {
6333     uint16_t ret;
6334     uint64_t timestamp;
6335 
6336     ret = nvme_h2c(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
6337     if (ret) {
6338         return ret;
6339     }
6340 
6341     nvme_set_timestamp(n, timestamp);
6342 
6343     return NVME_SUCCESS;
6344 }
6345 
6346 static uint16_t nvme_set_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
6347                                             NvmeRequest *req)
6348 {
6349     NvmeCmd *cmd = &req->cmd;
6350     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
6351     uint16_t ph = cdw11 & 0xffff;
6352     uint8_t noet = (cdw11 >> 16) & 0xff;
6353     uint16_t ret, ruhid;
6354     uint8_t enable = le32_to_cpu(cmd->cdw12) & 0x1;
6355     uint8_t event_mask = 0;
6356     unsigned int i;
6357     g_autofree uint8_t *events = g_malloc0(noet);
6358     NvmeRuHandle *ruh = NULL;
6359 
6360     assert(ns);
6361 
6362     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6363         return NVME_FDP_DISABLED | NVME_DNR;
6364     }
6365 
6366     if (!nvme_ph_valid(ns, ph)) {
6367         return NVME_INVALID_FIELD | NVME_DNR;
6368     }
6369 
6370     ruhid = ns->fdp.phs[ph];
6371     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
6372 
6373     ret = nvme_h2c(n, events, noet, req);
6374     if (ret) {
6375         return ret;
6376     }
6377 
6378     for (i = 0; i < noet; i++) {
6379         event_mask |= (1 << nvme_fdp_evf_shifts[events[i]]);
6380     }
6381 
6382     if (enable) {
6383         ruh->event_filter |= event_mask;
6384     } else {
6385         ruh->event_filter = ruh->event_filter & ~event_mask;
6386     }
6387 
6388     return NVME_SUCCESS;
6389 }
6390 
6391 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
6392 {
6393     NvmeNamespace *ns = NULL;
6394 
6395     NvmeCmd *cmd = &req->cmd;
6396     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
6397     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
6398     uint32_t nsid = le32_to_cpu(cmd->nsid);
6399     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
6400     uint8_t save = NVME_SETFEAT_SAVE(dw10);
6401     uint16_t status;
6402     int i;
6403     NvmeIdCtrl *id = &n->id_ctrl;
6404     NvmeAtomic *atomic = &n->atomic;
6405 
6406     trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
6407 
6408     if (save && !(nvme_feature_cap[fid] & NVME_FEAT_CAP_SAVE)) {
6409         return NVME_FID_NOT_SAVEABLE | NVME_DNR;
6410     }
6411 
6412     if (!nvme_feature_support[fid]) {
6413         return NVME_INVALID_FIELD | NVME_DNR;
6414     }
6415 
6416     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
6417         if (nsid != NVME_NSID_BROADCAST) {
6418             if (!nvme_nsid_valid(n, nsid)) {
6419                 return NVME_INVALID_NSID | NVME_DNR;
6420             }
6421 
6422             ns = nvme_ns(n, nsid);
6423             if (unlikely(!ns)) {
6424                 return NVME_INVALID_FIELD | NVME_DNR;
6425             }
6426         }
6427     } else if (nsid && nsid != NVME_NSID_BROADCAST) {
6428         if (!nvme_nsid_valid(n, nsid)) {
6429             return NVME_INVALID_NSID | NVME_DNR;
6430         }
6431 
6432         return NVME_FEAT_NOT_NS_SPEC | NVME_DNR;
6433     }
6434 
6435     if (!(nvme_feature_cap[fid] & NVME_FEAT_CAP_CHANGE)) {
6436         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6437     }
6438 
6439     switch (fid) {
6440     case NVME_TEMPERATURE_THRESHOLD:
6441         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6442             break;
6443         }
6444 
6445         switch (NVME_TEMP_THSEL(dw11)) {
6446         case NVME_TEMP_THSEL_OVER:
6447             n->features.temp_thresh_hi = NVME_TEMP_TMPTH(dw11);
6448             break;
6449         case NVME_TEMP_THSEL_UNDER:
6450             n->features.temp_thresh_low = NVME_TEMP_TMPTH(dw11);
6451             break;
6452         default:
6453             return NVME_INVALID_FIELD | NVME_DNR;
6454         }
6455 
6456         if ((n->temperature >= n->features.temp_thresh_hi) ||
6457             (n->temperature <= n->features.temp_thresh_low)) {
6458             nvme_smart_event(n, NVME_SMART_TEMPERATURE);
6459         }
6460 
6461         break;
6462     case NVME_ERROR_RECOVERY:
6463         if (nsid == NVME_NSID_BROADCAST) {
6464             for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6465                 ns = nvme_ns(n, i);
6466 
6467                 if (!ns) {
6468                     continue;
6469                 }
6470 
6471                 if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat)) {
6472                     ns->features.err_rec = dw11;
6473                 }
6474             }
6475 
6476             break;
6477         }
6478 
6479         assert(ns);
6480         if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat))  {
6481             ns->features.err_rec = dw11;
6482         }
6483         break;
6484     case NVME_VOLATILE_WRITE_CACHE:
6485         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6486             ns = nvme_ns(n, i);
6487             if (!ns) {
6488                 continue;
6489             }
6490 
6491             if (!(dw11 & 0x1) && blk_enable_write_cache(ns->blkconf.blk)) {
6492                 blk_flush(ns->blkconf.blk);
6493             }
6494 
6495             blk_set_enable_write_cache(ns->blkconf.blk, dw11 & 1);
6496         }
6497 
6498         break;
6499 
6500     case NVME_NUMBER_OF_QUEUES:
6501         if (n->qs_created) {
6502             return NVME_CMD_SEQ_ERROR | NVME_DNR;
6503         }
6504 
6505         /*
6506          * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
6507          * and NSQR.
6508          */
6509         if ((dw11 & 0xffff) == 0xffff || ((dw11 >> 16) & 0xffff) == 0xffff) {
6510             return NVME_INVALID_FIELD | NVME_DNR;
6511         }
6512 
6513         trace_pci_nvme_setfeat_numq((dw11 & 0xffff) + 1,
6514                                     ((dw11 >> 16) & 0xffff) + 1,
6515                                     n->conf_ioqpairs,
6516                                     n->conf_ioqpairs);
6517         req->cqe.result = cpu_to_le32((n->conf_ioqpairs - 1) |
6518                                       ((n->conf_ioqpairs - 1) << 16));
6519         break;
6520     case NVME_ASYNCHRONOUS_EVENT_CONF:
6521         n->features.async_config = dw11;
6522         break;
6523     case NVME_TIMESTAMP:
6524         return nvme_set_feature_timestamp(n, req);
6525     case NVME_HOST_BEHAVIOR_SUPPORT:
6526         status = nvme_h2c(n, (uint8_t *)&n->features.hbs,
6527                           sizeof(n->features.hbs), req);
6528         if (status) {
6529             return status;
6530         }
6531 
6532         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6533             ns = nvme_ns(n, i);
6534 
6535             if (!ns) {
6536                 continue;
6537             }
6538 
6539             ns->id_ns.nlbaf = ns->nlbaf - 1;
6540             if (!n->features.hbs.lbafee) {
6541                 ns->id_ns.nlbaf = MIN(ns->id_ns.nlbaf, 15);
6542             }
6543         }
6544 
6545         return status;
6546     case NVME_COMMAND_SET_PROFILE:
6547         if (dw11 & 0x1ff) {
6548             trace_pci_nvme_err_invalid_iocsci(dw11 & 0x1ff);
6549             return NVME_CMD_SET_CMB_REJECTED | NVME_DNR;
6550         }
6551         break;
6552     case NVME_FDP_MODE:
6553         /* spec: abort with cmd seq err if there's one or more NS' in endgrp */
6554         return NVME_CMD_SEQ_ERROR | NVME_DNR;
6555     case NVME_FDP_EVENTS:
6556         return nvme_set_feature_fdp_events(n, ns, req);
6557     case NVME_WRITE_ATOMICITY:
6558 
6559         n->dn = 0x1 & dw11;
6560 
6561         if (n->dn) {
6562             atomic->atomic_max_write_size = le16_to_cpu(id->awupf) + 1;
6563         } else {
6564             atomic->atomic_max_write_size = le16_to_cpu(id->awun) + 1;
6565         }
6566 
6567         if (atomic->atomic_max_write_size == 1) {
6568             atomic->atomic_writes = 0;
6569         } else {
6570             atomic->atomic_writes = 1;
6571         }
6572         break;
6573     default:
6574         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6575     }
6576     return NVME_SUCCESS;
6577 }
6578 
6579 static uint16_t nvme_aer(NvmeCtrl *n, NvmeRequest *req)
6580 {
6581     trace_pci_nvme_aer(nvme_cid(req));
6582 
6583     if (n->outstanding_aers > n->params.aerl) {
6584         trace_pci_nvme_aer_aerl_exceeded();
6585         return NVME_AER_LIMIT_EXCEEDED;
6586     }
6587 
6588     n->aer_reqs[n->outstanding_aers] = req;
6589     n->outstanding_aers++;
6590 
6591     if (!QTAILQ_EMPTY(&n->aer_queue)) {
6592         nvme_process_aers(n);
6593     }
6594 
6595     return NVME_NO_COMPLETE;
6596 }
6597 
6598 static void nvme_update_dmrsl(NvmeCtrl *n)
6599 {
6600     int nsid;
6601 
6602     for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
6603         NvmeNamespace *ns = nvme_ns(n, nsid);
6604         if (!ns) {
6605             continue;
6606         }
6607 
6608         n->dmrsl = MIN_NON_ZERO(n->dmrsl,
6609                                 BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
6610     }
6611 }
6612 
6613 static void nvme_select_iocs_ns(NvmeCtrl *n, NvmeNamespace *ns)
6614 {
6615     uint32_t cc = ldl_le_p(&n->bar.cc);
6616 
6617     ns->iocs = nvme_cse_iocs_none;
6618     switch (ns->csi) {
6619     case NVME_CSI_NVM:
6620         if (NVME_CC_CSS(cc) != NVME_CC_CSS_ADMIN_ONLY) {
6621             ns->iocs = nvme_cse_iocs_nvm;
6622         }
6623         break;
6624     case NVME_CSI_ZONED:
6625         if (NVME_CC_CSS(cc) == NVME_CC_CSS_CSI) {
6626             ns->iocs = nvme_cse_iocs_zoned;
6627         } else if (NVME_CC_CSS(cc) == NVME_CC_CSS_NVM) {
6628             ns->iocs = nvme_cse_iocs_nvm;
6629         }
6630         break;
6631     }
6632 }
6633 
6634 static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
6635 {
6636     NvmeNamespace *ns;
6637     NvmeCtrl *ctrl;
6638     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
6639     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6640     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6641     uint8_t sel = dw10 & 0xf;
6642     uint16_t *nr_ids = &list[0];
6643     uint16_t *ids = &list[1];
6644     uint16_t ret;
6645     int i;
6646 
6647     trace_pci_nvme_ns_attachment(nvme_cid(req), dw10 & 0xf);
6648 
6649     if (!nvme_nsid_valid(n, nsid)) {
6650         return NVME_INVALID_NSID | NVME_DNR;
6651     }
6652 
6653     ns = nvme_subsys_ns(n->subsys, nsid);
6654     if (!ns) {
6655         return NVME_INVALID_FIELD | NVME_DNR;
6656     }
6657 
6658     ret = nvme_h2c(n, (uint8_t *)list, 4096, req);
6659     if (ret) {
6660         return ret;
6661     }
6662 
6663     if (!*nr_ids) {
6664         return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6665     }
6666 
6667     *nr_ids = MIN(*nr_ids, NVME_CONTROLLER_LIST_SIZE - 1);
6668     for (i = 0; i < *nr_ids; i++) {
6669         ctrl = nvme_subsys_ctrl(n->subsys, ids[i]);
6670         if (!ctrl) {
6671             return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6672         }
6673 
6674         switch (sel) {
6675         case NVME_NS_ATTACHMENT_ATTACH:
6676             if (nvme_ns(ctrl, nsid)) {
6677                 return NVME_NS_ALREADY_ATTACHED | NVME_DNR;
6678             }
6679 
6680             if (ns->attached && !ns->params.shared) {
6681                 return NVME_NS_PRIVATE | NVME_DNR;
6682             }
6683 
6684             nvme_attach_ns(ctrl, ns);
6685             nvme_select_iocs_ns(ctrl, ns);
6686 
6687             break;
6688 
6689         case NVME_NS_ATTACHMENT_DETACH:
6690             if (!nvme_ns(ctrl, nsid)) {
6691                 return NVME_NS_NOT_ATTACHED | NVME_DNR;
6692             }
6693 
6694             ctrl->namespaces[nsid] = NULL;
6695             ns->attached--;
6696 
6697             nvme_update_dmrsl(ctrl);
6698 
6699             break;
6700 
6701         default:
6702             return NVME_INVALID_FIELD | NVME_DNR;
6703         }
6704 
6705         /*
6706          * Add namespace id to the changed namespace id list for event clearing
6707          * via Get Log Page command.
6708          */
6709         if (!test_and_set_bit(nsid, ctrl->changed_nsids)) {
6710             nvme_enqueue_event(ctrl, NVME_AER_TYPE_NOTICE,
6711                                NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED,
6712                                NVME_LOG_CHANGED_NSLIST);
6713         }
6714     }
6715 
6716     return NVME_SUCCESS;
6717 }
6718 
6719 typedef struct NvmeFormatAIOCB {
6720     BlockAIOCB common;
6721     BlockAIOCB *aiocb;
6722     NvmeRequest *req;
6723     int ret;
6724 
6725     NvmeNamespace *ns;
6726     uint32_t nsid;
6727     bool broadcast;
6728     int64_t offset;
6729 
6730     uint8_t lbaf;
6731     uint8_t mset;
6732     uint8_t pi;
6733     uint8_t pil;
6734 } NvmeFormatAIOCB;
6735 
6736 static void nvme_format_cancel(BlockAIOCB *aiocb)
6737 {
6738     NvmeFormatAIOCB *iocb = container_of(aiocb, NvmeFormatAIOCB, common);
6739 
6740     iocb->ret = -ECANCELED;
6741 
6742     if (iocb->aiocb) {
6743         blk_aio_cancel_async(iocb->aiocb);
6744         iocb->aiocb = NULL;
6745     }
6746 }
6747 
6748 static const AIOCBInfo nvme_format_aiocb_info = {
6749     .aiocb_size = sizeof(NvmeFormatAIOCB),
6750     .cancel_async = nvme_format_cancel,
6751 };
6752 
6753 static void nvme_format_set(NvmeNamespace *ns, uint8_t lbaf, uint8_t mset,
6754                             uint8_t pi, uint8_t pil)
6755 {
6756     uint8_t lbafl = lbaf & 0xf;
6757     uint8_t lbafu = lbaf >> 4;
6758 
6759     trace_pci_nvme_format_set(ns->params.nsid, lbaf, mset, pi, pil);
6760 
6761     ns->id_ns.dps = (pil << 3) | pi;
6762     ns->id_ns.flbas = (lbafu << 5) | (mset << 4) | lbafl;
6763 
6764     nvme_ns_init_format(ns);
6765 }
6766 
6767 static void nvme_do_format(NvmeFormatAIOCB *iocb);
6768 
6769 static void nvme_format_ns_cb(void *opaque, int ret)
6770 {
6771     NvmeFormatAIOCB *iocb = opaque;
6772     NvmeNamespace *ns = iocb->ns;
6773     int bytes;
6774 
6775     if (iocb->ret < 0) {
6776         goto done;
6777     } else if (ret < 0) {
6778         iocb->ret = ret;
6779         goto done;
6780     }
6781 
6782     assert(ns);
6783 
6784     if (iocb->offset < ns->size) {
6785         bytes = MIN(BDRV_REQUEST_MAX_BYTES, ns->size - iocb->offset);
6786 
6787         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, iocb->offset,
6788                                             bytes, BDRV_REQ_MAY_UNMAP,
6789                                             nvme_format_ns_cb, iocb);
6790 
6791         iocb->offset += bytes;
6792         return;
6793     }
6794 
6795     nvme_format_set(ns, iocb->lbaf, iocb->mset, iocb->pi, iocb->pil);
6796     ns->status = 0x0;
6797     iocb->ns = NULL;
6798     iocb->offset = 0;
6799 
6800 done:
6801     nvme_do_format(iocb);
6802 }
6803 
6804 static uint16_t nvme_format_check(NvmeNamespace *ns, uint8_t lbaf, uint8_t pi)
6805 {
6806     if (ns->params.zoned) {
6807         return NVME_INVALID_FORMAT | NVME_DNR;
6808     }
6809 
6810     if (lbaf > ns->id_ns.nlbaf) {
6811         return NVME_INVALID_FORMAT | NVME_DNR;
6812     }
6813 
6814     if (pi && (ns->id_ns.lbaf[lbaf].ms < nvme_pi_tuple_size(ns))) {
6815         return NVME_INVALID_FORMAT | NVME_DNR;
6816     }
6817 
6818     if (pi && pi > NVME_ID_NS_DPS_TYPE_3) {
6819         return NVME_INVALID_FIELD | NVME_DNR;
6820     }
6821 
6822     return NVME_SUCCESS;
6823 }
6824 
6825 static void nvme_do_format(NvmeFormatAIOCB *iocb)
6826 {
6827     NvmeRequest *req = iocb->req;
6828     NvmeCtrl *n = nvme_ctrl(req);
6829     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6830     uint8_t lbaf = dw10 & 0xf;
6831     uint8_t pi = (dw10 >> 5) & 0x7;
6832     uint16_t status;
6833     int i;
6834 
6835     if (iocb->ret < 0) {
6836         goto done;
6837     }
6838 
6839     if (iocb->broadcast) {
6840         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
6841             iocb->ns = nvme_ns(n, i);
6842             if (iocb->ns) {
6843                 iocb->nsid = i;
6844                 break;
6845             }
6846         }
6847     }
6848 
6849     if (!iocb->ns) {
6850         goto done;
6851     }
6852 
6853     status = nvme_format_check(iocb->ns, lbaf, pi);
6854     if (status) {
6855         req->status = status;
6856         goto done;
6857     }
6858 
6859     iocb->ns->status = NVME_FORMAT_IN_PROGRESS;
6860     nvme_format_ns_cb(iocb, 0);
6861     return;
6862 
6863 done:
6864     iocb->common.cb(iocb->common.opaque, iocb->ret);
6865     qemu_aio_unref(iocb);
6866 }
6867 
6868 static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
6869 {
6870     NvmeFormatAIOCB *iocb;
6871     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6872     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6873     uint8_t lbaf = dw10 & 0xf;
6874     uint8_t mset = (dw10 >> 4) & 0x1;
6875     uint8_t pi = (dw10 >> 5) & 0x7;
6876     uint8_t pil = (dw10 >> 8) & 0x1;
6877     uint8_t lbafu = (dw10 >> 12) & 0x3;
6878     uint16_t status;
6879 
6880     iocb = qemu_aio_get(&nvme_format_aiocb_info, NULL, nvme_misc_cb, req);
6881 
6882     iocb->req = req;
6883     iocb->ret = 0;
6884     iocb->ns = NULL;
6885     iocb->nsid = 0;
6886     iocb->lbaf = lbaf;
6887     iocb->mset = mset;
6888     iocb->pi = pi;
6889     iocb->pil = pil;
6890     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
6891     iocb->offset = 0;
6892 
6893     if (n->features.hbs.lbafee) {
6894         iocb->lbaf |= lbafu << 4;
6895     }
6896 
6897     if (!iocb->broadcast) {
6898         if (!nvme_nsid_valid(n, nsid)) {
6899             status = NVME_INVALID_NSID | NVME_DNR;
6900             goto out;
6901         }
6902 
6903         iocb->ns = nvme_ns(n, nsid);
6904         if (!iocb->ns) {
6905             status = NVME_INVALID_FIELD | NVME_DNR;
6906             goto out;
6907         }
6908     }
6909 
6910     req->aiocb = &iocb->common;
6911     nvme_do_format(iocb);
6912 
6913     return NVME_NO_COMPLETE;
6914 
6915 out:
6916     qemu_aio_unref(iocb);
6917 
6918     return status;
6919 }
6920 
6921 static void nvme_get_virt_res_num(NvmeCtrl *n, uint8_t rt, int *num_total,
6922                                   int *num_prim, int *num_sec)
6923 {
6924     *num_total = le32_to_cpu(rt ?
6925                              n->pri_ctrl_cap.vifrt : n->pri_ctrl_cap.vqfrt);
6926     *num_prim = le16_to_cpu(rt ?
6927                             n->pri_ctrl_cap.virfap : n->pri_ctrl_cap.vqrfap);
6928     *num_sec = le16_to_cpu(rt ? n->pri_ctrl_cap.virfa : n->pri_ctrl_cap.vqrfa);
6929 }
6930 
6931 static uint16_t nvme_assign_virt_res_to_prim(NvmeCtrl *n, NvmeRequest *req,
6932                                              uint16_t cntlid, uint8_t rt,
6933                                              int nr)
6934 {
6935     int num_total, num_prim, num_sec;
6936 
6937     if (cntlid != n->cntlid) {
6938         return NVME_INVALID_CTRL_ID | NVME_DNR;
6939     }
6940 
6941     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6942 
6943     if (nr > num_total) {
6944         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6945     }
6946 
6947     if (nr > num_total - num_sec) {
6948         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6949     }
6950 
6951     if (rt) {
6952         n->next_pri_ctrl_cap.virfap = cpu_to_le16(nr);
6953     } else {
6954         n->next_pri_ctrl_cap.vqrfap = cpu_to_le16(nr);
6955     }
6956 
6957     req->cqe.result = cpu_to_le32(nr);
6958     return req->status;
6959 }
6960 
6961 static void nvme_update_virt_res(NvmeCtrl *n, NvmeSecCtrlEntry *sctrl,
6962                                  uint8_t rt, int nr)
6963 {
6964     int prev_nr, prev_total;
6965 
6966     if (rt) {
6967         prev_nr = le16_to_cpu(sctrl->nvi);
6968         prev_total = le32_to_cpu(n->pri_ctrl_cap.virfa);
6969         sctrl->nvi = cpu_to_le16(nr);
6970         n->pri_ctrl_cap.virfa = cpu_to_le32(prev_total + nr - prev_nr);
6971     } else {
6972         prev_nr = le16_to_cpu(sctrl->nvq);
6973         prev_total = le32_to_cpu(n->pri_ctrl_cap.vqrfa);
6974         sctrl->nvq = cpu_to_le16(nr);
6975         n->pri_ctrl_cap.vqrfa = cpu_to_le32(prev_total + nr - prev_nr);
6976     }
6977 }
6978 
6979 static uint16_t nvme_assign_virt_res_to_sec(NvmeCtrl *n, NvmeRequest *req,
6980                                             uint16_t cntlid, uint8_t rt, int nr)
6981 {
6982     int num_total, num_prim, num_sec, num_free, diff, limit;
6983     NvmeSecCtrlEntry *sctrl;
6984 
6985     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6986     if (!sctrl) {
6987         return NVME_INVALID_CTRL_ID | NVME_DNR;
6988     }
6989 
6990     if (sctrl->scs) {
6991         return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6992     }
6993 
6994     limit = le16_to_cpu(rt ? n->pri_ctrl_cap.vifrsm : n->pri_ctrl_cap.vqfrsm);
6995     if (nr > limit) {
6996         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6997     }
6998 
6999     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
7000     num_free = num_total - num_prim - num_sec;
7001     diff = nr - le16_to_cpu(rt ? sctrl->nvi : sctrl->nvq);
7002 
7003     if (diff > num_free) {
7004         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
7005     }
7006 
7007     nvme_update_virt_res(n, sctrl, rt, nr);
7008     req->cqe.result = cpu_to_le32(nr);
7009 
7010     return req->status;
7011 }
7012 
7013 static uint16_t nvme_virt_set_state(NvmeCtrl *n, uint16_t cntlid, bool online)
7014 {
7015     PCIDevice *pci = PCI_DEVICE(n);
7016     NvmeCtrl *sn = NULL;
7017     NvmeSecCtrlEntry *sctrl;
7018     int vf_index;
7019 
7020     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
7021     if (!sctrl) {
7022         return NVME_INVALID_CTRL_ID | NVME_DNR;
7023     }
7024 
7025     if (!pci_is_vf(pci)) {
7026         vf_index = le16_to_cpu(sctrl->vfn) - 1;
7027         sn = NVME(pcie_sriov_get_vf_at_index(pci, vf_index));
7028     }
7029 
7030     if (online) {
7031         if (!sctrl->nvi || (le16_to_cpu(sctrl->nvq) < 2) || !sn) {
7032             return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
7033         }
7034 
7035         if (!sctrl->scs) {
7036             sctrl->scs = 0x1;
7037             nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
7038         }
7039     } else {
7040         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_INTERRUPT, 0);
7041         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_QUEUE, 0);
7042 
7043         if (sctrl->scs) {
7044             sctrl->scs = 0x0;
7045             if (sn) {
7046                 nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
7047             }
7048         }
7049     }
7050 
7051     return NVME_SUCCESS;
7052 }
7053 
7054 static uint16_t nvme_virt_mngmt(NvmeCtrl *n, NvmeRequest *req)
7055 {
7056     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
7057     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
7058     uint8_t act = dw10 & 0xf;
7059     uint8_t rt = (dw10 >> 8) & 0x7;
7060     uint16_t cntlid = (dw10 >> 16) & 0xffff;
7061     int nr = dw11 & 0xffff;
7062 
7063     trace_pci_nvme_virt_mngmt(nvme_cid(req), act, cntlid, rt ? "VI" : "VQ", nr);
7064 
7065     if (rt != NVME_VIRT_RES_QUEUE && rt != NVME_VIRT_RES_INTERRUPT) {
7066         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
7067     }
7068 
7069     switch (act) {
7070     case NVME_VIRT_MNGMT_ACTION_SEC_ASSIGN:
7071         return nvme_assign_virt_res_to_sec(n, req, cntlid, rt, nr);
7072     case NVME_VIRT_MNGMT_ACTION_PRM_ALLOC:
7073         return nvme_assign_virt_res_to_prim(n, req, cntlid, rt, nr);
7074     case NVME_VIRT_MNGMT_ACTION_SEC_ONLINE:
7075         return nvme_virt_set_state(n, cntlid, true);
7076     case NVME_VIRT_MNGMT_ACTION_SEC_OFFLINE:
7077         return nvme_virt_set_state(n, cntlid, false);
7078     default:
7079         return NVME_INVALID_FIELD | NVME_DNR;
7080     }
7081 }
7082 
7083 static uint16_t nvme_dbbuf_config(NvmeCtrl *n, const NvmeRequest *req)
7084 {
7085     PCIDevice *pci = PCI_DEVICE(n);
7086     uint64_t dbs_addr = le64_to_cpu(req->cmd.dptr.prp1);
7087     uint64_t eis_addr = le64_to_cpu(req->cmd.dptr.prp2);
7088     int i;
7089 
7090     /* Address should be page aligned */
7091     if (dbs_addr & (n->page_size - 1) || eis_addr & (n->page_size - 1)) {
7092         return NVME_INVALID_FIELD | NVME_DNR;
7093     }
7094 
7095     /* Save shadow buffer base addr for use during queue creation */
7096     n->dbbuf_dbs = dbs_addr;
7097     n->dbbuf_eis = eis_addr;
7098     n->dbbuf_enabled = true;
7099 
7100     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7101         NvmeSQueue *sq = n->sq[i];
7102         NvmeCQueue *cq = n->cq[i];
7103 
7104         if (sq) {
7105             /*
7106              * CAP.DSTRD is 0, so offset of ith sq db_addr is (i<<3)
7107              * nvme_process_db() uses this hard-coded way to calculate
7108              * doorbell offsets. Be consistent with that here.
7109              */
7110             sq->db_addr = dbs_addr + (i << 3);
7111             sq->ei_addr = eis_addr + (i << 3);
7112             stl_le_pci_dma(pci, sq->db_addr, sq->tail, MEMTXATTRS_UNSPECIFIED);
7113 
7114             if (n->params.ioeventfd && sq->sqid != 0) {
7115                 if (!nvme_init_sq_ioeventfd(sq)) {
7116                     sq->ioeventfd_enabled = true;
7117                 }
7118             }
7119         }
7120 
7121         if (cq) {
7122             /* CAP.DSTRD is 0, so offset of ith cq db_addr is (i<<3)+(1<<2) */
7123             cq->db_addr = dbs_addr + (i << 3) + (1 << 2);
7124             cq->ei_addr = eis_addr + (i << 3) + (1 << 2);
7125             stl_le_pci_dma(pci, cq->db_addr, cq->head, MEMTXATTRS_UNSPECIFIED);
7126 
7127             if (n->params.ioeventfd && cq->cqid != 0) {
7128                 if (!nvme_init_cq_ioeventfd(cq)) {
7129                     cq->ioeventfd_enabled = true;
7130                 }
7131             }
7132         }
7133     }
7134 
7135     trace_pci_nvme_dbbuf_config(dbs_addr, eis_addr);
7136 
7137     return NVME_SUCCESS;
7138 }
7139 
7140 static uint16_t nvme_directive_send(NvmeCtrl *n, NvmeRequest *req)
7141 {
7142     return NVME_INVALID_FIELD | NVME_DNR;
7143 }
7144 
7145 static uint16_t nvme_directive_receive(NvmeCtrl *n, NvmeRequest *req)
7146 {
7147     NvmeNamespace *ns;
7148     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
7149     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
7150     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
7151     uint8_t doper, dtype;
7152     uint32_t numd, trans_len;
7153     NvmeDirectiveIdentify id = {
7154         .supported = 1 << NVME_DIRECTIVE_IDENTIFY,
7155         .enabled = 1 << NVME_DIRECTIVE_IDENTIFY,
7156     };
7157 
7158     numd = dw10 + 1;
7159     doper = dw11 & 0xff;
7160     dtype = (dw11 >> 8) & 0xff;
7161 
7162     trans_len = MIN(sizeof(NvmeDirectiveIdentify), numd << 2);
7163 
7164     if (nsid == NVME_NSID_BROADCAST || dtype != NVME_DIRECTIVE_IDENTIFY ||
7165         doper != NVME_DIRECTIVE_RETURN_PARAMS) {
7166         return NVME_INVALID_FIELD | NVME_DNR;
7167     }
7168 
7169     ns = nvme_ns(n, nsid);
7170     if (!ns) {
7171         return NVME_INVALID_FIELD | NVME_DNR;
7172     }
7173 
7174     switch (dtype) {
7175     case NVME_DIRECTIVE_IDENTIFY:
7176         switch (doper) {
7177         case NVME_DIRECTIVE_RETURN_PARAMS:
7178             if (ns->endgrp && ns->endgrp->fdp.enabled) {
7179                 id.supported |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
7180                 id.enabled |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
7181                 id.persistent |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
7182             }
7183 
7184             return nvme_c2h(n, (uint8_t *)&id, trans_len, req);
7185 
7186         default:
7187             return NVME_INVALID_FIELD | NVME_DNR;
7188         }
7189 
7190     default:
7191         return NVME_INVALID_FIELD;
7192     }
7193 }
7194 
7195 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
7196 {
7197     trace_pci_nvme_admin_cmd(nvme_cid(req), nvme_sqid(req), req->cmd.opcode,
7198                              nvme_adm_opc_str(req->cmd.opcode));
7199 
7200     if (!(nvme_cse_acs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
7201         trace_pci_nvme_err_invalid_admin_opc(req->cmd.opcode);
7202         return NVME_INVALID_OPCODE | NVME_DNR;
7203     }
7204 
7205     /* SGLs shall not be used for Admin commands in NVMe over PCIe */
7206     if (NVME_CMD_FLAGS_PSDT(req->cmd.flags) != NVME_PSDT_PRP) {
7207         return NVME_INVALID_FIELD | NVME_DNR;
7208     }
7209 
7210     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
7211         return NVME_INVALID_FIELD;
7212     }
7213 
7214     switch (req->cmd.opcode) {
7215     case NVME_ADM_CMD_DELETE_SQ:
7216         return nvme_del_sq(n, req);
7217     case NVME_ADM_CMD_CREATE_SQ:
7218         return nvme_create_sq(n, req);
7219     case NVME_ADM_CMD_GET_LOG_PAGE:
7220         return nvme_get_log(n, req);
7221     case NVME_ADM_CMD_DELETE_CQ:
7222         return nvme_del_cq(n, req);
7223     case NVME_ADM_CMD_CREATE_CQ:
7224         return nvme_create_cq(n, req);
7225     case NVME_ADM_CMD_IDENTIFY:
7226         return nvme_identify(n, req);
7227     case NVME_ADM_CMD_ABORT:
7228         return nvme_abort(n, req);
7229     case NVME_ADM_CMD_SET_FEATURES:
7230         return nvme_set_feature(n, req);
7231     case NVME_ADM_CMD_GET_FEATURES:
7232         return nvme_get_feature(n, req);
7233     case NVME_ADM_CMD_ASYNC_EV_REQ:
7234         return nvme_aer(n, req);
7235     case NVME_ADM_CMD_NS_ATTACHMENT:
7236         return nvme_ns_attachment(n, req);
7237     case NVME_ADM_CMD_VIRT_MNGMT:
7238         return nvme_virt_mngmt(n, req);
7239     case NVME_ADM_CMD_DBBUF_CONFIG:
7240         return nvme_dbbuf_config(n, req);
7241     case NVME_ADM_CMD_FORMAT_NVM:
7242         return nvme_format(n, req);
7243     case NVME_ADM_CMD_DIRECTIVE_SEND:
7244         return nvme_directive_send(n, req);
7245     case NVME_ADM_CMD_DIRECTIVE_RECV:
7246         return nvme_directive_receive(n, req);
7247     default:
7248         g_assert_not_reached();
7249     }
7250 
7251     return NVME_INVALID_OPCODE | NVME_DNR;
7252 }
7253 
7254 static void nvme_update_sq_eventidx(const NvmeSQueue *sq)
7255 {
7256     trace_pci_nvme_update_sq_eventidx(sq->sqid, sq->tail);
7257 
7258     stl_le_pci_dma(PCI_DEVICE(sq->ctrl), sq->ei_addr, sq->tail,
7259                    MEMTXATTRS_UNSPECIFIED);
7260 }
7261 
7262 static void nvme_update_sq_tail(NvmeSQueue *sq)
7263 {
7264     ldl_le_pci_dma(PCI_DEVICE(sq->ctrl), sq->db_addr, &sq->tail,
7265                    MEMTXATTRS_UNSPECIFIED);
7266 
7267     trace_pci_nvme_update_sq_tail(sq->sqid, sq->tail);
7268 }
7269 
7270 #define NVME_ATOMIC_NO_START        0
7271 #define NVME_ATOMIC_START_ATOMIC    1
7272 #define NVME_ATOMIC_START_NONATOMIC 2
7273 
7274 static int nvme_atomic_write_check(NvmeCtrl *n, NvmeCmd *cmd,
7275     NvmeAtomic *atomic)
7276 {
7277     NvmeRwCmd *rw = (NvmeRwCmd *)cmd;
7278     uint64_t slba = le64_to_cpu(rw->slba);
7279     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb);
7280     uint64_t elba = slba + nlb;
7281     bool cmd_atomic_wr = true;
7282     int i;
7283 
7284     if ((cmd->opcode == NVME_CMD_READ) || ((cmd->opcode == NVME_CMD_WRITE) &&
7285         ((rw->nlb + 1) > atomic->atomic_max_write_size))) {
7286         cmd_atomic_wr = false;
7287     }
7288 
7289     /*
7290      * Walk the queues to see if there are any atomic conflicts.
7291      */
7292     for (i = 1; i < n->params.max_ioqpairs + 1; i++) {
7293         NvmeSQueue *sq;
7294         NvmeRequest *req;
7295         NvmeRwCmd *req_rw;
7296         uint64_t req_slba;
7297         uint32_t req_nlb;
7298         uint64_t req_elba;
7299 
7300         sq = n->sq[i];
7301         if (!sq) {
7302             continue;
7303         }
7304 
7305         /*
7306          * Walk all the requests on a given queue.
7307          */
7308         QTAILQ_FOREACH(req, &sq->out_req_list, entry) {
7309             req_rw = (NvmeRwCmd *)&req->cmd;
7310 
7311             if (((req_rw->opcode == NVME_CMD_WRITE) ||
7312                  (req_rw->opcode == NVME_CMD_READ)) &&
7313                 (cmd->nsid == req->ns->params.nsid)) {
7314                 req_slba = le64_to_cpu(req_rw->slba);
7315                 req_nlb = (uint32_t)le16_to_cpu(req_rw->nlb);
7316                 req_elba = req_slba + req_nlb;
7317 
7318                 if (cmd_atomic_wr) {
7319                     if ((elba >= req_slba) && (slba <= req_elba)) {
7320                         return NVME_ATOMIC_NO_START;
7321                     }
7322                 } else {
7323                     if (req->atomic_write && ((elba >= req_slba) &&
7324                         (slba <= req_elba))) {
7325                         return NVME_ATOMIC_NO_START;
7326                     }
7327                 }
7328             }
7329         }
7330     }
7331     if (cmd_atomic_wr) {
7332         return NVME_ATOMIC_START_ATOMIC;
7333     }
7334     return NVME_ATOMIC_START_NONATOMIC;
7335 }
7336 
7337 static NvmeAtomic *nvme_get_atomic(NvmeCtrl *n, NvmeCmd *cmd)
7338 {
7339     if (n->atomic.atomic_writes) {
7340         return &n->atomic;
7341     }
7342     return NULL;
7343 }
7344 
7345 static void nvme_process_sq(void *opaque)
7346 {
7347     NvmeSQueue *sq = opaque;
7348     NvmeCtrl *n = sq->ctrl;
7349     NvmeCQueue *cq = n->cq[sq->cqid];
7350 
7351     uint16_t status;
7352     hwaddr addr;
7353     NvmeCmd cmd;
7354     NvmeRequest *req;
7355 
7356     if (n->dbbuf_enabled) {
7357         nvme_update_sq_tail(sq);
7358     }
7359 
7360     while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
7361         NvmeAtomic *atomic;
7362         bool cmd_is_atomic;
7363 
7364         addr = sq->dma_addr + (sq->head << NVME_SQES);
7365         if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
7366             trace_pci_nvme_err_addr_read(addr);
7367             trace_pci_nvme_err_cfs();
7368             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
7369             break;
7370         }
7371 
7372         atomic = nvme_get_atomic(n, &cmd);
7373 
7374         cmd_is_atomic = false;
7375         if (sq->sqid && atomic) {
7376             int ret;
7377 
7378             ret = nvme_atomic_write_check(n, &cmd, atomic);
7379             switch (ret) {
7380             case NVME_ATOMIC_NO_START:
7381                 qemu_bh_schedule(sq->bh);
7382                 return;
7383             case NVME_ATOMIC_START_ATOMIC:
7384                 cmd_is_atomic = true;
7385                 break;
7386             case NVME_ATOMIC_START_NONATOMIC:
7387             default:
7388                 break;
7389             }
7390         }
7391         nvme_inc_sq_head(sq);
7392 
7393         req = QTAILQ_FIRST(&sq->req_list);
7394         QTAILQ_REMOVE(&sq->req_list, req, entry);
7395         QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
7396         nvme_req_clear(req);
7397         req->cqe.cid = cmd.cid;
7398         memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
7399 
7400         if (sq->sqid && atomic) {
7401             req->atomic_write = cmd_is_atomic;
7402         }
7403 
7404         status = sq->sqid ? nvme_io_cmd(n, req) :
7405             nvme_admin_cmd(n, req);
7406         if (status != NVME_NO_COMPLETE) {
7407             req->status = status;
7408             nvme_enqueue_req_completion(cq, req);
7409         }
7410 
7411         if (n->dbbuf_enabled) {
7412             nvme_update_sq_eventidx(sq);
7413             nvme_update_sq_tail(sq);
7414         }
7415     }
7416 }
7417 
7418 static void nvme_update_msixcap_ts(PCIDevice *pci_dev, uint32_t table_size)
7419 {
7420     uint8_t *config;
7421 
7422     if (!msix_present(pci_dev)) {
7423         return;
7424     }
7425 
7426     assert(table_size > 0 && table_size <= pci_dev->msix_entries_nr);
7427 
7428     config = pci_dev->config + pci_dev->msix_cap;
7429     pci_set_word_by_mask(config + PCI_MSIX_FLAGS, PCI_MSIX_FLAGS_QSIZE,
7430                          table_size - 1);
7431 }
7432 
7433 static void nvme_activate_virt_res(NvmeCtrl *n)
7434 {
7435     PCIDevice *pci_dev = PCI_DEVICE(n);
7436     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7437     NvmeSecCtrlEntry *sctrl;
7438 
7439     /* -1 to account for the admin queue */
7440     if (pci_is_vf(pci_dev)) {
7441         sctrl = nvme_sctrl(n);
7442         cap->vqprt = sctrl->nvq;
7443         cap->viprt = sctrl->nvi;
7444         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7445         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7446     } else {
7447         cap->vqrfap = n->next_pri_ctrl_cap.vqrfap;
7448         cap->virfap = n->next_pri_ctrl_cap.virfap;
7449         n->conf_ioqpairs = le16_to_cpu(cap->vqprt) +
7450                            le16_to_cpu(cap->vqrfap) - 1;
7451         n->conf_msix_qsize = le16_to_cpu(cap->viprt) +
7452                              le16_to_cpu(cap->virfap);
7453     }
7454 }
7455 
7456 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst)
7457 {
7458     PCIDevice *pci_dev = PCI_DEVICE(n);
7459     NvmeSecCtrlEntry *sctrl;
7460     NvmeNamespace *ns;
7461     int i;
7462 
7463     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7464         ns = nvme_ns(n, i);
7465         if (!ns) {
7466             continue;
7467         }
7468 
7469         nvme_ns_drain(ns);
7470     }
7471 
7472     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7473         if (n->sq[i] != NULL) {
7474             nvme_free_sq(n->sq[i], n);
7475         }
7476     }
7477     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7478         if (n->cq[i] != NULL) {
7479             nvme_free_cq(n->cq[i], n);
7480         }
7481     }
7482 
7483     while (!QTAILQ_EMPTY(&n->aer_queue)) {
7484         NvmeAsyncEvent *event = QTAILQ_FIRST(&n->aer_queue);
7485         QTAILQ_REMOVE(&n->aer_queue, event, entry);
7486         g_free(event);
7487     }
7488 
7489     if (n->params.sriov_max_vfs) {
7490         if (!pci_is_vf(pci_dev)) {
7491             for (i = 0; i < n->nr_sec_ctrls; i++) {
7492                 sctrl = &n->sec_ctrl_list[i];
7493                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
7494             }
7495         }
7496 
7497         if (rst != NVME_RESET_CONTROLLER) {
7498             nvme_activate_virt_res(n);
7499         }
7500     }
7501 
7502     n->aer_queued = 0;
7503     n->aer_mask = 0;
7504     n->outstanding_aers = 0;
7505     n->qs_created = false;
7506 
7507     n->dn = n->params.atomic_dn; /* Set Disable Normal */
7508 
7509     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
7510 
7511     if (pci_is_vf(pci_dev)) {
7512         sctrl = nvme_sctrl(n);
7513 
7514         stl_le_p(&n->bar.csts, sctrl->scs ? 0 : NVME_CSTS_FAILED);
7515     } else {
7516         stl_le_p(&n->bar.csts, 0);
7517     }
7518 
7519     stl_le_p(&n->bar.intms, 0);
7520     stl_le_p(&n->bar.intmc, 0);
7521     stl_le_p(&n->bar.cc, 0);
7522 
7523     n->dbbuf_dbs = 0;
7524     n->dbbuf_eis = 0;
7525     n->dbbuf_enabled = false;
7526 }
7527 
7528 static void nvme_ctrl_shutdown(NvmeCtrl *n)
7529 {
7530     NvmeNamespace *ns;
7531     int i;
7532 
7533     if (n->pmr.dev) {
7534         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7535     }
7536 
7537     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7538         ns = nvme_ns(n, i);
7539         if (!ns) {
7540             continue;
7541         }
7542 
7543         nvme_ns_shutdown(ns);
7544     }
7545 }
7546 
7547 static void nvme_select_iocs(NvmeCtrl *n)
7548 {
7549     NvmeNamespace *ns;
7550     int i;
7551 
7552     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7553         ns = nvme_ns(n, i);
7554         if (!ns) {
7555             continue;
7556         }
7557 
7558         nvme_select_iocs_ns(n, ns);
7559     }
7560 }
7561 
7562 static int nvme_start_ctrl(NvmeCtrl *n)
7563 {
7564     uint64_t cap = ldq_le_p(&n->bar.cap);
7565     uint32_t cc = ldl_le_p(&n->bar.cc);
7566     uint32_t aqa = ldl_le_p(&n->bar.aqa);
7567     uint64_t asq = ldq_le_p(&n->bar.asq);
7568     uint64_t acq = ldq_le_p(&n->bar.acq);
7569     uint32_t page_bits = NVME_CC_MPS(cc) + 12;
7570     uint32_t page_size = 1 << page_bits;
7571     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
7572 
7573     if (pci_is_vf(PCI_DEVICE(n)) && !sctrl->scs) {
7574         trace_pci_nvme_err_startfail_virt_state(le16_to_cpu(sctrl->nvi),
7575                                                 le16_to_cpu(sctrl->nvq));
7576         return -1;
7577     }
7578     if (unlikely(n->cq[0])) {
7579         trace_pci_nvme_err_startfail_cq();
7580         return -1;
7581     }
7582     if (unlikely(n->sq[0])) {
7583         trace_pci_nvme_err_startfail_sq();
7584         return -1;
7585     }
7586     if (unlikely(asq & (page_size - 1))) {
7587         trace_pci_nvme_err_startfail_asq_misaligned(asq);
7588         return -1;
7589     }
7590     if (unlikely(acq & (page_size - 1))) {
7591         trace_pci_nvme_err_startfail_acq_misaligned(acq);
7592         return -1;
7593     }
7594     if (unlikely(!(NVME_CAP_CSS(cap) & (1 << NVME_CC_CSS(cc))))) {
7595         trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc));
7596         return -1;
7597     }
7598     if (unlikely(NVME_CC_MPS(cc) < NVME_CAP_MPSMIN(cap))) {
7599         trace_pci_nvme_err_startfail_page_too_small(
7600                     NVME_CC_MPS(cc),
7601                     NVME_CAP_MPSMIN(cap));
7602         return -1;
7603     }
7604     if (unlikely(NVME_CC_MPS(cc) >
7605                  NVME_CAP_MPSMAX(cap))) {
7606         trace_pci_nvme_err_startfail_page_too_large(
7607                     NVME_CC_MPS(cc),
7608                     NVME_CAP_MPSMAX(cap));
7609         return -1;
7610     }
7611     if (unlikely(!NVME_AQA_ASQS(aqa))) {
7612         trace_pci_nvme_err_startfail_asqent_sz_zero();
7613         return -1;
7614     }
7615     if (unlikely(!NVME_AQA_ACQS(aqa))) {
7616         trace_pci_nvme_err_startfail_acqent_sz_zero();
7617         return -1;
7618     }
7619 
7620     n->page_bits = page_bits;
7621     n->page_size = page_size;
7622     n->max_prp_ents = n->page_size / sizeof(uint64_t);
7623     nvme_init_cq(&n->admin_cq, n, acq, 0, 0, NVME_AQA_ACQS(aqa) + 1, 1);
7624     nvme_init_sq(&n->admin_sq, n, asq, 0, 0, NVME_AQA_ASQS(aqa) + 1);
7625 
7626     nvme_set_timestamp(n, 0ULL);
7627 
7628     nvme_select_iocs(n);
7629 
7630     return 0;
7631 }
7632 
7633 static void nvme_cmb_enable_regs(NvmeCtrl *n)
7634 {
7635     uint32_t cmbloc = ldl_le_p(&n->bar.cmbloc);
7636     uint32_t cmbsz = ldl_le_p(&n->bar.cmbsz);
7637 
7638     NVME_CMBLOC_SET_CDPCILS(cmbloc, 1);
7639     NVME_CMBLOC_SET_CDPMLS(cmbloc, 1);
7640     NVME_CMBLOC_SET_BIR(cmbloc, NVME_CMB_BIR);
7641     stl_le_p(&n->bar.cmbloc, cmbloc);
7642 
7643     NVME_CMBSZ_SET_SQS(cmbsz, 1);
7644     NVME_CMBSZ_SET_CQS(cmbsz, 0);
7645     NVME_CMBSZ_SET_LISTS(cmbsz, 1);
7646     NVME_CMBSZ_SET_RDS(cmbsz, 1);
7647     NVME_CMBSZ_SET_WDS(cmbsz, 1);
7648     NVME_CMBSZ_SET_SZU(cmbsz, 2); /* MBs */
7649     NVME_CMBSZ_SET_SZ(cmbsz, n->params.cmb_size_mb);
7650     stl_le_p(&n->bar.cmbsz, cmbsz);
7651 }
7652 
7653 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
7654                            unsigned size)
7655 {
7656     PCIDevice *pci = PCI_DEVICE(n);
7657     uint64_t cap = ldq_le_p(&n->bar.cap);
7658     uint32_t cc = ldl_le_p(&n->bar.cc);
7659     uint32_t intms = ldl_le_p(&n->bar.intms);
7660     uint32_t csts = ldl_le_p(&n->bar.csts);
7661     uint32_t pmrsts = ldl_le_p(&n->bar.pmrsts);
7662 
7663     if (unlikely(offset & (sizeof(uint32_t) - 1))) {
7664         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
7665                        "MMIO write not 32-bit aligned,"
7666                        " offset=0x%"PRIx64"", offset);
7667         /* should be ignored, fall through for now */
7668     }
7669 
7670     if (unlikely(size < sizeof(uint32_t))) {
7671         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
7672                        "MMIO write smaller than 32-bits,"
7673                        " offset=0x%"PRIx64", size=%u",
7674                        offset, size);
7675         /* should be ignored, fall through for now */
7676     }
7677 
7678     switch (offset) {
7679     case NVME_REG_INTMS:
7680         if (unlikely(msix_enabled(pci))) {
7681             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7682                            "undefined access to interrupt mask set"
7683                            " when MSI-X is enabled");
7684             /* should be ignored, fall through for now */
7685         }
7686         intms |= data;
7687         stl_le_p(&n->bar.intms, intms);
7688         n->bar.intmc = n->bar.intms;
7689         trace_pci_nvme_mmio_intm_set(data & 0xffffffff, intms);
7690         nvme_irq_check(n);
7691         break;
7692     case NVME_REG_INTMC:
7693         if (unlikely(msix_enabled(pci))) {
7694             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7695                            "undefined access to interrupt mask clr"
7696                            " when MSI-X is enabled");
7697             /* should be ignored, fall through for now */
7698         }
7699         intms &= ~data;
7700         stl_le_p(&n->bar.intms, intms);
7701         n->bar.intmc = n->bar.intms;
7702         trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, intms);
7703         nvme_irq_check(n);
7704         break;
7705     case NVME_REG_CC:
7706         stl_le_p(&n->bar.cc, data);
7707 
7708         trace_pci_nvme_mmio_cfg(data & 0xffffffff);
7709 
7710         if (NVME_CC_SHN(data) && !(NVME_CC_SHN(cc))) {
7711             trace_pci_nvme_mmio_shutdown_set();
7712             nvme_ctrl_shutdown(n);
7713             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7714             csts |= NVME_CSTS_SHST_COMPLETE;
7715         } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(cc)) {
7716             trace_pci_nvme_mmio_shutdown_cleared();
7717             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7718         }
7719 
7720         if (NVME_CC_EN(data) && !NVME_CC_EN(cc)) {
7721             if (unlikely(nvme_start_ctrl(n))) {
7722                 trace_pci_nvme_err_startfail();
7723                 csts = NVME_CSTS_FAILED;
7724             } else {
7725                 trace_pci_nvme_mmio_start_success();
7726                 csts = NVME_CSTS_READY;
7727             }
7728         } else if (!NVME_CC_EN(data) && NVME_CC_EN(cc)) {
7729             trace_pci_nvme_mmio_stopped();
7730             nvme_ctrl_reset(n, NVME_RESET_CONTROLLER);
7731 
7732             break;
7733         }
7734 
7735         stl_le_p(&n->bar.csts, csts);
7736 
7737         break;
7738     case NVME_REG_CSTS:
7739         if (data & (1 << 4)) {
7740             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
7741                            "attempted to W1C CSTS.NSSRO"
7742                            " but CAP.NSSRS is zero (not supported)");
7743         } else if (data != 0) {
7744             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
7745                            "attempted to set a read only bit"
7746                            " of controller status");
7747         }
7748         break;
7749     case NVME_REG_NSSR:
7750         if (data == 0x4e564d65) {
7751             trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
7752         } else {
7753             /* The spec says that writes of other values have no effect */
7754             return;
7755         }
7756         break;
7757     case NVME_REG_AQA:
7758         stl_le_p(&n->bar.aqa, data);
7759         trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
7760         break;
7761     case NVME_REG_ASQ:
7762         stn_le_p(&n->bar.asq, size, data);
7763         trace_pci_nvme_mmio_asqaddr(data);
7764         break;
7765     case NVME_REG_ASQ + 4:
7766         stl_le_p((uint8_t *)&n->bar.asq + 4, data);
7767         trace_pci_nvme_mmio_asqaddr_hi(data, ldq_le_p(&n->bar.asq));
7768         break;
7769     case NVME_REG_ACQ:
7770         trace_pci_nvme_mmio_acqaddr(data);
7771         stn_le_p(&n->bar.acq, size, data);
7772         break;
7773     case NVME_REG_ACQ + 4:
7774         stl_le_p((uint8_t *)&n->bar.acq + 4, data);
7775         trace_pci_nvme_mmio_acqaddr_hi(data, ldq_le_p(&n->bar.acq));
7776         break;
7777     case NVME_REG_CMBLOC:
7778         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
7779                        "invalid write to reserved CMBLOC"
7780                        " when CMBSZ is zero, ignored");
7781         return;
7782     case NVME_REG_CMBSZ:
7783         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
7784                        "invalid write to read only CMBSZ, ignored");
7785         return;
7786     case NVME_REG_CMBMSC:
7787         if (!NVME_CAP_CMBS(cap)) {
7788             return;
7789         }
7790 
7791         stn_le_p(&n->bar.cmbmsc, size, data);
7792         n->cmb.cmse = false;
7793 
7794         if (NVME_CMBMSC_CRE(data)) {
7795             nvme_cmb_enable_regs(n);
7796 
7797             if (NVME_CMBMSC_CMSE(data)) {
7798                 uint64_t cmbmsc = ldq_le_p(&n->bar.cmbmsc);
7799                 hwaddr cba = NVME_CMBMSC_CBA(cmbmsc) << CMBMSC_CBA_SHIFT;
7800                 if (cba + int128_get64(n->cmb.mem.size) < cba) {
7801                     uint32_t cmbsts = ldl_le_p(&n->bar.cmbsts);
7802                     NVME_CMBSTS_SET_CBAI(cmbsts, 1);
7803                     stl_le_p(&n->bar.cmbsts, cmbsts);
7804                     return;
7805                 }
7806 
7807                 n->cmb.cba = cba;
7808                 n->cmb.cmse = true;
7809             }
7810         } else {
7811             n->bar.cmbsz = 0;
7812             n->bar.cmbloc = 0;
7813         }
7814 
7815         return;
7816     case NVME_REG_CMBMSC + 4:
7817         stl_le_p((uint8_t *)&n->bar.cmbmsc + 4, data);
7818         return;
7819 
7820     case NVME_REG_PMRCAP:
7821         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
7822                        "invalid write to PMRCAP register, ignored");
7823         return;
7824     case NVME_REG_PMRCTL:
7825         if (!NVME_CAP_PMRS(cap)) {
7826             return;
7827         }
7828 
7829         stl_le_p(&n->bar.pmrctl, data);
7830         if (NVME_PMRCTL_EN(data)) {
7831             memory_region_set_enabled(&n->pmr.dev->mr, true);
7832             pmrsts = 0;
7833         } else {
7834             memory_region_set_enabled(&n->pmr.dev->mr, false);
7835             NVME_PMRSTS_SET_NRDY(pmrsts, 1);
7836             n->pmr.cmse = false;
7837         }
7838         stl_le_p(&n->bar.pmrsts, pmrsts);
7839         return;
7840     case NVME_REG_PMRSTS:
7841         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
7842                        "invalid write to PMRSTS register, ignored");
7843         return;
7844     case NVME_REG_PMREBS:
7845         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
7846                        "invalid write to PMREBS register, ignored");
7847         return;
7848     case NVME_REG_PMRSWTP:
7849         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
7850                        "invalid write to PMRSWTP register, ignored");
7851         return;
7852     case NVME_REG_PMRMSCL:
7853         if (!NVME_CAP_PMRS(cap)) {
7854             return;
7855         }
7856 
7857         stl_le_p(&n->bar.pmrmscl, data);
7858         n->pmr.cmse = false;
7859 
7860         if (NVME_PMRMSCL_CMSE(data)) {
7861             uint64_t pmrmscu = ldl_le_p(&n->bar.pmrmscu);
7862             hwaddr cba = pmrmscu << 32 |
7863                 (NVME_PMRMSCL_CBA(data) << PMRMSCL_CBA_SHIFT);
7864             if (cba + int128_get64(n->pmr.dev->mr.size) < cba) {
7865                 NVME_PMRSTS_SET_CBAI(pmrsts, 1);
7866                 stl_le_p(&n->bar.pmrsts, pmrsts);
7867                 return;
7868             }
7869 
7870             n->pmr.cmse = true;
7871             n->pmr.cba = cba;
7872         }
7873 
7874         return;
7875     case NVME_REG_PMRMSCU:
7876         if (!NVME_CAP_PMRS(cap)) {
7877             return;
7878         }
7879 
7880         stl_le_p(&n->bar.pmrmscu, data);
7881         return;
7882     default:
7883         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
7884                        "invalid MMIO write,"
7885                        " offset=0x%"PRIx64", data=%"PRIx64"",
7886                        offset, data);
7887         break;
7888     }
7889 }
7890 
7891 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
7892 {
7893     NvmeCtrl *n = (NvmeCtrl *)opaque;
7894     uint8_t *ptr = (uint8_t *)&n->bar;
7895 
7896     trace_pci_nvme_mmio_read(addr, size);
7897 
7898     if (unlikely(addr & (sizeof(uint32_t) - 1))) {
7899         NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
7900                        "MMIO read not 32-bit aligned,"
7901                        " offset=0x%"PRIx64"", addr);
7902         /* should RAZ, fall through for now */
7903     } else if (unlikely(size < sizeof(uint32_t))) {
7904         NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
7905                        "MMIO read smaller than 32-bits,"
7906                        " offset=0x%"PRIx64"", addr);
7907         /* should RAZ, fall through for now */
7908     }
7909 
7910     if (addr > sizeof(n->bar) - size) {
7911         NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
7912                        "MMIO read beyond last register,"
7913                        " offset=0x%"PRIx64", returning 0", addr);
7914 
7915         return 0;
7916     }
7917 
7918     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7919         addr != NVME_REG_CSTS) {
7920         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7921         return 0;
7922     }
7923 
7924     /*
7925      * When PMRWBM bit 1 is set then read from
7926      * from PMRSTS should ensure prior writes
7927      * made it to persistent media
7928      */
7929     if (addr == NVME_REG_PMRSTS &&
7930         (NVME_PMRCAP_PMRWBM(ldl_le_p(&n->bar.pmrcap)) & 0x02)) {
7931         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7932     }
7933 
7934     return ldn_le_p(ptr + addr, size);
7935 }
7936 
7937 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
7938 {
7939     PCIDevice *pci = PCI_DEVICE(n);
7940     uint32_t qid;
7941 
7942     if (unlikely(addr & ((1 << 2) - 1))) {
7943         NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
7944                        "doorbell write not 32-bit aligned,"
7945                        " offset=0x%"PRIx64", ignoring", addr);
7946         return;
7947     }
7948 
7949     if (((addr - 0x1000) >> 2) & 1) {
7950         /* Completion queue doorbell write */
7951 
7952         uint16_t new_head = val & 0xffff;
7953         int start_sqs;
7954         NvmeCQueue *cq;
7955 
7956         qid = (addr - (0x1000 + (1 << 2))) >> 3;
7957         if (unlikely(nvme_check_cqid(n, qid))) {
7958             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
7959                            "completion queue doorbell write"
7960                            " for nonexistent queue,"
7961                            " sqid=%"PRIu32", ignoring", qid);
7962 
7963             /*
7964              * NVM Express v1.3d, Section 4.1 state: "If host software writes
7965              * an invalid value to the Submission Queue Tail Doorbell or
7966              * Completion Queue Head Doorbell register and an Asynchronous Event
7967              * Request command is outstanding, then an asynchronous event is
7968              * posted to the Admin Completion Queue with a status code of
7969              * Invalid Doorbell Write Value."
7970              *
7971              * Also note that the spec includes the "Invalid Doorbell Register"
7972              * status code, but nowhere does it specify when to use it.
7973              * However, it seems reasonable to use it here in a similar
7974              * fashion.
7975              */
7976             if (n->outstanding_aers) {
7977                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7978                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7979                                    NVME_LOG_ERROR_INFO);
7980             }
7981 
7982             return;
7983         }
7984 
7985         cq = n->cq[qid];
7986         if (unlikely(new_head >= cq->size)) {
7987             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
7988                            "completion queue doorbell write value"
7989                            " beyond queue size, sqid=%"PRIu32","
7990                            " new_head=%"PRIu16", ignoring",
7991                            qid, new_head);
7992 
7993             if (n->outstanding_aers) {
7994                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7995                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7996                                    NVME_LOG_ERROR_INFO);
7997             }
7998 
7999             return;
8000         }
8001 
8002         trace_pci_nvme_mmio_doorbell_cq(cq->cqid, new_head);
8003 
8004         start_sqs = nvme_cq_full(cq) ? 1 : 0;
8005         cq->head = new_head;
8006         if (!qid && n->dbbuf_enabled) {
8007             stl_le_pci_dma(pci, cq->db_addr, cq->head, MEMTXATTRS_UNSPECIFIED);
8008         }
8009         if (start_sqs) {
8010             NvmeSQueue *sq;
8011             QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
8012                 qemu_bh_schedule(sq->bh);
8013             }
8014             qemu_bh_schedule(cq->bh);
8015         }
8016 
8017         if (cq->tail == cq->head) {
8018             if (cq->irq_enabled) {
8019                 n->cq_pending--;
8020             }
8021 
8022             nvme_irq_deassert(n, cq);
8023         }
8024     } else {
8025         /* Submission queue doorbell write */
8026 
8027         uint16_t new_tail = val & 0xffff;
8028         NvmeSQueue *sq;
8029 
8030         qid = (addr - 0x1000) >> 3;
8031         if (unlikely(nvme_check_sqid(n, qid))) {
8032             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
8033                            "submission queue doorbell write"
8034                            " for nonexistent queue,"
8035                            " sqid=%"PRIu32", ignoring", qid);
8036 
8037             if (n->outstanding_aers) {
8038                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
8039                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
8040                                    NVME_LOG_ERROR_INFO);
8041             }
8042 
8043             return;
8044         }
8045 
8046         sq = n->sq[qid];
8047         if (unlikely(new_tail >= sq->size)) {
8048             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
8049                            "submission queue doorbell write value"
8050                            " beyond queue size, sqid=%"PRIu32","
8051                            " new_tail=%"PRIu16", ignoring",
8052                            qid, new_tail);
8053 
8054             if (n->outstanding_aers) {
8055                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
8056                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
8057                                    NVME_LOG_ERROR_INFO);
8058             }
8059 
8060             return;
8061         }
8062 
8063         trace_pci_nvme_mmio_doorbell_sq(sq->sqid, new_tail);
8064 
8065         sq->tail = new_tail;
8066         if (!qid && n->dbbuf_enabled) {
8067             /*
8068              * The spec states "the host shall also update the controller's
8069              * corresponding doorbell property to match the value of that entry
8070              * in the Shadow Doorbell buffer."
8071              *
8072              * Since this context is currently a VM trap, we can safely enforce
8073              * the requirement from the device side in case the host is
8074              * misbehaving.
8075              *
8076              * Note, we shouldn't have to do this, but various drivers
8077              * including ones that run on Linux, are not updating Admin Queues,
8078              * so we can't trust reading it for an appropriate sq tail.
8079              */
8080             stl_le_pci_dma(pci, sq->db_addr, sq->tail, MEMTXATTRS_UNSPECIFIED);
8081         }
8082 
8083         qemu_bh_schedule(sq->bh);
8084     }
8085 }
8086 
8087 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
8088                             unsigned size)
8089 {
8090     NvmeCtrl *n = (NvmeCtrl *)opaque;
8091 
8092     trace_pci_nvme_mmio_write(addr, data, size);
8093 
8094     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
8095         addr != NVME_REG_CSTS) {
8096         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
8097         return;
8098     }
8099 
8100     if (addr < sizeof(n->bar)) {
8101         nvme_write_bar(n, addr, data, size);
8102     } else {
8103         nvme_process_db(n, addr, data);
8104     }
8105 }
8106 
8107 static const MemoryRegionOps nvme_mmio_ops = {
8108     .read = nvme_mmio_read,
8109     .write = nvme_mmio_write,
8110     .endianness = DEVICE_LITTLE_ENDIAN,
8111     .impl = {
8112         .min_access_size = 2,
8113         .max_access_size = 8,
8114     },
8115 };
8116 
8117 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
8118                            unsigned size)
8119 {
8120     NvmeCtrl *n = (NvmeCtrl *)opaque;
8121     stn_le_p(&n->cmb.buf[addr], size, data);
8122 }
8123 
8124 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
8125 {
8126     NvmeCtrl *n = (NvmeCtrl *)opaque;
8127     return ldn_le_p(&n->cmb.buf[addr], size);
8128 }
8129 
8130 static const MemoryRegionOps nvme_cmb_ops = {
8131     .read = nvme_cmb_read,
8132     .write = nvme_cmb_write,
8133     .endianness = DEVICE_LITTLE_ENDIAN,
8134     .impl = {
8135         .min_access_size = 1,
8136         .max_access_size = 8,
8137     },
8138 };
8139 
8140 static bool nvme_check_params(NvmeCtrl *n, Error **errp)
8141 {
8142     NvmeParams *params = &n->params;
8143 
8144     if (params->num_queues) {
8145         warn_report("num_queues is deprecated; please use max_ioqpairs "
8146                     "instead");
8147 
8148         params->max_ioqpairs = params->num_queues - 1;
8149     }
8150 
8151     if (n->namespace.blkconf.blk && n->subsys) {
8152         error_setg(errp, "subsystem support is unavailable with legacy "
8153                    "namespace ('drive' property)");
8154         return false;
8155     }
8156 
8157     if (params->max_ioqpairs < 1 ||
8158         params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
8159         error_setg(errp, "max_ioqpairs must be between 1 and %d",
8160                    NVME_MAX_IOQPAIRS);
8161         return false;
8162     }
8163 
8164     if (params->msix_qsize < 1 ||
8165         params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
8166         error_setg(errp, "msix_qsize must be between 1 and %d",
8167                    PCI_MSIX_FLAGS_QSIZE + 1);
8168         return false;
8169     }
8170 
8171     if (!params->serial) {
8172         error_setg(errp, "serial property not set");
8173         return false;
8174     }
8175 
8176     if (params->mqes < 1) {
8177         error_setg(errp, "mqes property cannot be less than 1");
8178         return false;
8179     }
8180 
8181     if (n->pmr.dev) {
8182         if (params->msix_exclusive_bar) {
8183             error_setg(errp, "not enough BARs available to enable PMR");
8184             return false;
8185         }
8186 
8187         if (host_memory_backend_is_mapped(n->pmr.dev)) {
8188             error_setg(errp, "can't use already busy memdev: %s",
8189                        object_get_canonical_path_component(OBJECT(n->pmr.dev)));
8190             return false;
8191         }
8192 
8193         if (!is_power_of_2(n->pmr.dev->size)) {
8194             error_setg(errp, "pmr backend size needs to be power of 2 in size");
8195             return false;
8196         }
8197 
8198         host_memory_backend_set_mapped(n->pmr.dev, true);
8199     }
8200 
8201     if (n->params.zasl > n->params.mdts) {
8202         error_setg(errp, "zoned.zasl (Zone Append Size Limit) must be less "
8203                    "than or equal to mdts (Maximum Data Transfer Size)");
8204         return false;
8205     }
8206 
8207     if (!n->params.vsl) {
8208         error_setg(errp, "vsl must be non-zero");
8209         return false;
8210     }
8211 
8212     if (params->sriov_max_vfs) {
8213         if (!n->subsys) {
8214             error_setg(errp, "subsystem is required for the use of SR-IOV");
8215             return false;
8216         }
8217 
8218         if (params->cmb_size_mb) {
8219             error_setg(errp, "CMB is not supported with SR-IOV");
8220             return false;
8221         }
8222 
8223         if (n->pmr.dev) {
8224             error_setg(errp, "PMR is not supported with SR-IOV");
8225             return false;
8226         }
8227 
8228         if (!params->sriov_vq_flexible || !params->sriov_vi_flexible) {
8229             error_setg(errp, "both sriov_vq_flexible and sriov_vi_flexible"
8230                        " must be set for the use of SR-IOV");
8231             return false;
8232         }
8233 
8234         if (params->sriov_vq_flexible < params->sriov_max_vfs * 2) {
8235             error_setg(errp, "sriov_vq_flexible must be greater than or equal"
8236                        " to %d (sriov_max_vfs * 2)", params->sriov_max_vfs * 2);
8237             return false;
8238         }
8239 
8240         if (params->max_ioqpairs < params->sriov_vq_flexible + 2) {
8241             error_setg(errp, "(max_ioqpairs - sriov_vq_flexible) must be"
8242                        " greater than or equal to 2");
8243             return false;
8244         }
8245 
8246         if (params->sriov_vi_flexible < params->sriov_max_vfs) {
8247             error_setg(errp, "sriov_vi_flexible must be greater than or equal"
8248                        " to %d (sriov_max_vfs)", params->sriov_max_vfs);
8249             return false;
8250         }
8251 
8252         if (params->msix_qsize < params->sriov_vi_flexible + 1) {
8253             error_setg(errp, "(msix_qsize - sriov_vi_flexible) must be"
8254                        " greater than or equal to 1");
8255             return false;
8256         }
8257 
8258         if (params->sriov_max_vi_per_vf &&
8259             (params->sriov_max_vi_per_vf - 1) % NVME_VF_RES_GRANULARITY) {
8260             error_setg(errp, "sriov_max_vi_per_vf must meet:"
8261                        " (sriov_max_vi_per_vf - 1) %% %d == 0 and"
8262                        " sriov_max_vi_per_vf >= 1", NVME_VF_RES_GRANULARITY);
8263             return false;
8264         }
8265 
8266         if (params->sriov_max_vq_per_vf &&
8267             (params->sriov_max_vq_per_vf < 2 ||
8268              (params->sriov_max_vq_per_vf - 1) % NVME_VF_RES_GRANULARITY)) {
8269             error_setg(errp, "sriov_max_vq_per_vf must meet:"
8270                        " (sriov_max_vq_per_vf - 1) %% %d == 0 and"
8271                        " sriov_max_vq_per_vf >= 2", NVME_VF_RES_GRANULARITY);
8272             return false;
8273         }
8274     }
8275 
8276     return true;
8277 }
8278 
8279 static void nvme_init_state(NvmeCtrl *n)
8280 {
8281     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
8282     NvmeSecCtrlEntry *list = n->sec_ctrl_list;
8283     NvmeSecCtrlEntry *sctrl;
8284     PCIDevice *pci = PCI_DEVICE(n);
8285     NvmeAtomic *atomic = &n->atomic;
8286     NvmeIdCtrl *id = &n->id_ctrl;
8287     uint8_t max_vfs;
8288     int i;
8289 
8290     if (pci_is_vf(pci)) {
8291         sctrl = nvme_sctrl(n);
8292         max_vfs = 0;
8293         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
8294         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
8295     } else {
8296         max_vfs = n->params.sriov_max_vfs;
8297         n->conf_ioqpairs = n->params.max_ioqpairs;
8298         n->conf_msix_qsize = n->params.msix_qsize;
8299     }
8300 
8301     n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
8302     n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
8303     n->temperature = NVME_TEMPERATURE;
8304     n->features.temp_thresh_hi = NVME_TEMPERATURE_WARNING;
8305     n->starttime_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
8306     n->aer_reqs = g_new0(NvmeRequest *, n->params.aerl + 1);
8307     QTAILQ_INIT(&n->aer_queue);
8308 
8309     n->nr_sec_ctrls = max_vfs;
8310     for (i = 0; i < max_vfs; i++) {
8311         sctrl = &list[i];
8312         sctrl->pcid = cpu_to_le16(n->cntlid);
8313         sctrl->vfn = cpu_to_le16(i + 1);
8314     }
8315 
8316     cap->cntlid = cpu_to_le16(n->cntlid);
8317     cap->crt = NVME_CRT_VQ | NVME_CRT_VI;
8318 
8319     if (pci_is_vf(pci)) {
8320         cap->vqprt = cpu_to_le16(1 + n->conf_ioqpairs);
8321     } else {
8322         cap->vqprt = cpu_to_le16(1 + n->params.max_ioqpairs -
8323                                  n->params.sriov_vq_flexible);
8324         cap->vqfrt = cpu_to_le32(n->params.sriov_vq_flexible);
8325         cap->vqrfap = cap->vqfrt;
8326         cap->vqgran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
8327         cap->vqfrsm = n->params.sriov_max_vq_per_vf ?
8328                         cpu_to_le16(n->params.sriov_max_vq_per_vf) :
8329                         cap->vqfrt / MAX(max_vfs, 1);
8330     }
8331 
8332     if (pci_is_vf(pci)) {
8333         cap->viprt = cpu_to_le16(n->conf_msix_qsize);
8334     } else {
8335         cap->viprt = cpu_to_le16(n->params.msix_qsize -
8336                                  n->params.sriov_vi_flexible);
8337         cap->vifrt = cpu_to_le32(n->params.sriov_vi_flexible);
8338         cap->virfap = cap->vifrt;
8339         cap->vigran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
8340         cap->vifrsm = n->params.sriov_max_vi_per_vf ?
8341                         cpu_to_le16(n->params.sriov_max_vi_per_vf) :
8342                         cap->vifrt / MAX(max_vfs, 1);
8343     }
8344 
8345     /* Atomic Write */
8346     id->awun = cpu_to_le16(n->params.atomic_awun);
8347     id->awupf = cpu_to_le16(n->params.atomic_awupf);
8348     n->dn = n->params.atomic_dn;
8349 
8350     if (id->awun || id->awupf) {
8351         if (id->awupf > id->awun) {
8352             id->awupf = 0;
8353         }
8354 
8355         if (n->dn) {
8356             atomic->atomic_max_write_size = id->awupf + 1;
8357         } else {
8358             atomic->atomic_max_write_size = id->awun + 1;
8359         }
8360 
8361         if (atomic->atomic_max_write_size == 1) {
8362             atomic->atomic_writes = 0;
8363         } else {
8364             atomic->atomic_writes = 1;
8365         }
8366     }
8367 }
8368 
8369 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
8370 {
8371     uint64_t cmb_size = n->params.cmb_size_mb * MiB;
8372     uint64_t cap = ldq_le_p(&n->bar.cap);
8373 
8374     n->cmb.buf = g_malloc0(cmb_size);
8375     memory_region_init_io(&n->cmb.mem, OBJECT(n), &nvme_cmb_ops, n,
8376                           "nvme-cmb", cmb_size);
8377     pci_register_bar(pci_dev, NVME_CMB_BIR,
8378                      PCI_BASE_ADDRESS_SPACE_MEMORY |
8379                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
8380                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->cmb.mem);
8381 
8382     NVME_CAP_SET_CMBS(cap, 1);
8383     stq_le_p(&n->bar.cap, cap);
8384 
8385     if (n->params.legacy_cmb) {
8386         nvme_cmb_enable_regs(n);
8387         n->cmb.cmse = true;
8388     }
8389 }
8390 
8391 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
8392 {
8393     uint32_t pmrcap = ldl_le_p(&n->bar.pmrcap);
8394 
8395     NVME_PMRCAP_SET_RDS(pmrcap, 1);
8396     NVME_PMRCAP_SET_WDS(pmrcap, 1);
8397     NVME_PMRCAP_SET_BIR(pmrcap, NVME_PMR_BIR);
8398     /* Turn on bit 1 support */
8399     NVME_PMRCAP_SET_PMRWBM(pmrcap, 0x02);
8400     NVME_PMRCAP_SET_CMSS(pmrcap, 1);
8401     stl_le_p(&n->bar.pmrcap, pmrcap);
8402 
8403     pci_register_bar(pci_dev, NVME_PMR_BIR,
8404                      PCI_BASE_ADDRESS_SPACE_MEMORY |
8405                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
8406                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmr.dev->mr);
8407 
8408     memory_region_set_enabled(&n->pmr.dev->mr, false);
8409 }
8410 
8411 static uint64_t nvme_mbar_size(unsigned total_queues, unsigned total_irqs,
8412                                unsigned *msix_table_offset,
8413                                unsigned *msix_pba_offset)
8414 {
8415     uint64_t bar_size, msix_table_size;
8416 
8417     bar_size = sizeof(NvmeBar) + 2 * total_queues * NVME_DB_SIZE;
8418 
8419     if (total_irqs == 0) {
8420         goto out;
8421     }
8422 
8423     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8424 
8425     if (msix_table_offset) {
8426         *msix_table_offset = bar_size;
8427     }
8428 
8429     msix_table_size = PCI_MSIX_ENTRY_SIZE * total_irqs;
8430     bar_size += msix_table_size;
8431     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8432 
8433     if (msix_pba_offset) {
8434         *msix_pba_offset = bar_size;
8435     }
8436 
8437     bar_size += QEMU_ALIGN_UP(total_irqs, 64) / 8;
8438 
8439 out:
8440     return pow2ceil(bar_size);
8441 }
8442 
8443 static void nvme_init_sriov(NvmeCtrl *n, PCIDevice *pci_dev, uint16_t offset)
8444 {
8445     uint16_t vf_dev_id = n->params.use_intel_id ?
8446                          PCI_DEVICE_ID_INTEL_NVME : PCI_DEVICE_ID_REDHAT_NVME;
8447     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
8448     uint64_t bar_size = nvme_mbar_size(le16_to_cpu(cap->vqfrsm),
8449                                       le16_to_cpu(cap->vifrsm),
8450                                       NULL, NULL);
8451 
8452     pcie_sriov_pf_init(pci_dev, offset, "nvme", vf_dev_id,
8453                        n->params.sriov_max_vfs, n->params.sriov_max_vfs,
8454                        NVME_VF_OFFSET, NVME_VF_STRIDE);
8455 
8456     pcie_sriov_pf_init_vf_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8457                               PCI_BASE_ADDRESS_MEM_TYPE_64, bar_size);
8458 }
8459 
8460 static int nvme_add_pm_capability(PCIDevice *pci_dev, uint8_t offset)
8461 {
8462     Error *err = NULL;
8463     int ret;
8464 
8465     ret = pci_add_capability(pci_dev, PCI_CAP_ID_PM, offset,
8466                              PCI_PM_SIZEOF, &err);
8467     if (err) {
8468         error_report_err(err);
8469         return ret;
8470     }
8471 
8472     pci_set_word(pci_dev->config + offset + PCI_PM_PMC,
8473                  PCI_PM_CAP_VER_1_2);
8474     pci_set_word(pci_dev->config + offset + PCI_PM_CTRL,
8475                  PCI_PM_CTRL_NO_SOFT_RESET);
8476     pci_set_word(pci_dev->wmask + offset + PCI_PM_CTRL,
8477                  PCI_PM_CTRL_STATE_MASK);
8478 
8479     return 0;
8480 }
8481 
8482 static bool pcie_doe_spdm_rsp(DOECap *doe_cap)
8483 {
8484     void *req = pcie_doe_get_write_mbox_ptr(doe_cap);
8485     uint32_t req_len = pcie_doe_get_obj_len(req) * 4;
8486     void *rsp = doe_cap->read_mbox;
8487     uint32_t rsp_len = SPDM_SOCKET_MAX_MESSAGE_BUFFER_SIZE;
8488 
8489     uint32_t recvd = spdm_socket_rsp(doe_cap->spdm_socket,
8490                              SPDM_SOCKET_TRANSPORT_TYPE_PCI_DOE,
8491                              req, req_len, rsp, rsp_len);
8492     doe_cap->read_mbox_len += DIV_ROUND_UP(recvd, 4);
8493 
8494     return recvd != 0;
8495 }
8496 
8497 static DOEProtocol doe_spdm_prot[] = {
8498     { PCI_VENDOR_ID_PCI_SIG, PCI_SIG_DOE_CMA, pcie_doe_spdm_rsp },
8499     { PCI_VENDOR_ID_PCI_SIG, PCI_SIG_DOE_SECURED_CMA, pcie_doe_spdm_rsp },
8500     { }
8501 };
8502 
8503 static bool nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
8504 {
8505     ERRP_GUARD();
8506     uint8_t *pci_conf = pci_dev->config;
8507     uint64_t bar_size;
8508     unsigned msix_table_offset = 0, msix_pba_offset = 0;
8509     unsigned nr_vectors;
8510     int ret;
8511 
8512     pci_conf[PCI_INTERRUPT_PIN] = 1;
8513     pci_config_set_prog_interface(pci_conf, 0x2);
8514 
8515     if (n->params.use_intel_id) {
8516         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_INTEL);
8517         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_INTEL_NVME);
8518     } else {
8519         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_REDHAT);
8520         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_REDHAT_NVME);
8521     }
8522 
8523     pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
8524     nvme_add_pm_capability(pci_dev, 0x60);
8525     pcie_endpoint_cap_init(pci_dev, 0x80);
8526     pcie_cap_flr_init(pci_dev);
8527     if (n->params.sriov_max_vfs) {
8528         pcie_ari_init(pci_dev, 0x100);
8529     }
8530 
8531     if (n->params.msix_exclusive_bar && !pci_is_vf(pci_dev)) {
8532         bar_size = nvme_mbar_size(n->params.max_ioqpairs + 1, 0, NULL, NULL);
8533         memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
8534                               bar_size);
8535         pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8536                          PCI_BASE_ADDRESS_MEM_TYPE_64, &n->iomem);
8537         ret = msix_init_exclusive_bar(pci_dev, n->params.msix_qsize, 4, errp);
8538     } else {
8539         assert(n->params.msix_qsize >= 1);
8540 
8541         /* add one to max_ioqpairs to account for the admin queue pair */
8542         if (!pci_is_vf(pci_dev)) {
8543             nr_vectors = n->params.msix_qsize;
8544             bar_size = nvme_mbar_size(n->params.max_ioqpairs + 1,
8545                                       nr_vectors, &msix_table_offset,
8546                                       &msix_pba_offset);
8547         } else {
8548             NvmeCtrl *pn = NVME(pcie_sriov_get_pf(pci_dev));
8549             NvmePriCtrlCap *cap = &pn->pri_ctrl_cap;
8550 
8551             nr_vectors = le16_to_cpu(cap->vifrsm);
8552             bar_size = nvme_mbar_size(le16_to_cpu(cap->vqfrsm), nr_vectors,
8553                                       &msix_table_offset, &msix_pba_offset);
8554         }
8555 
8556         memory_region_init(&n->bar0, OBJECT(n), "nvme-bar0", bar_size);
8557         memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
8558                               msix_table_offset);
8559         memory_region_add_subregion(&n->bar0, 0, &n->iomem);
8560 
8561         if (pci_is_vf(pci_dev)) {
8562             pcie_sriov_vf_register_bar(pci_dev, 0, &n->bar0);
8563         } else {
8564             pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8565                              PCI_BASE_ADDRESS_MEM_TYPE_64, &n->bar0);
8566         }
8567 
8568         ret = msix_init(pci_dev, nr_vectors,
8569                         &n->bar0, 0, msix_table_offset,
8570                         &n->bar0, 0, msix_pba_offset, 0, errp);
8571     }
8572 
8573     if (ret == -ENOTSUP) {
8574         /* report that msix is not supported, but do not error out */
8575         warn_report_err(*errp);
8576         *errp = NULL;
8577     } else if (ret < 0) {
8578         /* propagate error to caller */
8579         return false;
8580     }
8581 
8582     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
8583 
8584     pcie_cap_deverr_init(pci_dev);
8585 
8586     /* DOE Initialisation */
8587     if (pci_dev->spdm_port) {
8588         uint16_t doe_offset = n->params.sriov_max_vfs ?
8589                                   PCI_CONFIG_SPACE_SIZE + PCI_ARI_SIZEOF
8590                                   : PCI_CONFIG_SPACE_SIZE;
8591 
8592         pcie_doe_init(pci_dev, &pci_dev->doe_spdm, doe_offset,
8593                       doe_spdm_prot, true, 0);
8594 
8595         pci_dev->doe_spdm.spdm_socket = spdm_socket_connect(pci_dev->spdm_port,
8596                                                             errp);
8597 
8598         if (pci_dev->doe_spdm.spdm_socket < 0) {
8599             return false;
8600         }
8601     }
8602 
8603     if (n->params.cmb_size_mb) {
8604         nvme_init_cmb(n, pci_dev);
8605     }
8606 
8607     if (n->pmr.dev) {
8608         nvme_init_pmr(n, pci_dev);
8609     }
8610 
8611     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8612         nvme_init_sriov(n, pci_dev, 0x120);
8613     }
8614 
8615     return true;
8616 }
8617 
8618 static void nvme_init_subnqn(NvmeCtrl *n)
8619 {
8620     NvmeSubsystem *subsys = n->subsys;
8621     NvmeIdCtrl *id = &n->id_ctrl;
8622 
8623     if (!subsys) {
8624         snprintf((char *)id->subnqn, sizeof(id->subnqn),
8625                  "nqn.2019-08.org.qemu:%s", n->params.serial);
8626     } else {
8627         pstrcpy((char *)id->subnqn, sizeof(id->subnqn), (char*)subsys->subnqn);
8628     }
8629 }
8630 
8631 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
8632 {
8633     NvmeIdCtrl *id = &n->id_ctrl;
8634     uint8_t *pci_conf = pci_dev->config;
8635     uint64_t cap = ldq_le_p(&n->bar.cap);
8636     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
8637     uint32_t ctratt;
8638 
8639     id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
8640     id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
8641     strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
8642     strpadcpy((char *)id->fr, sizeof(id->fr), QEMU_VERSION, ' ');
8643     strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
8644 
8645     id->cntlid = cpu_to_le16(n->cntlid);
8646 
8647     id->oaes = cpu_to_le32(NVME_OAES_NS_ATTR);
8648 
8649     ctratt = NVME_CTRATT_ELBAS;
8650     if (n->params.ctratt.mem) {
8651         ctratt |= NVME_CTRATT_MEM;
8652     }
8653 
8654     id->rab = 6;
8655 
8656     if (n->params.use_intel_id) {
8657         id->ieee[0] = 0xb3;
8658         id->ieee[1] = 0x02;
8659         id->ieee[2] = 0x00;
8660     } else {
8661         id->ieee[0] = 0x00;
8662         id->ieee[1] = 0x54;
8663         id->ieee[2] = 0x52;
8664     }
8665 
8666     id->mdts = n->params.mdts;
8667     id->ver = cpu_to_le32(NVME_SPEC_VER);
8668     id->oacs =
8669         cpu_to_le16(NVME_OACS_NS_MGMT | NVME_OACS_FORMAT | NVME_OACS_DBBUF |
8670                     NVME_OACS_DIRECTIVES);
8671     id->cntrltype = 0x1;
8672 
8673     /*
8674      * Because the controller always completes the Abort command immediately,
8675      * there can never be more than one concurrently executing Abort command,
8676      * so this value is never used for anything. Note that there can easily be
8677      * many Abort commands in the queues, but they are not considered
8678      * "executing" until processed by nvme_abort.
8679      *
8680      * The specification recommends a value of 3 for Abort Command Limit (four
8681      * concurrently outstanding Abort commands), so lets use that though it is
8682      * inconsequential.
8683      */
8684     id->acl = 3;
8685     id->aerl = n->params.aerl;
8686     id->frmw = (NVME_NUM_FW_SLOTS << 1) | NVME_FRMW_SLOT1_RO;
8687     id->lpa = NVME_LPA_NS_SMART | NVME_LPA_CSE | NVME_LPA_EXTENDED;
8688 
8689     /* recommended default value (~70 C) */
8690     id->wctemp = cpu_to_le16(NVME_TEMPERATURE_WARNING);
8691     id->cctemp = cpu_to_le16(NVME_TEMPERATURE_CRITICAL);
8692 
8693     id->sqes = (NVME_SQES << 4) | NVME_SQES;
8694     id->cqes = (NVME_CQES << 4) | NVME_CQES;
8695     id->nn = cpu_to_le32(NVME_MAX_NAMESPACES);
8696     id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP |
8697                            NVME_ONCS_FEATURES | NVME_ONCS_DSM |
8698                            NVME_ONCS_COMPARE | NVME_ONCS_COPY |
8699                            NVME_ONCS_NVMCSA | NVME_ONCS_NVMAFC);
8700 
8701     /*
8702      * NOTE: If this device ever supports a command set that does NOT use 0x0
8703      * as a Flush-equivalent operation, support for the broadcast NSID in Flush
8704      * should probably be removed.
8705      *
8706      * See comment in nvme_io_cmd.
8707      */
8708     id->vwc = NVME_VWC_NSID_BROADCAST_SUPPORT | NVME_VWC_PRESENT;
8709 
8710     id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0 | NVME_OCFS_COPY_FORMAT_1 |
8711                             NVME_OCFS_COPY_FORMAT_2 | NVME_OCFS_COPY_FORMAT_3);
8712     id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN |
8713                            NVME_CTRL_SGLS_MPTR_SGL);
8714 
8715     nvme_init_subnqn(n);
8716 
8717     id->psd[0].mp = cpu_to_le16(0x9c4);
8718     id->psd[0].enlat = cpu_to_le32(0x10);
8719     id->psd[0].exlat = cpu_to_le32(0x4);
8720 
8721     if (n->subsys) {
8722         id->cmic |= NVME_CMIC_MULTI_CTRL;
8723         ctratt |= NVME_CTRATT_ENDGRPS;
8724 
8725         id->endgidmax = cpu_to_le16(0x1);
8726 
8727         if (n->subsys->endgrp.fdp.enabled) {
8728             ctratt |= NVME_CTRATT_FDPS;
8729         }
8730     }
8731 
8732     id->ctratt = cpu_to_le32(ctratt);
8733 
8734     NVME_CAP_SET_MQES(cap, n->params.mqes);
8735     NVME_CAP_SET_CQR(cap, 1);
8736     NVME_CAP_SET_TO(cap, 0xf);
8737     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_NVM);
8738     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_CSI_SUPP);
8739     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_ADMIN_ONLY);
8740     NVME_CAP_SET_MPSMAX(cap, 4);
8741     NVME_CAP_SET_CMBS(cap, n->params.cmb_size_mb ? 1 : 0);
8742     NVME_CAP_SET_PMRS(cap, n->pmr.dev ? 1 : 0);
8743     stq_le_p(&n->bar.cap, cap);
8744 
8745     stl_le_p(&n->bar.vs, NVME_SPEC_VER);
8746     n->bar.intmc = n->bar.intms = 0;
8747 
8748     if (pci_is_vf(pci_dev) && !sctrl->scs) {
8749         stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
8750     }
8751 }
8752 
8753 static int nvme_init_subsys(NvmeCtrl *n, Error **errp)
8754 {
8755     int cntlid;
8756 
8757     if (!n->subsys) {
8758         return 0;
8759     }
8760 
8761     cntlid = nvme_subsys_register_ctrl(n, errp);
8762     if (cntlid < 0) {
8763         return -1;
8764     }
8765 
8766     n->cntlid = cntlid;
8767 
8768     return 0;
8769 }
8770 
8771 void nvme_attach_ns(NvmeCtrl *n, NvmeNamespace *ns)
8772 {
8773     uint32_t nsid = ns->params.nsid;
8774     assert(nsid && nsid <= NVME_MAX_NAMESPACES);
8775 
8776     n->namespaces[nsid] = ns;
8777     ns->attached++;
8778 
8779     n->dmrsl = MIN_NON_ZERO(n->dmrsl,
8780                             BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
8781 }
8782 
8783 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
8784 {
8785     NvmeCtrl *n = NVME(pci_dev);
8786     DeviceState *dev = DEVICE(pci_dev);
8787     NvmeNamespace *ns;
8788     NvmeCtrl *pn = NVME(pcie_sriov_get_pf(pci_dev));
8789 
8790     if (pci_is_vf(pci_dev)) {
8791         /*
8792          * VFs derive settings from the parent. PF's lifespan exceeds
8793          * that of VF's.
8794          */
8795         memcpy(&n->params, &pn->params, sizeof(NvmeParams));
8796 
8797         /*
8798          * Set PF's serial value to a new string memory to prevent 'serial'
8799          * property object release of PF when a VF is removed from the system.
8800          */
8801         n->params.serial = g_strdup(pn->params.serial);
8802         n->subsys = pn->subsys;
8803     }
8804 
8805     if (!nvme_check_params(n, errp)) {
8806         return;
8807     }
8808 
8809     qbus_init(&n->bus, sizeof(NvmeBus), TYPE_NVME_BUS, dev, dev->id);
8810 
8811     if (nvme_init_subsys(n, errp)) {
8812         return;
8813     }
8814     nvme_init_state(n);
8815     if (!nvme_init_pci(n, pci_dev, errp)) {
8816         return;
8817     }
8818     nvme_init_ctrl(n, pci_dev);
8819 
8820     /* setup a namespace if the controller drive property was given */
8821     if (n->namespace.blkconf.blk) {
8822         ns = &n->namespace;
8823         ns->params.nsid = 1;
8824 
8825         if (nvme_ns_setup(ns, errp)) {
8826             return;
8827         }
8828 
8829         nvme_attach_ns(n, ns);
8830     }
8831 }
8832 
8833 static void nvme_exit(PCIDevice *pci_dev)
8834 {
8835     NvmeCtrl *n = NVME(pci_dev);
8836     NvmeNamespace *ns;
8837     int i;
8838 
8839     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8840 
8841     if (n->subsys) {
8842         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
8843             ns = nvme_ns(n, i);
8844             if (ns) {
8845                 ns->attached--;
8846             }
8847         }
8848 
8849         nvme_subsys_unregister_ctrl(n->subsys, n);
8850     }
8851 
8852     g_free(n->cq);
8853     g_free(n->sq);
8854     g_free(n->aer_reqs);
8855 
8856     if (n->params.cmb_size_mb) {
8857         g_free(n->cmb.buf);
8858     }
8859 
8860     if (pci_dev->doe_spdm.spdm_socket > 0) {
8861         spdm_socket_close(pci_dev->doe_spdm.spdm_socket,
8862                           SPDM_SOCKET_TRANSPORT_TYPE_PCI_DOE);
8863     }
8864 
8865     if (n->pmr.dev) {
8866         host_memory_backend_set_mapped(n->pmr.dev, false);
8867     }
8868 
8869     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8870         pcie_sriov_pf_exit(pci_dev);
8871     }
8872 
8873     msix_uninit(pci_dev, &n->bar0, &n->bar0);
8874     memory_region_del_subregion(&n->bar0, &n->iomem);
8875 }
8876 
8877 static Property nvme_props[] = {
8878     DEFINE_BLOCK_PROPERTIES(NvmeCtrl, namespace.blkconf),
8879     DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmr.dev, TYPE_MEMORY_BACKEND,
8880                      HostMemoryBackend *),
8881     DEFINE_PROP_LINK("subsys", NvmeCtrl, subsys, TYPE_NVME_SUBSYS,
8882                      NvmeSubsystem *),
8883     DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
8884     DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
8885     DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
8886     DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
8887     DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
8888     DEFINE_PROP_UINT8("aerl", NvmeCtrl, params.aerl, 3),
8889     DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl, params.aer_max_queued, 64),
8890     DEFINE_PROP_UINT8("mdts", NvmeCtrl, params.mdts, 7),
8891     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
8892     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
8893     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
8894     DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl, params.ioeventfd, false),
8895     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
8896     DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl,
8897                      params.auto_transition_zones, true),
8898     DEFINE_PROP_UINT16("sriov_max_vfs", NvmeCtrl, params.sriov_max_vfs, 0),
8899     DEFINE_PROP_UINT16("sriov_vq_flexible", NvmeCtrl,
8900                        params.sriov_vq_flexible, 0),
8901     DEFINE_PROP_UINT16("sriov_vi_flexible", NvmeCtrl,
8902                        params.sriov_vi_flexible, 0),
8903     DEFINE_PROP_UINT32("sriov_max_vi_per_vf", NvmeCtrl,
8904                        params.sriov_max_vi_per_vf, 0),
8905     DEFINE_PROP_UINT32("sriov_max_vq_per_vf", NvmeCtrl,
8906                        params.sriov_max_vq_per_vf, 0),
8907     DEFINE_PROP_BOOL("msix-exclusive-bar", NvmeCtrl, params.msix_exclusive_bar,
8908                      false),
8909     DEFINE_PROP_UINT16("mqes", NvmeCtrl, params.mqes, 0x7ff),
8910     DEFINE_PROP_UINT16("spdm_port", PCIDevice, spdm_port, 0),
8911     DEFINE_PROP_BOOL("ctratt.mem", NvmeCtrl, params.ctratt.mem, false),
8912     DEFINE_PROP_BOOL("atomic.dn", NvmeCtrl, params.atomic_dn, 0),
8913     DEFINE_PROP_UINT16("atomic.awun", NvmeCtrl, params.atomic_awun, 0),
8914     DEFINE_PROP_UINT16("atomic.awupf", NvmeCtrl, params.atomic_awupf, 0),
8915     DEFINE_PROP_END_OF_LIST(),
8916 };
8917 
8918 static void nvme_get_smart_warning(Object *obj, Visitor *v, const char *name,
8919                                    void *opaque, Error **errp)
8920 {
8921     NvmeCtrl *n = NVME(obj);
8922     uint8_t value = n->smart_critical_warning;
8923 
8924     visit_type_uint8(v, name, &value, errp);
8925 }
8926 
8927 static void nvme_set_smart_warning(Object *obj, Visitor *v, const char *name,
8928                                    void *opaque, Error **errp)
8929 {
8930     NvmeCtrl *n = NVME(obj);
8931     uint8_t value, old_value, cap = 0, index, event;
8932 
8933     if (!visit_type_uint8(v, name, &value, errp)) {
8934         return;
8935     }
8936 
8937     cap = NVME_SMART_SPARE | NVME_SMART_TEMPERATURE | NVME_SMART_RELIABILITY
8938           | NVME_SMART_MEDIA_READ_ONLY | NVME_SMART_FAILED_VOLATILE_MEDIA;
8939     if (NVME_CAP_PMRS(ldq_le_p(&n->bar.cap))) {
8940         cap |= NVME_SMART_PMR_UNRELIABLE;
8941     }
8942 
8943     if ((value & cap) != value) {
8944         error_setg(errp, "unsupported smart critical warning bits: 0x%x",
8945                    value & ~cap);
8946         return;
8947     }
8948 
8949     old_value = n->smart_critical_warning;
8950     n->smart_critical_warning = value;
8951 
8952     /* only inject new bits of smart critical warning */
8953     for (index = 0; index < NVME_SMART_WARN_MAX; index++) {
8954         event = 1 << index;
8955         if (value & ~old_value & event)
8956             nvme_smart_event(n, event);
8957     }
8958 }
8959 
8960 static void nvme_pci_reset(DeviceState *qdev)
8961 {
8962     PCIDevice *pci_dev = PCI_DEVICE(qdev);
8963     NvmeCtrl *n = NVME(pci_dev);
8964 
8965     trace_pci_nvme_pci_reset();
8966     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8967 }
8968 
8969 static void nvme_sriov_post_write_config(PCIDevice *dev, uint16_t old_num_vfs)
8970 {
8971     NvmeCtrl *n = NVME(dev);
8972     NvmeSecCtrlEntry *sctrl;
8973     int i;
8974 
8975     for (i = pcie_sriov_num_vfs(dev); i < old_num_vfs; i++) {
8976         sctrl = &n->sec_ctrl_list[i];
8977         nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
8978     }
8979 }
8980 
8981 static void nvme_pci_write_config(PCIDevice *dev, uint32_t address,
8982                                   uint32_t val, int len)
8983 {
8984     uint16_t old_num_vfs = pcie_sriov_num_vfs(dev);
8985 
8986     if (pcie_find_capability(dev, PCI_EXT_CAP_ID_DOE)) {
8987         pcie_doe_write_config(&dev->doe_spdm, address, val, len);
8988     }
8989     pci_default_write_config(dev, address, val, len);
8990     pcie_cap_flr_write_config(dev, address, val, len);
8991     nvme_sriov_post_write_config(dev, old_num_vfs);
8992 }
8993 
8994 static uint32_t nvme_pci_read_config(PCIDevice *dev, uint32_t address, int len)
8995 {
8996     uint32_t val;
8997     if (dev->spdm_port && pcie_find_capability(dev, PCI_EXT_CAP_ID_DOE)) {
8998         if (pcie_doe_read_config(&dev->doe_spdm, address, len, &val)) {
8999             return val;
9000         }
9001     }
9002     return pci_default_read_config(dev, address, len);
9003 }
9004 
9005 static const VMStateDescription nvme_vmstate = {
9006     .name = "nvme",
9007     .unmigratable = 1,
9008 };
9009 
9010 static void nvme_class_init(ObjectClass *oc, void *data)
9011 {
9012     DeviceClass *dc = DEVICE_CLASS(oc);
9013     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
9014 
9015     pc->realize = nvme_realize;
9016     pc->config_write = nvme_pci_write_config;
9017     pc->config_read = nvme_pci_read_config;
9018     pc->exit = nvme_exit;
9019     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
9020     pc->revision = 2;
9021 
9022     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
9023     dc->desc = "Non-Volatile Memory Express";
9024     device_class_set_props(dc, nvme_props);
9025     dc->vmsd = &nvme_vmstate;
9026     device_class_set_legacy_reset(dc, nvme_pci_reset);
9027 }
9028 
9029 static void nvme_instance_init(Object *obj)
9030 {
9031     NvmeCtrl *n = NVME(obj);
9032 
9033     device_add_bootindex_property(obj, &n->namespace.blkconf.bootindex,
9034                                   "bootindex", "/namespace@1,0",
9035                                   DEVICE(obj));
9036 
9037     object_property_add(obj, "smart_critical_warning", "uint8",
9038                         nvme_get_smart_warning,
9039                         nvme_set_smart_warning, NULL, NULL);
9040 }
9041 
9042 static const TypeInfo nvme_info = {
9043     .name          = TYPE_NVME,
9044     .parent        = TYPE_PCI_DEVICE,
9045     .instance_size = sizeof(NvmeCtrl),
9046     .instance_init = nvme_instance_init,
9047     .class_init    = nvme_class_init,
9048     .interfaces = (InterfaceInfo[]) {
9049         { INTERFACE_PCIE_DEVICE },
9050         { }
9051     },
9052 };
9053 
9054 static const TypeInfo nvme_bus_info = {
9055     .name = TYPE_NVME_BUS,
9056     .parent = TYPE_BUS,
9057     .instance_size = sizeof(NvmeBus),
9058 };
9059 
9060 static void nvme_register_types(void)
9061 {
9062     type_register_static(&nvme_info);
9063     type_register_static(&nvme_bus_info);
9064 }
9065 
9066 type_init(nvme_register_types)
9067