xref: /openbmc/qemu/hw/nvme/ctrl.c (revision 83340193)
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  *              subsys=<subsys_id>
44  *      -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
45  *              zoned=<true|false[optional]>, \
46  *              subsys=<subsys_id>,shared=<true|false[optional]>, \
47  *              detached=<true|false[optional]>, \
48  *              zoned.zone_size=<N[optional]>, \
49  *              zoned.zone_capacity=<N[optional]>, \
50  *              zoned.descr_ext_size=<N[optional]>, \
51  *              zoned.max_active=<N[optional]>, \
52  *              zoned.max_open=<N[optional]>, \
53  *              zoned.cross_read=<true|false[optional]>
54  *
55  * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
56  * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. By default, the
57  * device will use the "v1.4 CMB scheme" - use the `legacy-cmb` parameter to
58  * always enable the CMBLOC and CMBSZ registers (v1.3 behavior).
59  *
60  * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
61  * For example:
62  * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
63  *  size=<size> .... -device nvme,...,pmrdev=<mem_id>
64  *
65  * The PMR will use BAR 4/5 exclusively.
66  *
67  * To place controller(s) and namespace(s) to a subsystem, then provide
68  * nvme-subsys device as above.
69  *
70  * nvme subsystem device parameters
71  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
72  * - `nqn`
73  *   This parameter provides the `<nqn_id>` part of the string
74  *   `nqn.2019-08.org.qemu:<nqn_id>` which will be reported in the SUBNQN field
75  *   of subsystem controllers. Note that `<nqn_id>` should be unique per
76  *   subsystem, but this is not enforced by QEMU. If not specified, it will
77  *   default to the value of the `id` parameter (`<subsys_id>`).
78  *
79  * nvme device parameters
80  * ~~~~~~~~~~~~~~~~~~~~~~
81  * - `subsys`
82  *   Specifying this parameter attaches the controller to the subsystem and
83  *   the SUBNQN field in the controller will report the NQN of the subsystem
84  *   device. This also enables multi controller capability represented in
85  *   Identify Controller data structure in CMIC (Controller Multi-path I/O and
86  *   Namespace Sharing Capabilities).
87  *
88  * - `aerl`
89  *   The Asynchronous Event Request Limit (AERL). Indicates the maximum number
90  *   of concurrently outstanding Asynchronous Event Request commands support
91  *   by the controller. This is a 0's based value.
92  *
93  * - `aer_max_queued`
94  *   This is the maximum number of events that the device will enqueue for
95  *   completion when there are no outstanding AERs. When the maximum number of
96  *   enqueued events are reached, subsequent events will be dropped.
97  *
98  * - `mdts`
99  *   Indicates the maximum data transfer size for a command that transfers data
100  *   between host-accessible memory and the controller. The value is specified
101  *   as a power of two (2^n) and is in units of the minimum memory page size
102  *   (CAP.MPSMIN). The default value is 7 (i.e. 512 KiB).
103  *
104  * - `vsl`
105  *   Indicates the maximum data size limit for the Verify command. Like `mdts`,
106  *   this value is specified as a power of two (2^n) and is in units of the
107  *   minimum memory page size (CAP.MPSMIN). The default value is 7 (i.e. 512
108  *   KiB).
109  *
110  * - `zoned.zasl`
111  *   Indicates the maximum data transfer size for the Zone Append command. Like
112  *   `mdts`, the value is specified as a power of two (2^n) and is in units of
113  *   the minimum memory page size (CAP.MPSMIN). The default value is 0 (i.e.
114  *   defaulting to the value of `mdts`).
115  *
116  * - `zoned.auto_transition`
117  *   Indicates if zones in zone state implicitly opened can be automatically
118  *   transitioned to zone state closed for resource management purposes.
119  *   Defaults to 'on'.
120  *
121  * - `sriov_max_vfs`
122  *   Indicates the maximum number of PCIe virtual functions supported
123  *   by the controller. The default value is 0. Specifying a non-zero value
124  *   enables reporting of both SR-IOV and ARI capabilities by the NVMe device.
125  *   Virtual function controllers will not report SR-IOV capability.
126  *
127  *   NOTE: Single Root I/O Virtualization support is experimental.
128  *   All the related parameters may be subject to change.
129  *
130  * - `sriov_vq_flexible`
131  *   Indicates the total number of flexible queue resources assignable to all
132  *   the secondary controllers. Implicitly sets the number of primary
133  *   controller's private resources to `(max_ioqpairs - sriov_vq_flexible)`.
134  *
135  * - `sriov_vi_flexible`
136  *   Indicates the total number of flexible interrupt resources assignable to
137  *   all the secondary controllers. Implicitly sets the number of primary
138  *   controller's private resources to `(msix_qsize - sriov_vi_flexible)`.
139  *
140  * - `sriov_max_vi_per_vf`
141  *   Indicates the maximum number of virtual interrupt resources assignable
142  *   to a secondary controller. The default 0 resolves to
143  *   `(sriov_vi_flexible / sriov_max_vfs)`.
144  *
145  * - `sriov_max_vq_per_vf`
146  *   Indicates the maximum number of virtual queue resources assignable to
147  *   a secondary controller. The default 0 resolves to
148  *   `(sriov_vq_flexible / sriov_max_vfs)`.
149  *
150  * nvme namespace device parameters
151  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
152  * - `shared`
153  *   When the parent nvme device (as defined explicitly by the 'bus' parameter
154  *   or implicitly by the most recently defined NvmeBus) is linked to an
155  *   nvme-subsys device, the namespace will be attached to all controllers in
156  *   the subsystem. If set to 'off' (the default), the namespace will remain a
157  *   private namespace and may only be attached to a single controller at a
158  *   time.
159  *
160  * - `detached`
161  *   This parameter is only valid together with the `subsys` parameter. If left
162  *   at the default value (`false/off`), the namespace will be attached to all
163  *   controllers in the NVMe subsystem at boot-up. If set to `true/on`, the
164  *   namespace will be available in the subsystem but not attached to any
165  *   controllers.
166  *
167  * Setting `zoned` to true selects Zoned Command Set at the namespace.
168  * In this case, the following namespace properties are available to configure
169  * zoned operation:
170  *     zoned.zone_size=<zone size in bytes, default: 128MiB>
171  *         The number may be followed by K, M, G as in kilo-, mega- or giga-.
172  *
173  *     zoned.zone_capacity=<zone capacity in bytes, default: zone size>
174  *         The value 0 (default) forces zone capacity to be the same as zone
175  *         size. The value of this property may not exceed zone size.
176  *
177  *     zoned.descr_ext_size=<zone descriptor extension size, default 0>
178  *         This value needs to be specified in 64B units. If it is zero,
179  *         namespace(s) will not support zone descriptor extensions.
180  *
181  *     zoned.max_active=<Maximum Active Resources (zones), default: 0>
182  *         The default value means there is no limit to the number of
183  *         concurrently active zones.
184  *
185  *     zoned.max_open=<Maximum Open Resources (zones), default: 0>
186  *         The default value means there is no limit to the number of
187  *         concurrently open zones.
188  *
189  *     zoned.cross_read=<enable RAZB, default: false>
190  *         Setting this property to true enables Read Across Zone Boundaries.
191  */
192 
193 #include "qemu/osdep.h"
194 #include "qemu/cutils.h"
195 #include "qemu/error-report.h"
196 #include "qemu/log.h"
197 #include "qemu/units.h"
198 #include "qemu/range.h"
199 #include "qapi/error.h"
200 #include "qapi/visitor.h"
201 #include "sysemu/sysemu.h"
202 #include "sysemu/block-backend.h"
203 #include "sysemu/hostmem.h"
204 #include "hw/pci/msix.h"
205 #include "hw/pci/pcie_sriov.h"
206 #include "sysemu/spdm-socket.h"
207 #include "migration/vmstate.h"
208 
209 #include "nvme.h"
210 #include "dif.h"
211 #include "trace.h"
212 
213 #define NVME_MAX_IOQPAIRS 0xffff
214 #define NVME_DB_SIZE  4
215 #define NVME_SPEC_VER 0x00010400
216 #define NVME_CMB_BIR 2
217 #define NVME_PMR_BIR 4
218 #define NVME_TEMPERATURE 0x143
219 #define NVME_TEMPERATURE_WARNING 0x157
220 #define NVME_TEMPERATURE_CRITICAL 0x175
221 #define NVME_NUM_FW_SLOTS 1
222 #define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
223 #define NVME_VF_RES_GRANULARITY 1
224 #define NVME_VF_OFFSET 0x1
225 #define NVME_VF_STRIDE 1
226 
227 #define NVME_GUEST_ERR(trace, fmt, ...) \
228     do { \
229         (trace_##trace)(__VA_ARGS__); \
230         qemu_log_mask(LOG_GUEST_ERROR, #trace \
231             " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
232     } while (0)
233 
234 static const bool nvme_feature_support[NVME_FID_MAX] = {
235     [NVME_ARBITRATION]              = true,
236     [NVME_POWER_MANAGEMENT]         = true,
237     [NVME_TEMPERATURE_THRESHOLD]    = true,
238     [NVME_ERROR_RECOVERY]           = true,
239     [NVME_VOLATILE_WRITE_CACHE]     = true,
240     [NVME_NUMBER_OF_QUEUES]         = true,
241     [NVME_INTERRUPT_COALESCING]     = true,
242     [NVME_INTERRUPT_VECTOR_CONF]    = true,
243     [NVME_WRITE_ATOMICITY]          = true,
244     [NVME_ASYNCHRONOUS_EVENT_CONF]  = true,
245     [NVME_TIMESTAMP]                = true,
246     [NVME_HOST_BEHAVIOR_SUPPORT]    = true,
247     [NVME_COMMAND_SET_PROFILE]      = true,
248     [NVME_FDP_MODE]                 = true,
249     [NVME_FDP_EVENTS]               = true,
250 };
251 
252 static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
253     [NVME_TEMPERATURE_THRESHOLD]    = NVME_FEAT_CAP_CHANGE,
254     [NVME_ERROR_RECOVERY]           = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
255     [NVME_VOLATILE_WRITE_CACHE]     = NVME_FEAT_CAP_CHANGE,
256     [NVME_NUMBER_OF_QUEUES]         = NVME_FEAT_CAP_CHANGE,
257     [NVME_ASYNCHRONOUS_EVENT_CONF]  = NVME_FEAT_CAP_CHANGE,
258     [NVME_TIMESTAMP]                = NVME_FEAT_CAP_CHANGE,
259     [NVME_HOST_BEHAVIOR_SUPPORT]    = NVME_FEAT_CAP_CHANGE,
260     [NVME_COMMAND_SET_PROFILE]      = NVME_FEAT_CAP_CHANGE,
261     [NVME_FDP_MODE]                 = NVME_FEAT_CAP_CHANGE,
262     [NVME_FDP_EVENTS]               = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
263 };
264 
265 static const uint32_t nvme_cse_acs[256] = {
266     [NVME_ADM_CMD_DELETE_SQ]        = NVME_CMD_EFF_CSUPP,
267     [NVME_ADM_CMD_CREATE_SQ]        = NVME_CMD_EFF_CSUPP,
268     [NVME_ADM_CMD_GET_LOG_PAGE]     = NVME_CMD_EFF_CSUPP,
269     [NVME_ADM_CMD_DELETE_CQ]        = NVME_CMD_EFF_CSUPP,
270     [NVME_ADM_CMD_CREATE_CQ]        = NVME_CMD_EFF_CSUPP,
271     [NVME_ADM_CMD_IDENTIFY]         = NVME_CMD_EFF_CSUPP,
272     [NVME_ADM_CMD_ABORT]            = NVME_CMD_EFF_CSUPP,
273     [NVME_ADM_CMD_SET_FEATURES]     = NVME_CMD_EFF_CSUPP,
274     [NVME_ADM_CMD_GET_FEATURES]     = NVME_CMD_EFF_CSUPP,
275     [NVME_ADM_CMD_ASYNC_EV_REQ]     = NVME_CMD_EFF_CSUPP,
276     [NVME_ADM_CMD_NS_ATTACHMENT]    = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_NIC,
277     [NVME_ADM_CMD_VIRT_MNGMT]       = NVME_CMD_EFF_CSUPP,
278     [NVME_ADM_CMD_DBBUF_CONFIG]     = NVME_CMD_EFF_CSUPP,
279     [NVME_ADM_CMD_FORMAT_NVM]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
280     [NVME_ADM_CMD_DIRECTIVE_RECV]   = NVME_CMD_EFF_CSUPP,
281     [NVME_ADM_CMD_DIRECTIVE_SEND]   = NVME_CMD_EFF_CSUPP,
282 };
283 
284 static const uint32_t nvme_cse_iocs_none[256];
285 
286 static const uint32_t nvme_cse_iocs_nvm[256] = {
287     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
288     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
289     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
290     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
291     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
292     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
293     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
294     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
295     [NVME_CMD_IO_MGMT_RECV]         = NVME_CMD_EFF_CSUPP,
296     [NVME_CMD_IO_MGMT_SEND]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
297 };
298 
299 static const uint32_t nvme_cse_iocs_zoned[256] = {
300     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
301     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
302     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
303     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
304     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
305     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
306     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
307     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
308     [NVME_CMD_ZONE_APPEND]          = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
309     [NVME_CMD_ZONE_MGMT_SEND]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
310     [NVME_CMD_ZONE_MGMT_RECV]       = NVME_CMD_EFF_CSUPP,
311 };
312 
313 static void nvme_process_sq(void *opaque);
314 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst);
315 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n);
316 
317 static uint16_t nvme_sqid(NvmeRequest *req)
318 {
319     return le16_to_cpu(req->sq->sqid);
320 }
321 
322 static inline uint16_t nvme_make_pid(NvmeNamespace *ns, uint16_t rg,
323                                      uint16_t ph)
324 {
325     uint16_t rgif = ns->endgrp->fdp.rgif;
326 
327     if (!rgif) {
328         return ph;
329     }
330 
331     return (rg << (16 - rgif)) | ph;
332 }
333 
334 static inline bool nvme_ph_valid(NvmeNamespace *ns, uint16_t ph)
335 {
336     return ph < ns->fdp.nphs;
337 }
338 
339 static inline bool nvme_rg_valid(NvmeEnduranceGroup *endgrp, uint16_t rg)
340 {
341     return rg < endgrp->fdp.nrg;
342 }
343 
344 static inline uint16_t nvme_pid2ph(NvmeNamespace *ns, uint16_t pid)
345 {
346     uint16_t rgif = ns->endgrp->fdp.rgif;
347 
348     if (!rgif) {
349         return pid;
350     }
351 
352     return pid & ((1 << (15 - rgif)) - 1);
353 }
354 
355 static inline uint16_t nvme_pid2rg(NvmeNamespace *ns, uint16_t pid)
356 {
357     uint16_t rgif = ns->endgrp->fdp.rgif;
358 
359     if (!rgif) {
360         return 0;
361     }
362 
363     return pid >> (16 - rgif);
364 }
365 
366 static inline bool nvme_parse_pid(NvmeNamespace *ns, uint16_t pid,
367                                   uint16_t *ph, uint16_t *rg)
368 {
369     *rg = nvme_pid2rg(ns, pid);
370     *ph = nvme_pid2ph(ns, pid);
371 
372     return nvme_ph_valid(ns, *ph) && nvme_rg_valid(ns->endgrp, *rg);
373 }
374 
375 static void nvme_assign_zone_state(NvmeNamespace *ns, NvmeZone *zone,
376                                    NvmeZoneState state)
377 {
378     if (QTAILQ_IN_USE(zone, entry)) {
379         switch (nvme_get_zone_state(zone)) {
380         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
381             QTAILQ_REMOVE(&ns->exp_open_zones, zone, entry);
382             break;
383         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
384             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
385             break;
386         case NVME_ZONE_STATE_CLOSED:
387             QTAILQ_REMOVE(&ns->closed_zones, zone, entry);
388             break;
389         case NVME_ZONE_STATE_FULL:
390             QTAILQ_REMOVE(&ns->full_zones, zone, entry);
391         default:
392             ;
393         }
394     }
395 
396     nvme_set_zone_state(zone, state);
397 
398     switch (state) {
399     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
400         QTAILQ_INSERT_TAIL(&ns->exp_open_zones, zone, entry);
401         break;
402     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
403         QTAILQ_INSERT_TAIL(&ns->imp_open_zones, zone, entry);
404         break;
405     case NVME_ZONE_STATE_CLOSED:
406         QTAILQ_INSERT_TAIL(&ns->closed_zones, zone, entry);
407         break;
408     case NVME_ZONE_STATE_FULL:
409         QTAILQ_INSERT_TAIL(&ns->full_zones, zone, entry);
410     case NVME_ZONE_STATE_READ_ONLY:
411         break;
412     default:
413         zone->d.za = 0;
414     }
415 }
416 
417 static uint16_t nvme_zns_check_resources(NvmeNamespace *ns, uint32_t act,
418                                          uint32_t opn, uint32_t zrwa)
419 {
420     if (ns->params.max_active_zones != 0 &&
421         ns->nr_active_zones + act > ns->params.max_active_zones) {
422         trace_pci_nvme_err_insuff_active_res(ns->params.max_active_zones);
423         return NVME_ZONE_TOO_MANY_ACTIVE | NVME_DNR;
424     }
425 
426     if (ns->params.max_open_zones != 0 &&
427         ns->nr_open_zones + opn > ns->params.max_open_zones) {
428         trace_pci_nvme_err_insuff_open_res(ns->params.max_open_zones);
429         return NVME_ZONE_TOO_MANY_OPEN | NVME_DNR;
430     }
431 
432     if (zrwa > ns->zns.numzrwa) {
433         return NVME_NOZRWA | NVME_DNR;
434     }
435 
436     return NVME_SUCCESS;
437 }
438 
439 /*
440  * Check if we can open a zone without exceeding open/active limits.
441  * AOR stands for "Active and Open Resources" (see TP 4053 section 2.5).
442  */
443 static uint16_t nvme_aor_check(NvmeNamespace *ns, uint32_t act, uint32_t opn)
444 {
445     return nvme_zns_check_resources(ns, act, opn, 0);
446 }
447 
448 static NvmeFdpEvent *nvme_fdp_alloc_event(NvmeCtrl *n, NvmeFdpEventBuffer *ebuf)
449 {
450     NvmeFdpEvent *ret = NULL;
451     bool is_full = ebuf->next == ebuf->start && ebuf->nelems;
452 
453     ret = &ebuf->events[ebuf->next++];
454     if (unlikely(ebuf->next == NVME_FDP_MAX_EVENTS)) {
455         ebuf->next = 0;
456     }
457     if (is_full) {
458         ebuf->start = ebuf->next;
459     } else {
460         ebuf->nelems++;
461     }
462 
463     memset(ret, 0, sizeof(NvmeFdpEvent));
464     ret->timestamp = nvme_get_timestamp(n);
465 
466     return ret;
467 }
468 
469 static inline int log_event(NvmeRuHandle *ruh, uint8_t event_type)
470 {
471     return (ruh->event_filter >> nvme_fdp_evf_shifts[event_type]) & 0x1;
472 }
473 
474 static bool nvme_update_ruh(NvmeCtrl *n, NvmeNamespace *ns, uint16_t pid)
475 {
476     NvmeEnduranceGroup *endgrp = ns->endgrp;
477     NvmeRuHandle *ruh;
478     NvmeReclaimUnit *ru;
479     NvmeFdpEvent *e = NULL;
480     uint16_t ph, rg, ruhid;
481 
482     if (!nvme_parse_pid(ns, pid, &ph, &rg)) {
483         return false;
484     }
485 
486     ruhid = ns->fdp.phs[ph];
487 
488     ruh = &endgrp->fdp.ruhs[ruhid];
489     ru = &ruh->rus[rg];
490 
491     if (ru->ruamw) {
492         if (log_event(ruh, FDP_EVT_RU_NOT_FULLY_WRITTEN)) {
493             e = nvme_fdp_alloc_event(n, &endgrp->fdp.host_events);
494             e->type = FDP_EVT_RU_NOT_FULLY_WRITTEN;
495             e->flags = FDPEF_PIV | FDPEF_NSIDV | FDPEF_LV;
496             e->pid = cpu_to_le16(pid);
497             e->nsid = cpu_to_le32(ns->params.nsid);
498             e->rgid = cpu_to_le16(rg);
499             e->ruhid = cpu_to_le16(ruhid);
500         }
501 
502         /* log (eventual) GC overhead of prematurely swapping the RU */
503         nvme_fdp_stat_inc(&endgrp->fdp.mbmw, nvme_l2b(ns, ru->ruamw));
504     }
505 
506     ru->ruamw = ruh->ruamw;
507 
508     return true;
509 }
510 
511 static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
512 {
513     hwaddr hi, lo;
514 
515     if (!n->cmb.cmse) {
516         return false;
517     }
518 
519     lo = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
520     hi = lo + int128_get64(n->cmb.mem.size);
521 
522     return addr >= lo && addr < hi;
523 }
524 
525 static inline void *nvme_addr_to_cmb(NvmeCtrl *n, hwaddr addr)
526 {
527     hwaddr base = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
528     return &n->cmb.buf[addr - base];
529 }
530 
531 static bool nvme_addr_is_pmr(NvmeCtrl *n, hwaddr addr)
532 {
533     hwaddr hi;
534 
535     if (!n->pmr.cmse) {
536         return false;
537     }
538 
539     hi = n->pmr.cba + int128_get64(n->pmr.dev->mr.size);
540 
541     return addr >= n->pmr.cba && addr < hi;
542 }
543 
544 static inline void *nvme_addr_to_pmr(NvmeCtrl *n, hwaddr addr)
545 {
546     return memory_region_get_ram_ptr(&n->pmr.dev->mr) + (addr - n->pmr.cba);
547 }
548 
549 static inline bool nvme_addr_is_iomem(NvmeCtrl *n, hwaddr addr)
550 {
551     hwaddr hi, lo;
552 
553     /*
554      * The purpose of this check is to guard against invalid "local" access to
555      * the iomem (i.e. controller registers). Thus, we check against the range
556      * covered by the 'bar0' MemoryRegion since that is currently composed of
557      * two subregions (the NVMe "MBAR" and the MSI-X table/pba). Note, however,
558      * that if the device model is ever changed to allow the CMB to be located
559      * in BAR0 as well, then this must be changed.
560      */
561     lo = n->bar0.addr;
562     hi = lo + int128_get64(n->bar0.size);
563 
564     return addr >= lo && addr < hi;
565 }
566 
567 static int nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
568 {
569     hwaddr hi = addr + size - 1;
570     if (hi < addr) {
571         return 1;
572     }
573 
574     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
575         memcpy(buf, nvme_addr_to_cmb(n, addr), size);
576         return 0;
577     }
578 
579     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
580         memcpy(buf, nvme_addr_to_pmr(n, addr), size);
581         return 0;
582     }
583 
584     return pci_dma_read(PCI_DEVICE(n), addr, buf, size);
585 }
586 
587 static int nvme_addr_write(NvmeCtrl *n, hwaddr addr, const void *buf, int size)
588 {
589     hwaddr hi = addr + size - 1;
590     if (hi < addr) {
591         return 1;
592     }
593 
594     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
595         memcpy(nvme_addr_to_cmb(n, addr), buf, size);
596         return 0;
597     }
598 
599     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
600         memcpy(nvme_addr_to_pmr(n, addr), buf, size);
601         return 0;
602     }
603 
604     return pci_dma_write(PCI_DEVICE(n), addr, buf, size);
605 }
606 
607 static bool nvme_nsid_valid(NvmeCtrl *n, uint32_t nsid)
608 {
609     return nsid &&
610         (nsid == NVME_NSID_BROADCAST || nsid <= NVME_MAX_NAMESPACES);
611 }
612 
613 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
614 {
615     return sqid < n->conf_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
616 }
617 
618 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
619 {
620     return cqid < n->conf_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
621 }
622 
623 static void nvme_inc_cq_tail(NvmeCQueue *cq)
624 {
625     cq->tail++;
626     if (cq->tail >= cq->size) {
627         cq->tail = 0;
628         cq->phase = !cq->phase;
629     }
630 }
631 
632 static void nvme_inc_sq_head(NvmeSQueue *sq)
633 {
634     sq->head = (sq->head + 1) % sq->size;
635 }
636 
637 static uint8_t nvme_cq_full(NvmeCQueue *cq)
638 {
639     return (cq->tail + 1) % cq->size == cq->head;
640 }
641 
642 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
643 {
644     return sq->head == sq->tail;
645 }
646 
647 static void nvme_irq_check(NvmeCtrl *n)
648 {
649     PCIDevice *pci = PCI_DEVICE(n);
650     uint32_t intms = ldl_le_p(&n->bar.intms);
651 
652     if (msix_enabled(pci)) {
653         return;
654     }
655     if (~intms & n->irq_status) {
656         pci_irq_assert(pci);
657     } else {
658         pci_irq_deassert(pci);
659     }
660 }
661 
662 static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
663 {
664     PCIDevice *pci = PCI_DEVICE(n);
665 
666     if (cq->irq_enabled) {
667         if (msix_enabled(pci)) {
668             trace_pci_nvme_irq_msix(cq->vector);
669             msix_notify(pci, cq->vector);
670         } else {
671             trace_pci_nvme_irq_pin();
672             assert(cq->vector < 32);
673             n->irq_status |= 1 << cq->vector;
674             nvme_irq_check(n);
675         }
676     } else {
677         trace_pci_nvme_irq_masked();
678     }
679 }
680 
681 static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
682 {
683     if (cq->irq_enabled) {
684         if (msix_enabled(PCI_DEVICE(n))) {
685             return;
686         } else {
687             assert(cq->vector < 32);
688             if (!n->cq_pending) {
689                 n->irq_status &= ~(1 << cq->vector);
690             }
691             nvme_irq_check(n);
692         }
693     }
694 }
695 
696 static void nvme_req_clear(NvmeRequest *req)
697 {
698     req->ns = NULL;
699     req->opaque = NULL;
700     req->aiocb = NULL;
701     memset(&req->cqe, 0x0, sizeof(req->cqe));
702     req->status = NVME_SUCCESS;
703 }
704 
705 static inline void nvme_sg_init(NvmeCtrl *n, NvmeSg *sg, bool dma)
706 {
707     if (dma) {
708         pci_dma_sglist_init(&sg->qsg, PCI_DEVICE(n), 0);
709         sg->flags = NVME_SG_DMA;
710     } else {
711         qemu_iovec_init(&sg->iov, 0);
712     }
713 
714     sg->flags |= NVME_SG_ALLOC;
715 }
716 
717 static inline void nvme_sg_unmap(NvmeSg *sg)
718 {
719     if (!(sg->flags & NVME_SG_ALLOC)) {
720         return;
721     }
722 
723     if (sg->flags & NVME_SG_DMA) {
724         qemu_sglist_destroy(&sg->qsg);
725     } else {
726         qemu_iovec_destroy(&sg->iov);
727     }
728 
729     memset(sg, 0x0, sizeof(*sg));
730 }
731 
732 /*
733  * When metadata is transferred as extended LBAs, the DPTR mapped into `sg`
734  * holds both data and metadata. This function splits the data and metadata
735  * into two separate QSG/IOVs.
736  */
737 static void nvme_sg_split(NvmeSg *sg, NvmeNamespace *ns, NvmeSg *data,
738                           NvmeSg *mdata)
739 {
740     NvmeSg *dst = data;
741     uint32_t trans_len, count = ns->lbasz;
742     uint64_t offset = 0;
743     bool dma = sg->flags & NVME_SG_DMA;
744     size_t sge_len;
745     size_t sg_len = dma ? sg->qsg.size : sg->iov.size;
746     int sg_idx = 0;
747 
748     assert(sg->flags & NVME_SG_ALLOC);
749 
750     while (sg_len) {
751         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
752 
753         trans_len = MIN(sg_len, count);
754         trans_len = MIN(trans_len, sge_len - offset);
755 
756         if (dst) {
757             if (dma) {
758                 qemu_sglist_add(&dst->qsg, sg->qsg.sg[sg_idx].base + offset,
759                                 trans_len);
760             } else {
761                 qemu_iovec_add(&dst->iov,
762                                sg->iov.iov[sg_idx].iov_base + offset,
763                                trans_len);
764             }
765         }
766 
767         sg_len -= trans_len;
768         count -= trans_len;
769         offset += trans_len;
770 
771         if (count == 0) {
772             dst = (dst == data) ? mdata : data;
773             count = (dst == data) ? ns->lbasz : ns->lbaf.ms;
774         }
775 
776         if (sge_len == offset) {
777             offset = 0;
778             sg_idx++;
779         }
780     }
781 }
782 
783 static uint16_t nvme_map_addr_cmb(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
784                                   size_t len)
785 {
786     if (!len) {
787         return NVME_SUCCESS;
788     }
789 
790     trace_pci_nvme_map_addr_cmb(addr, len);
791 
792     if (!nvme_addr_is_cmb(n, addr) || !nvme_addr_is_cmb(n, addr + len - 1)) {
793         return NVME_DATA_TRAS_ERROR;
794     }
795 
796     qemu_iovec_add(iov, nvme_addr_to_cmb(n, addr), len);
797 
798     return NVME_SUCCESS;
799 }
800 
801 static uint16_t nvme_map_addr_pmr(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
802                                   size_t len)
803 {
804     if (!len) {
805         return NVME_SUCCESS;
806     }
807 
808     if (!nvme_addr_is_pmr(n, addr) || !nvme_addr_is_pmr(n, addr + len - 1)) {
809         return NVME_DATA_TRAS_ERROR;
810     }
811 
812     qemu_iovec_add(iov, nvme_addr_to_pmr(n, addr), len);
813 
814     return NVME_SUCCESS;
815 }
816 
817 static uint16_t nvme_map_addr(NvmeCtrl *n, NvmeSg *sg, hwaddr addr, size_t len)
818 {
819     bool cmb = false, pmr = false;
820 
821     if (!len) {
822         return NVME_SUCCESS;
823     }
824 
825     trace_pci_nvme_map_addr(addr, len);
826 
827     if (nvme_addr_is_iomem(n, addr)) {
828         return NVME_DATA_TRAS_ERROR;
829     }
830 
831     if (nvme_addr_is_cmb(n, addr)) {
832         cmb = true;
833     } else if (nvme_addr_is_pmr(n, addr)) {
834         pmr = true;
835     }
836 
837     if (cmb || pmr) {
838         if (sg->flags & NVME_SG_DMA) {
839             return NVME_INVALID_USE_OF_CMB | NVME_DNR;
840         }
841 
842         if (sg->iov.niov + 1 > IOV_MAX) {
843             goto max_mappings_exceeded;
844         }
845 
846         if (cmb) {
847             return nvme_map_addr_cmb(n, &sg->iov, addr, len);
848         } else {
849             return nvme_map_addr_pmr(n, &sg->iov, addr, len);
850         }
851     }
852 
853     if (!(sg->flags & NVME_SG_DMA)) {
854         return NVME_INVALID_USE_OF_CMB | NVME_DNR;
855     }
856 
857     if (sg->qsg.nsg + 1 > IOV_MAX) {
858         goto max_mappings_exceeded;
859     }
860 
861     qemu_sglist_add(&sg->qsg, addr, len);
862 
863     return NVME_SUCCESS;
864 
865 max_mappings_exceeded:
866     NVME_GUEST_ERR(pci_nvme_ub_too_many_mappings,
867                    "number of mappings exceed 1024");
868     return NVME_INTERNAL_DEV_ERROR | NVME_DNR;
869 }
870 
871 static inline bool nvme_addr_is_dma(NvmeCtrl *n, hwaddr addr)
872 {
873     return !(nvme_addr_is_cmb(n, addr) || nvme_addr_is_pmr(n, addr));
874 }
875 
876 static uint16_t nvme_map_prp(NvmeCtrl *n, NvmeSg *sg, uint64_t prp1,
877                              uint64_t prp2, uint32_t len)
878 {
879     hwaddr trans_len = n->page_size - (prp1 % n->page_size);
880     trans_len = MIN(len, trans_len);
881     int num_prps = (len >> n->page_bits) + 1;
882     uint16_t status;
883     int ret;
884 
885     trace_pci_nvme_map_prp(trans_len, len, prp1, prp2, num_prps);
886 
887     nvme_sg_init(n, sg, nvme_addr_is_dma(n, prp1));
888 
889     status = nvme_map_addr(n, sg, prp1, trans_len);
890     if (status) {
891         goto unmap;
892     }
893 
894     len -= trans_len;
895     if (len) {
896         if (len > n->page_size) {
897             g_autofree uint64_t *prp_list = g_new(uint64_t, n->max_prp_ents);
898             uint32_t nents, prp_trans;
899             int i = 0;
900 
901             /*
902              * The first PRP list entry, pointed to by PRP2 may contain offset.
903              * Hence, we need to calculate the number of entries in based on
904              * that offset.
905              */
906             nents = (n->page_size - (prp2 & (n->page_size - 1))) >> 3;
907             prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
908             ret = nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
909             if (ret) {
910                 trace_pci_nvme_err_addr_read(prp2);
911                 status = NVME_DATA_TRAS_ERROR;
912                 goto unmap;
913             }
914             while (len != 0) {
915                 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
916 
917                 if (i == nents - 1 && len > n->page_size) {
918                     if (unlikely(prp_ent & (n->page_size - 1))) {
919                         trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
920                         status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
921                         goto unmap;
922                     }
923 
924                     i = 0;
925                     nents = (len + n->page_size - 1) >> n->page_bits;
926                     nents = MIN(nents, n->max_prp_ents);
927                     prp_trans = nents * sizeof(uint64_t);
928                     ret = nvme_addr_read(n, prp_ent, (void *)prp_list,
929                                          prp_trans);
930                     if (ret) {
931                         trace_pci_nvme_err_addr_read(prp_ent);
932                         status = NVME_DATA_TRAS_ERROR;
933                         goto unmap;
934                     }
935                     prp_ent = le64_to_cpu(prp_list[i]);
936                 }
937 
938                 if (unlikely(prp_ent & (n->page_size - 1))) {
939                     trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
940                     status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
941                     goto unmap;
942                 }
943 
944                 trans_len = MIN(len, n->page_size);
945                 status = nvme_map_addr(n, sg, prp_ent, trans_len);
946                 if (status) {
947                     goto unmap;
948                 }
949 
950                 len -= trans_len;
951                 i++;
952             }
953         } else {
954             if (unlikely(prp2 & (n->page_size - 1))) {
955                 trace_pci_nvme_err_invalid_prp2_align(prp2);
956                 status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
957                 goto unmap;
958             }
959             status = nvme_map_addr(n, sg, prp2, len);
960             if (status) {
961                 goto unmap;
962             }
963         }
964     }
965 
966     return NVME_SUCCESS;
967 
968 unmap:
969     nvme_sg_unmap(sg);
970     return status;
971 }
972 
973 /*
974  * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
975  * number of bytes mapped in len.
976  */
977 static uint16_t nvme_map_sgl_data(NvmeCtrl *n, NvmeSg *sg,
978                                   NvmeSglDescriptor *segment, uint64_t nsgld,
979                                   size_t *len, NvmeCmd *cmd)
980 {
981     dma_addr_t addr, trans_len;
982     uint32_t dlen;
983     uint16_t status;
984 
985     for (int i = 0; i < nsgld; i++) {
986         uint8_t type = NVME_SGL_TYPE(segment[i].type);
987 
988         switch (type) {
989         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
990             break;
991         case NVME_SGL_DESCR_TYPE_SEGMENT:
992         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
993             return NVME_INVALID_NUM_SGL_DESCRS | NVME_DNR;
994         default:
995             return NVME_SGL_DESCR_TYPE_INVALID | NVME_DNR;
996         }
997 
998         dlen = le32_to_cpu(segment[i].len);
999 
1000         if (!dlen) {
1001             continue;
1002         }
1003 
1004         if (*len == 0) {
1005             /*
1006              * All data has been mapped, but the SGL contains additional
1007              * segments and/or descriptors. The controller might accept
1008              * ignoring the rest of the SGL.
1009              */
1010             uint32_t sgls = le32_to_cpu(n->id_ctrl.sgls);
1011             if (sgls & NVME_CTRL_SGLS_EXCESS_LENGTH) {
1012                 break;
1013             }
1014 
1015             trace_pci_nvme_err_invalid_sgl_excess_length(dlen);
1016             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1017         }
1018 
1019         trans_len = MIN(*len, dlen);
1020 
1021         addr = le64_to_cpu(segment[i].addr);
1022 
1023         if (UINT64_MAX - addr < dlen) {
1024             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1025         }
1026 
1027         status = nvme_map_addr(n, sg, addr, trans_len);
1028         if (status) {
1029             return status;
1030         }
1031 
1032         *len -= trans_len;
1033     }
1034 
1035     return NVME_SUCCESS;
1036 }
1037 
1038 static uint16_t nvme_map_sgl(NvmeCtrl *n, NvmeSg *sg, NvmeSglDescriptor sgl,
1039                              size_t len, NvmeCmd *cmd)
1040 {
1041     /*
1042      * Read the segment in chunks of 256 descriptors (one 4k page) to avoid
1043      * dynamically allocating a potentially huge SGL. The spec allows the SGL
1044      * to be larger (as in number of bytes required to describe the SGL
1045      * descriptors and segment chain) than the command transfer size, so it is
1046      * not bounded by MDTS.
1047      */
1048 #define SEG_CHUNK_SIZE 256
1049 
1050     NvmeSglDescriptor segment[SEG_CHUNK_SIZE], *sgld, *last_sgld;
1051     uint64_t nsgld;
1052     uint32_t seg_len;
1053     uint16_t status;
1054     hwaddr addr;
1055     int ret;
1056 
1057     sgld = &sgl;
1058     addr = le64_to_cpu(sgl.addr);
1059 
1060     trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl.type), len);
1061 
1062     nvme_sg_init(n, sg, nvme_addr_is_dma(n, addr));
1063 
1064     /*
1065      * If the entire transfer can be described with a single data block it can
1066      * be mapped directly.
1067      */
1068     if (NVME_SGL_TYPE(sgl.type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1069         status = nvme_map_sgl_data(n, sg, sgld, 1, &len, cmd);
1070         if (status) {
1071             goto unmap;
1072         }
1073 
1074         goto out;
1075     }
1076 
1077     for (;;) {
1078         switch (NVME_SGL_TYPE(sgld->type)) {
1079         case NVME_SGL_DESCR_TYPE_SEGMENT:
1080         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
1081             break;
1082         default:
1083             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1084         }
1085 
1086         seg_len = le32_to_cpu(sgld->len);
1087 
1088         /* check the length of the (Last) Segment descriptor */
1089         if (!seg_len || seg_len & 0xf) {
1090             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1091         }
1092 
1093         if (UINT64_MAX - addr < seg_len) {
1094             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1095         }
1096 
1097         nsgld = seg_len / sizeof(NvmeSglDescriptor);
1098 
1099         while (nsgld > SEG_CHUNK_SIZE) {
1100             if (nvme_addr_read(n, addr, segment, sizeof(segment))) {
1101                 trace_pci_nvme_err_addr_read(addr);
1102                 status = NVME_DATA_TRAS_ERROR;
1103                 goto unmap;
1104             }
1105 
1106             status = nvme_map_sgl_data(n, sg, segment, SEG_CHUNK_SIZE,
1107                                        &len, cmd);
1108             if (status) {
1109                 goto unmap;
1110             }
1111 
1112             nsgld -= SEG_CHUNK_SIZE;
1113             addr += SEG_CHUNK_SIZE * sizeof(NvmeSglDescriptor);
1114         }
1115 
1116         ret = nvme_addr_read(n, addr, segment, nsgld *
1117                              sizeof(NvmeSglDescriptor));
1118         if (ret) {
1119             trace_pci_nvme_err_addr_read(addr);
1120             status = NVME_DATA_TRAS_ERROR;
1121             goto unmap;
1122         }
1123 
1124         last_sgld = &segment[nsgld - 1];
1125 
1126         /*
1127          * If the segment ends with a Data Block, then we are done.
1128          */
1129         if (NVME_SGL_TYPE(last_sgld->type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1130             status = nvme_map_sgl_data(n, sg, segment, nsgld, &len, cmd);
1131             if (status) {
1132                 goto unmap;
1133             }
1134 
1135             goto out;
1136         }
1137 
1138         /*
1139          * If the last descriptor was not a Data Block, then the current
1140          * segment must not be a Last Segment.
1141          */
1142         if (NVME_SGL_TYPE(sgld->type) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT) {
1143             status = NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1144             goto unmap;
1145         }
1146 
1147         sgld = last_sgld;
1148         addr = le64_to_cpu(sgld->addr);
1149 
1150         /*
1151          * Do not map the last descriptor; it will be a Segment or Last Segment
1152          * descriptor and is handled by the next iteration.
1153          */
1154         status = nvme_map_sgl_data(n, sg, segment, nsgld - 1, &len, cmd);
1155         if (status) {
1156             goto unmap;
1157         }
1158     }
1159 
1160 out:
1161     /* if there is any residual left in len, the SGL was too short */
1162     if (len) {
1163         status = NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1164         goto unmap;
1165     }
1166 
1167     return NVME_SUCCESS;
1168 
1169 unmap:
1170     nvme_sg_unmap(sg);
1171     return status;
1172 }
1173 
1174 uint16_t nvme_map_dptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1175                        NvmeCmd *cmd)
1176 {
1177     uint64_t prp1, prp2;
1178 
1179     switch (NVME_CMD_FLAGS_PSDT(cmd->flags)) {
1180     case NVME_PSDT_PRP:
1181         prp1 = le64_to_cpu(cmd->dptr.prp1);
1182         prp2 = le64_to_cpu(cmd->dptr.prp2);
1183 
1184         return nvme_map_prp(n, sg, prp1, prp2, len);
1185     case NVME_PSDT_SGL_MPTR_CONTIGUOUS:
1186     case NVME_PSDT_SGL_MPTR_SGL:
1187         return nvme_map_sgl(n, sg, cmd->dptr.sgl, len, cmd);
1188     default:
1189         return NVME_INVALID_FIELD;
1190     }
1191 }
1192 
1193 static uint16_t nvme_map_mptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1194                               NvmeCmd *cmd)
1195 {
1196     int psdt = NVME_CMD_FLAGS_PSDT(cmd->flags);
1197     hwaddr mptr = le64_to_cpu(cmd->mptr);
1198     uint16_t status;
1199 
1200     if (psdt == NVME_PSDT_SGL_MPTR_SGL) {
1201         NvmeSglDescriptor sgl;
1202 
1203         if (nvme_addr_read(n, mptr, &sgl, sizeof(sgl))) {
1204             return NVME_DATA_TRAS_ERROR;
1205         }
1206 
1207         status = nvme_map_sgl(n, sg, sgl, len, cmd);
1208         if (status && (status & 0x7ff) == NVME_DATA_SGL_LEN_INVALID) {
1209             status = NVME_MD_SGL_LEN_INVALID | NVME_DNR;
1210         }
1211 
1212         return status;
1213     }
1214 
1215     nvme_sg_init(n, sg, nvme_addr_is_dma(n, mptr));
1216     status = nvme_map_addr(n, sg, mptr, len);
1217     if (status) {
1218         nvme_sg_unmap(sg);
1219     }
1220 
1221     return status;
1222 }
1223 
1224 static uint16_t nvme_map_data(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1225 {
1226     NvmeNamespace *ns = req->ns;
1227     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1228     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1229     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1230     size_t len = nvme_l2b(ns, nlb);
1231     uint16_t status;
1232 
1233     if (nvme_ns_ext(ns) &&
1234         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1235         NvmeSg sg;
1236 
1237         len += nvme_m2b(ns, nlb);
1238 
1239         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1240         if (status) {
1241             return status;
1242         }
1243 
1244         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1245         nvme_sg_split(&sg, ns, &req->sg, NULL);
1246         nvme_sg_unmap(&sg);
1247 
1248         return NVME_SUCCESS;
1249     }
1250 
1251     return nvme_map_dptr(n, &req->sg, len, &req->cmd);
1252 }
1253 
1254 static uint16_t nvme_map_mdata(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1255 {
1256     NvmeNamespace *ns = req->ns;
1257     size_t len = nvme_m2b(ns, nlb);
1258     uint16_t status;
1259 
1260     if (nvme_ns_ext(ns)) {
1261         NvmeSg sg;
1262 
1263         len += nvme_l2b(ns, nlb);
1264 
1265         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1266         if (status) {
1267             return status;
1268         }
1269 
1270         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1271         nvme_sg_split(&sg, ns, NULL, &req->sg);
1272         nvme_sg_unmap(&sg);
1273 
1274         return NVME_SUCCESS;
1275     }
1276 
1277     return nvme_map_mptr(n, &req->sg, len, &req->cmd);
1278 }
1279 
1280 static uint16_t nvme_tx_interleaved(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr,
1281                                     uint32_t len, uint32_t bytes,
1282                                     int32_t skip_bytes, int64_t offset,
1283                                     NvmeTxDirection dir)
1284 {
1285     hwaddr addr;
1286     uint32_t trans_len, count = bytes;
1287     bool dma = sg->flags & NVME_SG_DMA;
1288     int64_t sge_len;
1289     int sg_idx = 0;
1290     int ret;
1291 
1292     assert(sg->flags & NVME_SG_ALLOC);
1293 
1294     while (len) {
1295         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
1296 
1297         if (sge_len - offset < 0) {
1298             offset -= sge_len;
1299             sg_idx++;
1300             continue;
1301         }
1302 
1303         if (sge_len == offset) {
1304             offset = 0;
1305             sg_idx++;
1306             continue;
1307         }
1308 
1309         trans_len = MIN(len, count);
1310         trans_len = MIN(trans_len, sge_len - offset);
1311 
1312         if (dma) {
1313             addr = sg->qsg.sg[sg_idx].base + offset;
1314         } else {
1315             addr = (hwaddr)(uintptr_t)sg->iov.iov[sg_idx].iov_base + offset;
1316         }
1317 
1318         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1319             ret = nvme_addr_read(n, addr, ptr, trans_len);
1320         } else {
1321             ret = nvme_addr_write(n, addr, ptr, trans_len);
1322         }
1323 
1324         if (ret) {
1325             return NVME_DATA_TRAS_ERROR;
1326         }
1327 
1328         ptr += trans_len;
1329         len -= trans_len;
1330         count -= trans_len;
1331         offset += trans_len;
1332 
1333         if (count == 0) {
1334             count = bytes;
1335             offset += skip_bytes;
1336         }
1337     }
1338 
1339     return NVME_SUCCESS;
1340 }
1341 
1342 static uint16_t nvme_tx(NvmeCtrl *n, NvmeSg *sg, void *ptr, uint32_t len,
1343                         NvmeTxDirection dir)
1344 {
1345     assert(sg->flags & NVME_SG_ALLOC);
1346 
1347     if (sg->flags & NVME_SG_DMA) {
1348         const MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
1349         dma_addr_t residual;
1350 
1351         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1352             dma_buf_write(ptr, len, &residual, &sg->qsg, attrs);
1353         } else {
1354             dma_buf_read(ptr, len, &residual, &sg->qsg, attrs);
1355         }
1356 
1357         if (unlikely(residual)) {
1358             trace_pci_nvme_err_invalid_dma();
1359             return NVME_INVALID_FIELD | NVME_DNR;
1360         }
1361     } else {
1362         size_t bytes;
1363 
1364         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1365             bytes = qemu_iovec_to_buf(&sg->iov, 0, ptr, len);
1366         } else {
1367             bytes = qemu_iovec_from_buf(&sg->iov, 0, ptr, len);
1368         }
1369 
1370         if (unlikely(bytes != len)) {
1371             trace_pci_nvme_err_invalid_dma();
1372             return NVME_INVALID_FIELD | NVME_DNR;
1373         }
1374     }
1375 
1376     return NVME_SUCCESS;
1377 }
1378 
1379 static inline uint16_t nvme_c2h(NvmeCtrl *n, void *ptr, uint32_t len,
1380                                 NvmeRequest *req)
1381 {
1382     uint16_t status;
1383 
1384     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1385     if (status) {
1386         return status;
1387     }
1388 
1389     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_FROM_DEVICE);
1390 }
1391 
1392 static inline uint16_t nvme_h2c(NvmeCtrl *n, void *ptr, uint32_t len,
1393                                 NvmeRequest *req)
1394 {
1395     uint16_t status;
1396 
1397     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1398     if (status) {
1399         return status;
1400     }
1401 
1402     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_TO_DEVICE);
1403 }
1404 
1405 uint16_t nvme_bounce_data(NvmeCtrl *n, void *ptr, uint32_t len,
1406                           NvmeTxDirection dir, NvmeRequest *req)
1407 {
1408     NvmeNamespace *ns = req->ns;
1409     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1410     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1411     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1412 
1413     if (nvme_ns_ext(ns) &&
1414         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1415         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbasz,
1416                                    ns->lbaf.ms, 0, dir);
1417     }
1418 
1419     return nvme_tx(n, &req->sg, ptr, len, dir);
1420 }
1421 
1422 uint16_t nvme_bounce_mdata(NvmeCtrl *n, void *ptr, uint32_t len,
1423                            NvmeTxDirection dir, NvmeRequest *req)
1424 {
1425     NvmeNamespace *ns = req->ns;
1426     uint16_t status;
1427 
1428     if (nvme_ns_ext(ns)) {
1429         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbaf.ms,
1430                                    ns->lbasz, ns->lbasz, dir);
1431     }
1432 
1433     nvme_sg_unmap(&req->sg);
1434 
1435     status = nvme_map_mptr(n, &req->sg, len, &req->cmd);
1436     if (status) {
1437         return status;
1438     }
1439 
1440     return nvme_tx(n, &req->sg, ptr, len, dir);
1441 }
1442 
1443 static inline void nvme_blk_read(BlockBackend *blk, int64_t offset,
1444                                  uint32_t align, BlockCompletionFunc *cb,
1445                                  NvmeRequest *req)
1446 {
1447     assert(req->sg.flags & NVME_SG_ALLOC);
1448 
1449     if (req->sg.flags & NVME_SG_DMA) {
1450         req->aiocb = dma_blk_read(blk, &req->sg.qsg, offset, align, cb, req);
1451     } else {
1452         req->aiocb = blk_aio_preadv(blk, offset, &req->sg.iov, 0, cb, req);
1453     }
1454 }
1455 
1456 static inline void nvme_blk_write(BlockBackend *blk, int64_t offset,
1457                                   uint32_t align, BlockCompletionFunc *cb,
1458                                   NvmeRequest *req)
1459 {
1460     assert(req->sg.flags & NVME_SG_ALLOC);
1461 
1462     if (req->sg.flags & NVME_SG_DMA) {
1463         req->aiocb = dma_blk_write(blk, &req->sg.qsg, offset, align, cb, req);
1464     } else {
1465         req->aiocb = blk_aio_pwritev(blk, offset, &req->sg.iov, 0, cb, req);
1466     }
1467 }
1468 
1469 static void nvme_update_cq_eventidx(const NvmeCQueue *cq)
1470 {
1471     trace_pci_nvme_update_cq_eventidx(cq->cqid, cq->head);
1472 
1473     stl_le_pci_dma(PCI_DEVICE(cq->ctrl), cq->ei_addr, cq->head,
1474                    MEMTXATTRS_UNSPECIFIED);
1475 }
1476 
1477 static void nvme_update_cq_head(NvmeCQueue *cq)
1478 {
1479     ldl_le_pci_dma(PCI_DEVICE(cq->ctrl), cq->db_addr, &cq->head,
1480                    MEMTXATTRS_UNSPECIFIED);
1481 
1482     trace_pci_nvme_update_cq_head(cq->cqid, cq->head);
1483 }
1484 
1485 static void nvme_post_cqes(void *opaque)
1486 {
1487     NvmeCQueue *cq = opaque;
1488     NvmeCtrl *n = cq->ctrl;
1489     NvmeRequest *req, *next;
1490     bool pending = cq->head != cq->tail;
1491     int ret;
1492 
1493     QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
1494         NvmeSQueue *sq;
1495         hwaddr addr;
1496 
1497         if (n->dbbuf_enabled) {
1498             nvme_update_cq_eventidx(cq);
1499             nvme_update_cq_head(cq);
1500         }
1501 
1502         if (nvme_cq_full(cq)) {
1503             break;
1504         }
1505 
1506         sq = req->sq;
1507         req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
1508         req->cqe.sq_id = cpu_to_le16(sq->sqid);
1509         req->cqe.sq_head = cpu_to_le16(sq->head);
1510         addr = cq->dma_addr + (cq->tail << NVME_CQES);
1511         ret = pci_dma_write(PCI_DEVICE(n), addr, (void *)&req->cqe,
1512                             sizeof(req->cqe));
1513         if (ret) {
1514             trace_pci_nvme_err_addr_write(addr);
1515             trace_pci_nvme_err_cfs();
1516             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
1517             break;
1518         }
1519         QTAILQ_REMOVE(&cq->req_list, req, entry);
1520         nvme_inc_cq_tail(cq);
1521         nvme_sg_unmap(&req->sg);
1522         QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
1523     }
1524     if (cq->tail != cq->head) {
1525         if (cq->irq_enabled && !pending) {
1526             n->cq_pending++;
1527         }
1528 
1529         nvme_irq_assert(n, cq);
1530     }
1531 }
1532 
1533 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
1534 {
1535     assert(cq->cqid == req->sq->cqid);
1536     trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
1537                                           le32_to_cpu(req->cqe.result),
1538                                           le32_to_cpu(req->cqe.dw1),
1539                                           req->status);
1540 
1541     if (req->status) {
1542         trace_pci_nvme_err_req_status(nvme_cid(req), nvme_nsid(req->ns),
1543                                       req->status, req->cmd.opcode);
1544     }
1545 
1546     QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
1547     QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
1548 
1549     qemu_bh_schedule(cq->bh);
1550 }
1551 
1552 static void nvme_process_aers(void *opaque)
1553 {
1554     NvmeCtrl *n = opaque;
1555     NvmeAsyncEvent *event, *next;
1556 
1557     trace_pci_nvme_process_aers(n->aer_queued);
1558 
1559     QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
1560         NvmeRequest *req;
1561         NvmeAerResult *result;
1562 
1563         /* can't post cqe if there is nothing to complete */
1564         if (!n->outstanding_aers) {
1565             trace_pci_nvme_no_outstanding_aers();
1566             break;
1567         }
1568 
1569         /* ignore if masked (cqe posted, but event not cleared) */
1570         if (n->aer_mask & (1 << event->result.event_type)) {
1571             trace_pci_nvme_aer_masked(event->result.event_type, n->aer_mask);
1572             continue;
1573         }
1574 
1575         QTAILQ_REMOVE(&n->aer_queue, event, entry);
1576         n->aer_queued--;
1577 
1578         n->aer_mask |= 1 << event->result.event_type;
1579         n->outstanding_aers--;
1580 
1581         req = n->aer_reqs[n->outstanding_aers];
1582 
1583         result = (NvmeAerResult *) &req->cqe.result;
1584         result->event_type = event->result.event_type;
1585         result->event_info = event->result.event_info;
1586         result->log_page = event->result.log_page;
1587         g_free(event);
1588 
1589         trace_pci_nvme_aer_post_cqe(result->event_type, result->event_info,
1590                                     result->log_page);
1591 
1592         nvme_enqueue_req_completion(&n->admin_cq, req);
1593     }
1594 }
1595 
1596 static void nvme_enqueue_event(NvmeCtrl *n, uint8_t event_type,
1597                                uint8_t event_info, uint8_t log_page)
1598 {
1599     NvmeAsyncEvent *event;
1600 
1601     trace_pci_nvme_enqueue_event(event_type, event_info, log_page);
1602 
1603     if (n->aer_queued == n->params.aer_max_queued) {
1604         trace_pci_nvme_enqueue_event_noqueue(n->aer_queued);
1605         return;
1606     }
1607 
1608     event = g_new(NvmeAsyncEvent, 1);
1609     event->result = (NvmeAerResult) {
1610         .event_type = event_type,
1611         .event_info = event_info,
1612         .log_page   = log_page,
1613     };
1614 
1615     QTAILQ_INSERT_TAIL(&n->aer_queue, event, entry);
1616     n->aer_queued++;
1617 
1618     nvme_process_aers(n);
1619 }
1620 
1621 static void nvme_smart_event(NvmeCtrl *n, uint8_t event)
1622 {
1623     uint8_t aer_info;
1624 
1625     /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
1626     if (!(NVME_AEC_SMART(n->features.async_config) & event)) {
1627         return;
1628     }
1629 
1630     switch (event) {
1631     case NVME_SMART_SPARE:
1632         aer_info = NVME_AER_INFO_SMART_SPARE_THRESH;
1633         break;
1634     case NVME_SMART_TEMPERATURE:
1635         aer_info = NVME_AER_INFO_SMART_TEMP_THRESH;
1636         break;
1637     case NVME_SMART_RELIABILITY:
1638     case NVME_SMART_MEDIA_READ_ONLY:
1639     case NVME_SMART_FAILED_VOLATILE_MEDIA:
1640     case NVME_SMART_PMR_UNRELIABLE:
1641         aer_info = NVME_AER_INFO_SMART_RELIABILITY;
1642         break;
1643     default:
1644         return;
1645     }
1646 
1647     nvme_enqueue_event(n, NVME_AER_TYPE_SMART, aer_info, NVME_LOG_SMART_INFO);
1648 }
1649 
1650 static void nvme_clear_events(NvmeCtrl *n, uint8_t event_type)
1651 {
1652     n->aer_mask &= ~(1 << event_type);
1653     if (!QTAILQ_EMPTY(&n->aer_queue)) {
1654         nvme_process_aers(n);
1655     }
1656 }
1657 
1658 static inline uint16_t nvme_check_mdts(NvmeCtrl *n, size_t len)
1659 {
1660     uint8_t mdts = n->params.mdts;
1661 
1662     if (mdts && len > n->page_size << mdts) {
1663         trace_pci_nvme_err_mdts(len);
1664         return NVME_INVALID_FIELD | NVME_DNR;
1665     }
1666 
1667     return NVME_SUCCESS;
1668 }
1669 
1670 static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
1671                                          uint32_t nlb)
1672 {
1673     uint64_t nsze = le64_to_cpu(ns->id_ns.nsze);
1674 
1675     if (unlikely(UINT64_MAX - slba < nlb || slba + nlb > nsze)) {
1676         trace_pci_nvme_err_invalid_lba_range(slba, nlb, nsze);
1677         return NVME_LBA_RANGE | NVME_DNR;
1678     }
1679 
1680     return NVME_SUCCESS;
1681 }
1682 
1683 static int nvme_block_status_all(NvmeNamespace *ns, uint64_t slba,
1684                                  uint32_t nlb, int flags)
1685 {
1686     BlockDriverState *bs = blk_bs(ns->blkconf.blk);
1687 
1688     int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
1689     int64_t offset = nvme_l2b(ns, slba);
1690     int ret;
1691 
1692     /*
1693      * `pnum` holds the number of bytes after offset that shares the same
1694      * allocation status as the byte at offset. If `pnum` is different from
1695      * `bytes`, we should check the allocation status of the next range and
1696      * continue this until all bytes have been checked.
1697      */
1698     do {
1699         bytes -= pnum;
1700 
1701         ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
1702         if (ret < 0) {
1703             return ret;
1704         }
1705 
1706 
1707         trace_pci_nvme_block_status(offset, bytes, pnum, ret,
1708                                     !!(ret & BDRV_BLOCK_ZERO));
1709 
1710         if (!(ret & flags)) {
1711             return 1;
1712         }
1713 
1714         offset += pnum;
1715     } while (pnum != bytes);
1716 
1717     return 0;
1718 }
1719 
1720 static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
1721                                  uint32_t nlb)
1722 {
1723     int ret;
1724     Error *err = NULL;
1725 
1726     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_DATA);
1727     if (ret) {
1728         if (ret < 0) {
1729             error_setg_errno(&err, -ret, "unable to get block status");
1730             error_report_err(err);
1731 
1732             return NVME_INTERNAL_DEV_ERROR;
1733         }
1734 
1735         return NVME_DULB;
1736     }
1737 
1738     return NVME_SUCCESS;
1739 }
1740 
1741 static void nvme_aio_err(NvmeRequest *req, int ret)
1742 {
1743     uint16_t status = NVME_SUCCESS;
1744     Error *local_err = NULL;
1745 
1746     switch (req->cmd.opcode) {
1747     case NVME_CMD_READ:
1748         status = NVME_UNRECOVERED_READ;
1749         break;
1750     case NVME_CMD_FLUSH:
1751     case NVME_CMD_WRITE:
1752     case NVME_CMD_WRITE_ZEROES:
1753     case NVME_CMD_ZONE_APPEND:
1754     case NVME_CMD_COPY:
1755         status = NVME_WRITE_FAULT;
1756         break;
1757     default:
1758         status = NVME_INTERNAL_DEV_ERROR;
1759         break;
1760     }
1761 
1762     if (ret == -ECANCELED) {
1763         status = NVME_CMD_ABORT_REQ;
1764     }
1765 
1766     trace_pci_nvme_err_aio(nvme_cid(req), strerror(-ret), status);
1767 
1768     error_setg_errno(&local_err, -ret, "aio failed");
1769     error_report_err(local_err);
1770 
1771     /*
1772      * Set the command status code to the first encountered error but allow a
1773      * subsequent Internal Device Error to trump it.
1774      */
1775     if (req->status && status != NVME_INTERNAL_DEV_ERROR) {
1776         return;
1777     }
1778 
1779     req->status = status;
1780 }
1781 
1782 static inline uint32_t nvme_zone_idx(NvmeNamespace *ns, uint64_t slba)
1783 {
1784     return ns->zone_size_log2 > 0 ? slba >> ns->zone_size_log2 :
1785                                     slba / ns->zone_size;
1786 }
1787 
1788 static inline NvmeZone *nvme_get_zone_by_slba(NvmeNamespace *ns, uint64_t slba)
1789 {
1790     uint32_t zone_idx = nvme_zone_idx(ns, slba);
1791 
1792     if (zone_idx >= ns->num_zones) {
1793         return NULL;
1794     }
1795 
1796     return &ns->zone_array[zone_idx];
1797 }
1798 
1799 static uint16_t nvme_check_zone_state_for_write(NvmeZone *zone)
1800 {
1801     uint64_t zslba = zone->d.zslba;
1802 
1803     switch (nvme_get_zone_state(zone)) {
1804     case NVME_ZONE_STATE_EMPTY:
1805     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1806     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1807     case NVME_ZONE_STATE_CLOSED:
1808         return NVME_SUCCESS;
1809     case NVME_ZONE_STATE_FULL:
1810         trace_pci_nvme_err_zone_is_full(zslba);
1811         return NVME_ZONE_FULL;
1812     case NVME_ZONE_STATE_OFFLINE:
1813         trace_pci_nvme_err_zone_is_offline(zslba);
1814         return NVME_ZONE_OFFLINE;
1815     case NVME_ZONE_STATE_READ_ONLY:
1816         trace_pci_nvme_err_zone_is_read_only(zslba);
1817         return NVME_ZONE_READ_ONLY;
1818     default:
1819         assert(false);
1820     }
1821 
1822     return NVME_INTERNAL_DEV_ERROR;
1823 }
1824 
1825 static uint16_t nvme_check_zone_write(NvmeNamespace *ns, NvmeZone *zone,
1826                                       uint64_t slba, uint32_t nlb)
1827 {
1828     uint64_t zcap = nvme_zone_wr_boundary(zone);
1829     uint16_t status;
1830 
1831     status = nvme_check_zone_state_for_write(zone);
1832     if (status) {
1833         return status;
1834     }
1835 
1836     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1837         uint64_t ezrwa = zone->w_ptr + 2 * ns->zns.zrwas;
1838 
1839         if (slba < zone->w_ptr || slba + nlb > ezrwa) {
1840             trace_pci_nvme_err_zone_invalid_write(slba, zone->w_ptr);
1841             return NVME_ZONE_INVALID_WRITE;
1842         }
1843     } else {
1844         if (unlikely(slba != zone->w_ptr)) {
1845             trace_pci_nvme_err_write_not_at_wp(slba, zone->d.zslba,
1846                                                zone->w_ptr);
1847             return NVME_ZONE_INVALID_WRITE;
1848         }
1849     }
1850 
1851     if (unlikely((slba + nlb) > zcap)) {
1852         trace_pci_nvme_err_zone_boundary(slba, nlb, zcap);
1853         return NVME_ZONE_BOUNDARY_ERROR;
1854     }
1855 
1856     return NVME_SUCCESS;
1857 }
1858 
1859 static uint16_t nvme_check_zone_state_for_read(NvmeZone *zone)
1860 {
1861     switch (nvme_get_zone_state(zone)) {
1862     case NVME_ZONE_STATE_EMPTY:
1863     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1864     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1865     case NVME_ZONE_STATE_FULL:
1866     case NVME_ZONE_STATE_CLOSED:
1867     case NVME_ZONE_STATE_READ_ONLY:
1868         return NVME_SUCCESS;
1869     case NVME_ZONE_STATE_OFFLINE:
1870         trace_pci_nvme_err_zone_is_offline(zone->d.zslba);
1871         return NVME_ZONE_OFFLINE;
1872     default:
1873         assert(false);
1874     }
1875 
1876     return NVME_INTERNAL_DEV_ERROR;
1877 }
1878 
1879 static uint16_t nvme_check_zone_read(NvmeNamespace *ns, uint64_t slba,
1880                                      uint32_t nlb)
1881 {
1882     NvmeZone *zone;
1883     uint64_t bndry, end;
1884     uint16_t status;
1885 
1886     zone = nvme_get_zone_by_slba(ns, slba);
1887     assert(zone);
1888 
1889     bndry = nvme_zone_rd_boundary(ns, zone);
1890     end = slba + nlb;
1891 
1892     status = nvme_check_zone_state_for_read(zone);
1893     if (status) {
1894         ;
1895     } else if (unlikely(end > bndry)) {
1896         if (!ns->params.cross_zone_read) {
1897             status = NVME_ZONE_BOUNDARY_ERROR;
1898         } else {
1899             /*
1900              * Read across zone boundary - check that all subsequent
1901              * zones that are being read have an appropriate state.
1902              */
1903             do {
1904                 zone++;
1905                 status = nvme_check_zone_state_for_read(zone);
1906                 if (status) {
1907                     break;
1908                 }
1909             } while (end > nvme_zone_rd_boundary(ns, zone));
1910         }
1911     }
1912 
1913     return status;
1914 }
1915 
1916 static uint16_t nvme_zrm_finish(NvmeNamespace *ns, NvmeZone *zone)
1917 {
1918     switch (nvme_get_zone_state(zone)) {
1919     case NVME_ZONE_STATE_FULL:
1920         return NVME_SUCCESS;
1921 
1922     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1923     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1924         nvme_aor_dec_open(ns);
1925         /* fallthrough */
1926     case NVME_ZONE_STATE_CLOSED:
1927         nvme_aor_dec_active(ns);
1928 
1929         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1930             zone->d.za &= ~NVME_ZA_ZRWA_VALID;
1931             if (ns->params.numzrwa) {
1932                 ns->zns.numzrwa++;
1933             }
1934         }
1935 
1936         /* fallthrough */
1937     case NVME_ZONE_STATE_EMPTY:
1938         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_FULL);
1939         return NVME_SUCCESS;
1940 
1941     default:
1942         return NVME_ZONE_INVAL_TRANSITION;
1943     }
1944 }
1945 
1946 static uint16_t nvme_zrm_close(NvmeNamespace *ns, NvmeZone *zone)
1947 {
1948     switch (nvme_get_zone_state(zone)) {
1949     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1950     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1951         nvme_aor_dec_open(ns);
1952         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
1953         /* fall through */
1954     case NVME_ZONE_STATE_CLOSED:
1955         return NVME_SUCCESS;
1956 
1957     default:
1958         return NVME_ZONE_INVAL_TRANSITION;
1959     }
1960 }
1961 
1962 static uint16_t nvme_zrm_reset(NvmeNamespace *ns, NvmeZone *zone)
1963 {
1964     switch (nvme_get_zone_state(zone)) {
1965     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1966     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1967         nvme_aor_dec_open(ns);
1968         /* fallthrough */
1969     case NVME_ZONE_STATE_CLOSED:
1970         nvme_aor_dec_active(ns);
1971 
1972         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1973             if (ns->params.numzrwa) {
1974                 ns->zns.numzrwa++;
1975             }
1976         }
1977 
1978         /* fallthrough */
1979     case NVME_ZONE_STATE_FULL:
1980         zone->w_ptr = zone->d.zslba;
1981         zone->d.wp = zone->w_ptr;
1982         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
1983         /* fallthrough */
1984     case NVME_ZONE_STATE_EMPTY:
1985         return NVME_SUCCESS;
1986 
1987     default:
1988         return NVME_ZONE_INVAL_TRANSITION;
1989     }
1990 }
1991 
1992 static void nvme_zrm_auto_transition_zone(NvmeNamespace *ns)
1993 {
1994     NvmeZone *zone;
1995 
1996     if (ns->params.max_open_zones &&
1997         ns->nr_open_zones == ns->params.max_open_zones) {
1998         zone = QTAILQ_FIRST(&ns->imp_open_zones);
1999         if (zone) {
2000             /*
2001              * Automatically close this implicitly open zone.
2002              */
2003             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
2004             nvme_zrm_close(ns, zone);
2005         }
2006     }
2007 }
2008 
2009 enum {
2010     NVME_ZRM_AUTO = 1 << 0,
2011     NVME_ZRM_ZRWA = 1 << 1,
2012 };
2013 
2014 static uint16_t nvme_zrm_open_flags(NvmeCtrl *n, NvmeNamespace *ns,
2015                                     NvmeZone *zone, int flags)
2016 {
2017     int act = 0;
2018     uint16_t status;
2019 
2020     switch (nvme_get_zone_state(zone)) {
2021     case NVME_ZONE_STATE_EMPTY:
2022         act = 1;
2023 
2024         /* fallthrough */
2025 
2026     case NVME_ZONE_STATE_CLOSED:
2027         if (n->params.auto_transition_zones) {
2028             nvme_zrm_auto_transition_zone(ns);
2029         }
2030         status = nvme_zns_check_resources(ns, act, 1,
2031                                           (flags & NVME_ZRM_ZRWA) ? 1 : 0);
2032         if (status) {
2033             return status;
2034         }
2035 
2036         if (act) {
2037             nvme_aor_inc_active(ns);
2038         }
2039 
2040         nvme_aor_inc_open(ns);
2041 
2042         if (flags & NVME_ZRM_AUTO) {
2043             nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_IMPLICITLY_OPEN);
2044             return NVME_SUCCESS;
2045         }
2046 
2047         /* fallthrough */
2048 
2049     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
2050         if (flags & NVME_ZRM_AUTO) {
2051             return NVME_SUCCESS;
2052         }
2053 
2054         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EXPLICITLY_OPEN);
2055 
2056         /* fallthrough */
2057 
2058     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
2059         if (flags & NVME_ZRM_ZRWA) {
2060             ns->zns.numzrwa--;
2061 
2062             zone->d.za |= NVME_ZA_ZRWA_VALID;
2063         }
2064 
2065         return NVME_SUCCESS;
2066 
2067     default:
2068         return NVME_ZONE_INVAL_TRANSITION;
2069     }
2070 }
2071 
2072 static inline uint16_t nvme_zrm_auto(NvmeCtrl *n, NvmeNamespace *ns,
2073                                      NvmeZone *zone)
2074 {
2075     return nvme_zrm_open_flags(n, ns, zone, NVME_ZRM_AUTO);
2076 }
2077 
2078 static void nvme_advance_zone_wp(NvmeNamespace *ns, NvmeZone *zone,
2079                                  uint32_t nlb)
2080 {
2081     zone->d.wp += nlb;
2082 
2083     if (zone->d.wp == nvme_zone_wr_boundary(zone)) {
2084         nvme_zrm_finish(ns, zone);
2085     }
2086 }
2087 
2088 static void nvme_zoned_zrwa_implicit_flush(NvmeNamespace *ns, NvmeZone *zone,
2089                                            uint32_t nlbc)
2090 {
2091     uint16_t nzrwafgs = DIV_ROUND_UP(nlbc, ns->zns.zrwafg);
2092 
2093     nlbc = nzrwafgs * ns->zns.zrwafg;
2094 
2095     trace_pci_nvme_zoned_zrwa_implicit_flush(zone->d.zslba, nlbc);
2096 
2097     zone->w_ptr += nlbc;
2098 
2099     nvme_advance_zone_wp(ns, zone, nlbc);
2100 }
2101 
2102 static void nvme_finalize_zoned_write(NvmeNamespace *ns, NvmeRequest *req)
2103 {
2104     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2105     NvmeZone *zone;
2106     uint64_t slba;
2107     uint32_t nlb;
2108 
2109     slba = le64_to_cpu(rw->slba);
2110     nlb = le16_to_cpu(rw->nlb) + 1;
2111     zone = nvme_get_zone_by_slba(ns, slba);
2112     assert(zone);
2113 
2114     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
2115         uint64_t ezrwa = zone->w_ptr + ns->zns.zrwas - 1;
2116         uint64_t elba = slba + nlb - 1;
2117 
2118         if (elba > ezrwa) {
2119             nvme_zoned_zrwa_implicit_flush(ns, zone, elba - ezrwa);
2120         }
2121 
2122         return;
2123     }
2124 
2125     nvme_advance_zone_wp(ns, zone, nlb);
2126 }
2127 
2128 static inline bool nvme_is_write(NvmeRequest *req)
2129 {
2130     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2131 
2132     return rw->opcode == NVME_CMD_WRITE ||
2133            rw->opcode == NVME_CMD_ZONE_APPEND ||
2134            rw->opcode == NVME_CMD_WRITE_ZEROES;
2135 }
2136 
2137 static void nvme_misc_cb(void *opaque, int ret)
2138 {
2139     NvmeRequest *req = opaque;
2140 
2141     trace_pci_nvme_misc_cb(nvme_cid(req));
2142 
2143     if (ret) {
2144         nvme_aio_err(req, ret);
2145     }
2146 
2147     nvme_enqueue_req_completion(nvme_cq(req), req);
2148 }
2149 
2150 void nvme_rw_complete_cb(void *opaque, int ret)
2151 {
2152     NvmeRequest *req = opaque;
2153     NvmeNamespace *ns = req->ns;
2154     BlockBackend *blk = ns->blkconf.blk;
2155     BlockAcctCookie *acct = &req->acct;
2156     BlockAcctStats *stats = blk_get_stats(blk);
2157 
2158     trace_pci_nvme_rw_complete_cb(nvme_cid(req), blk_name(blk));
2159 
2160     if (ret) {
2161         block_acct_failed(stats, acct);
2162         nvme_aio_err(req, ret);
2163     } else {
2164         block_acct_done(stats, acct);
2165     }
2166 
2167     if (ns->params.zoned && nvme_is_write(req)) {
2168         nvme_finalize_zoned_write(ns, req);
2169     }
2170 
2171     nvme_enqueue_req_completion(nvme_cq(req), req);
2172 }
2173 
2174 static void nvme_rw_cb(void *opaque, int ret)
2175 {
2176     NvmeRequest *req = opaque;
2177     NvmeNamespace *ns = req->ns;
2178 
2179     BlockBackend *blk = ns->blkconf.blk;
2180 
2181     trace_pci_nvme_rw_cb(nvme_cid(req), blk_name(blk));
2182 
2183     if (ret) {
2184         goto out;
2185     }
2186 
2187     if (ns->lbaf.ms) {
2188         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2189         uint64_t slba = le64_to_cpu(rw->slba);
2190         uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
2191         uint64_t offset = nvme_moff(ns, slba);
2192 
2193         if (req->cmd.opcode == NVME_CMD_WRITE_ZEROES) {
2194             size_t mlen = nvme_m2b(ns, nlb);
2195 
2196             req->aiocb = blk_aio_pwrite_zeroes(blk, offset, mlen,
2197                                                BDRV_REQ_MAY_UNMAP,
2198                                                nvme_rw_complete_cb, req);
2199             return;
2200         }
2201 
2202         if (nvme_ns_ext(ns) || req->cmd.mptr) {
2203             uint16_t status;
2204 
2205             nvme_sg_unmap(&req->sg);
2206             status = nvme_map_mdata(nvme_ctrl(req), nlb, req);
2207             if (status) {
2208                 ret = -EFAULT;
2209                 goto out;
2210             }
2211 
2212             if (req->cmd.opcode == NVME_CMD_READ) {
2213                 return nvme_blk_read(blk, offset, 1, nvme_rw_complete_cb, req);
2214             }
2215 
2216             return nvme_blk_write(blk, offset, 1, nvme_rw_complete_cb, req);
2217         }
2218     }
2219 
2220 out:
2221     nvme_rw_complete_cb(req, ret);
2222 }
2223 
2224 static void nvme_verify_cb(void *opaque, int ret)
2225 {
2226     NvmeBounceContext *ctx = opaque;
2227     NvmeRequest *req = ctx->req;
2228     NvmeNamespace *ns = req->ns;
2229     BlockBackend *blk = ns->blkconf.blk;
2230     BlockAcctCookie *acct = &req->acct;
2231     BlockAcctStats *stats = blk_get_stats(blk);
2232     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2233     uint64_t slba = le64_to_cpu(rw->slba);
2234     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2235     uint16_t apptag = le16_to_cpu(rw->apptag);
2236     uint16_t appmask = le16_to_cpu(rw->appmask);
2237     uint64_t reftag = le32_to_cpu(rw->reftag);
2238     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2239     uint16_t status;
2240 
2241     reftag |= cdw3 << 32;
2242 
2243     trace_pci_nvme_verify_cb(nvme_cid(req), prinfo, apptag, appmask, reftag);
2244 
2245     if (ret) {
2246         block_acct_failed(stats, acct);
2247         nvme_aio_err(req, ret);
2248         goto out;
2249     }
2250 
2251     block_acct_done(stats, acct);
2252 
2253     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2254         status = nvme_dif_mangle_mdata(ns, ctx->mdata.bounce,
2255                                        ctx->mdata.iov.size, slba);
2256         if (status) {
2257             req->status = status;
2258             goto out;
2259         }
2260 
2261         req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2262                                      ctx->mdata.bounce, ctx->mdata.iov.size,
2263                                      prinfo, slba, apptag, appmask, &reftag);
2264     }
2265 
2266 out:
2267     qemu_iovec_destroy(&ctx->data.iov);
2268     g_free(ctx->data.bounce);
2269 
2270     qemu_iovec_destroy(&ctx->mdata.iov);
2271     g_free(ctx->mdata.bounce);
2272 
2273     g_free(ctx);
2274 
2275     nvme_enqueue_req_completion(nvme_cq(req), req);
2276 }
2277 
2278 
2279 static void nvme_verify_mdata_in_cb(void *opaque, int ret)
2280 {
2281     NvmeBounceContext *ctx = opaque;
2282     NvmeRequest *req = ctx->req;
2283     NvmeNamespace *ns = req->ns;
2284     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2285     uint64_t slba = le64_to_cpu(rw->slba);
2286     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2287     size_t mlen = nvme_m2b(ns, nlb);
2288     uint64_t offset = nvme_moff(ns, slba);
2289     BlockBackend *blk = ns->blkconf.blk;
2290 
2291     trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req), blk_name(blk));
2292 
2293     if (ret) {
2294         goto out;
2295     }
2296 
2297     ctx->mdata.bounce = g_malloc(mlen);
2298 
2299     qemu_iovec_reset(&ctx->mdata.iov);
2300     qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2301 
2302     req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2303                                 nvme_verify_cb, ctx);
2304     return;
2305 
2306 out:
2307     nvme_verify_cb(ctx, ret);
2308 }
2309 
2310 struct nvme_compare_ctx {
2311     struct {
2312         QEMUIOVector iov;
2313         uint8_t *bounce;
2314     } data;
2315 
2316     struct {
2317         QEMUIOVector iov;
2318         uint8_t *bounce;
2319     } mdata;
2320 };
2321 
2322 static void nvme_compare_mdata_cb(void *opaque, int ret)
2323 {
2324     NvmeRequest *req = opaque;
2325     NvmeNamespace *ns = req->ns;
2326     NvmeCtrl *n = nvme_ctrl(req);
2327     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2328     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2329     uint16_t apptag = le16_to_cpu(rw->apptag);
2330     uint16_t appmask = le16_to_cpu(rw->appmask);
2331     uint64_t reftag = le32_to_cpu(rw->reftag);
2332     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2333     struct nvme_compare_ctx *ctx = req->opaque;
2334     g_autofree uint8_t *buf = NULL;
2335     BlockBackend *blk = ns->blkconf.blk;
2336     BlockAcctCookie *acct = &req->acct;
2337     BlockAcctStats *stats = blk_get_stats(blk);
2338     uint16_t status = NVME_SUCCESS;
2339 
2340     reftag |= cdw3 << 32;
2341 
2342     trace_pci_nvme_compare_mdata_cb(nvme_cid(req));
2343 
2344     if (ret) {
2345         block_acct_failed(stats, acct);
2346         nvme_aio_err(req, ret);
2347         goto out;
2348     }
2349 
2350     buf = g_malloc(ctx->mdata.iov.size);
2351 
2352     status = nvme_bounce_mdata(n, buf, ctx->mdata.iov.size,
2353                                NVME_TX_DIRECTION_TO_DEVICE, req);
2354     if (status) {
2355         req->status = status;
2356         goto out;
2357     }
2358 
2359     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2360         uint64_t slba = le64_to_cpu(rw->slba);
2361         uint8_t *bufp;
2362         uint8_t *mbufp = ctx->mdata.bounce;
2363         uint8_t *end = mbufp + ctx->mdata.iov.size;
2364         int16_t pil = 0;
2365 
2366         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2367                                 ctx->mdata.bounce, ctx->mdata.iov.size, prinfo,
2368                                 slba, apptag, appmask, &reftag);
2369         if (status) {
2370             req->status = status;
2371             goto out;
2372         }
2373 
2374         /*
2375          * When formatted with protection information, do not compare the DIF
2376          * tuple.
2377          */
2378         if (!(ns->id_ns.dps & NVME_ID_NS_DPS_FIRST_EIGHT)) {
2379             pil = ns->lbaf.ms - nvme_pi_tuple_size(ns);
2380         }
2381 
2382         for (bufp = buf; mbufp < end; bufp += ns->lbaf.ms, mbufp += ns->lbaf.ms) {
2383             if (memcmp(bufp + pil, mbufp + pil, ns->lbaf.ms - pil)) {
2384                 req->status = NVME_CMP_FAILURE | NVME_DNR;
2385                 goto out;
2386             }
2387         }
2388 
2389         goto out;
2390     }
2391 
2392     if (memcmp(buf, ctx->mdata.bounce, ctx->mdata.iov.size)) {
2393         req->status = NVME_CMP_FAILURE | NVME_DNR;
2394         goto out;
2395     }
2396 
2397     block_acct_done(stats, acct);
2398 
2399 out:
2400     qemu_iovec_destroy(&ctx->data.iov);
2401     g_free(ctx->data.bounce);
2402 
2403     qemu_iovec_destroy(&ctx->mdata.iov);
2404     g_free(ctx->mdata.bounce);
2405 
2406     g_free(ctx);
2407 
2408     nvme_enqueue_req_completion(nvme_cq(req), req);
2409 }
2410 
2411 static void nvme_compare_data_cb(void *opaque, int ret)
2412 {
2413     NvmeRequest *req = opaque;
2414     NvmeCtrl *n = nvme_ctrl(req);
2415     NvmeNamespace *ns = req->ns;
2416     BlockBackend *blk = ns->blkconf.blk;
2417     BlockAcctCookie *acct = &req->acct;
2418     BlockAcctStats *stats = blk_get_stats(blk);
2419 
2420     struct nvme_compare_ctx *ctx = req->opaque;
2421     g_autofree uint8_t *buf = NULL;
2422     uint16_t status;
2423 
2424     trace_pci_nvme_compare_data_cb(nvme_cid(req));
2425 
2426     if (ret) {
2427         block_acct_failed(stats, acct);
2428         nvme_aio_err(req, ret);
2429         goto out;
2430     }
2431 
2432     buf = g_malloc(ctx->data.iov.size);
2433 
2434     status = nvme_bounce_data(n, buf, ctx->data.iov.size,
2435                               NVME_TX_DIRECTION_TO_DEVICE, req);
2436     if (status) {
2437         req->status = status;
2438         goto out;
2439     }
2440 
2441     if (memcmp(buf, ctx->data.bounce, ctx->data.iov.size)) {
2442         req->status = NVME_CMP_FAILURE | NVME_DNR;
2443         goto out;
2444     }
2445 
2446     if (ns->lbaf.ms) {
2447         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2448         uint64_t slba = le64_to_cpu(rw->slba);
2449         uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2450         size_t mlen = nvme_m2b(ns, nlb);
2451         uint64_t offset = nvme_moff(ns, slba);
2452 
2453         ctx->mdata.bounce = g_malloc(mlen);
2454 
2455         qemu_iovec_init(&ctx->mdata.iov, 1);
2456         qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2457 
2458         req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2459                                     nvme_compare_mdata_cb, req);
2460         return;
2461     }
2462 
2463     block_acct_done(stats, acct);
2464 
2465 out:
2466     qemu_iovec_destroy(&ctx->data.iov);
2467     g_free(ctx->data.bounce);
2468     g_free(ctx);
2469 
2470     nvme_enqueue_req_completion(nvme_cq(req), req);
2471 }
2472 
2473 typedef struct NvmeDSMAIOCB {
2474     BlockAIOCB common;
2475     BlockAIOCB *aiocb;
2476     NvmeRequest *req;
2477     int ret;
2478 
2479     NvmeDsmRange *range;
2480     unsigned int nr;
2481     unsigned int idx;
2482 } NvmeDSMAIOCB;
2483 
2484 static void nvme_dsm_cancel(BlockAIOCB *aiocb)
2485 {
2486     NvmeDSMAIOCB *iocb = container_of(aiocb, NvmeDSMAIOCB, common);
2487 
2488     /* break nvme_dsm_cb loop */
2489     iocb->idx = iocb->nr;
2490     iocb->ret = -ECANCELED;
2491 
2492     if (iocb->aiocb) {
2493         blk_aio_cancel_async(iocb->aiocb);
2494         iocb->aiocb = NULL;
2495     } else {
2496         /*
2497          * We only reach this if nvme_dsm_cancel() has already been called or
2498          * the command ran to completion.
2499          */
2500         assert(iocb->idx == iocb->nr);
2501     }
2502 }
2503 
2504 static const AIOCBInfo nvme_dsm_aiocb_info = {
2505     .aiocb_size   = sizeof(NvmeDSMAIOCB),
2506     .cancel_async = nvme_dsm_cancel,
2507 };
2508 
2509 static void nvme_dsm_cb(void *opaque, int ret);
2510 
2511 static void nvme_dsm_md_cb(void *opaque, int ret)
2512 {
2513     NvmeDSMAIOCB *iocb = opaque;
2514     NvmeRequest *req = iocb->req;
2515     NvmeNamespace *ns = req->ns;
2516     NvmeDsmRange *range;
2517     uint64_t slba;
2518     uint32_t nlb;
2519 
2520     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
2521         goto done;
2522     }
2523 
2524     range = &iocb->range[iocb->idx - 1];
2525     slba = le64_to_cpu(range->slba);
2526     nlb = le32_to_cpu(range->nlb);
2527 
2528     /*
2529      * Check that all block were discarded (zeroed); otherwise we do not zero
2530      * the metadata.
2531      */
2532 
2533     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_ZERO);
2534     if (ret) {
2535         if (ret < 0) {
2536             goto done;
2537         }
2538 
2539         nvme_dsm_cb(iocb, 0);
2540         return;
2541     }
2542 
2543     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_moff(ns, slba),
2544                                         nvme_m2b(ns, nlb), BDRV_REQ_MAY_UNMAP,
2545                                         nvme_dsm_cb, iocb);
2546     return;
2547 
2548 done:
2549     nvme_dsm_cb(iocb, ret);
2550 }
2551 
2552 static void nvme_dsm_cb(void *opaque, int ret)
2553 {
2554     NvmeDSMAIOCB *iocb = opaque;
2555     NvmeRequest *req = iocb->req;
2556     NvmeCtrl *n = nvme_ctrl(req);
2557     NvmeNamespace *ns = req->ns;
2558     NvmeDsmRange *range;
2559     uint64_t slba;
2560     uint32_t nlb;
2561 
2562     if (iocb->ret < 0) {
2563         goto done;
2564     } else if (ret < 0) {
2565         iocb->ret = ret;
2566         goto done;
2567     }
2568 
2569 next:
2570     if (iocb->idx == iocb->nr) {
2571         goto done;
2572     }
2573 
2574     range = &iocb->range[iocb->idx++];
2575     slba = le64_to_cpu(range->slba);
2576     nlb = le32_to_cpu(range->nlb);
2577 
2578     trace_pci_nvme_dsm_deallocate(slba, nlb);
2579 
2580     if (nlb > n->dmrsl) {
2581         trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
2582         goto next;
2583     }
2584 
2585     if (nvme_check_bounds(ns, slba, nlb)) {
2586         trace_pci_nvme_err_invalid_lba_range(slba, nlb,
2587                                              ns->id_ns.nsze);
2588         goto next;
2589     }
2590 
2591     iocb->aiocb = blk_aio_pdiscard(ns->blkconf.blk, nvme_l2b(ns, slba),
2592                                    nvme_l2b(ns, nlb),
2593                                    nvme_dsm_md_cb, iocb);
2594     return;
2595 
2596 done:
2597     iocb->aiocb = NULL;
2598     iocb->common.cb(iocb->common.opaque, iocb->ret);
2599     g_free(iocb->range);
2600     qemu_aio_unref(iocb);
2601 }
2602 
2603 static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
2604 {
2605     NvmeNamespace *ns = req->ns;
2606     NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
2607     uint32_t attr = le32_to_cpu(dsm->attributes);
2608     uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
2609     uint16_t status = NVME_SUCCESS;
2610 
2611     trace_pci_nvme_dsm(nr, attr);
2612 
2613     if (attr & NVME_DSMGMT_AD) {
2614         NvmeDSMAIOCB *iocb = blk_aio_get(&nvme_dsm_aiocb_info, ns->blkconf.blk,
2615                                          nvme_misc_cb, req);
2616 
2617         iocb->req = req;
2618         iocb->ret = 0;
2619         iocb->range = g_new(NvmeDsmRange, nr);
2620         iocb->nr = nr;
2621         iocb->idx = 0;
2622 
2623         status = nvme_h2c(n, (uint8_t *)iocb->range, sizeof(NvmeDsmRange) * nr,
2624                           req);
2625         if (status) {
2626             g_free(iocb->range);
2627             qemu_aio_unref(iocb);
2628 
2629             return status;
2630         }
2631 
2632         req->aiocb = &iocb->common;
2633         nvme_dsm_cb(iocb, 0);
2634 
2635         return NVME_NO_COMPLETE;
2636     }
2637 
2638     return status;
2639 }
2640 
2641 static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
2642 {
2643     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2644     NvmeNamespace *ns = req->ns;
2645     BlockBackend *blk = ns->blkconf.blk;
2646     uint64_t slba = le64_to_cpu(rw->slba);
2647     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2648     size_t len = nvme_l2b(ns, nlb);
2649     int64_t offset = nvme_l2b(ns, slba);
2650     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2651     uint32_t reftag = le32_to_cpu(rw->reftag);
2652     NvmeBounceContext *ctx = NULL;
2653     uint16_t status;
2654 
2655     trace_pci_nvme_verify(nvme_cid(req), nvme_nsid(ns), slba, nlb);
2656 
2657     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2658         status = nvme_check_prinfo(ns, prinfo, slba, reftag);
2659         if (status) {
2660             return status;
2661         }
2662 
2663         if (prinfo & NVME_PRINFO_PRACT) {
2664             return NVME_INVALID_PROT_INFO | NVME_DNR;
2665         }
2666     }
2667 
2668     if (len > n->page_size << n->params.vsl) {
2669         return NVME_INVALID_FIELD | NVME_DNR;
2670     }
2671 
2672     status = nvme_check_bounds(ns, slba, nlb);
2673     if (status) {
2674         return status;
2675     }
2676 
2677     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2678         status = nvme_check_dulbe(ns, slba, nlb);
2679         if (status) {
2680             return status;
2681         }
2682     }
2683 
2684     ctx = g_new0(NvmeBounceContext, 1);
2685     ctx->req = req;
2686 
2687     ctx->data.bounce = g_malloc(len);
2688 
2689     qemu_iovec_init(&ctx->data.iov, 1);
2690     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, len);
2691 
2692     block_acct_start(blk_get_stats(blk), &req->acct, ctx->data.iov.size,
2693                      BLOCK_ACCT_READ);
2694 
2695     req->aiocb = blk_aio_preadv(ns->blkconf.blk, offset, &ctx->data.iov, 0,
2696                                 nvme_verify_mdata_in_cb, ctx);
2697     return NVME_NO_COMPLETE;
2698 }
2699 
2700 typedef struct NvmeCopyAIOCB {
2701     BlockAIOCB common;
2702     BlockAIOCB *aiocb;
2703     NvmeRequest *req;
2704     NvmeCtrl *n;
2705     int ret;
2706 
2707     void *ranges;
2708     unsigned int format;
2709     int nr;
2710     int idx;
2711 
2712     uint8_t *bounce;
2713     QEMUIOVector iov;
2714     struct {
2715         BlockAcctCookie read;
2716         BlockAcctCookie write;
2717     } acct;
2718 
2719     uint64_t reftag;
2720     uint64_t slba;
2721 
2722     NvmeZone *zone;
2723     NvmeNamespace *sns;
2724     uint32_t tcl;
2725 } NvmeCopyAIOCB;
2726 
2727 static void nvme_copy_cancel(BlockAIOCB *aiocb)
2728 {
2729     NvmeCopyAIOCB *iocb = container_of(aiocb, NvmeCopyAIOCB, common);
2730 
2731     iocb->ret = -ECANCELED;
2732 
2733     if (iocb->aiocb) {
2734         blk_aio_cancel_async(iocb->aiocb);
2735         iocb->aiocb = NULL;
2736     }
2737 }
2738 
2739 static const AIOCBInfo nvme_copy_aiocb_info = {
2740     .aiocb_size   = sizeof(NvmeCopyAIOCB),
2741     .cancel_async = nvme_copy_cancel,
2742 };
2743 
2744 static void nvme_copy_done(NvmeCopyAIOCB *iocb)
2745 {
2746     NvmeRequest *req = iocb->req;
2747     NvmeNamespace *ns = req->ns;
2748     BlockAcctStats *stats = blk_get_stats(ns->blkconf.blk);
2749 
2750     if (iocb->idx != iocb->nr) {
2751         req->cqe.result = cpu_to_le32(iocb->idx);
2752     }
2753 
2754     qemu_iovec_destroy(&iocb->iov);
2755     g_free(iocb->bounce);
2756 
2757     if (iocb->ret < 0) {
2758         block_acct_failed(stats, &iocb->acct.read);
2759         block_acct_failed(stats, &iocb->acct.write);
2760     } else {
2761         block_acct_done(stats, &iocb->acct.read);
2762         block_acct_done(stats, &iocb->acct.write);
2763     }
2764 
2765     iocb->common.cb(iocb->common.opaque, iocb->ret);
2766     qemu_aio_unref(iocb);
2767 }
2768 
2769 static void nvme_do_copy(NvmeCopyAIOCB *iocb);
2770 
2771 static void nvme_copy_source_range_parse_format0_2(void *ranges,
2772                                                    int idx, uint64_t *slba,
2773                                                    uint32_t *nlb,
2774                                                    uint32_t *snsid,
2775                                                    uint16_t *apptag,
2776                                                    uint16_t *appmask,
2777                                                    uint64_t *reftag)
2778 {
2779     NvmeCopySourceRangeFormat0_2 *_ranges = ranges;
2780 
2781     if (snsid) {
2782         *snsid = le32_to_cpu(_ranges[idx].sparams);
2783     }
2784 
2785     if (slba) {
2786         *slba = le64_to_cpu(_ranges[idx].slba);
2787     }
2788 
2789     if (nlb) {
2790         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2791     }
2792 
2793     if (apptag) {
2794         *apptag = le16_to_cpu(_ranges[idx].apptag);
2795     }
2796 
2797     if (appmask) {
2798         *appmask = le16_to_cpu(_ranges[idx].appmask);
2799     }
2800 
2801     if (reftag) {
2802         *reftag = le32_to_cpu(_ranges[idx].reftag);
2803     }
2804 }
2805 
2806 static void nvme_copy_source_range_parse_format1_3(void *ranges, int idx,
2807                                                    uint64_t *slba,
2808                                                    uint32_t *nlb,
2809                                                    uint32_t *snsid,
2810                                                    uint16_t *apptag,
2811                                                    uint16_t *appmask,
2812                                                    uint64_t *reftag)
2813 {
2814     NvmeCopySourceRangeFormat1_3 *_ranges = ranges;
2815 
2816     if (snsid) {
2817         *snsid = le32_to_cpu(_ranges[idx].sparams);
2818     }
2819 
2820     if (slba) {
2821         *slba = le64_to_cpu(_ranges[idx].slba);
2822     }
2823 
2824     if (nlb) {
2825         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2826     }
2827 
2828     if (apptag) {
2829         *apptag = le16_to_cpu(_ranges[idx].apptag);
2830     }
2831 
2832     if (appmask) {
2833         *appmask = le16_to_cpu(_ranges[idx].appmask);
2834     }
2835 
2836     if (reftag) {
2837         *reftag = 0;
2838 
2839         *reftag |= (uint64_t)_ranges[idx].sr[4] << 40;
2840         *reftag |= (uint64_t)_ranges[idx].sr[5] << 32;
2841         *reftag |= (uint64_t)_ranges[idx].sr[6] << 24;
2842         *reftag |= (uint64_t)_ranges[idx].sr[7] << 16;
2843         *reftag |= (uint64_t)_ranges[idx].sr[8] << 8;
2844         *reftag |= (uint64_t)_ranges[idx].sr[9];
2845     }
2846 }
2847 
2848 static void nvme_copy_source_range_parse(void *ranges, int idx, uint8_t format,
2849                                          uint64_t *slba, uint32_t *nlb,
2850                                          uint32_t *snsid, uint16_t *apptag,
2851                                          uint16_t *appmask, uint64_t *reftag)
2852 {
2853     switch (format) {
2854     case NVME_COPY_FORMAT_0:
2855     case NVME_COPY_FORMAT_2:
2856         nvme_copy_source_range_parse_format0_2(ranges, idx, slba, nlb, snsid,
2857                                                apptag, appmask, reftag);
2858         break;
2859 
2860     case NVME_COPY_FORMAT_1:
2861     case NVME_COPY_FORMAT_3:
2862         nvme_copy_source_range_parse_format1_3(ranges, idx, slba, nlb, snsid,
2863                                                apptag, appmask, reftag);
2864         break;
2865 
2866     default:
2867         abort();
2868     }
2869 }
2870 
2871 static inline uint16_t nvme_check_copy_mcl(NvmeNamespace *ns,
2872                                            NvmeCopyAIOCB *iocb, uint16_t nr)
2873 {
2874     uint32_t copy_len = 0;
2875 
2876     for (int idx = 0; idx < nr; idx++) {
2877         uint32_t nlb;
2878         nvme_copy_source_range_parse(iocb->ranges, idx, iocb->format, NULL,
2879                                      &nlb, NULL, NULL, NULL, NULL);
2880         copy_len += nlb;
2881     }
2882     iocb->tcl = copy_len;
2883     if (copy_len > ns->id_ns.mcl) {
2884         return NVME_CMD_SIZE_LIMIT | NVME_DNR;
2885     }
2886 
2887     return NVME_SUCCESS;
2888 }
2889 
2890 static void nvme_copy_out_completed_cb(void *opaque, int ret)
2891 {
2892     NvmeCopyAIOCB *iocb = opaque;
2893     NvmeRequest *req = iocb->req;
2894     NvmeNamespace *dns = req->ns;
2895     uint32_t nlb;
2896 
2897     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2898                                  &nlb, NULL, NULL, NULL, NULL);
2899 
2900     if (ret < 0) {
2901         iocb->ret = ret;
2902         goto out;
2903     } else if (iocb->ret < 0) {
2904         goto out;
2905     }
2906 
2907     if (dns->params.zoned) {
2908         nvme_advance_zone_wp(dns, iocb->zone, nlb);
2909     }
2910 
2911     iocb->idx++;
2912     iocb->slba += nlb;
2913 out:
2914     nvme_do_copy(iocb);
2915 }
2916 
2917 static void nvme_copy_out_cb(void *opaque, int ret)
2918 {
2919     NvmeCopyAIOCB *iocb = opaque;
2920     NvmeRequest *req = iocb->req;
2921     NvmeNamespace *dns = req->ns;
2922     uint32_t nlb;
2923     size_t mlen;
2924     uint8_t *mbounce;
2925 
2926     if (ret < 0 || iocb->ret < 0 || !dns->lbaf.ms) {
2927         goto out;
2928     }
2929 
2930     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2931                                  &nlb, NULL, NULL, NULL, NULL);
2932 
2933     mlen = nvme_m2b(dns, nlb);
2934     mbounce = iocb->bounce + nvme_l2b(dns, nlb);
2935 
2936     qemu_iovec_reset(&iocb->iov);
2937     qemu_iovec_add(&iocb->iov, mbounce, mlen);
2938 
2939     iocb->aiocb = blk_aio_pwritev(dns->blkconf.blk, nvme_moff(dns, iocb->slba),
2940                                   &iocb->iov, 0, nvme_copy_out_completed_cb,
2941                                   iocb);
2942 
2943     return;
2944 
2945 out:
2946     nvme_copy_out_completed_cb(iocb, ret);
2947 }
2948 
2949 static void nvme_copy_in_completed_cb(void *opaque, int ret)
2950 {
2951     NvmeCopyAIOCB *iocb = opaque;
2952     NvmeRequest *req = iocb->req;
2953     NvmeNamespace *sns = iocb->sns;
2954     NvmeNamespace *dns = req->ns;
2955     NvmeCopyCmd *copy = NULL;
2956     uint8_t *mbounce = NULL;
2957     uint32_t nlb;
2958     uint64_t slba;
2959     uint16_t apptag, appmask;
2960     uint64_t reftag;
2961     size_t len, mlen;
2962     uint16_t status;
2963 
2964     if (ret < 0) {
2965         iocb->ret = ret;
2966         goto out;
2967     } else if (iocb->ret < 0) {
2968         goto out;
2969     }
2970 
2971     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
2972                                  &nlb, NULL, &apptag, &appmask, &reftag);
2973 
2974     trace_pci_nvme_copy_out(iocb->slba, nlb);
2975 
2976     len = nvme_l2b(sns, nlb);
2977 
2978     if (NVME_ID_NS_DPS_TYPE(sns->id_ns.dps)) {
2979         copy = (NvmeCopyCmd *)&req->cmd;
2980 
2981         uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
2982 
2983         mlen = nvme_m2b(sns, nlb);
2984         mbounce = iocb->bounce + nvme_l2b(sns, nlb);
2985 
2986         status = nvme_dif_mangle_mdata(sns, mbounce, mlen, slba);
2987         if (status) {
2988             goto invalid;
2989         }
2990         status = nvme_dif_check(sns, iocb->bounce, len, mbounce, mlen, prinfor,
2991                                 slba, apptag, appmask, &reftag);
2992         if (status) {
2993             goto invalid;
2994         }
2995     }
2996 
2997     if (NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
2998         copy = (NvmeCopyCmd *)&req->cmd;
2999         uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
3000 
3001         mlen = nvme_m2b(dns, nlb);
3002         mbounce = iocb->bounce + nvme_l2b(dns, nlb);
3003 
3004         apptag = le16_to_cpu(copy->apptag);
3005         appmask = le16_to_cpu(copy->appmask);
3006 
3007         if (prinfow & NVME_PRINFO_PRACT) {
3008             status = nvme_check_prinfo(dns, prinfow, iocb->slba, iocb->reftag);
3009             if (status) {
3010                 goto invalid;
3011             }
3012 
3013             nvme_dif_pract_generate_dif(dns, iocb->bounce, len, mbounce, mlen,
3014                                         apptag, &iocb->reftag);
3015         } else {
3016             status = nvme_dif_check(dns, iocb->bounce, len, mbounce, mlen,
3017                                     prinfow, iocb->slba, apptag, appmask,
3018                                     &iocb->reftag);
3019             if (status) {
3020                 goto invalid;
3021             }
3022         }
3023     }
3024 
3025     status = nvme_check_bounds(dns, iocb->slba, nlb);
3026     if (status) {
3027         goto invalid;
3028     }
3029 
3030     if (dns->params.zoned) {
3031         status = nvme_check_zone_write(dns, iocb->zone, iocb->slba, nlb);
3032         if (status) {
3033             goto invalid;
3034         }
3035 
3036         if (!(iocb->zone->d.za & NVME_ZA_ZRWA_VALID)) {
3037             iocb->zone->w_ptr += nlb;
3038         }
3039     }
3040 
3041     qemu_iovec_reset(&iocb->iov);
3042     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3043 
3044     block_acct_start(blk_get_stats(dns->blkconf.blk), &iocb->acct.write, 0,
3045                      BLOCK_ACCT_WRITE);
3046 
3047     iocb->aiocb = blk_aio_pwritev(dns->blkconf.blk, nvme_l2b(dns, iocb->slba),
3048                                   &iocb->iov, 0, nvme_copy_out_cb, iocb);
3049 
3050     return;
3051 
3052 invalid:
3053     req->status = status;
3054     iocb->ret = -1;
3055 out:
3056     nvme_do_copy(iocb);
3057 }
3058 
3059 static void nvme_copy_in_cb(void *opaque, int ret)
3060 {
3061     NvmeCopyAIOCB *iocb = opaque;
3062     NvmeNamespace *sns = iocb->sns;
3063     uint64_t slba;
3064     uint32_t nlb;
3065 
3066     if (ret < 0 || iocb->ret < 0 || !sns->lbaf.ms) {
3067         goto out;
3068     }
3069 
3070     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
3071                                  &nlb, NULL, NULL, NULL, NULL);
3072 
3073     qemu_iovec_reset(&iocb->iov);
3074     qemu_iovec_add(&iocb->iov, iocb->bounce + nvme_l2b(sns, nlb),
3075                    nvme_m2b(sns, nlb));
3076 
3077     iocb->aiocb = blk_aio_preadv(sns->blkconf.blk, nvme_moff(sns, slba),
3078                                  &iocb->iov, 0, nvme_copy_in_completed_cb,
3079                                  iocb);
3080     return;
3081 
3082 out:
3083     nvme_copy_in_completed_cb(iocb, ret);
3084 }
3085 
3086 static inline bool nvme_csi_supports_copy(uint8_t csi)
3087 {
3088     return csi == NVME_CSI_NVM || csi == NVME_CSI_ZONED;
3089 }
3090 
3091 static inline bool nvme_copy_ns_format_match(NvmeNamespace *sns,
3092                                              NvmeNamespace *dns)
3093 {
3094     return sns->lbaf.ds == dns->lbaf.ds && sns->lbaf.ms == dns->lbaf.ms;
3095 }
3096 
3097 static bool nvme_copy_matching_ns_format(NvmeNamespace *sns, NvmeNamespace *dns,
3098                                          bool pi_enable)
3099 {
3100     if (!nvme_csi_supports_copy(sns->csi) ||
3101         !nvme_csi_supports_copy(dns->csi)) {
3102         return false;
3103     }
3104 
3105     if (!pi_enable && !nvme_copy_ns_format_match(sns, dns)) {
3106             return false;
3107     }
3108 
3109     if (pi_enable && (!nvme_copy_ns_format_match(sns, dns) ||
3110         sns->id_ns.dps != dns->id_ns.dps)) {
3111             return false;
3112     }
3113 
3114     return true;
3115 }
3116 
3117 static inline bool nvme_copy_corresp_pi_match(NvmeNamespace *sns,
3118                                               NvmeNamespace *dns)
3119 {
3120     return sns->lbaf.ms == 0 &&
3121            ((dns->lbaf.ms == 8 && dns->pif == 0) ||
3122            (dns->lbaf.ms == 16 && dns->pif == 1));
3123 }
3124 
3125 static bool nvme_copy_corresp_pi_format(NvmeNamespace *sns, NvmeNamespace *dns,
3126                                         bool sns_pi_en)
3127 {
3128     if (!nvme_csi_supports_copy(sns->csi) ||
3129         !nvme_csi_supports_copy(dns->csi)) {
3130         return false;
3131     }
3132 
3133     if (!sns_pi_en && !nvme_copy_corresp_pi_match(sns, dns)) {
3134         return false;
3135     }
3136 
3137     if (sns_pi_en && !nvme_copy_corresp_pi_match(dns, sns)) {
3138         return false;
3139     }
3140 
3141     return true;
3142 }
3143 
3144 static void nvme_do_copy(NvmeCopyAIOCB *iocb)
3145 {
3146     NvmeRequest *req = iocb->req;
3147     NvmeNamespace *sns;
3148     NvmeNamespace *dns = req->ns;
3149     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
3150     uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
3151     uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
3152     uint64_t slba;
3153     uint32_t nlb;
3154     size_t len;
3155     uint16_t status;
3156     uint32_t dnsid = le32_to_cpu(req->cmd.nsid);
3157     uint32_t snsid = dnsid;
3158 
3159     if (iocb->ret < 0) {
3160         goto done;
3161     }
3162 
3163     if (iocb->idx == iocb->nr) {
3164         goto done;
3165     }
3166 
3167     if (iocb->format == 2 || iocb->format == 3) {
3168         nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format,
3169                                      &slba, &nlb, &snsid, NULL, NULL, NULL);
3170         if (snsid != dnsid) {
3171             if (snsid == NVME_NSID_BROADCAST ||
3172                 !nvme_nsid_valid(iocb->n, snsid)) {
3173                 status = NVME_INVALID_NSID | NVME_DNR;
3174                 goto invalid;
3175             }
3176             iocb->sns = nvme_ns(iocb->n, snsid);
3177             if (unlikely(!iocb->sns)) {
3178                 status = NVME_INVALID_FIELD | NVME_DNR;
3179                 goto invalid;
3180             }
3181         } else {
3182             if (((slba + nlb) > iocb->slba) &&
3183                 ((slba + nlb) < (iocb->slba + iocb->tcl))) {
3184                 status = NVME_CMD_OVERLAP_IO_RANGE | NVME_DNR;
3185                 goto invalid;
3186             }
3187         }
3188     } else {
3189         nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format,
3190                                      &slba, &nlb, NULL, NULL, NULL, NULL);
3191     }
3192 
3193     sns = iocb->sns;
3194     if ((snsid == dnsid) && NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3195         ((prinfor & NVME_PRINFO_PRACT) != (prinfow & NVME_PRINFO_PRACT))) {
3196         status = NVME_INVALID_FIELD | NVME_DNR;
3197         goto invalid;
3198     } else if (snsid != dnsid) {
3199         if (!NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3200             !NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3201             if (!nvme_copy_matching_ns_format(sns, dns, false)) {
3202                 status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3203                 goto invalid;
3204             }
3205         }
3206         if (NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3207             NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3208             if ((prinfor & NVME_PRINFO_PRACT) !=
3209                 (prinfow & NVME_PRINFO_PRACT)) {
3210                 status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3211                 goto invalid;
3212             } else {
3213                 if (!nvme_copy_matching_ns_format(sns, dns, true)) {
3214                     status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3215                     goto invalid;
3216                 }
3217             }
3218         }
3219 
3220         if (!NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3221             NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3222             if (!(prinfow & NVME_PRINFO_PRACT)) {
3223                 status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3224                 goto invalid;
3225             } else {
3226                 if (!nvme_copy_corresp_pi_format(sns, dns, false)) {
3227                     status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3228                     goto invalid;
3229                 }
3230             }
3231         }
3232 
3233         if (NVME_ID_NS_DPS_TYPE(sns->id_ns.dps) &&
3234             !NVME_ID_NS_DPS_TYPE(dns->id_ns.dps)) {
3235             if (!(prinfor & NVME_PRINFO_PRACT)) {
3236                 status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3237                 goto invalid;
3238             } else {
3239                 if (!nvme_copy_corresp_pi_format(sns, dns, true)) {
3240                     status = NVME_CMD_INCOMP_NS_OR_FMT | NVME_DNR;
3241                     goto invalid;
3242                 }
3243             }
3244         }
3245     }
3246     len = nvme_l2b(sns, nlb);
3247 
3248     trace_pci_nvme_copy_source_range(slba, nlb);
3249 
3250     if (nlb > le16_to_cpu(sns->id_ns.mssrl)) {
3251         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3252         goto invalid;
3253     }
3254 
3255     status = nvme_check_bounds(sns, slba, nlb);
3256     if (status) {
3257         goto invalid;
3258     }
3259 
3260     if (NVME_ERR_REC_DULBE(sns->features.err_rec)) {
3261         status = nvme_check_dulbe(sns, slba, nlb);
3262         if (status) {
3263             goto invalid;
3264         }
3265     }
3266 
3267     if (sns->params.zoned) {
3268         status = nvme_check_zone_read(sns, slba, nlb);
3269         if (status) {
3270             goto invalid;
3271         }
3272     }
3273 
3274     g_free(iocb->bounce);
3275     iocb->bounce = g_malloc_n(le16_to_cpu(sns->id_ns.mssrl),
3276                               sns->lbasz + sns->lbaf.ms);
3277 
3278     qemu_iovec_reset(&iocb->iov);
3279     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3280 
3281     block_acct_start(blk_get_stats(sns->blkconf.blk), &iocb->acct.read, 0,
3282                      BLOCK_ACCT_READ);
3283 
3284     iocb->aiocb = blk_aio_preadv(sns->blkconf.blk, nvme_l2b(sns, slba),
3285                                  &iocb->iov, 0, nvme_copy_in_cb, iocb);
3286     return;
3287 
3288 invalid:
3289     req->status = status;
3290     iocb->ret = -1;
3291 done:
3292     nvme_copy_done(iocb);
3293 }
3294 
3295 static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
3296 {
3297     NvmeNamespace *ns = req->ns;
3298     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
3299     NvmeCopyAIOCB *iocb = blk_aio_get(&nvme_copy_aiocb_info, ns->blkconf.blk,
3300                                       nvme_misc_cb, req);
3301     uint16_t nr = copy->nr + 1;
3302     uint8_t format = copy->control[0] & 0xf;
3303     size_t len = sizeof(NvmeCopySourceRangeFormat0_2);
3304 
3305     uint16_t status;
3306 
3307     trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);
3308 
3309     iocb->ranges = NULL;
3310     iocb->zone = NULL;
3311 
3312     if (!(n->id_ctrl.ocfs & (1 << format)) ||
3313         ((format == 2 || format == 3) &&
3314          !(n->features.hbs.cdfe & (1 << format)))) {
3315         trace_pci_nvme_err_copy_invalid_format(format);
3316         status = NVME_INVALID_FIELD | NVME_DNR;
3317         goto invalid;
3318     }
3319 
3320     if (nr > ns->id_ns.msrc + 1) {
3321         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3322         goto invalid;
3323     }
3324 
3325     if ((ns->pif == 0x0 && (format != 0x0 && format != 0x2)) ||
3326         (ns->pif != 0x0 && (format != 0x1 && format != 0x3))) {
3327         status = NVME_INVALID_FORMAT | NVME_DNR;
3328         goto invalid;
3329     }
3330 
3331     if (ns->pif) {
3332         len = sizeof(NvmeCopySourceRangeFormat1_3);
3333     }
3334 
3335     iocb->format = format;
3336     iocb->ranges = g_malloc_n(nr, len);
3337     status = nvme_h2c(n, (uint8_t *)iocb->ranges, len * nr, req);
3338     if (status) {
3339         goto invalid;
3340     }
3341 
3342     iocb->slba = le64_to_cpu(copy->sdlba);
3343 
3344     if (ns->params.zoned) {
3345         iocb->zone = nvme_get_zone_by_slba(ns, iocb->slba);
3346         if (!iocb->zone) {
3347             status = NVME_LBA_RANGE | NVME_DNR;
3348             goto invalid;
3349         }
3350 
3351         status = nvme_zrm_auto(n, ns, iocb->zone);
3352         if (status) {
3353             goto invalid;
3354         }
3355     }
3356 
3357     status = nvme_check_copy_mcl(ns, iocb, nr);
3358     if (status) {
3359         goto invalid;
3360     }
3361 
3362     iocb->req = req;
3363     iocb->ret = 0;
3364     iocb->nr = nr;
3365     iocb->idx = 0;
3366     iocb->reftag = le32_to_cpu(copy->reftag);
3367     iocb->reftag |= (uint64_t)le32_to_cpu(copy->cdw3) << 32;
3368 
3369     qemu_iovec_init(&iocb->iov, 1);
3370 
3371     req->aiocb = &iocb->common;
3372     iocb->sns = req->ns;
3373     iocb->n = n;
3374     iocb->bounce = NULL;
3375     nvme_do_copy(iocb);
3376 
3377     return NVME_NO_COMPLETE;
3378 
3379 invalid:
3380     g_free(iocb->ranges);
3381     qemu_aio_unref(iocb);
3382     return status;
3383 }
3384 
3385 static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req)
3386 {
3387     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3388     NvmeNamespace *ns = req->ns;
3389     BlockBackend *blk = ns->blkconf.blk;
3390     uint64_t slba = le64_to_cpu(rw->slba);
3391     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
3392     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3393     size_t data_len = nvme_l2b(ns, nlb);
3394     size_t len = data_len;
3395     int64_t offset = nvme_l2b(ns, slba);
3396     struct nvme_compare_ctx *ctx = NULL;
3397     uint16_t status;
3398 
3399     trace_pci_nvme_compare(nvme_cid(req), nvme_nsid(ns), slba, nlb);
3400 
3401     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) && (prinfo & NVME_PRINFO_PRACT)) {
3402         return NVME_INVALID_PROT_INFO | NVME_DNR;
3403     }
3404 
3405     if (nvme_ns_ext(ns)) {
3406         len += nvme_m2b(ns, nlb);
3407     }
3408 
3409     status = nvme_check_mdts(n, len);
3410     if (status) {
3411         return status;
3412     }
3413 
3414     status = nvme_check_bounds(ns, slba, nlb);
3415     if (status) {
3416         return status;
3417     }
3418 
3419     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3420         status = nvme_check_dulbe(ns, slba, nlb);
3421         if (status) {
3422             return status;
3423         }
3424     }
3425 
3426     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
3427     if (status) {
3428         return status;
3429     }
3430 
3431     ctx = g_new(struct nvme_compare_ctx, 1);
3432     ctx->data.bounce = g_malloc(data_len);
3433 
3434     req->opaque = ctx;
3435 
3436     qemu_iovec_init(&ctx->data.iov, 1);
3437     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, data_len);
3438 
3439     block_acct_start(blk_get_stats(blk), &req->acct, data_len,
3440                      BLOCK_ACCT_READ);
3441     req->aiocb = blk_aio_preadv(blk, offset, &ctx->data.iov, 0,
3442                                 nvme_compare_data_cb, req);
3443 
3444     return NVME_NO_COMPLETE;
3445 }
3446 
3447 typedef struct NvmeFlushAIOCB {
3448     BlockAIOCB common;
3449     BlockAIOCB *aiocb;
3450     NvmeRequest *req;
3451     int ret;
3452 
3453     NvmeNamespace *ns;
3454     uint32_t nsid;
3455     bool broadcast;
3456 } NvmeFlushAIOCB;
3457 
3458 static void nvme_flush_cancel(BlockAIOCB *acb)
3459 {
3460     NvmeFlushAIOCB *iocb = container_of(acb, NvmeFlushAIOCB, common);
3461 
3462     iocb->ret = -ECANCELED;
3463 
3464     if (iocb->aiocb) {
3465         blk_aio_cancel_async(iocb->aiocb);
3466         iocb->aiocb = NULL;
3467     }
3468 }
3469 
3470 static const AIOCBInfo nvme_flush_aiocb_info = {
3471     .aiocb_size = sizeof(NvmeFlushAIOCB),
3472     .cancel_async = nvme_flush_cancel,
3473 };
3474 
3475 static void nvme_do_flush(NvmeFlushAIOCB *iocb);
3476 
3477 static void nvme_flush_ns_cb(void *opaque, int ret)
3478 {
3479     NvmeFlushAIOCB *iocb = opaque;
3480     NvmeNamespace *ns = iocb->ns;
3481 
3482     if (ret < 0) {
3483         iocb->ret = ret;
3484         goto out;
3485     } else if (iocb->ret < 0) {
3486         goto out;
3487     }
3488 
3489     if (ns) {
3490         trace_pci_nvme_flush_ns(iocb->nsid);
3491 
3492         iocb->ns = NULL;
3493         iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);
3494         return;
3495     }
3496 
3497 out:
3498     nvme_do_flush(iocb);
3499 }
3500 
3501 static void nvme_do_flush(NvmeFlushAIOCB *iocb)
3502 {
3503     NvmeRequest *req = iocb->req;
3504     NvmeCtrl *n = nvme_ctrl(req);
3505     int i;
3506 
3507     if (iocb->ret < 0) {
3508         goto done;
3509     }
3510 
3511     if (iocb->broadcast) {
3512         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
3513             iocb->ns = nvme_ns(n, i);
3514             if (iocb->ns) {
3515                 iocb->nsid = i;
3516                 break;
3517             }
3518         }
3519     }
3520 
3521     if (!iocb->ns) {
3522         goto done;
3523     }
3524 
3525     nvme_flush_ns_cb(iocb, 0);
3526     return;
3527 
3528 done:
3529     iocb->common.cb(iocb->common.opaque, iocb->ret);
3530     qemu_aio_unref(iocb);
3531 }
3532 
3533 static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req)
3534 {
3535     NvmeFlushAIOCB *iocb;
3536     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
3537     uint16_t status;
3538 
3539     iocb = qemu_aio_get(&nvme_flush_aiocb_info, NULL, nvme_misc_cb, req);
3540 
3541     iocb->req = req;
3542     iocb->ret = 0;
3543     iocb->ns = NULL;
3544     iocb->nsid = 0;
3545     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
3546 
3547     if (!iocb->broadcast) {
3548         if (!nvme_nsid_valid(n, nsid)) {
3549             status = NVME_INVALID_NSID | NVME_DNR;
3550             goto out;
3551         }
3552 
3553         iocb->ns = nvme_ns(n, nsid);
3554         if (!iocb->ns) {
3555             status = NVME_INVALID_FIELD | NVME_DNR;
3556             goto out;
3557         }
3558 
3559         iocb->nsid = nsid;
3560     }
3561 
3562     req->aiocb = &iocb->common;
3563     nvme_do_flush(iocb);
3564 
3565     return NVME_NO_COMPLETE;
3566 
3567 out:
3568     qemu_aio_unref(iocb);
3569 
3570     return status;
3571 }
3572 
3573 static uint16_t nvme_read(NvmeCtrl *n, NvmeRequest *req)
3574 {
3575     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3576     NvmeNamespace *ns = req->ns;
3577     uint64_t slba = le64_to_cpu(rw->slba);
3578     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3579     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3580     uint64_t data_size = nvme_l2b(ns, nlb);
3581     uint64_t mapped_size = data_size;
3582     uint64_t data_offset;
3583     BlockBackend *blk = ns->blkconf.blk;
3584     uint16_t status;
3585 
3586     if (nvme_ns_ext(ns)) {
3587         mapped_size += nvme_m2b(ns, nlb);
3588 
3589         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3590             bool pract = prinfo & NVME_PRINFO_PRACT;
3591 
3592             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3593                 mapped_size = data_size;
3594             }
3595         }
3596     }
3597 
3598     trace_pci_nvme_read(nvme_cid(req), nvme_nsid(ns), nlb, mapped_size, slba);
3599 
3600     status = nvme_check_mdts(n, mapped_size);
3601     if (status) {
3602         goto invalid;
3603     }
3604 
3605     status = nvme_check_bounds(ns, slba, nlb);
3606     if (status) {
3607         goto invalid;
3608     }
3609 
3610     if (ns->params.zoned) {
3611         status = nvme_check_zone_read(ns, slba, nlb);
3612         if (status) {
3613             trace_pci_nvme_err_zone_read_not_ok(slba, nlb, status);
3614             goto invalid;
3615         }
3616     }
3617 
3618     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3619         status = nvme_check_dulbe(ns, slba, nlb);
3620         if (status) {
3621             goto invalid;
3622         }
3623     }
3624 
3625     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3626         return nvme_dif_rw(n, req);
3627     }
3628 
3629     status = nvme_map_data(n, nlb, req);
3630     if (status) {
3631         goto invalid;
3632     }
3633 
3634     data_offset = nvme_l2b(ns, slba);
3635 
3636     block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3637                      BLOCK_ACCT_READ);
3638     nvme_blk_read(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3639     return NVME_NO_COMPLETE;
3640 
3641 invalid:
3642     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_READ);
3643     return status | NVME_DNR;
3644 }
3645 
3646 static void nvme_do_write_fdp(NvmeCtrl *n, NvmeRequest *req, uint64_t slba,
3647                               uint32_t nlb)
3648 {
3649     NvmeNamespace *ns = req->ns;
3650     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3651     uint64_t data_size = nvme_l2b(ns, nlb);
3652     uint32_t dw12 = le32_to_cpu(req->cmd.cdw12);
3653     uint8_t dtype = (dw12 >> 20) & 0xf;
3654     uint16_t pid = le16_to_cpu(rw->dspec);
3655     uint16_t ph, rg, ruhid;
3656     NvmeReclaimUnit *ru;
3657 
3658     if (dtype != NVME_DIRECTIVE_DATA_PLACEMENT ||
3659         !nvme_parse_pid(ns, pid, &ph, &rg)) {
3660         ph = 0;
3661         rg = 0;
3662     }
3663 
3664     ruhid = ns->fdp.phs[ph];
3665     ru = &ns->endgrp->fdp.ruhs[ruhid].rus[rg];
3666 
3667     nvme_fdp_stat_inc(&ns->endgrp->fdp.hbmw, data_size);
3668     nvme_fdp_stat_inc(&ns->endgrp->fdp.mbmw, data_size);
3669 
3670     while (nlb) {
3671         if (nlb < ru->ruamw) {
3672             ru->ruamw -= nlb;
3673             break;
3674         }
3675 
3676         nlb -= ru->ruamw;
3677         nvme_update_ruh(n, ns, pid);
3678     }
3679 }
3680 
3681 static uint16_t nvme_do_write(NvmeCtrl *n, NvmeRequest *req, bool append,
3682                               bool wrz)
3683 {
3684     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3685     NvmeNamespace *ns = req->ns;
3686     uint64_t slba = le64_to_cpu(rw->slba);
3687     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3688     uint16_t ctrl = le16_to_cpu(rw->control);
3689     uint8_t prinfo = NVME_RW_PRINFO(ctrl);
3690     uint64_t data_size = nvme_l2b(ns, nlb);
3691     uint64_t mapped_size = data_size;
3692     uint64_t data_offset;
3693     NvmeZone *zone;
3694     NvmeZonedResult *res = (NvmeZonedResult *)&req->cqe;
3695     BlockBackend *blk = ns->blkconf.blk;
3696     uint16_t status;
3697 
3698     if (nvme_ns_ext(ns)) {
3699         mapped_size += nvme_m2b(ns, nlb);
3700 
3701         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3702             bool pract = prinfo & NVME_PRINFO_PRACT;
3703 
3704             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3705                 mapped_size -= nvme_m2b(ns, nlb);
3706             }
3707         }
3708     }
3709 
3710     trace_pci_nvme_write(nvme_cid(req), nvme_io_opc_str(rw->opcode),
3711                          nvme_nsid(ns), nlb, mapped_size, slba);
3712 
3713     if (!wrz) {
3714         status = nvme_check_mdts(n, mapped_size);
3715         if (status) {
3716             goto invalid;
3717         }
3718     }
3719 
3720     status = nvme_check_bounds(ns, slba, nlb);
3721     if (status) {
3722         goto invalid;
3723     }
3724 
3725     if (ns->params.zoned) {
3726         zone = nvme_get_zone_by_slba(ns, slba);
3727         assert(zone);
3728 
3729         if (append) {
3730             bool piremap = !!(ctrl & NVME_RW_PIREMAP);
3731 
3732             if (unlikely(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3733                 return NVME_INVALID_ZONE_OP | NVME_DNR;
3734             }
3735 
3736             if (unlikely(slba != zone->d.zslba)) {
3737                 trace_pci_nvme_err_append_not_at_start(slba, zone->d.zslba);
3738                 status = NVME_INVALID_FIELD;
3739                 goto invalid;
3740             }
3741 
3742             if (n->params.zasl &&
3743                 data_size > (uint64_t)n->page_size << n->params.zasl) {
3744                 trace_pci_nvme_err_zasl(data_size);
3745                 return NVME_INVALID_FIELD | NVME_DNR;
3746             }
3747 
3748             slba = zone->w_ptr;
3749             rw->slba = cpu_to_le64(slba);
3750             res->slba = cpu_to_le64(slba);
3751 
3752             switch (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3753             case NVME_ID_NS_DPS_TYPE_1:
3754                 if (!piremap) {
3755                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3756                 }
3757 
3758                 /* fallthrough */
3759 
3760             case NVME_ID_NS_DPS_TYPE_2:
3761                 if (piremap) {
3762                     uint32_t reftag = le32_to_cpu(rw->reftag);
3763                     rw->reftag = cpu_to_le32(reftag + (slba - zone->d.zslba));
3764                 }
3765 
3766                 break;
3767 
3768             case NVME_ID_NS_DPS_TYPE_3:
3769                 if (piremap) {
3770                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3771                 }
3772 
3773                 break;
3774             }
3775         }
3776 
3777         status = nvme_check_zone_write(ns, zone, slba, nlb);
3778         if (status) {
3779             goto invalid;
3780         }
3781 
3782         status = nvme_zrm_auto(n, ns, zone);
3783         if (status) {
3784             goto invalid;
3785         }
3786 
3787         if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3788             zone->w_ptr += nlb;
3789         }
3790     } else if (ns->endgrp && ns->endgrp->fdp.enabled) {
3791         nvme_do_write_fdp(n, req, slba, nlb);
3792     }
3793 
3794     data_offset = nvme_l2b(ns, slba);
3795 
3796     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3797         return nvme_dif_rw(n, req);
3798     }
3799 
3800     if (!wrz) {
3801         status = nvme_map_data(n, nlb, req);
3802         if (status) {
3803             goto invalid;
3804         }
3805 
3806         block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3807                          BLOCK_ACCT_WRITE);
3808         nvme_blk_write(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3809     } else {
3810         req->aiocb = blk_aio_pwrite_zeroes(blk, data_offset, data_size,
3811                                            BDRV_REQ_MAY_UNMAP, nvme_rw_cb,
3812                                            req);
3813     }
3814 
3815     return NVME_NO_COMPLETE;
3816 
3817 invalid:
3818     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_WRITE);
3819     return status | NVME_DNR;
3820 }
3821 
3822 static inline uint16_t nvme_write(NvmeCtrl *n, NvmeRequest *req)
3823 {
3824     return nvme_do_write(n, req, false, false);
3825 }
3826 
3827 static inline uint16_t nvme_write_zeroes(NvmeCtrl *n, NvmeRequest *req)
3828 {
3829     return nvme_do_write(n, req, false, true);
3830 }
3831 
3832 static inline uint16_t nvme_zone_append(NvmeCtrl *n, NvmeRequest *req)
3833 {
3834     return nvme_do_write(n, req, true, false);
3835 }
3836 
3837 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace *ns, NvmeCmd *c,
3838                                             uint64_t *slba, uint32_t *zone_idx)
3839 {
3840     uint32_t dw10 = le32_to_cpu(c->cdw10);
3841     uint32_t dw11 = le32_to_cpu(c->cdw11);
3842 
3843     if (!ns->params.zoned) {
3844         trace_pci_nvme_err_invalid_opc(c->opcode);
3845         return NVME_INVALID_OPCODE | NVME_DNR;
3846     }
3847 
3848     *slba = ((uint64_t)dw11) << 32 | dw10;
3849     if (unlikely(*slba >= ns->id_ns.nsze)) {
3850         trace_pci_nvme_err_invalid_lba_range(*slba, 0, ns->id_ns.nsze);
3851         *slba = 0;
3852         return NVME_LBA_RANGE | NVME_DNR;
3853     }
3854 
3855     *zone_idx = nvme_zone_idx(ns, *slba);
3856     assert(*zone_idx < ns->num_zones);
3857 
3858     return NVME_SUCCESS;
3859 }
3860 
3861 typedef uint16_t (*op_handler_t)(NvmeNamespace *, NvmeZone *, NvmeZoneState,
3862                                  NvmeRequest *);
3863 
3864 enum NvmeZoneProcessingMask {
3865     NVME_PROC_CURRENT_ZONE    = 0,
3866     NVME_PROC_OPENED_ZONES    = 1 << 0,
3867     NVME_PROC_CLOSED_ZONES    = 1 << 1,
3868     NVME_PROC_READ_ONLY_ZONES = 1 << 2,
3869     NVME_PROC_FULL_ZONES      = 1 << 3,
3870 };
3871 
3872 static uint16_t nvme_open_zone(NvmeNamespace *ns, NvmeZone *zone,
3873                                NvmeZoneState state, NvmeRequest *req)
3874 {
3875     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
3876     int flags = 0;
3877 
3878     if (cmd->zsflags & NVME_ZSFLAG_ZRWA_ALLOC) {
3879         uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3880 
3881         if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3882             return NVME_INVALID_ZONE_OP | NVME_DNR;
3883         }
3884 
3885         if (zone->w_ptr % ns->zns.zrwafg) {
3886             return NVME_NOZRWA | NVME_DNR;
3887         }
3888 
3889         flags = NVME_ZRM_ZRWA;
3890     }
3891 
3892     return nvme_zrm_open_flags(nvme_ctrl(req), ns, zone, flags);
3893 }
3894 
3895 static uint16_t nvme_close_zone(NvmeNamespace *ns, NvmeZone *zone,
3896                                 NvmeZoneState state, NvmeRequest *req)
3897 {
3898     return nvme_zrm_close(ns, zone);
3899 }
3900 
3901 static uint16_t nvme_finish_zone(NvmeNamespace *ns, NvmeZone *zone,
3902                                  NvmeZoneState state, NvmeRequest *req)
3903 {
3904     return nvme_zrm_finish(ns, zone);
3905 }
3906 
3907 static uint16_t nvme_offline_zone(NvmeNamespace *ns, NvmeZone *zone,
3908                                   NvmeZoneState state, NvmeRequest *req)
3909 {
3910     switch (state) {
3911     case NVME_ZONE_STATE_READ_ONLY:
3912         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_OFFLINE);
3913         /* fall through */
3914     case NVME_ZONE_STATE_OFFLINE:
3915         return NVME_SUCCESS;
3916     default:
3917         return NVME_ZONE_INVAL_TRANSITION;
3918     }
3919 }
3920 
3921 static uint16_t nvme_set_zd_ext(NvmeNamespace *ns, NvmeZone *zone)
3922 {
3923     uint16_t status;
3924     uint8_t state = nvme_get_zone_state(zone);
3925 
3926     if (state == NVME_ZONE_STATE_EMPTY) {
3927         status = nvme_aor_check(ns, 1, 0);
3928         if (status) {
3929             return status;
3930         }
3931         nvme_aor_inc_active(ns);
3932         zone->d.za |= NVME_ZA_ZD_EXT_VALID;
3933         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
3934         return NVME_SUCCESS;
3935     }
3936 
3937     return NVME_ZONE_INVAL_TRANSITION;
3938 }
3939 
3940 static uint16_t nvme_bulk_proc_zone(NvmeNamespace *ns, NvmeZone *zone,
3941                                     enum NvmeZoneProcessingMask proc_mask,
3942                                     op_handler_t op_hndlr, NvmeRequest *req)
3943 {
3944     uint16_t status = NVME_SUCCESS;
3945     NvmeZoneState zs = nvme_get_zone_state(zone);
3946     bool proc_zone;
3947 
3948     switch (zs) {
3949     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3950     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3951         proc_zone = proc_mask & NVME_PROC_OPENED_ZONES;
3952         break;
3953     case NVME_ZONE_STATE_CLOSED:
3954         proc_zone = proc_mask & NVME_PROC_CLOSED_ZONES;
3955         break;
3956     case NVME_ZONE_STATE_READ_ONLY:
3957         proc_zone = proc_mask & NVME_PROC_READ_ONLY_ZONES;
3958         break;
3959     case NVME_ZONE_STATE_FULL:
3960         proc_zone = proc_mask & NVME_PROC_FULL_ZONES;
3961         break;
3962     default:
3963         proc_zone = false;
3964     }
3965 
3966     if (proc_zone) {
3967         status = op_hndlr(ns, zone, zs, req);
3968     }
3969 
3970     return status;
3971 }
3972 
3973 static uint16_t nvme_do_zone_op(NvmeNamespace *ns, NvmeZone *zone,
3974                                 enum NvmeZoneProcessingMask proc_mask,
3975                                 op_handler_t op_hndlr, NvmeRequest *req)
3976 {
3977     NvmeZone *next;
3978     uint16_t status = NVME_SUCCESS;
3979     int i;
3980 
3981     if (!proc_mask) {
3982         status = op_hndlr(ns, zone, nvme_get_zone_state(zone), req);
3983     } else {
3984         if (proc_mask & NVME_PROC_CLOSED_ZONES) {
3985             QTAILQ_FOREACH_SAFE(zone, &ns->closed_zones, entry, next) {
3986                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3987                                              req);
3988                 if (status && status != NVME_NO_COMPLETE) {
3989                     goto out;
3990                 }
3991             }
3992         }
3993         if (proc_mask & NVME_PROC_OPENED_ZONES) {
3994             QTAILQ_FOREACH_SAFE(zone, &ns->imp_open_zones, entry, next) {
3995                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3996                                              req);
3997                 if (status && status != NVME_NO_COMPLETE) {
3998                     goto out;
3999                 }
4000             }
4001 
4002             QTAILQ_FOREACH_SAFE(zone, &ns->exp_open_zones, entry, next) {
4003                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
4004                                              req);
4005                 if (status && status != NVME_NO_COMPLETE) {
4006                     goto out;
4007                 }
4008             }
4009         }
4010         if (proc_mask & NVME_PROC_FULL_ZONES) {
4011             QTAILQ_FOREACH_SAFE(zone, &ns->full_zones, entry, next) {
4012                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
4013                                              req);
4014                 if (status && status != NVME_NO_COMPLETE) {
4015                     goto out;
4016                 }
4017             }
4018         }
4019 
4020         if (proc_mask & NVME_PROC_READ_ONLY_ZONES) {
4021             for (i = 0; i < ns->num_zones; i++, zone++) {
4022                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
4023                                              req);
4024                 if (status && status != NVME_NO_COMPLETE) {
4025                     goto out;
4026                 }
4027             }
4028         }
4029     }
4030 
4031 out:
4032     return status;
4033 }
4034 
4035 typedef struct NvmeZoneResetAIOCB {
4036     BlockAIOCB common;
4037     BlockAIOCB *aiocb;
4038     NvmeRequest *req;
4039     int ret;
4040 
4041     bool all;
4042     int idx;
4043     NvmeZone *zone;
4044 } NvmeZoneResetAIOCB;
4045 
4046 static void nvme_zone_reset_cancel(BlockAIOCB *aiocb)
4047 {
4048     NvmeZoneResetAIOCB *iocb = container_of(aiocb, NvmeZoneResetAIOCB, common);
4049     NvmeRequest *req = iocb->req;
4050     NvmeNamespace *ns = req->ns;
4051 
4052     iocb->idx = ns->num_zones;
4053 
4054     iocb->ret = -ECANCELED;
4055 
4056     if (iocb->aiocb) {
4057         blk_aio_cancel_async(iocb->aiocb);
4058         iocb->aiocb = NULL;
4059     }
4060 }
4061 
4062 static const AIOCBInfo nvme_zone_reset_aiocb_info = {
4063     .aiocb_size = sizeof(NvmeZoneResetAIOCB),
4064     .cancel_async = nvme_zone_reset_cancel,
4065 };
4066 
4067 static void nvme_zone_reset_cb(void *opaque, int ret);
4068 
4069 static void nvme_zone_reset_epilogue_cb(void *opaque, int ret)
4070 {
4071     NvmeZoneResetAIOCB *iocb = opaque;
4072     NvmeRequest *req = iocb->req;
4073     NvmeNamespace *ns = req->ns;
4074     int64_t moff;
4075     int count;
4076 
4077     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
4078         goto out;
4079     }
4080 
4081     moff = nvme_moff(ns, iocb->zone->d.zslba);
4082     count = nvme_m2b(ns, ns->zone_size);
4083 
4084     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, moff, count,
4085                                         BDRV_REQ_MAY_UNMAP,
4086                                         nvme_zone_reset_cb, iocb);
4087     return;
4088 
4089 out:
4090     nvme_zone_reset_cb(iocb, ret);
4091 }
4092 
4093 static void nvme_zone_reset_cb(void *opaque, int ret)
4094 {
4095     NvmeZoneResetAIOCB *iocb = opaque;
4096     NvmeRequest *req = iocb->req;
4097     NvmeNamespace *ns = req->ns;
4098 
4099     if (iocb->ret < 0) {
4100         goto done;
4101     } else if (ret < 0) {
4102         iocb->ret = ret;
4103         goto done;
4104     }
4105 
4106     if (iocb->zone) {
4107         nvme_zrm_reset(ns, iocb->zone);
4108 
4109         if (!iocb->all) {
4110             goto done;
4111         }
4112     }
4113 
4114     while (iocb->idx < ns->num_zones) {
4115         NvmeZone *zone = &ns->zone_array[iocb->idx++];
4116 
4117         switch (nvme_get_zone_state(zone)) {
4118         case NVME_ZONE_STATE_EMPTY:
4119             if (!iocb->all) {
4120                 goto done;
4121             }
4122 
4123             continue;
4124 
4125         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
4126         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
4127         case NVME_ZONE_STATE_CLOSED:
4128         case NVME_ZONE_STATE_FULL:
4129             iocb->zone = zone;
4130             break;
4131 
4132         default:
4133             continue;
4134         }
4135 
4136         trace_pci_nvme_zns_zone_reset(zone->d.zslba);
4137 
4138         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk,
4139                                             nvme_l2b(ns, zone->d.zslba),
4140                                             nvme_l2b(ns, ns->zone_size),
4141                                             BDRV_REQ_MAY_UNMAP,
4142                                             nvme_zone_reset_epilogue_cb,
4143                                             iocb);
4144         return;
4145     }
4146 
4147 done:
4148     iocb->aiocb = NULL;
4149 
4150     iocb->common.cb(iocb->common.opaque, iocb->ret);
4151     qemu_aio_unref(iocb);
4152 }
4153 
4154 static uint16_t nvme_zone_mgmt_send_zrwa_flush(NvmeCtrl *n, NvmeZone *zone,
4155                                                uint64_t elba, NvmeRequest *req)
4156 {
4157     NvmeNamespace *ns = req->ns;
4158     uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
4159     uint64_t wp = zone->d.wp;
4160     uint32_t nlb = elba - wp + 1;
4161     uint16_t status;
4162 
4163 
4164     if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
4165         return NVME_INVALID_ZONE_OP | NVME_DNR;
4166     }
4167 
4168     if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
4169         return NVME_INVALID_FIELD | NVME_DNR;
4170     }
4171 
4172     if (elba < wp || elba > wp + ns->zns.zrwas) {
4173         return NVME_ZONE_BOUNDARY_ERROR | NVME_DNR;
4174     }
4175 
4176     if (nlb % ns->zns.zrwafg) {
4177         return NVME_INVALID_FIELD | NVME_DNR;
4178     }
4179 
4180     status = nvme_zrm_auto(n, ns, zone);
4181     if (status) {
4182         return status;
4183     }
4184 
4185     zone->w_ptr += nlb;
4186 
4187     nvme_advance_zone_wp(ns, zone, nlb);
4188 
4189     return NVME_SUCCESS;
4190 }
4191 
4192 static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4193 {
4194     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
4195     NvmeNamespace *ns = req->ns;
4196     NvmeZone *zone;
4197     NvmeZoneResetAIOCB *iocb;
4198     uint8_t *zd_ext;
4199     uint64_t slba = 0;
4200     uint32_t zone_idx = 0;
4201     uint16_t status;
4202     uint8_t action = cmd->zsa;
4203     bool all;
4204     enum NvmeZoneProcessingMask proc_mask = NVME_PROC_CURRENT_ZONE;
4205 
4206     all = cmd->zsflags & NVME_ZSFLAG_SELECT_ALL;
4207 
4208     req->status = NVME_SUCCESS;
4209 
4210     if (!all) {
4211         status = nvme_get_mgmt_zone_slba_idx(ns, &req->cmd, &slba, &zone_idx);
4212         if (status) {
4213             return status;
4214         }
4215     }
4216 
4217     zone = &ns->zone_array[zone_idx];
4218     if (slba != zone->d.zslba && action != NVME_ZONE_ACTION_ZRWA_FLUSH) {
4219         trace_pci_nvme_err_unaligned_zone_cmd(action, slba, zone->d.zslba);
4220         return NVME_INVALID_FIELD | NVME_DNR;
4221     }
4222 
4223     switch (action) {
4224 
4225     case NVME_ZONE_ACTION_OPEN:
4226         if (all) {
4227             proc_mask = NVME_PROC_CLOSED_ZONES;
4228         }
4229         trace_pci_nvme_open_zone(slba, zone_idx, all);
4230         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_open_zone, req);
4231         break;
4232 
4233     case NVME_ZONE_ACTION_CLOSE:
4234         if (all) {
4235             proc_mask = NVME_PROC_OPENED_ZONES;
4236         }
4237         trace_pci_nvme_close_zone(slba, zone_idx, all);
4238         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_close_zone, req);
4239         break;
4240 
4241     case NVME_ZONE_ACTION_FINISH:
4242         if (all) {
4243             proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES;
4244         }
4245         trace_pci_nvme_finish_zone(slba, zone_idx, all);
4246         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_finish_zone, req);
4247         break;
4248 
4249     case NVME_ZONE_ACTION_RESET:
4250         trace_pci_nvme_reset_zone(slba, zone_idx, all);
4251 
4252         iocb = blk_aio_get(&nvme_zone_reset_aiocb_info, ns->blkconf.blk,
4253                            nvme_misc_cb, req);
4254 
4255         iocb->req = req;
4256         iocb->ret = 0;
4257         iocb->all = all;
4258         iocb->idx = zone_idx;
4259         iocb->zone = NULL;
4260 
4261         req->aiocb = &iocb->common;
4262         nvme_zone_reset_cb(iocb, 0);
4263 
4264         return NVME_NO_COMPLETE;
4265 
4266     case NVME_ZONE_ACTION_OFFLINE:
4267         if (all) {
4268             proc_mask = NVME_PROC_READ_ONLY_ZONES;
4269         }
4270         trace_pci_nvme_offline_zone(slba, zone_idx, all);
4271         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_offline_zone, req);
4272         break;
4273 
4274     case NVME_ZONE_ACTION_SET_ZD_EXT:
4275         trace_pci_nvme_set_descriptor_extension(slba, zone_idx);
4276         if (all || !ns->params.zd_extension_size) {
4277             return NVME_INVALID_FIELD | NVME_DNR;
4278         }
4279         zd_ext = nvme_get_zd_extension(ns, zone_idx);
4280         status = nvme_h2c(n, zd_ext, ns->params.zd_extension_size, req);
4281         if (status) {
4282             trace_pci_nvme_err_zd_extension_map_error(zone_idx);
4283             return status;
4284         }
4285 
4286         status = nvme_set_zd_ext(ns, zone);
4287         if (status == NVME_SUCCESS) {
4288             trace_pci_nvme_zd_extension_set(zone_idx);
4289             return status;
4290         }
4291         break;
4292 
4293     case NVME_ZONE_ACTION_ZRWA_FLUSH:
4294         if (all) {
4295             return NVME_INVALID_FIELD | NVME_DNR;
4296         }
4297 
4298         return nvme_zone_mgmt_send_zrwa_flush(n, zone, slba, req);
4299 
4300     default:
4301         trace_pci_nvme_err_invalid_mgmt_action(action);
4302         status = NVME_INVALID_FIELD;
4303     }
4304 
4305     if (status == NVME_ZONE_INVAL_TRANSITION) {
4306         trace_pci_nvme_err_invalid_zone_state_transition(action, slba,
4307                                                          zone->d.za);
4308     }
4309     if (status) {
4310         status |= NVME_DNR;
4311     }
4312 
4313     return status;
4314 }
4315 
4316 static bool nvme_zone_matches_filter(uint32_t zafs, NvmeZone *zl)
4317 {
4318     NvmeZoneState zs = nvme_get_zone_state(zl);
4319 
4320     switch (zafs) {
4321     case NVME_ZONE_REPORT_ALL:
4322         return true;
4323     case NVME_ZONE_REPORT_EMPTY:
4324         return zs == NVME_ZONE_STATE_EMPTY;
4325     case NVME_ZONE_REPORT_IMPLICITLY_OPEN:
4326         return zs == NVME_ZONE_STATE_IMPLICITLY_OPEN;
4327     case NVME_ZONE_REPORT_EXPLICITLY_OPEN:
4328         return zs == NVME_ZONE_STATE_EXPLICITLY_OPEN;
4329     case NVME_ZONE_REPORT_CLOSED:
4330         return zs == NVME_ZONE_STATE_CLOSED;
4331     case NVME_ZONE_REPORT_FULL:
4332         return zs == NVME_ZONE_STATE_FULL;
4333     case NVME_ZONE_REPORT_READ_ONLY:
4334         return zs == NVME_ZONE_STATE_READ_ONLY;
4335     case NVME_ZONE_REPORT_OFFLINE:
4336         return zs == NVME_ZONE_STATE_OFFLINE;
4337     default:
4338         return false;
4339     }
4340 }
4341 
4342 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4343 {
4344     NvmeCmd *cmd = &req->cmd;
4345     NvmeNamespace *ns = req->ns;
4346     /* cdw12 is zero-based number of dwords to return. Convert to bytes */
4347     uint32_t data_size = (le32_to_cpu(cmd->cdw12) + 1) << 2;
4348     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
4349     uint32_t zone_idx, zra, zrasf, partial;
4350     uint64_t max_zones, nr_zones = 0;
4351     uint16_t status;
4352     uint64_t slba;
4353     NvmeZoneDescr *z;
4354     NvmeZone *zone;
4355     NvmeZoneReportHeader *header;
4356     void *buf, *buf_p;
4357     size_t zone_entry_sz;
4358     int i;
4359 
4360     req->status = NVME_SUCCESS;
4361 
4362     status = nvme_get_mgmt_zone_slba_idx(ns, cmd, &slba, &zone_idx);
4363     if (status) {
4364         return status;
4365     }
4366 
4367     zra = dw13 & 0xff;
4368     if (zra != NVME_ZONE_REPORT && zra != NVME_ZONE_REPORT_EXTENDED) {
4369         return NVME_INVALID_FIELD | NVME_DNR;
4370     }
4371     if (zra == NVME_ZONE_REPORT_EXTENDED && !ns->params.zd_extension_size) {
4372         return NVME_INVALID_FIELD | NVME_DNR;
4373     }
4374 
4375     zrasf = (dw13 >> 8) & 0xff;
4376     if (zrasf > NVME_ZONE_REPORT_OFFLINE) {
4377         return NVME_INVALID_FIELD | NVME_DNR;
4378     }
4379 
4380     if (data_size < sizeof(NvmeZoneReportHeader)) {
4381         return NVME_INVALID_FIELD | NVME_DNR;
4382     }
4383 
4384     status = nvme_check_mdts(n, data_size);
4385     if (status) {
4386         return status;
4387     }
4388 
4389     partial = (dw13 >> 16) & 0x01;
4390 
4391     zone_entry_sz = sizeof(NvmeZoneDescr);
4392     if (zra == NVME_ZONE_REPORT_EXTENDED) {
4393         zone_entry_sz += ns->params.zd_extension_size;
4394     }
4395 
4396     max_zones = (data_size - sizeof(NvmeZoneReportHeader)) / zone_entry_sz;
4397     buf = g_malloc0(data_size);
4398 
4399     zone = &ns->zone_array[zone_idx];
4400     for (i = zone_idx; i < ns->num_zones; i++) {
4401         if (partial && nr_zones >= max_zones) {
4402             break;
4403         }
4404         if (nvme_zone_matches_filter(zrasf, zone++)) {
4405             nr_zones++;
4406         }
4407     }
4408     header = buf;
4409     header->nr_zones = cpu_to_le64(nr_zones);
4410 
4411     buf_p = buf + sizeof(NvmeZoneReportHeader);
4412     for (; zone_idx < ns->num_zones && max_zones > 0; zone_idx++) {
4413         zone = &ns->zone_array[zone_idx];
4414         if (nvme_zone_matches_filter(zrasf, zone)) {
4415             z = buf_p;
4416             buf_p += sizeof(NvmeZoneDescr);
4417 
4418             z->zt = zone->d.zt;
4419             z->zs = zone->d.zs;
4420             z->zcap = cpu_to_le64(zone->d.zcap);
4421             z->zslba = cpu_to_le64(zone->d.zslba);
4422             z->za = zone->d.za;
4423 
4424             if (nvme_wp_is_valid(zone)) {
4425                 z->wp = cpu_to_le64(zone->d.wp);
4426             } else {
4427                 z->wp = cpu_to_le64(~0ULL);
4428             }
4429 
4430             if (zra == NVME_ZONE_REPORT_EXTENDED) {
4431                 if (zone->d.za & NVME_ZA_ZD_EXT_VALID) {
4432                     memcpy(buf_p, nvme_get_zd_extension(ns, zone_idx),
4433                            ns->params.zd_extension_size);
4434                 }
4435                 buf_p += ns->params.zd_extension_size;
4436             }
4437 
4438             max_zones--;
4439         }
4440     }
4441 
4442     status = nvme_c2h(n, (uint8_t *)buf, data_size, req);
4443 
4444     g_free(buf);
4445 
4446     return status;
4447 }
4448 
4449 static uint16_t nvme_io_mgmt_recv_ruhs(NvmeCtrl *n, NvmeRequest *req,
4450                                        size_t len)
4451 {
4452     NvmeNamespace *ns = req->ns;
4453     NvmeEnduranceGroup *endgrp;
4454     NvmeRuhStatus *hdr;
4455     NvmeRuhStatusDescr *ruhsd;
4456     unsigned int nruhsd;
4457     uint16_t rg, ph, *ruhid;
4458     size_t trans_len;
4459     g_autofree uint8_t *buf = NULL;
4460 
4461     if (!n->subsys) {
4462         return NVME_INVALID_FIELD | NVME_DNR;
4463     }
4464 
4465     if (ns->params.nsid == 0 || ns->params.nsid == 0xffffffff) {
4466         return NVME_INVALID_NSID | NVME_DNR;
4467     }
4468 
4469     if (!n->subsys->endgrp.fdp.enabled) {
4470         return NVME_FDP_DISABLED | NVME_DNR;
4471     }
4472 
4473     endgrp = ns->endgrp;
4474 
4475     nruhsd = ns->fdp.nphs * endgrp->fdp.nrg;
4476     trans_len = sizeof(NvmeRuhStatus) + nruhsd * sizeof(NvmeRuhStatusDescr);
4477     buf = g_malloc(trans_len);
4478 
4479     trans_len = MIN(trans_len, len);
4480 
4481     hdr = (NvmeRuhStatus *)buf;
4482     ruhsd = (NvmeRuhStatusDescr *)(buf + sizeof(NvmeRuhStatus));
4483 
4484     hdr->nruhsd = cpu_to_le16(nruhsd);
4485 
4486     ruhid = ns->fdp.phs;
4487 
4488     for (ph = 0; ph < ns->fdp.nphs; ph++, ruhid++) {
4489         NvmeRuHandle *ruh = &endgrp->fdp.ruhs[*ruhid];
4490 
4491         for (rg = 0; rg < endgrp->fdp.nrg; rg++, ruhsd++) {
4492             uint16_t pid = nvme_make_pid(ns, rg, ph);
4493 
4494             ruhsd->pid = cpu_to_le16(pid);
4495             ruhsd->ruhid = *ruhid;
4496             ruhsd->earutr = 0;
4497             ruhsd->ruamw = cpu_to_le64(ruh->rus[rg].ruamw);
4498         }
4499     }
4500 
4501     return nvme_c2h(n, buf, trans_len, req);
4502 }
4503 
4504 static uint16_t nvme_io_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4505 {
4506     NvmeCmd *cmd = &req->cmd;
4507     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4508     uint32_t numd = le32_to_cpu(cmd->cdw11);
4509     uint8_t mo = (cdw10 & 0xff);
4510     size_t len = (numd + 1) << 2;
4511 
4512     switch (mo) {
4513     case NVME_IOMR_MO_NOP:
4514         return 0;
4515     case NVME_IOMR_MO_RUH_STATUS:
4516         return nvme_io_mgmt_recv_ruhs(n, req, len);
4517     default:
4518         return NVME_INVALID_FIELD | NVME_DNR;
4519     };
4520 }
4521 
4522 static uint16_t nvme_io_mgmt_send_ruh_update(NvmeCtrl *n, NvmeRequest *req)
4523 {
4524     NvmeCmd *cmd = &req->cmd;
4525     NvmeNamespace *ns = req->ns;
4526     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4527     uint16_t ret = NVME_SUCCESS;
4528     uint32_t npid = (cdw10 >> 16) + 1;
4529     unsigned int i = 0;
4530     g_autofree uint16_t *pids = NULL;
4531     uint32_t maxnpid;
4532 
4533     if (!ns->endgrp || !ns->endgrp->fdp.enabled) {
4534         return NVME_FDP_DISABLED | NVME_DNR;
4535     }
4536 
4537     maxnpid = n->subsys->endgrp.fdp.nrg * n->subsys->endgrp.fdp.nruh;
4538 
4539     if (unlikely(npid >= MIN(NVME_FDP_MAXPIDS, maxnpid))) {
4540         return NVME_INVALID_FIELD | NVME_DNR;
4541     }
4542 
4543     pids = g_new(uint16_t, npid);
4544 
4545     ret = nvme_h2c(n, pids, npid * sizeof(uint16_t), req);
4546     if (ret) {
4547         return ret;
4548     }
4549 
4550     for (; i < npid; i++) {
4551         if (!nvme_update_ruh(n, ns, pids[i])) {
4552             return NVME_INVALID_FIELD | NVME_DNR;
4553         }
4554     }
4555 
4556     return ret;
4557 }
4558 
4559 static uint16_t nvme_io_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4560 {
4561     NvmeCmd *cmd = &req->cmd;
4562     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4563     uint8_t mo = (cdw10 & 0xff);
4564 
4565     switch (mo) {
4566     case NVME_IOMS_MO_NOP:
4567         return 0;
4568     case NVME_IOMS_MO_RUH_UPDATE:
4569         return nvme_io_mgmt_send_ruh_update(n, req);
4570     default:
4571         return NVME_INVALID_FIELD | NVME_DNR;
4572     };
4573 }
4574 
4575 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
4576 {
4577     NvmeNamespace *ns;
4578     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4579 
4580     trace_pci_nvme_io_cmd(nvme_cid(req), nsid, nvme_sqid(req),
4581                           req->cmd.opcode, nvme_io_opc_str(req->cmd.opcode));
4582 
4583     /*
4584      * In the base NVM command set, Flush may apply to all namespaces
4585      * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4586      * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4587      *
4588      * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4589      * opcode with a specific command since we cannot determine a unique I/O
4590      * command set. Opcode 0h could have any other meaning than something
4591      * equivalent to flushing and say it DOES have completely different
4592      * semantics in some other command set - does an NSID of FFFFFFFFh then
4593      * mean "for all namespaces, apply whatever command set specific command
4594      * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4595      * whatever command that uses the 0h opcode if, and only if, it allows NSID
4596      * to be FFFFFFFFh"?
4597      *
4598      * Anyway (and luckily), for now, we do not care about this since the
4599      * device only supports namespace types that includes the NVM Flush command
4600      * (NVM and Zoned), so always do an NVM Flush.
4601      */
4602 
4603     if (req->cmd.opcode == NVME_CMD_FLUSH) {
4604         return nvme_flush(n, req);
4605     }
4606 
4607     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4608         return NVME_INVALID_NSID | NVME_DNR;
4609     }
4610 
4611     ns = nvme_ns(n, nsid);
4612     if (unlikely(!ns)) {
4613         return NVME_INVALID_FIELD | NVME_DNR;
4614     }
4615 
4616     if (!(ns->iocs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
4617         trace_pci_nvme_err_invalid_opc(req->cmd.opcode);
4618         return NVME_INVALID_OPCODE | NVME_DNR;
4619     }
4620 
4621     if (ns->status) {
4622         return ns->status;
4623     }
4624 
4625     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
4626         return NVME_INVALID_FIELD;
4627     }
4628 
4629     req->ns = ns;
4630 
4631     switch (req->cmd.opcode) {
4632     case NVME_CMD_WRITE_ZEROES:
4633         return nvme_write_zeroes(n, req);
4634     case NVME_CMD_ZONE_APPEND:
4635         return nvme_zone_append(n, req);
4636     case NVME_CMD_WRITE:
4637         return nvme_write(n, req);
4638     case NVME_CMD_READ:
4639         return nvme_read(n, req);
4640     case NVME_CMD_COMPARE:
4641         return nvme_compare(n, req);
4642     case NVME_CMD_DSM:
4643         return nvme_dsm(n, req);
4644     case NVME_CMD_VERIFY:
4645         return nvme_verify(n, req);
4646     case NVME_CMD_COPY:
4647         return nvme_copy(n, req);
4648     case NVME_CMD_ZONE_MGMT_SEND:
4649         return nvme_zone_mgmt_send(n, req);
4650     case NVME_CMD_ZONE_MGMT_RECV:
4651         return nvme_zone_mgmt_recv(n, req);
4652     case NVME_CMD_IO_MGMT_RECV:
4653         return nvme_io_mgmt_recv(n, req);
4654     case NVME_CMD_IO_MGMT_SEND:
4655         return nvme_io_mgmt_send(n, req);
4656     default:
4657         assert(false);
4658     }
4659 
4660     return NVME_INVALID_OPCODE | NVME_DNR;
4661 }
4662 
4663 static void nvme_cq_notifier(EventNotifier *e)
4664 {
4665     NvmeCQueue *cq = container_of(e, NvmeCQueue, notifier);
4666     NvmeCtrl *n = cq->ctrl;
4667 
4668     if (!event_notifier_test_and_clear(e)) {
4669         return;
4670     }
4671 
4672     nvme_update_cq_head(cq);
4673 
4674     if (cq->tail == cq->head) {
4675         if (cq->irq_enabled) {
4676             n->cq_pending--;
4677         }
4678 
4679         nvme_irq_deassert(n, cq);
4680     }
4681 
4682     qemu_bh_schedule(cq->bh);
4683 }
4684 
4685 static int nvme_init_cq_ioeventfd(NvmeCQueue *cq)
4686 {
4687     NvmeCtrl *n = cq->ctrl;
4688     uint16_t offset = (cq->cqid << 3) + (1 << 2);
4689     int ret;
4690 
4691     ret = event_notifier_init(&cq->notifier, 0);
4692     if (ret < 0) {
4693         return ret;
4694     }
4695 
4696     event_notifier_set_handler(&cq->notifier, nvme_cq_notifier);
4697     memory_region_add_eventfd(&n->iomem,
4698                               0x1000 + offset, 4, false, 0, &cq->notifier);
4699 
4700     return 0;
4701 }
4702 
4703 static void nvme_sq_notifier(EventNotifier *e)
4704 {
4705     NvmeSQueue *sq = container_of(e, NvmeSQueue, notifier);
4706 
4707     if (!event_notifier_test_and_clear(e)) {
4708         return;
4709     }
4710 
4711     nvme_process_sq(sq);
4712 }
4713 
4714 static int nvme_init_sq_ioeventfd(NvmeSQueue *sq)
4715 {
4716     NvmeCtrl *n = sq->ctrl;
4717     uint16_t offset = sq->sqid << 3;
4718     int ret;
4719 
4720     ret = event_notifier_init(&sq->notifier, 0);
4721     if (ret < 0) {
4722         return ret;
4723     }
4724 
4725     event_notifier_set_handler(&sq->notifier, nvme_sq_notifier);
4726     memory_region_add_eventfd(&n->iomem,
4727                               0x1000 + offset, 4, false, 0, &sq->notifier);
4728 
4729     return 0;
4730 }
4731 
4732 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
4733 {
4734     uint16_t offset = sq->sqid << 3;
4735 
4736     n->sq[sq->sqid] = NULL;
4737     qemu_bh_delete(sq->bh);
4738     if (sq->ioeventfd_enabled) {
4739         memory_region_del_eventfd(&n->iomem,
4740                                   0x1000 + offset, 4, false, 0, &sq->notifier);
4741         event_notifier_set_handler(&sq->notifier, NULL);
4742         event_notifier_cleanup(&sq->notifier);
4743     }
4744     g_free(sq->io_req);
4745     if (sq->sqid) {
4746         g_free(sq);
4747     }
4748 }
4749 
4750 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
4751 {
4752     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4753     NvmeRequest *r, *next;
4754     NvmeSQueue *sq;
4755     NvmeCQueue *cq;
4756     uint16_t qid = le16_to_cpu(c->qid);
4757 
4758     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
4759         trace_pci_nvme_err_invalid_del_sq(qid);
4760         return NVME_INVALID_QID | NVME_DNR;
4761     }
4762 
4763     trace_pci_nvme_del_sq(qid);
4764 
4765     sq = n->sq[qid];
4766     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
4767         r = QTAILQ_FIRST(&sq->out_req_list);
4768         assert(r->aiocb);
4769         blk_aio_cancel(r->aiocb);
4770     }
4771 
4772     assert(QTAILQ_EMPTY(&sq->out_req_list));
4773 
4774     if (!nvme_check_cqid(n, sq->cqid)) {
4775         cq = n->cq[sq->cqid];
4776         QTAILQ_REMOVE(&cq->sq_list, sq, entry);
4777 
4778         nvme_post_cqes(cq);
4779         QTAILQ_FOREACH_SAFE(r, &cq->req_list, entry, next) {
4780             if (r->sq == sq) {
4781                 QTAILQ_REMOVE(&cq->req_list, r, entry);
4782                 QTAILQ_INSERT_TAIL(&sq->req_list, r, entry);
4783             }
4784         }
4785     }
4786 
4787     nvme_free_sq(sq, n);
4788     return NVME_SUCCESS;
4789 }
4790 
4791 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
4792                          uint16_t sqid, uint16_t cqid, uint16_t size)
4793 {
4794     int i;
4795     NvmeCQueue *cq;
4796 
4797     sq->ctrl = n;
4798     sq->dma_addr = dma_addr;
4799     sq->sqid = sqid;
4800     sq->size = size;
4801     sq->cqid = cqid;
4802     sq->head = sq->tail = 0;
4803     sq->io_req = g_new0(NvmeRequest, sq->size);
4804 
4805     QTAILQ_INIT(&sq->req_list);
4806     QTAILQ_INIT(&sq->out_req_list);
4807     for (i = 0; i < sq->size; i++) {
4808         sq->io_req[i].sq = sq;
4809         QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
4810     }
4811 
4812     sq->bh = qemu_bh_new_guarded(nvme_process_sq, sq,
4813                                  &DEVICE(sq->ctrl)->mem_reentrancy_guard);
4814 
4815     if (n->dbbuf_enabled) {
4816         sq->db_addr = n->dbbuf_dbs + (sqid << 3);
4817         sq->ei_addr = n->dbbuf_eis + (sqid << 3);
4818 
4819         if (n->params.ioeventfd && sq->sqid != 0) {
4820             if (!nvme_init_sq_ioeventfd(sq)) {
4821                 sq->ioeventfd_enabled = true;
4822             }
4823         }
4824     }
4825 
4826     assert(n->cq[cqid]);
4827     cq = n->cq[cqid];
4828     QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
4829     n->sq[sqid] = sq;
4830 }
4831 
4832 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeRequest *req)
4833 {
4834     NvmeSQueue *sq;
4835     NvmeCreateSq *c = (NvmeCreateSq *)&req->cmd;
4836 
4837     uint16_t cqid = le16_to_cpu(c->cqid);
4838     uint16_t sqid = le16_to_cpu(c->sqid);
4839     uint16_t qsize = le16_to_cpu(c->qsize);
4840     uint16_t qflags = le16_to_cpu(c->sq_flags);
4841     uint64_t prp1 = le64_to_cpu(c->prp1);
4842 
4843     trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
4844 
4845     if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
4846         trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
4847         return NVME_INVALID_CQID | NVME_DNR;
4848     }
4849     if (unlikely(!sqid || sqid > n->conf_ioqpairs || n->sq[sqid] != NULL)) {
4850         trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
4851         return NVME_INVALID_QID | NVME_DNR;
4852     }
4853     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
4854         trace_pci_nvme_err_invalid_create_sq_size(qsize);
4855         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4856     }
4857     if (unlikely(prp1 & (n->page_size - 1))) {
4858         trace_pci_nvme_err_invalid_create_sq_addr(prp1);
4859         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4860     }
4861     if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
4862         trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
4863         return NVME_INVALID_FIELD | NVME_DNR;
4864     }
4865     sq = g_malloc0(sizeof(*sq));
4866     nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
4867     return NVME_SUCCESS;
4868 }
4869 
4870 struct nvme_stats {
4871     uint64_t units_read;
4872     uint64_t units_written;
4873     uint64_t read_commands;
4874     uint64_t write_commands;
4875 };
4876 
4877 static void nvme_set_blk_stats(NvmeNamespace *ns, struct nvme_stats *stats)
4878 {
4879     BlockAcctStats *s = blk_get_stats(ns->blkconf.blk);
4880 
4881     stats->units_read += s->nr_bytes[BLOCK_ACCT_READ];
4882     stats->units_written += s->nr_bytes[BLOCK_ACCT_WRITE];
4883     stats->read_commands += s->nr_ops[BLOCK_ACCT_READ];
4884     stats->write_commands += s->nr_ops[BLOCK_ACCT_WRITE];
4885 }
4886 
4887 static uint16_t nvme_smart_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4888                                 uint64_t off, NvmeRequest *req)
4889 {
4890     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4891     struct nvme_stats stats = { 0 };
4892     NvmeSmartLog smart = { 0 };
4893     uint32_t trans_len;
4894     NvmeNamespace *ns;
4895     time_t current_ms;
4896     uint64_t u_read, u_written;
4897 
4898     if (off >= sizeof(smart)) {
4899         return NVME_INVALID_FIELD | NVME_DNR;
4900     }
4901 
4902     if (nsid != 0xffffffff) {
4903         ns = nvme_ns(n, nsid);
4904         if (!ns) {
4905             return NVME_INVALID_NSID | NVME_DNR;
4906         }
4907         nvme_set_blk_stats(ns, &stats);
4908     } else {
4909         int i;
4910 
4911         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4912             ns = nvme_ns(n, i);
4913             if (!ns) {
4914                 continue;
4915             }
4916             nvme_set_blk_stats(ns, &stats);
4917         }
4918     }
4919 
4920     trans_len = MIN(sizeof(smart) - off, buf_len);
4921     smart.critical_warning = n->smart_critical_warning;
4922 
4923     u_read = DIV_ROUND_UP(stats.units_read >> BDRV_SECTOR_BITS, 1000);
4924     u_written = DIV_ROUND_UP(stats.units_written >> BDRV_SECTOR_BITS, 1000);
4925 
4926     smart.data_units_read[0] = cpu_to_le64(u_read);
4927     smart.data_units_written[0] = cpu_to_le64(u_written);
4928     smart.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4929     smart.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4930 
4931     smart.temperature = cpu_to_le16(n->temperature);
4932 
4933     if ((n->temperature >= n->features.temp_thresh_hi) ||
4934         (n->temperature <= n->features.temp_thresh_low)) {
4935         smart.critical_warning |= NVME_SMART_TEMPERATURE;
4936     }
4937 
4938     current_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4939     smart.power_on_hours[0] =
4940         cpu_to_le64((((current_ms - n->starttime_ms) / 1000) / 60) / 60);
4941 
4942     if (!rae) {
4943         nvme_clear_events(n, NVME_AER_TYPE_SMART);
4944     }
4945 
4946     return nvme_c2h(n, (uint8_t *) &smart + off, trans_len, req);
4947 }
4948 
4949 static uint16_t nvme_endgrp_info(NvmeCtrl *n,  uint8_t rae, uint32_t buf_len,
4950                                  uint64_t off, NvmeRequest *req)
4951 {
4952     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
4953     uint16_t endgrpid = (dw11 >> 16) & 0xffff;
4954     struct nvme_stats stats = {};
4955     NvmeEndGrpLog info = {};
4956     int i;
4957 
4958     if (!n->subsys || endgrpid != 0x1) {
4959         return NVME_INVALID_FIELD | NVME_DNR;
4960     }
4961 
4962     if (off >= sizeof(info)) {
4963         return NVME_INVALID_FIELD | NVME_DNR;
4964     }
4965 
4966     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4967         NvmeNamespace *ns = nvme_subsys_ns(n->subsys, i);
4968         if (!ns) {
4969             continue;
4970         }
4971 
4972         nvme_set_blk_stats(ns, &stats);
4973     }
4974 
4975     info.data_units_read[0] =
4976         cpu_to_le64(DIV_ROUND_UP(stats.units_read / 1000000000, 1000000000));
4977     info.data_units_written[0] =
4978         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4979     info.media_units_written[0] =
4980         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4981 
4982     info.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4983     info.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4984 
4985     buf_len = MIN(sizeof(info) - off, buf_len);
4986 
4987     return nvme_c2h(n, (uint8_t *)&info + off, buf_len, req);
4988 }
4989 
4990 
4991 static uint16_t nvme_fw_log_info(NvmeCtrl *n, uint32_t buf_len, uint64_t off,
4992                                  NvmeRequest *req)
4993 {
4994     uint32_t trans_len;
4995     NvmeFwSlotInfoLog fw_log = {
4996         .afi = 0x1,
4997     };
4998 
4999     if (off >= sizeof(fw_log)) {
5000         return NVME_INVALID_FIELD | NVME_DNR;
5001     }
5002 
5003     strpadcpy((char *)&fw_log.frs1, sizeof(fw_log.frs1), "1.0", ' ');
5004     trans_len = MIN(sizeof(fw_log) - off, buf_len);
5005 
5006     return nvme_c2h(n, (uint8_t *) &fw_log + off, trans_len, req);
5007 }
5008 
5009 static uint16_t nvme_error_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
5010                                 uint64_t off, NvmeRequest *req)
5011 {
5012     uint32_t trans_len;
5013     NvmeErrorLog errlog;
5014 
5015     if (off >= sizeof(errlog)) {
5016         return NVME_INVALID_FIELD | NVME_DNR;
5017     }
5018 
5019     if (!rae) {
5020         nvme_clear_events(n, NVME_AER_TYPE_ERROR);
5021     }
5022 
5023     memset(&errlog, 0x0, sizeof(errlog));
5024     trans_len = MIN(sizeof(errlog) - off, buf_len);
5025 
5026     return nvme_c2h(n, (uint8_t *)&errlog, trans_len, req);
5027 }
5028 
5029 static uint16_t nvme_changed_nslist(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
5030                                     uint64_t off, NvmeRequest *req)
5031 {
5032     uint32_t nslist[1024];
5033     uint32_t trans_len;
5034     int i = 0;
5035     uint32_t nsid;
5036 
5037     if (off >= sizeof(nslist)) {
5038         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(nslist));
5039         return NVME_INVALID_FIELD | NVME_DNR;
5040     }
5041 
5042     memset(nslist, 0x0, sizeof(nslist));
5043     trans_len = MIN(sizeof(nslist) - off, buf_len);
5044 
5045     while ((nsid = find_first_bit(n->changed_nsids, NVME_CHANGED_NSID_SIZE)) !=
5046             NVME_CHANGED_NSID_SIZE) {
5047         /*
5048          * If more than 1024 namespaces, the first entry in the log page should
5049          * be set to FFFFFFFFh and the others to 0 as spec.
5050          */
5051         if (i == ARRAY_SIZE(nslist)) {
5052             memset(nslist, 0x0, sizeof(nslist));
5053             nslist[0] = 0xffffffff;
5054             break;
5055         }
5056 
5057         nslist[i++] = nsid;
5058         clear_bit(nsid, n->changed_nsids);
5059     }
5060 
5061     /*
5062      * Remove all the remaining list entries in case returns directly due to
5063      * more than 1024 namespaces.
5064      */
5065     if (nslist[0] == 0xffffffff) {
5066         bitmap_zero(n->changed_nsids, NVME_CHANGED_NSID_SIZE);
5067     }
5068 
5069     if (!rae) {
5070         nvme_clear_events(n, NVME_AER_TYPE_NOTICE);
5071     }
5072 
5073     return nvme_c2h(n, ((uint8_t *)nslist) + off, trans_len, req);
5074 }
5075 
5076 static uint16_t nvme_cmd_effects(NvmeCtrl *n, uint8_t csi, uint32_t buf_len,
5077                                  uint64_t off, NvmeRequest *req)
5078 {
5079     NvmeEffectsLog log = {};
5080     const uint32_t *src_iocs = NULL;
5081     uint32_t trans_len;
5082 
5083     if (off >= sizeof(log)) {
5084         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(log));
5085         return NVME_INVALID_FIELD | NVME_DNR;
5086     }
5087 
5088     switch (NVME_CC_CSS(ldl_le_p(&n->bar.cc))) {
5089     case NVME_CC_CSS_NVM:
5090         src_iocs = nvme_cse_iocs_nvm;
5091         /* fall through */
5092     case NVME_CC_CSS_ADMIN_ONLY:
5093         break;
5094     case NVME_CC_CSS_CSI:
5095         switch (csi) {
5096         case NVME_CSI_NVM:
5097             src_iocs = nvme_cse_iocs_nvm;
5098             break;
5099         case NVME_CSI_ZONED:
5100             src_iocs = nvme_cse_iocs_zoned;
5101             break;
5102         }
5103     }
5104 
5105     memcpy(log.acs, nvme_cse_acs, sizeof(nvme_cse_acs));
5106 
5107     if (src_iocs) {
5108         memcpy(log.iocs, src_iocs, sizeof(log.iocs));
5109     }
5110 
5111     trans_len = MIN(sizeof(log) - off, buf_len);
5112 
5113     return nvme_c2h(n, ((uint8_t *)&log) + off, trans_len, req);
5114 }
5115 
5116 static size_t sizeof_fdp_conf_descr(size_t nruh, size_t vss)
5117 {
5118     size_t entry_siz = sizeof(NvmeFdpDescrHdr) + nruh * sizeof(NvmeRuhDescr)
5119                        + vss;
5120     return ROUND_UP(entry_siz, 8);
5121 }
5122 
5123 static uint16_t nvme_fdp_confs(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
5124                                uint64_t off, NvmeRequest *req)
5125 {
5126     uint32_t log_size, trans_len;
5127     g_autofree uint8_t *buf = NULL;
5128     NvmeFdpDescrHdr *hdr;
5129     NvmeRuhDescr *ruhd;
5130     NvmeEnduranceGroup *endgrp;
5131     NvmeFdpConfsHdr *log;
5132     size_t nruh, fdp_descr_size;
5133     int i;
5134 
5135     if (endgrpid != 1 || !n->subsys) {
5136         return NVME_INVALID_FIELD | NVME_DNR;
5137     }
5138 
5139     endgrp = &n->subsys->endgrp;
5140 
5141     if (endgrp->fdp.enabled) {
5142         nruh = endgrp->fdp.nruh;
5143     } else {
5144         nruh = 1;
5145     }
5146 
5147     fdp_descr_size = sizeof_fdp_conf_descr(nruh, FDPVSS);
5148     log_size = sizeof(NvmeFdpConfsHdr) + fdp_descr_size;
5149 
5150     if (off >= log_size) {
5151         return NVME_INVALID_FIELD | NVME_DNR;
5152     }
5153 
5154     trans_len = MIN(log_size - off, buf_len);
5155 
5156     buf = g_malloc0(log_size);
5157     log = (NvmeFdpConfsHdr *)buf;
5158     hdr = (NvmeFdpDescrHdr *)(log + 1);
5159     ruhd = (NvmeRuhDescr *)(buf + sizeof(*log) + sizeof(*hdr));
5160 
5161     log->num_confs = cpu_to_le16(0);
5162     log->size = cpu_to_le32(log_size);
5163 
5164     hdr->descr_size = cpu_to_le16(fdp_descr_size);
5165     if (endgrp->fdp.enabled) {
5166         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, VALID, 1);
5167         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, RGIF, endgrp->fdp.rgif);
5168         hdr->nrg = cpu_to_le16(endgrp->fdp.nrg);
5169         hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5170         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5171         hdr->nnss = cpu_to_le32(NVME_MAX_NAMESPACES);
5172         hdr->runs = cpu_to_le64(endgrp->fdp.runs);
5173 
5174         for (i = 0; i < nruh; i++) {
5175             ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5176             ruhd++;
5177         }
5178     } else {
5179         /* 1 bit for RUH in PIF -> 2 RUHs max. */
5180         hdr->nrg = cpu_to_le16(1);
5181         hdr->nruh = cpu_to_le16(1);
5182         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5183         hdr->nnss = cpu_to_le32(1);
5184         hdr->runs = cpu_to_le64(96 * MiB);
5185 
5186         ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5187     }
5188 
5189     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5190 }
5191 
5192 static uint16_t nvme_fdp_ruh_usage(NvmeCtrl *n, uint32_t endgrpid,
5193                                    uint32_t dw10, uint32_t dw12,
5194                                    uint32_t buf_len, uint64_t off,
5195                                    NvmeRequest *req)
5196 {
5197     NvmeRuHandle *ruh;
5198     NvmeRuhuLog *hdr;
5199     NvmeRuhuDescr *ruhud;
5200     NvmeEnduranceGroup *endgrp;
5201     g_autofree uint8_t *buf = NULL;
5202     uint32_t log_size, trans_len;
5203     uint16_t i;
5204 
5205     if (endgrpid != 1 || !n->subsys) {
5206         return NVME_INVALID_FIELD | NVME_DNR;
5207     }
5208 
5209     endgrp = &n->subsys->endgrp;
5210 
5211     if (!endgrp->fdp.enabled) {
5212         return NVME_FDP_DISABLED | NVME_DNR;
5213     }
5214 
5215     log_size = sizeof(NvmeRuhuLog) + endgrp->fdp.nruh * sizeof(NvmeRuhuDescr);
5216 
5217     if (off >= log_size) {
5218         return NVME_INVALID_FIELD | NVME_DNR;
5219     }
5220 
5221     trans_len = MIN(log_size - off, buf_len);
5222 
5223     buf = g_malloc0(log_size);
5224     hdr = (NvmeRuhuLog *)buf;
5225     ruhud = (NvmeRuhuDescr *)(hdr + 1);
5226 
5227     ruh = endgrp->fdp.ruhs;
5228     hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5229 
5230     for (i = 0; i < endgrp->fdp.nruh; i++, ruhud++, ruh++) {
5231         ruhud->ruha = ruh->ruha;
5232     }
5233 
5234     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5235 }
5236 
5237 static uint16_t nvme_fdp_stats(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
5238                                uint64_t off, NvmeRequest *req)
5239 {
5240     NvmeEnduranceGroup *endgrp;
5241     NvmeFdpStatsLog log = {};
5242     uint32_t trans_len;
5243 
5244     if (off >= sizeof(NvmeFdpStatsLog)) {
5245         return NVME_INVALID_FIELD | NVME_DNR;
5246     }
5247 
5248     if (endgrpid != 1 || !n->subsys) {
5249         return NVME_INVALID_FIELD | NVME_DNR;
5250     }
5251 
5252     if (!n->subsys->endgrp.fdp.enabled) {
5253         return NVME_FDP_DISABLED | NVME_DNR;
5254     }
5255 
5256     endgrp = &n->subsys->endgrp;
5257 
5258     trans_len = MIN(sizeof(log) - off, buf_len);
5259 
5260     /* spec value is 128 bit, we only use 64 bit */
5261     log.hbmw[0] = cpu_to_le64(endgrp->fdp.hbmw);
5262     log.mbmw[0] = cpu_to_le64(endgrp->fdp.mbmw);
5263     log.mbe[0] = cpu_to_le64(endgrp->fdp.mbe);
5264 
5265     return nvme_c2h(n, (uint8_t *)&log + off, trans_len, req);
5266 }
5267 
5268 static uint16_t nvme_fdp_events(NvmeCtrl *n, uint32_t endgrpid,
5269                                 uint32_t buf_len, uint64_t off,
5270                                 NvmeRequest *req)
5271 {
5272     NvmeEnduranceGroup *endgrp;
5273     NvmeCmd *cmd = &req->cmd;
5274     bool host_events = (cmd->cdw10 >> 8) & 0x1;
5275     uint32_t log_size, trans_len;
5276     NvmeFdpEventBuffer *ebuf;
5277     g_autofree NvmeFdpEventsLog *elog = NULL;
5278     NvmeFdpEvent *event;
5279 
5280     if (endgrpid != 1 || !n->subsys) {
5281         return NVME_INVALID_FIELD | NVME_DNR;
5282     }
5283 
5284     endgrp = &n->subsys->endgrp;
5285 
5286     if (!endgrp->fdp.enabled) {
5287         return NVME_FDP_DISABLED | NVME_DNR;
5288     }
5289 
5290     if (host_events) {
5291         ebuf = &endgrp->fdp.host_events;
5292     } else {
5293         ebuf = &endgrp->fdp.ctrl_events;
5294     }
5295 
5296     log_size = sizeof(NvmeFdpEventsLog) + ebuf->nelems * sizeof(NvmeFdpEvent);
5297 
5298     if (off >= log_size) {
5299         return NVME_INVALID_FIELD | NVME_DNR;
5300     }
5301 
5302     trans_len = MIN(log_size - off, buf_len);
5303     elog = g_malloc0(log_size);
5304     elog->num_events = cpu_to_le32(ebuf->nelems);
5305     event = (NvmeFdpEvent *)(elog + 1);
5306 
5307     if (ebuf->nelems && ebuf->start == ebuf->next) {
5308         unsigned int nelems = (NVME_FDP_MAX_EVENTS - ebuf->start);
5309         /* wrap over, copy [start;NVME_FDP_MAX_EVENTS[ and [0; next[ */
5310         memcpy(event, &ebuf->events[ebuf->start],
5311                sizeof(NvmeFdpEvent) * nelems);
5312         memcpy(event + nelems, ebuf->events,
5313                sizeof(NvmeFdpEvent) * ebuf->next);
5314     } else if (ebuf->start < ebuf->next) {
5315         memcpy(event, &ebuf->events[ebuf->start],
5316                sizeof(NvmeFdpEvent) * (ebuf->next - ebuf->start));
5317     }
5318 
5319     return nvme_c2h(n, (uint8_t *)elog + off, trans_len, req);
5320 }
5321 
5322 static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
5323 {
5324     NvmeCmd *cmd = &req->cmd;
5325 
5326     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5327     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5328     uint32_t dw12 = le32_to_cpu(cmd->cdw12);
5329     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
5330     uint8_t  lid = dw10 & 0xff;
5331     uint8_t  lsp = (dw10 >> 8) & 0xf;
5332     uint8_t  rae = (dw10 >> 15) & 0x1;
5333     uint8_t  csi = le32_to_cpu(cmd->cdw14) >> 24;
5334     uint32_t numdl, numdu, lspi;
5335     uint64_t off, lpol, lpou;
5336     size_t   len;
5337     uint16_t status;
5338 
5339     numdl = (dw10 >> 16);
5340     numdu = (dw11 & 0xffff);
5341     lspi = (dw11 >> 16);
5342     lpol = dw12;
5343     lpou = dw13;
5344 
5345     len = (((numdu << 16) | numdl) + 1) << 2;
5346     off = (lpou << 32ULL) | lpol;
5347 
5348     if (off & 0x3) {
5349         return NVME_INVALID_FIELD | NVME_DNR;
5350     }
5351 
5352     trace_pci_nvme_get_log(nvme_cid(req), lid, lsp, rae, len, off);
5353 
5354     status = nvme_check_mdts(n, len);
5355     if (status) {
5356         return status;
5357     }
5358 
5359     switch (lid) {
5360     case NVME_LOG_ERROR_INFO:
5361         return nvme_error_info(n, rae, len, off, req);
5362     case NVME_LOG_SMART_INFO:
5363         return nvme_smart_info(n, rae, len, off, req);
5364     case NVME_LOG_FW_SLOT_INFO:
5365         return nvme_fw_log_info(n, len, off, req);
5366     case NVME_LOG_CHANGED_NSLIST:
5367         return nvme_changed_nslist(n, rae, len, off, req);
5368     case NVME_LOG_CMD_EFFECTS:
5369         return nvme_cmd_effects(n, csi, len, off, req);
5370     case NVME_LOG_ENDGRP:
5371         return nvme_endgrp_info(n, rae, len, off, req);
5372     case NVME_LOG_FDP_CONFS:
5373         return nvme_fdp_confs(n, lspi, len, off, req);
5374     case NVME_LOG_FDP_RUH_USAGE:
5375         return nvme_fdp_ruh_usage(n, lspi, dw10, dw12, len, off, req);
5376     case NVME_LOG_FDP_STATS:
5377         return nvme_fdp_stats(n, lspi, len, off, req);
5378     case NVME_LOG_FDP_EVENTS:
5379         return nvme_fdp_events(n, lspi, len, off, req);
5380     default:
5381         trace_pci_nvme_err_invalid_log_page(nvme_cid(req), lid);
5382         return NVME_INVALID_FIELD | NVME_DNR;
5383     }
5384 }
5385 
5386 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
5387 {
5388     PCIDevice *pci = PCI_DEVICE(n);
5389     uint16_t offset = (cq->cqid << 3) + (1 << 2);
5390 
5391     n->cq[cq->cqid] = NULL;
5392     qemu_bh_delete(cq->bh);
5393     if (cq->ioeventfd_enabled) {
5394         memory_region_del_eventfd(&n->iomem,
5395                                   0x1000 + offset, 4, false, 0, &cq->notifier);
5396         event_notifier_set_handler(&cq->notifier, NULL);
5397         event_notifier_cleanup(&cq->notifier);
5398     }
5399     if (msix_enabled(pci)) {
5400         msix_vector_unuse(pci, cq->vector);
5401     }
5402     if (cq->cqid) {
5403         g_free(cq);
5404     }
5405 }
5406 
5407 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)
5408 {
5409     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
5410     NvmeCQueue *cq;
5411     uint16_t qid = le16_to_cpu(c->qid);
5412 
5413     if (unlikely(!qid || nvme_check_cqid(n, qid))) {
5414         trace_pci_nvme_err_invalid_del_cq_cqid(qid);
5415         return NVME_INVALID_CQID | NVME_DNR;
5416     }
5417 
5418     cq = n->cq[qid];
5419     if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
5420         trace_pci_nvme_err_invalid_del_cq_notempty(qid);
5421         return NVME_INVALID_QUEUE_DEL;
5422     }
5423 
5424     if (cq->irq_enabled && cq->tail != cq->head) {
5425         n->cq_pending--;
5426     }
5427 
5428     nvme_irq_deassert(n, cq);
5429     trace_pci_nvme_del_cq(qid);
5430     nvme_free_cq(cq, n);
5431     return NVME_SUCCESS;
5432 }
5433 
5434 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
5435                          uint16_t cqid, uint16_t vector, uint16_t size,
5436                          uint16_t irq_enabled)
5437 {
5438     PCIDevice *pci = PCI_DEVICE(n);
5439 
5440     if (msix_enabled(pci)) {
5441         msix_vector_use(pci, vector);
5442     }
5443     cq->ctrl = n;
5444     cq->cqid = cqid;
5445     cq->size = size;
5446     cq->dma_addr = dma_addr;
5447     cq->phase = 1;
5448     cq->irq_enabled = irq_enabled;
5449     cq->vector = vector;
5450     cq->head = cq->tail = 0;
5451     QTAILQ_INIT(&cq->req_list);
5452     QTAILQ_INIT(&cq->sq_list);
5453     if (n->dbbuf_enabled) {
5454         cq->db_addr = n->dbbuf_dbs + (cqid << 3) + (1 << 2);
5455         cq->ei_addr = n->dbbuf_eis + (cqid << 3) + (1 << 2);
5456 
5457         if (n->params.ioeventfd && cqid != 0) {
5458             if (!nvme_init_cq_ioeventfd(cq)) {
5459                 cq->ioeventfd_enabled = true;
5460             }
5461         }
5462     }
5463     n->cq[cqid] = cq;
5464     cq->bh = qemu_bh_new_guarded(nvme_post_cqes, cq,
5465                                  &DEVICE(cq->ctrl)->mem_reentrancy_guard);
5466 }
5467 
5468 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeRequest *req)
5469 {
5470     NvmeCQueue *cq;
5471     NvmeCreateCq *c = (NvmeCreateCq *)&req->cmd;
5472     uint16_t cqid = le16_to_cpu(c->cqid);
5473     uint16_t vector = le16_to_cpu(c->irq_vector);
5474     uint16_t qsize = le16_to_cpu(c->qsize);
5475     uint16_t qflags = le16_to_cpu(c->cq_flags);
5476     uint64_t prp1 = le64_to_cpu(c->prp1);
5477     uint32_t cc = ldq_le_p(&n->bar.cc);
5478     uint8_t iocqes = NVME_CC_IOCQES(cc);
5479     uint8_t iosqes = NVME_CC_IOSQES(cc);
5480 
5481     trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
5482                              NVME_CQ_FLAGS_IEN(qflags) != 0);
5483 
5484     if (iosqes != NVME_SQES || iocqes != NVME_CQES) {
5485         trace_pci_nvme_err_invalid_create_cq_entry_size(iosqes, iocqes);
5486         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5487     }
5488 
5489     if (unlikely(!cqid || cqid > n->conf_ioqpairs || n->cq[cqid] != NULL)) {
5490         trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
5491         return NVME_INVALID_QID | NVME_DNR;
5492     }
5493     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
5494         trace_pci_nvme_err_invalid_create_cq_size(qsize);
5495         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5496     }
5497     if (unlikely(prp1 & (n->page_size - 1))) {
5498         trace_pci_nvme_err_invalid_create_cq_addr(prp1);
5499         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
5500     }
5501     if (unlikely(!msix_enabled(PCI_DEVICE(n)) && vector)) {
5502         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5503         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5504     }
5505     if (unlikely(vector >= n->conf_msix_qsize)) {
5506         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5507         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5508     }
5509     if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
5510         trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
5511         return NVME_INVALID_FIELD | NVME_DNR;
5512     }
5513 
5514     cq = g_malloc0(sizeof(*cq));
5515     nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
5516                  NVME_CQ_FLAGS_IEN(qflags));
5517 
5518     /*
5519      * It is only required to set qs_created when creating a completion queue;
5520      * creating a submission queue without a matching completion queue will
5521      * fail.
5522      */
5523     n->qs_created = true;
5524     return NVME_SUCCESS;
5525 }
5526 
5527 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl *n, NvmeRequest *req)
5528 {
5529     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5530 
5531     return nvme_c2h(n, id, sizeof(id), req);
5532 }
5533 
5534 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeRequest *req)
5535 {
5536     trace_pci_nvme_identify_ctrl();
5537 
5538     return nvme_c2h(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl), req);
5539 }
5540 
5541 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl *n, NvmeRequest *req)
5542 {
5543     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5544     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5545     NvmeIdCtrlNvm *id_nvm = (NvmeIdCtrlNvm *)&id;
5546 
5547     trace_pci_nvme_identify_ctrl_csi(c->csi);
5548 
5549     switch (c->csi) {
5550     case NVME_CSI_NVM:
5551         id_nvm->vsl = n->params.vsl;
5552         id_nvm->dmrsl = cpu_to_le32(n->dmrsl);
5553         break;
5554 
5555     case NVME_CSI_ZONED:
5556         ((NvmeIdCtrlZoned *)&id)->zasl = n->params.zasl;
5557         break;
5558 
5559     default:
5560         return NVME_INVALID_FIELD | NVME_DNR;
5561     }
5562 
5563     return nvme_c2h(n, id, sizeof(id), req);
5564 }
5565 
5566 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeRequest *req, bool active)
5567 {
5568     NvmeNamespace *ns;
5569     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5570     uint32_t nsid = le32_to_cpu(c->nsid);
5571 
5572     trace_pci_nvme_identify_ns(nsid);
5573 
5574     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5575         return NVME_INVALID_NSID | NVME_DNR;
5576     }
5577 
5578     ns = nvme_ns(n, nsid);
5579     if (unlikely(!ns)) {
5580         if (!active) {
5581             ns = nvme_subsys_ns(n->subsys, nsid);
5582             if (!ns) {
5583                 return nvme_rpt_empty_id_struct(n, req);
5584             }
5585         } else {
5586             return nvme_rpt_empty_id_struct(n, req);
5587         }
5588     }
5589 
5590     if (active || ns->csi == NVME_CSI_NVM) {
5591         return nvme_c2h(n, (uint8_t *)&ns->id_ns, sizeof(NvmeIdNs), req);
5592     }
5593 
5594     return NVME_INVALID_CMD_SET | NVME_DNR;
5595 }
5596 
5597 static uint16_t nvme_identify_ctrl_list(NvmeCtrl *n, NvmeRequest *req,
5598                                         bool attached)
5599 {
5600     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5601     uint32_t nsid = le32_to_cpu(c->nsid);
5602     uint16_t min_id = le16_to_cpu(c->ctrlid);
5603     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5604     uint16_t *ids = &list[1];
5605     NvmeNamespace *ns;
5606     NvmeCtrl *ctrl;
5607     int cntlid, nr_ids = 0;
5608 
5609     trace_pci_nvme_identify_ctrl_list(c->cns, min_id);
5610 
5611     if (!n->subsys) {
5612         return NVME_INVALID_FIELD | NVME_DNR;
5613     }
5614 
5615     if (attached) {
5616         if (nsid == NVME_NSID_BROADCAST) {
5617             return NVME_INVALID_FIELD | NVME_DNR;
5618         }
5619 
5620         ns = nvme_subsys_ns(n->subsys, nsid);
5621         if (!ns) {
5622             return NVME_INVALID_FIELD | NVME_DNR;
5623         }
5624     }
5625 
5626     for (cntlid = min_id; cntlid < ARRAY_SIZE(n->subsys->ctrls); cntlid++) {
5627         ctrl = nvme_subsys_ctrl(n->subsys, cntlid);
5628         if (!ctrl) {
5629             continue;
5630         }
5631 
5632         if (attached && !nvme_ns(ctrl, nsid)) {
5633             continue;
5634         }
5635 
5636         ids[nr_ids++] = cntlid;
5637     }
5638 
5639     list[0] = nr_ids;
5640 
5641     return nvme_c2h(n, (uint8_t *)list, sizeof(list), req);
5642 }
5643 
5644 static uint16_t nvme_identify_pri_ctrl_cap(NvmeCtrl *n, NvmeRequest *req)
5645 {
5646     trace_pci_nvme_identify_pri_ctrl_cap(le16_to_cpu(n->pri_ctrl_cap.cntlid));
5647 
5648     return nvme_c2h(n, (uint8_t *)&n->pri_ctrl_cap,
5649                     sizeof(NvmePriCtrlCap), req);
5650 }
5651 
5652 static uint16_t nvme_identify_sec_ctrl_list(NvmeCtrl *n, NvmeRequest *req)
5653 {
5654     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5655     uint16_t pri_ctrl_id = le16_to_cpu(n->pri_ctrl_cap.cntlid);
5656     uint16_t min_id = le16_to_cpu(c->ctrlid);
5657     uint8_t num_sec_ctrl = n->nr_sec_ctrls;
5658     NvmeSecCtrlList list = {0};
5659     uint8_t i;
5660 
5661     for (i = 0; i < num_sec_ctrl; i++) {
5662         if (n->sec_ctrl_list[i].scid >= min_id) {
5663             list.numcntl = MIN(num_sec_ctrl - i, 127);
5664             memcpy(&list.sec, n->sec_ctrl_list + i,
5665                    list.numcntl * sizeof(NvmeSecCtrlEntry));
5666             break;
5667         }
5668     }
5669 
5670     trace_pci_nvme_identify_sec_ctrl_list(pri_ctrl_id, list.numcntl);
5671 
5672     return nvme_c2h(n, (uint8_t *)&list, sizeof(list), req);
5673 }
5674 
5675 static uint16_t nvme_identify_ns_csi(NvmeCtrl *n, NvmeRequest *req,
5676                                      bool active)
5677 {
5678     NvmeNamespace *ns;
5679     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5680     uint32_t nsid = le32_to_cpu(c->nsid);
5681 
5682     trace_pci_nvme_identify_ns_csi(nsid, c->csi);
5683 
5684     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5685         return NVME_INVALID_NSID | NVME_DNR;
5686     }
5687 
5688     ns = nvme_ns(n, nsid);
5689     if (unlikely(!ns)) {
5690         if (!active) {
5691             ns = nvme_subsys_ns(n->subsys, nsid);
5692             if (!ns) {
5693                 return nvme_rpt_empty_id_struct(n, req);
5694             }
5695         } else {
5696             return nvme_rpt_empty_id_struct(n, req);
5697         }
5698     }
5699 
5700     if (c->csi == NVME_CSI_NVM) {
5701         return nvme_c2h(n, (uint8_t *)&ns->id_ns_nvm, sizeof(NvmeIdNsNvm),
5702                         req);
5703     } else if (c->csi == NVME_CSI_ZONED && ns->csi == NVME_CSI_ZONED) {
5704         return nvme_c2h(n, (uint8_t *)ns->id_ns_zoned, sizeof(NvmeIdNsZoned),
5705                         req);
5706     }
5707 
5708     return NVME_INVALID_FIELD | NVME_DNR;
5709 }
5710 
5711 static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeRequest *req,
5712                                      bool active)
5713 {
5714     NvmeNamespace *ns;
5715     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5716     uint32_t min_nsid = le32_to_cpu(c->nsid);
5717     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5718     static const int data_len = sizeof(list);
5719     uint32_t *list_ptr = (uint32_t *)list;
5720     int i, j = 0;
5721 
5722     trace_pci_nvme_identify_nslist(min_nsid);
5723 
5724     /*
5725      * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
5726      * since the Active Namespace ID List should return namespaces with ids
5727      * *higher* than the NSID specified in the command. This is also specified
5728      * in the spec (NVM Express v1.3d, Section 5.15.4).
5729      */
5730     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5731         return NVME_INVALID_NSID | NVME_DNR;
5732     }
5733 
5734     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5735         ns = nvme_ns(n, i);
5736         if (!ns) {
5737             if (!active) {
5738                 ns = nvme_subsys_ns(n->subsys, i);
5739                 if (!ns) {
5740                     continue;
5741                 }
5742             } else {
5743                 continue;
5744             }
5745         }
5746         if (ns->params.nsid <= min_nsid) {
5747             continue;
5748         }
5749         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5750         if (j == data_len / sizeof(uint32_t)) {
5751             break;
5752         }
5753     }
5754 
5755     return nvme_c2h(n, list, data_len, req);
5756 }
5757 
5758 static uint16_t nvme_identify_nslist_csi(NvmeCtrl *n, NvmeRequest *req,
5759                                          bool active)
5760 {
5761     NvmeNamespace *ns;
5762     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5763     uint32_t min_nsid = le32_to_cpu(c->nsid);
5764     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5765     static const int data_len = sizeof(list);
5766     uint32_t *list_ptr = (uint32_t *)list;
5767     int i, j = 0;
5768 
5769     trace_pci_nvme_identify_nslist_csi(min_nsid, c->csi);
5770 
5771     /*
5772      * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
5773      */
5774     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5775         return NVME_INVALID_NSID | NVME_DNR;
5776     }
5777 
5778     if (c->csi != NVME_CSI_NVM && c->csi != NVME_CSI_ZONED) {
5779         return NVME_INVALID_FIELD | NVME_DNR;
5780     }
5781 
5782     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5783         ns = nvme_ns(n, i);
5784         if (!ns) {
5785             if (!active) {
5786                 ns = nvme_subsys_ns(n->subsys, i);
5787                 if (!ns) {
5788                     continue;
5789                 }
5790             } else {
5791                 continue;
5792             }
5793         }
5794         if (ns->params.nsid <= min_nsid || c->csi != ns->csi) {
5795             continue;
5796         }
5797         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5798         if (j == data_len / sizeof(uint32_t)) {
5799             break;
5800         }
5801     }
5802 
5803     return nvme_c2h(n, list, data_len, req);
5804 }
5805 
5806 static uint16_t nvme_endurance_group_list(NvmeCtrl *n, NvmeRequest *req)
5807 {
5808     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5809     uint16_t *nr_ids = &list[0];
5810     uint16_t *ids = &list[1];
5811     uint16_t endgid = le32_to_cpu(req->cmd.cdw11) & 0xffff;
5812 
5813     /*
5814      * The current nvme-subsys only supports Endurance Group #1.
5815      */
5816     if (!endgid) {
5817         *nr_ids = 1;
5818         ids[0] = 1;
5819     } else {
5820         *nr_ids = 0;
5821     }
5822 
5823     return nvme_c2h(n, list, sizeof(list), req);
5824 }
5825 
5826 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl *n, NvmeRequest *req)
5827 {
5828     NvmeNamespace *ns;
5829     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5830     uint32_t nsid = le32_to_cpu(c->nsid);
5831     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5832     uint8_t *pos = list;
5833     struct {
5834         NvmeIdNsDescr hdr;
5835         uint8_t v[NVME_NIDL_UUID];
5836     } QEMU_PACKED uuid = {};
5837     struct {
5838         NvmeIdNsDescr hdr;
5839         uint8_t v[NVME_NIDL_NGUID];
5840     } QEMU_PACKED nguid = {};
5841     struct {
5842         NvmeIdNsDescr hdr;
5843         uint64_t v;
5844     } QEMU_PACKED eui64 = {};
5845     struct {
5846         NvmeIdNsDescr hdr;
5847         uint8_t v;
5848     } QEMU_PACKED csi = {};
5849 
5850     trace_pci_nvme_identify_ns_descr_list(nsid);
5851 
5852     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5853         return NVME_INVALID_NSID | NVME_DNR;
5854     }
5855 
5856     ns = nvme_ns(n, nsid);
5857     if (unlikely(!ns)) {
5858         return NVME_INVALID_FIELD | NVME_DNR;
5859     }
5860 
5861     if (!qemu_uuid_is_null(&ns->params.uuid)) {
5862         uuid.hdr.nidt = NVME_NIDT_UUID;
5863         uuid.hdr.nidl = NVME_NIDL_UUID;
5864         memcpy(uuid.v, ns->params.uuid.data, NVME_NIDL_UUID);
5865         memcpy(pos, &uuid, sizeof(uuid));
5866         pos += sizeof(uuid);
5867     }
5868 
5869     if (!nvme_nguid_is_null(&ns->params.nguid)) {
5870         nguid.hdr.nidt = NVME_NIDT_NGUID;
5871         nguid.hdr.nidl = NVME_NIDL_NGUID;
5872         memcpy(nguid.v, ns->params.nguid.data, NVME_NIDL_NGUID);
5873         memcpy(pos, &nguid, sizeof(nguid));
5874         pos += sizeof(nguid);
5875     }
5876 
5877     if (ns->params.eui64) {
5878         eui64.hdr.nidt = NVME_NIDT_EUI64;
5879         eui64.hdr.nidl = NVME_NIDL_EUI64;
5880         eui64.v = cpu_to_be64(ns->params.eui64);
5881         memcpy(pos, &eui64, sizeof(eui64));
5882         pos += sizeof(eui64);
5883     }
5884 
5885     csi.hdr.nidt = NVME_NIDT_CSI;
5886     csi.hdr.nidl = NVME_NIDL_CSI;
5887     csi.v = ns->csi;
5888     memcpy(pos, &csi, sizeof(csi));
5889     pos += sizeof(csi);
5890 
5891     return nvme_c2h(n, list, sizeof(list), req);
5892 }
5893 
5894 static uint16_t nvme_identify_cmd_set(NvmeCtrl *n, NvmeRequest *req)
5895 {
5896     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5897     static const int data_len = sizeof(list);
5898 
5899     trace_pci_nvme_identify_cmd_set();
5900 
5901     NVME_SET_CSI(*list, NVME_CSI_NVM);
5902     NVME_SET_CSI(*list, NVME_CSI_ZONED);
5903 
5904     return nvme_c2h(n, list, data_len, req);
5905 }
5906 
5907 static uint16_t nvme_identify(NvmeCtrl *n, NvmeRequest *req)
5908 {
5909     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5910 
5911     trace_pci_nvme_identify(nvme_cid(req), c->cns, le16_to_cpu(c->ctrlid),
5912                             c->csi);
5913 
5914     switch (c->cns) {
5915     case NVME_ID_CNS_NS:
5916         return nvme_identify_ns(n, req, true);
5917     case NVME_ID_CNS_NS_PRESENT:
5918         return nvme_identify_ns(n, req, false);
5919     case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST:
5920         return nvme_identify_ctrl_list(n, req, true);
5921     case NVME_ID_CNS_CTRL_LIST:
5922         return nvme_identify_ctrl_list(n, req, false);
5923     case NVME_ID_CNS_PRIMARY_CTRL_CAP:
5924         return nvme_identify_pri_ctrl_cap(n, req);
5925     case NVME_ID_CNS_SECONDARY_CTRL_LIST:
5926         return nvme_identify_sec_ctrl_list(n, req);
5927     case NVME_ID_CNS_CS_NS:
5928         return nvme_identify_ns_csi(n, req, true);
5929     case NVME_ID_CNS_CS_NS_PRESENT:
5930         return nvme_identify_ns_csi(n, req, false);
5931     case NVME_ID_CNS_CTRL:
5932         return nvme_identify_ctrl(n, req);
5933     case NVME_ID_CNS_CS_CTRL:
5934         return nvme_identify_ctrl_csi(n, req);
5935     case NVME_ID_CNS_NS_ACTIVE_LIST:
5936         return nvme_identify_nslist(n, req, true);
5937     case NVME_ID_CNS_NS_PRESENT_LIST:
5938         return nvme_identify_nslist(n, req, false);
5939     case NVME_ID_CNS_CS_NS_ACTIVE_LIST:
5940         return nvme_identify_nslist_csi(n, req, true);
5941     case NVME_ID_CNS_ENDURANCE_GROUP_LIST:
5942         return nvme_endurance_group_list(n, req);
5943     case NVME_ID_CNS_CS_NS_PRESENT_LIST:
5944         return nvme_identify_nslist_csi(n, req, false);
5945     case NVME_ID_CNS_NS_DESCR_LIST:
5946         return nvme_identify_ns_descr_list(n, req);
5947     case NVME_ID_CNS_IO_COMMAND_SET:
5948         return nvme_identify_cmd_set(n, req);
5949     default:
5950         trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
5951         return NVME_INVALID_FIELD | NVME_DNR;
5952     }
5953 }
5954 
5955 static uint16_t nvme_abort(NvmeCtrl *n, NvmeRequest *req)
5956 {
5957     uint16_t sqid = le32_to_cpu(req->cmd.cdw10) & 0xffff;
5958     uint16_t cid  = (le32_to_cpu(req->cmd.cdw10) >> 16) & 0xffff;
5959     NvmeSQueue *sq = n->sq[sqid];
5960     NvmeRequest *r, *next;
5961     int i;
5962 
5963     req->cqe.result = 1;
5964     if (nvme_check_sqid(n, sqid)) {
5965         return NVME_INVALID_FIELD | NVME_DNR;
5966     }
5967 
5968     if (sqid == 0) {
5969         for (i = 0; i < n->outstanding_aers; i++) {
5970             NvmeRequest *re = n->aer_reqs[i];
5971             if (re->cqe.cid == cid) {
5972                 memmove(n->aer_reqs + i, n->aer_reqs + i + 1,
5973                          (n->outstanding_aers - i - 1) * sizeof(NvmeRequest *));
5974                 n->outstanding_aers--;
5975                 re->status = NVME_CMD_ABORT_REQ;
5976                 req->cqe.result = 0;
5977                 nvme_enqueue_req_completion(&n->admin_cq, re);
5978                 return NVME_SUCCESS;
5979             }
5980         }
5981     }
5982 
5983     QTAILQ_FOREACH_SAFE(r, &sq->out_req_list, entry, next) {
5984         if (r->cqe.cid == cid) {
5985             if (r->aiocb) {
5986                 blk_aio_cancel_async(r->aiocb);
5987             }
5988             break;
5989         }
5990     }
5991 
5992     return NVME_SUCCESS;
5993 }
5994 
5995 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
5996 {
5997     trace_pci_nvme_setfeat_timestamp(ts);
5998 
5999     n->host_timestamp = le64_to_cpu(ts);
6000     n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
6001 }
6002 
6003 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
6004 {
6005     uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
6006     uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
6007 
6008     union nvme_timestamp {
6009         struct {
6010             uint64_t timestamp:48;
6011             uint64_t sync:1;
6012             uint64_t origin:3;
6013             uint64_t rsvd1:12;
6014         };
6015         uint64_t all;
6016     };
6017 
6018     union nvme_timestamp ts;
6019     ts.all = 0;
6020     ts.timestamp = n->host_timestamp + elapsed_time;
6021 
6022     /* If the host timestamp is non-zero, set the timestamp origin */
6023     ts.origin = n->host_timestamp ? 0x01 : 0x00;
6024 
6025     trace_pci_nvme_getfeat_timestamp(ts.all);
6026 
6027     return cpu_to_le64(ts.all);
6028 }
6029 
6030 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
6031 {
6032     uint64_t timestamp = nvme_get_timestamp(n);
6033 
6034     return nvme_c2h(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
6035 }
6036 
6037 static int nvme_get_feature_fdp(NvmeCtrl *n, uint32_t endgrpid,
6038                                 uint32_t *result)
6039 {
6040     *result = 0;
6041 
6042     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6043         return NVME_INVALID_FIELD | NVME_DNR;
6044     }
6045 
6046     *result = FIELD_DP16(0, FEAT_FDP, FDPE, 1);
6047     *result = FIELD_DP16(*result, FEAT_FDP, CONF_NDX, 0);
6048 
6049     return NVME_SUCCESS;
6050 }
6051 
6052 static uint16_t nvme_get_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
6053                                             NvmeRequest *req, uint32_t *result)
6054 {
6055     NvmeCmd *cmd = &req->cmd;
6056     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
6057     uint16_t ph = cdw11 & 0xffff;
6058     uint8_t noet = (cdw11 >> 16) & 0xff;
6059     uint16_t ruhid, ret;
6060     uint32_t nentries = 0;
6061     uint8_t s_events_ndx = 0;
6062     size_t s_events_siz = sizeof(NvmeFdpEventDescr) * noet;
6063     g_autofree NvmeFdpEventDescr *s_events = g_malloc0(s_events_siz);
6064     NvmeRuHandle *ruh;
6065     NvmeFdpEventDescr *s_event;
6066 
6067     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6068         return NVME_FDP_DISABLED | NVME_DNR;
6069     }
6070 
6071     if (!nvme_ph_valid(ns, ph)) {
6072         return NVME_INVALID_FIELD | NVME_DNR;
6073     }
6074 
6075     ruhid = ns->fdp.phs[ph];
6076     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
6077 
6078     assert(ruh);
6079 
6080     if (unlikely(noet == 0)) {
6081         return NVME_INVALID_FIELD | NVME_DNR;
6082     }
6083 
6084     for (uint8_t event_type = 0; event_type < FDP_EVT_MAX; event_type++) {
6085         uint8_t shift = nvme_fdp_evf_shifts[event_type];
6086         if (!shift && event_type) {
6087             /*
6088              * only first entry (event_type == 0) has a shift value of 0
6089              * other entries are simply unpopulated.
6090              */
6091             continue;
6092         }
6093 
6094         nentries++;
6095 
6096         s_event = &s_events[s_events_ndx];
6097         s_event->evt = event_type;
6098         s_event->evta = (ruh->event_filter >> shift) & 0x1;
6099 
6100         /* break if all `noet` entries are filled */
6101         if ((++s_events_ndx) == noet) {
6102             break;
6103         }
6104     }
6105 
6106     ret = nvme_c2h(n, s_events, s_events_siz, req);
6107     if (ret) {
6108         return ret;
6109     }
6110 
6111     *result = nentries;
6112     return NVME_SUCCESS;
6113 }
6114 
6115 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeRequest *req)
6116 {
6117     NvmeCmd *cmd = &req->cmd;
6118     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
6119     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
6120     uint32_t nsid = le32_to_cpu(cmd->nsid);
6121     uint32_t result = 0;
6122     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
6123     NvmeGetFeatureSelect sel = NVME_GETFEAT_SELECT(dw10);
6124     uint16_t iv;
6125     NvmeNamespace *ns;
6126     int i;
6127     uint16_t endgrpid = 0, ret = NVME_SUCCESS;
6128 
6129     static const uint32_t nvme_feature_default[NVME_FID_MAX] = {
6130         [NVME_ARBITRATION] = NVME_ARB_AB_NOLIMIT,
6131     };
6132 
6133     trace_pci_nvme_getfeat(nvme_cid(req), nsid, fid, sel, dw11);
6134 
6135     if (!nvme_feature_support[fid]) {
6136         return NVME_INVALID_FIELD | NVME_DNR;
6137     }
6138 
6139     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
6140         if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
6141             /*
6142              * The Reservation Notification Mask and Reservation Persistence
6143              * features require a status code of Invalid Field in Command when
6144              * NSID is FFFFFFFFh. Since the device does not support those
6145              * features we can always return Invalid Namespace or Format as we
6146              * should do for all other features.
6147              */
6148             return NVME_INVALID_NSID | NVME_DNR;
6149         }
6150 
6151         if (!nvme_ns(n, nsid)) {
6152             return NVME_INVALID_FIELD | NVME_DNR;
6153         }
6154     }
6155 
6156     switch (sel) {
6157     case NVME_GETFEAT_SELECT_CURRENT:
6158         break;
6159     case NVME_GETFEAT_SELECT_SAVED:
6160         /* no features are saveable by the controller; fallthrough */
6161     case NVME_GETFEAT_SELECT_DEFAULT:
6162         goto defaults;
6163     case NVME_GETFEAT_SELECT_CAP:
6164         result = nvme_feature_cap[fid];
6165         goto out;
6166     }
6167 
6168     switch (fid) {
6169     case NVME_TEMPERATURE_THRESHOLD:
6170         result = 0;
6171 
6172         /*
6173          * The controller only implements the Composite Temperature sensor, so
6174          * return 0 for all other sensors.
6175          */
6176         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6177             goto out;
6178         }
6179 
6180         switch (NVME_TEMP_THSEL(dw11)) {
6181         case NVME_TEMP_THSEL_OVER:
6182             result = n->features.temp_thresh_hi;
6183             goto out;
6184         case NVME_TEMP_THSEL_UNDER:
6185             result = n->features.temp_thresh_low;
6186             goto out;
6187         }
6188 
6189         return NVME_INVALID_FIELD | NVME_DNR;
6190     case NVME_ERROR_RECOVERY:
6191         if (!nvme_nsid_valid(n, nsid)) {
6192             return NVME_INVALID_NSID | NVME_DNR;
6193         }
6194 
6195         ns = nvme_ns(n, nsid);
6196         if (unlikely(!ns)) {
6197             return NVME_INVALID_FIELD | NVME_DNR;
6198         }
6199 
6200         result = ns->features.err_rec;
6201         goto out;
6202     case NVME_VOLATILE_WRITE_CACHE:
6203         result = 0;
6204         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6205             ns = nvme_ns(n, i);
6206             if (!ns) {
6207                 continue;
6208             }
6209 
6210             result = blk_enable_write_cache(ns->blkconf.blk);
6211             if (result) {
6212                 break;
6213             }
6214         }
6215         trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
6216         goto out;
6217     case NVME_ASYNCHRONOUS_EVENT_CONF:
6218         result = n->features.async_config;
6219         goto out;
6220     case NVME_TIMESTAMP:
6221         return nvme_get_feature_timestamp(n, req);
6222     case NVME_HOST_BEHAVIOR_SUPPORT:
6223         return nvme_c2h(n, (uint8_t *)&n->features.hbs,
6224                         sizeof(n->features.hbs), req);
6225     case NVME_FDP_MODE:
6226         endgrpid = dw11 & 0xff;
6227 
6228         if (endgrpid != 0x1) {
6229             return NVME_INVALID_FIELD | NVME_DNR;
6230         }
6231 
6232         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6233         if (ret) {
6234             return ret;
6235         }
6236         goto out;
6237     case NVME_FDP_EVENTS:
6238         if (!nvme_nsid_valid(n, nsid)) {
6239             return NVME_INVALID_NSID | NVME_DNR;
6240         }
6241 
6242         ns = nvme_ns(n, nsid);
6243         if (unlikely(!ns)) {
6244             return NVME_INVALID_FIELD | NVME_DNR;
6245         }
6246 
6247         ret = nvme_get_feature_fdp_events(n, ns, req, &result);
6248         if (ret) {
6249             return ret;
6250         }
6251         goto out;
6252     default:
6253         break;
6254     }
6255 
6256 defaults:
6257     switch (fid) {
6258     case NVME_TEMPERATURE_THRESHOLD:
6259         result = 0;
6260 
6261         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6262             break;
6263         }
6264 
6265         if (NVME_TEMP_THSEL(dw11) == NVME_TEMP_THSEL_OVER) {
6266             result = NVME_TEMPERATURE_WARNING;
6267         }
6268 
6269         break;
6270     case NVME_NUMBER_OF_QUEUES:
6271         result = (n->conf_ioqpairs - 1) | ((n->conf_ioqpairs - 1) << 16);
6272         trace_pci_nvme_getfeat_numq(result);
6273         break;
6274     case NVME_INTERRUPT_VECTOR_CONF:
6275         iv = dw11 & 0xffff;
6276         if (iv >= n->conf_ioqpairs + 1) {
6277             return NVME_INVALID_FIELD | NVME_DNR;
6278         }
6279 
6280         result = iv;
6281         if (iv == n->admin_cq.vector) {
6282             result |= NVME_INTVC_NOCOALESCING;
6283         }
6284         break;
6285     case NVME_FDP_MODE:
6286         endgrpid = dw11 & 0xff;
6287 
6288         if (endgrpid != 0x1) {
6289             return NVME_INVALID_FIELD | NVME_DNR;
6290         }
6291 
6292         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6293         if (ret) {
6294             return ret;
6295         }
6296         goto out;
6297 
6298         break;
6299     default:
6300         result = nvme_feature_default[fid];
6301         break;
6302     }
6303 
6304 out:
6305     req->cqe.result = cpu_to_le32(result);
6306     return ret;
6307 }
6308 
6309 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
6310 {
6311     uint16_t ret;
6312     uint64_t timestamp;
6313 
6314     ret = nvme_h2c(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
6315     if (ret) {
6316         return ret;
6317     }
6318 
6319     nvme_set_timestamp(n, timestamp);
6320 
6321     return NVME_SUCCESS;
6322 }
6323 
6324 static uint16_t nvme_set_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
6325                                             NvmeRequest *req)
6326 {
6327     NvmeCmd *cmd = &req->cmd;
6328     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
6329     uint16_t ph = cdw11 & 0xffff;
6330     uint8_t noet = (cdw11 >> 16) & 0xff;
6331     uint16_t ret, ruhid;
6332     uint8_t enable = le32_to_cpu(cmd->cdw12) & 0x1;
6333     uint8_t event_mask = 0;
6334     unsigned int i;
6335     g_autofree uint8_t *events = g_malloc0(noet);
6336     NvmeRuHandle *ruh = NULL;
6337 
6338     assert(ns);
6339 
6340     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6341         return NVME_FDP_DISABLED | NVME_DNR;
6342     }
6343 
6344     if (!nvme_ph_valid(ns, ph)) {
6345         return NVME_INVALID_FIELD | NVME_DNR;
6346     }
6347 
6348     ruhid = ns->fdp.phs[ph];
6349     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
6350 
6351     ret = nvme_h2c(n, events, noet, req);
6352     if (ret) {
6353         return ret;
6354     }
6355 
6356     for (i = 0; i < noet; i++) {
6357         event_mask |= (1 << nvme_fdp_evf_shifts[events[i]]);
6358     }
6359 
6360     if (enable) {
6361         ruh->event_filter |= event_mask;
6362     } else {
6363         ruh->event_filter = ruh->event_filter & ~event_mask;
6364     }
6365 
6366     return NVME_SUCCESS;
6367 }
6368 
6369 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
6370 {
6371     NvmeNamespace *ns = NULL;
6372 
6373     NvmeCmd *cmd = &req->cmd;
6374     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
6375     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
6376     uint32_t nsid = le32_to_cpu(cmd->nsid);
6377     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
6378     uint8_t save = NVME_SETFEAT_SAVE(dw10);
6379     uint16_t status;
6380     int i;
6381 
6382     trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
6383 
6384     if (save && !(nvme_feature_cap[fid] & NVME_FEAT_CAP_SAVE)) {
6385         return NVME_FID_NOT_SAVEABLE | NVME_DNR;
6386     }
6387 
6388     if (!nvme_feature_support[fid]) {
6389         return NVME_INVALID_FIELD | NVME_DNR;
6390     }
6391 
6392     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
6393         if (nsid != NVME_NSID_BROADCAST) {
6394             if (!nvme_nsid_valid(n, nsid)) {
6395                 return NVME_INVALID_NSID | NVME_DNR;
6396             }
6397 
6398             ns = nvme_ns(n, nsid);
6399             if (unlikely(!ns)) {
6400                 return NVME_INVALID_FIELD | NVME_DNR;
6401             }
6402         }
6403     } else if (nsid && nsid != NVME_NSID_BROADCAST) {
6404         if (!nvme_nsid_valid(n, nsid)) {
6405             return NVME_INVALID_NSID | NVME_DNR;
6406         }
6407 
6408         return NVME_FEAT_NOT_NS_SPEC | NVME_DNR;
6409     }
6410 
6411     if (!(nvme_feature_cap[fid] & NVME_FEAT_CAP_CHANGE)) {
6412         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6413     }
6414 
6415     switch (fid) {
6416     case NVME_TEMPERATURE_THRESHOLD:
6417         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6418             break;
6419         }
6420 
6421         switch (NVME_TEMP_THSEL(dw11)) {
6422         case NVME_TEMP_THSEL_OVER:
6423             n->features.temp_thresh_hi = NVME_TEMP_TMPTH(dw11);
6424             break;
6425         case NVME_TEMP_THSEL_UNDER:
6426             n->features.temp_thresh_low = NVME_TEMP_TMPTH(dw11);
6427             break;
6428         default:
6429             return NVME_INVALID_FIELD | NVME_DNR;
6430         }
6431 
6432         if ((n->temperature >= n->features.temp_thresh_hi) ||
6433             (n->temperature <= n->features.temp_thresh_low)) {
6434             nvme_smart_event(n, NVME_SMART_TEMPERATURE);
6435         }
6436 
6437         break;
6438     case NVME_ERROR_RECOVERY:
6439         if (nsid == NVME_NSID_BROADCAST) {
6440             for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6441                 ns = nvme_ns(n, i);
6442 
6443                 if (!ns) {
6444                     continue;
6445                 }
6446 
6447                 if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat)) {
6448                     ns->features.err_rec = dw11;
6449                 }
6450             }
6451 
6452             break;
6453         }
6454 
6455         assert(ns);
6456         if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat))  {
6457             ns->features.err_rec = dw11;
6458         }
6459         break;
6460     case NVME_VOLATILE_WRITE_CACHE:
6461         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6462             ns = nvme_ns(n, i);
6463             if (!ns) {
6464                 continue;
6465             }
6466 
6467             if (!(dw11 & 0x1) && blk_enable_write_cache(ns->blkconf.blk)) {
6468                 blk_flush(ns->blkconf.blk);
6469             }
6470 
6471             blk_set_enable_write_cache(ns->blkconf.blk, dw11 & 1);
6472         }
6473 
6474         break;
6475 
6476     case NVME_NUMBER_OF_QUEUES:
6477         if (n->qs_created) {
6478             return NVME_CMD_SEQ_ERROR | NVME_DNR;
6479         }
6480 
6481         /*
6482          * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
6483          * and NSQR.
6484          */
6485         if ((dw11 & 0xffff) == 0xffff || ((dw11 >> 16) & 0xffff) == 0xffff) {
6486             return NVME_INVALID_FIELD | NVME_DNR;
6487         }
6488 
6489         trace_pci_nvme_setfeat_numq((dw11 & 0xffff) + 1,
6490                                     ((dw11 >> 16) & 0xffff) + 1,
6491                                     n->conf_ioqpairs,
6492                                     n->conf_ioqpairs);
6493         req->cqe.result = cpu_to_le32((n->conf_ioqpairs - 1) |
6494                                       ((n->conf_ioqpairs - 1) << 16));
6495         break;
6496     case NVME_ASYNCHRONOUS_EVENT_CONF:
6497         n->features.async_config = dw11;
6498         break;
6499     case NVME_TIMESTAMP:
6500         return nvme_set_feature_timestamp(n, req);
6501     case NVME_HOST_BEHAVIOR_SUPPORT:
6502         status = nvme_h2c(n, (uint8_t *)&n->features.hbs,
6503                           sizeof(n->features.hbs), req);
6504         if (status) {
6505             return status;
6506         }
6507 
6508         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6509             ns = nvme_ns(n, i);
6510 
6511             if (!ns) {
6512                 continue;
6513             }
6514 
6515             ns->id_ns.nlbaf = ns->nlbaf - 1;
6516             if (!n->features.hbs.lbafee) {
6517                 ns->id_ns.nlbaf = MIN(ns->id_ns.nlbaf, 15);
6518             }
6519         }
6520 
6521         return status;
6522     case NVME_COMMAND_SET_PROFILE:
6523         if (dw11 & 0x1ff) {
6524             trace_pci_nvme_err_invalid_iocsci(dw11 & 0x1ff);
6525             return NVME_CMD_SET_CMB_REJECTED | NVME_DNR;
6526         }
6527         break;
6528     case NVME_FDP_MODE:
6529         /* spec: abort with cmd seq err if there's one or more NS' in endgrp */
6530         return NVME_CMD_SEQ_ERROR | NVME_DNR;
6531     case NVME_FDP_EVENTS:
6532         return nvme_set_feature_fdp_events(n, ns, req);
6533     default:
6534         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6535     }
6536     return NVME_SUCCESS;
6537 }
6538 
6539 static uint16_t nvme_aer(NvmeCtrl *n, NvmeRequest *req)
6540 {
6541     trace_pci_nvme_aer(nvme_cid(req));
6542 
6543     if (n->outstanding_aers > n->params.aerl) {
6544         trace_pci_nvme_aer_aerl_exceeded();
6545         return NVME_AER_LIMIT_EXCEEDED;
6546     }
6547 
6548     n->aer_reqs[n->outstanding_aers] = req;
6549     n->outstanding_aers++;
6550 
6551     if (!QTAILQ_EMPTY(&n->aer_queue)) {
6552         nvme_process_aers(n);
6553     }
6554 
6555     return NVME_NO_COMPLETE;
6556 }
6557 
6558 static void nvme_update_dmrsl(NvmeCtrl *n)
6559 {
6560     int nsid;
6561 
6562     for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
6563         NvmeNamespace *ns = nvme_ns(n, nsid);
6564         if (!ns) {
6565             continue;
6566         }
6567 
6568         n->dmrsl = MIN_NON_ZERO(n->dmrsl,
6569                                 BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
6570     }
6571 }
6572 
6573 static void nvme_select_iocs_ns(NvmeCtrl *n, NvmeNamespace *ns)
6574 {
6575     uint32_t cc = ldl_le_p(&n->bar.cc);
6576 
6577     ns->iocs = nvme_cse_iocs_none;
6578     switch (ns->csi) {
6579     case NVME_CSI_NVM:
6580         if (NVME_CC_CSS(cc) != NVME_CC_CSS_ADMIN_ONLY) {
6581             ns->iocs = nvme_cse_iocs_nvm;
6582         }
6583         break;
6584     case NVME_CSI_ZONED:
6585         if (NVME_CC_CSS(cc) == NVME_CC_CSS_CSI) {
6586             ns->iocs = nvme_cse_iocs_zoned;
6587         } else if (NVME_CC_CSS(cc) == NVME_CC_CSS_NVM) {
6588             ns->iocs = nvme_cse_iocs_nvm;
6589         }
6590         break;
6591     }
6592 }
6593 
6594 static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
6595 {
6596     NvmeNamespace *ns;
6597     NvmeCtrl *ctrl;
6598     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
6599     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6600     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6601     uint8_t sel = dw10 & 0xf;
6602     uint16_t *nr_ids = &list[0];
6603     uint16_t *ids = &list[1];
6604     uint16_t ret;
6605     int i;
6606 
6607     trace_pci_nvme_ns_attachment(nvme_cid(req), dw10 & 0xf);
6608 
6609     if (!nvme_nsid_valid(n, nsid)) {
6610         return NVME_INVALID_NSID | NVME_DNR;
6611     }
6612 
6613     ns = nvme_subsys_ns(n->subsys, nsid);
6614     if (!ns) {
6615         return NVME_INVALID_FIELD | NVME_DNR;
6616     }
6617 
6618     ret = nvme_h2c(n, (uint8_t *)list, 4096, req);
6619     if (ret) {
6620         return ret;
6621     }
6622 
6623     if (!*nr_ids) {
6624         return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6625     }
6626 
6627     *nr_ids = MIN(*nr_ids, NVME_CONTROLLER_LIST_SIZE - 1);
6628     for (i = 0; i < *nr_ids; i++) {
6629         ctrl = nvme_subsys_ctrl(n->subsys, ids[i]);
6630         if (!ctrl) {
6631             return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6632         }
6633 
6634         switch (sel) {
6635         case NVME_NS_ATTACHMENT_ATTACH:
6636             if (nvme_ns(ctrl, nsid)) {
6637                 return NVME_NS_ALREADY_ATTACHED | NVME_DNR;
6638             }
6639 
6640             if (ns->attached && !ns->params.shared) {
6641                 return NVME_NS_PRIVATE | NVME_DNR;
6642             }
6643 
6644             nvme_attach_ns(ctrl, ns);
6645             nvme_select_iocs_ns(ctrl, ns);
6646 
6647             break;
6648 
6649         case NVME_NS_ATTACHMENT_DETACH:
6650             if (!nvme_ns(ctrl, nsid)) {
6651                 return NVME_NS_NOT_ATTACHED | NVME_DNR;
6652             }
6653 
6654             ctrl->namespaces[nsid] = NULL;
6655             ns->attached--;
6656 
6657             nvme_update_dmrsl(ctrl);
6658 
6659             break;
6660 
6661         default:
6662             return NVME_INVALID_FIELD | NVME_DNR;
6663         }
6664 
6665         /*
6666          * Add namespace id to the changed namespace id list for event clearing
6667          * via Get Log Page command.
6668          */
6669         if (!test_and_set_bit(nsid, ctrl->changed_nsids)) {
6670             nvme_enqueue_event(ctrl, NVME_AER_TYPE_NOTICE,
6671                                NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED,
6672                                NVME_LOG_CHANGED_NSLIST);
6673         }
6674     }
6675 
6676     return NVME_SUCCESS;
6677 }
6678 
6679 typedef struct NvmeFormatAIOCB {
6680     BlockAIOCB common;
6681     BlockAIOCB *aiocb;
6682     NvmeRequest *req;
6683     int ret;
6684 
6685     NvmeNamespace *ns;
6686     uint32_t nsid;
6687     bool broadcast;
6688     int64_t offset;
6689 
6690     uint8_t lbaf;
6691     uint8_t mset;
6692     uint8_t pi;
6693     uint8_t pil;
6694 } NvmeFormatAIOCB;
6695 
6696 static void nvme_format_cancel(BlockAIOCB *aiocb)
6697 {
6698     NvmeFormatAIOCB *iocb = container_of(aiocb, NvmeFormatAIOCB, common);
6699 
6700     iocb->ret = -ECANCELED;
6701 
6702     if (iocb->aiocb) {
6703         blk_aio_cancel_async(iocb->aiocb);
6704         iocb->aiocb = NULL;
6705     }
6706 }
6707 
6708 static const AIOCBInfo nvme_format_aiocb_info = {
6709     .aiocb_size = sizeof(NvmeFormatAIOCB),
6710     .cancel_async = nvme_format_cancel,
6711 };
6712 
6713 static void nvme_format_set(NvmeNamespace *ns, uint8_t lbaf, uint8_t mset,
6714                             uint8_t pi, uint8_t pil)
6715 {
6716     uint8_t lbafl = lbaf & 0xf;
6717     uint8_t lbafu = lbaf >> 4;
6718 
6719     trace_pci_nvme_format_set(ns->params.nsid, lbaf, mset, pi, pil);
6720 
6721     ns->id_ns.dps = (pil << 3) | pi;
6722     ns->id_ns.flbas = (lbafu << 5) | (mset << 4) | lbafl;
6723 
6724     nvme_ns_init_format(ns);
6725 }
6726 
6727 static void nvme_do_format(NvmeFormatAIOCB *iocb);
6728 
6729 static void nvme_format_ns_cb(void *opaque, int ret)
6730 {
6731     NvmeFormatAIOCB *iocb = opaque;
6732     NvmeNamespace *ns = iocb->ns;
6733     int bytes;
6734 
6735     if (iocb->ret < 0) {
6736         goto done;
6737     } else if (ret < 0) {
6738         iocb->ret = ret;
6739         goto done;
6740     }
6741 
6742     assert(ns);
6743 
6744     if (iocb->offset < ns->size) {
6745         bytes = MIN(BDRV_REQUEST_MAX_BYTES, ns->size - iocb->offset);
6746 
6747         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, iocb->offset,
6748                                             bytes, BDRV_REQ_MAY_UNMAP,
6749                                             nvme_format_ns_cb, iocb);
6750 
6751         iocb->offset += bytes;
6752         return;
6753     }
6754 
6755     nvme_format_set(ns, iocb->lbaf, iocb->mset, iocb->pi, iocb->pil);
6756     ns->status = 0x0;
6757     iocb->ns = NULL;
6758     iocb->offset = 0;
6759 
6760 done:
6761     nvme_do_format(iocb);
6762 }
6763 
6764 static uint16_t nvme_format_check(NvmeNamespace *ns, uint8_t lbaf, uint8_t pi)
6765 {
6766     if (ns->params.zoned) {
6767         return NVME_INVALID_FORMAT | NVME_DNR;
6768     }
6769 
6770     if (lbaf > ns->id_ns.nlbaf) {
6771         return NVME_INVALID_FORMAT | NVME_DNR;
6772     }
6773 
6774     if (pi && (ns->id_ns.lbaf[lbaf].ms < nvme_pi_tuple_size(ns))) {
6775         return NVME_INVALID_FORMAT | NVME_DNR;
6776     }
6777 
6778     if (pi && pi > NVME_ID_NS_DPS_TYPE_3) {
6779         return NVME_INVALID_FIELD | NVME_DNR;
6780     }
6781 
6782     return NVME_SUCCESS;
6783 }
6784 
6785 static void nvme_do_format(NvmeFormatAIOCB *iocb)
6786 {
6787     NvmeRequest *req = iocb->req;
6788     NvmeCtrl *n = nvme_ctrl(req);
6789     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6790     uint8_t lbaf = dw10 & 0xf;
6791     uint8_t pi = (dw10 >> 5) & 0x7;
6792     uint16_t status;
6793     int i;
6794 
6795     if (iocb->ret < 0) {
6796         goto done;
6797     }
6798 
6799     if (iocb->broadcast) {
6800         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
6801             iocb->ns = nvme_ns(n, i);
6802             if (iocb->ns) {
6803                 iocb->nsid = i;
6804                 break;
6805             }
6806         }
6807     }
6808 
6809     if (!iocb->ns) {
6810         goto done;
6811     }
6812 
6813     status = nvme_format_check(iocb->ns, lbaf, pi);
6814     if (status) {
6815         req->status = status;
6816         goto done;
6817     }
6818 
6819     iocb->ns->status = NVME_FORMAT_IN_PROGRESS;
6820     nvme_format_ns_cb(iocb, 0);
6821     return;
6822 
6823 done:
6824     iocb->common.cb(iocb->common.opaque, iocb->ret);
6825     qemu_aio_unref(iocb);
6826 }
6827 
6828 static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
6829 {
6830     NvmeFormatAIOCB *iocb;
6831     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6832     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6833     uint8_t lbaf = dw10 & 0xf;
6834     uint8_t mset = (dw10 >> 4) & 0x1;
6835     uint8_t pi = (dw10 >> 5) & 0x7;
6836     uint8_t pil = (dw10 >> 8) & 0x1;
6837     uint8_t lbafu = (dw10 >> 12) & 0x3;
6838     uint16_t status;
6839 
6840     iocb = qemu_aio_get(&nvme_format_aiocb_info, NULL, nvme_misc_cb, req);
6841 
6842     iocb->req = req;
6843     iocb->ret = 0;
6844     iocb->ns = NULL;
6845     iocb->nsid = 0;
6846     iocb->lbaf = lbaf;
6847     iocb->mset = mset;
6848     iocb->pi = pi;
6849     iocb->pil = pil;
6850     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
6851     iocb->offset = 0;
6852 
6853     if (n->features.hbs.lbafee) {
6854         iocb->lbaf |= lbafu << 4;
6855     }
6856 
6857     if (!iocb->broadcast) {
6858         if (!nvme_nsid_valid(n, nsid)) {
6859             status = NVME_INVALID_NSID | NVME_DNR;
6860             goto out;
6861         }
6862 
6863         iocb->ns = nvme_ns(n, nsid);
6864         if (!iocb->ns) {
6865             status = NVME_INVALID_FIELD | NVME_DNR;
6866             goto out;
6867         }
6868     }
6869 
6870     req->aiocb = &iocb->common;
6871     nvme_do_format(iocb);
6872 
6873     return NVME_NO_COMPLETE;
6874 
6875 out:
6876     qemu_aio_unref(iocb);
6877 
6878     return status;
6879 }
6880 
6881 static void nvme_get_virt_res_num(NvmeCtrl *n, uint8_t rt, int *num_total,
6882                                   int *num_prim, int *num_sec)
6883 {
6884     *num_total = le32_to_cpu(rt ?
6885                              n->pri_ctrl_cap.vifrt : n->pri_ctrl_cap.vqfrt);
6886     *num_prim = le16_to_cpu(rt ?
6887                             n->pri_ctrl_cap.virfap : n->pri_ctrl_cap.vqrfap);
6888     *num_sec = le16_to_cpu(rt ? n->pri_ctrl_cap.virfa : n->pri_ctrl_cap.vqrfa);
6889 }
6890 
6891 static uint16_t nvme_assign_virt_res_to_prim(NvmeCtrl *n, NvmeRequest *req,
6892                                              uint16_t cntlid, uint8_t rt,
6893                                              int nr)
6894 {
6895     int num_total, num_prim, num_sec;
6896 
6897     if (cntlid != n->cntlid) {
6898         return NVME_INVALID_CTRL_ID | NVME_DNR;
6899     }
6900 
6901     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6902 
6903     if (nr > num_total) {
6904         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6905     }
6906 
6907     if (nr > num_total - num_sec) {
6908         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6909     }
6910 
6911     if (rt) {
6912         n->next_pri_ctrl_cap.virfap = cpu_to_le16(nr);
6913     } else {
6914         n->next_pri_ctrl_cap.vqrfap = cpu_to_le16(nr);
6915     }
6916 
6917     req->cqe.result = cpu_to_le32(nr);
6918     return req->status;
6919 }
6920 
6921 static void nvme_update_virt_res(NvmeCtrl *n, NvmeSecCtrlEntry *sctrl,
6922                                  uint8_t rt, int nr)
6923 {
6924     int prev_nr, prev_total;
6925 
6926     if (rt) {
6927         prev_nr = le16_to_cpu(sctrl->nvi);
6928         prev_total = le32_to_cpu(n->pri_ctrl_cap.virfa);
6929         sctrl->nvi = cpu_to_le16(nr);
6930         n->pri_ctrl_cap.virfa = cpu_to_le32(prev_total + nr - prev_nr);
6931     } else {
6932         prev_nr = le16_to_cpu(sctrl->nvq);
6933         prev_total = le32_to_cpu(n->pri_ctrl_cap.vqrfa);
6934         sctrl->nvq = cpu_to_le16(nr);
6935         n->pri_ctrl_cap.vqrfa = cpu_to_le32(prev_total + nr - prev_nr);
6936     }
6937 }
6938 
6939 static uint16_t nvme_assign_virt_res_to_sec(NvmeCtrl *n, NvmeRequest *req,
6940                                             uint16_t cntlid, uint8_t rt, int nr)
6941 {
6942     int num_total, num_prim, num_sec, num_free, diff, limit;
6943     NvmeSecCtrlEntry *sctrl;
6944 
6945     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6946     if (!sctrl) {
6947         return NVME_INVALID_CTRL_ID | NVME_DNR;
6948     }
6949 
6950     if (sctrl->scs) {
6951         return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6952     }
6953 
6954     limit = le16_to_cpu(rt ? n->pri_ctrl_cap.vifrsm : n->pri_ctrl_cap.vqfrsm);
6955     if (nr > limit) {
6956         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6957     }
6958 
6959     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6960     num_free = num_total - num_prim - num_sec;
6961     diff = nr - le16_to_cpu(rt ? sctrl->nvi : sctrl->nvq);
6962 
6963     if (diff > num_free) {
6964         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6965     }
6966 
6967     nvme_update_virt_res(n, sctrl, rt, nr);
6968     req->cqe.result = cpu_to_le32(nr);
6969 
6970     return req->status;
6971 }
6972 
6973 static uint16_t nvme_virt_set_state(NvmeCtrl *n, uint16_t cntlid, bool online)
6974 {
6975     PCIDevice *pci = PCI_DEVICE(n);
6976     NvmeCtrl *sn = NULL;
6977     NvmeSecCtrlEntry *sctrl;
6978     int vf_index;
6979 
6980     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6981     if (!sctrl) {
6982         return NVME_INVALID_CTRL_ID | NVME_DNR;
6983     }
6984 
6985     if (!pci_is_vf(pci)) {
6986         vf_index = le16_to_cpu(sctrl->vfn) - 1;
6987         sn = NVME(pcie_sriov_get_vf_at_index(pci, vf_index));
6988     }
6989 
6990     if (online) {
6991         if (!sctrl->nvi || (le16_to_cpu(sctrl->nvq) < 2) || !sn) {
6992             return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6993         }
6994 
6995         if (!sctrl->scs) {
6996             sctrl->scs = 0x1;
6997             nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6998         }
6999     } else {
7000         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_INTERRUPT, 0);
7001         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_QUEUE, 0);
7002 
7003         if (sctrl->scs) {
7004             sctrl->scs = 0x0;
7005             if (sn) {
7006                 nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
7007             }
7008         }
7009     }
7010 
7011     return NVME_SUCCESS;
7012 }
7013 
7014 static uint16_t nvme_virt_mngmt(NvmeCtrl *n, NvmeRequest *req)
7015 {
7016     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
7017     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
7018     uint8_t act = dw10 & 0xf;
7019     uint8_t rt = (dw10 >> 8) & 0x7;
7020     uint16_t cntlid = (dw10 >> 16) & 0xffff;
7021     int nr = dw11 & 0xffff;
7022 
7023     trace_pci_nvme_virt_mngmt(nvme_cid(req), act, cntlid, rt ? "VI" : "VQ", nr);
7024 
7025     if (rt != NVME_VIRT_RES_QUEUE && rt != NVME_VIRT_RES_INTERRUPT) {
7026         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
7027     }
7028 
7029     switch (act) {
7030     case NVME_VIRT_MNGMT_ACTION_SEC_ASSIGN:
7031         return nvme_assign_virt_res_to_sec(n, req, cntlid, rt, nr);
7032     case NVME_VIRT_MNGMT_ACTION_PRM_ALLOC:
7033         return nvme_assign_virt_res_to_prim(n, req, cntlid, rt, nr);
7034     case NVME_VIRT_MNGMT_ACTION_SEC_ONLINE:
7035         return nvme_virt_set_state(n, cntlid, true);
7036     case NVME_VIRT_MNGMT_ACTION_SEC_OFFLINE:
7037         return nvme_virt_set_state(n, cntlid, false);
7038     default:
7039         return NVME_INVALID_FIELD | NVME_DNR;
7040     }
7041 }
7042 
7043 static uint16_t nvme_dbbuf_config(NvmeCtrl *n, const NvmeRequest *req)
7044 {
7045     PCIDevice *pci = PCI_DEVICE(n);
7046     uint64_t dbs_addr = le64_to_cpu(req->cmd.dptr.prp1);
7047     uint64_t eis_addr = le64_to_cpu(req->cmd.dptr.prp2);
7048     int i;
7049 
7050     /* Address should be page aligned */
7051     if (dbs_addr & (n->page_size - 1) || eis_addr & (n->page_size - 1)) {
7052         return NVME_INVALID_FIELD | NVME_DNR;
7053     }
7054 
7055     /* Save shadow buffer base addr for use during queue creation */
7056     n->dbbuf_dbs = dbs_addr;
7057     n->dbbuf_eis = eis_addr;
7058     n->dbbuf_enabled = true;
7059 
7060     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7061         NvmeSQueue *sq = n->sq[i];
7062         NvmeCQueue *cq = n->cq[i];
7063 
7064         if (sq) {
7065             /*
7066              * CAP.DSTRD is 0, so offset of ith sq db_addr is (i<<3)
7067              * nvme_process_db() uses this hard-coded way to calculate
7068              * doorbell offsets. Be consistent with that here.
7069              */
7070             sq->db_addr = dbs_addr + (i << 3);
7071             sq->ei_addr = eis_addr + (i << 3);
7072             stl_le_pci_dma(pci, sq->db_addr, sq->tail, MEMTXATTRS_UNSPECIFIED);
7073 
7074             if (n->params.ioeventfd && sq->sqid != 0) {
7075                 if (!nvme_init_sq_ioeventfd(sq)) {
7076                     sq->ioeventfd_enabled = true;
7077                 }
7078             }
7079         }
7080 
7081         if (cq) {
7082             /* CAP.DSTRD is 0, so offset of ith cq db_addr is (i<<3)+(1<<2) */
7083             cq->db_addr = dbs_addr + (i << 3) + (1 << 2);
7084             cq->ei_addr = eis_addr + (i << 3) + (1 << 2);
7085             stl_le_pci_dma(pci, cq->db_addr, cq->head, MEMTXATTRS_UNSPECIFIED);
7086 
7087             if (n->params.ioeventfd && cq->cqid != 0) {
7088                 if (!nvme_init_cq_ioeventfd(cq)) {
7089                     cq->ioeventfd_enabled = true;
7090                 }
7091             }
7092         }
7093     }
7094 
7095     trace_pci_nvme_dbbuf_config(dbs_addr, eis_addr);
7096 
7097     return NVME_SUCCESS;
7098 }
7099 
7100 static uint16_t nvme_directive_send(NvmeCtrl *n, NvmeRequest *req)
7101 {
7102     return NVME_INVALID_FIELD | NVME_DNR;
7103 }
7104 
7105 static uint16_t nvme_directive_receive(NvmeCtrl *n, NvmeRequest *req)
7106 {
7107     NvmeNamespace *ns;
7108     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
7109     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
7110     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
7111     uint8_t doper, dtype;
7112     uint32_t numd, trans_len;
7113     NvmeDirectiveIdentify id = {
7114         .supported = 1 << NVME_DIRECTIVE_IDENTIFY,
7115         .enabled = 1 << NVME_DIRECTIVE_IDENTIFY,
7116     };
7117 
7118     numd = dw10 + 1;
7119     doper = dw11 & 0xff;
7120     dtype = (dw11 >> 8) & 0xff;
7121 
7122     trans_len = MIN(sizeof(NvmeDirectiveIdentify), numd << 2);
7123 
7124     if (nsid == NVME_NSID_BROADCAST || dtype != NVME_DIRECTIVE_IDENTIFY ||
7125         doper != NVME_DIRECTIVE_RETURN_PARAMS) {
7126         return NVME_INVALID_FIELD | NVME_DNR;
7127     }
7128 
7129     ns = nvme_ns(n, nsid);
7130     if (!ns) {
7131         return NVME_INVALID_FIELD | NVME_DNR;
7132     }
7133 
7134     switch (dtype) {
7135     case NVME_DIRECTIVE_IDENTIFY:
7136         switch (doper) {
7137         case NVME_DIRECTIVE_RETURN_PARAMS:
7138             if (ns->endgrp && ns->endgrp->fdp.enabled) {
7139                 id.supported |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
7140                 id.enabled |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
7141                 id.persistent |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
7142             }
7143 
7144             return nvme_c2h(n, (uint8_t *)&id, trans_len, req);
7145 
7146         default:
7147             return NVME_INVALID_FIELD | NVME_DNR;
7148         }
7149 
7150     default:
7151         return NVME_INVALID_FIELD;
7152     }
7153 }
7154 
7155 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
7156 {
7157     trace_pci_nvme_admin_cmd(nvme_cid(req), nvme_sqid(req), req->cmd.opcode,
7158                              nvme_adm_opc_str(req->cmd.opcode));
7159 
7160     if (!(nvme_cse_acs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
7161         trace_pci_nvme_err_invalid_admin_opc(req->cmd.opcode);
7162         return NVME_INVALID_OPCODE | NVME_DNR;
7163     }
7164 
7165     /* SGLs shall not be used for Admin commands in NVMe over PCIe */
7166     if (NVME_CMD_FLAGS_PSDT(req->cmd.flags) != NVME_PSDT_PRP) {
7167         return NVME_INVALID_FIELD | NVME_DNR;
7168     }
7169 
7170     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
7171         return NVME_INVALID_FIELD;
7172     }
7173 
7174     switch (req->cmd.opcode) {
7175     case NVME_ADM_CMD_DELETE_SQ:
7176         return nvme_del_sq(n, req);
7177     case NVME_ADM_CMD_CREATE_SQ:
7178         return nvme_create_sq(n, req);
7179     case NVME_ADM_CMD_GET_LOG_PAGE:
7180         return nvme_get_log(n, req);
7181     case NVME_ADM_CMD_DELETE_CQ:
7182         return nvme_del_cq(n, req);
7183     case NVME_ADM_CMD_CREATE_CQ:
7184         return nvme_create_cq(n, req);
7185     case NVME_ADM_CMD_IDENTIFY:
7186         return nvme_identify(n, req);
7187     case NVME_ADM_CMD_ABORT:
7188         return nvme_abort(n, req);
7189     case NVME_ADM_CMD_SET_FEATURES:
7190         return nvme_set_feature(n, req);
7191     case NVME_ADM_CMD_GET_FEATURES:
7192         return nvme_get_feature(n, req);
7193     case NVME_ADM_CMD_ASYNC_EV_REQ:
7194         return nvme_aer(n, req);
7195     case NVME_ADM_CMD_NS_ATTACHMENT:
7196         return nvme_ns_attachment(n, req);
7197     case NVME_ADM_CMD_VIRT_MNGMT:
7198         return nvme_virt_mngmt(n, req);
7199     case NVME_ADM_CMD_DBBUF_CONFIG:
7200         return nvme_dbbuf_config(n, req);
7201     case NVME_ADM_CMD_FORMAT_NVM:
7202         return nvme_format(n, req);
7203     case NVME_ADM_CMD_DIRECTIVE_SEND:
7204         return nvme_directive_send(n, req);
7205     case NVME_ADM_CMD_DIRECTIVE_RECV:
7206         return nvme_directive_receive(n, req);
7207     default:
7208         assert(false);
7209     }
7210 
7211     return NVME_INVALID_OPCODE | NVME_DNR;
7212 }
7213 
7214 static void nvme_update_sq_eventidx(const NvmeSQueue *sq)
7215 {
7216     trace_pci_nvme_update_sq_eventidx(sq->sqid, sq->tail);
7217 
7218     stl_le_pci_dma(PCI_DEVICE(sq->ctrl), sq->ei_addr, sq->tail,
7219                    MEMTXATTRS_UNSPECIFIED);
7220 }
7221 
7222 static void nvme_update_sq_tail(NvmeSQueue *sq)
7223 {
7224     ldl_le_pci_dma(PCI_DEVICE(sq->ctrl), sq->db_addr, &sq->tail,
7225                    MEMTXATTRS_UNSPECIFIED);
7226 
7227     trace_pci_nvme_update_sq_tail(sq->sqid, sq->tail);
7228 }
7229 
7230 static void nvme_process_sq(void *opaque)
7231 {
7232     NvmeSQueue *sq = opaque;
7233     NvmeCtrl *n = sq->ctrl;
7234     NvmeCQueue *cq = n->cq[sq->cqid];
7235 
7236     uint16_t status;
7237     hwaddr addr;
7238     NvmeCmd cmd;
7239     NvmeRequest *req;
7240 
7241     if (n->dbbuf_enabled) {
7242         nvme_update_sq_tail(sq);
7243     }
7244 
7245     while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
7246         addr = sq->dma_addr + (sq->head << NVME_SQES);
7247         if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
7248             trace_pci_nvme_err_addr_read(addr);
7249             trace_pci_nvme_err_cfs();
7250             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
7251             break;
7252         }
7253         nvme_inc_sq_head(sq);
7254 
7255         req = QTAILQ_FIRST(&sq->req_list);
7256         QTAILQ_REMOVE(&sq->req_list, req, entry);
7257         QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
7258         nvme_req_clear(req);
7259         req->cqe.cid = cmd.cid;
7260         memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
7261 
7262         status = sq->sqid ? nvme_io_cmd(n, req) :
7263             nvme_admin_cmd(n, req);
7264         if (status != NVME_NO_COMPLETE) {
7265             req->status = status;
7266             nvme_enqueue_req_completion(cq, req);
7267         }
7268 
7269         if (n->dbbuf_enabled) {
7270             nvme_update_sq_eventidx(sq);
7271             nvme_update_sq_tail(sq);
7272         }
7273     }
7274 }
7275 
7276 static void nvme_update_msixcap_ts(PCIDevice *pci_dev, uint32_t table_size)
7277 {
7278     uint8_t *config;
7279 
7280     if (!msix_present(pci_dev)) {
7281         return;
7282     }
7283 
7284     assert(table_size > 0 && table_size <= pci_dev->msix_entries_nr);
7285 
7286     config = pci_dev->config + pci_dev->msix_cap;
7287     pci_set_word_by_mask(config + PCI_MSIX_FLAGS, PCI_MSIX_FLAGS_QSIZE,
7288                          table_size - 1);
7289 }
7290 
7291 static void nvme_activate_virt_res(NvmeCtrl *n)
7292 {
7293     PCIDevice *pci_dev = PCI_DEVICE(n);
7294     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7295     NvmeSecCtrlEntry *sctrl;
7296 
7297     /* -1 to account for the admin queue */
7298     if (pci_is_vf(pci_dev)) {
7299         sctrl = nvme_sctrl(n);
7300         cap->vqprt = sctrl->nvq;
7301         cap->viprt = sctrl->nvi;
7302         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7303         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7304     } else {
7305         cap->vqrfap = n->next_pri_ctrl_cap.vqrfap;
7306         cap->virfap = n->next_pri_ctrl_cap.virfap;
7307         n->conf_ioqpairs = le16_to_cpu(cap->vqprt) +
7308                            le16_to_cpu(cap->vqrfap) - 1;
7309         n->conf_msix_qsize = le16_to_cpu(cap->viprt) +
7310                              le16_to_cpu(cap->virfap);
7311     }
7312 }
7313 
7314 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst)
7315 {
7316     PCIDevice *pci_dev = PCI_DEVICE(n);
7317     NvmeSecCtrlEntry *sctrl;
7318     NvmeNamespace *ns;
7319     int i;
7320 
7321     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7322         ns = nvme_ns(n, i);
7323         if (!ns) {
7324             continue;
7325         }
7326 
7327         nvme_ns_drain(ns);
7328     }
7329 
7330     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7331         if (n->sq[i] != NULL) {
7332             nvme_free_sq(n->sq[i], n);
7333         }
7334     }
7335     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7336         if (n->cq[i] != NULL) {
7337             nvme_free_cq(n->cq[i], n);
7338         }
7339     }
7340 
7341     while (!QTAILQ_EMPTY(&n->aer_queue)) {
7342         NvmeAsyncEvent *event = QTAILQ_FIRST(&n->aer_queue);
7343         QTAILQ_REMOVE(&n->aer_queue, event, entry);
7344         g_free(event);
7345     }
7346 
7347     if (n->params.sriov_max_vfs) {
7348         if (!pci_is_vf(pci_dev)) {
7349             for (i = 0; i < n->nr_sec_ctrls; i++) {
7350                 sctrl = &n->sec_ctrl_list[i];
7351                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
7352             }
7353         }
7354 
7355         if (rst != NVME_RESET_CONTROLLER) {
7356             nvme_activate_virt_res(n);
7357         }
7358     }
7359 
7360     n->aer_queued = 0;
7361     n->aer_mask = 0;
7362     n->outstanding_aers = 0;
7363     n->qs_created = false;
7364 
7365     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
7366 
7367     if (pci_is_vf(pci_dev)) {
7368         sctrl = nvme_sctrl(n);
7369 
7370         stl_le_p(&n->bar.csts, sctrl->scs ? 0 : NVME_CSTS_FAILED);
7371     } else {
7372         stl_le_p(&n->bar.csts, 0);
7373     }
7374 
7375     stl_le_p(&n->bar.intms, 0);
7376     stl_le_p(&n->bar.intmc, 0);
7377     stl_le_p(&n->bar.cc, 0);
7378 
7379     n->dbbuf_dbs = 0;
7380     n->dbbuf_eis = 0;
7381     n->dbbuf_enabled = false;
7382 }
7383 
7384 static void nvme_ctrl_shutdown(NvmeCtrl *n)
7385 {
7386     NvmeNamespace *ns;
7387     int i;
7388 
7389     if (n->pmr.dev) {
7390         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7391     }
7392 
7393     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7394         ns = nvme_ns(n, i);
7395         if (!ns) {
7396             continue;
7397         }
7398 
7399         nvme_ns_shutdown(ns);
7400     }
7401 }
7402 
7403 static void nvme_select_iocs(NvmeCtrl *n)
7404 {
7405     NvmeNamespace *ns;
7406     int i;
7407 
7408     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7409         ns = nvme_ns(n, i);
7410         if (!ns) {
7411             continue;
7412         }
7413 
7414         nvme_select_iocs_ns(n, ns);
7415     }
7416 }
7417 
7418 static int nvme_start_ctrl(NvmeCtrl *n)
7419 {
7420     uint64_t cap = ldq_le_p(&n->bar.cap);
7421     uint32_t cc = ldl_le_p(&n->bar.cc);
7422     uint32_t aqa = ldl_le_p(&n->bar.aqa);
7423     uint64_t asq = ldq_le_p(&n->bar.asq);
7424     uint64_t acq = ldq_le_p(&n->bar.acq);
7425     uint32_t page_bits = NVME_CC_MPS(cc) + 12;
7426     uint32_t page_size = 1 << page_bits;
7427     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
7428 
7429     if (pci_is_vf(PCI_DEVICE(n)) && !sctrl->scs) {
7430         trace_pci_nvme_err_startfail_virt_state(le16_to_cpu(sctrl->nvi),
7431                                                 le16_to_cpu(sctrl->nvq));
7432         return -1;
7433     }
7434     if (unlikely(n->cq[0])) {
7435         trace_pci_nvme_err_startfail_cq();
7436         return -1;
7437     }
7438     if (unlikely(n->sq[0])) {
7439         trace_pci_nvme_err_startfail_sq();
7440         return -1;
7441     }
7442     if (unlikely(asq & (page_size - 1))) {
7443         trace_pci_nvme_err_startfail_asq_misaligned(asq);
7444         return -1;
7445     }
7446     if (unlikely(acq & (page_size - 1))) {
7447         trace_pci_nvme_err_startfail_acq_misaligned(acq);
7448         return -1;
7449     }
7450     if (unlikely(!(NVME_CAP_CSS(cap) & (1 << NVME_CC_CSS(cc))))) {
7451         trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc));
7452         return -1;
7453     }
7454     if (unlikely(NVME_CC_MPS(cc) < NVME_CAP_MPSMIN(cap))) {
7455         trace_pci_nvme_err_startfail_page_too_small(
7456                     NVME_CC_MPS(cc),
7457                     NVME_CAP_MPSMIN(cap));
7458         return -1;
7459     }
7460     if (unlikely(NVME_CC_MPS(cc) >
7461                  NVME_CAP_MPSMAX(cap))) {
7462         trace_pci_nvme_err_startfail_page_too_large(
7463                     NVME_CC_MPS(cc),
7464                     NVME_CAP_MPSMAX(cap));
7465         return -1;
7466     }
7467     if (unlikely(!NVME_AQA_ASQS(aqa))) {
7468         trace_pci_nvme_err_startfail_asqent_sz_zero();
7469         return -1;
7470     }
7471     if (unlikely(!NVME_AQA_ACQS(aqa))) {
7472         trace_pci_nvme_err_startfail_acqent_sz_zero();
7473         return -1;
7474     }
7475 
7476     n->page_bits = page_bits;
7477     n->page_size = page_size;
7478     n->max_prp_ents = n->page_size / sizeof(uint64_t);
7479     nvme_init_cq(&n->admin_cq, n, acq, 0, 0, NVME_AQA_ACQS(aqa) + 1, 1);
7480     nvme_init_sq(&n->admin_sq, n, asq, 0, 0, NVME_AQA_ASQS(aqa) + 1);
7481 
7482     nvme_set_timestamp(n, 0ULL);
7483 
7484     nvme_select_iocs(n);
7485 
7486     return 0;
7487 }
7488 
7489 static void nvme_cmb_enable_regs(NvmeCtrl *n)
7490 {
7491     uint32_t cmbloc = ldl_le_p(&n->bar.cmbloc);
7492     uint32_t cmbsz = ldl_le_p(&n->bar.cmbsz);
7493 
7494     NVME_CMBLOC_SET_CDPCILS(cmbloc, 1);
7495     NVME_CMBLOC_SET_CDPMLS(cmbloc, 1);
7496     NVME_CMBLOC_SET_BIR(cmbloc, NVME_CMB_BIR);
7497     stl_le_p(&n->bar.cmbloc, cmbloc);
7498 
7499     NVME_CMBSZ_SET_SQS(cmbsz, 1);
7500     NVME_CMBSZ_SET_CQS(cmbsz, 0);
7501     NVME_CMBSZ_SET_LISTS(cmbsz, 1);
7502     NVME_CMBSZ_SET_RDS(cmbsz, 1);
7503     NVME_CMBSZ_SET_WDS(cmbsz, 1);
7504     NVME_CMBSZ_SET_SZU(cmbsz, 2); /* MBs */
7505     NVME_CMBSZ_SET_SZ(cmbsz, n->params.cmb_size_mb);
7506     stl_le_p(&n->bar.cmbsz, cmbsz);
7507 }
7508 
7509 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
7510                            unsigned size)
7511 {
7512     PCIDevice *pci = PCI_DEVICE(n);
7513     uint64_t cap = ldq_le_p(&n->bar.cap);
7514     uint32_t cc = ldl_le_p(&n->bar.cc);
7515     uint32_t intms = ldl_le_p(&n->bar.intms);
7516     uint32_t csts = ldl_le_p(&n->bar.csts);
7517     uint32_t pmrsts = ldl_le_p(&n->bar.pmrsts);
7518 
7519     if (unlikely(offset & (sizeof(uint32_t) - 1))) {
7520         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
7521                        "MMIO write not 32-bit aligned,"
7522                        " offset=0x%"PRIx64"", offset);
7523         /* should be ignored, fall through for now */
7524     }
7525 
7526     if (unlikely(size < sizeof(uint32_t))) {
7527         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
7528                        "MMIO write smaller than 32-bits,"
7529                        " offset=0x%"PRIx64", size=%u",
7530                        offset, size);
7531         /* should be ignored, fall through for now */
7532     }
7533 
7534     switch (offset) {
7535     case NVME_REG_INTMS:
7536         if (unlikely(msix_enabled(pci))) {
7537             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7538                            "undefined access to interrupt mask set"
7539                            " when MSI-X is enabled");
7540             /* should be ignored, fall through for now */
7541         }
7542         intms |= data;
7543         stl_le_p(&n->bar.intms, intms);
7544         n->bar.intmc = n->bar.intms;
7545         trace_pci_nvme_mmio_intm_set(data & 0xffffffff, intms);
7546         nvme_irq_check(n);
7547         break;
7548     case NVME_REG_INTMC:
7549         if (unlikely(msix_enabled(pci))) {
7550             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7551                            "undefined access to interrupt mask clr"
7552                            " when MSI-X is enabled");
7553             /* should be ignored, fall through for now */
7554         }
7555         intms &= ~data;
7556         stl_le_p(&n->bar.intms, intms);
7557         n->bar.intmc = n->bar.intms;
7558         trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, intms);
7559         nvme_irq_check(n);
7560         break;
7561     case NVME_REG_CC:
7562         stl_le_p(&n->bar.cc, data);
7563 
7564         trace_pci_nvme_mmio_cfg(data & 0xffffffff);
7565 
7566         if (NVME_CC_SHN(data) && !(NVME_CC_SHN(cc))) {
7567             trace_pci_nvme_mmio_shutdown_set();
7568             nvme_ctrl_shutdown(n);
7569             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7570             csts |= NVME_CSTS_SHST_COMPLETE;
7571         } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(cc)) {
7572             trace_pci_nvme_mmio_shutdown_cleared();
7573             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7574         }
7575 
7576         if (NVME_CC_EN(data) && !NVME_CC_EN(cc)) {
7577             if (unlikely(nvme_start_ctrl(n))) {
7578                 trace_pci_nvme_err_startfail();
7579                 csts = NVME_CSTS_FAILED;
7580             } else {
7581                 trace_pci_nvme_mmio_start_success();
7582                 csts = NVME_CSTS_READY;
7583             }
7584         } else if (!NVME_CC_EN(data) && NVME_CC_EN(cc)) {
7585             trace_pci_nvme_mmio_stopped();
7586             nvme_ctrl_reset(n, NVME_RESET_CONTROLLER);
7587 
7588             break;
7589         }
7590 
7591         stl_le_p(&n->bar.csts, csts);
7592 
7593         break;
7594     case NVME_REG_CSTS:
7595         if (data & (1 << 4)) {
7596             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
7597                            "attempted to W1C CSTS.NSSRO"
7598                            " but CAP.NSSRS is zero (not supported)");
7599         } else if (data != 0) {
7600             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
7601                            "attempted to set a read only bit"
7602                            " of controller status");
7603         }
7604         break;
7605     case NVME_REG_NSSR:
7606         if (data == 0x4e564d65) {
7607             trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
7608         } else {
7609             /* The spec says that writes of other values have no effect */
7610             return;
7611         }
7612         break;
7613     case NVME_REG_AQA:
7614         stl_le_p(&n->bar.aqa, data);
7615         trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
7616         break;
7617     case NVME_REG_ASQ:
7618         stn_le_p(&n->bar.asq, size, data);
7619         trace_pci_nvme_mmio_asqaddr(data);
7620         break;
7621     case NVME_REG_ASQ + 4:
7622         stl_le_p((uint8_t *)&n->bar.asq + 4, data);
7623         trace_pci_nvme_mmio_asqaddr_hi(data, ldq_le_p(&n->bar.asq));
7624         break;
7625     case NVME_REG_ACQ:
7626         trace_pci_nvme_mmio_acqaddr(data);
7627         stn_le_p(&n->bar.acq, size, data);
7628         break;
7629     case NVME_REG_ACQ + 4:
7630         stl_le_p((uint8_t *)&n->bar.acq + 4, data);
7631         trace_pci_nvme_mmio_acqaddr_hi(data, ldq_le_p(&n->bar.acq));
7632         break;
7633     case NVME_REG_CMBLOC:
7634         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
7635                        "invalid write to reserved CMBLOC"
7636                        " when CMBSZ is zero, ignored");
7637         return;
7638     case NVME_REG_CMBSZ:
7639         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
7640                        "invalid write to read only CMBSZ, ignored");
7641         return;
7642     case NVME_REG_CMBMSC:
7643         if (!NVME_CAP_CMBS(cap)) {
7644             return;
7645         }
7646 
7647         stn_le_p(&n->bar.cmbmsc, size, data);
7648         n->cmb.cmse = false;
7649 
7650         if (NVME_CMBMSC_CRE(data)) {
7651             nvme_cmb_enable_regs(n);
7652 
7653             if (NVME_CMBMSC_CMSE(data)) {
7654                 uint64_t cmbmsc = ldq_le_p(&n->bar.cmbmsc);
7655                 hwaddr cba = NVME_CMBMSC_CBA(cmbmsc) << CMBMSC_CBA_SHIFT;
7656                 if (cba + int128_get64(n->cmb.mem.size) < cba) {
7657                     uint32_t cmbsts = ldl_le_p(&n->bar.cmbsts);
7658                     NVME_CMBSTS_SET_CBAI(cmbsts, 1);
7659                     stl_le_p(&n->bar.cmbsts, cmbsts);
7660                     return;
7661                 }
7662 
7663                 n->cmb.cba = cba;
7664                 n->cmb.cmse = true;
7665             }
7666         } else {
7667             n->bar.cmbsz = 0;
7668             n->bar.cmbloc = 0;
7669         }
7670 
7671         return;
7672     case NVME_REG_CMBMSC + 4:
7673         stl_le_p((uint8_t *)&n->bar.cmbmsc + 4, data);
7674         return;
7675 
7676     case NVME_REG_PMRCAP:
7677         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
7678                        "invalid write to PMRCAP register, ignored");
7679         return;
7680     case NVME_REG_PMRCTL:
7681         if (!NVME_CAP_PMRS(cap)) {
7682             return;
7683         }
7684 
7685         stl_le_p(&n->bar.pmrctl, data);
7686         if (NVME_PMRCTL_EN(data)) {
7687             memory_region_set_enabled(&n->pmr.dev->mr, true);
7688             pmrsts = 0;
7689         } else {
7690             memory_region_set_enabled(&n->pmr.dev->mr, false);
7691             NVME_PMRSTS_SET_NRDY(pmrsts, 1);
7692             n->pmr.cmse = false;
7693         }
7694         stl_le_p(&n->bar.pmrsts, pmrsts);
7695         return;
7696     case NVME_REG_PMRSTS:
7697         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
7698                        "invalid write to PMRSTS register, ignored");
7699         return;
7700     case NVME_REG_PMREBS:
7701         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
7702                        "invalid write to PMREBS register, ignored");
7703         return;
7704     case NVME_REG_PMRSWTP:
7705         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
7706                        "invalid write to PMRSWTP register, ignored");
7707         return;
7708     case NVME_REG_PMRMSCL:
7709         if (!NVME_CAP_PMRS(cap)) {
7710             return;
7711         }
7712 
7713         stl_le_p(&n->bar.pmrmscl, data);
7714         n->pmr.cmse = false;
7715 
7716         if (NVME_PMRMSCL_CMSE(data)) {
7717             uint64_t pmrmscu = ldl_le_p(&n->bar.pmrmscu);
7718             hwaddr cba = pmrmscu << 32 |
7719                 (NVME_PMRMSCL_CBA(data) << PMRMSCL_CBA_SHIFT);
7720             if (cba + int128_get64(n->pmr.dev->mr.size) < cba) {
7721                 NVME_PMRSTS_SET_CBAI(pmrsts, 1);
7722                 stl_le_p(&n->bar.pmrsts, pmrsts);
7723                 return;
7724             }
7725 
7726             n->pmr.cmse = true;
7727             n->pmr.cba = cba;
7728         }
7729 
7730         return;
7731     case NVME_REG_PMRMSCU:
7732         if (!NVME_CAP_PMRS(cap)) {
7733             return;
7734         }
7735 
7736         stl_le_p(&n->bar.pmrmscu, data);
7737         return;
7738     default:
7739         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
7740                        "invalid MMIO write,"
7741                        " offset=0x%"PRIx64", data=%"PRIx64"",
7742                        offset, data);
7743         break;
7744     }
7745 }
7746 
7747 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
7748 {
7749     NvmeCtrl *n = (NvmeCtrl *)opaque;
7750     uint8_t *ptr = (uint8_t *)&n->bar;
7751 
7752     trace_pci_nvme_mmio_read(addr, size);
7753 
7754     if (unlikely(addr & (sizeof(uint32_t) - 1))) {
7755         NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
7756                        "MMIO read not 32-bit aligned,"
7757                        " offset=0x%"PRIx64"", addr);
7758         /* should RAZ, fall through for now */
7759     } else if (unlikely(size < sizeof(uint32_t))) {
7760         NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
7761                        "MMIO read smaller than 32-bits,"
7762                        " offset=0x%"PRIx64"", addr);
7763         /* should RAZ, fall through for now */
7764     }
7765 
7766     if (addr > sizeof(n->bar) - size) {
7767         NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
7768                        "MMIO read beyond last register,"
7769                        " offset=0x%"PRIx64", returning 0", addr);
7770 
7771         return 0;
7772     }
7773 
7774     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7775         addr != NVME_REG_CSTS) {
7776         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7777         return 0;
7778     }
7779 
7780     /*
7781      * When PMRWBM bit 1 is set then read from
7782      * from PMRSTS should ensure prior writes
7783      * made it to persistent media
7784      */
7785     if (addr == NVME_REG_PMRSTS &&
7786         (NVME_PMRCAP_PMRWBM(ldl_le_p(&n->bar.pmrcap)) & 0x02)) {
7787         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7788     }
7789 
7790     return ldn_le_p(ptr + addr, size);
7791 }
7792 
7793 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
7794 {
7795     PCIDevice *pci = PCI_DEVICE(n);
7796     uint32_t qid;
7797 
7798     if (unlikely(addr & ((1 << 2) - 1))) {
7799         NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
7800                        "doorbell write not 32-bit aligned,"
7801                        " offset=0x%"PRIx64", ignoring", addr);
7802         return;
7803     }
7804 
7805     if (((addr - 0x1000) >> 2) & 1) {
7806         /* Completion queue doorbell write */
7807 
7808         uint16_t new_head = val & 0xffff;
7809         int start_sqs;
7810         NvmeCQueue *cq;
7811 
7812         qid = (addr - (0x1000 + (1 << 2))) >> 3;
7813         if (unlikely(nvme_check_cqid(n, qid))) {
7814             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
7815                            "completion queue doorbell write"
7816                            " for nonexistent queue,"
7817                            " sqid=%"PRIu32", ignoring", qid);
7818 
7819             /*
7820              * NVM Express v1.3d, Section 4.1 state: "If host software writes
7821              * an invalid value to the Submission Queue Tail Doorbell or
7822              * Completion Queue Head Doorbell register and an Asynchronous Event
7823              * Request command is outstanding, then an asynchronous event is
7824              * posted to the Admin Completion Queue with a status code of
7825              * Invalid Doorbell Write Value."
7826              *
7827              * Also note that the spec includes the "Invalid Doorbell Register"
7828              * status code, but nowhere does it specify when to use it.
7829              * However, it seems reasonable to use it here in a similar
7830              * fashion.
7831              */
7832             if (n->outstanding_aers) {
7833                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7834                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7835                                    NVME_LOG_ERROR_INFO);
7836             }
7837 
7838             return;
7839         }
7840 
7841         cq = n->cq[qid];
7842         if (unlikely(new_head >= cq->size)) {
7843             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
7844                            "completion queue doorbell write value"
7845                            " beyond queue size, sqid=%"PRIu32","
7846                            " new_head=%"PRIu16", ignoring",
7847                            qid, new_head);
7848 
7849             if (n->outstanding_aers) {
7850                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7851                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7852                                    NVME_LOG_ERROR_INFO);
7853             }
7854 
7855             return;
7856         }
7857 
7858         trace_pci_nvme_mmio_doorbell_cq(cq->cqid, new_head);
7859 
7860         start_sqs = nvme_cq_full(cq) ? 1 : 0;
7861         cq->head = new_head;
7862         if (!qid && n->dbbuf_enabled) {
7863             stl_le_pci_dma(pci, cq->db_addr, cq->head, MEMTXATTRS_UNSPECIFIED);
7864         }
7865         if (start_sqs) {
7866             NvmeSQueue *sq;
7867             QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
7868                 qemu_bh_schedule(sq->bh);
7869             }
7870             qemu_bh_schedule(cq->bh);
7871         }
7872 
7873         if (cq->tail == cq->head) {
7874             if (cq->irq_enabled) {
7875                 n->cq_pending--;
7876             }
7877 
7878             nvme_irq_deassert(n, cq);
7879         }
7880     } else {
7881         /* Submission queue doorbell write */
7882 
7883         uint16_t new_tail = val & 0xffff;
7884         NvmeSQueue *sq;
7885 
7886         qid = (addr - 0x1000) >> 3;
7887         if (unlikely(nvme_check_sqid(n, qid))) {
7888             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
7889                            "submission queue doorbell write"
7890                            " for nonexistent queue,"
7891                            " sqid=%"PRIu32", ignoring", qid);
7892 
7893             if (n->outstanding_aers) {
7894                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7895                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7896                                    NVME_LOG_ERROR_INFO);
7897             }
7898 
7899             return;
7900         }
7901 
7902         sq = n->sq[qid];
7903         if (unlikely(new_tail >= sq->size)) {
7904             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
7905                            "submission queue doorbell write value"
7906                            " beyond queue size, sqid=%"PRIu32","
7907                            " new_tail=%"PRIu16", ignoring",
7908                            qid, new_tail);
7909 
7910             if (n->outstanding_aers) {
7911                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7912                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7913                                    NVME_LOG_ERROR_INFO);
7914             }
7915 
7916             return;
7917         }
7918 
7919         trace_pci_nvme_mmio_doorbell_sq(sq->sqid, new_tail);
7920 
7921         sq->tail = new_tail;
7922         if (!qid && n->dbbuf_enabled) {
7923             /*
7924              * The spec states "the host shall also update the controller's
7925              * corresponding doorbell property to match the value of that entry
7926              * in the Shadow Doorbell buffer."
7927              *
7928              * Since this context is currently a VM trap, we can safely enforce
7929              * the requirement from the device side in case the host is
7930              * misbehaving.
7931              *
7932              * Note, we shouldn't have to do this, but various drivers
7933              * including ones that run on Linux, are not updating Admin Queues,
7934              * so we can't trust reading it for an appropriate sq tail.
7935              */
7936             stl_le_pci_dma(pci, sq->db_addr, sq->tail, MEMTXATTRS_UNSPECIFIED);
7937         }
7938 
7939         qemu_bh_schedule(sq->bh);
7940     }
7941 }
7942 
7943 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
7944                             unsigned size)
7945 {
7946     NvmeCtrl *n = (NvmeCtrl *)opaque;
7947 
7948     trace_pci_nvme_mmio_write(addr, data, size);
7949 
7950     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7951         addr != NVME_REG_CSTS) {
7952         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7953         return;
7954     }
7955 
7956     if (addr < sizeof(n->bar)) {
7957         nvme_write_bar(n, addr, data, size);
7958     } else {
7959         nvme_process_db(n, addr, data);
7960     }
7961 }
7962 
7963 static const MemoryRegionOps nvme_mmio_ops = {
7964     .read = nvme_mmio_read,
7965     .write = nvme_mmio_write,
7966     .endianness = DEVICE_LITTLE_ENDIAN,
7967     .impl = {
7968         .min_access_size = 2,
7969         .max_access_size = 8,
7970     },
7971 };
7972 
7973 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
7974                            unsigned size)
7975 {
7976     NvmeCtrl *n = (NvmeCtrl *)opaque;
7977     stn_le_p(&n->cmb.buf[addr], size, data);
7978 }
7979 
7980 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
7981 {
7982     NvmeCtrl *n = (NvmeCtrl *)opaque;
7983     return ldn_le_p(&n->cmb.buf[addr], size);
7984 }
7985 
7986 static const MemoryRegionOps nvme_cmb_ops = {
7987     .read = nvme_cmb_read,
7988     .write = nvme_cmb_write,
7989     .endianness = DEVICE_LITTLE_ENDIAN,
7990     .impl = {
7991         .min_access_size = 1,
7992         .max_access_size = 8,
7993     },
7994 };
7995 
7996 static bool nvme_check_params(NvmeCtrl *n, Error **errp)
7997 {
7998     NvmeParams *params = &n->params;
7999 
8000     if (params->num_queues) {
8001         warn_report("num_queues is deprecated; please use max_ioqpairs "
8002                     "instead");
8003 
8004         params->max_ioqpairs = params->num_queues - 1;
8005     }
8006 
8007     if (n->namespace.blkconf.blk && n->subsys) {
8008         error_setg(errp, "subsystem support is unavailable with legacy "
8009                    "namespace ('drive' property)");
8010         return false;
8011     }
8012 
8013     if (params->max_ioqpairs < 1 ||
8014         params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
8015         error_setg(errp, "max_ioqpairs must be between 1 and %d",
8016                    NVME_MAX_IOQPAIRS);
8017         return false;
8018     }
8019 
8020     if (params->msix_qsize < 1 ||
8021         params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
8022         error_setg(errp, "msix_qsize must be between 1 and %d",
8023                    PCI_MSIX_FLAGS_QSIZE + 1);
8024         return false;
8025     }
8026 
8027     if (!params->serial) {
8028         error_setg(errp, "serial property not set");
8029         return false;
8030     }
8031 
8032     if (params->mqes < 1) {
8033         error_setg(errp, "mqes property cannot be less than 1");
8034         return false;
8035     }
8036 
8037     if (n->pmr.dev) {
8038         if (params->msix_exclusive_bar) {
8039             error_setg(errp, "not enough BARs available to enable PMR");
8040             return false;
8041         }
8042 
8043         if (host_memory_backend_is_mapped(n->pmr.dev)) {
8044             error_setg(errp, "can't use already busy memdev: %s",
8045                        object_get_canonical_path_component(OBJECT(n->pmr.dev)));
8046             return false;
8047         }
8048 
8049         if (!is_power_of_2(n->pmr.dev->size)) {
8050             error_setg(errp, "pmr backend size needs to be power of 2 in size");
8051             return false;
8052         }
8053 
8054         host_memory_backend_set_mapped(n->pmr.dev, true);
8055     }
8056 
8057     if (n->params.zasl > n->params.mdts) {
8058         error_setg(errp, "zoned.zasl (Zone Append Size Limit) must be less "
8059                    "than or equal to mdts (Maximum Data Transfer Size)");
8060         return false;
8061     }
8062 
8063     if (!n->params.vsl) {
8064         error_setg(errp, "vsl must be non-zero");
8065         return false;
8066     }
8067 
8068     if (params->sriov_max_vfs) {
8069         if (!n->subsys) {
8070             error_setg(errp, "subsystem is required for the use of SR-IOV");
8071             return false;
8072         }
8073 
8074         if (params->cmb_size_mb) {
8075             error_setg(errp, "CMB is not supported with SR-IOV");
8076             return false;
8077         }
8078 
8079         if (n->pmr.dev) {
8080             error_setg(errp, "PMR is not supported with SR-IOV");
8081             return false;
8082         }
8083 
8084         if (!params->sriov_vq_flexible || !params->sriov_vi_flexible) {
8085             error_setg(errp, "both sriov_vq_flexible and sriov_vi_flexible"
8086                        " must be set for the use of SR-IOV");
8087             return false;
8088         }
8089 
8090         if (params->sriov_vq_flexible < params->sriov_max_vfs * 2) {
8091             error_setg(errp, "sriov_vq_flexible must be greater than or equal"
8092                        " to %d (sriov_max_vfs * 2)", params->sriov_max_vfs * 2);
8093             return false;
8094         }
8095 
8096         if (params->max_ioqpairs < params->sriov_vq_flexible + 2) {
8097             error_setg(errp, "(max_ioqpairs - sriov_vq_flexible) must be"
8098                        " greater than or equal to 2");
8099             return false;
8100         }
8101 
8102         if (params->sriov_vi_flexible < params->sriov_max_vfs) {
8103             error_setg(errp, "sriov_vi_flexible must be greater than or equal"
8104                        " to %d (sriov_max_vfs)", params->sriov_max_vfs);
8105             return false;
8106         }
8107 
8108         if (params->msix_qsize < params->sriov_vi_flexible + 1) {
8109             error_setg(errp, "(msix_qsize - sriov_vi_flexible) must be"
8110                        " greater than or equal to 1");
8111             return false;
8112         }
8113 
8114         if (params->sriov_max_vi_per_vf &&
8115             (params->sriov_max_vi_per_vf - 1) % NVME_VF_RES_GRANULARITY) {
8116             error_setg(errp, "sriov_max_vi_per_vf must meet:"
8117                        " (sriov_max_vi_per_vf - 1) %% %d == 0 and"
8118                        " sriov_max_vi_per_vf >= 1", NVME_VF_RES_GRANULARITY);
8119             return false;
8120         }
8121 
8122         if (params->sriov_max_vq_per_vf &&
8123             (params->sriov_max_vq_per_vf < 2 ||
8124              (params->sriov_max_vq_per_vf - 1) % NVME_VF_RES_GRANULARITY)) {
8125             error_setg(errp, "sriov_max_vq_per_vf must meet:"
8126                        " (sriov_max_vq_per_vf - 1) %% %d == 0 and"
8127                        " sriov_max_vq_per_vf >= 2", NVME_VF_RES_GRANULARITY);
8128             return false;
8129         }
8130     }
8131 
8132     return true;
8133 }
8134 
8135 static void nvme_init_state(NvmeCtrl *n)
8136 {
8137     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
8138     NvmeSecCtrlEntry *list = n->sec_ctrl_list;
8139     NvmeSecCtrlEntry *sctrl;
8140     PCIDevice *pci = PCI_DEVICE(n);
8141     uint8_t max_vfs;
8142     int i;
8143 
8144     if (pci_is_vf(pci)) {
8145         sctrl = nvme_sctrl(n);
8146         max_vfs = 0;
8147         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
8148         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
8149     } else {
8150         max_vfs = n->params.sriov_max_vfs;
8151         n->conf_ioqpairs = n->params.max_ioqpairs;
8152         n->conf_msix_qsize = n->params.msix_qsize;
8153     }
8154 
8155     n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
8156     n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
8157     n->temperature = NVME_TEMPERATURE;
8158     n->features.temp_thresh_hi = NVME_TEMPERATURE_WARNING;
8159     n->starttime_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
8160     n->aer_reqs = g_new0(NvmeRequest *, n->params.aerl + 1);
8161     QTAILQ_INIT(&n->aer_queue);
8162 
8163     n->nr_sec_ctrls = max_vfs;
8164     for (i = 0; i < max_vfs; i++) {
8165         sctrl = &list[i];
8166         sctrl->pcid = cpu_to_le16(n->cntlid);
8167         sctrl->vfn = cpu_to_le16(i + 1);
8168     }
8169 
8170     cap->cntlid = cpu_to_le16(n->cntlid);
8171     cap->crt = NVME_CRT_VQ | NVME_CRT_VI;
8172 
8173     if (pci_is_vf(pci)) {
8174         cap->vqprt = cpu_to_le16(1 + n->conf_ioqpairs);
8175     } else {
8176         cap->vqprt = cpu_to_le16(1 + n->params.max_ioqpairs -
8177                                  n->params.sriov_vq_flexible);
8178         cap->vqfrt = cpu_to_le32(n->params.sriov_vq_flexible);
8179         cap->vqrfap = cap->vqfrt;
8180         cap->vqgran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
8181         cap->vqfrsm = n->params.sriov_max_vq_per_vf ?
8182                         cpu_to_le16(n->params.sriov_max_vq_per_vf) :
8183                         cap->vqfrt / MAX(max_vfs, 1);
8184     }
8185 
8186     if (pci_is_vf(pci)) {
8187         cap->viprt = cpu_to_le16(n->conf_msix_qsize);
8188     } else {
8189         cap->viprt = cpu_to_le16(n->params.msix_qsize -
8190                                  n->params.sriov_vi_flexible);
8191         cap->vifrt = cpu_to_le32(n->params.sriov_vi_flexible);
8192         cap->virfap = cap->vifrt;
8193         cap->vigran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
8194         cap->vifrsm = n->params.sriov_max_vi_per_vf ?
8195                         cpu_to_le16(n->params.sriov_max_vi_per_vf) :
8196                         cap->vifrt / MAX(max_vfs, 1);
8197     }
8198 }
8199 
8200 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
8201 {
8202     uint64_t cmb_size = n->params.cmb_size_mb * MiB;
8203     uint64_t cap = ldq_le_p(&n->bar.cap);
8204 
8205     n->cmb.buf = g_malloc0(cmb_size);
8206     memory_region_init_io(&n->cmb.mem, OBJECT(n), &nvme_cmb_ops, n,
8207                           "nvme-cmb", cmb_size);
8208     pci_register_bar(pci_dev, NVME_CMB_BIR,
8209                      PCI_BASE_ADDRESS_SPACE_MEMORY |
8210                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
8211                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->cmb.mem);
8212 
8213     NVME_CAP_SET_CMBS(cap, 1);
8214     stq_le_p(&n->bar.cap, cap);
8215 
8216     if (n->params.legacy_cmb) {
8217         nvme_cmb_enable_regs(n);
8218         n->cmb.cmse = true;
8219     }
8220 }
8221 
8222 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
8223 {
8224     uint32_t pmrcap = ldl_le_p(&n->bar.pmrcap);
8225 
8226     NVME_PMRCAP_SET_RDS(pmrcap, 1);
8227     NVME_PMRCAP_SET_WDS(pmrcap, 1);
8228     NVME_PMRCAP_SET_BIR(pmrcap, NVME_PMR_BIR);
8229     /* Turn on bit 1 support */
8230     NVME_PMRCAP_SET_PMRWBM(pmrcap, 0x02);
8231     NVME_PMRCAP_SET_CMSS(pmrcap, 1);
8232     stl_le_p(&n->bar.pmrcap, pmrcap);
8233 
8234     pci_register_bar(pci_dev, NVME_PMR_BIR,
8235                      PCI_BASE_ADDRESS_SPACE_MEMORY |
8236                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
8237                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmr.dev->mr);
8238 
8239     memory_region_set_enabled(&n->pmr.dev->mr, false);
8240 }
8241 
8242 static uint64_t nvme_mbar_size(unsigned total_queues, unsigned total_irqs,
8243                                unsigned *msix_table_offset,
8244                                unsigned *msix_pba_offset)
8245 {
8246     uint64_t bar_size, msix_table_size;
8247 
8248     bar_size = sizeof(NvmeBar) + 2 * total_queues * NVME_DB_SIZE;
8249 
8250     if (total_irqs == 0) {
8251         goto out;
8252     }
8253 
8254     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8255 
8256     if (msix_table_offset) {
8257         *msix_table_offset = bar_size;
8258     }
8259 
8260     msix_table_size = PCI_MSIX_ENTRY_SIZE * total_irqs;
8261     bar_size += msix_table_size;
8262     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8263 
8264     if (msix_pba_offset) {
8265         *msix_pba_offset = bar_size;
8266     }
8267 
8268     bar_size += QEMU_ALIGN_UP(total_irqs, 64) / 8;
8269 
8270 out:
8271     return pow2ceil(bar_size);
8272 }
8273 
8274 static bool nvme_init_sriov(NvmeCtrl *n, PCIDevice *pci_dev, uint16_t offset,
8275                             Error **errp)
8276 {
8277     uint16_t vf_dev_id = n->params.use_intel_id ?
8278                          PCI_DEVICE_ID_INTEL_NVME : PCI_DEVICE_ID_REDHAT_NVME;
8279     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
8280     uint64_t bar_size = nvme_mbar_size(le16_to_cpu(cap->vqfrsm),
8281                                       le16_to_cpu(cap->vifrsm),
8282                                       NULL, NULL);
8283 
8284     if (!pcie_sriov_pf_init(pci_dev, offset, "nvme", vf_dev_id,
8285                             n->params.sriov_max_vfs, n->params.sriov_max_vfs,
8286                             NVME_VF_OFFSET, NVME_VF_STRIDE,
8287                             errp)) {
8288         return false;
8289     }
8290 
8291     pcie_sriov_pf_init_vf_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8292                               PCI_BASE_ADDRESS_MEM_TYPE_64, bar_size);
8293 
8294     return true;
8295 }
8296 
8297 static int nvme_add_pm_capability(PCIDevice *pci_dev, uint8_t offset)
8298 {
8299     Error *err = NULL;
8300     int ret;
8301 
8302     ret = pci_add_capability(pci_dev, PCI_CAP_ID_PM, offset,
8303                              PCI_PM_SIZEOF, &err);
8304     if (err) {
8305         error_report_err(err);
8306         return ret;
8307     }
8308 
8309     pci_set_word(pci_dev->config + offset + PCI_PM_PMC,
8310                  PCI_PM_CAP_VER_1_2);
8311     pci_set_word(pci_dev->config + offset + PCI_PM_CTRL,
8312                  PCI_PM_CTRL_NO_SOFT_RESET);
8313     pci_set_word(pci_dev->wmask + offset + PCI_PM_CTRL,
8314                  PCI_PM_CTRL_STATE_MASK);
8315 
8316     return 0;
8317 }
8318 
8319 static bool pcie_doe_spdm_rsp(DOECap *doe_cap)
8320 {
8321     void *req = pcie_doe_get_write_mbox_ptr(doe_cap);
8322     uint32_t req_len = pcie_doe_get_obj_len(req) * 4;
8323     void *rsp = doe_cap->read_mbox;
8324     uint32_t rsp_len = SPDM_SOCKET_MAX_MESSAGE_BUFFER_SIZE;
8325 
8326     uint32_t recvd = spdm_socket_rsp(doe_cap->spdm_socket,
8327                              SPDM_SOCKET_TRANSPORT_TYPE_PCI_DOE,
8328                              req, req_len, rsp, rsp_len);
8329     doe_cap->read_mbox_len += DIV_ROUND_UP(recvd, 4);
8330 
8331     return recvd != 0;
8332 }
8333 
8334 static DOEProtocol doe_spdm_prot[] = {
8335     { PCI_VENDOR_ID_PCI_SIG, PCI_SIG_DOE_CMA, pcie_doe_spdm_rsp },
8336     { PCI_VENDOR_ID_PCI_SIG, PCI_SIG_DOE_SECURED_CMA, pcie_doe_spdm_rsp },
8337     { }
8338 };
8339 
8340 static bool nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
8341 {
8342     ERRP_GUARD();
8343     uint8_t *pci_conf = pci_dev->config;
8344     uint64_t bar_size;
8345     unsigned msix_table_offset = 0, msix_pba_offset = 0;
8346     unsigned nr_vectors;
8347     int ret;
8348 
8349     pci_conf[PCI_INTERRUPT_PIN] = 1;
8350     pci_config_set_prog_interface(pci_conf, 0x2);
8351 
8352     if (n->params.use_intel_id) {
8353         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_INTEL);
8354         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_INTEL_NVME);
8355     } else {
8356         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_REDHAT);
8357         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_REDHAT_NVME);
8358     }
8359 
8360     pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
8361     nvme_add_pm_capability(pci_dev, 0x60);
8362     pcie_endpoint_cap_init(pci_dev, 0x80);
8363     pcie_cap_flr_init(pci_dev);
8364     if (n->params.sriov_max_vfs) {
8365         pcie_ari_init(pci_dev, 0x100);
8366     }
8367 
8368     if (n->params.msix_exclusive_bar && !pci_is_vf(pci_dev)) {
8369         bar_size = nvme_mbar_size(n->params.max_ioqpairs + 1, 0, NULL, NULL);
8370         memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
8371                               bar_size);
8372         pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8373                          PCI_BASE_ADDRESS_MEM_TYPE_64, &n->iomem);
8374         ret = msix_init_exclusive_bar(pci_dev, n->params.msix_qsize, 4, errp);
8375     } else {
8376         assert(n->params.msix_qsize >= 1);
8377 
8378         /* add one to max_ioqpairs to account for the admin queue pair */
8379         if (!pci_is_vf(pci_dev)) {
8380             nr_vectors = n->params.msix_qsize;
8381             bar_size = nvme_mbar_size(n->params.max_ioqpairs + 1,
8382                                       nr_vectors, &msix_table_offset,
8383                                       &msix_pba_offset);
8384         } else {
8385             NvmeCtrl *pn = NVME(pcie_sriov_get_pf(pci_dev));
8386             NvmePriCtrlCap *cap = &pn->pri_ctrl_cap;
8387 
8388             nr_vectors = le16_to_cpu(cap->vifrsm);
8389             bar_size = nvme_mbar_size(le16_to_cpu(cap->vqfrsm), nr_vectors,
8390                                       &msix_table_offset, &msix_pba_offset);
8391         }
8392 
8393         memory_region_init(&n->bar0, OBJECT(n), "nvme-bar0", bar_size);
8394         memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
8395                               msix_table_offset);
8396         memory_region_add_subregion(&n->bar0, 0, &n->iomem);
8397 
8398         if (pci_is_vf(pci_dev)) {
8399             pcie_sriov_vf_register_bar(pci_dev, 0, &n->bar0);
8400         } else {
8401             pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8402                              PCI_BASE_ADDRESS_MEM_TYPE_64, &n->bar0);
8403         }
8404 
8405         ret = msix_init(pci_dev, nr_vectors,
8406                         &n->bar0, 0, msix_table_offset,
8407                         &n->bar0, 0, msix_pba_offset, 0, errp);
8408     }
8409 
8410     if (ret == -ENOTSUP) {
8411         /* report that msix is not supported, but do not error out */
8412         warn_report_err(*errp);
8413         *errp = NULL;
8414     } else if (ret < 0) {
8415         /* propagate error to caller */
8416         return false;
8417     }
8418 
8419     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs &&
8420         !nvme_init_sriov(n, pci_dev, 0x120, errp)) {
8421         msix_uninit(pci_dev, &n->bar0, &n->bar0);
8422         return false;
8423     }
8424 
8425     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
8426 
8427     pcie_cap_deverr_init(pci_dev);
8428 
8429     /* DOE Initialisation */
8430     if (pci_dev->spdm_port) {
8431         uint16_t doe_offset = n->params.sriov_max_vfs ?
8432                                   PCI_CONFIG_SPACE_SIZE + PCI_ARI_SIZEOF
8433                                   : PCI_CONFIG_SPACE_SIZE;
8434 
8435         pcie_doe_init(pci_dev, &pci_dev->doe_spdm, doe_offset,
8436                       doe_spdm_prot, true, 0);
8437 
8438         pci_dev->doe_spdm.spdm_socket = spdm_socket_connect(pci_dev->spdm_port,
8439                                                             errp);
8440 
8441         if (pci_dev->doe_spdm.spdm_socket < 0) {
8442             return false;
8443         }
8444     }
8445 
8446     if (n->params.cmb_size_mb) {
8447         nvme_init_cmb(n, pci_dev);
8448     }
8449 
8450     if (n->pmr.dev) {
8451         nvme_init_pmr(n, pci_dev);
8452     }
8453 
8454     return true;
8455 }
8456 
8457 static void nvme_init_subnqn(NvmeCtrl *n)
8458 {
8459     NvmeSubsystem *subsys = n->subsys;
8460     NvmeIdCtrl *id = &n->id_ctrl;
8461 
8462     if (!subsys) {
8463         snprintf((char *)id->subnqn, sizeof(id->subnqn),
8464                  "nqn.2019-08.org.qemu:%s", n->params.serial);
8465     } else {
8466         pstrcpy((char *)id->subnqn, sizeof(id->subnqn), (char*)subsys->subnqn);
8467     }
8468 }
8469 
8470 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
8471 {
8472     NvmeIdCtrl *id = &n->id_ctrl;
8473     uint8_t *pci_conf = pci_dev->config;
8474     uint64_t cap = ldq_le_p(&n->bar.cap);
8475     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
8476     uint32_t ctratt;
8477 
8478     id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
8479     id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
8480     strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
8481     strpadcpy((char *)id->fr, sizeof(id->fr), QEMU_VERSION, ' ');
8482     strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
8483 
8484     id->cntlid = cpu_to_le16(n->cntlid);
8485 
8486     id->oaes = cpu_to_le32(NVME_OAES_NS_ATTR);
8487     ctratt = NVME_CTRATT_ELBAS;
8488 
8489     id->rab = 6;
8490 
8491     if (n->params.use_intel_id) {
8492         id->ieee[0] = 0xb3;
8493         id->ieee[1] = 0x02;
8494         id->ieee[2] = 0x00;
8495     } else {
8496         id->ieee[0] = 0x00;
8497         id->ieee[1] = 0x54;
8498         id->ieee[2] = 0x52;
8499     }
8500 
8501     id->mdts = n->params.mdts;
8502     id->ver = cpu_to_le32(NVME_SPEC_VER);
8503     id->oacs =
8504         cpu_to_le16(NVME_OACS_NS_MGMT | NVME_OACS_FORMAT | NVME_OACS_DBBUF |
8505                     NVME_OACS_DIRECTIVES);
8506     id->cntrltype = 0x1;
8507 
8508     /*
8509      * Because the controller always completes the Abort command immediately,
8510      * there can never be more than one concurrently executing Abort command,
8511      * so this value is never used for anything. Note that there can easily be
8512      * many Abort commands in the queues, but they are not considered
8513      * "executing" until processed by nvme_abort.
8514      *
8515      * The specification recommends a value of 3 for Abort Command Limit (four
8516      * concurrently outstanding Abort commands), so lets use that though it is
8517      * inconsequential.
8518      */
8519     id->acl = 3;
8520     id->aerl = n->params.aerl;
8521     id->frmw = (NVME_NUM_FW_SLOTS << 1) | NVME_FRMW_SLOT1_RO;
8522     id->lpa = NVME_LPA_NS_SMART | NVME_LPA_CSE | NVME_LPA_EXTENDED;
8523 
8524     /* recommended default value (~70 C) */
8525     id->wctemp = cpu_to_le16(NVME_TEMPERATURE_WARNING);
8526     id->cctemp = cpu_to_le16(NVME_TEMPERATURE_CRITICAL);
8527 
8528     id->sqes = (NVME_SQES << 4) | NVME_SQES;
8529     id->cqes = (NVME_CQES << 4) | NVME_CQES;
8530     id->nn = cpu_to_le32(NVME_MAX_NAMESPACES);
8531     id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP |
8532                            NVME_ONCS_FEATURES | NVME_ONCS_DSM |
8533                            NVME_ONCS_COMPARE | NVME_ONCS_COPY |
8534                            NVME_ONCS_NVMCSA | NVME_ONCS_NVMAFC);
8535 
8536     /*
8537      * NOTE: If this device ever supports a command set that does NOT use 0x0
8538      * as a Flush-equivalent operation, support for the broadcast NSID in Flush
8539      * should probably be removed.
8540      *
8541      * See comment in nvme_io_cmd.
8542      */
8543     id->vwc = NVME_VWC_NSID_BROADCAST_SUPPORT | NVME_VWC_PRESENT;
8544 
8545     id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0 | NVME_OCFS_COPY_FORMAT_1 |
8546                             NVME_OCFS_COPY_FORMAT_2 | NVME_OCFS_COPY_FORMAT_3);
8547     id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN);
8548 
8549     nvme_init_subnqn(n);
8550 
8551     id->psd[0].mp = cpu_to_le16(0x9c4);
8552     id->psd[0].enlat = cpu_to_le32(0x10);
8553     id->psd[0].exlat = cpu_to_le32(0x4);
8554 
8555     if (n->subsys) {
8556         id->cmic |= NVME_CMIC_MULTI_CTRL;
8557         ctratt |= NVME_CTRATT_ENDGRPS;
8558 
8559         id->endgidmax = cpu_to_le16(0x1);
8560 
8561         if (n->subsys->endgrp.fdp.enabled) {
8562             ctratt |= NVME_CTRATT_FDPS;
8563         }
8564     }
8565 
8566     id->ctratt = cpu_to_le32(ctratt);
8567 
8568     NVME_CAP_SET_MQES(cap, n->params.mqes);
8569     NVME_CAP_SET_CQR(cap, 1);
8570     NVME_CAP_SET_TO(cap, 0xf);
8571     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_NVM);
8572     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_CSI_SUPP);
8573     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_ADMIN_ONLY);
8574     NVME_CAP_SET_MPSMAX(cap, 4);
8575     NVME_CAP_SET_CMBS(cap, n->params.cmb_size_mb ? 1 : 0);
8576     NVME_CAP_SET_PMRS(cap, n->pmr.dev ? 1 : 0);
8577     stq_le_p(&n->bar.cap, cap);
8578 
8579     stl_le_p(&n->bar.vs, NVME_SPEC_VER);
8580     n->bar.intmc = n->bar.intms = 0;
8581 
8582     if (pci_is_vf(pci_dev) && !sctrl->scs) {
8583         stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
8584     }
8585 }
8586 
8587 static int nvme_init_subsys(NvmeCtrl *n, Error **errp)
8588 {
8589     int cntlid;
8590 
8591     if (!n->subsys) {
8592         return 0;
8593     }
8594 
8595     cntlid = nvme_subsys_register_ctrl(n, errp);
8596     if (cntlid < 0) {
8597         return -1;
8598     }
8599 
8600     n->cntlid = cntlid;
8601 
8602     return 0;
8603 }
8604 
8605 void nvme_attach_ns(NvmeCtrl *n, NvmeNamespace *ns)
8606 {
8607     uint32_t nsid = ns->params.nsid;
8608     assert(nsid && nsid <= NVME_MAX_NAMESPACES);
8609 
8610     n->namespaces[nsid] = ns;
8611     ns->attached++;
8612 
8613     n->dmrsl = MIN_NON_ZERO(n->dmrsl,
8614                             BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
8615 }
8616 
8617 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
8618 {
8619     NvmeCtrl *n = NVME(pci_dev);
8620     DeviceState *dev = DEVICE(pci_dev);
8621     NvmeNamespace *ns;
8622     NvmeCtrl *pn = NVME(pcie_sriov_get_pf(pci_dev));
8623 
8624     if (pci_is_vf(pci_dev)) {
8625         /*
8626          * VFs derive settings from the parent. PF's lifespan exceeds
8627          * that of VF's.
8628          */
8629         memcpy(&n->params, &pn->params, sizeof(NvmeParams));
8630 
8631         /*
8632          * Set PF's serial value to a new string memory to prevent 'serial'
8633          * property object release of PF when a VF is removed from the system.
8634          */
8635         n->params.serial = g_strdup(pn->params.serial);
8636         n->subsys = pn->subsys;
8637     }
8638 
8639     if (!nvme_check_params(n, errp)) {
8640         return;
8641     }
8642 
8643     qbus_init(&n->bus, sizeof(NvmeBus), TYPE_NVME_BUS, dev, dev->id);
8644 
8645     if (nvme_init_subsys(n, errp)) {
8646         return;
8647     }
8648     nvme_init_state(n);
8649     if (!nvme_init_pci(n, pci_dev, errp)) {
8650         return;
8651     }
8652     nvme_init_ctrl(n, pci_dev);
8653 
8654     /* setup a namespace if the controller drive property was given */
8655     if (n->namespace.blkconf.blk) {
8656         ns = &n->namespace;
8657         ns->params.nsid = 1;
8658 
8659         if (nvme_ns_setup(ns, errp)) {
8660             return;
8661         }
8662 
8663         nvme_attach_ns(n, ns);
8664     }
8665 }
8666 
8667 static void nvme_exit(PCIDevice *pci_dev)
8668 {
8669     NvmeCtrl *n = NVME(pci_dev);
8670     NvmeNamespace *ns;
8671     int i;
8672 
8673     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8674 
8675     if (n->subsys) {
8676         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
8677             ns = nvme_ns(n, i);
8678             if (ns) {
8679                 ns->attached--;
8680             }
8681         }
8682 
8683         nvme_subsys_unregister_ctrl(n->subsys, n);
8684     }
8685 
8686     g_free(n->cq);
8687     g_free(n->sq);
8688     g_free(n->aer_reqs);
8689 
8690     if (n->params.cmb_size_mb) {
8691         g_free(n->cmb.buf);
8692     }
8693 
8694     if (pci_dev->doe_spdm.spdm_socket > 0) {
8695         spdm_socket_close(pci_dev->doe_spdm.spdm_socket,
8696                           SPDM_SOCKET_TRANSPORT_TYPE_PCI_DOE);
8697     }
8698 
8699     if (n->pmr.dev) {
8700         host_memory_backend_set_mapped(n->pmr.dev, false);
8701     }
8702 
8703     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8704         pcie_sriov_pf_exit(pci_dev);
8705     }
8706 
8707     msix_uninit(pci_dev, &n->bar0, &n->bar0);
8708     memory_region_del_subregion(&n->bar0, &n->iomem);
8709 }
8710 
8711 static Property nvme_props[] = {
8712     DEFINE_BLOCK_PROPERTIES(NvmeCtrl, namespace.blkconf),
8713     DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmr.dev, TYPE_MEMORY_BACKEND,
8714                      HostMemoryBackend *),
8715     DEFINE_PROP_LINK("subsys", NvmeCtrl, subsys, TYPE_NVME_SUBSYS,
8716                      NvmeSubsystem *),
8717     DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
8718     DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
8719     DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
8720     DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
8721     DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
8722     DEFINE_PROP_UINT8("aerl", NvmeCtrl, params.aerl, 3),
8723     DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl, params.aer_max_queued, 64),
8724     DEFINE_PROP_UINT8("mdts", NvmeCtrl, params.mdts, 7),
8725     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
8726     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
8727     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
8728     DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl, params.ioeventfd, false),
8729     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
8730     DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl,
8731                      params.auto_transition_zones, true),
8732     DEFINE_PROP_UINT16("sriov_max_vfs", NvmeCtrl, params.sriov_max_vfs, 0),
8733     DEFINE_PROP_UINT16("sriov_vq_flexible", NvmeCtrl,
8734                        params.sriov_vq_flexible, 0),
8735     DEFINE_PROP_UINT16("sriov_vi_flexible", NvmeCtrl,
8736                        params.sriov_vi_flexible, 0),
8737     DEFINE_PROP_UINT32("sriov_max_vi_per_vf", NvmeCtrl,
8738                        params.sriov_max_vi_per_vf, 0),
8739     DEFINE_PROP_UINT32("sriov_max_vq_per_vf", NvmeCtrl,
8740                        params.sriov_max_vq_per_vf, 0),
8741     DEFINE_PROP_BOOL("msix-exclusive-bar", NvmeCtrl, params.msix_exclusive_bar,
8742                      false),
8743     DEFINE_PROP_UINT16("mqes", NvmeCtrl, params.mqes, 0x7ff),
8744     DEFINE_PROP_UINT16("spdm_port", PCIDevice, spdm_port, 0),
8745     DEFINE_PROP_END_OF_LIST(),
8746 };
8747 
8748 static void nvme_get_smart_warning(Object *obj, Visitor *v, const char *name,
8749                                    void *opaque, Error **errp)
8750 {
8751     NvmeCtrl *n = NVME(obj);
8752     uint8_t value = n->smart_critical_warning;
8753 
8754     visit_type_uint8(v, name, &value, errp);
8755 }
8756 
8757 static void nvme_set_smart_warning(Object *obj, Visitor *v, const char *name,
8758                                    void *opaque, Error **errp)
8759 {
8760     NvmeCtrl *n = NVME(obj);
8761     uint8_t value, old_value, cap = 0, index, event;
8762 
8763     if (!visit_type_uint8(v, name, &value, errp)) {
8764         return;
8765     }
8766 
8767     cap = NVME_SMART_SPARE | NVME_SMART_TEMPERATURE | NVME_SMART_RELIABILITY
8768           | NVME_SMART_MEDIA_READ_ONLY | NVME_SMART_FAILED_VOLATILE_MEDIA;
8769     if (NVME_CAP_PMRS(ldq_le_p(&n->bar.cap))) {
8770         cap |= NVME_SMART_PMR_UNRELIABLE;
8771     }
8772 
8773     if ((value & cap) != value) {
8774         error_setg(errp, "unsupported smart critical warning bits: 0x%x",
8775                    value & ~cap);
8776         return;
8777     }
8778 
8779     old_value = n->smart_critical_warning;
8780     n->smart_critical_warning = value;
8781 
8782     /* only inject new bits of smart critical warning */
8783     for (index = 0; index < NVME_SMART_WARN_MAX; index++) {
8784         event = 1 << index;
8785         if (value & ~old_value & event)
8786             nvme_smart_event(n, event);
8787     }
8788 }
8789 
8790 static void nvme_pci_reset(DeviceState *qdev)
8791 {
8792     PCIDevice *pci_dev = PCI_DEVICE(qdev);
8793     NvmeCtrl *n = NVME(pci_dev);
8794 
8795     trace_pci_nvme_pci_reset();
8796     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8797 }
8798 
8799 static void nvme_sriov_post_write_config(PCIDevice *dev, uint16_t old_num_vfs)
8800 {
8801     NvmeCtrl *n = NVME(dev);
8802     NvmeSecCtrlEntry *sctrl;
8803     int i;
8804 
8805     for (i = pcie_sriov_num_vfs(dev); i < old_num_vfs; i++) {
8806         sctrl = &n->sec_ctrl_list[i];
8807         nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
8808     }
8809 }
8810 
8811 static void nvme_pci_write_config(PCIDevice *dev, uint32_t address,
8812                                   uint32_t val, int len)
8813 {
8814     uint16_t old_num_vfs = pcie_sriov_num_vfs(dev);
8815 
8816     if (pcie_find_capability(dev, PCI_EXT_CAP_ID_DOE)) {
8817         pcie_doe_write_config(&dev->doe_spdm, address, val, len);
8818     }
8819     pci_default_write_config(dev, address, val, len);
8820     pcie_cap_flr_write_config(dev, address, val, len);
8821     nvme_sriov_post_write_config(dev, old_num_vfs);
8822 }
8823 
8824 static uint32_t nvme_pci_read_config(PCIDevice *dev, uint32_t address, int len)
8825 {
8826     uint32_t val;
8827     if (dev->spdm_port && pcie_find_capability(dev, PCI_EXT_CAP_ID_DOE)) {
8828         if (pcie_doe_read_config(&dev->doe_spdm, address, len, &val)) {
8829             return val;
8830         }
8831     }
8832     return pci_default_read_config(dev, address, len);
8833 }
8834 
8835 static const VMStateDescription nvme_vmstate = {
8836     .name = "nvme",
8837     .unmigratable = 1,
8838 };
8839 
8840 static void nvme_class_init(ObjectClass *oc, void *data)
8841 {
8842     DeviceClass *dc = DEVICE_CLASS(oc);
8843     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
8844 
8845     pc->realize = nvme_realize;
8846     pc->config_write = nvme_pci_write_config;
8847     pc->config_read = nvme_pci_read_config;
8848     pc->exit = nvme_exit;
8849     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
8850     pc->revision = 2;
8851 
8852     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
8853     dc->desc = "Non-Volatile Memory Express";
8854     device_class_set_props(dc, nvme_props);
8855     dc->vmsd = &nvme_vmstate;
8856     dc->reset = nvme_pci_reset;
8857 }
8858 
8859 static void nvme_instance_init(Object *obj)
8860 {
8861     NvmeCtrl *n = NVME(obj);
8862 
8863     device_add_bootindex_property(obj, &n->namespace.blkconf.bootindex,
8864                                   "bootindex", "/namespace@1,0",
8865                                   DEVICE(obj));
8866 
8867     object_property_add(obj, "smart_critical_warning", "uint8",
8868                         nvme_get_smart_warning,
8869                         nvme_set_smart_warning, NULL, NULL);
8870 }
8871 
8872 static const TypeInfo nvme_info = {
8873     .name          = TYPE_NVME,
8874     .parent        = TYPE_PCI_DEVICE,
8875     .instance_size = sizeof(NvmeCtrl),
8876     .instance_init = nvme_instance_init,
8877     .class_init    = nvme_class_init,
8878     .interfaces = (InterfaceInfo[]) {
8879         { INTERFACE_PCIE_DEVICE },
8880         { }
8881     },
8882 };
8883 
8884 static const TypeInfo nvme_bus_info = {
8885     .name = TYPE_NVME_BUS,
8886     .parent = TYPE_BUS,
8887     .instance_size = sizeof(NvmeBus),
8888 };
8889 
8890 static void nvme_register_types(void)
8891 {
8892     type_register_static(&nvme_info);
8893     type_register_static(&nvme_bus_info);
8894 }
8895 
8896 type_init(nvme_register_types)
8897