xref: /openbmc/qemu/hw/nvme/ctrl.c (revision 9ea2e69f)
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 thes 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 "migration/vmstate.h"
207 
208 #include "nvme.h"
209 #include "dif.h"
210 #include "trace.h"
211 
212 #define NVME_MAX_IOQPAIRS 0xffff
213 #define NVME_DB_SIZE  4
214 #define NVME_SPEC_VER 0x00010400
215 #define NVME_CMB_BIR 2
216 #define NVME_PMR_BIR 4
217 #define NVME_TEMPERATURE 0x143
218 #define NVME_TEMPERATURE_WARNING 0x157
219 #define NVME_TEMPERATURE_CRITICAL 0x175
220 #define NVME_NUM_FW_SLOTS 1
221 #define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
222 #define NVME_MAX_VFS 127
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 transfered 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             uint64_t prp_list[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     const int 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     uint32_t v = cpu_to_le32(cq->head);
1472 
1473     trace_pci_nvme_update_cq_eventidx(cq->cqid, cq->head);
1474 
1475     pci_dma_write(PCI_DEVICE(cq->ctrl), cq->ei_addr, &v, sizeof(v));
1476 }
1477 
1478 static void nvme_update_cq_head(NvmeCQueue *cq)
1479 {
1480     uint32_t v;
1481 
1482     pci_dma_read(PCI_DEVICE(cq->ctrl), cq->db_addr, &v, sizeof(v));
1483 
1484     cq->head = le32_to_cpu(v);
1485 
1486     trace_pci_nvme_update_cq_head(cq->cqid, cq->head);
1487 }
1488 
1489 static void nvme_post_cqes(void *opaque)
1490 {
1491     NvmeCQueue *cq = opaque;
1492     NvmeCtrl *n = cq->ctrl;
1493     NvmeRequest *req, *next;
1494     bool pending = cq->head != cq->tail;
1495     int ret;
1496 
1497     QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
1498         NvmeSQueue *sq;
1499         hwaddr addr;
1500 
1501         if (n->dbbuf_enabled) {
1502             nvme_update_cq_eventidx(cq);
1503             nvme_update_cq_head(cq);
1504         }
1505 
1506         if (nvme_cq_full(cq)) {
1507             break;
1508         }
1509 
1510         sq = req->sq;
1511         req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
1512         req->cqe.sq_id = cpu_to_le16(sq->sqid);
1513         req->cqe.sq_head = cpu_to_le16(sq->head);
1514         addr = cq->dma_addr + cq->tail * n->cqe_size;
1515         ret = pci_dma_write(PCI_DEVICE(n), addr, (void *)&req->cqe,
1516                             sizeof(req->cqe));
1517         if (ret) {
1518             trace_pci_nvme_err_addr_write(addr);
1519             trace_pci_nvme_err_cfs();
1520             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
1521             break;
1522         }
1523         QTAILQ_REMOVE(&cq->req_list, req, entry);
1524         nvme_inc_cq_tail(cq);
1525         nvme_sg_unmap(&req->sg);
1526         QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
1527     }
1528     if (cq->tail != cq->head) {
1529         if (cq->irq_enabled && !pending) {
1530             n->cq_pending++;
1531         }
1532 
1533         nvme_irq_assert(n, cq);
1534     }
1535 }
1536 
1537 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
1538 {
1539     assert(cq->cqid == req->sq->cqid);
1540     trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
1541                                           le32_to_cpu(req->cqe.result),
1542                                           le32_to_cpu(req->cqe.dw1),
1543                                           req->status);
1544 
1545     if (req->status) {
1546         trace_pci_nvme_err_req_status(nvme_cid(req), nvme_nsid(req->ns),
1547                                       req->status, req->cmd.opcode);
1548     }
1549 
1550     QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
1551     QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
1552 
1553     qemu_bh_schedule(cq->bh);
1554 }
1555 
1556 static void nvme_process_aers(void *opaque)
1557 {
1558     NvmeCtrl *n = opaque;
1559     NvmeAsyncEvent *event, *next;
1560 
1561     trace_pci_nvme_process_aers(n->aer_queued);
1562 
1563     QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
1564         NvmeRequest *req;
1565         NvmeAerResult *result;
1566 
1567         /* can't post cqe if there is nothing to complete */
1568         if (!n->outstanding_aers) {
1569             trace_pci_nvme_no_outstanding_aers();
1570             break;
1571         }
1572 
1573         /* ignore if masked (cqe posted, but event not cleared) */
1574         if (n->aer_mask & (1 << event->result.event_type)) {
1575             trace_pci_nvme_aer_masked(event->result.event_type, n->aer_mask);
1576             continue;
1577         }
1578 
1579         QTAILQ_REMOVE(&n->aer_queue, event, entry);
1580         n->aer_queued--;
1581 
1582         n->aer_mask |= 1 << event->result.event_type;
1583         n->outstanding_aers--;
1584 
1585         req = n->aer_reqs[n->outstanding_aers];
1586 
1587         result = (NvmeAerResult *) &req->cqe.result;
1588         result->event_type = event->result.event_type;
1589         result->event_info = event->result.event_info;
1590         result->log_page = event->result.log_page;
1591         g_free(event);
1592 
1593         trace_pci_nvme_aer_post_cqe(result->event_type, result->event_info,
1594                                     result->log_page);
1595 
1596         nvme_enqueue_req_completion(&n->admin_cq, req);
1597     }
1598 }
1599 
1600 static void nvme_enqueue_event(NvmeCtrl *n, uint8_t event_type,
1601                                uint8_t event_info, uint8_t log_page)
1602 {
1603     NvmeAsyncEvent *event;
1604 
1605     trace_pci_nvme_enqueue_event(event_type, event_info, log_page);
1606 
1607     if (n->aer_queued == n->params.aer_max_queued) {
1608         trace_pci_nvme_enqueue_event_noqueue(n->aer_queued);
1609         return;
1610     }
1611 
1612     event = g_new(NvmeAsyncEvent, 1);
1613     event->result = (NvmeAerResult) {
1614         .event_type = event_type,
1615         .event_info = event_info,
1616         .log_page   = log_page,
1617     };
1618 
1619     QTAILQ_INSERT_TAIL(&n->aer_queue, event, entry);
1620     n->aer_queued++;
1621 
1622     nvme_process_aers(n);
1623 }
1624 
1625 static void nvme_smart_event(NvmeCtrl *n, uint8_t event)
1626 {
1627     uint8_t aer_info;
1628 
1629     /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
1630     if (!(NVME_AEC_SMART(n->features.async_config) & event)) {
1631         return;
1632     }
1633 
1634     switch (event) {
1635     case NVME_SMART_SPARE:
1636         aer_info = NVME_AER_INFO_SMART_SPARE_THRESH;
1637         break;
1638     case NVME_SMART_TEMPERATURE:
1639         aer_info = NVME_AER_INFO_SMART_TEMP_THRESH;
1640         break;
1641     case NVME_SMART_RELIABILITY:
1642     case NVME_SMART_MEDIA_READ_ONLY:
1643     case NVME_SMART_FAILED_VOLATILE_MEDIA:
1644     case NVME_SMART_PMR_UNRELIABLE:
1645         aer_info = NVME_AER_INFO_SMART_RELIABILITY;
1646         break;
1647     default:
1648         return;
1649     }
1650 
1651     nvme_enqueue_event(n, NVME_AER_TYPE_SMART, aer_info, NVME_LOG_SMART_INFO);
1652 }
1653 
1654 static void nvme_clear_events(NvmeCtrl *n, uint8_t event_type)
1655 {
1656     n->aer_mask &= ~(1 << event_type);
1657     if (!QTAILQ_EMPTY(&n->aer_queue)) {
1658         nvme_process_aers(n);
1659     }
1660 }
1661 
1662 static inline uint16_t nvme_check_mdts(NvmeCtrl *n, size_t len)
1663 {
1664     uint8_t mdts = n->params.mdts;
1665 
1666     if (mdts && len > n->page_size << mdts) {
1667         trace_pci_nvme_err_mdts(len);
1668         return NVME_INVALID_FIELD | NVME_DNR;
1669     }
1670 
1671     return NVME_SUCCESS;
1672 }
1673 
1674 static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
1675                                          uint32_t nlb)
1676 {
1677     uint64_t nsze = le64_to_cpu(ns->id_ns.nsze);
1678 
1679     if (unlikely(UINT64_MAX - slba < nlb || slba + nlb > nsze)) {
1680         trace_pci_nvme_err_invalid_lba_range(slba, nlb, nsze);
1681         return NVME_LBA_RANGE | NVME_DNR;
1682     }
1683 
1684     return NVME_SUCCESS;
1685 }
1686 
1687 static int nvme_block_status_all(NvmeNamespace *ns, uint64_t slba,
1688                                  uint32_t nlb, int flags)
1689 {
1690     BlockDriverState *bs = blk_bs(ns->blkconf.blk);
1691 
1692     int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
1693     int64_t offset = nvme_l2b(ns, slba);
1694     int ret;
1695 
1696     /*
1697      * `pnum` holds the number of bytes after offset that shares the same
1698      * allocation status as the byte at offset. If `pnum` is different from
1699      * `bytes`, we should check the allocation status of the next range and
1700      * continue this until all bytes have been checked.
1701      */
1702     do {
1703         bytes -= pnum;
1704 
1705         ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
1706         if (ret < 0) {
1707             return ret;
1708         }
1709 
1710 
1711         trace_pci_nvme_block_status(offset, bytes, pnum, ret,
1712                                     !!(ret & BDRV_BLOCK_ZERO));
1713 
1714         if (!(ret & flags)) {
1715             return 1;
1716         }
1717 
1718         offset += pnum;
1719     } while (pnum != bytes);
1720 
1721     return 0;
1722 }
1723 
1724 static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
1725                                  uint32_t nlb)
1726 {
1727     int ret;
1728     Error *err = NULL;
1729 
1730     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_DATA);
1731     if (ret) {
1732         if (ret < 0) {
1733             error_setg_errno(&err, -ret, "unable to get block status");
1734             error_report_err(err);
1735 
1736             return NVME_INTERNAL_DEV_ERROR;
1737         }
1738 
1739         return NVME_DULB;
1740     }
1741 
1742     return NVME_SUCCESS;
1743 }
1744 
1745 static void nvme_aio_err(NvmeRequest *req, int ret)
1746 {
1747     uint16_t status = NVME_SUCCESS;
1748     Error *local_err = NULL;
1749 
1750     switch (req->cmd.opcode) {
1751     case NVME_CMD_READ:
1752         status = NVME_UNRECOVERED_READ;
1753         break;
1754     case NVME_CMD_FLUSH:
1755     case NVME_CMD_WRITE:
1756     case NVME_CMD_WRITE_ZEROES:
1757     case NVME_CMD_ZONE_APPEND:
1758     case NVME_CMD_COPY:
1759         status = NVME_WRITE_FAULT;
1760         break;
1761     default:
1762         status = NVME_INTERNAL_DEV_ERROR;
1763         break;
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 AioContext *nvme_get_aio_context(BlockAIOCB *acb)
2138 {
2139     return qemu_get_aio_context();
2140 }
2141 
2142 static void nvme_misc_cb(void *opaque, int ret)
2143 {
2144     NvmeRequest *req = opaque;
2145 
2146     trace_pci_nvme_misc_cb(nvme_cid(req));
2147 
2148     if (ret) {
2149         nvme_aio_err(req, ret);
2150     }
2151 
2152     nvme_enqueue_req_completion(nvme_cq(req), req);
2153 }
2154 
2155 void nvme_rw_complete_cb(void *opaque, int ret)
2156 {
2157     NvmeRequest *req = opaque;
2158     NvmeNamespace *ns = req->ns;
2159     BlockBackend *blk = ns->blkconf.blk;
2160     BlockAcctCookie *acct = &req->acct;
2161     BlockAcctStats *stats = blk_get_stats(blk);
2162 
2163     trace_pci_nvme_rw_complete_cb(nvme_cid(req), blk_name(blk));
2164 
2165     if (ret) {
2166         block_acct_failed(stats, acct);
2167         nvme_aio_err(req, ret);
2168     } else {
2169         block_acct_done(stats, acct);
2170     }
2171 
2172     if (ns->params.zoned && nvme_is_write(req)) {
2173         nvme_finalize_zoned_write(ns, req);
2174     }
2175 
2176     nvme_enqueue_req_completion(nvme_cq(req), req);
2177 }
2178 
2179 static void nvme_rw_cb(void *opaque, int ret)
2180 {
2181     NvmeRequest *req = opaque;
2182     NvmeNamespace *ns = req->ns;
2183 
2184     BlockBackend *blk = ns->blkconf.blk;
2185 
2186     trace_pci_nvme_rw_cb(nvme_cid(req), blk_name(blk));
2187 
2188     if (ret) {
2189         goto out;
2190     }
2191 
2192     if (ns->lbaf.ms) {
2193         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2194         uint64_t slba = le64_to_cpu(rw->slba);
2195         uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
2196         uint64_t offset = nvme_moff(ns, slba);
2197 
2198         if (req->cmd.opcode == NVME_CMD_WRITE_ZEROES) {
2199             size_t mlen = nvme_m2b(ns, nlb);
2200 
2201             req->aiocb = blk_aio_pwrite_zeroes(blk, offset, mlen,
2202                                                BDRV_REQ_MAY_UNMAP,
2203                                                nvme_rw_complete_cb, req);
2204             return;
2205         }
2206 
2207         if (nvme_ns_ext(ns) || req->cmd.mptr) {
2208             uint16_t status;
2209 
2210             nvme_sg_unmap(&req->sg);
2211             status = nvme_map_mdata(nvme_ctrl(req), nlb, req);
2212             if (status) {
2213                 ret = -EFAULT;
2214                 goto out;
2215             }
2216 
2217             if (req->cmd.opcode == NVME_CMD_READ) {
2218                 return nvme_blk_read(blk, offset, 1, nvme_rw_complete_cb, req);
2219             }
2220 
2221             return nvme_blk_write(blk, offset, 1, nvme_rw_complete_cb, req);
2222         }
2223     }
2224 
2225 out:
2226     nvme_rw_complete_cb(req, ret);
2227 }
2228 
2229 static void nvme_verify_cb(void *opaque, int ret)
2230 {
2231     NvmeBounceContext *ctx = opaque;
2232     NvmeRequest *req = ctx->req;
2233     NvmeNamespace *ns = req->ns;
2234     BlockBackend *blk = ns->blkconf.blk;
2235     BlockAcctCookie *acct = &req->acct;
2236     BlockAcctStats *stats = blk_get_stats(blk);
2237     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2238     uint64_t slba = le64_to_cpu(rw->slba);
2239     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2240     uint16_t apptag = le16_to_cpu(rw->apptag);
2241     uint16_t appmask = le16_to_cpu(rw->appmask);
2242     uint64_t reftag = le32_to_cpu(rw->reftag);
2243     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2244     uint16_t status;
2245 
2246     reftag |= cdw3 << 32;
2247 
2248     trace_pci_nvme_verify_cb(nvme_cid(req), prinfo, apptag, appmask, reftag);
2249 
2250     if (ret) {
2251         block_acct_failed(stats, acct);
2252         nvme_aio_err(req, ret);
2253         goto out;
2254     }
2255 
2256     block_acct_done(stats, acct);
2257 
2258     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2259         status = nvme_dif_mangle_mdata(ns, ctx->mdata.bounce,
2260                                        ctx->mdata.iov.size, slba);
2261         if (status) {
2262             req->status = status;
2263             goto out;
2264         }
2265 
2266         req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2267                                      ctx->mdata.bounce, ctx->mdata.iov.size,
2268                                      prinfo, slba, apptag, appmask, &reftag);
2269     }
2270 
2271 out:
2272     qemu_iovec_destroy(&ctx->data.iov);
2273     g_free(ctx->data.bounce);
2274 
2275     qemu_iovec_destroy(&ctx->mdata.iov);
2276     g_free(ctx->mdata.bounce);
2277 
2278     g_free(ctx);
2279 
2280     nvme_enqueue_req_completion(nvme_cq(req), req);
2281 }
2282 
2283 
2284 static void nvme_verify_mdata_in_cb(void *opaque, int ret)
2285 {
2286     NvmeBounceContext *ctx = opaque;
2287     NvmeRequest *req = ctx->req;
2288     NvmeNamespace *ns = req->ns;
2289     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2290     uint64_t slba = le64_to_cpu(rw->slba);
2291     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2292     size_t mlen = nvme_m2b(ns, nlb);
2293     uint64_t offset = nvme_moff(ns, slba);
2294     BlockBackend *blk = ns->blkconf.blk;
2295 
2296     trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req), blk_name(blk));
2297 
2298     if (ret) {
2299         goto out;
2300     }
2301 
2302     ctx->mdata.bounce = g_malloc(mlen);
2303 
2304     qemu_iovec_reset(&ctx->mdata.iov);
2305     qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2306 
2307     req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2308                                 nvme_verify_cb, ctx);
2309     return;
2310 
2311 out:
2312     nvme_verify_cb(ctx, ret);
2313 }
2314 
2315 struct nvme_compare_ctx {
2316     struct {
2317         QEMUIOVector iov;
2318         uint8_t *bounce;
2319     } data;
2320 
2321     struct {
2322         QEMUIOVector iov;
2323         uint8_t *bounce;
2324     } mdata;
2325 };
2326 
2327 static void nvme_compare_mdata_cb(void *opaque, int ret)
2328 {
2329     NvmeRequest *req = opaque;
2330     NvmeNamespace *ns = req->ns;
2331     NvmeCtrl *n = nvme_ctrl(req);
2332     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2333     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2334     uint16_t apptag = le16_to_cpu(rw->apptag);
2335     uint16_t appmask = le16_to_cpu(rw->appmask);
2336     uint64_t reftag = le32_to_cpu(rw->reftag);
2337     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2338     struct nvme_compare_ctx *ctx = req->opaque;
2339     g_autofree uint8_t *buf = NULL;
2340     BlockBackend *blk = ns->blkconf.blk;
2341     BlockAcctCookie *acct = &req->acct;
2342     BlockAcctStats *stats = blk_get_stats(blk);
2343     uint16_t status = NVME_SUCCESS;
2344 
2345     reftag |= cdw3 << 32;
2346 
2347     trace_pci_nvme_compare_mdata_cb(nvme_cid(req));
2348 
2349     if (ret) {
2350         block_acct_failed(stats, acct);
2351         nvme_aio_err(req, ret);
2352         goto out;
2353     }
2354 
2355     buf = g_malloc(ctx->mdata.iov.size);
2356 
2357     status = nvme_bounce_mdata(n, buf, ctx->mdata.iov.size,
2358                                NVME_TX_DIRECTION_TO_DEVICE, req);
2359     if (status) {
2360         req->status = status;
2361         goto out;
2362     }
2363 
2364     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2365         uint64_t slba = le64_to_cpu(rw->slba);
2366         uint8_t *bufp;
2367         uint8_t *mbufp = ctx->mdata.bounce;
2368         uint8_t *end = mbufp + ctx->mdata.iov.size;
2369         int16_t pil = 0;
2370 
2371         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2372                                 ctx->mdata.bounce, ctx->mdata.iov.size, prinfo,
2373                                 slba, apptag, appmask, &reftag);
2374         if (status) {
2375             req->status = status;
2376             goto out;
2377         }
2378 
2379         /*
2380          * When formatted with protection information, do not compare the DIF
2381          * tuple.
2382          */
2383         if (!(ns->id_ns.dps & NVME_ID_NS_DPS_FIRST_EIGHT)) {
2384             pil = ns->lbaf.ms - nvme_pi_tuple_size(ns);
2385         }
2386 
2387         for (bufp = buf; mbufp < end; bufp += ns->lbaf.ms, mbufp += ns->lbaf.ms) {
2388             if (memcmp(bufp + pil, mbufp + pil, ns->lbaf.ms - pil)) {
2389                 req->status = NVME_CMP_FAILURE | NVME_DNR;
2390                 goto out;
2391             }
2392         }
2393 
2394         goto out;
2395     }
2396 
2397     if (memcmp(buf, ctx->mdata.bounce, ctx->mdata.iov.size)) {
2398         req->status = NVME_CMP_FAILURE | NVME_DNR;
2399         goto out;
2400     }
2401 
2402     block_acct_done(stats, acct);
2403 
2404 out:
2405     qemu_iovec_destroy(&ctx->data.iov);
2406     g_free(ctx->data.bounce);
2407 
2408     qemu_iovec_destroy(&ctx->mdata.iov);
2409     g_free(ctx->mdata.bounce);
2410 
2411     g_free(ctx);
2412 
2413     nvme_enqueue_req_completion(nvme_cq(req), req);
2414 }
2415 
2416 static void nvme_compare_data_cb(void *opaque, int ret)
2417 {
2418     NvmeRequest *req = opaque;
2419     NvmeCtrl *n = nvme_ctrl(req);
2420     NvmeNamespace *ns = req->ns;
2421     BlockBackend *blk = ns->blkconf.blk;
2422     BlockAcctCookie *acct = &req->acct;
2423     BlockAcctStats *stats = blk_get_stats(blk);
2424 
2425     struct nvme_compare_ctx *ctx = req->opaque;
2426     g_autofree uint8_t *buf = NULL;
2427     uint16_t status;
2428 
2429     trace_pci_nvme_compare_data_cb(nvme_cid(req));
2430 
2431     if (ret) {
2432         block_acct_failed(stats, acct);
2433         nvme_aio_err(req, ret);
2434         goto out;
2435     }
2436 
2437     buf = g_malloc(ctx->data.iov.size);
2438 
2439     status = nvme_bounce_data(n, buf, ctx->data.iov.size,
2440                               NVME_TX_DIRECTION_TO_DEVICE, req);
2441     if (status) {
2442         req->status = status;
2443         goto out;
2444     }
2445 
2446     if (memcmp(buf, ctx->data.bounce, ctx->data.iov.size)) {
2447         req->status = NVME_CMP_FAILURE | NVME_DNR;
2448         goto out;
2449     }
2450 
2451     if (ns->lbaf.ms) {
2452         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2453         uint64_t slba = le64_to_cpu(rw->slba);
2454         uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2455         size_t mlen = nvme_m2b(ns, nlb);
2456         uint64_t offset = nvme_moff(ns, slba);
2457 
2458         ctx->mdata.bounce = g_malloc(mlen);
2459 
2460         qemu_iovec_init(&ctx->mdata.iov, 1);
2461         qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2462 
2463         req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2464                                     nvme_compare_mdata_cb, req);
2465         return;
2466     }
2467 
2468     block_acct_done(stats, acct);
2469 
2470 out:
2471     qemu_iovec_destroy(&ctx->data.iov);
2472     g_free(ctx->data.bounce);
2473     g_free(ctx);
2474 
2475     nvme_enqueue_req_completion(nvme_cq(req), req);
2476 }
2477 
2478 typedef struct NvmeDSMAIOCB {
2479     BlockAIOCB common;
2480     BlockAIOCB *aiocb;
2481     NvmeRequest *req;
2482     int ret;
2483 
2484     NvmeDsmRange *range;
2485     unsigned int nr;
2486     unsigned int idx;
2487 } NvmeDSMAIOCB;
2488 
2489 static void nvme_dsm_cancel(BlockAIOCB *aiocb)
2490 {
2491     NvmeDSMAIOCB *iocb = container_of(aiocb, NvmeDSMAIOCB, common);
2492 
2493     /* break nvme_dsm_cb loop */
2494     iocb->idx = iocb->nr;
2495     iocb->ret = -ECANCELED;
2496 
2497     if (iocb->aiocb) {
2498         blk_aio_cancel_async(iocb->aiocb);
2499         iocb->aiocb = NULL;
2500     } else {
2501         /*
2502          * We only reach this if nvme_dsm_cancel() has already been called or
2503          * the command ran to completion.
2504          */
2505         assert(iocb->idx == iocb->nr);
2506     }
2507 }
2508 
2509 static const AIOCBInfo nvme_dsm_aiocb_info = {
2510     .aiocb_size   = sizeof(NvmeDSMAIOCB),
2511     .cancel_async = nvme_dsm_cancel,
2512 };
2513 
2514 static void nvme_dsm_cb(void *opaque, int ret);
2515 
2516 static void nvme_dsm_md_cb(void *opaque, int ret)
2517 {
2518     NvmeDSMAIOCB *iocb = opaque;
2519     NvmeRequest *req = iocb->req;
2520     NvmeNamespace *ns = req->ns;
2521     NvmeDsmRange *range;
2522     uint64_t slba;
2523     uint32_t nlb;
2524 
2525     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
2526         goto done;
2527     }
2528 
2529     range = &iocb->range[iocb->idx - 1];
2530     slba = le64_to_cpu(range->slba);
2531     nlb = le32_to_cpu(range->nlb);
2532 
2533     /*
2534      * Check that all block were discarded (zeroed); otherwise we do not zero
2535      * the metadata.
2536      */
2537 
2538     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_ZERO);
2539     if (ret) {
2540         if (ret < 0) {
2541             goto done;
2542         }
2543 
2544         nvme_dsm_cb(iocb, 0);
2545         return;
2546     }
2547 
2548     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_moff(ns, slba),
2549                                         nvme_m2b(ns, nlb), BDRV_REQ_MAY_UNMAP,
2550                                         nvme_dsm_cb, iocb);
2551     return;
2552 
2553 done:
2554     nvme_dsm_cb(iocb, ret);
2555 }
2556 
2557 static void nvme_dsm_cb(void *opaque, int ret)
2558 {
2559     NvmeDSMAIOCB *iocb = opaque;
2560     NvmeRequest *req = iocb->req;
2561     NvmeCtrl *n = nvme_ctrl(req);
2562     NvmeNamespace *ns = req->ns;
2563     NvmeDsmRange *range;
2564     uint64_t slba;
2565     uint32_t nlb;
2566 
2567     if (iocb->ret < 0) {
2568         goto done;
2569     } else if (ret < 0) {
2570         iocb->ret = ret;
2571         goto done;
2572     }
2573 
2574 next:
2575     if (iocb->idx == iocb->nr) {
2576         goto done;
2577     }
2578 
2579     range = &iocb->range[iocb->idx++];
2580     slba = le64_to_cpu(range->slba);
2581     nlb = le32_to_cpu(range->nlb);
2582 
2583     trace_pci_nvme_dsm_deallocate(slba, nlb);
2584 
2585     if (nlb > n->dmrsl) {
2586         trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
2587         goto next;
2588     }
2589 
2590     if (nvme_check_bounds(ns, slba, nlb)) {
2591         trace_pci_nvme_err_invalid_lba_range(slba, nlb,
2592                                              ns->id_ns.nsze);
2593         goto next;
2594     }
2595 
2596     iocb->aiocb = blk_aio_pdiscard(ns->blkconf.blk, nvme_l2b(ns, slba),
2597                                    nvme_l2b(ns, nlb),
2598                                    nvme_dsm_md_cb, iocb);
2599     return;
2600 
2601 done:
2602     iocb->aiocb = NULL;
2603     iocb->common.cb(iocb->common.opaque, iocb->ret);
2604     qemu_aio_unref(iocb);
2605 }
2606 
2607 static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
2608 {
2609     NvmeNamespace *ns = req->ns;
2610     NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
2611     uint32_t attr = le32_to_cpu(dsm->attributes);
2612     uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
2613     uint16_t status = NVME_SUCCESS;
2614 
2615     trace_pci_nvme_dsm(nr, attr);
2616 
2617     if (attr & NVME_DSMGMT_AD) {
2618         NvmeDSMAIOCB *iocb = blk_aio_get(&nvme_dsm_aiocb_info, ns->blkconf.blk,
2619                                          nvme_misc_cb, req);
2620 
2621         iocb->req = req;
2622         iocb->ret = 0;
2623         iocb->range = g_new(NvmeDsmRange, nr);
2624         iocb->nr = nr;
2625         iocb->idx = 0;
2626 
2627         status = nvme_h2c(n, (uint8_t *)iocb->range, sizeof(NvmeDsmRange) * nr,
2628                           req);
2629         if (status) {
2630             g_free(iocb->range);
2631             qemu_aio_unref(iocb);
2632 
2633             return status;
2634         }
2635 
2636         req->aiocb = &iocb->common;
2637         nvme_dsm_cb(iocb, 0);
2638 
2639         return NVME_NO_COMPLETE;
2640     }
2641 
2642     return status;
2643 }
2644 
2645 static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
2646 {
2647     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2648     NvmeNamespace *ns = req->ns;
2649     BlockBackend *blk = ns->blkconf.blk;
2650     uint64_t slba = le64_to_cpu(rw->slba);
2651     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2652     size_t len = nvme_l2b(ns, nlb);
2653     int64_t offset = nvme_l2b(ns, slba);
2654     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2655     uint32_t reftag = le32_to_cpu(rw->reftag);
2656     NvmeBounceContext *ctx = NULL;
2657     uint16_t status;
2658 
2659     trace_pci_nvme_verify(nvme_cid(req), nvme_nsid(ns), slba, nlb);
2660 
2661     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2662         status = nvme_check_prinfo(ns, prinfo, slba, reftag);
2663         if (status) {
2664             return status;
2665         }
2666 
2667         if (prinfo & NVME_PRINFO_PRACT) {
2668             return NVME_INVALID_PROT_INFO | NVME_DNR;
2669         }
2670     }
2671 
2672     if (len > n->page_size << n->params.vsl) {
2673         return NVME_INVALID_FIELD | NVME_DNR;
2674     }
2675 
2676     status = nvme_check_bounds(ns, slba, nlb);
2677     if (status) {
2678         return status;
2679     }
2680 
2681     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2682         status = nvme_check_dulbe(ns, slba, nlb);
2683         if (status) {
2684             return status;
2685         }
2686     }
2687 
2688     ctx = g_new0(NvmeBounceContext, 1);
2689     ctx->req = req;
2690 
2691     ctx->data.bounce = g_malloc(len);
2692 
2693     qemu_iovec_init(&ctx->data.iov, 1);
2694     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, len);
2695 
2696     block_acct_start(blk_get_stats(blk), &req->acct, ctx->data.iov.size,
2697                      BLOCK_ACCT_READ);
2698 
2699     req->aiocb = blk_aio_preadv(ns->blkconf.blk, offset, &ctx->data.iov, 0,
2700                                 nvme_verify_mdata_in_cb, ctx);
2701     return NVME_NO_COMPLETE;
2702 }
2703 
2704 typedef struct NvmeCopyAIOCB {
2705     BlockAIOCB common;
2706     BlockAIOCB *aiocb;
2707     NvmeRequest *req;
2708     int ret;
2709 
2710     void *ranges;
2711     unsigned int format;
2712     int nr;
2713     int idx;
2714 
2715     uint8_t *bounce;
2716     QEMUIOVector iov;
2717     struct {
2718         BlockAcctCookie read;
2719         BlockAcctCookie write;
2720     } acct;
2721 
2722     uint64_t reftag;
2723     uint64_t slba;
2724 
2725     NvmeZone *zone;
2726 } NvmeCopyAIOCB;
2727 
2728 static void nvme_copy_cancel(BlockAIOCB *aiocb)
2729 {
2730     NvmeCopyAIOCB *iocb = container_of(aiocb, NvmeCopyAIOCB, common);
2731 
2732     iocb->ret = -ECANCELED;
2733 
2734     if (iocb->aiocb) {
2735         blk_aio_cancel_async(iocb->aiocb);
2736         iocb->aiocb = NULL;
2737     }
2738 }
2739 
2740 static const AIOCBInfo nvme_copy_aiocb_info = {
2741     .aiocb_size   = sizeof(NvmeCopyAIOCB),
2742     .cancel_async = nvme_copy_cancel,
2743 };
2744 
2745 static void nvme_copy_done(NvmeCopyAIOCB *iocb)
2746 {
2747     NvmeRequest *req = iocb->req;
2748     NvmeNamespace *ns = req->ns;
2749     BlockAcctStats *stats = blk_get_stats(ns->blkconf.blk);
2750 
2751     if (iocb->idx != iocb->nr) {
2752         req->cqe.result = cpu_to_le32(iocb->idx);
2753     }
2754 
2755     qemu_iovec_destroy(&iocb->iov);
2756     g_free(iocb->bounce);
2757 
2758     if (iocb->ret < 0) {
2759         block_acct_failed(stats, &iocb->acct.read);
2760         block_acct_failed(stats, &iocb->acct.write);
2761     } else {
2762         block_acct_done(stats, &iocb->acct.read);
2763         block_acct_done(stats, &iocb->acct.write);
2764     }
2765 
2766     iocb->common.cb(iocb->common.opaque, iocb->ret);
2767     qemu_aio_unref(iocb);
2768 }
2769 
2770 static void nvme_do_copy(NvmeCopyAIOCB *iocb);
2771 
2772 static void nvme_copy_source_range_parse_format0(void *ranges, int idx,
2773                                                  uint64_t *slba, uint32_t *nlb,
2774                                                  uint16_t *apptag,
2775                                                  uint16_t *appmask,
2776                                                  uint64_t *reftag)
2777 {
2778     NvmeCopySourceRangeFormat0 *_ranges = ranges;
2779 
2780     if (slba) {
2781         *slba = le64_to_cpu(_ranges[idx].slba);
2782     }
2783 
2784     if (nlb) {
2785         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2786     }
2787 
2788     if (apptag) {
2789         *apptag = le16_to_cpu(_ranges[idx].apptag);
2790     }
2791 
2792     if (appmask) {
2793         *appmask = le16_to_cpu(_ranges[idx].appmask);
2794     }
2795 
2796     if (reftag) {
2797         *reftag = le32_to_cpu(_ranges[idx].reftag);
2798     }
2799 }
2800 
2801 static void nvme_copy_source_range_parse_format1(void *ranges, int idx,
2802                                                  uint64_t *slba, uint32_t *nlb,
2803                                                  uint16_t *apptag,
2804                                                  uint16_t *appmask,
2805                                                  uint64_t *reftag)
2806 {
2807     NvmeCopySourceRangeFormat1 *_ranges = ranges;
2808 
2809     if (slba) {
2810         *slba = le64_to_cpu(_ranges[idx].slba);
2811     }
2812 
2813     if (nlb) {
2814         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2815     }
2816 
2817     if (apptag) {
2818         *apptag = le16_to_cpu(_ranges[idx].apptag);
2819     }
2820 
2821     if (appmask) {
2822         *appmask = le16_to_cpu(_ranges[idx].appmask);
2823     }
2824 
2825     if (reftag) {
2826         *reftag = 0;
2827 
2828         *reftag |= (uint64_t)_ranges[idx].sr[4] << 40;
2829         *reftag |= (uint64_t)_ranges[idx].sr[5] << 32;
2830         *reftag |= (uint64_t)_ranges[idx].sr[6] << 24;
2831         *reftag |= (uint64_t)_ranges[idx].sr[7] << 16;
2832         *reftag |= (uint64_t)_ranges[idx].sr[8] << 8;
2833         *reftag |= (uint64_t)_ranges[idx].sr[9];
2834     }
2835 }
2836 
2837 static void nvme_copy_source_range_parse(void *ranges, int idx, uint8_t format,
2838                                          uint64_t *slba, uint32_t *nlb,
2839                                          uint16_t *apptag, uint16_t *appmask,
2840                                          uint64_t *reftag)
2841 {
2842     switch (format) {
2843     case NVME_COPY_FORMAT_0:
2844         nvme_copy_source_range_parse_format0(ranges, idx, slba, nlb, apptag,
2845                                              appmask, reftag);
2846         break;
2847 
2848     case NVME_COPY_FORMAT_1:
2849         nvme_copy_source_range_parse_format1(ranges, idx, slba, nlb, apptag,
2850                                              appmask, reftag);
2851         break;
2852 
2853     default:
2854         abort();
2855     }
2856 }
2857 
2858 static inline uint16_t nvme_check_copy_mcl(NvmeNamespace *ns,
2859                                            NvmeCopyAIOCB *iocb, uint16_t nr)
2860 {
2861     uint32_t copy_len = 0;
2862 
2863     for (int idx = 0; idx < nr; idx++) {
2864         uint32_t nlb;
2865         nvme_copy_source_range_parse(iocb->ranges, idx, iocb->format, NULL,
2866                                      &nlb, NULL, NULL, NULL);
2867         copy_len += nlb + 1;
2868     }
2869 
2870     if (copy_len > ns->id_ns.mcl) {
2871         return NVME_CMD_SIZE_LIMIT | NVME_DNR;
2872     }
2873 
2874     return NVME_SUCCESS;
2875 }
2876 
2877 static void nvme_copy_out_completed_cb(void *opaque, int ret)
2878 {
2879     NvmeCopyAIOCB *iocb = opaque;
2880     NvmeRequest *req = iocb->req;
2881     NvmeNamespace *ns = req->ns;
2882     uint32_t nlb;
2883 
2884     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2885                                  &nlb, NULL, NULL, NULL);
2886 
2887     if (ret < 0) {
2888         iocb->ret = ret;
2889         goto out;
2890     } else if (iocb->ret < 0) {
2891         goto out;
2892     }
2893 
2894     if (ns->params.zoned) {
2895         nvme_advance_zone_wp(ns, iocb->zone, nlb);
2896     }
2897 
2898     iocb->idx++;
2899     iocb->slba += nlb;
2900 out:
2901     nvme_do_copy(iocb);
2902 }
2903 
2904 static void nvme_copy_out_cb(void *opaque, int ret)
2905 {
2906     NvmeCopyAIOCB *iocb = opaque;
2907     NvmeRequest *req = iocb->req;
2908     NvmeNamespace *ns = req->ns;
2909     uint32_t nlb;
2910     size_t mlen;
2911     uint8_t *mbounce;
2912 
2913     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
2914         goto out;
2915     }
2916 
2917     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2918                                  &nlb, NULL, NULL, NULL);
2919 
2920     mlen = nvme_m2b(ns, nlb);
2921     mbounce = iocb->bounce + nvme_l2b(ns, nlb);
2922 
2923     qemu_iovec_reset(&iocb->iov);
2924     qemu_iovec_add(&iocb->iov, mbounce, mlen);
2925 
2926     iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_moff(ns, iocb->slba),
2927                                   &iocb->iov, 0, nvme_copy_out_completed_cb,
2928                                   iocb);
2929 
2930     return;
2931 
2932 out:
2933     nvme_copy_out_completed_cb(iocb, ret);
2934 }
2935 
2936 static void nvme_copy_in_completed_cb(void *opaque, int ret)
2937 {
2938     NvmeCopyAIOCB *iocb = opaque;
2939     NvmeRequest *req = iocb->req;
2940     NvmeNamespace *ns = req->ns;
2941     uint32_t nlb;
2942     uint64_t slba;
2943     uint16_t apptag, appmask;
2944     uint64_t reftag;
2945     size_t len;
2946     uint16_t status;
2947 
2948     if (ret < 0) {
2949         iocb->ret = ret;
2950         goto out;
2951     } else if (iocb->ret < 0) {
2952         goto out;
2953     }
2954 
2955     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
2956                                  &nlb, &apptag, &appmask, &reftag);
2957     len = nvme_l2b(ns, nlb);
2958 
2959     trace_pci_nvme_copy_out(iocb->slba, nlb);
2960 
2961     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2962         NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2963 
2964         uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
2965         uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
2966 
2967         size_t mlen = nvme_m2b(ns, nlb);
2968         uint8_t *mbounce = iocb->bounce + nvme_l2b(ns, nlb);
2969 
2970         status = nvme_dif_mangle_mdata(ns, mbounce, mlen, slba);
2971         if (status) {
2972             goto invalid;
2973         }
2974         status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen, prinfor,
2975                                 slba, apptag, appmask, &reftag);
2976         if (status) {
2977             goto invalid;
2978         }
2979 
2980         apptag = le16_to_cpu(copy->apptag);
2981         appmask = le16_to_cpu(copy->appmask);
2982 
2983         if (prinfow & NVME_PRINFO_PRACT) {
2984             status = nvme_check_prinfo(ns, prinfow, iocb->slba, iocb->reftag);
2985             if (status) {
2986                 goto invalid;
2987             }
2988 
2989             nvme_dif_pract_generate_dif(ns, iocb->bounce, len, mbounce, mlen,
2990                                         apptag, &iocb->reftag);
2991         } else {
2992             status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen,
2993                                     prinfow, iocb->slba, apptag, appmask,
2994                                     &iocb->reftag);
2995             if (status) {
2996                 goto invalid;
2997             }
2998         }
2999     }
3000 
3001     status = nvme_check_bounds(ns, iocb->slba, nlb);
3002     if (status) {
3003         goto invalid;
3004     }
3005 
3006     if (ns->params.zoned) {
3007         status = nvme_check_zone_write(ns, iocb->zone, iocb->slba, nlb);
3008         if (status) {
3009             goto invalid;
3010         }
3011 
3012         if (!(iocb->zone->d.za & NVME_ZA_ZRWA_VALID)) {
3013             iocb->zone->w_ptr += nlb;
3014         }
3015     }
3016 
3017     qemu_iovec_reset(&iocb->iov);
3018     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3019 
3020     iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, iocb->slba),
3021                                   &iocb->iov, 0, nvme_copy_out_cb, iocb);
3022 
3023     return;
3024 
3025 invalid:
3026     req->status = status;
3027     iocb->ret = -1;
3028 out:
3029     nvme_do_copy(iocb);
3030 }
3031 
3032 static void nvme_copy_in_cb(void *opaque, int ret)
3033 {
3034     NvmeCopyAIOCB *iocb = opaque;
3035     NvmeRequest *req = iocb->req;
3036     NvmeNamespace *ns = req->ns;
3037     uint64_t slba;
3038     uint32_t nlb;
3039 
3040     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
3041         goto out;
3042     }
3043 
3044     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
3045                                  &nlb, NULL, NULL, NULL);
3046 
3047     qemu_iovec_reset(&iocb->iov);
3048     qemu_iovec_add(&iocb->iov, iocb->bounce + nvme_l2b(ns, nlb),
3049                    nvme_m2b(ns, nlb));
3050 
3051     iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_moff(ns, slba),
3052                                  &iocb->iov, 0, nvme_copy_in_completed_cb,
3053                                  iocb);
3054     return;
3055 
3056 out:
3057     nvme_copy_in_completed_cb(iocb, ret);
3058 }
3059 
3060 static void nvme_do_copy(NvmeCopyAIOCB *iocb)
3061 {
3062     NvmeRequest *req = iocb->req;
3063     NvmeNamespace *ns = req->ns;
3064     uint64_t slba;
3065     uint32_t nlb;
3066     size_t len;
3067     uint16_t status;
3068 
3069     if (iocb->ret < 0) {
3070         goto done;
3071     }
3072 
3073     if (iocb->idx == iocb->nr) {
3074         goto done;
3075     }
3076 
3077     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
3078                                  &nlb, NULL, NULL, NULL);
3079     len = nvme_l2b(ns, nlb);
3080 
3081     trace_pci_nvme_copy_source_range(slba, nlb);
3082 
3083     if (nlb > le16_to_cpu(ns->id_ns.mssrl)) {
3084         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3085         goto invalid;
3086     }
3087 
3088     status = nvme_check_bounds(ns, slba, nlb);
3089     if (status) {
3090         goto invalid;
3091     }
3092 
3093     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3094         status = nvme_check_dulbe(ns, slba, nlb);
3095         if (status) {
3096             goto invalid;
3097         }
3098     }
3099 
3100     if (ns->params.zoned) {
3101         status = nvme_check_zone_read(ns, slba, nlb);
3102         if (status) {
3103             goto invalid;
3104         }
3105     }
3106 
3107     qemu_iovec_reset(&iocb->iov);
3108     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3109 
3110     iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_l2b(ns, slba),
3111                                  &iocb->iov, 0, nvme_copy_in_cb, iocb);
3112     return;
3113 
3114 invalid:
3115     req->status = status;
3116     iocb->ret = -1;
3117 done:
3118     nvme_copy_done(iocb);
3119 }
3120 
3121 static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
3122 {
3123     NvmeNamespace *ns = req->ns;
3124     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
3125     NvmeCopyAIOCB *iocb = blk_aio_get(&nvme_copy_aiocb_info, ns->blkconf.blk,
3126                                       nvme_misc_cb, req);
3127     uint16_t nr = copy->nr + 1;
3128     uint8_t format = copy->control[0] & 0xf;
3129     uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
3130     uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
3131     size_t len = sizeof(NvmeCopySourceRangeFormat0);
3132 
3133     uint16_t status;
3134 
3135     trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);
3136 
3137     iocb->ranges = NULL;
3138     iocb->zone = NULL;
3139 
3140     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) &&
3141         ((prinfor & NVME_PRINFO_PRACT) != (prinfow & NVME_PRINFO_PRACT))) {
3142         status = NVME_INVALID_FIELD | NVME_DNR;
3143         goto invalid;
3144     }
3145 
3146     if (!(n->id_ctrl.ocfs & (1 << format))) {
3147         trace_pci_nvme_err_copy_invalid_format(format);
3148         status = NVME_INVALID_FIELD | NVME_DNR;
3149         goto invalid;
3150     }
3151 
3152     if (nr > ns->id_ns.msrc + 1) {
3153         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3154         goto invalid;
3155     }
3156 
3157     if ((ns->pif == 0x0 && format != 0x0) ||
3158         (ns->pif != 0x0 && format != 0x1)) {
3159         status = NVME_INVALID_FORMAT | NVME_DNR;
3160         goto invalid;
3161     }
3162 
3163     if (ns->pif) {
3164         len = sizeof(NvmeCopySourceRangeFormat1);
3165     }
3166 
3167     iocb->format = format;
3168     iocb->ranges = g_malloc_n(nr, len);
3169     status = nvme_h2c(n, (uint8_t *)iocb->ranges, len * nr, req);
3170     if (status) {
3171         goto invalid;
3172     }
3173 
3174     iocb->slba = le64_to_cpu(copy->sdlba);
3175 
3176     if (ns->params.zoned) {
3177         iocb->zone = nvme_get_zone_by_slba(ns, iocb->slba);
3178         if (!iocb->zone) {
3179             status = NVME_LBA_RANGE | NVME_DNR;
3180             goto invalid;
3181         }
3182 
3183         status = nvme_zrm_auto(n, ns, iocb->zone);
3184         if (status) {
3185             goto invalid;
3186         }
3187     }
3188 
3189     status = nvme_check_copy_mcl(ns, iocb, nr);
3190     if (status) {
3191         goto invalid;
3192     }
3193 
3194     iocb->req = req;
3195     iocb->ret = 0;
3196     iocb->nr = nr;
3197     iocb->idx = 0;
3198     iocb->reftag = le32_to_cpu(copy->reftag);
3199     iocb->reftag |= (uint64_t)le32_to_cpu(copy->cdw3) << 32;
3200     iocb->bounce = g_malloc_n(le16_to_cpu(ns->id_ns.mssrl),
3201                               ns->lbasz + ns->lbaf.ms);
3202 
3203     qemu_iovec_init(&iocb->iov, 1);
3204 
3205     block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.read, 0,
3206                      BLOCK_ACCT_READ);
3207     block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.write, 0,
3208                      BLOCK_ACCT_WRITE);
3209 
3210     req->aiocb = &iocb->common;
3211     nvme_do_copy(iocb);
3212 
3213     return NVME_NO_COMPLETE;
3214 
3215 invalid:
3216     g_free(iocb->ranges);
3217     qemu_aio_unref(iocb);
3218     return status;
3219 }
3220 
3221 static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req)
3222 {
3223     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3224     NvmeNamespace *ns = req->ns;
3225     BlockBackend *blk = ns->blkconf.blk;
3226     uint64_t slba = le64_to_cpu(rw->slba);
3227     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
3228     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3229     size_t data_len = nvme_l2b(ns, nlb);
3230     size_t len = data_len;
3231     int64_t offset = nvme_l2b(ns, slba);
3232     struct nvme_compare_ctx *ctx = NULL;
3233     uint16_t status;
3234 
3235     trace_pci_nvme_compare(nvme_cid(req), nvme_nsid(ns), slba, nlb);
3236 
3237     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) && (prinfo & NVME_PRINFO_PRACT)) {
3238         return NVME_INVALID_PROT_INFO | NVME_DNR;
3239     }
3240 
3241     if (nvme_ns_ext(ns)) {
3242         len += nvme_m2b(ns, nlb);
3243     }
3244 
3245     status = nvme_check_mdts(n, len);
3246     if (status) {
3247         return status;
3248     }
3249 
3250     status = nvme_check_bounds(ns, slba, nlb);
3251     if (status) {
3252         return status;
3253     }
3254 
3255     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3256         status = nvme_check_dulbe(ns, slba, nlb);
3257         if (status) {
3258             return status;
3259         }
3260     }
3261 
3262     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
3263     if (status) {
3264         return status;
3265     }
3266 
3267     ctx = g_new(struct nvme_compare_ctx, 1);
3268     ctx->data.bounce = g_malloc(data_len);
3269 
3270     req->opaque = ctx;
3271 
3272     qemu_iovec_init(&ctx->data.iov, 1);
3273     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, data_len);
3274 
3275     block_acct_start(blk_get_stats(blk), &req->acct, data_len,
3276                      BLOCK_ACCT_READ);
3277     req->aiocb = blk_aio_preadv(blk, offset, &ctx->data.iov, 0,
3278                                 nvme_compare_data_cb, req);
3279 
3280     return NVME_NO_COMPLETE;
3281 }
3282 
3283 typedef struct NvmeFlushAIOCB {
3284     BlockAIOCB common;
3285     BlockAIOCB *aiocb;
3286     NvmeRequest *req;
3287     int ret;
3288 
3289     NvmeNamespace *ns;
3290     uint32_t nsid;
3291     bool broadcast;
3292 } NvmeFlushAIOCB;
3293 
3294 static void nvme_flush_cancel(BlockAIOCB *acb)
3295 {
3296     NvmeFlushAIOCB *iocb = container_of(acb, NvmeFlushAIOCB, common);
3297 
3298     iocb->ret = -ECANCELED;
3299 
3300     if (iocb->aiocb) {
3301         blk_aio_cancel_async(iocb->aiocb);
3302         iocb->aiocb = NULL;
3303     }
3304 }
3305 
3306 static const AIOCBInfo nvme_flush_aiocb_info = {
3307     .aiocb_size = sizeof(NvmeFlushAIOCB),
3308     .cancel_async = nvme_flush_cancel,
3309     .get_aio_context = nvme_get_aio_context,
3310 };
3311 
3312 static void nvme_do_flush(NvmeFlushAIOCB *iocb);
3313 
3314 static void nvme_flush_ns_cb(void *opaque, int ret)
3315 {
3316     NvmeFlushAIOCB *iocb = opaque;
3317     NvmeNamespace *ns = iocb->ns;
3318 
3319     if (ret < 0) {
3320         iocb->ret = ret;
3321         goto out;
3322     } else if (iocb->ret < 0) {
3323         goto out;
3324     }
3325 
3326     if (ns) {
3327         trace_pci_nvme_flush_ns(iocb->nsid);
3328 
3329         iocb->ns = NULL;
3330         iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);
3331         return;
3332     }
3333 
3334 out:
3335     nvme_do_flush(iocb);
3336 }
3337 
3338 static void nvme_do_flush(NvmeFlushAIOCB *iocb)
3339 {
3340     NvmeRequest *req = iocb->req;
3341     NvmeCtrl *n = nvme_ctrl(req);
3342     int i;
3343 
3344     if (iocb->ret < 0) {
3345         goto done;
3346     }
3347 
3348     if (iocb->broadcast) {
3349         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
3350             iocb->ns = nvme_ns(n, i);
3351             if (iocb->ns) {
3352                 iocb->nsid = i;
3353                 break;
3354             }
3355         }
3356     }
3357 
3358     if (!iocb->ns) {
3359         goto done;
3360     }
3361 
3362     nvme_flush_ns_cb(iocb, 0);
3363     return;
3364 
3365 done:
3366     iocb->common.cb(iocb->common.opaque, iocb->ret);
3367     qemu_aio_unref(iocb);
3368 }
3369 
3370 static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req)
3371 {
3372     NvmeFlushAIOCB *iocb;
3373     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
3374     uint16_t status;
3375 
3376     iocb = qemu_aio_get(&nvme_flush_aiocb_info, NULL, nvme_misc_cb, req);
3377 
3378     iocb->req = req;
3379     iocb->ret = 0;
3380     iocb->ns = NULL;
3381     iocb->nsid = 0;
3382     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
3383 
3384     if (!iocb->broadcast) {
3385         if (!nvme_nsid_valid(n, nsid)) {
3386             status = NVME_INVALID_NSID | NVME_DNR;
3387             goto out;
3388         }
3389 
3390         iocb->ns = nvme_ns(n, nsid);
3391         if (!iocb->ns) {
3392             status = NVME_INVALID_FIELD | NVME_DNR;
3393             goto out;
3394         }
3395 
3396         iocb->nsid = nsid;
3397     }
3398 
3399     req->aiocb = &iocb->common;
3400     nvme_do_flush(iocb);
3401 
3402     return NVME_NO_COMPLETE;
3403 
3404 out:
3405     qemu_aio_unref(iocb);
3406 
3407     return status;
3408 }
3409 
3410 static uint16_t nvme_read(NvmeCtrl *n, NvmeRequest *req)
3411 {
3412     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3413     NvmeNamespace *ns = req->ns;
3414     uint64_t slba = le64_to_cpu(rw->slba);
3415     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3416     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3417     uint64_t data_size = nvme_l2b(ns, nlb);
3418     uint64_t mapped_size = data_size;
3419     uint64_t data_offset;
3420     BlockBackend *blk = ns->blkconf.blk;
3421     uint16_t status;
3422 
3423     if (nvme_ns_ext(ns)) {
3424         mapped_size += nvme_m2b(ns, nlb);
3425 
3426         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3427             bool pract = prinfo & NVME_PRINFO_PRACT;
3428 
3429             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3430                 mapped_size = data_size;
3431             }
3432         }
3433     }
3434 
3435     trace_pci_nvme_read(nvme_cid(req), nvme_nsid(ns), nlb, mapped_size, slba);
3436 
3437     status = nvme_check_mdts(n, mapped_size);
3438     if (status) {
3439         goto invalid;
3440     }
3441 
3442     status = nvme_check_bounds(ns, slba, nlb);
3443     if (status) {
3444         goto invalid;
3445     }
3446 
3447     if (ns->params.zoned) {
3448         status = nvme_check_zone_read(ns, slba, nlb);
3449         if (status) {
3450             trace_pci_nvme_err_zone_read_not_ok(slba, nlb, status);
3451             goto invalid;
3452         }
3453     }
3454 
3455     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3456         status = nvme_check_dulbe(ns, slba, nlb);
3457         if (status) {
3458             goto invalid;
3459         }
3460     }
3461 
3462     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3463         return nvme_dif_rw(n, req);
3464     }
3465 
3466     status = nvme_map_data(n, nlb, req);
3467     if (status) {
3468         goto invalid;
3469     }
3470 
3471     data_offset = nvme_l2b(ns, slba);
3472 
3473     block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3474                      BLOCK_ACCT_READ);
3475     nvme_blk_read(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3476     return NVME_NO_COMPLETE;
3477 
3478 invalid:
3479     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_READ);
3480     return status | NVME_DNR;
3481 }
3482 
3483 static void nvme_do_write_fdp(NvmeCtrl *n, NvmeRequest *req, uint64_t slba,
3484                               uint32_t nlb)
3485 {
3486     NvmeNamespace *ns = req->ns;
3487     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3488     uint64_t data_size = nvme_l2b(ns, nlb);
3489     uint32_t dw12 = le32_to_cpu(req->cmd.cdw12);
3490     uint8_t dtype = (dw12 >> 20) & 0xf;
3491     uint16_t pid = le16_to_cpu(rw->dspec);
3492     uint16_t ph, rg, ruhid;
3493     NvmeReclaimUnit *ru;
3494 
3495     if (dtype != NVME_DIRECTIVE_DATA_PLACEMENT ||
3496         !nvme_parse_pid(ns, pid, &ph, &rg)) {
3497         ph = 0;
3498         rg = 0;
3499     }
3500 
3501     ruhid = ns->fdp.phs[ph];
3502     ru = &ns->endgrp->fdp.ruhs[ruhid].rus[rg];
3503 
3504     nvme_fdp_stat_inc(&ns->endgrp->fdp.hbmw, data_size);
3505     nvme_fdp_stat_inc(&ns->endgrp->fdp.mbmw, data_size);
3506 
3507     while (nlb) {
3508         if (nlb < ru->ruamw) {
3509             ru->ruamw -= nlb;
3510             break;
3511         }
3512 
3513         nlb -= ru->ruamw;
3514         nvme_update_ruh(n, ns, pid);
3515     }
3516 }
3517 
3518 static uint16_t nvme_do_write(NvmeCtrl *n, NvmeRequest *req, bool append,
3519                               bool wrz)
3520 {
3521     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3522     NvmeNamespace *ns = req->ns;
3523     uint64_t slba = le64_to_cpu(rw->slba);
3524     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3525     uint16_t ctrl = le16_to_cpu(rw->control);
3526     uint8_t prinfo = NVME_RW_PRINFO(ctrl);
3527     uint64_t data_size = nvme_l2b(ns, nlb);
3528     uint64_t mapped_size = data_size;
3529     uint64_t data_offset;
3530     NvmeZone *zone;
3531     NvmeZonedResult *res = (NvmeZonedResult *)&req->cqe;
3532     BlockBackend *blk = ns->blkconf.blk;
3533     uint16_t status;
3534 
3535     if (nvme_ns_ext(ns)) {
3536         mapped_size += nvme_m2b(ns, nlb);
3537 
3538         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3539             bool pract = prinfo & NVME_PRINFO_PRACT;
3540 
3541             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3542                 mapped_size -= nvme_m2b(ns, nlb);
3543             }
3544         }
3545     }
3546 
3547     trace_pci_nvme_write(nvme_cid(req), nvme_io_opc_str(rw->opcode),
3548                          nvme_nsid(ns), nlb, mapped_size, slba);
3549 
3550     if (!wrz) {
3551         status = nvme_check_mdts(n, mapped_size);
3552         if (status) {
3553             goto invalid;
3554         }
3555     }
3556 
3557     status = nvme_check_bounds(ns, slba, nlb);
3558     if (status) {
3559         goto invalid;
3560     }
3561 
3562     if (ns->params.zoned) {
3563         zone = nvme_get_zone_by_slba(ns, slba);
3564         assert(zone);
3565 
3566         if (append) {
3567             bool piremap = !!(ctrl & NVME_RW_PIREMAP);
3568 
3569             if (unlikely(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3570                 return NVME_INVALID_ZONE_OP | NVME_DNR;
3571             }
3572 
3573             if (unlikely(slba != zone->d.zslba)) {
3574                 trace_pci_nvme_err_append_not_at_start(slba, zone->d.zslba);
3575                 status = NVME_INVALID_FIELD;
3576                 goto invalid;
3577             }
3578 
3579             if (n->params.zasl &&
3580                 data_size > (uint64_t)n->page_size << n->params.zasl) {
3581                 trace_pci_nvme_err_zasl(data_size);
3582                 return NVME_INVALID_FIELD | NVME_DNR;
3583             }
3584 
3585             slba = zone->w_ptr;
3586             rw->slba = cpu_to_le64(slba);
3587             res->slba = cpu_to_le64(slba);
3588 
3589             switch (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3590             case NVME_ID_NS_DPS_TYPE_1:
3591                 if (!piremap) {
3592                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3593                 }
3594 
3595                 /* fallthrough */
3596 
3597             case NVME_ID_NS_DPS_TYPE_2:
3598                 if (piremap) {
3599                     uint32_t reftag = le32_to_cpu(rw->reftag);
3600                     rw->reftag = cpu_to_le32(reftag + (slba - zone->d.zslba));
3601                 }
3602 
3603                 break;
3604 
3605             case NVME_ID_NS_DPS_TYPE_3:
3606                 if (piremap) {
3607                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3608                 }
3609 
3610                 break;
3611             }
3612         }
3613 
3614         status = nvme_check_zone_write(ns, zone, slba, nlb);
3615         if (status) {
3616             goto invalid;
3617         }
3618 
3619         status = nvme_zrm_auto(n, ns, zone);
3620         if (status) {
3621             goto invalid;
3622         }
3623 
3624         if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3625             zone->w_ptr += nlb;
3626         }
3627     } else if (ns->endgrp && ns->endgrp->fdp.enabled) {
3628         nvme_do_write_fdp(n, req, slba, nlb);
3629     }
3630 
3631     data_offset = nvme_l2b(ns, slba);
3632 
3633     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3634         return nvme_dif_rw(n, req);
3635     }
3636 
3637     if (!wrz) {
3638         status = nvme_map_data(n, nlb, req);
3639         if (status) {
3640             goto invalid;
3641         }
3642 
3643         block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3644                          BLOCK_ACCT_WRITE);
3645         nvme_blk_write(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3646     } else {
3647         req->aiocb = blk_aio_pwrite_zeroes(blk, data_offset, data_size,
3648                                            BDRV_REQ_MAY_UNMAP, nvme_rw_cb,
3649                                            req);
3650     }
3651 
3652     return NVME_NO_COMPLETE;
3653 
3654 invalid:
3655     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_WRITE);
3656     return status | NVME_DNR;
3657 }
3658 
3659 static inline uint16_t nvme_write(NvmeCtrl *n, NvmeRequest *req)
3660 {
3661     return nvme_do_write(n, req, false, false);
3662 }
3663 
3664 static inline uint16_t nvme_write_zeroes(NvmeCtrl *n, NvmeRequest *req)
3665 {
3666     return nvme_do_write(n, req, false, true);
3667 }
3668 
3669 static inline uint16_t nvme_zone_append(NvmeCtrl *n, NvmeRequest *req)
3670 {
3671     return nvme_do_write(n, req, true, false);
3672 }
3673 
3674 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace *ns, NvmeCmd *c,
3675                                             uint64_t *slba, uint32_t *zone_idx)
3676 {
3677     uint32_t dw10 = le32_to_cpu(c->cdw10);
3678     uint32_t dw11 = le32_to_cpu(c->cdw11);
3679 
3680     if (!ns->params.zoned) {
3681         trace_pci_nvme_err_invalid_opc(c->opcode);
3682         return NVME_INVALID_OPCODE | NVME_DNR;
3683     }
3684 
3685     *slba = ((uint64_t)dw11) << 32 | dw10;
3686     if (unlikely(*slba >= ns->id_ns.nsze)) {
3687         trace_pci_nvme_err_invalid_lba_range(*slba, 0, ns->id_ns.nsze);
3688         *slba = 0;
3689         return NVME_LBA_RANGE | NVME_DNR;
3690     }
3691 
3692     *zone_idx = nvme_zone_idx(ns, *slba);
3693     assert(*zone_idx < ns->num_zones);
3694 
3695     return NVME_SUCCESS;
3696 }
3697 
3698 typedef uint16_t (*op_handler_t)(NvmeNamespace *, NvmeZone *, NvmeZoneState,
3699                                  NvmeRequest *);
3700 
3701 enum NvmeZoneProcessingMask {
3702     NVME_PROC_CURRENT_ZONE    = 0,
3703     NVME_PROC_OPENED_ZONES    = 1 << 0,
3704     NVME_PROC_CLOSED_ZONES    = 1 << 1,
3705     NVME_PROC_READ_ONLY_ZONES = 1 << 2,
3706     NVME_PROC_FULL_ZONES      = 1 << 3,
3707 };
3708 
3709 static uint16_t nvme_open_zone(NvmeNamespace *ns, NvmeZone *zone,
3710                                NvmeZoneState state, NvmeRequest *req)
3711 {
3712     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
3713     int flags = 0;
3714 
3715     if (cmd->zsflags & NVME_ZSFLAG_ZRWA_ALLOC) {
3716         uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3717 
3718         if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3719             return NVME_INVALID_ZONE_OP | NVME_DNR;
3720         }
3721 
3722         if (zone->w_ptr % ns->zns.zrwafg) {
3723             return NVME_NOZRWA | NVME_DNR;
3724         }
3725 
3726         flags = NVME_ZRM_ZRWA;
3727     }
3728 
3729     return nvme_zrm_open_flags(nvme_ctrl(req), ns, zone, flags);
3730 }
3731 
3732 static uint16_t nvme_close_zone(NvmeNamespace *ns, NvmeZone *zone,
3733                                 NvmeZoneState state, NvmeRequest *req)
3734 {
3735     return nvme_zrm_close(ns, zone);
3736 }
3737 
3738 static uint16_t nvme_finish_zone(NvmeNamespace *ns, NvmeZone *zone,
3739                                  NvmeZoneState state, NvmeRequest *req)
3740 {
3741     return nvme_zrm_finish(ns, zone);
3742 }
3743 
3744 static uint16_t nvme_offline_zone(NvmeNamespace *ns, NvmeZone *zone,
3745                                   NvmeZoneState state, NvmeRequest *req)
3746 {
3747     switch (state) {
3748     case NVME_ZONE_STATE_READ_ONLY:
3749         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_OFFLINE);
3750         /* fall through */
3751     case NVME_ZONE_STATE_OFFLINE:
3752         return NVME_SUCCESS;
3753     default:
3754         return NVME_ZONE_INVAL_TRANSITION;
3755     }
3756 }
3757 
3758 static uint16_t nvme_set_zd_ext(NvmeNamespace *ns, NvmeZone *zone)
3759 {
3760     uint16_t status;
3761     uint8_t state = nvme_get_zone_state(zone);
3762 
3763     if (state == NVME_ZONE_STATE_EMPTY) {
3764         status = nvme_aor_check(ns, 1, 0);
3765         if (status) {
3766             return status;
3767         }
3768         nvme_aor_inc_active(ns);
3769         zone->d.za |= NVME_ZA_ZD_EXT_VALID;
3770         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
3771         return NVME_SUCCESS;
3772     }
3773 
3774     return NVME_ZONE_INVAL_TRANSITION;
3775 }
3776 
3777 static uint16_t nvme_bulk_proc_zone(NvmeNamespace *ns, NvmeZone *zone,
3778                                     enum NvmeZoneProcessingMask proc_mask,
3779                                     op_handler_t op_hndlr, NvmeRequest *req)
3780 {
3781     uint16_t status = NVME_SUCCESS;
3782     NvmeZoneState zs = nvme_get_zone_state(zone);
3783     bool proc_zone;
3784 
3785     switch (zs) {
3786     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3787     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3788         proc_zone = proc_mask & NVME_PROC_OPENED_ZONES;
3789         break;
3790     case NVME_ZONE_STATE_CLOSED:
3791         proc_zone = proc_mask & NVME_PROC_CLOSED_ZONES;
3792         break;
3793     case NVME_ZONE_STATE_READ_ONLY:
3794         proc_zone = proc_mask & NVME_PROC_READ_ONLY_ZONES;
3795         break;
3796     case NVME_ZONE_STATE_FULL:
3797         proc_zone = proc_mask & NVME_PROC_FULL_ZONES;
3798         break;
3799     default:
3800         proc_zone = false;
3801     }
3802 
3803     if (proc_zone) {
3804         status = op_hndlr(ns, zone, zs, req);
3805     }
3806 
3807     return status;
3808 }
3809 
3810 static uint16_t nvme_do_zone_op(NvmeNamespace *ns, NvmeZone *zone,
3811                                 enum NvmeZoneProcessingMask proc_mask,
3812                                 op_handler_t op_hndlr, NvmeRequest *req)
3813 {
3814     NvmeZone *next;
3815     uint16_t status = NVME_SUCCESS;
3816     int i;
3817 
3818     if (!proc_mask) {
3819         status = op_hndlr(ns, zone, nvme_get_zone_state(zone), req);
3820     } else {
3821         if (proc_mask & NVME_PROC_CLOSED_ZONES) {
3822             QTAILQ_FOREACH_SAFE(zone, &ns->closed_zones, entry, next) {
3823                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3824                                              req);
3825                 if (status && status != NVME_NO_COMPLETE) {
3826                     goto out;
3827                 }
3828             }
3829         }
3830         if (proc_mask & NVME_PROC_OPENED_ZONES) {
3831             QTAILQ_FOREACH_SAFE(zone, &ns->imp_open_zones, entry, next) {
3832                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3833                                              req);
3834                 if (status && status != NVME_NO_COMPLETE) {
3835                     goto out;
3836                 }
3837             }
3838 
3839             QTAILQ_FOREACH_SAFE(zone, &ns->exp_open_zones, entry, next) {
3840                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3841                                              req);
3842                 if (status && status != NVME_NO_COMPLETE) {
3843                     goto out;
3844                 }
3845             }
3846         }
3847         if (proc_mask & NVME_PROC_FULL_ZONES) {
3848             QTAILQ_FOREACH_SAFE(zone, &ns->full_zones, entry, next) {
3849                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3850                                              req);
3851                 if (status && status != NVME_NO_COMPLETE) {
3852                     goto out;
3853                 }
3854             }
3855         }
3856 
3857         if (proc_mask & NVME_PROC_READ_ONLY_ZONES) {
3858             for (i = 0; i < ns->num_zones; i++, zone++) {
3859                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3860                                              req);
3861                 if (status && status != NVME_NO_COMPLETE) {
3862                     goto out;
3863                 }
3864             }
3865         }
3866     }
3867 
3868 out:
3869     return status;
3870 }
3871 
3872 typedef struct NvmeZoneResetAIOCB {
3873     BlockAIOCB common;
3874     BlockAIOCB *aiocb;
3875     NvmeRequest *req;
3876     int ret;
3877 
3878     bool all;
3879     int idx;
3880     NvmeZone *zone;
3881 } NvmeZoneResetAIOCB;
3882 
3883 static void nvme_zone_reset_cancel(BlockAIOCB *aiocb)
3884 {
3885     NvmeZoneResetAIOCB *iocb = container_of(aiocb, NvmeZoneResetAIOCB, common);
3886     NvmeRequest *req = iocb->req;
3887     NvmeNamespace *ns = req->ns;
3888 
3889     iocb->idx = ns->num_zones;
3890 
3891     iocb->ret = -ECANCELED;
3892 
3893     if (iocb->aiocb) {
3894         blk_aio_cancel_async(iocb->aiocb);
3895         iocb->aiocb = NULL;
3896     }
3897 }
3898 
3899 static const AIOCBInfo nvme_zone_reset_aiocb_info = {
3900     .aiocb_size = sizeof(NvmeZoneResetAIOCB),
3901     .cancel_async = nvme_zone_reset_cancel,
3902 };
3903 
3904 static void nvme_zone_reset_cb(void *opaque, int ret);
3905 
3906 static void nvme_zone_reset_epilogue_cb(void *opaque, int ret)
3907 {
3908     NvmeZoneResetAIOCB *iocb = opaque;
3909     NvmeRequest *req = iocb->req;
3910     NvmeNamespace *ns = req->ns;
3911     int64_t moff;
3912     int count;
3913 
3914     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
3915         goto out;
3916     }
3917 
3918     moff = nvme_moff(ns, iocb->zone->d.zslba);
3919     count = nvme_m2b(ns, ns->zone_size);
3920 
3921     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, moff, count,
3922                                         BDRV_REQ_MAY_UNMAP,
3923                                         nvme_zone_reset_cb, iocb);
3924     return;
3925 
3926 out:
3927     nvme_zone_reset_cb(iocb, ret);
3928 }
3929 
3930 static void nvme_zone_reset_cb(void *opaque, int ret)
3931 {
3932     NvmeZoneResetAIOCB *iocb = opaque;
3933     NvmeRequest *req = iocb->req;
3934     NvmeNamespace *ns = req->ns;
3935 
3936     if (iocb->ret < 0) {
3937         goto done;
3938     } else if (ret < 0) {
3939         iocb->ret = ret;
3940         goto done;
3941     }
3942 
3943     if (iocb->zone) {
3944         nvme_zrm_reset(ns, iocb->zone);
3945 
3946         if (!iocb->all) {
3947             goto done;
3948         }
3949     }
3950 
3951     while (iocb->idx < ns->num_zones) {
3952         NvmeZone *zone = &ns->zone_array[iocb->idx++];
3953 
3954         switch (nvme_get_zone_state(zone)) {
3955         case NVME_ZONE_STATE_EMPTY:
3956             if (!iocb->all) {
3957                 goto done;
3958             }
3959 
3960             continue;
3961 
3962         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3963         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3964         case NVME_ZONE_STATE_CLOSED:
3965         case NVME_ZONE_STATE_FULL:
3966             iocb->zone = zone;
3967             break;
3968 
3969         default:
3970             continue;
3971         }
3972 
3973         trace_pci_nvme_zns_zone_reset(zone->d.zslba);
3974 
3975         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk,
3976                                             nvme_l2b(ns, zone->d.zslba),
3977                                             nvme_l2b(ns, ns->zone_size),
3978                                             BDRV_REQ_MAY_UNMAP,
3979                                             nvme_zone_reset_epilogue_cb,
3980                                             iocb);
3981         return;
3982     }
3983 
3984 done:
3985     iocb->aiocb = NULL;
3986 
3987     iocb->common.cb(iocb->common.opaque, iocb->ret);
3988     qemu_aio_unref(iocb);
3989 }
3990 
3991 static uint16_t nvme_zone_mgmt_send_zrwa_flush(NvmeCtrl *n, NvmeZone *zone,
3992                                                uint64_t elba, NvmeRequest *req)
3993 {
3994     NvmeNamespace *ns = req->ns;
3995     uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3996     uint64_t wp = zone->d.wp;
3997     uint32_t nlb = elba - wp + 1;
3998     uint16_t status;
3999 
4000 
4001     if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
4002         return NVME_INVALID_ZONE_OP | NVME_DNR;
4003     }
4004 
4005     if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
4006         return NVME_INVALID_FIELD | NVME_DNR;
4007     }
4008 
4009     if (elba < wp || elba > wp + ns->zns.zrwas) {
4010         return NVME_ZONE_BOUNDARY_ERROR | NVME_DNR;
4011     }
4012 
4013     if (nlb % ns->zns.zrwafg) {
4014         return NVME_INVALID_FIELD | NVME_DNR;
4015     }
4016 
4017     status = nvme_zrm_auto(n, ns, zone);
4018     if (status) {
4019         return status;
4020     }
4021 
4022     zone->w_ptr += nlb;
4023 
4024     nvme_advance_zone_wp(ns, zone, nlb);
4025 
4026     return NVME_SUCCESS;
4027 }
4028 
4029 static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4030 {
4031     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
4032     NvmeNamespace *ns = req->ns;
4033     NvmeZone *zone;
4034     NvmeZoneResetAIOCB *iocb;
4035     uint8_t *zd_ext;
4036     uint64_t slba = 0;
4037     uint32_t zone_idx = 0;
4038     uint16_t status;
4039     uint8_t action = cmd->zsa;
4040     bool all;
4041     enum NvmeZoneProcessingMask proc_mask = NVME_PROC_CURRENT_ZONE;
4042 
4043     all = cmd->zsflags & NVME_ZSFLAG_SELECT_ALL;
4044 
4045     req->status = NVME_SUCCESS;
4046 
4047     if (!all) {
4048         status = nvme_get_mgmt_zone_slba_idx(ns, &req->cmd, &slba, &zone_idx);
4049         if (status) {
4050             return status;
4051         }
4052     }
4053 
4054     zone = &ns->zone_array[zone_idx];
4055     if (slba != zone->d.zslba && action != NVME_ZONE_ACTION_ZRWA_FLUSH) {
4056         trace_pci_nvme_err_unaligned_zone_cmd(action, slba, zone->d.zslba);
4057         return NVME_INVALID_FIELD | NVME_DNR;
4058     }
4059 
4060     switch (action) {
4061 
4062     case NVME_ZONE_ACTION_OPEN:
4063         if (all) {
4064             proc_mask = NVME_PROC_CLOSED_ZONES;
4065         }
4066         trace_pci_nvme_open_zone(slba, zone_idx, all);
4067         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_open_zone, req);
4068         break;
4069 
4070     case NVME_ZONE_ACTION_CLOSE:
4071         if (all) {
4072             proc_mask = NVME_PROC_OPENED_ZONES;
4073         }
4074         trace_pci_nvme_close_zone(slba, zone_idx, all);
4075         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_close_zone, req);
4076         break;
4077 
4078     case NVME_ZONE_ACTION_FINISH:
4079         if (all) {
4080             proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES;
4081         }
4082         trace_pci_nvme_finish_zone(slba, zone_idx, all);
4083         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_finish_zone, req);
4084         break;
4085 
4086     case NVME_ZONE_ACTION_RESET:
4087         trace_pci_nvme_reset_zone(slba, zone_idx, all);
4088 
4089         iocb = blk_aio_get(&nvme_zone_reset_aiocb_info, ns->blkconf.blk,
4090                            nvme_misc_cb, req);
4091 
4092         iocb->req = req;
4093         iocb->ret = 0;
4094         iocb->all = all;
4095         iocb->idx = zone_idx;
4096         iocb->zone = NULL;
4097 
4098         req->aiocb = &iocb->common;
4099         nvme_zone_reset_cb(iocb, 0);
4100 
4101         return NVME_NO_COMPLETE;
4102 
4103     case NVME_ZONE_ACTION_OFFLINE:
4104         if (all) {
4105             proc_mask = NVME_PROC_READ_ONLY_ZONES;
4106         }
4107         trace_pci_nvme_offline_zone(slba, zone_idx, all);
4108         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_offline_zone, req);
4109         break;
4110 
4111     case NVME_ZONE_ACTION_SET_ZD_EXT:
4112         trace_pci_nvme_set_descriptor_extension(slba, zone_idx);
4113         if (all || !ns->params.zd_extension_size) {
4114             return NVME_INVALID_FIELD | NVME_DNR;
4115         }
4116         zd_ext = nvme_get_zd_extension(ns, zone_idx);
4117         status = nvme_h2c(n, zd_ext, ns->params.zd_extension_size, req);
4118         if (status) {
4119             trace_pci_nvme_err_zd_extension_map_error(zone_idx);
4120             return status;
4121         }
4122 
4123         status = nvme_set_zd_ext(ns, zone);
4124         if (status == NVME_SUCCESS) {
4125             trace_pci_nvme_zd_extension_set(zone_idx);
4126             return status;
4127         }
4128         break;
4129 
4130     case NVME_ZONE_ACTION_ZRWA_FLUSH:
4131         if (all) {
4132             return NVME_INVALID_FIELD | NVME_DNR;
4133         }
4134 
4135         return nvme_zone_mgmt_send_zrwa_flush(n, zone, slba, req);
4136 
4137     default:
4138         trace_pci_nvme_err_invalid_mgmt_action(action);
4139         status = NVME_INVALID_FIELD;
4140     }
4141 
4142     if (status == NVME_ZONE_INVAL_TRANSITION) {
4143         trace_pci_nvme_err_invalid_zone_state_transition(action, slba,
4144                                                          zone->d.za);
4145     }
4146     if (status) {
4147         status |= NVME_DNR;
4148     }
4149 
4150     return status;
4151 }
4152 
4153 static bool nvme_zone_matches_filter(uint32_t zafs, NvmeZone *zl)
4154 {
4155     NvmeZoneState zs = nvme_get_zone_state(zl);
4156 
4157     switch (zafs) {
4158     case NVME_ZONE_REPORT_ALL:
4159         return true;
4160     case NVME_ZONE_REPORT_EMPTY:
4161         return zs == NVME_ZONE_STATE_EMPTY;
4162     case NVME_ZONE_REPORT_IMPLICITLY_OPEN:
4163         return zs == NVME_ZONE_STATE_IMPLICITLY_OPEN;
4164     case NVME_ZONE_REPORT_EXPLICITLY_OPEN:
4165         return zs == NVME_ZONE_STATE_EXPLICITLY_OPEN;
4166     case NVME_ZONE_REPORT_CLOSED:
4167         return zs == NVME_ZONE_STATE_CLOSED;
4168     case NVME_ZONE_REPORT_FULL:
4169         return zs == NVME_ZONE_STATE_FULL;
4170     case NVME_ZONE_REPORT_READ_ONLY:
4171         return zs == NVME_ZONE_STATE_READ_ONLY;
4172     case NVME_ZONE_REPORT_OFFLINE:
4173         return zs == NVME_ZONE_STATE_OFFLINE;
4174     default:
4175         return false;
4176     }
4177 }
4178 
4179 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4180 {
4181     NvmeCmd *cmd = (NvmeCmd *)&req->cmd;
4182     NvmeNamespace *ns = req->ns;
4183     /* cdw12 is zero-based number of dwords to return. Convert to bytes */
4184     uint32_t data_size = (le32_to_cpu(cmd->cdw12) + 1) << 2;
4185     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
4186     uint32_t zone_idx, zra, zrasf, partial;
4187     uint64_t max_zones, nr_zones = 0;
4188     uint16_t status;
4189     uint64_t slba;
4190     NvmeZoneDescr *z;
4191     NvmeZone *zone;
4192     NvmeZoneReportHeader *header;
4193     void *buf, *buf_p;
4194     size_t zone_entry_sz;
4195     int i;
4196 
4197     req->status = NVME_SUCCESS;
4198 
4199     status = nvme_get_mgmt_zone_slba_idx(ns, cmd, &slba, &zone_idx);
4200     if (status) {
4201         return status;
4202     }
4203 
4204     zra = dw13 & 0xff;
4205     if (zra != NVME_ZONE_REPORT && zra != NVME_ZONE_REPORT_EXTENDED) {
4206         return NVME_INVALID_FIELD | NVME_DNR;
4207     }
4208     if (zra == NVME_ZONE_REPORT_EXTENDED && !ns->params.zd_extension_size) {
4209         return NVME_INVALID_FIELD | NVME_DNR;
4210     }
4211 
4212     zrasf = (dw13 >> 8) & 0xff;
4213     if (zrasf > NVME_ZONE_REPORT_OFFLINE) {
4214         return NVME_INVALID_FIELD | NVME_DNR;
4215     }
4216 
4217     if (data_size < sizeof(NvmeZoneReportHeader)) {
4218         return NVME_INVALID_FIELD | NVME_DNR;
4219     }
4220 
4221     status = nvme_check_mdts(n, data_size);
4222     if (status) {
4223         return status;
4224     }
4225 
4226     partial = (dw13 >> 16) & 0x01;
4227 
4228     zone_entry_sz = sizeof(NvmeZoneDescr);
4229     if (zra == NVME_ZONE_REPORT_EXTENDED) {
4230         zone_entry_sz += ns->params.zd_extension_size;
4231     }
4232 
4233     max_zones = (data_size - sizeof(NvmeZoneReportHeader)) / zone_entry_sz;
4234     buf = g_malloc0(data_size);
4235 
4236     zone = &ns->zone_array[zone_idx];
4237     for (i = zone_idx; i < ns->num_zones; i++) {
4238         if (partial && nr_zones >= max_zones) {
4239             break;
4240         }
4241         if (nvme_zone_matches_filter(zrasf, zone++)) {
4242             nr_zones++;
4243         }
4244     }
4245     header = buf;
4246     header->nr_zones = cpu_to_le64(nr_zones);
4247 
4248     buf_p = buf + sizeof(NvmeZoneReportHeader);
4249     for (; zone_idx < ns->num_zones && max_zones > 0; zone_idx++) {
4250         zone = &ns->zone_array[zone_idx];
4251         if (nvme_zone_matches_filter(zrasf, zone)) {
4252             z = buf_p;
4253             buf_p += sizeof(NvmeZoneDescr);
4254 
4255             z->zt = zone->d.zt;
4256             z->zs = zone->d.zs;
4257             z->zcap = cpu_to_le64(zone->d.zcap);
4258             z->zslba = cpu_to_le64(zone->d.zslba);
4259             z->za = zone->d.za;
4260 
4261             if (nvme_wp_is_valid(zone)) {
4262                 z->wp = cpu_to_le64(zone->d.wp);
4263             } else {
4264                 z->wp = cpu_to_le64(~0ULL);
4265             }
4266 
4267             if (zra == NVME_ZONE_REPORT_EXTENDED) {
4268                 if (zone->d.za & NVME_ZA_ZD_EXT_VALID) {
4269                     memcpy(buf_p, nvme_get_zd_extension(ns, zone_idx),
4270                            ns->params.zd_extension_size);
4271                 }
4272                 buf_p += ns->params.zd_extension_size;
4273             }
4274 
4275             max_zones--;
4276         }
4277     }
4278 
4279     status = nvme_c2h(n, (uint8_t *)buf, data_size, req);
4280 
4281     g_free(buf);
4282 
4283     return status;
4284 }
4285 
4286 static uint16_t nvme_io_mgmt_recv_ruhs(NvmeCtrl *n, NvmeRequest *req,
4287                                        size_t len)
4288 {
4289     NvmeNamespace *ns = req->ns;
4290     NvmeEnduranceGroup *endgrp;
4291     NvmeRuhStatus *hdr;
4292     NvmeRuhStatusDescr *ruhsd;
4293     unsigned int nruhsd;
4294     uint16_t rg, ph, *ruhid;
4295     size_t trans_len;
4296     g_autofree uint8_t *buf = NULL;
4297 
4298     if (!n->subsys) {
4299         return NVME_INVALID_FIELD | NVME_DNR;
4300     }
4301 
4302     if (ns->params.nsid == 0 || ns->params.nsid == 0xffffffff) {
4303         return NVME_INVALID_NSID | NVME_DNR;
4304     }
4305 
4306     if (!n->subsys->endgrp.fdp.enabled) {
4307         return NVME_FDP_DISABLED | NVME_DNR;
4308     }
4309 
4310     endgrp = ns->endgrp;
4311 
4312     nruhsd = ns->fdp.nphs * endgrp->fdp.nrg;
4313     trans_len = sizeof(NvmeRuhStatus) + nruhsd * sizeof(NvmeRuhStatusDescr);
4314     buf = g_malloc(trans_len);
4315 
4316     trans_len = MIN(trans_len, len);
4317 
4318     hdr = (NvmeRuhStatus *)buf;
4319     ruhsd = (NvmeRuhStatusDescr *)(buf + sizeof(NvmeRuhStatus));
4320 
4321     hdr->nruhsd = cpu_to_le16(nruhsd);
4322 
4323     ruhid = ns->fdp.phs;
4324 
4325     for (ph = 0; ph < ns->fdp.nphs; ph++, ruhid++) {
4326         NvmeRuHandle *ruh = &endgrp->fdp.ruhs[*ruhid];
4327 
4328         for (rg = 0; rg < endgrp->fdp.nrg; rg++, ruhsd++) {
4329             uint16_t pid = nvme_make_pid(ns, rg, ph);
4330 
4331             ruhsd->pid = cpu_to_le16(pid);
4332             ruhsd->ruhid = *ruhid;
4333             ruhsd->earutr = 0;
4334             ruhsd->ruamw = cpu_to_le64(ruh->rus[rg].ruamw);
4335         }
4336     }
4337 
4338     return nvme_c2h(n, buf, trans_len, req);
4339 }
4340 
4341 static uint16_t nvme_io_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4342 {
4343     NvmeCmd *cmd = &req->cmd;
4344     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4345     uint32_t numd = le32_to_cpu(cmd->cdw11);
4346     uint8_t mo = (cdw10 & 0xff);
4347     size_t len = (numd + 1) << 2;
4348 
4349     switch (mo) {
4350     case NVME_IOMR_MO_NOP:
4351         return 0;
4352     case NVME_IOMR_MO_RUH_STATUS:
4353         return nvme_io_mgmt_recv_ruhs(n, req, len);
4354     default:
4355         return NVME_INVALID_FIELD | NVME_DNR;
4356     };
4357 }
4358 
4359 static uint16_t nvme_io_mgmt_send_ruh_update(NvmeCtrl *n, NvmeRequest *req)
4360 {
4361     NvmeCmd *cmd = &req->cmd;
4362     NvmeNamespace *ns = req->ns;
4363     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4364     uint16_t ret = NVME_SUCCESS;
4365     uint32_t npid = (cdw10 >> 1) + 1;
4366     unsigned int i = 0;
4367     g_autofree uint16_t *pids = NULL;
4368     uint32_t maxnpid = n->subsys->endgrp.fdp.nrg * n->subsys->endgrp.fdp.nruh;
4369 
4370     if (unlikely(npid >= MIN(NVME_FDP_MAXPIDS, maxnpid))) {
4371         return NVME_INVALID_FIELD | NVME_DNR;
4372     }
4373 
4374     pids = g_new(uint16_t, npid);
4375 
4376     ret = nvme_h2c(n, pids, npid * sizeof(uint16_t), req);
4377     if (ret) {
4378         return ret;
4379     }
4380 
4381     for (; i < npid; i++) {
4382         if (!nvme_update_ruh(n, ns, pids[i])) {
4383             return NVME_INVALID_FIELD | NVME_DNR;
4384         }
4385     }
4386 
4387     return ret;
4388 }
4389 
4390 static uint16_t nvme_io_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4391 {
4392     NvmeCmd *cmd = &req->cmd;
4393     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4394     uint8_t mo = (cdw10 & 0xff);
4395 
4396     switch (mo) {
4397     case NVME_IOMS_MO_NOP:
4398         return 0;
4399     case NVME_IOMS_MO_RUH_UPDATE:
4400         return nvme_io_mgmt_send_ruh_update(n, req);
4401     default:
4402         return NVME_INVALID_FIELD | NVME_DNR;
4403     };
4404 }
4405 
4406 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
4407 {
4408     NvmeNamespace *ns;
4409     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4410 
4411     trace_pci_nvme_io_cmd(nvme_cid(req), nsid, nvme_sqid(req),
4412                           req->cmd.opcode, nvme_io_opc_str(req->cmd.opcode));
4413 
4414     if (!nvme_nsid_valid(n, nsid)) {
4415         return NVME_INVALID_NSID | NVME_DNR;
4416     }
4417 
4418     /*
4419      * In the base NVM command set, Flush may apply to all namespaces
4420      * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4421      * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4422      *
4423      * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4424      * opcode with a specific command since we cannot determine a unique I/O
4425      * command set. Opcode 0h could have any other meaning than something
4426      * equivalent to flushing and say it DOES have completely different
4427      * semantics in some other command set - does an NSID of FFFFFFFFh then
4428      * mean "for all namespaces, apply whatever command set specific command
4429      * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4430      * whatever command that uses the 0h opcode if, and only if, it allows NSID
4431      * to be FFFFFFFFh"?
4432      *
4433      * Anyway (and luckily), for now, we do not care about this since the
4434      * device only supports namespace types that includes the NVM Flush command
4435      * (NVM and Zoned), so always do an NVM Flush.
4436      */
4437     if (req->cmd.opcode == NVME_CMD_FLUSH) {
4438         return nvme_flush(n, req);
4439     }
4440 
4441     ns = nvme_ns(n, nsid);
4442     if (unlikely(!ns)) {
4443         return NVME_INVALID_FIELD | NVME_DNR;
4444     }
4445 
4446     if (!(ns->iocs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
4447         trace_pci_nvme_err_invalid_opc(req->cmd.opcode);
4448         return NVME_INVALID_OPCODE | NVME_DNR;
4449     }
4450 
4451     if (ns->status) {
4452         return ns->status;
4453     }
4454 
4455     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
4456         return NVME_INVALID_FIELD;
4457     }
4458 
4459     req->ns = ns;
4460 
4461     switch (req->cmd.opcode) {
4462     case NVME_CMD_WRITE_ZEROES:
4463         return nvme_write_zeroes(n, req);
4464     case NVME_CMD_ZONE_APPEND:
4465         return nvme_zone_append(n, req);
4466     case NVME_CMD_WRITE:
4467         return nvme_write(n, req);
4468     case NVME_CMD_READ:
4469         return nvme_read(n, req);
4470     case NVME_CMD_COMPARE:
4471         return nvme_compare(n, req);
4472     case NVME_CMD_DSM:
4473         return nvme_dsm(n, req);
4474     case NVME_CMD_VERIFY:
4475         return nvme_verify(n, req);
4476     case NVME_CMD_COPY:
4477         return nvme_copy(n, req);
4478     case NVME_CMD_ZONE_MGMT_SEND:
4479         return nvme_zone_mgmt_send(n, req);
4480     case NVME_CMD_ZONE_MGMT_RECV:
4481         return nvme_zone_mgmt_recv(n, req);
4482     case NVME_CMD_IO_MGMT_RECV:
4483         return nvme_io_mgmt_recv(n, req);
4484     case NVME_CMD_IO_MGMT_SEND:
4485         return nvme_io_mgmt_send(n, req);
4486     default:
4487         assert(false);
4488     }
4489 
4490     return NVME_INVALID_OPCODE | NVME_DNR;
4491 }
4492 
4493 static void nvme_cq_notifier(EventNotifier *e)
4494 {
4495     NvmeCQueue *cq = container_of(e, NvmeCQueue, notifier);
4496     NvmeCtrl *n = cq->ctrl;
4497 
4498     if (!event_notifier_test_and_clear(e)) {
4499         return;
4500     }
4501 
4502     nvme_update_cq_head(cq);
4503 
4504     if (cq->tail == cq->head) {
4505         if (cq->irq_enabled) {
4506             n->cq_pending--;
4507         }
4508 
4509         nvme_irq_deassert(n, cq);
4510     }
4511 
4512     qemu_bh_schedule(cq->bh);
4513 }
4514 
4515 static int nvme_init_cq_ioeventfd(NvmeCQueue *cq)
4516 {
4517     NvmeCtrl *n = cq->ctrl;
4518     uint16_t offset = (cq->cqid << 3) + (1 << 2);
4519     int ret;
4520 
4521     ret = event_notifier_init(&cq->notifier, 0);
4522     if (ret < 0) {
4523         return ret;
4524     }
4525 
4526     event_notifier_set_handler(&cq->notifier, nvme_cq_notifier);
4527     memory_region_add_eventfd(&n->iomem,
4528                               0x1000 + offset, 4, false, 0, &cq->notifier);
4529 
4530     return 0;
4531 }
4532 
4533 static void nvme_sq_notifier(EventNotifier *e)
4534 {
4535     NvmeSQueue *sq = container_of(e, NvmeSQueue, notifier);
4536 
4537     if (!event_notifier_test_and_clear(e)) {
4538         return;
4539     }
4540 
4541     nvme_process_sq(sq);
4542 }
4543 
4544 static int nvme_init_sq_ioeventfd(NvmeSQueue *sq)
4545 {
4546     NvmeCtrl *n = sq->ctrl;
4547     uint16_t offset = sq->sqid << 3;
4548     int ret;
4549 
4550     ret = event_notifier_init(&sq->notifier, 0);
4551     if (ret < 0) {
4552         return ret;
4553     }
4554 
4555     event_notifier_set_handler(&sq->notifier, nvme_sq_notifier);
4556     memory_region_add_eventfd(&n->iomem,
4557                               0x1000 + offset, 4, false, 0, &sq->notifier);
4558 
4559     return 0;
4560 }
4561 
4562 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
4563 {
4564     uint16_t offset = sq->sqid << 3;
4565 
4566     n->sq[sq->sqid] = NULL;
4567     qemu_bh_delete(sq->bh);
4568     if (sq->ioeventfd_enabled) {
4569         memory_region_del_eventfd(&n->iomem,
4570                                   0x1000 + offset, 4, false, 0, &sq->notifier);
4571         event_notifier_set_handler(&sq->notifier, NULL);
4572         event_notifier_cleanup(&sq->notifier);
4573     }
4574     g_free(sq->io_req);
4575     if (sq->sqid) {
4576         g_free(sq);
4577     }
4578 }
4579 
4580 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
4581 {
4582     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4583     NvmeRequest *r, *next;
4584     NvmeSQueue *sq;
4585     NvmeCQueue *cq;
4586     uint16_t qid = le16_to_cpu(c->qid);
4587 
4588     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
4589         trace_pci_nvme_err_invalid_del_sq(qid);
4590         return NVME_INVALID_QID | NVME_DNR;
4591     }
4592 
4593     trace_pci_nvme_del_sq(qid);
4594 
4595     sq = n->sq[qid];
4596     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
4597         r = QTAILQ_FIRST(&sq->out_req_list);
4598         assert(r->aiocb);
4599         blk_aio_cancel(r->aiocb);
4600     }
4601 
4602     assert(QTAILQ_EMPTY(&sq->out_req_list));
4603 
4604     if (!nvme_check_cqid(n, sq->cqid)) {
4605         cq = n->cq[sq->cqid];
4606         QTAILQ_REMOVE(&cq->sq_list, sq, entry);
4607 
4608         nvme_post_cqes(cq);
4609         QTAILQ_FOREACH_SAFE(r, &cq->req_list, entry, next) {
4610             if (r->sq == sq) {
4611                 QTAILQ_REMOVE(&cq->req_list, r, entry);
4612                 QTAILQ_INSERT_TAIL(&sq->req_list, r, entry);
4613             }
4614         }
4615     }
4616 
4617     nvme_free_sq(sq, n);
4618     return NVME_SUCCESS;
4619 }
4620 
4621 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
4622                          uint16_t sqid, uint16_t cqid, uint16_t size)
4623 {
4624     int i;
4625     NvmeCQueue *cq;
4626 
4627     sq->ctrl = n;
4628     sq->dma_addr = dma_addr;
4629     sq->sqid = sqid;
4630     sq->size = size;
4631     sq->cqid = cqid;
4632     sq->head = sq->tail = 0;
4633     sq->io_req = g_new0(NvmeRequest, sq->size);
4634 
4635     QTAILQ_INIT(&sq->req_list);
4636     QTAILQ_INIT(&sq->out_req_list);
4637     for (i = 0; i < sq->size; i++) {
4638         sq->io_req[i].sq = sq;
4639         QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
4640     }
4641 
4642     sq->bh = qemu_bh_new_guarded(nvme_process_sq, sq,
4643                                  &DEVICE(sq->ctrl)->mem_reentrancy_guard);
4644 
4645     if (n->dbbuf_enabled) {
4646         sq->db_addr = n->dbbuf_dbs + (sqid << 3);
4647         sq->ei_addr = n->dbbuf_eis + (sqid << 3);
4648 
4649         if (n->params.ioeventfd && sq->sqid != 0) {
4650             if (!nvme_init_sq_ioeventfd(sq)) {
4651                 sq->ioeventfd_enabled = true;
4652             }
4653         }
4654     }
4655 
4656     assert(n->cq[cqid]);
4657     cq = n->cq[cqid];
4658     QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
4659     n->sq[sqid] = sq;
4660 }
4661 
4662 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeRequest *req)
4663 {
4664     NvmeSQueue *sq;
4665     NvmeCreateSq *c = (NvmeCreateSq *)&req->cmd;
4666 
4667     uint16_t cqid = le16_to_cpu(c->cqid);
4668     uint16_t sqid = le16_to_cpu(c->sqid);
4669     uint16_t qsize = le16_to_cpu(c->qsize);
4670     uint16_t qflags = le16_to_cpu(c->sq_flags);
4671     uint64_t prp1 = le64_to_cpu(c->prp1);
4672 
4673     trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
4674 
4675     if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
4676         trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
4677         return NVME_INVALID_CQID | NVME_DNR;
4678     }
4679     if (unlikely(!sqid || sqid > n->conf_ioqpairs || n->sq[sqid] != NULL)) {
4680         trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
4681         return NVME_INVALID_QID | NVME_DNR;
4682     }
4683     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
4684         trace_pci_nvme_err_invalid_create_sq_size(qsize);
4685         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4686     }
4687     if (unlikely(prp1 & (n->page_size - 1))) {
4688         trace_pci_nvme_err_invalid_create_sq_addr(prp1);
4689         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4690     }
4691     if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
4692         trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
4693         return NVME_INVALID_FIELD | NVME_DNR;
4694     }
4695     sq = g_malloc0(sizeof(*sq));
4696     nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
4697     return NVME_SUCCESS;
4698 }
4699 
4700 struct nvme_stats {
4701     uint64_t units_read;
4702     uint64_t units_written;
4703     uint64_t read_commands;
4704     uint64_t write_commands;
4705 };
4706 
4707 static void nvme_set_blk_stats(NvmeNamespace *ns, struct nvme_stats *stats)
4708 {
4709     BlockAcctStats *s = blk_get_stats(ns->blkconf.blk);
4710 
4711     stats->units_read += s->nr_bytes[BLOCK_ACCT_READ];
4712     stats->units_written += s->nr_bytes[BLOCK_ACCT_WRITE];
4713     stats->read_commands += s->nr_ops[BLOCK_ACCT_READ];
4714     stats->write_commands += s->nr_ops[BLOCK_ACCT_WRITE];
4715 }
4716 
4717 static uint16_t nvme_smart_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4718                                 uint64_t off, NvmeRequest *req)
4719 {
4720     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4721     struct nvme_stats stats = { 0 };
4722     NvmeSmartLog smart = { 0 };
4723     uint32_t trans_len;
4724     NvmeNamespace *ns;
4725     time_t current_ms;
4726     uint64_t u_read, u_written;
4727 
4728     if (off >= sizeof(smart)) {
4729         return NVME_INVALID_FIELD | NVME_DNR;
4730     }
4731 
4732     if (nsid != 0xffffffff) {
4733         ns = nvme_ns(n, nsid);
4734         if (!ns) {
4735             return NVME_INVALID_NSID | NVME_DNR;
4736         }
4737         nvme_set_blk_stats(ns, &stats);
4738     } else {
4739         int i;
4740 
4741         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4742             ns = nvme_ns(n, i);
4743             if (!ns) {
4744                 continue;
4745             }
4746             nvme_set_blk_stats(ns, &stats);
4747         }
4748     }
4749 
4750     trans_len = MIN(sizeof(smart) - off, buf_len);
4751     smart.critical_warning = n->smart_critical_warning;
4752 
4753     u_read = DIV_ROUND_UP(stats.units_read >> BDRV_SECTOR_BITS, 1000);
4754     u_written = DIV_ROUND_UP(stats.units_written >> BDRV_SECTOR_BITS, 1000);
4755 
4756     smart.data_units_read[0] = cpu_to_le64(u_read);
4757     smart.data_units_written[0] = cpu_to_le64(u_written);
4758     smart.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4759     smart.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4760 
4761     smart.temperature = cpu_to_le16(n->temperature);
4762 
4763     if ((n->temperature >= n->features.temp_thresh_hi) ||
4764         (n->temperature <= n->features.temp_thresh_low)) {
4765         smart.critical_warning |= NVME_SMART_TEMPERATURE;
4766     }
4767 
4768     current_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4769     smart.power_on_hours[0] =
4770         cpu_to_le64((((current_ms - n->starttime_ms) / 1000) / 60) / 60);
4771 
4772     if (!rae) {
4773         nvme_clear_events(n, NVME_AER_TYPE_SMART);
4774     }
4775 
4776     return nvme_c2h(n, (uint8_t *) &smart + off, trans_len, req);
4777 }
4778 
4779 static uint16_t nvme_endgrp_info(NvmeCtrl *n,  uint8_t rae, uint32_t buf_len,
4780                                  uint64_t off, NvmeRequest *req)
4781 {
4782     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
4783     uint16_t endgrpid = (dw11 >> 16) & 0xffff;
4784     struct nvme_stats stats = {};
4785     NvmeEndGrpLog info = {};
4786     int i;
4787 
4788     if (!n->subsys || endgrpid != 0x1) {
4789         return NVME_INVALID_FIELD | NVME_DNR;
4790     }
4791 
4792     if (off >= sizeof(info)) {
4793         return NVME_INVALID_FIELD | NVME_DNR;
4794     }
4795 
4796     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4797         NvmeNamespace *ns = nvme_subsys_ns(n->subsys, i);
4798         if (!ns) {
4799             continue;
4800         }
4801 
4802         nvme_set_blk_stats(ns, &stats);
4803     }
4804 
4805     info.data_units_read[0] =
4806         cpu_to_le64(DIV_ROUND_UP(stats.units_read / 1000000000, 1000000000));
4807     info.data_units_written[0] =
4808         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4809     info.media_units_written[0] =
4810         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4811 
4812     info.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4813     info.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4814 
4815     buf_len = MIN(sizeof(info) - off, buf_len);
4816 
4817     return nvme_c2h(n, (uint8_t *)&info + off, buf_len, req);
4818 }
4819 
4820 
4821 static uint16_t nvme_fw_log_info(NvmeCtrl *n, uint32_t buf_len, uint64_t off,
4822                                  NvmeRequest *req)
4823 {
4824     uint32_t trans_len;
4825     NvmeFwSlotInfoLog fw_log = {
4826         .afi = 0x1,
4827     };
4828 
4829     if (off >= sizeof(fw_log)) {
4830         return NVME_INVALID_FIELD | NVME_DNR;
4831     }
4832 
4833     strpadcpy((char *)&fw_log.frs1, sizeof(fw_log.frs1), "1.0", ' ');
4834     trans_len = MIN(sizeof(fw_log) - off, buf_len);
4835 
4836     return nvme_c2h(n, (uint8_t *) &fw_log + off, trans_len, req);
4837 }
4838 
4839 static uint16_t nvme_error_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4840                                 uint64_t off, NvmeRequest *req)
4841 {
4842     uint32_t trans_len;
4843     NvmeErrorLog errlog;
4844 
4845     if (off >= sizeof(errlog)) {
4846         return NVME_INVALID_FIELD | NVME_DNR;
4847     }
4848 
4849     if (!rae) {
4850         nvme_clear_events(n, NVME_AER_TYPE_ERROR);
4851     }
4852 
4853     memset(&errlog, 0x0, sizeof(errlog));
4854     trans_len = MIN(sizeof(errlog) - off, buf_len);
4855 
4856     return nvme_c2h(n, (uint8_t *)&errlog, trans_len, req);
4857 }
4858 
4859 static uint16_t nvme_changed_nslist(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4860                                     uint64_t off, NvmeRequest *req)
4861 {
4862     uint32_t nslist[1024];
4863     uint32_t trans_len;
4864     int i = 0;
4865     uint32_t nsid;
4866 
4867     if (off >= sizeof(nslist)) {
4868         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(nslist));
4869         return NVME_INVALID_FIELD | NVME_DNR;
4870     }
4871 
4872     memset(nslist, 0x0, sizeof(nslist));
4873     trans_len = MIN(sizeof(nslist) - off, buf_len);
4874 
4875     while ((nsid = find_first_bit(n->changed_nsids, NVME_CHANGED_NSID_SIZE)) !=
4876             NVME_CHANGED_NSID_SIZE) {
4877         /*
4878          * If more than 1024 namespaces, the first entry in the log page should
4879          * be set to FFFFFFFFh and the others to 0 as spec.
4880          */
4881         if (i == ARRAY_SIZE(nslist)) {
4882             memset(nslist, 0x0, sizeof(nslist));
4883             nslist[0] = 0xffffffff;
4884             break;
4885         }
4886 
4887         nslist[i++] = nsid;
4888         clear_bit(nsid, n->changed_nsids);
4889     }
4890 
4891     /*
4892      * Remove all the remaining list entries in case returns directly due to
4893      * more than 1024 namespaces.
4894      */
4895     if (nslist[0] == 0xffffffff) {
4896         bitmap_zero(n->changed_nsids, NVME_CHANGED_NSID_SIZE);
4897     }
4898 
4899     if (!rae) {
4900         nvme_clear_events(n, NVME_AER_TYPE_NOTICE);
4901     }
4902 
4903     return nvme_c2h(n, ((uint8_t *)nslist) + off, trans_len, req);
4904 }
4905 
4906 static uint16_t nvme_cmd_effects(NvmeCtrl *n, uint8_t csi, uint32_t buf_len,
4907                                  uint64_t off, NvmeRequest *req)
4908 {
4909     NvmeEffectsLog log = {};
4910     const uint32_t *src_iocs = NULL;
4911     uint32_t trans_len;
4912 
4913     if (off >= sizeof(log)) {
4914         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(log));
4915         return NVME_INVALID_FIELD | NVME_DNR;
4916     }
4917 
4918     switch (NVME_CC_CSS(ldl_le_p(&n->bar.cc))) {
4919     case NVME_CC_CSS_NVM:
4920         src_iocs = nvme_cse_iocs_nvm;
4921         /* fall through */
4922     case NVME_CC_CSS_ADMIN_ONLY:
4923         break;
4924     case NVME_CC_CSS_CSI:
4925         switch (csi) {
4926         case NVME_CSI_NVM:
4927             src_iocs = nvme_cse_iocs_nvm;
4928             break;
4929         case NVME_CSI_ZONED:
4930             src_iocs = nvme_cse_iocs_zoned;
4931             break;
4932         }
4933     }
4934 
4935     memcpy(log.acs, nvme_cse_acs, sizeof(nvme_cse_acs));
4936 
4937     if (src_iocs) {
4938         memcpy(log.iocs, src_iocs, sizeof(log.iocs));
4939     }
4940 
4941     trans_len = MIN(sizeof(log) - off, buf_len);
4942 
4943     return nvme_c2h(n, ((uint8_t *)&log) + off, trans_len, req);
4944 }
4945 
4946 static size_t sizeof_fdp_conf_descr(size_t nruh, size_t vss)
4947 {
4948     size_t entry_siz = sizeof(NvmeFdpDescrHdr) + nruh * sizeof(NvmeRuhDescr)
4949                        + vss;
4950     return ROUND_UP(entry_siz, 8);
4951 }
4952 
4953 static uint16_t nvme_fdp_confs(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
4954                                uint64_t off, NvmeRequest *req)
4955 {
4956     uint32_t log_size, trans_len;
4957     g_autofree uint8_t *buf = NULL;
4958     NvmeFdpDescrHdr *hdr;
4959     NvmeRuhDescr *ruhd;
4960     NvmeEnduranceGroup *endgrp;
4961     NvmeFdpConfsHdr *log;
4962     size_t nruh, fdp_descr_size;
4963     int i;
4964 
4965     if (endgrpid != 1 || !n->subsys) {
4966         return NVME_INVALID_FIELD | NVME_DNR;
4967     }
4968 
4969     endgrp = &n->subsys->endgrp;
4970 
4971     if (endgrp->fdp.enabled) {
4972         nruh = endgrp->fdp.nruh;
4973     } else {
4974         nruh = 1;
4975     }
4976 
4977     fdp_descr_size = sizeof_fdp_conf_descr(nruh, FDPVSS);
4978     log_size = sizeof(NvmeFdpConfsHdr) + fdp_descr_size;
4979 
4980     if (off >= log_size) {
4981         return NVME_INVALID_FIELD | NVME_DNR;
4982     }
4983 
4984     trans_len = MIN(log_size - off, buf_len);
4985 
4986     buf = g_malloc0(log_size);
4987     log = (NvmeFdpConfsHdr *)buf;
4988     hdr = (NvmeFdpDescrHdr *)(log + 1);
4989     ruhd = (NvmeRuhDescr *)(buf + sizeof(*log) + sizeof(*hdr));
4990 
4991     log->num_confs = cpu_to_le16(0);
4992     log->size = cpu_to_le32(log_size);
4993 
4994     hdr->descr_size = cpu_to_le16(fdp_descr_size);
4995     if (endgrp->fdp.enabled) {
4996         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, VALID, 1);
4997         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, RGIF, endgrp->fdp.rgif);
4998         hdr->nrg = cpu_to_le16(endgrp->fdp.nrg);
4999         hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5000         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5001         hdr->nnss = cpu_to_le32(NVME_MAX_NAMESPACES);
5002         hdr->runs = cpu_to_le64(endgrp->fdp.runs);
5003 
5004         for (i = 0; i < nruh; i++) {
5005             ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5006             ruhd++;
5007         }
5008     } else {
5009         /* 1 bit for RUH in PIF -> 2 RUHs max. */
5010         hdr->nrg = cpu_to_le16(1);
5011         hdr->nruh = cpu_to_le16(1);
5012         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5013         hdr->nnss = cpu_to_le32(1);
5014         hdr->runs = cpu_to_le64(96 * MiB);
5015 
5016         ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5017     }
5018 
5019     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5020 }
5021 
5022 static uint16_t nvme_fdp_ruh_usage(NvmeCtrl *n, uint32_t endgrpid,
5023                                    uint32_t dw10, uint32_t dw12,
5024                                    uint32_t buf_len, uint64_t off,
5025                                    NvmeRequest *req)
5026 {
5027     NvmeRuHandle *ruh;
5028     NvmeRuhuLog *hdr;
5029     NvmeRuhuDescr *ruhud;
5030     NvmeEnduranceGroup *endgrp;
5031     g_autofree uint8_t *buf = NULL;
5032     uint32_t log_size, trans_len;
5033     uint16_t i;
5034 
5035     if (endgrpid != 1 || !n->subsys) {
5036         return NVME_INVALID_FIELD | NVME_DNR;
5037     }
5038 
5039     endgrp = &n->subsys->endgrp;
5040 
5041     if (!endgrp->fdp.enabled) {
5042         return NVME_FDP_DISABLED | NVME_DNR;
5043     }
5044 
5045     log_size = sizeof(NvmeRuhuLog) + endgrp->fdp.nruh * sizeof(NvmeRuhuDescr);
5046 
5047     if (off >= log_size) {
5048         return NVME_INVALID_FIELD | NVME_DNR;
5049     }
5050 
5051     trans_len = MIN(log_size - off, buf_len);
5052 
5053     buf = g_malloc0(log_size);
5054     hdr = (NvmeRuhuLog *)buf;
5055     ruhud = (NvmeRuhuDescr *)(hdr + 1);
5056 
5057     ruh = endgrp->fdp.ruhs;
5058     hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5059 
5060     for (i = 0; i < endgrp->fdp.nruh; i++, ruhud++, ruh++) {
5061         ruhud->ruha = ruh->ruha;
5062     }
5063 
5064     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5065 }
5066 
5067 static uint16_t nvme_fdp_stats(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
5068                                uint64_t off, NvmeRequest *req)
5069 {
5070     NvmeEnduranceGroup *endgrp;
5071     NvmeFdpStatsLog log = {};
5072     uint32_t trans_len;
5073 
5074     if (off >= sizeof(NvmeFdpStatsLog)) {
5075         return NVME_INVALID_FIELD | NVME_DNR;
5076     }
5077 
5078     if (endgrpid != 1 || !n->subsys) {
5079         return NVME_INVALID_FIELD | NVME_DNR;
5080     }
5081 
5082     if (!n->subsys->endgrp.fdp.enabled) {
5083         return NVME_FDP_DISABLED | NVME_DNR;
5084     }
5085 
5086     endgrp = &n->subsys->endgrp;
5087 
5088     trans_len = MIN(sizeof(log) - off, buf_len);
5089 
5090     /* spec value is 128 bit, we only use 64 bit */
5091     log.hbmw[0] = cpu_to_le64(endgrp->fdp.hbmw);
5092     log.mbmw[0] = cpu_to_le64(endgrp->fdp.mbmw);
5093     log.mbe[0] = cpu_to_le64(endgrp->fdp.mbe);
5094 
5095     return nvme_c2h(n, (uint8_t *)&log + off, trans_len, req);
5096 }
5097 
5098 static uint16_t nvme_fdp_events(NvmeCtrl *n, uint32_t endgrpid,
5099                                 uint32_t buf_len, uint64_t off,
5100                                 NvmeRequest *req)
5101 {
5102     NvmeEnduranceGroup *endgrp;
5103     NvmeCmd *cmd = &req->cmd;
5104     bool host_events = (cmd->cdw10 >> 8) & 0x1;
5105     uint32_t log_size, trans_len;
5106     NvmeFdpEventBuffer *ebuf;
5107     g_autofree NvmeFdpEventsLog *elog = NULL;
5108     NvmeFdpEvent *event;
5109 
5110     if (endgrpid != 1 || !n->subsys) {
5111         return NVME_INVALID_FIELD | NVME_DNR;
5112     }
5113 
5114     endgrp = &n->subsys->endgrp;
5115 
5116     if (!endgrp->fdp.enabled) {
5117         return NVME_FDP_DISABLED | NVME_DNR;
5118     }
5119 
5120     if (host_events) {
5121         ebuf = &endgrp->fdp.host_events;
5122     } else {
5123         ebuf = &endgrp->fdp.ctrl_events;
5124     }
5125 
5126     log_size = sizeof(NvmeFdpEventsLog) + ebuf->nelems * sizeof(NvmeFdpEvent);
5127     trans_len = MIN(log_size - off, buf_len);
5128     elog = g_malloc0(log_size);
5129     elog->num_events = cpu_to_le32(ebuf->nelems);
5130     event = (NvmeFdpEvent *)(elog + 1);
5131 
5132     if (ebuf->nelems && ebuf->start == ebuf->next) {
5133         unsigned int nelems = (NVME_FDP_MAX_EVENTS - ebuf->start);
5134         /* wrap over, copy [start;NVME_FDP_MAX_EVENTS[ and [0; next[ */
5135         memcpy(event, &ebuf->events[ebuf->start],
5136                sizeof(NvmeFdpEvent) * nelems);
5137         memcpy(event + nelems, ebuf->events,
5138                sizeof(NvmeFdpEvent) * ebuf->next);
5139     } else if (ebuf->start < ebuf->next) {
5140         memcpy(event, &ebuf->events[ebuf->start],
5141                sizeof(NvmeFdpEvent) * (ebuf->next - ebuf->start));
5142     }
5143 
5144     return nvme_c2h(n, (uint8_t *)elog + off, trans_len, req);
5145 }
5146 
5147 static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
5148 {
5149     NvmeCmd *cmd = &req->cmd;
5150 
5151     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5152     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5153     uint32_t dw12 = le32_to_cpu(cmd->cdw12);
5154     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
5155     uint8_t  lid = dw10 & 0xff;
5156     uint8_t  lsp = (dw10 >> 8) & 0xf;
5157     uint8_t  rae = (dw10 >> 15) & 0x1;
5158     uint8_t  csi = le32_to_cpu(cmd->cdw14) >> 24;
5159     uint32_t numdl, numdu, lspi;
5160     uint64_t off, lpol, lpou;
5161     size_t   len;
5162     uint16_t status;
5163 
5164     numdl = (dw10 >> 16);
5165     numdu = (dw11 & 0xffff);
5166     lspi = (dw11 >> 16);
5167     lpol = dw12;
5168     lpou = dw13;
5169 
5170     len = (((numdu << 16) | numdl) + 1) << 2;
5171     off = (lpou << 32ULL) | lpol;
5172 
5173     if (off & 0x3) {
5174         return NVME_INVALID_FIELD | NVME_DNR;
5175     }
5176 
5177     trace_pci_nvme_get_log(nvme_cid(req), lid, lsp, rae, len, off);
5178 
5179     status = nvme_check_mdts(n, len);
5180     if (status) {
5181         return status;
5182     }
5183 
5184     switch (lid) {
5185     case NVME_LOG_ERROR_INFO:
5186         return nvme_error_info(n, rae, len, off, req);
5187     case NVME_LOG_SMART_INFO:
5188         return nvme_smart_info(n, rae, len, off, req);
5189     case NVME_LOG_FW_SLOT_INFO:
5190         return nvme_fw_log_info(n, len, off, req);
5191     case NVME_LOG_CHANGED_NSLIST:
5192         return nvme_changed_nslist(n, rae, len, off, req);
5193     case NVME_LOG_CMD_EFFECTS:
5194         return nvme_cmd_effects(n, csi, len, off, req);
5195     case NVME_LOG_ENDGRP:
5196         return nvme_endgrp_info(n, rae, len, off, req);
5197     case NVME_LOG_FDP_CONFS:
5198         return nvme_fdp_confs(n, lspi, len, off, req);
5199     case NVME_LOG_FDP_RUH_USAGE:
5200         return nvme_fdp_ruh_usage(n, lspi, dw10, dw12, len, off, req);
5201     case NVME_LOG_FDP_STATS:
5202         return nvme_fdp_stats(n, lspi, len, off, req);
5203     case NVME_LOG_FDP_EVENTS:
5204         return nvme_fdp_events(n, lspi, len, off, req);
5205     default:
5206         trace_pci_nvme_err_invalid_log_page(nvme_cid(req), lid);
5207         return NVME_INVALID_FIELD | NVME_DNR;
5208     }
5209 }
5210 
5211 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
5212 {
5213     PCIDevice *pci = PCI_DEVICE(n);
5214     uint16_t offset = (cq->cqid << 3) + (1 << 2);
5215 
5216     n->cq[cq->cqid] = NULL;
5217     qemu_bh_delete(cq->bh);
5218     if (cq->ioeventfd_enabled) {
5219         memory_region_del_eventfd(&n->iomem,
5220                                   0x1000 + offset, 4, false, 0, &cq->notifier);
5221         event_notifier_set_handler(&cq->notifier, NULL);
5222         event_notifier_cleanup(&cq->notifier);
5223     }
5224     if (msix_enabled(pci)) {
5225         msix_vector_unuse(pci, cq->vector);
5226     }
5227     if (cq->cqid) {
5228         g_free(cq);
5229     }
5230 }
5231 
5232 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)
5233 {
5234     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
5235     NvmeCQueue *cq;
5236     uint16_t qid = le16_to_cpu(c->qid);
5237 
5238     if (unlikely(!qid || nvme_check_cqid(n, qid))) {
5239         trace_pci_nvme_err_invalid_del_cq_cqid(qid);
5240         return NVME_INVALID_CQID | NVME_DNR;
5241     }
5242 
5243     cq = n->cq[qid];
5244     if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
5245         trace_pci_nvme_err_invalid_del_cq_notempty(qid);
5246         return NVME_INVALID_QUEUE_DEL;
5247     }
5248 
5249     if (cq->irq_enabled && cq->tail != cq->head) {
5250         n->cq_pending--;
5251     }
5252 
5253     nvme_irq_deassert(n, cq);
5254     trace_pci_nvme_del_cq(qid);
5255     nvme_free_cq(cq, n);
5256     return NVME_SUCCESS;
5257 }
5258 
5259 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
5260                          uint16_t cqid, uint16_t vector, uint16_t size,
5261                          uint16_t irq_enabled)
5262 {
5263     PCIDevice *pci = PCI_DEVICE(n);
5264 
5265     if (msix_enabled(pci)) {
5266         msix_vector_use(pci, vector);
5267     }
5268     cq->ctrl = n;
5269     cq->cqid = cqid;
5270     cq->size = size;
5271     cq->dma_addr = dma_addr;
5272     cq->phase = 1;
5273     cq->irq_enabled = irq_enabled;
5274     cq->vector = vector;
5275     cq->head = cq->tail = 0;
5276     QTAILQ_INIT(&cq->req_list);
5277     QTAILQ_INIT(&cq->sq_list);
5278     if (n->dbbuf_enabled) {
5279         cq->db_addr = n->dbbuf_dbs + (cqid << 3) + (1 << 2);
5280         cq->ei_addr = n->dbbuf_eis + (cqid << 3) + (1 << 2);
5281 
5282         if (n->params.ioeventfd && cqid != 0) {
5283             if (!nvme_init_cq_ioeventfd(cq)) {
5284                 cq->ioeventfd_enabled = true;
5285             }
5286         }
5287     }
5288     n->cq[cqid] = cq;
5289     cq->bh = qemu_bh_new_guarded(nvme_post_cqes, cq,
5290                                  &DEVICE(cq->ctrl)->mem_reentrancy_guard);
5291 }
5292 
5293 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeRequest *req)
5294 {
5295     NvmeCQueue *cq;
5296     NvmeCreateCq *c = (NvmeCreateCq *)&req->cmd;
5297     uint16_t cqid = le16_to_cpu(c->cqid);
5298     uint16_t vector = le16_to_cpu(c->irq_vector);
5299     uint16_t qsize = le16_to_cpu(c->qsize);
5300     uint16_t qflags = le16_to_cpu(c->cq_flags);
5301     uint64_t prp1 = le64_to_cpu(c->prp1);
5302 
5303     trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
5304                              NVME_CQ_FLAGS_IEN(qflags) != 0);
5305 
5306     if (unlikely(!cqid || cqid > n->conf_ioqpairs || n->cq[cqid] != NULL)) {
5307         trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
5308         return NVME_INVALID_QID | NVME_DNR;
5309     }
5310     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
5311         trace_pci_nvme_err_invalid_create_cq_size(qsize);
5312         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5313     }
5314     if (unlikely(prp1 & (n->page_size - 1))) {
5315         trace_pci_nvme_err_invalid_create_cq_addr(prp1);
5316         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
5317     }
5318     if (unlikely(!msix_enabled(PCI_DEVICE(n)) && vector)) {
5319         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5320         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5321     }
5322     if (unlikely(vector >= n->conf_msix_qsize)) {
5323         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5324         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5325     }
5326     if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
5327         trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
5328         return NVME_INVALID_FIELD | NVME_DNR;
5329     }
5330 
5331     cq = g_malloc0(sizeof(*cq));
5332     nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
5333                  NVME_CQ_FLAGS_IEN(qflags));
5334 
5335     /*
5336      * It is only required to set qs_created when creating a completion queue;
5337      * creating a submission queue without a matching completion queue will
5338      * fail.
5339      */
5340     n->qs_created = true;
5341     return NVME_SUCCESS;
5342 }
5343 
5344 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl *n, NvmeRequest *req)
5345 {
5346     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5347 
5348     return nvme_c2h(n, id, sizeof(id), req);
5349 }
5350 
5351 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeRequest *req)
5352 {
5353     trace_pci_nvme_identify_ctrl();
5354 
5355     return nvme_c2h(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl), req);
5356 }
5357 
5358 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl *n, NvmeRequest *req)
5359 {
5360     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5361     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5362     NvmeIdCtrlNvm *id_nvm = (NvmeIdCtrlNvm *)&id;
5363 
5364     trace_pci_nvme_identify_ctrl_csi(c->csi);
5365 
5366     switch (c->csi) {
5367     case NVME_CSI_NVM:
5368         id_nvm->vsl = n->params.vsl;
5369         id_nvm->dmrsl = cpu_to_le32(n->dmrsl);
5370         break;
5371 
5372     case NVME_CSI_ZONED:
5373         ((NvmeIdCtrlZoned *)&id)->zasl = n->params.zasl;
5374         break;
5375 
5376     default:
5377         return NVME_INVALID_FIELD | NVME_DNR;
5378     }
5379 
5380     return nvme_c2h(n, id, sizeof(id), req);
5381 }
5382 
5383 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeRequest *req, bool active)
5384 {
5385     NvmeNamespace *ns;
5386     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5387     uint32_t nsid = le32_to_cpu(c->nsid);
5388 
5389     trace_pci_nvme_identify_ns(nsid);
5390 
5391     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5392         return NVME_INVALID_NSID | NVME_DNR;
5393     }
5394 
5395     ns = nvme_ns(n, nsid);
5396     if (unlikely(!ns)) {
5397         if (!active) {
5398             ns = nvme_subsys_ns(n->subsys, nsid);
5399             if (!ns) {
5400                 return nvme_rpt_empty_id_struct(n, req);
5401             }
5402         } else {
5403             return nvme_rpt_empty_id_struct(n, req);
5404         }
5405     }
5406 
5407     if (active || ns->csi == NVME_CSI_NVM) {
5408         return nvme_c2h(n, (uint8_t *)&ns->id_ns, sizeof(NvmeIdNs), req);
5409     }
5410 
5411     return NVME_INVALID_CMD_SET | NVME_DNR;
5412 }
5413 
5414 static uint16_t nvme_identify_ctrl_list(NvmeCtrl *n, NvmeRequest *req,
5415                                         bool attached)
5416 {
5417     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5418     uint32_t nsid = le32_to_cpu(c->nsid);
5419     uint16_t min_id = le16_to_cpu(c->ctrlid);
5420     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5421     uint16_t *ids = &list[1];
5422     NvmeNamespace *ns;
5423     NvmeCtrl *ctrl;
5424     int cntlid, nr_ids = 0;
5425 
5426     trace_pci_nvme_identify_ctrl_list(c->cns, min_id);
5427 
5428     if (!n->subsys) {
5429         return NVME_INVALID_FIELD | NVME_DNR;
5430     }
5431 
5432     if (attached) {
5433         if (nsid == NVME_NSID_BROADCAST) {
5434             return NVME_INVALID_FIELD | NVME_DNR;
5435         }
5436 
5437         ns = nvme_subsys_ns(n->subsys, nsid);
5438         if (!ns) {
5439             return NVME_INVALID_FIELD | NVME_DNR;
5440         }
5441     }
5442 
5443     for (cntlid = min_id; cntlid < ARRAY_SIZE(n->subsys->ctrls); cntlid++) {
5444         ctrl = nvme_subsys_ctrl(n->subsys, cntlid);
5445         if (!ctrl) {
5446             continue;
5447         }
5448 
5449         if (attached && !nvme_ns(ctrl, nsid)) {
5450             continue;
5451         }
5452 
5453         ids[nr_ids++] = cntlid;
5454     }
5455 
5456     list[0] = nr_ids;
5457 
5458     return nvme_c2h(n, (uint8_t *)list, sizeof(list), req);
5459 }
5460 
5461 static uint16_t nvme_identify_pri_ctrl_cap(NvmeCtrl *n, NvmeRequest *req)
5462 {
5463     trace_pci_nvme_identify_pri_ctrl_cap(le16_to_cpu(n->pri_ctrl_cap.cntlid));
5464 
5465     return nvme_c2h(n, (uint8_t *)&n->pri_ctrl_cap,
5466                     sizeof(NvmePriCtrlCap), req);
5467 }
5468 
5469 static uint16_t nvme_identify_sec_ctrl_list(NvmeCtrl *n, NvmeRequest *req)
5470 {
5471     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5472     uint16_t pri_ctrl_id = le16_to_cpu(n->pri_ctrl_cap.cntlid);
5473     uint16_t min_id = le16_to_cpu(c->ctrlid);
5474     uint8_t num_sec_ctrl = n->sec_ctrl_list.numcntl;
5475     NvmeSecCtrlList list = {0};
5476     uint8_t i;
5477 
5478     for (i = 0; i < num_sec_ctrl; i++) {
5479         if (n->sec_ctrl_list.sec[i].scid >= min_id) {
5480             list.numcntl = num_sec_ctrl - i;
5481             memcpy(&list.sec, n->sec_ctrl_list.sec + i,
5482                    list.numcntl * sizeof(NvmeSecCtrlEntry));
5483             break;
5484         }
5485     }
5486 
5487     trace_pci_nvme_identify_sec_ctrl_list(pri_ctrl_id, list.numcntl);
5488 
5489     return nvme_c2h(n, (uint8_t *)&list, sizeof(list), req);
5490 }
5491 
5492 static uint16_t nvme_identify_ns_csi(NvmeCtrl *n, NvmeRequest *req,
5493                                      bool active)
5494 {
5495     NvmeNamespace *ns;
5496     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5497     uint32_t nsid = le32_to_cpu(c->nsid);
5498 
5499     trace_pci_nvme_identify_ns_csi(nsid, c->csi);
5500 
5501     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5502         return NVME_INVALID_NSID | NVME_DNR;
5503     }
5504 
5505     ns = nvme_ns(n, nsid);
5506     if (unlikely(!ns)) {
5507         if (!active) {
5508             ns = nvme_subsys_ns(n->subsys, nsid);
5509             if (!ns) {
5510                 return nvme_rpt_empty_id_struct(n, req);
5511             }
5512         } else {
5513             return nvme_rpt_empty_id_struct(n, req);
5514         }
5515     }
5516 
5517     if (c->csi == NVME_CSI_NVM) {
5518         return nvme_c2h(n, (uint8_t *)&ns->id_ns_nvm, sizeof(NvmeIdNsNvm),
5519                         req);
5520     } else if (c->csi == NVME_CSI_ZONED && ns->csi == NVME_CSI_ZONED) {
5521         return nvme_c2h(n, (uint8_t *)ns->id_ns_zoned, sizeof(NvmeIdNsZoned),
5522                         req);
5523     }
5524 
5525     return NVME_INVALID_FIELD | NVME_DNR;
5526 }
5527 
5528 static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeRequest *req,
5529                                      bool active)
5530 {
5531     NvmeNamespace *ns;
5532     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5533     uint32_t min_nsid = le32_to_cpu(c->nsid);
5534     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5535     static const int data_len = sizeof(list);
5536     uint32_t *list_ptr = (uint32_t *)list;
5537     int i, j = 0;
5538 
5539     trace_pci_nvme_identify_nslist(min_nsid);
5540 
5541     /*
5542      * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
5543      * since the Active Namespace ID List should return namespaces with ids
5544      * *higher* than the NSID specified in the command. This is also specified
5545      * in the spec (NVM Express v1.3d, Section 5.15.4).
5546      */
5547     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5548         return NVME_INVALID_NSID | NVME_DNR;
5549     }
5550 
5551     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5552         ns = nvme_ns(n, i);
5553         if (!ns) {
5554             if (!active) {
5555                 ns = nvme_subsys_ns(n->subsys, i);
5556                 if (!ns) {
5557                     continue;
5558                 }
5559             } else {
5560                 continue;
5561             }
5562         }
5563         if (ns->params.nsid <= min_nsid) {
5564             continue;
5565         }
5566         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5567         if (j == data_len / sizeof(uint32_t)) {
5568             break;
5569         }
5570     }
5571 
5572     return nvme_c2h(n, list, data_len, req);
5573 }
5574 
5575 static uint16_t nvme_identify_nslist_csi(NvmeCtrl *n, NvmeRequest *req,
5576                                          bool active)
5577 {
5578     NvmeNamespace *ns;
5579     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5580     uint32_t min_nsid = le32_to_cpu(c->nsid);
5581     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5582     static const int data_len = sizeof(list);
5583     uint32_t *list_ptr = (uint32_t *)list;
5584     int i, j = 0;
5585 
5586     trace_pci_nvme_identify_nslist_csi(min_nsid, c->csi);
5587 
5588     /*
5589      * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
5590      */
5591     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5592         return NVME_INVALID_NSID | NVME_DNR;
5593     }
5594 
5595     if (c->csi != NVME_CSI_NVM && c->csi != NVME_CSI_ZONED) {
5596         return NVME_INVALID_FIELD | NVME_DNR;
5597     }
5598 
5599     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5600         ns = nvme_ns(n, i);
5601         if (!ns) {
5602             if (!active) {
5603                 ns = nvme_subsys_ns(n->subsys, i);
5604                 if (!ns) {
5605                     continue;
5606                 }
5607             } else {
5608                 continue;
5609             }
5610         }
5611         if (ns->params.nsid <= min_nsid || c->csi != ns->csi) {
5612             continue;
5613         }
5614         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5615         if (j == data_len / sizeof(uint32_t)) {
5616             break;
5617         }
5618     }
5619 
5620     return nvme_c2h(n, list, data_len, req);
5621 }
5622 
5623 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl *n, NvmeRequest *req)
5624 {
5625     NvmeNamespace *ns;
5626     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5627     uint32_t nsid = le32_to_cpu(c->nsid);
5628     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5629     uint8_t *pos = list;
5630     struct {
5631         NvmeIdNsDescr hdr;
5632         uint8_t v[NVME_NIDL_UUID];
5633     } QEMU_PACKED uuid = {};
5634     struct {
5635         NvmeIdNsDescr hdr;
5636         uint64_t v;
5637     } QEMU_PACKED eui64 = {};
5638     struct {
5639         NvmeIdNsDescr hdr;
5640         uint8_t v;
5641     } QEMU_PACKED csi = {};
5642 
5643     trace_pci_nvme_identify_ns_descr_list(nsid);
5644 
5645     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5646         return NVME_INVALID_NSID | NVME_DNR;
5647     }
5648 
5649     ns = nvme_ns(n, nsid);
5650     if (unlikely(!ns)) {
5651         return NVME_INVALID_FIELD | NVME_DNR;
5652     }
5653 
5654     if (!qemu_uuid_is_null(&ns->params.uuid)) {
5655         uuid.hdr.nidt = NVME_NIDT_UUID;
5656         uuid.hdr.nidl = NVME_NIDL_UUID;
5657         memcpy(uuid.v, ns->params.uuid.data, NVME_NIDL_UUID);
5658         memcpy(pos, &uuid, sizeof(uuid));
5659         pos += sizeof(uuid);
5660     }
5661 
5662     if (ns->params.eui64) {
5663         eui64.hdr.nidt = NVME_NIDT_EUI64;
5664         eui64.hdr.nidl = NVME_NIDL_EUI64;
5665         eui64.v = cpu_to_be64(ns->params.eui64);
5666         memcpy(pos, &eui64, sizeof(eui64));
5667         pos += sizeof(eui64);
5668     }
5669 
5670     csi.hdr.nidt = NVME_NIDT_CSI;
5671     csi.hdr.nidl = NVME_NIDL_CSI;
5672     csi.v = ns->csi;
5673     memcpy(pos, &csi, sizeof(csi));
5674     pos += sizeof(csi);
5675 
5676     return nvme_c2h(n, list, sizeof(list), req);
5677 }
5678 
5679 static uint16_t nvme_identify_cmd_set(NvmeCtrl *n, NvmeRequest *req)
5680 {
5681     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5682     static const int data_len = sizeof(list);
5683 
5684     trace_pci_nvme_identify_cmd_set();
5685 
5686     NVME_SET_CSI(*list, NVME_CSI_NVM);
5687     NVME_SET_CSI(*list, NVME_CSI_ZONED);
5688 
5689     return nvme_c2h(n, list, data_len, req);
5690 }
5691 
5692 static uint16_t nvme_identify(NvmeCtrl *n, NvmeRequest *req)
5693 {
5694     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5695 
5696     trace_pci_nvme_identify(nvme_cid(req), c->cns, le16_to_cpu(c->ctrlid),
5697                             c->csi);
5698 
5699     switch (c->cns) {
5700     case NVME_ID_CNS_NS:
5701         return nvme_identify_ns(n, req, true);
5702     case NVME_ID_CNS_NS_PRESENT:
5703         return nvme_identify_ns(n, req, false);
5704     case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST:
5705         return nvme_identify_ctrl_list(n, req, true);
5706     case NVME_ID_CNS_CTRL_LIST:
5707         return nvme_identify_ctrl_list(n, req, false);
5708     case NVME_ID_CNS_PRIMARY_CTRL_CAP:
5709         return nvme_identify_pri_ctrl_cap(n, req);
5710     case NVME_ID_CNS_SECONDARY_CTRL_LIST:
5711         return nvme_identify_sec_ctrl_list(n, req);
5712     case NVME_ID_CNS_CS_NS:
5713         return nvme_identify_ns_csi(n, req, true);
5714     case NVME_ID_CNS_CS_NS_PRESENT:
5715         return nvme_identify_ns_csi(n, req, false);
5716     case NVME_ID_CNS_CTRL:
5717         return nvme_identify_ctrl(n, req);
5718     case NVME_ID_CNS_CS_CTRL:
5719         return nvme_identify_ctrl_csi(n, req);
5720     case NVME_ID_CNS_NS_ACTIVE_LIST:
5721         return nvme_identify_nslist(n, req, true);
5722     case NVME_ID_CNS_NS_PRESENT_LIST:
5723         return nvme_identify_nslist(n, req, false);
5724     case NVME_ID_CNS_CS_NS_ACTIVE_LIST:
5725         return nvme_identify_nslist_csi(n, req, true);
5726     case NVME_ID_CNS_CS_NS_PRESENT_LIST:
5727         return nvme_identify_nslist_csi(n, req, false);
5728     case NVME_ID_CNS_NS_DESCR_LIST:
5729         return nvme_identify_ns_descr_list(n, req);
5730     case NVME_ID_CNS_IO_COMMAND_SET:
5731         return nvme_identify_cmd_set(n, req);
5732     default:
5733         trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
5734         return NVME_INVALID_FIELD | NVME_DNR;
5735     }
5736 }
5737 
5738 static uint16_t nvme_abort(NvmeCtrl *n, NvmeRequest *req)
5739 {
5740     uint16_t sqid = le32_to_cpu(req->cmd.cdw10) & 0xffff;
5741 
5742     req->cqe.result = 1;
5743     if (nvme_check_sqid(n, sqid)) {
5744         return NVME_INVALID_FIELD | NVME_DNR;
5745     }
5746 
5747     return NVME_SUCCESS;
5748 }
5749 
5750 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
5751 {
5752     trace_pci_nvme_setfeat_timestamp(ts);
5753 
5754     n->host_timestamp = le64_to_cpu(ts);
5755     n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5756 }
5757 
5758 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
5759 {
5760     uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5761     uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
5762 
5763     union nvme_timestamp {
5764         struct {
5765             uint64_t timestamp:48;
5766             uint64_t sync:1;
5767             uint64_t origin:3;
5768             uint64_t rsvd1:12;
5769         };
5770         uint64_t all;
5771     };
5772 
5773     union nvme_timestamp ts;
5774     ts.all = 0;
5775     ts.timestamp = n->host_timestamp + elapsed_time;
5776 
5777     /* If the host timestamp is non-zero, set the timestamp origin */
5778     ts.origin = n->host_timestamp ? 0x01 : 0x00;
5779 
5780     trace_pci_nvme_getfeat_timestamp(ts.all);
5781 
5782     return cpu_to_le64(ts.all);
5783 }
5784 
5785 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
5786 {
5787     uint64_t timestamp = nvme_get_timestamp(n);
5788 
5789     return nvme_c2h(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
5790 }
5791 
5792 static int nvme_get_feature_fdp(NvmeCtrl *n, uint32_t endgrpid,
5793                                 uint32_t *result)
5794 {
5795     *result = 0;
5796 
5797     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
5798         return NVME_INVALID_FIELD | NVME_DNR;
5799     }
5800 
5801     *result = FIELD_DP16(0, FEAT_FDP, FDPE, 1);
5802     *result = FIELD_DP16(*result, FEAT_FDP, CONF_NDX, 0);
5803 
5804     return NVME_SUCCESS;
5805 }
5806 
5807 static uint16_t nvme_get_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
5808                                             NvmeRequest *req, uint32_t *result)
5809 {
5810     NvmeCmd *cmd = &req->cmd;
5811     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
5812     uint16_t ph = cdw11 & 0xffff;
5813     uint8_t noet = (cdw11 >> 16) & 0xff;
5814     uint16_t ruhid, ret;
5815     uint32_t nentries = 0;
5816     uint8_t s_events_ndx = 0;
5817     size_t s_events_siz = sizeof(NvmeFdpEventDescr) * noet;
5818     g_autofree NvmeFdpEventDescr *s_events = g_malloc0(s_events_siz);
5819     NvmeRuHandle *ruh;
5820     NvmeFdpEventDescr *s_event;
5821 
5822     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
5823         return NVME_FDP_DISABLED | NVME_DNR;
5824     }
5825 
5826     if (!nvme_ph_valid(ns, ph)) {
5827         return NVME_INVALID_FIELD | NVME_DNR;
5828     }
5829 
5830     ruhid = ns->fdp.phs[ph];
5831     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
5832 
5833     assert(ruh);
5834 
5835     if (unlikely(noet == 0)) {
5836         return NVME_INVALID_FIELD | NVME_DNR;
5837     }
5838 
5839     for (uint8_t event_type = 0; event_type < FDP_EVT_MAX; event_type++) {
5840         uint8_t shift = nvme_fdp_evf_shifts[event_type];
5841         if (!shift && event_type) {
5842             /*
5843              * only first entry (event_type == 0) has a shift value of 0
5844              * other entries are simply unpopulated.
5845              */
5846             continue;
5847         }
5848 
5849         nentries++;
5850 
5851         s_event = &s_events[s_events_ndx];
5852         s_event->evt = event_type;
5853         s_event->evta = (ruh->event_filter >> shift) & 0x1;
5854 
5855         /* break if all `noet` entries are filled */
5856         if ((++s_events_ndx) == noet) {
5857             break;
5858         }
5859     }
5860 
5861     ret = nvme_c2h(n, s_events, s_events_siz, req);
5862     if (ret) {
5863         return ret;
5864     }
5865 
5866     *result = nentries;
5867     return NVME_SUCCESS;
5868 }
5869 
5870 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeRequest *req)
5871 {
5872     NvmeCmd *cmd = &req->cmd;
5873     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5874     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5875     uint32_t nsid = le32_to_cpu(cmd->nsid);
5876     uint32_t result;
5877     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
5878     NvmeGetFeatureSelect sel = NVME_GETFEAT_SELECT(dw10);
5879     uint16_t iv;
5880     NvmeNamespace *ns;
5881     int i;
5882     uint16_t endgrpid = 0, ret = NVME_SUCCESS;
5883 
5884     static const uint32_t nvme_feature_default[NVME_FID_MAX] = {
5885         [NVME_ARBITRATION] = NVME_ARB_AB_NOLIMIT,
5886     };
5887 
5888     trace_pci_nvme_getfeat(nvme_cid(req), nsid, fid, sel, dw11);
5889 
5890     if (!nvme_feature_support[fid]) {
5891         return NVME_INVALID_FIELD | NVME_DNR;
5892     }
5893 
5894     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
5895         if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5896             /*
5897              * The Reservation Notification Mask and Reservation Persistence
5898              * features require a status code of Invalid Field in Command when
5899              * NSID is FFFFFFFFh. Since the device does not support those
5900              * features we can always return Invalid Namespace or Format as we
5901              * should do for all other features.
5902              */
5903             return NVME_INVALID_NSID | NVME_DNR;
5904         }
5905 
5906         if (!nvme_ns(n, nsid)) {
5907             return NVME_INVALID_FIELD | NVME_DNR;
5908         }
5909     }
5910 
5911     switch (sel) {
5912     case NVME_GETFEAT_SELECT_CURRENT:
5913         break;
5914     case NVME_GETFEAT_SELECT_SAVED:
5915         /* no features are saveable by the controller; fallthrough */
5916     case NVME_GETFEAT_SELECT_DEFAULT:
5917         goto defaults;
5918     case NVME_GETFEAT_SELECT_CAP:
5919         result = nvme_feature_cap[fid];
5920         goto out;
5921     }
5922 
5923     switch (fid) {
5924     case NVME_TEMPERATURE_THRESHOLD:
5925         result = 0;
5926 
5927         /*
5928          * The controller only implements the Composite Temperature sensor, so
5929          * return 0 for all other sensors.
5930          */
5931         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
5932             goto out;
5933         }
5934 
5935         switch (NVME_TEMP_THSEL(dw11)) {
5936         case NVME_TEMP_THSEL_OVER:
5937             result = n->features.temp_thresh_hi;
5938             goto out;
5939         case NVME_TEMP_THSEL_UNDER:
5940             result = n->features.temp_thresh_low;
5941             goto out;
5942         }
5943 
5944         return NVME_INVALID_FIELD | NVME_DNR;
5945     case NVME_ERROR_RECOVERY:
5946         if (!nvme_nsid_valid(n, nsid)) {
5947             return NVME_INVALID_NSID | NVME_DNR;
5948         }
5949 
5950         ns = nvme_ns(n, nsid);
5951         if (unlikely(!ns)) {
5952             return NVME_INVALID_FIELD | NVME_DNR;
5953         }
5954 
5955         result = ns->features.err_rec;
5956         goto out;
5957     case NVME_VOLATILE_WRITE_CACHE:
5958         result = 0;
5959         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5960             ns = nvme_ns(n, i);
5961             if (!ns) {
5962                 continue;
5963             }
5964 
5965             result = blk_enable_write_cache(ns->blkconf.blk);
5966             if (result) {
5967                 break;
5968             }
5969         }
5970         trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
5971         goto out;
5972     case NVME_ASYNCHRONOUS_EVENT_CONF:
5973         result = n->features.async_config;
5974         goto out;
5975     case NVME_TIMESTAMP:
5976         return nvme_get_feature_timestamp(n, req);
5977     case NVME_HOST_BEHAVIOR_SUPPORT:
5978         return nvme_c2h(n, (uint8_t *)&n->features.hbs,
5979                         sizeof(n->features.hbs), req);
5980     case NVME_FDP_MODE:
5981         endgrpid = dw11 & 0xff;
5982 
5983         if (endgrpid != 0x1) {
5984             return NVME_INVALID_FIELD | NVME_DNR;
5985         }
5986 
5987         ret = nvme_get_feature_fdp(n, endgrpid, &result);
5988         if (ret) {
5989             return ret;
5990         }
5991         goto out;
5992     case NVME_FDP_EVENTS:
5993         if (!nvme_nsid_valid(n, nsid)) {
5994             return NVME_INVALID_NSID | NVME_DNR;
5995         }
5996 
5997         ns = nvme_ns(n, nsid);
5998         if (unlikely(!ns)) {
5999             return NVME_INVALID_FIELD | NVME_DNR;
6000         }
6001 
6002         ret = nvme_get_feature_fdp_events(n, ns, req, &result);
6003         if (ret) {
6004             return ret;
6005         }
6006         goto out;
6007     default:
6008         break;
6009     }
6010 
6011 defaults:
6012     switch (fid) {
6013     case NVME_TEMPERATURE_THRESHOLD:
6014         result = 0;
6015 
6016         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6017             break;
6018         }
6019 
6020         if (NVME_TEMP_THSEL(dw11) == NVME_TEMP_THSEL_OVER) {
6021             result = NVME_TEMPERATURE_WARNING;
6022         }
6023 
6024         break;
6025     case NVME_NUMBER_OF_QUEUES:
6026         result = (n->conf_ioqpairs - 1) | ((n->conf_ioqpairs - 1) << 16);
6027         trace_pci_nvme_getfeat_numq(result);
6028         break;
6029     case NVME_INTERRUPT_VECTOR_CONF:
6030         iv = dw11 & 0xffff;
6031         if (iv >= n->conf_ioqpairs + 1) {
6032             return NVME_INVALID_FIELD | NVME_DNR;
6033         }
6034 
6035         result = iv;
6036         if (iv == n->admin_cq.vector) {
6037             result |= NVME_INTVC_NOCOALESCING;
6038         }
6039         break;
6040     case NVME_FDP_MODE:
6041         endgrpid = dw11 & 0xff;
6042 
6043         if (endgrpid != 0x1) {
6044             return NVME_INVALID_FIELD | NVME_DNR;
6045         }
6046 
6047         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6048         if (ret) {
6049             return ret;
6050         }
6051         goto out;
6052 
6053         break;
6054     default:
6055         result = nvme_feature_default[fid];
6056         break;
6057     }
6058 
6059 out:
6060     req->cqe.result = cpu_to_le32(result);
6061     return ret;
6062 }
6063 
6064 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
6065 {
6066     uint16_t ret;
6067     uint64_t timestamp;
6068 
6069     ret = nvme_h2c(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
6070     if (ret) {
6071         return ret;
6072     }
6073 
6074     nvme_set_timestamp(n, timestamp);
6075 
6076     return NVME_SUCCESS;
6077 }
6078 
6079 static uint16_t nvme_set_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
6080                                             NvmeRequest *req)
6081 {
6082     NvmeCmd *cmd = &req->cmd;
6083     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
6084     uint16_t ph = cdw11 & 0xffff;
6085     uint8_t noet = (cdw11 >> 16) & 0xff;
6086     uint16_t ret, ruhid;
6087     uint8_t enable = le32_to_cpu(cmd->cdw12) & 0x1;
6088     uint8_t event_mask = 0;
6089     unsigned int i;
6090     g_autofree uint8_t *events = g_malloc0(noet);
6091     NvmeRuHandle *ruh = NULL;
6092 
6093     assert(ns);
6094 
6095     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6096         return NVME_FDP_DISABLED | NVME_DNR;
6097     }
6098 
6099     if (!nvme_ph_valid(ns, ph)) {
6100         return NVME_INVALID_FIELD | NVME_DNR;
6101     }
6102 
6103     ruhid = ns->fdp.phs[ph];
6104     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
6105 
6106     ret = nvme_h2c(n, events, noet, req);
6107     if (ret) {
6108         return ret;
6109     }
6110 
6111     for (i = 0; i < noet; i++) {
6112         event_mask |= (1 << nvme_fdp_evf_shifts[events[i]]);
6113     }
6114 
6115     if (enable) {
6116         ruh->event_filter |= event_mask;
6117     } else {
6118         ruh->event_filter = ruh->event_filter & ~event_mask;
6119     }
6120 
6121     return NVME_SUCCESS;
6122 }
6123 
6124 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
6125 {
6126     NvmeNamespace *ns = NULL;
6127 
6128     NvmeCmd *cmd = &req->cmd;
6129     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
6130     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
6131     uint32_t nsid = le32_to_cpu(cmd->nsid);
6132     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
6133     uint8_t save = NVME_SETFEAT_SAVE(dw10);
6134     uint16_t status;
6135     int i;
6136 
6137     trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
6138 
6139     if (save && !(nvme_feature_cap[fid] & NVME_FEAT_CAP_SAVE)) {
6140         return NVME_FID_NOT_SAVEABLE | NVME_DNR;
6141     }
6142 
6143     if (!nvme_feature_support[fid]) {
6144         return NVME_INVALID_FIELD | NVME_DNR;
6145     }
6146 
6147     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
6148         if (nsid != NVME_NSID_BROADCAST) {
6149             if (!nvme_nsid_valid(n, nsid)) {
6150                 return NVME_INVALID_NSID | NVME_DNR;
6151             }
6152 
6153             ns = nvme_ns(n, nsid);
6154             if (unlikely(!ns)) {
6155                 return NVME_INVALID_FIELD | NVME_DNR;
6156             }
6157         }
6158     } else if (nsid && nsid != NVME_NSID_BROADCAST) {
6159         if (!nvme_nsid_valid(n, nsid)) {
6160             return NVME_INVALID_NSID | NVME_DNR;
6161         }
6162 
6163         return NVME_FEAT_NOT_NS_SPEC | NVME_DNR;
6164     }
6165 
6166     if (!(nvme_feature_cap[fid] & NVME_FEAT_CAP_CHANGE)) {
6167         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6168     }
6169 
6170     switch (fid) {
6171     case NVME_TEMPERATURE_THRESHOLD:
6172         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6173             break;
6174         }
6175 
6176         switch (NVME_TEMP_THSEL(dw11)) {
6177         case NVME_TEMP_THSEL_OVER:
6178             n->features.temp_thresh_hi = NVME_TEMP_TMPTH(dw11);
6179             break;
6180         case NVME_TEMP_THSEL_UNDER:
6181             n->features.temp_thresh_low = NVME_TEMP_TMPTH(dw11);
6182             break;
6183         default:
6184             return NVME_INVALID_FIELD | NVME_DNR;
6185         }
6186 
6187         if ((n->temperature >= n->features.temp_thresh_hi) ||
6188             (n->temperature <= n->features.temp_thresh_low)) {
6189             nvme_smart_event(n, NVME_SMART_TEMPERATURE);
6190         }
6191 
6192         break;
6193     case NVME_ERROR_RECOVERY:
6194         if (nsid == NVME_NSID_BROADCAST) {
6195             for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6196                 ns = nvme_ns(n, i);
6197 
6198                 if (!ns) {
6199                     continue;
6200                 }
6201 
6202                 if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat)) {
6203                     ns->features.err_rec = dw11;
6204                 }
6205             }
6206 
6207             break;
6208         }
6209 
6210         assert(ns);
6211         if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat))  {
6212             ns->features.err_rec = dw11;
6213         }
6214         break;
6215     case NVME_VOLATILE_WRITE_CACHE:
6216         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6217             ns = nvme_ns(n, i);
6218             if (!ns) {
6219                 continue;
6220             }
6221 
6222             if (!(dw11 & 0x1) && blk_enable_write_cache(ns->blkconf.blk)) {
6223                 blk_flush(ns->blkconf.blk);
6224             }
6225 
6226             blk_set_enable_write_cache(ns->blkconf.blk, dw11 & 1);
6227         }
6228 
6229         break;
6230 
6231     case NVME_NUMBER_OF_QUEUES:
6232         if (n->qs_created) {
6233             return NVME_CMD_SEQ_ERROR | NVME_DNR;
6234         }
6235 
6236         /*
6237          * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
6238          * and NSQR.
6239          */
6240         if ((dw11 & 0xffff) == 0xffff || ((dw11 >> 16) & 0xffff) == 0xffff) {
6241             return NVME_INVALID_FIELD | NVME_DNR;
6242         }
6243 
6244         trace_pci_nvme_setfeat_numq((dw11 & 0xffff) + 1,
6245                                     ((dw11 >> 16) & 0xffff) + 1,
6246                                     n->conf_ioqpairs,
6247                                     n->conf_ioqpairs);
6248         req->cqe.result = cpu_to_le32((n->conf_ioqpairs - 1) |
6249                                       ((n->conf_ioqpairs - 1) << 16));
6250         break;
6251     case NVME_ASYNCHRONOUS_EVENT_CONF:
6252         n->features.async_config = dw11;
6253         break;
6254     case NVME_TIMESTAMP:
6255         return nvme_set_feature_timestamp(n, req);
6256     case NVME_HOST_BEHAVIOR_SUPPORT:
6257         status = nvme_h2c(n, (uint8_t *)&n->features.hbs,
6258                           sizeof(n->features.hbs), req);
6259         if (status) {
6260             return status;
6261         }
6262 
6263         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6264             ns = nvme_ns(n, i);
6265 
6266             if (!ns) {
6267                 continue;
6268             }
6269 
6270             ns->id_ns.nlbaf = ns->nlbaf - 1;
6271             if (!n->features.hbs.lbafee) {
6272                 ns->id_ns.nlbaf = MIN(ns->id_ns.nlbaf, 15);
6273             }
6274         }
6275 
6276         return status;
6277     case NVME_COMMAND_SET_PROFILE:
6278         if (dw11 & 0x1ff) {
6279             trace_pci_nvme_err_invalid_iocsci(dw11 & 0x1ff);
6280             return NVME_CMD_SET_CMB_REJECTED | NVME_DNR;
6281         }
6282         break;
6283     case NVME_FDP_MODE:
6284         /* spec: abort with cmd seq err if there's one or more NS' in endgrp */
6285         return NVME_CMD_SEQ_ERROR | NVME_DNR;
6286     case NVME_FDP_EVENTS:
6287         return nvme_set_feature_fdp_events(n, ns, req);
6288     default:
6289         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6290     }
6291     return NVME_SUCCESS;
6292 }
6293 
6294 static uint16_t nvme_aer(NvmeCtrl *n, NvmeRequest *req)
6295 {
6296     trace_pci_nvme_aer(nvme_cid(req));
6297 
6298     if (n->outstanding_aers > n->params.aerl) {
6299         trace_pci_nvme_aer_aerl_exceeded();
6300         return NVME_AER_LIMIT_EXCEEDED;
6301     }
6302 
6303     n->aer_reqs[n->outstanding_aers] = req;
6304     n->outstanding_aers++;
6305 
6306     if (!QTAILQ_EMPTY(&n->aer_queue)) {
6307         nvme_process_aers(n);
6308     }
6309 
6310     return NVME_NO_COMPLETE;
6311 }
6312 
6313 static void nvme_update_dmrsl(NvmeCtrl *n)
6314 {
6315     int nsid;
6316 
6317     for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
6318         NvmeNamespace *ns = nvme_ns(n, nsid);
6319         if (!ns) {
6320             continue;
6321         }
6322 
6323         n->dmrsl = MIN_NON_ZERO(n->dmrsl,
6324                                 BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
6325     }
6326 }
6327 
6328 static void nvme_select_iocs_ns(NvmeCtrl *n, NvmeNamespace *ns)
6329 {
6330     uint32_t cc = ldl_le_p(&n->bar.cc);
6331 
6332     ns->iocs = nvme_cse_iocs_none;
6333     switch (ns->csi) {
6334     case NVME_CSI_NVM:
6335         if (NVME_CC_CSS(cc) != NVME_CC_CSS_ADMIN_ONLY) {
6336             ns->iocs = nvme_cse_iocs_nvm;
6337         }
6338         break;
6339     case NVME_CSI_ZONED:
6340         if (NVME_CC_CSS(cc) == NVME_CC_CSS_CSI) {
6341             ns->iocs = nvme_cse_iocs_zoned;
6342         } else if (NVME_CC_CSS(cc) == NVME_CC_CSS_NVM) {
6343             ns->iocs = nvme_cse_iocs_nvm;
6344         }
6345         break;
6346     }
6347 }
6348 
6349 static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
6350 {
6351     NvmeNamespace *ns;
6352     NvmeCtrl *ctrl;
6353     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
6354     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6355     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6356     uint8_t sel = dw10 & 0xf;
6357     uint16_t *nr_ids = &list[0];
6358     uint16_t *ids = &list[1];
6359     uint16_t ret;
6360     int i;
6361 
6362     trace_pci_nvme_ns_attachment(nvme_cid(req), dw10 & 0xf);
6363 
6364     if (!nvme_nsid_valid(n, nsid)) {
6365         return NVME_INVALID_NSID | NVME_DNR;
6366     }
6367 
6368     ns = nvme_subsys_ns(n->subsys, nsid);
6369     if (!ns) {
6370         return NVME_INVALID_FIELD | NVME_DNR;
6371     }
6372 
6373     ret = nvme_h2c(n, (uint8_t *)list, 4096, req);
6374     if (ret) {
6375         return ret;
6376     }
6377 
6378     if (!*nr_ids) {
6379         return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6380     }
6381 
6382     *nr_ids = MIN(*nr_ids, NVME_CONTROLLER_LIST_SIZE - 1);
6383     for (i = 0; i < *nr_ids; i++) {
6384         ctrl = nvme_subsys_ctrl(n->subsys, ids[i]);
6385         if (!ctrl) {
6386             return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6387         }
6388 
6389         switch (sel) {
6390         case NVME_NS_ATTACHMENT_ATTACH:
6391             if (nvme_ns(ctrl, nsid)) {
6392                 return NVME_NS_ALREADY_ATTACHED | NVME_DNR;
6393             }
6394 
6395             if (ns->attached && !ns->params.shared) {
6396                 return NVME_NS_PRIVATE | NVME_DNR;
6397             }
6398 
6399             nvme_attach_ns(ctrl, ns);
6400             nvme_select_iocs_ns(ctrl, ns);
6401 
6402             break;
6403 
6404         case NVME_NS_ATTACHMENT_DETACH:
6405             if (!nvme_ns(ctrl, nsid)) {
6406                 return NVME_NS_NOT_ATTACHED | NVME_DNR;
6407             }
6408 
6409             ctrl->namespaces[nsid] = NULL;
6410             ns->attached--;
6411 
6412             nvme_update_dmrsl(ctrl);
6413 
6414             break;
6415 
6416         default:
6417             return NVME_INVALID_FIELD | NVME_DNR;
6418         }
6419 
6420         /*
6421          * Add namespace id to the changed namespace id list for event clearing
6422          * via Get Log Page command.
6423          */
6424         if (!test_and_set_bit(nsid, ctrl->changed_nsids)) {
6425             nvme_enqueue_event(ctrl, NVME_AER_TYPE_NOTICE,
6426                                NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED,
6427                                NVME_LOG_CHANGED_NSLIST);
6428         }
6429     }
6430 
6431     return NVME_SUCCESS;
6432 }
6433 
6434 typedef struct NvmeFormatAIOCB {
6435     BlockAIOCB common;
6436     BlockAIOCB *aiocb;
6437     NvmeRequest *req;
6438     int ret;
6439 
6440     NvmeNamespace *ns;
6441     uint32_t nsid;
6442     bool broadcast;
6443     int64_t offset;
6444 
6445     uint8_t lbaf;
6446     uint8_t mset;
6447     uint8_t pi;
6448     uint8_t pil;
6449 } NvmeFormatAIOCB;
6450 
6451 static void nvme_format_cancel(BlockAIOCB *aiocb)
6452 {
6453     NvmeFormatAIOCB *iocb = container_of(aiocb, NvmeFormatAIOCB, common);
6454 
6455     iocb->ret = -ECANCELED;
6456 
6457     if (iocb->aiocb) {
6458         blk_aio_cancel_async(iocb->aiocb);
6459         iocb->aiocb = NULL;
6460     }
6461 }
6462 
6463 static const AIOCBInfo nvme_format_aiocb_info = {
6464     .aiocb_size = sizeof(NvmeFormatAIOCB),
6465     .cancel_async = nvme_format_cancel,
6466     .get_aio_context = nvme_get_aio_context,
6467 };
6468 
6469 static void nvme_format_set(NvmeNamespace *ns, uint8_t lbaf, uint8_t mset,
6470                             uint8_t pi, uint8_t pil)
6471 {
6472     uint8_t lbafl = lbaf & 0xf;
6473     uint8_t lbafu = lbaf >> 4;
6474 
6475     trace_pci_nvme_format_set(ns->params.nsid, lbaf, mset, pi, pil);
6476 
6477     ns->id_ns.dps = (pil << 3) | pi;
6478     ns->id_ns.flbas = (lbafu << 5) | (mset << 4) | lbafl;
6479 
6480     nvme_ns_init_format(ns);
6481 }
6482 
6483 static void nvme_do_format(NvmeFormatAIOCB *iocb);
6484 
6485 static void nvme_format_ns_cb(void *opaque, int ret)
6486 {
6487     NvmeFormatAIOCB *iocb = opaque;
6488     NvmeNamespace *ns = iocb->ns;
6489     int bytes;
6490 
6491     if (iocb->ret < 0) {
6492         goto done;
6493     } else if (ret < 0) {
6494         iocb->ret = ret;
6495         goto done;
6496     }
6497 
6498     assert(ns);
6499 
6500     if (iocb->offset < ns->size) {
6501         bytes = MIN(BDRV_REQUEST_MAX_BYTES, ns->size - iocb->offset);
6502 
6503         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, iocb->offset,
6504                                             bytes, BDRV_REQ_MAY_UNMAP,
6505                                             nvme_format_ns_cb, iocb);
6506 
6507         iocb->offset += bytes;
6508         return;
6509     }
6510 
6511     nvme_format_set(ns, iocb->lbaf, iocb->mset, iocb->pi, iocb->pil);
6512     ns->status = 0x0;
6513     iocb->ns = NULL;
6514     iocb->offset = 0;
6515 
6516 done:
6517     nvme_do_format(iocb);
6518 }
6519 
6520 static uint16_t nvme_format_check(NvmeNamespace *ns, uint8_t lbaf, uint8_t pi)
6521 {
6522     if (ns->params.zoned) {
6523         return NVME_INVALID_FORMAT | NVME_DNR;
6524     }
6525 
6526     if (lbaf > ns->id_ns.nlbaf) {
6527         return NVME_INVALID_FORMAT | NVME_DNR;
6528     }
6529 
6530     if (pi && (ns->id_ns.lbaf[lbaf].ms < nvme_pi_tuple_size(ns))) {
6531         return NVME_INVALID_FORMAT | NVME_DNR;
6532     }
6533 
6534     if (pi && pi > NVME_ID_NS_DPS_TYPE_3) {
6535         return NVME_INVALID_FIELD | NVME_DNR;
6536     }
6537 
6538     return NVME_SUCCESS;
6539 }
6540 
6541 static void nvme_do_format(NvmeFormatAIOCB *iocb)
6542 {
6543     NvmeRequest *req = iocb->req;
6544     NvmeCtrl *n = nvme_ctrl(req);
6545     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6546     uint8_t lbaf = dw10 & 0xf;
6547     uint8_t pi = (dw10 >> 5) & 0x7;
6548     uint16_t status;
6549     int i;
6550 
6551     if (iocb->ret < 0) {
6552         goto done;
6553     }
6554 
6555     if (iocb->broadcast) {
6556         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
6557             iocb->ns = nvme_ns(n, i);
6558             if (iocb->ns) {
6559                 iocb->nsid = i;
6560                 break;
6561             }
6562         }
6563     }
6564 
6565     if (!iocb->ns) {
6566         goto done;
6567     }
6568 
6569     status = nvme_format_check(iocb->ns, lbaf, pi);
6570     if (status) {
6571         req->status = status;
6572         goto done;
6573     }
6574 
6575     iocb->ns->status = NVME_FORMAT_IN_PROGRESS;
6576     nvme_format_ns_cb(iocb, 0);
6577     return;
6578 
6579 done:
6580     iocb->common.cb(iocb->common.opaque, iocb->ret);
6581     qemu_aio_unref(iocb);
6582 }
6583 
6584 static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
6585 {
6586     NvmeFormatAIOCB *iocb;
6587     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6588     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6589     uint8_t lbaf = dw10 & 0xf;
6590     uint8_t mset = (dw10 >> 4) & 0x1;
6591     uint8_t pi = (dw10 >> 5) & 0x7;
6592     uint8_t pil = (dw10 >> 8) & 0x1;
6593     uint8_t lbafu = (dw10 >> 12) & 0x3;
6594     uint16_t status;
6595 
6596     iocb = qemu_aio_get(&nvme_format_aiocb_info, NULL, nvme_misc_cb, req);
6597 
6598     iocb->req = req;
6599     iocb->ret = 0;
6600     iocb->ns = NULL;
6601     iocb->nsid = 0;
6602     iocb->lbaf = lbaf;
6603     iocb->mset = mset;
6604     iocb->pi = pi;
6605     iocb->pil = pil;
6606     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
6607     iocb->offset = 0;
6608 
6609     if (n->features.hbs.lbafee) {
6610         iocb->lbaf |= lbafu << 4;
6611     }
6612 
6613     if (!iocb->broadcast) {
6614         if (!nvme_nsid_valid(n, nsid)) {
6615             status = NVME_INVALID_NSID | NVME_DNR;
6616             goto out;
6617         }
6618 
6619         iocb->ns = nvme_ns(n, nsid);
6620         if (!iocb->ns) {
6621             status = NVME_INVALID_FIELD | NVME_DNR;
6622             goto out;
6623         }
6624     }
6625 
6626     req->aiocb = &iocb->common;
6627     nvme_do_format(iocb);
6628 
6629     return NVME_NO_COMPLETE;
6630 
6631 out:
6632     qemu_aio_unref(iocb);
6633 
6634     return status;
6635 }
6636 
6637 static void nvme_get_virt_res_num(NvmeCtrl *n, uint8_t rt, int *num_total,
6638                                   int *num_prim, int *num_sec)
6639 {
6640     *num_total = le32_to_cpu(rt ?
6641                              n->pri_ctrl_cap.vifrt : n->pri_ctrl_cap.vqfrt);
6642     *num_prim = le16_to_cpu(rt ?
6643                             n->pri_ctrl_cap.virfap : n->pri_ctrl_cap.vqrfap);
6644     *num_sec = le16_to_cpu(rt ? n->pri_ctrl_cap.virfa : n->pri_ctrl_cap.vqrfa);
6645 }
6646 
6647 static uint16_t nvme_assign_virt_res_to_prim(NvmeCtrl *n, NvmeRequest *req,
6648                                              uint16_t cntlid, uint8_t rt,
6649                                              int nr)
6650 {
6651     int num_total, num_prim, num_sec;
6652 
6653     if (cntlid != n->cntlid) {
6654         return NVME_INVALID_CTRL_ID | NVME_DNR;
6655     }
6656 
6657     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6658 
6659     if (nr > num_total) {
6660         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6661     }
6662 
6663     if (nr > num_total - num_sec) {
6664         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6665     }
6666 
6667     if (rt) {
6668         n->next_pri_ctrl_cap.virfap = cpu_to_le16(nr);
6669     } else {
6670         n->next_pri_ctrl_cap.vqrfap = cpu_to_le16(nr);
6671     }
6672 
6673     req->cqe.result = cpu_to_le32(nr);
6674     return req->status;
6675 }
6676 
6677 static void nvme_update_virt_res(NvmeCtrl *n, NvmeSecCtrlEntry *sctrl,
6678                                  uint8_t rt, int nr)
6679 {
6680     int prev_nr, prev_total;
6681 
6682     if (rt) {
6683         prev_nr = le16_to_cpu(sctrl->nvi);
6684         prev_total = le32_to_cpu(n->pri_ctrl_cap.virfa);
6685         sctrl->nvi = cpu_to_le16(nr);
6686         n->pri_ctrl_cap.virfa = cpu_to_le32(prev_total + nr - prev_nr);
6687     } else {
6688         prev_nr = le16_to_cpu(sctrl->nvq);
6689         prev_total = le32_to_cpu(n->pri_ctrl_cap.vqrfa);
6690         sctrl->nvq = cpu_to_le16(nr);
6691         n->pri_ctrl_cap.vqrfa = cpu_to_le32(prev_total + nr - prev_nr);
6692     }
6693 }
6694 
6695 static uint16_t nvme_assign_virt_res_to_sec(NvmeCtrl *n, NvmeRequest *req,
6696                                             uint16_t cntlid, uint8_t rt, int nr)
6697 {
6698     int num_total, num_prim, num_sec, num_free, diff, limit;
6699     NvmeSecCtrlEntry *sctrl;
6700 
6701     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6702     if (!sctrl) {
6703         return NVME_INVALID_CTRL_ID | NVME_DNR;
6704     }
6705 
6706     if (sctrl->scs) {
6707         return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6708     }
6709 
6710     limit = le16_to_cpu(rt ? n->pri_ctrl_cap.vifrsm : n->pri_ctrl_cap.vqfrsm);
6711     if (nr > limit) {
6712         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6713     }
6714 
6715     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6716     num_free = num_total - num_prim - num_sec;
6717     diff = nr - le16_to_cpu(rt ? sctrl->nvi : sctrl->nvq);
6718 
6719     if (diff > num_free) {
6720         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6721     }
6722 
6723     nvme_update_virt_res(n, sctrl, rt, nr);
6724     req->cqe.result = cpu_to_le32(nr);
6725 
6726     return req->status;
6727 }
6728 
6729 static uint16_t nvme_virt_set_state(NvmeCtrl *n, uint16_t cntlid, bool online)
6730 {
6731     PCIDevice *pci = PCI_DEVICE(n);
6732     NvmeCtrl *sn = NULL;
6733     NvmeSecCtrlEntry *sctrl;
6734     int vf_index;
6735 
6736     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6737     if (!sctrl) {
6738         return NVME_INVALID_CTRL_ID | NVME_DNR;
6739     }
6740 
6741     if (!pci_is_vf(pci)) {
6742         vf_index = le16_to_cpu(sctrl->vfn) - 1;
6743         sn = NVME(pcie_sriov_get_vf_at_index(pci, vf_index));
6744     }
6745 
6746     if (online) {
6747         if (!sctrl->nvi || (le16_to_cpu(sctrl->nvq) < 2) || !sn) {
6748             return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6749         }
6750 
6751         if (!sctrl->scs) {
6752             sctrl->scs = 0x1;
6753             nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6754         }
6755     } else {
6756         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_INTERRUPT, 0);
6757         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_QUEUE, 0);
6758 
6759         if (sctrl->scs) {
6760             sctrl->scs = 0x0;
6761             if (sn) {
6762                 nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6763             }
6764         }
6765     }
6766 
6767     return NVME_SUCCESS;
6768 }
6769 
6770 static uint16_t nvme_virt_mngmt(NvmeCtrl *n, NvmeRequest *req)
6771 {
6772     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6773     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
6774     uint8_t act = dw10 & 0xf;
6775     uint8_t rt = (dw10 >> 8) & 0x7;
6776     uint16_t cntlid = (dw10 >> 16) & 0xffff;
6777     int nr = dw11 & 0xffff;
6778 
6779     trace_pci_nvme_virt_mngmt(nvme_cid(req), act, cntlid, rt ? "VI" : "VQ", nr);
6780 
6781     if (rt != NVME_VIRT_RES_QUEUE && rt != NVME_VIRT_RES_INTERRUPT) {
6782         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6783     }
6784 
6785     switch (act) {
6786     case NVME_VIRT_MNGMT_ACTION_SEC_ASSIGN:
6787         return nvme_assign_virt_res_to_sec(n, req, cntlid, rt, nr);
6788     case NVME_VIRT_MNGMT_ACTION_PRM_ALLOC:
6789         return nvme_assign_virt_res_to_prim(n, req, cntlid, rt, nr);
6790     case NVME_VIRT_MNGMT_ACTION_SEC_ONLINE:
6791         return nvme_virt_set_state(n, cntlid, true);
6792     case NVME_VIRT_MNGMT_ACTION_SEC_OFFLINE:
6793         return nvme_virt_set_state(n, cntlid, false);
6794     default:
6795         return NVME_INVALID_FIELD | NVME_DNR;
6796     }
6797 }
6798 
6799 static uint16_t nvme_dbbuf_config(NvmeCtrl *n, const NvmeRequest *req)
6800 {
6801     PCIDevice *pci = PCI_DEVICE(n);
6802     uint64_t dbs_addr = le64_to_cpu(req->cmd.dptr.prp1);
6803     uint64_t eis_addr = le64_to_cpu(req->cmd.dptr.prp2);
6804     int i;
6805 
6806     /* Address should be page aligned */
6807     if (dbs_addr & (n->page_size - 1) || eis_addr & (n->page_size - 1)) {
6808         return NVME_INVALID_FIELD | NVME_DNR;
6809     }
6810 
6811     /* Save shadow buffer base addr for use during queue creation */
6812     n->dbbuf_dbs = dbs_addr;
6813     n->dbbuf_eis = eis_addr;
6814     n->dbbuf_enabled = true;
6815 
6816     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
6817         NvmeSQueue *sq = n->sq[i];
6818         NvmeCQueue *cq = n->cq[i];
6819 
6820         if (sq) {
6821             /*
6822              * CAP.DSTRD is 0, so offset of ith sq db_addr is (i<<3)
6823              * nvme_process_db() uses this hard-coded way to calculate
6824              * doorbell offsets. Be consistent with that here.
6825              */
6826             sq->db_addr = dbs_addr + (i << 3);
6827             sq->ei_addr = eis_addr + (i << 3);
6828             pci_dma_write(pci, sq->db_addr, &sq->tail, sizeof(sq->tail));
6829 
6830             if (n->params.ioeventfd && sq->sqid != 0) {
6831                 if (!nvme_init_sq_ioeventfd(sq)) {
6832                     sq->ioeventfd_enabled = true;
6833                 }
6834             }
6835         }
6836 
6837         if (cq) {
6838             /* CAP.DSTRD is 0, so offset of ith cq db_addr is (i<<3)+(1<<2) */
6839             cq->db_addr = dbs_addr + (i << 3) + (1 << 2);
6840             cq->ei_addr = eis_addr + (i << 3) + (1 << 2);
6841             pci_dma_write(pci, cq->db_addr, &cq->head, sizeof(cq->head));
6842 
6843             if (n->params.ioeventfd && cq->cqid != 0) {
6844                 if (!nvme_init_cq_ioeventfd(cq)) {
6845                     cq->ioeventfd_enabled = true;
6846                 }
6847             }
6848         }
6849     }
6850 
6851     trace_pci_nvme_dbbuf_config(dbs_addr, eis_addr);
6852 
6853     return NVME_SUCCESS;
6854 }
6855 
6856 static uint16_t nvme_directive_send(NvmeCtrl *n, NvmeRequest *req)
6857 {
6858     return NVME_INVALID_FIELD | NVME_DNR;
6859 }
6860 
6861 static uint16_t nvme_directive_receive(NvmeCtrl *n, NvmeRequest *req)
6862 {
6863     NvmeNamespace *ns;
6864     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6865     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
6866     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6867     uint8_t doper, dtype;
6868     uint32_t numd, trans_len;
6869     NvmeDirectiveIdentify id = {
6870         .supported = 1 << NVME_DIRECTIVE_IDENTIFY,
6871         .enabled = 1 << NVME_DIRECTIVE_IDENTIFY,
6872     };
6873 
6874     numd = dw10 + 1;
6875     doper = dw11 & 0xff;
6876     dtype = (dw11 >> 8) & 0xff;
6877 
6878     trans_len = MIN(sizeof(NvmeDirectiveIdentify), numd << 2);
6879 
6880     if (nsid == NVME_NSID_BROADCAST || dtype != NVME_DIRECTIVE_IDENTIFY ||
6881         doper != NVME_DIRECTIVE_RETURN_PARAMS) {
6882         return NVME_INVALID_FIELD | NVME_DNR;
6883     }
6884 
6885     ns = nvme_ns(n, nsid);
6886     if (!ns) {
6887         return NVME_INVALID_FIELD | NVME_DNR;
6888     }
6889 
6890     switch (dtype) {
6891     case NVME_DIRECTIVE_IDENTIFY:
6892         switch (doper) {
6893         case NVME_DIRECTIVE_RETURN_PARAMS:
6894             if (ns->endgrp->fdp.enabled) {
6895                 id.supported |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6896                 id.enabled |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6897                 id.persistent |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6898             }
6899 
6900             return nvme_c2h(n, (uint8_t *)&id, trans_len, req);
6901 
6902         default:
6903             return NVME_INVALID_FIELD | NVME_DNR;
6904         }
6905 
6906     default:
6907         return NVME_INVALID_FIELD;
6908     }
6909 }
6910 
6911 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
6912 {
6913     trace_pci_nvme_admin_cmd(nvme_cid(req), nvme_sqid(req), req->cmd.opcode,
6914                              nvme_adm_opc_str(req->cmd.opcode));
6915 
6916     if (!(nvme_cse_acs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
6917         trace_pci_nvme_err_invalid_admin_opc(req->cmd.opcode);
6918         return NVME_INVALID_OPCODE | NVME_DNR;
6919     }
6920 
6921     /* SGLs shall not be used for Admin commands in NVMe over PCIe */
6922     if (NVME_CMD_FLAGS_PSDT(req->cmd.flags) != NVME_PSDT_PRP) {
6923         return NVME_INVALID_FIELD | NVME_DNR;
6924     }
6925 
6926     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
6927         return NVME_INVALID_FIELD;
6928     }
6929 
6930     switch (req->cmd.opcode) {
6931     case NVME_ADM_CMD_DELETE_SQ:
6932         return nvme_del_sq(n, req);
6933     case NVME_ADM_CMD_CREATE_SQ:
6934         return nvme_create_sq(n, req);
6935     case NVME_ADM_CMD_GET_LOG_PAGE:
6936         return nvme_get_log(n, req);
6937     case NVME_ADM_CMD_DELETE_CQ:
6938         return nvme_del_cq(n, req);
6939     case NVME_ADM_CMD_CREATE_CQ:
6940         return nvme_create_cq(n, req);
6941     case NVME_ADM_CMD_IDENTIFY:
6942         return nvme_identify(n, req);
6943     case NVME_ADM_CMD_ABORT:
6944         return nvme_abort(n, req);
6945     case NVME_ADM_CMD_SET_FEATURES:
6946         return nvme_set_feature(n, req);
6947     case NVME_ADM_CMD_GET_FEATURES:
6948         return nvme_get_feature(n, req);
6949     case NVME_ADM_CMD_ASYNC_EV_REQ:
6950         return nvme_aer(n, req);
6951     case NVME_ADM_CMD_NS_ATTACHMENT:
6952         return nvme_ns_attachment(n, req);
6953     case NVME_ADM_CMD_VIRT_MNGMT:
6954         return nvme_virt_mngmt(n, req);
6955     case NVME_ADM_CMD_DBBUF_CONFIG:
6956         return nvme_dbbuf_config(n, req);
6957     case NVME_ADM_CMD_FORMAT_NVM:
6958         return nvme_format(n, req);
6959     case NVME_ADM_CMD_DIRECTIVE_SEND:
6960         return nvme_directive_send(n, req);
6961     case NVME_ADM_CMD_DIRECTIVE_RECV:
6962         return nvme_directive_receive(n, req);
6963     default:
6964         assert(false);
6965     }
6966 
6967     return NVME_INVALID_OPCODE | NVME_DNR;
6968 }
6969 
6970 static void nvme_update_sq_eventidx(const NvmeSQueue *sq)
6971 {
6972     uint32_t v = cpu_to_le32(sq->tail);
6973 
6974     trace_pci_nvme_update_sq_eventidx(sq->sqid, sq->tail);
6975 
6976     pci_dma_write(PCI_DEVICE(sq->ctrl), sq->ei_addr, &v, sizeof(v));
6977 }
6978 
6979 static void nvme_update_sq_tail(NvmeSQueue *sq)
6980 {
6981     uint32_t v;
6982 
6983     pci_dma_read(PCI_DEVICE(sq->ctrl), sq->db_addr, &v, sizeof(v));
6984 
6985     sq->tail = le32_to_cpu(v);
6986 
6987     trace_pci_nvme_update_sq_tail(sq->sqid, sq->tail);
6988 }
6989 
6990 static void nvme_process_sq(void *opaque)
6991 {
6992     NvmeSQueue *sq = opaque;
6993     NvmeCtrl *n = sq->ctrl;
6994     NvmeCQueue *cq = n->cq[sq->cqid];
6995 
6996     uint16_t status;
6997     hwaddr addr;
6998     NvmeCmd cmd;
6999     NvmeRequest *req;
7000 
7001     if (n->dbbuf_enabled) {
7002         nvme_update_sq_tail(sq);
7003     }
7004 
7005     while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
7006         addr = sq->dma_addr + sq->head * n->sqe_size;
7007         if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
7008             trace_pci_nvme_err_addr_read(addr);
7009             trace_pci_nvme_err_cfs();
7010             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
7011             break;
7012         }
7013         nvme_inc_sq_head(sq);
7014 
7015         req = QTAILQ_FIRST(&sq->req_list);
7016         QTAILQ_REMOVE(&sq->req_list, req, entry);
7017         QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
7018         nvme_req_clear(req);
7019         req->cqe.cid = cmd.cid;
7020         memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
7021 
7022         status = sq->sqid ? nvme_io_cmd(n, req) :
7023             nvme_admin_cmd(n, req);
7024         if (status != NVME_NO_COMPLETE) {
7025             req->status = status;
7026             nvme_enqueue_req_completion(cq, req);
7027         }
7028 
7029         if (n->dbbuf_enabled) {
7030             nvme_update_sq_eventidx(sq);
7031             nvme_update_sq_tail(sq);
7032         }
7033     }
7034 }
7035 
7036 static void nvme_update_msixcap_ts(PCIDevice *pci_dev, uint32_t table_size)
7037 {
7038     uint8_t *config;
7039 
7040     if (!msix_present(pci_dev)) {
7041         return;
7042     }
7043 
7044     assert(table_size > 0 && table_size <= pci_dev->msix_entries_nr);
7045 
7046     config = pci_dev->config + pci_dev->msix_cap;
7047     pci_set_word_by_mask(config + PCI_MSIX_FLAGS, PCI_MSIX_FLAGS_QSIZE,
7048                          table_size - 1);
7049 }
7050 
7051 static void nvme_activate_virt_res(NvmeCtrl *n)
7052 {
7053     PCIDevice *pci_dev = PCI_DEVICE(n);
7054     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7055     NvmeSecCtrlEntry *sctrl;
7056 
7057     /* -1 to account for the admin queue */
7058     if (pci_is_vf(pci_dev)) {
7059         sctrl = nvme_sctrl(n);
7060         cap->vqprt = sctrl->nvq;
7061         cap->viprt = sctrl->nvi;
7062         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7063         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7064     } else {
7065         cap->vqrfap = n->next_pri_ctrl_cap.vqrfap;
7066         cap->virfap = n->next_pri_ctrl_cap.virfap;
7067         n->conf_ioqpairs = le16_to_cpu(cap->vqprt) +
7068                            le16_to_cpu(cap->vqrfap) - 1;
7069         n->conf_msix_qsize = le16_to_cpu(cap->viprt) +
7070                              le16_to_cpu(cap->virfap);
7071     }
7072 }
7073 
7074 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst)
7075 {
7076     PCIDevice *pci_dev = PCI_DEVICE(n);
7077     NvmeSecCtrlEntry *sctrl;
7078     NvmeNamespace *ns;
7079     int i;
7080 
7081     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7082         ns = nvme_ns(n, i);
7083         if (!ns) {
7084             continue;
7085         }
7086 
7087         nvme_ns_drain(ns);
7088     }
7089 
7090     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7091         if (n->sq[i] != NULL) {
7092             nvme_free_sq(n->sq[i], n);
7093         }
7094     }
7095     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7096         if (n->cq[i] != NULL) {
7097             nvme_free_cq(n->cq[i], n);
7098         }
7099     }
7100 
7101     while (!QTAILQ_EMPTY(&n->aer_queue)) {
7102         NvmeAsyncEvent *event = QTAILQ_FIRST(&n->aer_queue);
7103         QTAILQ_REMOVE(&n->aer_queue, event, entry);
7104         g_free(event);
7105     }
7106 
7107     if (n->params.sriov_max_vfs) {
7108         if (!pci_is_vf(pci_dev)) {
7109             for (i = 0; i < n->sec_ctrl_list.numcntl; i++) {
7110                 sctrl = &n->sec_ctrl_list.sec[i];
7111                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
7112             }
7113 
7114             if (rst != NVME_RESET_CONTROLLER) {
7115                 pcie_sriov_pf_disable_vfs(pci_dev);
7116             }
7117         }
7118 
7119         if (rst != NVME_RESET_CONTROLLER) {
7120             nvme_activate_virt_res(n);
7121         }
7122     }
7123 
7124     n->aer_queued = 0;
7125     n->aer_mask = 0;
7126     n->outstanding_aers = 0;
7127     n->qs_created = false;
7128 
7129     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
7130 
7131     if (pci_is_vf(pci_dev)) {
7132         sctrl = nvme_sctrl(n);
7133 
7134         stl_le_p(&n->bar.csts, sctrl->scs ? 0 : NVME_CSTS_FAILED);
7135     } else {
7136         stl_le_p(&n->bar.csts, 0);
7137     }
7138 
7139     stl_le_p(&n->bar.intms, 0);
7140     stl_le_p(&n->bar.intmc, 0);
7141     stl_le_p(&n->bar.cc, 0);
7142 
7143     n->dbbuf_dbs = 0;
7144     n->dbbuf_eis = 0;
7145     n->dbbuf_enabled = false;
7146 }
7147 
7148 static void nvme_ctrl_shutdown(NvmeCtrl *n)
7149 {
7150     NvmeNamespace *ns;
7151     int i;
7152 
7153     if (n->pmr.dev) {
7154         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7155     }
7156 
7157     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7158         ns = nvme_ns(n, i);
7159         if (!ns) {
7160             continue;
7161         }
7162 
7163         nvme_ns_shutdown(ns);
7164     }
7165 }
7166 
7167 static void nvme_select_iocs(NvmeCtrl *n)
7168 {
7169     NvmeNamespace *ns;
7170     int i;
7171 
7172     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7173         ns = nvme_ns(n, i);
7174         if (!ns) {
7175             continue;
7176         }
7177 
7178         nvme_select_iocs_ns(n, ns);
7179     }
7180 }
7181 
7182 static int nvme_start_ctrl(NvmeCtrl *n)
7183 {
7184     uint64_t cap = ldq_le_p(&n->bar.cap);
7185     uint32_t cc = ldl_le_p(&n->bar.cc);
7186     uint32_t aqa = ldl_le_p(&n->bar.aqa);
7187     uint64_t asq = ldq_le_p(&n->bar.asq);
7188     uint64_t acq = ldq_le_p(&n->bar.acq);
7189     uint32_t page_bits = NVME_CC_MPS(cc) + 12;
7190     uint32_t page_size = 1 << page_bits;
7191     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
7192 
7193     if (pci_is_vf(PCI_DEVICE(n)) && !sctrl->scs) {
7194         trace_pci_nvme_err_startfail_virt_state(le16_to_cpu(sctrl->nvi),
7195                                                 le16_to_cpu(sctrl->nvq));
7196         return -1;
7197     }
7198     if (unlikely(n->cq[0])) {
7199         trace_pci_nvme_err_startfail_cq();
7200         return -1;
7201     }
7202     if (unlikely(n->sq[0])) {
7203         trace_pci_nvme_err_startfail_sq();
7204         return -1;
7205     }
7206     if (unlikely(asq & (page_size - 1))) {
7207         trace_pci_nvme_err_startfail_asq_misaligned(asq);
7208         return -1;
7209     }
7210     if (unlikely(acq & (page_size - 1))) {
7211         trace_pci_nvme_err_startfail_acq_misaligned(acq);
7212         return -1;
7213     }
7214     if (unlikely(!(NVME_CAP_CSS(cap) & (1 << NVME_CC_CSS(cc))))) {
7215         trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc));
7216         return -1;
7217     }
7218     if (unlikely(NVME_CC_MPS(cc) < NVME_CAP_MPSMIN(cap))) {
7219         trace_pci_nvme_err_startfail_page_too_small(
7220                     NVME_CC_MPS(cc),
7221                     NVME_CAP_MPSMIN(cap));
7222         return -1;
7223     }
7224     if (unlikely(NVME_CC_MPS(cc) >
7225                  NVME_CAP_MPSMAX(cap))) {
7226         trace_pci_nvme_err_startfail_page_too_large(
7227                     NVME_CC_MPS(cc),
7228                     NVME_CAP_MPSMAX(cap));
7229         return -1;
7230     }
7231     if (unlikely(NVME_CC_IOCQES(cc) <
7232                  NVME_CTRL_CQES_MIN(n->id_ctrl.cqes))) {
7233         trace_pci_nvme_err_startfail_cqent_too_small(
7234                     NVME_CC_IOCQES(cc),
7235                     NVME_CTRL_CQES_MIN(cap));
7236         return -1;
7237     }
7238     if (unlikely(NVME_CC_IOCQES(cc) >
7239                  NVME_CTRL_CQES_MAX(n->id_ctrl.cqes))) {
7240         trace_pci_nvme_err_startfail_cqent_too_large(
7241                     NVME_CC_IOCQES(cc),
7242                     NVME_CTRL_CQES_MAX(cap));
7243         return -1;
7244     }
7245     if (unlikely(NVME_CC_IOSQES(cc) <
7246                  NVME_CTRL_SQES_MIN(n->id_ctrl.sqes))) {
7247         trace_pci_nvme_err_startfail_sqent_too_small(
7248                     NVME_CC_IOSQES(cc),
7249                     NVME_CTRL_SQES_MIN(cap));
7250         return -1;
7251     }
7252     if (unlikely(NVME_CC_IOSQES(cc) >
7253                  NVME_CTRL_SQES_MAX(n->id_ctrl.sqes))) {
7254         trace_pci_nvme_err_startfail_sqent_too_large(
7255                     NVME_CC_IOSQES(cc),
7256                     NVME_CTRL_SQES_MAX(cap));
7257         return -1;
7258     }
7259     if (unlikely(!NVME_AQA_ASQS(aqa))) {
7260         trace_pci_nvme_err_startfail_asqent_sz_zero();
7261         return -1;
7262     }
7263     if (unlikely(!NVME_AQA_ACQS(aqa))) {
7264         trace_pci_nvme_err_startfail_acqent_sz_zero();
7265         return -1;
7266     }
7267 
7268     n->page_bits = page_bits;
7269     n->page_size = page_size;
7270     n->max_prp_ents = n->page_size / sizeof(uint64_t);
7271     n->cqe_size = 1 << NVME_CC_IOCQES(cc);
7272     n->sqe_size = 1 << NVME_CC_IOSQES(cc);
7273     nvme_init_cq(&n->admin_cq, n, acq, 0, 0, NVME_AQA_ACQS(aqa) + 1, 1);
7274     nvme_init_sq(&n->admin_sq, n, asq, 0, 0, NVME_AQA_ASQS(aqa) + 1);
7275 
7276     nvme_set_timestamp(n, 0ULL);
7277 
7278     nvme_select_iocs(n);
7279 
7280     return 0;
7281 }
7282 
7283 static void nvme_cmb_enable_regs(NvmeCtrl *n)
7284 {
7285     uint32_t cmbloc = ldl_le_p(&n->bar.cmbloc);
7286     uint32_t cmbsz = ldl_le_p(&n->bar.cmbsz);
7287 
7288     NVME_CMBLOC_SET_CDPCILS(cmbloc, 1);
7289     NVME_CMBLOC_SET_CDPMLS(cmbloc, 1);
7290     NVME_CMBLOC_SET_BIR(cmbloc, NVME_CMB_BIR);
7291     stl_le_p(&n->bar.cmbloc, cmbloc);
7292 
7293     NVME_CMBSZ_SET_SQS(cmbsz, 1);
7294     NVME_CMBSZ_SET_CQS(cmbsz, 0);
7295     NVME_CMBSZ_SET_LISTS(cmbsz, 1);
7296     NVME_CMBSZ_SET_RDS(cmbsz, 1);
7297     NVME_CMBSZ_SET_WDS(cmbsz, 1);
7298     NVME_CMBSZ_SET_SZU(cmbsz, 2); /* MBs */
7299     NVME_CMBSZ_SET_SZ(cmbsz, n->params.cmb_size_mb);
7300     stl_le_p(&n->bar.cmbsz, cmbsz);
7301 }
7302 
7303 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
7304                            unsigned size)
7305 {
7306     PCIDevice *pci = PCI_DEVICE(n);
7307     uint64_t cap = ldq_le_p(&n->bar.cap);
7308     uint32_t cc = ldl_le_p(&n->bar.cc);
7309     uint32_t intms = ldl_le_p(&n->bar.intms);
7310     uint32_t csts = ldl_le_p(&n->bar.csts);
7311     uint32_t pmrsts = ldl_le_p(&n->bar.pmrsts);
7312 
7313     if (unlikely(offset & (sizeof(uint32_t) - 1))) {
7314         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
7315                        "MMIO write not 32-bit aligned,"
7316                        " offset=0x%"PRIx64"", offset);
7317         /* should be ignored, fall through for now */
7318     }
7319 
7320     if (unlikely(size < sizeof(uint32_t))) {
7321         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
7322                        "MMIO write smaller than 32-bits,"
7323                        " offset=0x%"PRIx64", size=%u",
7324                        offset, size);
7325         /* should be ignored, fall through for now */
7326     }
7327 
7328     switch (offset) {
7329     case NVME_REG_INTMS:
7330         if (unlikely(msix_enabled(pci))) {
7331             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7332                            "undefined access to interrupt mask set"
7333                            " when MSI-X is enabled");
7334             /* should be ignored, fall through for now */
7335         }
7336         intms |= data;
7337         stl_le_p(&n->bar.intms, intms);
7338         n->bar.intmc = n->bar.intms;
7339         trace_pci_nvme_mmio_intm_set(data & 0xffffffff, intms);
7340         nvme_irq_check(n);
7341         break;
7342     case NVME_REG_INTMC:
7343         if (unlikely(msix_enabled(pci))) {
7344             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7345                            "undefined access to interrupt mask clr"
7346                            " when MSI-X is enabled");
7347             /* should be ignored, fall through for now */
7348         }
7349         intms &= ~data;
7350         stl_le_p(&n->bar.intms, intms);
7351         n->bar.intmc = n->bar.intms;
7352         trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, intms);
7353         nvme_irq_check(n);
7354         break;
7355     case NVME_REG_CC:
7356         stl_le_p(&n->bar.cc, data);
7357 
7358         trace_pci_nvme_mmio_cfg(data & 0xffffffff);
7359 
7360         if (NVME_CC_SHN(data) && !(NVME_CC_SHN(cc))) {
7361             trace_pci_nvme_mmio_shutdown_set();
7362             nvme_ctrl_shutdown(n);
7363             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7364             csts |= NVME_CSTS_SHST_COMPLETE;
7365         } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(cc)) {
7366             trace_pci_nvme_mmio_shutdown_cleared();
7367             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7368         }
7369 
7370         if (NVME_CC_EN(data) && !NVME_CC_EN(cc)) {
7371             if (unlikely(nvme_start_ctrl(n))) {
7372                 trace_pci_nvme_err_startfail();
7373                 csts = NVME_CSTS_FAILED;
7374             } else {
7375                 trace_pci_nvme_mmio_start_success();
7376                 csts = NVME_CSTS_READY;
7377             }
7378         } else if (!NVME_CC_EN(data) && NVME_CC_EN(cc)) {
7379             trace_pci_nvme_mmio_stopped();
7380             nvme_ctrl_reset(n, NVME_RESET_CONTROLLER);
7381 
7382             break;
7383         }
7384 
7385         stl_le_p(&n->bar.csts, csts);
7386 
7387         break;
7388     case NVME_REG_CSTS:
7389         if (data & (1 << 4)) {
7390             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
7391                            "attempted to W1C CSTS.NSSRO"
7392                            " but CAP.NSSRS is zero (not supported)");
7393         } else if (data != 0) {
7394             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
7395                            "attempted to set a read only bit"
7396                            " of controller status");
7397         }
7398         break;
7399     case NVME_REG_NSSR:
7400         if (data == 0x4e564d65) {
7401             trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
7402         } else {
7403             /* The spec says that writes of other values have no effect */
7404             return;
7405         }
7406         break;
7407     case NVME_REG_AQA:
7408         stl_le_p(&n->bar.aqa, data);
7409         trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
7410         break;
7411     case NVME_REG_ASQ:
7412         stn_le_p(&n->bar.asq, size, data);
7413         trace_pci_nvme_mmio_asqaddr(data);
7414         break;
7415     case NVME_REG_ASQ + 4:
7416         stl_le_p((uint8_t *)&n->bar.asq + 4, data);
7417         trace_pci_nvme_mmio_asqaddr_hi(data, ldq_le_p(&n->bar.asq));
7418         break;
7419     case NVME_REG_ACQ:
7420         trace_pci_nvme_mmio_acqaddr(data);
7421         stn_le_p(&n->bar.acq, size, data);
7422         break;
7423     case NVME_REG_ACQ + 4:
7424         stl_le_p((uint8_t *)&n->bar.acq + 4, data);
7425         trace_pci_nvme_mmio_acqaddr_hi(data, ldq_le_p(&n->bar.acq));
7426         break;
7427     case NVME_REG_CMBLOC:
7428         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
7429                        "invalid write to reserved CMBLOC"
7430                        " when CMBSZ is zero, ignored");
7431         return;
7432     case NVME_REG_CMBSZ:
7433         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
7434                        "invalid write to read only CMBSZ, ignored");
7435         return;
7436     case NVME_REG_CMBMSC:
7437         if (!NVME_CAP_CMBS(cap)) {
7438             return;
7439         }
7440 
7441         stn_le_p(&n->bar.cmbmsc, size, data);
7442         n->cmb.cmse = false;
7443 
7444         if (NVME_CMBMSC_CRE(data)) {
7445             nvme_cmb_enable_regs(n);
7446 
7447             if (NVME_CMBMSC_CMSE(data)) {
7448                 uint64_t cmbmsc = ldq_le_p(&n->bar.cmbmsc);
7449                 hwaddr cba = NVME_CMBMSC_CBA(cmbmsc) << CMBMSC_CBA_SHIFT;
7450                 if (cba + int128_get64(n->cmb.mem.size) < cba) {
7451                     uint32_t cmbsts = ldl_le_p(&n->bar.cmbsts);
7452                     NVME_CMBSTS_SET_CBAI(cmbsts, 1);
7453                     stl_le_p(&n->bar.cmbsts, cmbsts);
7454                     return;
7455                 }
7456 
7457                 n->cmb.cba = cba;
7458                 n->cmb.cmse = true;
7459             }
7460         } else {
7461             n->bar.cmbsz = 0;
7462             n->bar.cmbloc = 0;
7463         }
7464 
7465         return;
7466     case NVME_REG_CMBMSC + 4:
7467         stl_le_p((uint8_t *)&n->bar.cmbmsc + 4, data);
7468         return;
7469 
7470     case NVME_REG_PMRCAP:
7471         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
7472                        "invalid write to PMRCAP register, ignored");
7473         return;
7474     case NVME_REG_PMRCTL:
7475         if (!NVME_CAP_PMRS(cap)) {
7476             return;
7477         }
7478 
7479         stl_le_p(&n->bar.pmrctl, data);
7480         if (NVME_PMRCTL_EN(data)) {
7481             memory_region_set_enabled(&n->pmr.dev->mr, true);
7482             pmrsts = 0;
7483         } else {
7484             memory_region_set_enabled(&n->pmr.dev->mr, false);
7485             NVME_PMRSTS_SET_NRDY(pmrsts, 1);
7486             n->pmr.cmse = false;
7487         }
7488         stl_le_p(&n->bar.pmrsts, pmrsts);
7489         return;
7490     case NVME_REG_PMRSTS:
7491         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
7492                        "invalid write to PMRSTS register, ignored");
7493         return;
7494     case NVME_REG_PMREBS:
7495         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
7496                        "invalid write to PMREBS register, ignored");
7497         return;
7498     case NVME_REG_PMRSWTP:
7499         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
7500                        "invalid write to PMRSWTP register, ignored");
7501         return;
7502     case NVME_REG_PMRMSCL:
7503         if (!NVME_CAP_PMRS(cap)) {
7504             return;
7505         }
7506 
7507         stl_le_p(&n->bar.pmrmscl, data);
7508         n->pmr.cmse = false;
7509 
7510         if (NVME_PMRMSCL_CMSE(data)) {
7511             uint64_t pmrmscu = ldl_le_p(&n->bar.pmrmscu);
7512             hwaddr cba = pmrmscu << 32 |
7513                 (NVME_PMRMSCL_CBA(data) << PMRMSCL_CBA_SHIFT);
7514             if (cba + int128_get64(n->pmr.dev->mr.size) < cba) {
7515                 NVME_PMRSTS_SET_CBAI(pmrsts, 1);
7516                 stl_le_p(&n->bar.pmrsts, pmrsts);
7517                 return;
7518             }
7519 
7520             n->pmr.cmse = true;
7521             n->pmr.cba = cba;
7522         }
7523 
7524         return;
7525     case NVME_REG_PMRMSCU:
7526         if (!NVME_CAP_PMRS(cap)) {
7527             return;
7528         }
7529 
7530         stl_le_p(&n->bar.pmrmscu, data);
7531         return;
7532     default:
7533         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
7534                        "invalid MMIO write,"
7535                        " offset=0x%"PRIx64", data=%"PRIx64"",
7536                        offset, data);
7537         break;
7538     }
7539 }
7540 
7541 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
7542 {
7543     NvmeCtrl *n = (NvmeCtrl *)opaque;
7544     uint8_t *ptr = (uint8_t *)&n->bar;
7545 
7546     trace_pci_nvme_mmio_read(addr, size);
7547 
7548     if (unlikely(addr & (sizeof(uint32_t) - 1))) {
7549         NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
7550                        "MMIO read not 32-bit aligned,"
7551                        " offset=0x%"PRIx64"", addr);
7552         /* should RAZ, fall through for now */
7553     } else if (unlikely(size < sizeof(uint32_t))) {
7554         NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
7555                        "MMIO read smaller than 32-bits,"
7556                        " offset=0x%"PRIx64"", addr);
7557         /* should RAZ, fall through for now */
7558     }
7559 
7560     if (addr > sizeof(n->bar) - size) {
7561         NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
7562                        "MMIO read beyond last register,"
7563                        " offset=0x%"PRIx64", returning 0", addr);
7564 
7565         return 0;
7566     }
7567 
7568     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7569         addr != NVME_REG_CSTS) {
7570         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7571         return 0;
7572     }
7573 
7574     /*
7575      * When PMRWBM bit 1 is set then read from
7576      * from PMRSTS should ensure prior writes
7577      * made it to persistent media
7578      */
7579     if (addr == NVME_REG_PMRSTS &&
7580         (NVME_PMRCAP_PMRWBM(ldl_le_p(&n->bar.pmrcap)) & 0x02)) {
7581         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7582     }
7583 
7584     return ldn_le_p(ptr + addr, size);
7585 }
7586 
7587 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
7588 {
7589     PCIDevice *pci = PCI_DEVICE(n);
7590     uint32_t qid;
7591 
7592     if (unlikely(addr & ((1 << 2) - 1))) {
7593         NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
7594                        "doorbell write not 32-bit aligned,"
7595                        " offset=0x%"PRIx64", ignoring", addr);
7596         return;
7597     }
7598 
7599     if (((addr - 0x1000) >> 2) & 1) {
7600         /* Completion queue doorbell write */
7601 
7602         uint16_t new_head = val & 0xffff;
7603         int start_sqs;
7604         NvmeCQueue *cq;
7605 
7606         qid = (addr - (0x1000 + (1 << 2))) >> 3;
7607         if (unlikely(nvme_check_cqid(n, qid))) {
7608             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
7609                            "completion queue doorbell write"
7610                            " for nonexistent queue,"
7611                            " sqid=%"PRIu32", ignoring", qid);
7612 
7613             /*
7614              * NVM Express v1.3d, Section 4.1 state: "If host software writes
7615              * an invalid value to the Submission Queue Tail Doorbell or
7616              * Completion Queue Head Doorbell regiter and an Asynchronous Event
7617              * Request command is outstanding, then an asynchronous event is
7618              * posted to the Admin Completion Queue with a status code of
7619              * Invalid Doorbell Write Value."
7620              *
7621              * Also note that the spec includes the "Invalid Doorbell Register"
7622              * status code, but nowhere does it specify when to use it.
7623              * However, it seems reasonable to use it here in a similar
7624              * fashion.
7625              */
7626             if (n->outstanding_aers) {
7627                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7628                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7629                                    NVME_LOG_ERROR_INFO);
7630             }
7631 
7632             return;
7633         }
7634 
7635         cq = n->cq[qid];
7636         if (unlikely(new_head >= cq->size)) {
7637             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
7638                            "completion queue doorbell write value"
7639                            " beyond queue size, sqid=%"PRIu32","
7640                            " new_head=%"PRIu16", ignoring",
7641                            qid, new_head);
7642 
7643             if (n->outstanding_aers) {
7644                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7645                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7646                                    NVME_LOG_ERROR_INFO);
7647             }
7648 
7649             return;
7650         }
7651 
7652         trace_pci_nvme_mmio_doorbell_cq(cq->cqid, new_head);
7653 
7654         start_sqs = nvme_cq_full(cq) ? 1 : 0;
7655         cq->head = new_head;
7656         if (!qid && n->dbbuf_enabled) {
7657             pci_dma_write(pci, cq->db_addr, &cq->head, sizeof(cq->head));
7658         }
7659         if (start_sqs) {
7660             NvmeSQueue *sq;
7661             QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
7662                 qemu_bh_schedule(sq->bh);
7663             }
7664             qemu_bh_schedule(cq->bh);
7665         }
7666 
7667         if (cq->tail == cq->head) {
7668             if (cq->irq_enabled) {
7669                 n->cq_pending--;
7670             }
7671 
7672             nvme_irq_deassert(n, cq);
7673         }
7674     } else {
7675         /* Submission queue doorbell write */
7676 
7677         uint16_t new_tail = val & 0xffff;
7678         NvmeSQueue *sq;
7679 
7680         qid = (addr - 0x1000) >> 3;
7681         if (unlikely(nvme_check_sqid(n, qid))) {
7682             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
7683                            "submission queue doorbell write"
7684                            " for nonexistent queue,"
7685                            " sqid=%"PRIu32", ignoring", qid);
7686 
7687             if (n->outstanding_aers) {
7688                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7689                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7690                                    NVME_LOG_ERROR_INFO);
7691             }
7692 
7693             return;
7694         }
7695 
7696         sq = n->sq[qid];
7697         if (unlikely(new_tail >= sq->size)) {
7698             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
7699                            "submission queue doorbell write value"
7700                            " beyond queue size, sqid=%"PRIu32","
7701                            " new_tail=%"PRIu16", ignoring",
7702                            qid, new_tail);
7703 
7704             if (n->outstanding_aers) {
7705                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7706                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7707                                    NVME_LOG_ERROR_INFO);
7708             }
7709 
7710             return;
7711         }
7712 
7713         trace_pci_nvme_mmio_doorbell_sq(sq->sqid, new_tail);
7714 
7715         sq->tail = new_tail;
7716         if (!qid && n->dbbuf_enabled) {
7717             /*
7718              * The spec states "the host shall also update the controller's
7719              * corresponding doorbell property to match the value of that entry
7720              * in the Shadow Doorbell buffer."
7721              *
7722              * Since this context is currently a VM trap, we can safely enforce
7723              * the requirement from the device side in case the host is
7724              * misbehaving.
7725              *
7726              * Note, we shouldn't have to do this, but various drivers
7727              * including ones that run on Linux, are not updating Admin Queues,
7728              * so we can't trust reading it for an appropriate sq tail.
7729              */
7730             pci_dma_write(pci, sq->db_addr, &sq->tail, sizeof(sq->tail));
7731         }
7732 
7733         qemu_bh_schedule(sq->bh);
7734     }
7735 }
7736 
7737 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
7738                             unsigned size)
7739 {
7740     NvmeCtrl *n = (NvmeCtrl *)opaque;
7741 
7742     trace_pci_nvme_mmio_write(addr, data, size);
7743 
7744     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7745         addr != NVME_REG_CSTS) {
7746         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7747         return;
7748     }
7749 
7750     if (addr < sizeof(n->bar)) {
7751         nvme_write_bar(n, addr, data, size);
7752     } else {
7753         nvme_process_db(n, addr, data);
7754     }
7755 }
7756 
7757 static const MemoryRegionOps nvme_mmio_ops = {
7758     .read = nvme_mmio_read,
7759     .write = nvme_mmio_write,
7760     .endianness = DEVICE_LITTLE_ENDIAN,
7761     .impl = {
7762         .min_access_size = 2,
7763         .max_access_size = 8,
7764     },
7765 };
7766 
7767 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
7768                            unsigned size)
7769 {
7770     NvmeCtrl *n = (NvmeCtrl *)opaque;
7771     stn_le_p(&n->cmb.buf[addr], size, data);
7772 }
7773 
7774 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
7775 {
7776     NvmeCtrl *n = (NvmeCtrl *)opaque;
7777     return ldn_le_p(&n->cmb.buf[addr], size);
7778 }
7779 
7780 static const MemoryRegionOps nvme_cmb_ops = {
7781     .read = nvme_cmb_read,
7782     .write = nvme_cmb_write,
7783     .endianness = DEVICE_LITTLE_ENDIAN,
7784     .impl = {
7785         .min_access_size = 1,
7786         .max_access_size = 8,
7787     },
7788 };
7789 
7790 static bool nvme_check_params(NvmeCtrl *n, Error **errp)
7791 {
7792     NvmeParams *params = &n->params;
7793 
7794     if (params->num_queues) {
7795         warn_report("num_queues is deprecated; please use max_ioqpairs "
7796                     "instead");
7797 
7798         params->max_ioqpairs = params->num_queues - 1;
7799     }
7800 
7801     if (n->namespace.blkconf.blk && n->subsys) {
7802         error_setg(errp, "subsystem support is unavailable with legacy "
7803                    "namespace ('drive' property)");
7804         return false;
7805     }
7806 
7807     if (params->max_ioqpairs < 1 ||
7808         params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
7809         error_setg(errp, "max_ioqpairs must be between 1 and %d",
7810                    NVME_MAX_IOQPAIRS);
7811         return false;
7812     }
7813 
7814     if (params->msix_qsize < 1 ||
7815         params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
7816         error_setg(errp, "msix_qsize must be between 1 and %d",
7817                    PCI_MSIX_FLAGS_QSIZE + 1);
7818         return false;
7819     }
7820 
7821     if (!params->serial) {
7822         error_setg(errp, "serial property not set");
7823         return false;
7824     }
7825 
7826     if (n->pmr.dev) {
7827         if (host_memory_backend_is_mapped(n->pmr.dev)) {
7828             error_setg(errp, "can't use already busy memdev: %s",
7829                        object_get_canonical_path_component(OBJECT(n->pmr.dev)));
7830             return false;
7831         }
7832 
7833         if (!is_power_of_2(n->pmr.dev->size)) {
7834             error_setg(errp, "pmr backend size needs to be power of 2 in size");
7835             return false;
7836         }
7837 
7838         host_memory_backend_set_mapped(n->pmr.dev, true);
7839     }
7840 
7841     if (n->params.zasl > n->params.mdts) {
7842         error_setg(errp, "zoned.zasl (Zone Append Size Limit) must be less "
7843                    "than or equal to mdts (Maximum Data Transfer Size)");
7844         return false;
7845     }
7846 
7847     if (!n->params.vsl) {
7848         error_setg(errp, "vsl must be non-zero");
7849         return false;
7850     }
7851 
7852     if (params->sriov_max_vfs) {
7853         if (!n->subsys) {
7854             error_setg(errp, "subsystem is required for the use of SR-IOV");
7855             return false;
7856         }
7857 
7858         if (params->sriov_max_vfs > NVME_MAX_VFS) {
7859             error_setg(errp, "sriov_max_vfs must be between 0 and %d",
7860                        NVME_MAX_VFS);
7861             return false;
7862         }
7863 
7864         if (params->cmb_size_mb) {
7865             error_setg(errp, "CMB is not supported with SR-IOV");
7866             return false;
7867         }
7868 
7869         if (n->pmr.dev) {
7870             error_setg(errp, "PMR is not supported with SR-IOV");
7871             return false;
7872         }
7873 
7874         if (!params->sriov_vq_flexible || !params->sriov_vi_flexible) {
7875             error_setg(errp, "both sriov_vq_flexible and sriov_vi_flexible"
7876                        " must be set for the use of SR-IOV");
7877             return false;
7878         }
7879 
7880         if (params->sriov_vq_flexible < params->sriov_max_vfs * 2) {
7881             error_setg(errp, "sriov_vq_flexible must be greater than or equal"
7882                        " to %d (sriov_max_vfs * 2)", params->sriov_max_vfs * 2);
7883             return false;
7884         }
7885 
7886         if (params->max_ioqpairs < params->sriov_vq_flexible + 2) {
7887             error_setg(errp, "(max_ioqpairs - sriov_vq_flexible) must be"
7888                        " greater than or equal to 2");
7889             return false;
7890         }
7891 
7892         if (params->sriov_vi_flexible < params->sriov_max_vfs) {
7893             error_setg(errp, "sriov_vi_flexible must be greater than or equal"
7894                        " to %d (sriov_max_vfs)", params->sriov_max_vfs);
7895             return false;
7896         }
7897 
7898         if (params->msix_qsize < params->sriov_vi_flexible + 1) {
7899             error_setg(errp, "(msix_qsize - sriov_vi_flexible) must be"
7900                        " greater than or equal to 1");
7901             return false;
7902         }
7903 
7904         if (params->sriov_max_vi_per_vf &&
7905             (params->sriov_max_vi_per_vf - 1) % NVME_VF_RES_GRANULARITY) {
7906             error_setg(errp, "sriov_max_vi_per_vf must meet:"
7907                        " (sriov_max_vi_per_vf - 1) %% %d == 0 and"
7908                        " sriov_max_vi_per_vf >= 1", NVME_VF_RES_GRANULARITY);
7909             return false;
7910         }
7911 
7912         if (params->sriov_max_vq_per_vf &&
7913             (params->sriov_max_vq_per_vf < 2 ||
7914              (params->sriov_max_vq_per_vf - 1) % NVME_VF_RES_GRANULARITY)) {
7915             error_setg(errp, "sriov_max_vq_per_vf must meet:"
7916                        " (sriov_max_vq_per_vf - 1) %% %d == 0 and"
7917                        " sriov_max_vq_per_vf >= 2", NVME_VF_RES_GRANULARITY);
7918             return false;
7919         }
7920     }
7921 
7922     return true;
7923 }
7924 
7925 static void nvme_init_state(NvmeCtrl *n)
7926 {
7927     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7928     NvmeSecCtrlList *list = &n->sec_ctrl_list;
7929     NvmeSecCtrlEntry *sctrl;
7930     PCIDevice *pci = PCI_DEVICE(n);
7931     uint8_t max_vfs;
7932     int i;
7933 
7934     if (pci_is_vf(pci)) {
7935         sctrl = nvme_sctrl(n);
7936         max_vfs = 0;
7937         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7938         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7939     } else {
7940         max_vfs = n->params.sriov_max_vfs;
7941         n->conf_ioqpairs = n->params.max_ioqpairs;
7942         n->conf_msix_qsize = n->params.msix_qsize;
7943     }
7944 
7945     n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
7946     n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
7947     n->temperature = NVME_TEMPERATURE;
7948     n->features.temp_thresh_hi = NVME_TEMPERATURE_WARNING;
7949     n->starttime_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
7950     n->aer_reqs = g_new0(NvmeRequest *, n->params.aerl + 1);
7951     QTAILQ_INIT(&n->aer_queue);
7952 
7953     list->numcntl = cpu_to_le16(max_vfs);
7954     for (i = 0; i < max_vfs; i++) {
7955         sctrl = &list->sec[i];
7956         sctrl->pcid = cpu_to_le16(n->cntlid);
7957         sctrl->vfn = cpu_to_le16(i + 1);
7958     }
7959 
7960     cap->cntlid = cpu_to_le16(n->cntlid);
7961     cap->crt = NVME_CRT_VQ | NVME_CRT_VI;
7962 
7963     if (pci_is_vf(pci)) {
7964         cap->vqprt = cpu_to_le16(1 + n->conf_ioqpairs);
7965     } else {
7966         cap->vqprt = cpu_to_le16(1 + n->params.max_ioqpairs -
7967                                  n->params.sriov_vq_flexible);
7968         cap->vqfrt = cpu_to_le32(n->params.sriov_vq_flexible);
7969         cap->vqrfap = cap->vqfrt;
7970         cap->vqgran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
7971         cap->vqfrsm = n->params.sriov_max_vq_per_vf ?
7972                         cpu_to_le16(n->params.sriov_max_vq_per_vf) :
7973                         cap->vqfrt / MAX(max_vfs, 1);
7974     }
7975 
7976     if (pci_is_vf(pci)) {
7977         cap->viprt = cpu_to_le16(n->conf_msix_qsize);
7978     } else {
7979         cap->viprt = cpu_to_le16(n->params.msix_qsize -
7980                                  n->params.sriov_vi_flexible);
7981         cap->vifrt = cpu_to_le32(n->params.sriov_vi_flexible);
7982         cap->virfap = cap->vifrt;
7983         cap->vigran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
7984         cap->vifrsm = n->params.sriov_max_vi_per_vf ?
7985                         cpu_to_le16(n->params.sriov_max_vi_per_vf) :
7986                         cap->vifrt / MAX(max_vfs, 1);
7987     }
7988 }
7989 
7990 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
7991 {
7992     uint64_t cmb_size = n->params.cmb_size_mb * MiB;
7993     uint64_t cap = ldq_le_p(&n->bar.cap);
7994 
7995     n->cmb.buf = g_malloc0(cmb_size);
7996     memory_region_init_io(&n->cmb.mem, OBJECT(n), &nvme_cmb_ops, n,
7997                           "nvme-cmb", cmb_size);
7998     pci_register_bar(pci_dev, NVME_CMB_BIR,
7999                      PCI_BASE_ADDRESS_SPACE_MEMORY |
8000                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
8001                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->cmb.mem);
8002 
8003     NVME_CAP_SET_CMBS(cap, 1);
8004     stq_le_p(&n->bar.cap, cap);
8005 
8006     if (n->params.legacy_cmb) {
8007         nvme_cmb_enable_regs(n);
8008         n->cmb.cmse = true;
8009     }
8010 }
8011 
8012 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
8013 {
8014     uint32_t pmrcap = ldl_le_p(&n->bar.pmrcap);
8015 
8016     NVME_PMRCAP_SET_RDS(pmrcap, 1);
8017     NVME_PMRCAP_SET_WDS(pmrcap, 1);
8018     NVME_PMRCAP_SET_BIR(pmrcap, NVME_PMR_BIR);
8019     /* Turn on bit 1 support */
8020     NVME_PMRCAP_SET_PMRWBM(pmrcap, 0x02);
8021     NVME_PMRCAP_SET_CMSS(pmrcap, 1);
8022     stl_le_p(&n->bar.pmrcap, pmrcap);
8023 
8024     pci_register_bar(pci_dev, NVME_PMR_BIR,
8025                      PCI_BASE_ADDRESS_SPACE_MEMORY |
8026                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
8027                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmr.dev->mr);
8028 
8029     memory_region_set_enabled(&n->pmr.dev->mr, false);
8030 }
8031 
8032 static uint64_t nvme_bar_size(unsigned total_queues, unsigned total_irqs,
8033                               unsigned *msix_table_offset,
8034                               unsigned *msix_pba_offset)
8035 {
8036     uint64_t bar_size, msix_table_size, msix_pba_size;
8037 
8038     bar_size = sizeof(NvmeBar) + 2 * total_queues * NVME_DB_SIZE;
8039     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8040 
8041     if (msix_table_offset) {
8042         *msix_table_offset = bar_size;
8043     }
8044 
8045     msix_table_size = PCI_MSIX_ENTRY_SIZE * total_irqs;
8046     bar_size += msix_table_size;
8047     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8048 
8049     if (msix_pba_offset) {
8050         *msix_pba_offset = bar_size;
8051     }
8052 
8053     msix_pba_size = QEMU_ALIGN_UP(total_irqs, 64) / 8;
8054     bar_size += msix_pba_size;
8055 
8056     bar_size = pow2ceil(bar_size);
8057     return bar_size;
8058 }
8059 
8060 static void nvme_init_sriov(NvmeCtrl *n, PCIDevice *pci_dev, uint16_t offset)
8061 {
8062     uint16_t vf_dev_id = n->params.use_intel_id ?
8063                          PCI_DEVICE_ID_INTEL_NVME : PCI_DEVICE_ID_REDHAT_NVME;
8064     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
8065     uint64_t bar_size = nvme_bar_size(le16_to_cpu(cap->vqfrsm),
8066                                       le16_to_cpu(cap->vifrsm),
8067                                       NULL, NULL);
8068 
8069     pcie_sriov_pf_init(pci_dev, offset, "nvme", vf_dev_id,
8070                        n->params.sriov_max_vfs, n->params.sriov_max_vfs,
8071                        NVME_VF_OFFSET, NVME_VF_STRIDE);
8072 
8073     pcie_sriov_pf_init_vf_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8074                               PCI_BASE_ADDRESS_MEM_TYPE_64, bar_size);
8075 }
8076 
8077 static int nvme_add_pm_capability(PCIDevice *pci_dev, uint8_t offset)
8078 {
8079     Error *err = NULL;
8080     int ret;
8081 
8082     ret = pci_add_capability(pci_dev, PCI_CAP_ID_PM, offset,
8083                              PCI_PM_SIZEOF, &err);
8084     if (err) {
8085         error_report_err(err);
8086         return ret;
8087     }
8088 
8089     pci_set_word(pci_dev->config + offset + PCI_PM_PMC,
8090                  PCI_PM_CAP_VER_1_2);
8091     pci_set_word(pci_dev->config + offset + PCI_PM_CTRL,
8092                  PCI_PM_CTRL_NO_SOFT_RESET);
8093     pci_set_word(pci_dev->wmask + offset + PCI_PM_CTRL,
8094                  PCI_PM_CTRL_STATE_MASK);
8095 
8096     return 0;
8097 }
8098 
8099 static bool nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
8100 {
8101     ERRP_GUARD();
8102     uint8_t *pci_conf = pci_dev->config;
8103     uint64_t bar_size;
8104     unsigned msix_table_offset, msix_pba_offset;
8105     int ret;
8106 
8107     pci_conf[PCI_INTERRUPT_PIN] = 1;
8108     pci_config_set_prog_interface(pci_conf, 0x2);
8109 
8110     if (n->params.use_intel_id) {
8111         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_INTEL);
8112         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_INTEL_NVME);
8113     } else {
8114         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_REDHAT);
8115         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_REDHAT_NVME);
8116     }
8117 
8118     pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
8119     nvme_add_pm_capability(pci_dev, 0x60);
8120     pcie_endpoint_cap_init(pci_dev, 0x80);
8121     pcie_cap_flr_init(pci_dev);
8122     if (n->params.sriov_max_vfs) {
8123         pcie_ari_init(pci_dev, 0x100, 1);
8124     }
8125 
8126     /* add one to max_ioqpairs to account for the admin queue pair */
8127     bar_size = nvme_bar_size(n->params.max_ioqpairs + 1, n->params.msix_qsize,
8128                              &msix_table_offset, &msix_pba_offset);
8129 
8130     memory_region_init(&n->bar0, OBJECT(n), "nvme-bar0", bar_size);
8131     memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
8132                           msix_table_offset);
8133     memory_region_add_subregion(&n->bar0, 0, &n->iomem);
8134 
8135     if (pci_is_vf(pci_dev)) {
8136         pcie_sriov_vf_register_bar(pci_dev, 0, &n->bar0);
8137     } else {
8138         pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8139                          PCI_BASE_ADDRESS_MEM_TYPE_64, &n->bar0);
8140     }
8141     ret = msix_init(pci_dev, n->params.msix_qsize,
8142                     &n->bar0, 0, msix_table_offset,
8143                     &n->bar0, 0, msix_pba_offset, 0, errp);
8144     if (ret == -ENOTSUP) {
8145         /* report that msix is not supported, but do not error out */
8146         warn_report_err(*errp);
8147         *errp = NULL;
8148     } else if (ret < 0) {
8149         /* propagate error to caller */
8150         return false;
8151     }
8152 
8153     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
8154 
8155     if (n->params.cmb_size_mb) {
8156         nvme_init_cmb(n, pci_dev);
8157     }
8158 
8159     if (n->pmr.dev) {
8160         nvme_init_pmr(n, pci_dev);
8161     }
8162 
8163     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8164         nvme_init_sriov(n, pci_dev, 0x120);
8165     }
8166 
8167     return true;
8168 }
8169 
8170 static void nvme_init_subnqn(NvmeCtrl *n)
8171 {
8172     NvmeSubsystem *subsys = n->subsys;
8173     NvmeIdCtrl *id = &n->id_ctrl;
8174 
8175     if (!subsys) {
8176         snprintf((char *)id->subnqn, sizeof(id->subnqn),
8177                  "nqn.2019-08.org.qemu:%s", n->params.serial);
8178     } else {
8179         pstrcpy((char *)id->subnqn, sizeof(id->subnqn), (char*)subsys->subnqn);
8180     }
8181 }
8182 
8183 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
8184 {
8185     NvmeIdCtrl *id = &n->id_ctrl;
8186     uint8_t *pci_conf = pci_dev->config;
8187     uint64_t cap = ldq_le_p(&n->bar.cap);
8188     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
8189     uint32_t ctratt;
8190 
8191     id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
8192     id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
8193     strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
8194     strpadcpy((char *)id->fr, sizeof(id->fr), QEMU_VERSION, ' ');
8195     strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
8196 
8197     id->cntlid = cpu_to_le16(n->cntlid);
8198 
8199     id->oaes = cpu_to_le32(NVME_OAES_NS_ATTR);
8200     ctratt = NVME_CTRATT_ELBAS;
8201 
8202     id->rab = 6;
8203 
8204     if (n->params.use_intel_id) {
8205         id->ieee[0] = 0xb3;
8206         id->ieee[1] = 0x02;
8207         id->ieee[2] = 0x00;
8208     } else {
8209         id->ieee[0] = 0x00;
8210         id->ieee[1] = 0x54;
8211         id->ieee[2] = 0x52;
8212     }
8213 
8214     id->mdts = n->params.mdts;
8215     id->ver = cpu_to_le32(NVME_SPEC_VER);
8216     id->oacs =
8217         cpu_to_le16(NVME_OACS_NS_MGMT | NVME_OACS_FORMAT | NVME_OACS_DBBUF |
8218                     NVME_OACS_DIRECTIVES);
8219     id->cntrltype = 0x1;
8220 
8221     /*
8222      * Because the controller always completes the Abort command immediately,
8223      * there can never be more than one concurrently executing Abort command,
8224      * so this value is never used for anything. Note that there can easily be
8225      * many Abort commands in the queues, but they are not considered
8226      * "executing" until processed by nvme_abort.
8227      *
8228      * The specification recommends a value of 3 for Abort Command Limit (four
8229      * concurrently outstanding Abort commands), so lets use that though it is
8230      * inconsequential.
8231      */
8232     id->acl = 3;
8233     id->aerl = n->params.aerl;
8234     id->frmw = (NVME_NUM_FW_SLOTS << 1) | NVME_FRMW_SLOT1_RO;
8235     id->lpa = NVME_LPA_NS_SMART | NVME_LPA_CSE | NVME_LPA_EXTENDED;
8236 
8237     /* recommended default value (~70 C) */
8238     id->wctemp = cpu_to_le16(NVME_TEMPERATURE_WARNING);
8239     id->cctemp = cpu_to_le16(NVME_TEMPERATURE_CRITICAL);
8240 
8241     id->sqes = (0x6 << 4) | 0x6;
8242     id->cqes = (0x4 << 4) | 0x4;
8243     id->nn = cpu_to_le32(NVME_MAX_NAMESPACES);
8244     id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP |
8245                            NVME_ONCS_FEATURES | NVME_ONCS_DSM |
8246                            NVME_ONCS_COMPARE | NVME_ONCS_COPY);
8247 
8248     /*
8249      * NOTE: If this device ever supports a command set that does NOT use 0x0
8250      * as a Flush-equivalent operation, support for the broadcast NSID in Flush
8251      * should probably be removed.
8252      *
8253      * See comment in nvme_io_cmd.
8254      */
8255     id->vwc = NVME_VWC_NSID_BROADCAST_SUPPORT | NVME_VWC_PRESENT;
8256 
8257     id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0 | NVME_OCFS_COPY_FORMAT_1);
8258     id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN);
8259 
8260     nvme_init_subnqn(n);
8261 
8262     id->psd[0].mp = cpu_to_le16(0x9c4);
8263     id->psd[0].enlat = cpu_to_le32(0x10);
8264     id->psd[0].exlat = cpu_to_le32(0x4);
8265 
8266     if (n->subsys) {
8267         id->cmic |= NVME_CMIC_MULTI_CTRL;
8268         ctratt |= NVME_CTRATT_ENDGRPS;
8269 
8270         id->endgidmax = cpu_to_le16(0x1);
8271 
8272         if (n->subsys->endgrp.fdp.enabled) {
8273             ctratt |= NVME_CTRATT_FDPS;
8274         }
8275     }
8276 
8277     id->ctratt = cpu_to_le32(ctratt);
8278 
8279     NVME_CAP_SET_MQES(cap, 0x7ff);
8280     NVME_CAP_SET_CQR(cap, 1);
8281     NVME_CAP_SET_TO(cap, 0xf);
8282     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_NVM);
8283     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_CSI_SUPP);
8284     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_ADMIN_ONLY);
8285     NVME_CAP_SET_MPSMAX(cap, 4);
8286     NVME_CAP_SET_CMBS(cap, n->params.cmb_size_mb ? 1 : 0);
8287     NVME_CAP_SET_PMRS(cap, n->pmr.dev ? 1 : 0);
8288     stq_le_p(&n->bar.cap, cap);
8289 
8290     stl_le_p(&n->bar.vs, NVME_SPEC_VER);
8291     n->bar.intmc = n->bar.intms = 0;
8292 
8293     if (pci_is_vf(pci_dev) && !sctrl->scs) {
8294         stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
8295     }
8296 }
8297 
8298 static int nvme_init_subsys(NvmeCtrl *n, Error **errp)
8299 {
8300     int cntlid;
8301 
8302     if (!n->subsys) {
8303         return 0;
8304     }
8305 
8306     cntlid = nvme_subsys_register_ctrl(n, errp);
8307     if (cntlid < 0) {
8308         return -1;
8309     }
8310 
8311     n->cntlid = cntlid;
8312 
8313     return 0;
8314 }
8315 
8316 void nvme_attach_ns(NvmeCtrl *n, NvmeNamespace *ns)
8317 {
8318     uint32_t nsid = ns->params.nsid;
8319     assert(nsid && nsid <= NVME_MAX_NAMESPACES);
8320 
8321     n->namespaces[nsid] = ns;
8322     ns->attached++;
8323 
8324     n->dmrsl = MIN_NON_ZERO(n->dmrsl,
8325                             BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
8326 }
8327 
8328 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
8329 {
8330     NvmeCtrl *n = NVME(pci_dev);
8331     DeviceState *dev = DEVICE(pci_dev);
8332     NvmeNamespace *ns;
8333     NvmeCtrl *pn = NVME(pcie_sriov_get_pf(pci_dev));
8334 
8335     if (pci_is_vf(pci_dev)) {
8336         /*
8337          * VFs derive settings from the parent. PF's lifespan exceeds
8338          * that of VF's, so it's safe to share params.serial.
8339          */
8340         memcpy(&n->params, &pn->params, sizeof(NvmeParams));
8341         n->subsys = pn->subsys;
8342     }
8343 
8344     if (!nvme_check_params(n, errp)) {
8345         return;
8346     }
8347 
8348     qbus_init(&n->bus, sizeof(NvmeBus), TYPE_NVME_BUS, dev, dev->id);
8349 
8350     if (nvme_init_subsys(n, errp)) {
8351         return;
8352     }
8353     nvme_init_state(n);
8354     if (!nvme_init_pci(n, pci_dev, errp)) {
8355         return;
8356     }
8357     nvme_init_ctrl(n, pci_dev);
8358 
8359     /* setup a namespace if the controller drive property was given */
8360     if (n->namespace.blkconf.blk) {
8361         ns = &n->namespace;
8362         ns->params.nsid = 1;
8363 
8364         if (nvme_ns_setup(ns, errp)) {
8365             return;
8366         }
8367 
8368         nvme_attach_ns(n, ns);
8369     }
8370 }
8371 
8372 static void nvme_exit(PCIDevice *pci_dev)
8373 {
8374     NvmeCtrl *n = NVME(pci_dev);
8375     NvmeNamespace *ns;
8376     int i;
8377 
8378     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8379 
8380     if (n->subsys) {
8381         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
8382             ns = nvme_ns(n, i);
8383             if (ns) {
8384                 ns->attached--;
8385             }
8386         }
8387 
8388         nvme_subsys_unregister_ctrl(n->subsys, n);
8389     }
8390 
8391     g_free(n->cq);
8392     g_free(n->sq);
8393     g_free(n->aer_reqs);
8394 
8395     if (n->params.cmb_size_mb) {
8396         g_free(n->cmb.buf);
8397     }
8398 
8399     if (n->pmr.dev) {
8400         host_memory_backend_set_mapped(n->pmr.dev, false);
8401     }
8402 
8403     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8404         pcie_sriov_pf_exit(pci_dev);
8405     }
8406 
8407     msix_uninit(pci_dev, &n->bar0, &n->bar0);
8408     memory_region_del_subregion(&n->bar0, &n->iomem);
8409 }
8410 
8411 static Property nvme_props[] = {
8412     DEFINE_BLOCK_PROPERTIES(NvmeCtrl, namespace.blkconf),
8413     DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmr.dev, TYPE_MEMORY_BACKEND,
8414                      HostMemoryBackend *),
8415     DEFINE_PROP_LINK("subsys", NvmeCtrl, subsys, TYPE_NVME_SUBSYS,
8416                      NvmeSubsystem *),
8417     DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
8418     DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
8419     DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
8420     DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
8421     DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
8422     DEFINE_PROP_UINT8("aerl", NvmeCtrl, params.aerl, 3),
8423     DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl, params.aer_max_queued, 64),
8424     DEFINE_PROP_UINT8("mdts", NvmeCtrl, params.mdts, 7),
8425     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
8426     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
8427     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
8428     DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl, params.ioeventfd, false),
8429     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
8430     DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl,
8431                      params.auto_transition_zones, true),
8432     DEFINE_PROP_UINT8("sriov_max_vfs", NvmeCtrl, params.sriov_max_vfs, 0),
8433     DEFINE_PROP_UINT16("sriov_vq_flexible", NvmeCtrl,
8434                        params.sriov_vq_flexible, 0),
8435     DEFINE_PROP_UINT16("sriov_vi_flexible", NvmeCtrl,
8436                        params.sriov_vi_flexible, 0),
8437     DEFINE_PROP_UINT8("sriov_max_vi_per_vf", NvmeCtrl,
8438                       params.sriov_max_vi_per_vf, 0),
8439     DEFINE_PROP_UINT8("sriov_max_vq_per_vf", NvmeCtrl,
8440                       params.sriov_max_vq_per_vf, 0),
8441     DEFINE_PROP_END_OF_LIST(),
8442 };
8443 
8444 static void nvme_get_smart_warning(Object *obj, Visitor *v, const char *name,
8445                                    void *opaque, Error **errp)
8446 {
8447     NvmeCtrl *n = NVME(obj);
8448     uint8_t value = n->smart_critical_warning;
8449 
8450     visit_type_uint8(v, name, &value, errp);
8451 }
8452 
8453 static void nvme_set_smart_warning(Object *obj, Visitor *v, const char *name,
8454                                    void *opaque, Error **errp)
8455 {
8456     NvmeCtrl *n = NVME(obj);
8457     uint8_t value, old_value, cap = 0, index, event;
8458 
8459     if (!visit_type_uint8(v, name, &value, errp)) {
8460         return;
8461     }
8462 
8463     cap = NVME_SMART_SPARE | NVME_SMART_TEMPERATURE | NVME_SMART_RELIABILITY
8464           | NVME_SMART_MEDIA_READ_ONLY | NVME_SMART_FAILED_VOLATILE_MEDIA;
8465     if (NVME_CAP_PMRS(ldq_le_p(&n->bar.cap))) {
8466         cap |= NVME_SMART_PMR_UNRELIABLE;
8467     }
8468 
8469     if ((value & cap) != value) {
8470         error_setg(errp, "unsupported smart critical warning bits: 0x%x",
8471                    value & ~cap);
8472         return;
8473     }
8474 
8475     old_value = n->smart_critical_warning;
8476     n->smart_critical_warning = value;
8477 
8478     /* only inject new bits of smart critical warning */
8479     for (index = 0; index < NVME_SMART_WARN_MAX; index++) {
8480         event = 1 << index;
8481         if (value & ~old_value & event)
8482             nvme_smart_event(n, event);
8483     }
8484 }
8485 
8486 static void nvme_pci_reset(DeviceState *qdev)
8487 {
8488     PCIDevice *pci_dev = PCI_DEVICE(qdev);
8489     NvmeCtrl *n = NVME(pci_dev);
8490 
8491     trace_pci_nvme_pci_reset();
8492     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8493 }
8494 
8495 static void nvme_sriov_pre_write_ctrl(PCIDevice *dev, uint32_t address,
8496                                       uint32_t val, int len)
8497 {
8498     NvmeCtrl *n = NVME(dev);
8499     NvmeSecCtrlEntry *sctrl;
8500     uint16_t sriov_cap = dev->exp.sriov_cap;
8501     uint32_t off = address - sriov_cap;
8502     int i, num_vfs;
8503 
8504     if (!sriov_cap) {
8505         return;
8506     }
8507 
8508     if (range_covers_byte(off, len, PCI_SRIOV_CTRL)) {
8509         if (!(val & PCI_SRIOV_CTRL_VFE)) {
8510             num_vfs = pci_get_word(dev->config + sriov_cap + PCI_SRIOV_NUM_VF);
8511             for (i = 0; i < num_vfs; i++) {
8512                 sctrl = &n->sec_ctrl_list.sec[i];
8513                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
8514             }
8515         }
8516     }
8517 }
8518 
8519 static void nvme_pci_write_config(PCIDevice *dev, uint32_t address,
8520                                   uint32_t val, int len)
8521 {
8522     nvme_sriov_pre_write_ctrl(dev, address, val, len);
8523     pci_default_write_config(dev, address, val, len);
8524     pcie_cap_flr_write_config(dev, address, val, len);
8525 }
8526 
8527 static const VMStateDescription nvme_vmstate = {
8528     .name = "nvme",
8529     .unmigratable = 1,
8530 };
8531 
8532 static void nvme_class_init(ObjectClass *oc, void *data)
8533 {
8534     DeviceClass *dc = DEVICE_CLASS(oc);
8535     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
8536 
8537     pc->realize = nvme_realize;
8538     pc->config_write = nvme_pci_write_config;
8539     pc->exit = nvme_exit;
8540     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
8541     pc->revision = 2;
8542 
8543     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
8544     dc->desc = "Non-Volatile Memory Express";
8545     device_class_set_props(dc, nvme_props);
8546     dc->vmsd = &nvme_vmstate;
8547     dc->reset = nvme_pci_reset;
8548 }
8549 
8550 static void nvme_instance_init(Object *obj)
8551 {
8552     NvmeCtrl *n = NVME(obj);
8553 
8554     device_add_bootindex_property(obj, &n->namespace.blkconf.bootindex,
8555                                   "bootindex", "/namespace@1,0",
8556                                   DEVICE(obj));
8557 
8558     object_property_add(obj, "smart_critical_warning", "uint8",
8559                         nvme_get_smart_warning,
8560                         nvme_set_smart_warning, NULL, NULL);
8561 }
8562 
8563 static const TypeInfo nvme_info = {
8564     .name          = TYPE_NVME,
8565     .parent        = TYPE_PCI_DEVICE,
8566     .instance_size = sizeof(NvmeCtrl),
8567     .instance_init = nvme_instance_init,
8568     .class_init    = nvme_class_init,
8569     .interfaces = (InterfaceInfo[]) {
8570         { INTERFACE_PCIE_DEVICE },
8571         { }
8572     },
8573 };
8574 
8575 static const TypeInfo nvme_bus_info = {
8576     .name = TYPE_NVME_BUS,
8577     .parent = TYPE_BUS,
8578     .instance_size = sizeof(NvmeBus),
8579 };
8580 
8581 static void nvme_register_types(void)
8582 {
8583     type_register_static(&nvme_info);
8584     type_register_static(&nvme_bus_info);
8585 }
8586 
8587 type_init(nvme_register_types)
8588