xref: /openbmc/qemu/hw/nvme/ctrl.c (revision 278f064e)
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  *              subsys=<subsys_id>
38  *      -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
39  *              zoned=<true|false[optional]>, \
40  *              subsys=<subsys_id>,detached=<true|false[optional]>
41  *
42  * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
43  * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. By default, the
44  * device will use the "v1.4 CMB scheme" - use the `legacy-cmb` parameter to
45  * always enable the CMBLOC and CMBSZ registers (v1.3 behavior).
46  *
47  * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
48  * For example:
49  * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
50  *  size=<size> .... -device nvme,...,pmrdev=<mem_id>
51  *
52  * The PMR will use BAR 4/5 exclusively.
53  *
54  * To place controller(s) and namespace(s) to a subsystem, then provide
55  * nvme-subsys device as above.
56  *
57  * nvme subsystem device parameters
58  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
59  * - `nqn`
60  *   This parameter provides the `<nqn_id>` part of the string
61  *   `nqn.2019-08.org.qemu:<nqn_id>` which will be reported in the SUBNQN field
62  *   of subsystem controllers. Note that `<nqn_id>` should be unique per
63  *   subsystem, but this is not enforced by QEMU. If not specified, it will
64  *   default to the value of the `id` parameter (`<subsys_id>`).
65  *
66  * nvme device parameters
67  * ~~~~~~~~~~~~~~~~~~~~~~
68  * - `subsys`
69  *   Specifying this parameter attaches the controller to the subsystem and
70  *   the SUBNQN field in the controller will report the NQN of the subsystem
71  *   device. This also enables multi controller capability represented in
72  *   Identify Controller data structure in CMIC (Controller Multi-path I/O and
73  *   Namesapce Sharing Capabilities).
74  *
75  * - `aerl`
76  *   The Asynchronous Event Request Limit (AERL). Indicates the maximum number
77  *   of concurrently outstanding Asynchronous Event Request commands support
78  *   by the controller. This is a 0's based value.
79  *
80  * - `aer_max_queued`
81  *   This is the maximum number of events that the device will enqueue for
82  *   completion when there are no outstanding AERs. When the maximum number of
83  *   enqueued events are reached, subsequent events will be dropped.
84  *
85  * - `mdts`
86  *   Indicates the maximum data transfer size for a command that transfers data
87  *   between host-accessible memory and the controller. The value is specified
88  *   as a power of two (2^n) and is in units of the minimum memory page size
89  *   (CAP.MPSMIN). The default value is 7 (i.e. 512 KiB).
90  *
91  * - `vsl`
92  *   Indicates the maximum data size limit for the Verify command. Like `mdts`,
93  *   this value is specified as a power of two (2^n) and is in units of the
94  *   minimum memory page size (CAP.MPSMIN). The default value is 7 (i.e. 512
95  *   KiB).
96  *
97  * - `zoned.zasl`
98  *   Indicates the maximum data transfer size for the Zone Append command. Like
99  *   `mdts`, the value is specified as a power of two (2^n) and is in units of
100  *   the minimum memory page size (CAP.MPSMIN). The default value is 0 (i.e.
101  *   defaulting to the value of `mdts`).
102  *
103  * nvme namespace device parameters
104  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
105  * - `shared`
106  *   When the parent nvme device (as defined explicitly by the 'bus' parameter
107  *   or implicitly by the most recently defined NvmeBus) is linked to an
108  *   nvme-subsys device, the namespace will be attached to all controllers in
109  *   the subsystem. If set to 'off' (the default), the namespace will remain a
110  *   private namespace and may only be attached to a single controller at a
111  *   time.
112  *
113  * - `detached`
114  *   This parameter is only valid together with the `subsys` parameter. If left
115  *   at the default value (`false/off`), the namespace will be attached to all
116  *   controllers in the NVMe subsystem at boot-up. If set to `true/on`, the
117  *   namespace will be be available in the subsystem not not attached to any
118  *   controllers.
119  *
120  * Setting `zoned` to true selects Zoned Command Set at the namespace.
121  * In this case, the following namespace properties are available to configure
122  * zoned operation:
123  *     zoned.zone_size=<zone size in bytes, default: 128MiB>
124  *         The number may be followed by K, M, G as in kilo-, mega- or giga-.
125  *
126  *     zoned.zone_capacity=<zone capacity in bytes, default: zone size>
127  *         The value 0 (default) forces zone capacity to be the same as zone
128  *         size. The value of this property may not exceed zone size.
129  *
130  *     zoned.descr_ext_size=<zone descriptor extension size, default 0>
131  *         This value needs to be specified in 64B units. If it is zero,
132  *         namespace(s) will not support zone descriptor extensions.
133  *
134  *     zoned.max_active=<Maximum Active Resources (zones), default: 0>
135  *         The default value means there is no limit to the number of
136  *         concurrently active zones.
137  *
138  *     zoned.max_open=<Maximum Open Resources (zones), default: 0>
139  *         The default value means there is no limit to the number of
140  *         concurrently open zones.
141  *
142  *     zoned.cross_read=<enable RAZB, default: false>
143  *         Setting this property to true enables Read Across Zone Boundaries.
144  */
145 
146 #include "qemu/osdep.h"
147 #include "qemu/cutils.h"
148 #include "qemu/error-report.h"
149 #include "qemu/log.h"
150 #include "qemu/units.h"
151 #include "qapi/error.h"
152 #include "qapi/visitor.h"
153 #include "sysemu/sysemu.h"
154 #include "sysemu/block-backend.h"
155 #include "sysemu/hostmem.h"
156 #include "hw/pci/msix.h"
157 #include "migration/vmstate.h"
158 
159 #include "nvme.h"
160 #include "trace.h"
161 
162 #define NVME_MAX_IOQPAIRS 0xffff
163 #define NVME_DB_SIZE  4
164 #define NVME_SPEC_VER 0x00010400
165 #define NVME_CMB_BIR 2
166 #define NVME_PMR_BIR 4
167 #define NVME_TEMPERATURE 0x143
168 #define NVME_TEMPERATURE_WARNING 0x157
169 #define NVME_TEMPERATURE_CRITICAL 0x175
170 #define NVME_NUM_FW_SLOTS 1
171 #define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
172 
173 #define NVME_GUEST_ERR(trace, fmt, ...) \
174     do { \
175         (trace_##trace)(__VA_ARGS__); \
176         qemu_log_mask(LOG_GUEST_ERROR, #trace \
177             " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
178     } while (0)
179 
180 static const bool nvme_feature_support[NVME_FID_MAX] = {
181     [NVME_ARBITRATION]              = true,
182     [NVME_POWER_MANAGEMENT]         = true,
183     [NVME_TEMPERATURE_THRESHOLD]    = true,
184     [NVME_ERROR_RECOVERY]           = true,
185     [NVME_VOLATILE_WRITE_CACHE]     = true,
186     [NVME_NUMBER_OF_QUEUES]         = true,
187     [NVME_INTERRUPT_COALESCING]     = true,
188     [NVME_INTERRUPT_VECTOR_CONF]    = true,
189     [NVME_WRITE_ATOMICITY]          = true,
190     [NVME_ASYNCHRONOUS_EVENT_CONF]  = true,
191     [NVME_TIMESTAMP]                = true,
192     [NVME_COMMAND_SET_PROFILE]      = true,
193 };
194 
195 static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
196     [NVME_TEMPERATURE_THRESHOLD]    = NVME_FEAT_CAP_CHANGE,
197     [NVME_ERROR_RECOVERY]           = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
198     [NVME_VOLATILE_WRITE_CACHE]     = NVME_FEAT_CAP_CHANGE,
199     [NVME_NUMBER_OF_QUEUES]         = NVME_FEAT_CAP_CHANGE,
200     [NVME_ASYNCHRONOUS_EVENT_CONF]  = NVME_FEAT_CAP_CHANGE,
201     [NVME_TIMESTAMP]                = NVME_FEAT_CAP_CHANGE,
202     [NVME_COMMAND_SET_PROFILE]      = NVME_FEAT_CAP_CHANGE,
203 };
204 
205 static const uint32_t nvme_cse_acs[256] = {
206     [NVME_ADM_CMD_DELETE_SQ]        = NVME_CMD_EFF_CSUPP,
207     [NVME_ADM_CMD_CREATE_SQ]        = NVME_CMD_EFF_CSUPP,
208     [NVME_ADM_CMD_GET_LOG_PAGE]     = NVME_CMD_EFF_CSUPP,
209     [NVME_ADM_CMD_DELETE_CQ]        = NVME_CMD_EFF_CSUPP,
210     [NVME_ADM_CMD_CREATE_CQ]        = NVME_CMD_EFF_CSUPP,
211     [NVME_ADM_CMD_IDENTIFY]         = NVME_CMD_EFF_CSUPP,
212     [NVME_ADM_CMD_ABORT]            = NVME_CMD_EFF_CSUPP,
213     [NVME_ADM_CMD_SET_FEATURES]     = NVME_CMD_EFF_CSUPP,
214     [NVME_ADM_CMD_GET_FEATURES]     = NVME_CMD_EFF_CSUPP,
215     [NVME_ADM_CMD_ASYNC_EV_REQ]     = NVME_CMD_EFF_CSUPP,
216     [NVME_ADM_CMD_NS_ATTACHMENT]    = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_NIC,
217     [NVME_ADM_CMD_FORMAT_NVM]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
218 };
219 
220 static const uint32_t nvme_cse_iocs_none[256];
221 
222 static const uint32_t nvme_cse_iocs_nvm[256] = {
223     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
224     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
225     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
226     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
227     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
228     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
229     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
230     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
231 };
232 
233 static const uint32_t nvme_cse_iocs_zoned[256] = {
234     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
235     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
236     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
237     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
238     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
239     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
240     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
241     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
242     [NVME_CMD_ZONE_APPEND]          = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
243     [NVME_CMD_ZONE_MGMT_SEND]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
244     [NVME_CMD_ZONE_MGMT_RECV]       = NVME_CMD_EFF_CSUPP,
245 };
246 
247 static void nvme_process_sq(void *opaque);
248 
249 static uint16_t nvme_sqid(NvmeRequest *req)
250 {
251     return le16_to_cpu(req->sq->sqid);
252 }
253 
254 static void nvme_assign_zone_state(NvmeNamespace *ns, NvmeZone *zone,
255                                    NvmeZoneState state)
256 {
257     if (QTAILQ_IN_USE(zone, entry)) {
258         switch (nvme_get_zone_state(zone)) {
259         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
260             QTAILQ_REMOVE(&ns->exp_open_zones, zone, entry);
261             break;
262         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
263             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
264             break;
265         case NVME_ZONE_STATE_CLOSED:
266             QTAILQ_REMOVE(&ns->closed_zones, zone, entry);
267             break;
268         case NVME_ZONE_STATE_FULL:
269             QTAILQ_REMOVE(&ns->full_zones, zone, entry);
270         default:
271             ;
272         }
273     }
274 
275     nvme_set_zone_state(zone, state);
276 
277     switch (state) {
278     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
279         QTAILQ_INSERT_TAIL(&ns->exp_open_zones, zone, entry);
280         break;
281     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
282         QTAILQ_INSERT_TAIL(&ns->imp_open_zones, zone, entry);
283         break;
284     case NVME_ZONE_STATE_CLOSED:
285         QTAILQ_INSERT_TAIL(&ns->closed_zones, zone, entry);
286         break;
287     case NVME_ZONE_STATE_FULL:
288         QTAILQ_INSERT_TAIL(&ns->full_zones, zone, entry);
289     case NVME_ZONE_STATE_READ_ONLY:
290         break;
291     default:
292         zone->d.za = 0;
293     }
294 }
295 
296 /*
297  * Check if we can open a zone without exceeding open/active limits.
298  * AOR stands for "Active and Open Resources" (see TP 4053 section 2.5).
299  */
300 static int nvme_aor_check(NvmeNamespace *ns, uint32_t act, uint32_t opn)
301 {
302     if (ns->params.max_active_zones != 0 &&
303         ns->nr_active_zones + act > ns->params.max_active_zones) {
304         trace_pci_nvme_err_insuff_active_res(ns->params.max_active_zones);
305         return NVME_ZONE_TOO_MANY_ACTIVE | NVME_DNR;
306     }
307     if (ns->params.max_open_zones != 0 &&
308         ns->nr_open_zones + opn > ns->params.max_open_zones) {
309         trace_pci_nvme_err_insuff_open_res(ns->params.max_open_zones);
310         return NVME_ZONE_TOO_MANY_OPEN | NVME_DNR;
311     }
312 
313     return NVME_SUCCESS;
314 }
315 
316 static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
317 {
318     hwaddr hi, lo;
319 
320     if (!n->cmb.cmse) {
321         return false;
322     }
323 
324     lo = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
325     hi = lo + int128_get64(n->cmb.mem.size);
326 
327     return addr >= lo && addr < hi;
328 }
329 
330 static inline void *nvme_addr_to_cmb(NvmeCtrl *n, hwaddr addr)
331 {
332     hwaddr base = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
333     return &n->cmb.buf[addr - base];
334 }
335 
336 static bool nvme_addr_is_pmr(NvmeCtrl *n, hwaddr addr)
337 {
338     hwaddr hi;
339 
340     if (!n->pmr.cmse) {
341         return false;
342     }
343 
344     hi = n->pmr.cba + int128_get64(n->pmr.dev->mr.size);
345 
346     return addr >= n->pmr.cba && addr < hi;
347 }
348 
349 static inline void *nvme_addr_to_pmr(NvmeCtrl *n, hwaddr addr)
350 {
351     return memory_region_get_ram_ptr(&n->pmr.dev->mr) + (addr - n->pmr.cba);
352 }
353 
354 static int nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
355 {
356     hwaddr hi = addr + size - 1;
357     if (hi < addr) {
358         return 1;
359     }
360 
361     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
362         memcpy(buf, nvme_addr_to_cmb(n, addr), size);
363         return 0;
364     }
365 
366     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
367         memcpy(buf, nvme_addr_to_pmr(n, addr), size);
368         return 0;
369     }
370 
371     return pci_dma_read(&n->parent_obj, addr, buf, size);
372 }
373 
374 static int nvme_addr_write(NvmeCtrl *n, hwaddr addr, void *buf, int size)
375 {
376     hwaddr hi = addr + size - 1;
377     if (hi < addr) {
378         return 1;
379     }
380 
381     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
382         memcpy(nvme_addr_to_cmb(n, addr), buf, size);
383         return 0;
384     }
385 
386     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
387         memcpy(nvme_addr_to_pmr(n, addr), buf, size);
388         return 0;
389     }
390 
391     return pci_dma_write(&n->parent_obj, addr, buf, size);
392 }
393 
394 static bool nvme_nsid_valid(NvmeCtrl *n, uint32_t nsid)
395 {
396     return nsid &&
397         (nsid == NVME_NSID_BROADCAST || nsid <= NVME_MAX_NAMESPACES);
398 }
399 
400 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
401 {
402     return sqid < n->params.max_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
403 }
404 
405 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
406 {
407     return cqid < n->params.max_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
408 }
409 
410 static void nvme_inc_cq_tail(NvmeCQueue *cq)
411 {
412     cq->tail++;
413     if (cq->tail >= cq->size) {
414         cq->tail = 0;
415         cq->phase = !cq->phase;
416     }
417 }
418 
419 static void nvme_inc_sq_head(NvmeSQueue *sq)
420 {
421     sq->head = (sq->head + 1) % sq->size;
422 }
423 
424 static uint8_t nvme_cq_full(NvmeCQueue *cq)
425 {
426     return (cq->tail + 1) % cq->size == cq->head;
427 }
428 
429 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
430 {
431     return sq->head == sq->tail;
432 }
433 
434 static void nvme_irq_check(NvmeCtrl *n)
435 {
436     if (msix_enabled(&(n->parent_obj))) {
437         return;
438     }
439     if (~n->bar.intms & n->irq_status) {
440         pci_irq_assert(&n->parent_obj);
441     } else {
442         pci_irq_deassert(&n->parent_obj);
443     }
444 }
445 
446 static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
447 {
448     if (cq->irq_enabled) {
449         if (msix_enabled(&(n->parent_obj))) {
450             trace_pci_nvme_irq_msix(cq->vector);
451             msix_notify(&(n->parent_obj), cq->vector);
452         } else {
453             trace_pci_nvme_irq_pin();
454             assert(cq->vector < 32);
455             n->irq_status |= 1 << cq->vector;
456             nvme_irq_check(n);
457         }
458     } else {
459         trace_pci_nvme_irq_masked();
460     }
461 }
462 
463 static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
464 {
465     if (cq->irq_enabled) {
466         if (msix_enabled(&(n->parent_obj))) {
467             return;
468         } else {
469             assert(cq->vector < 32);
470             n->irq_status &= ~(1 << cq->vector);
471             nvme_irq_check(n);
472         }
473     }
474 }
475 
476 static void nvme_req_clear(NvmeRequest *req)
477 {
478     req->ns = NULL;
479     req->opaque = NULL;
480     req->aiocb = NULL;
481     memset(&req->cqe, 0x0, sizeof(req->cqe));
482     req->status = NVME_SUCCESS;
483 }
484 
485 static inline void nvme_sg_init(NvmeCtrl *n, NvmeSg *sg, bool dma)
486 {
487     if (dma) {
488         pci_dma_sglist_init(&sg->qsg, &n->parent_obj, 0);
489         sg->flags = NVME_SG_DMA;
490     } else {
491         qemu_iovec_init(&sg->iov, 0);
492     }
493 
494     sg->flags |= NVME_SG_ALLOC;
495 }
496 
497 static inline void nvme_sg_unmap(NvmeSg *sg)
498 {
499     if (!(sg->flags & NVME_SG_ALLOC)) {
500         return;
501     }
502 
503     if (sg->flags & NVME_SG_DMA) {
504         qemu_sglist_destroy(&sg->qsg);
505     } else {
506         qemu_iovec_destroy(&sg->iov);
507     }
508 
509     memset(sg, 0x0, sizeof(*sg));
510 }
511 
512 /*
513  * When metadata is transfered as extended LBAs, the DPTR mapped into `sg`
514  * holds both data and metadata. This function splits the data and metadata
515  * into two separate QSG/IOVs.
516  */
517 static void nvme_sg_split(NvmeSg *sg, NvmeNamespace *ns, NvmeSg *data,
518                           NvmeSg *mdata)
519 {
520     NvmeSg *dst = data;
521     uint32_t trans_len, count = ns->lbasz;
522     uint64_t offset = 0;
523     bool dma = sg->flags & NVME_SG_DMA;
524     size_t sge_len;
525     size_t sg_len = dma ? sg->qsg.size : sg->iov.size;
526     int sg_idx = 0;
527 
528     assert(sg->flags & NVME_SG_ALLOC);
529 
530     while (sg_len) {
531         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
532 
533         trans_len = MIN(sg_len, count);
534         trans_len = MIN(trans_len, sge_len - offset);
535 
536         if (dst) {
537             if (dma) {
538                 qemu_sglist_add(&dst->qsg, sg->qsg.sg[sg_idx].base + offset,
539                                 trans_len);
540             } else {
541                 qemu_iovec_add(&dst->iov,
542                                sg->iov.iov[sg_idx].iov_base + offset,
543                                trans_len);
544             }
545         }
546 
547         sg_len -= trans_len;
548         count -= trans_len;
549         offset += trans_len;
550 
551         if (count == 0) {
552             dst = (dst == data) ? mdata : data;
553             count = (dst == data) ? ns->lbasz : ns->lbaf.ms;
554         }
555 
556         if (sge_len == offset) {
557             offset = 0;
558             sg_idx++;
559         }
560     }
561 }
562 
563 static uint16_t nvme_map_addr_cmb(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
564                                   size_t len)
565 {
566     if (!len) {
567         return NVME_SUCCESS;
568     }
569 
570     trace_pci_nvme_map_addr_cmb(addr, len);
571 
572     if (!nvme_addr_is_cmb(n, addr) || !nvme_addr_is_cmb(n, addr + len - 1)) {
573         return NVME_DATA_TRAS_ERROR;
574     }
575 
576     qemu_iovec_add(iov, nvme_addr_to_cmb(n, addr), len);
577 
578     return NVME_SUCCESS;
579 }
580 
581 static uint16_t nvme_map_addr_pmr(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
582                                   size_t len)
583 {
584     if (!len) {
585         return NVME_SUCCESS;
586     }
587 
588     if (!nvme_addr_is_pmr(n, addr) || !nvme_addr_is_pmr(n, addr + len - 1)) {
589         return NVME_DATA_TRAS_ERROR;
590     }
591 
592     qemu_iovec_add(iov, nvme_addr_to_pmr(n, addr), len);
593 
594     return NVME_SUCCESS;
595 }
596 
597 static uint16_t nvme_map_addr(NvmeCtrl *n, NvmeSg *sg, hwaddr addr, size_t len)
598 {
599     bool cmb = false, pmr = false;
600 
601     if (!len) {
602         return NVME_SUCCESS;
603     }
604 
605     trace_pci_nvme_map_addr(addr, len);
606 
607     if (nvme_addr_is_cmb(n, addr)) {
608         cmb = true;
609     } else if (nvme_addr_is_pmr(n, addr)) {
610         pmr = true;
611     }
612 
613     if (cmb || pmr) {
614         if (sg->flags & NVME_SG_DMA) {
615             return NVME_INVALID_USE_OF_CMB | NVME_DNR;
616         }
617 
618         if (cmb) {
619             return nvme_map_addr_cmb(n, &sg->iov, addr, len);
620         } else {
621             return nvme_map_addr_pmr(n, &sg->iov, addr, len);
622         }
623     }
624 
625     if (!(sg->flags & NVME_SG_DMA)) {
626         return NVME_INVALID_USE_OF_CMB | NVME_DNR;
627     }
628 
629     qemu_sglist_add(&sg->qsg, addr, len);
630 
631     return NVME_SUCCESS;
632 }
633 
634 static inline bool nvme_addr_is_dma(NvmeCtrl *n, hwaddr addr)
635 {
636     return !(nvme_addr_is_cmb(n, addr) || nvme_addr_is_pmr(n, addr));
637 }
638 
639 static uint16_t nvme_map_prp(NvmeCtrl *n, NvmeSg *sg, uint64_t prp1,
640                              uint64_t prp2, uint32_t len)
641 {
642     hwaddr trans_len = n->page_size - (prp1 % n->page_size);
643     trans_len = MIN(len, trans_len);
644     int num_prps = (len >> n->page_bits) + 1;
645     uint16_t status;
646     int ret;
647 
648     trace_pci_nvme_map_prp(trans_len, len, prp1, prp2, num_prps);
649 
650     nvme_sg_init(n, sg, nvme_addr_is_dma(n, prp1));
651 
652     status = nvme_map_addr(n, sg, prp1, trans_len);
653     if (status) {
654         goto unmap;
655     }
656 
657     len -= trans_len;
658     if (len) {
659         if (len > n->page_size) {
660             uint64_t prp_list[n->max_prp_ents];
661             uint32_t nents, prp_trans;
662             int i = 0;
663 
664             /*
665              * The first PRP list entry, pointed to by PRP2 may contain offset.
666              * Hence, we need to calculate the number of entries in based on
667              * that offset.
668              */
669             nents = (n->page_size - (prp2 & (n->page_size - 1))) >> 3;
670             prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
671             ret = nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
672             if (ret) {
673                 trace_pci_nvme_err_addr_read(prp2);
674                 status = NVME_DATA_TRAS_ERROR;
675                 goto unmap;
676             }
677             while (len != 0) {
678                 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
679 
680                 if (i == nents - 1 && len > n->page_size) {
681                     if (unlikely(prp_ent & (n->page_size - 1))) {
682                         trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
683                         status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
684                         goto unmap;
685                     }
686 
687                     i = 0;
688                     nents = (len + n->page_size - 1) >> n->page_bits;
689                     nents = MIN(nents, n->max_prp_ents);
690                     prp_trans = nents * sizeof(uint64_t);
691                     ret = nvme_addr_read(n, prp_ent, (void *)prp_list,
692                                          prp_trans);
693                     if (ret) {
694                         trace_pci_nvme_err_addr_read(prp_ent);
695                         status = NVME_DATA_TRAS_ERROR;
696                         goto unmap;
697                     }
698                     prp_ent = le64_to_cpu(prp_list[i]);
699                 }
700 
701                 if (unlikely(prp_ent & (n->page_size - 1))) {
702                     trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
703                     status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
704                     goto unmap;
705                 }
706 
707                 trans_len = MIN(len, n->page_size);
708                 status = nvme_map_addr(n, sg, prp_ent, trans_len);
709                 if (status) {
710                     goto unmap;
711                 }
712 
713                 len -= trans_len;
714                 i++;
715             }
716         } else {
717             if (unlikely(prp2 & (n->page_size - 1))) {
718                 trace_pci_nvme_err_invalid_prp2_align(prp2);
719                 status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
720                 goto unmap;
721             }
722             status = nvme_map_addr(n, sg, prp2, len);
723             if (status) {
724                 goto unmap;
725             }
726         }
727     }
728 
729     return NVME_SUCCESS;
730 
731 unmap:
732     nvme_sg_unmap(sg);
733     return status;
734 }
735 
736 /*
737  * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
738  * number of bytes mapped in len.
739  */
740 static uint16_t nvme_map_sgl_data(NvmeCtrl *n, NvmeSg *sg,
741                                   NvmeSglDescriptor *segment, uint64_t nsgld,
742                                   size_t *len, NvmeCmd *cmd)
743 {
744     dma_addr_t addr, trans_len;
745     uint32_t dlen;
746     uint16_t status;
747 
748     for (int i = 0; i < nsgld; i++) {
749         uint8_t type = NVME_SGL_TYPE(segment[i].type);
750 
751         switch (type) {
752         case NVME_SGL_DESCR_TYPE_BIT_BUCKET:
753             if (cmd->opcode == NVME_CMD_WRITE) {
754                 continue;
755             }
756         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
757             break;
758         case NVME_SGL_DESCR_TYPE_SEGMENT:
759         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
760             return NVME_INVALID_NUM_SGL_DESCRS | NVME_DNR;
761         default:
762             return NVME_SGL_DESCR_TYPE_INVALID | NVME_DNR;
763         }
764 
765         dlen = le32_to_cpu(segment[i].len);
766 
767         if (!dlen) {
768             continue;
769         }
770 
771         if (*len == 0) {
772             /*
773              * All data has been mapped, but the SGL contains additional
774              * segments and/or descriptors. The controller might accept
775              * ignoring the rest of the SGL.
776              */
777             uint32_t sgls = le32_to_cpu(n->id_ctrl.sgls);
778             if (sgls & NVME_CTRL_SGLS_EXCESS_LENGTH) {
779                 break;
780             }
781 
782             trace_pci_nvme_err_invalid_sgl_excess_length(dlen);
783             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
784         }
785 
786         trans_len = MIN(*len, dlen);
787 
788         if (type == NVME_SGL_DESCR_TYPE_BIT_BUCKET) {
789             goto next;
790         }
791 
792         addr = le64_to_cpu(segment[i].addr);
793 
794         if (UINT64_MAX - addr < dlen) {
795             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
796         }
797 
798         status = nvme_map_addr(n, sg, addr, trans_len);
799         if (status) {
800             return status;
801         }
802 
803 next:
804         *len -= trans_len;
805     }
806 
807     return NVME_SUCCESS;
808 }
809 
810 static uint16_t nvme_map_sgl(NvmeCtrl *n, NvmeSg *sg, NvmeSglDescriptor sgl,
811                              size_t len, NvmeCmd *cmd)
812 {
813     /*
814      * Read the segment in chunks of 256 descriptors (one 4k page) to avoid
815      * dynamically allocating a potentially huge SGL. The spec allows the SGL
816      * to be larger (as in number of bytes required to describe the SGL
817      * descriptors and segment chain) than the command transfer size, so it is
818      * not bounded by MDTS.
819      */
820     const int SEG_CHUNK_SIZE = 256;
821 
822     NvmeSglDescriptor segment[SEG_CHUNK_SIZE], *sgld, *last_sgld;
823     uint64_t nsgld;
824     uint32_t seg_len;
825     uint16_t status;
826     hwaddr addr;
827     int ret;
828 
829     sgld = &sgl;
830     addr = le64_to_cpu(sgl.addr);
831 
832     trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl.type), len);
833 
834     nvme_sg_init(n, sg, nvme_addr_is_dma(n, addr));
835 
836     /*
837      * If the entire transfer can be described with a single data block it can
838      * be mapped directly.
839      */
840     if (NVME_SGL_TYPE(sgl.type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
841         status = nvme_map_sgl_data(n, sg, sgld, 1, &len, cmd);
842         if (status) {
843             goto unmap;
844         }
845 
846         goto out;
847     }
848 
849     for (;;) {
850         switch (NVME_SGL_TYPE(sgld->type)) {
851         case NVME_SGL_DESCR_TYPE_SEGMENT:
852         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
853             break;
854         default:
855             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
856         }
857 
858         seg_len = le32_to_cpu(sgld->len);
859 
860         /* check the length of the (Last) Segment descriptor */
861         if ((!seg_len || seg_len & 0xf) &&
862             (NVME_SGL_TYPE(sgld->type) != NVME_SGL_DESCR_TYPE_BIT_BUCKET)) {
863             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
864         }
865 
866         if (UINT64_MAX - addr < seg_len) {
867             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
868         }
869 
870         nsgld = seg_len / sizeof(NvmeSglDescriptor);
871 
872         while (nsgld > SEG_CHUNK_SIZE) {
873             if (nvme_addr_read(n, addr, segment, sizeof(segment))) {
874                 trace_pci_nvme_err_addr_read(addr);
875                 status = NVME_DATA_TRAS_ERROR;
876                 goto unmap;
877             }
878 
879             status = nvme_map_sgl_data(n, sg, segment, SEG_CHUNK_SIZE,
880                                        &len, cmd);
881             if (status) {
882                 goto unmap;
883             }
884 
885             nsgld -= SEG_CHUNK_SIZE;
886             addr += SEG_CHUNK_SIZE * sizeof(NvmeSglDescriptor);
887         }
888 
889         ret = nvme_addr_read(n, addr, segment, nsgld *
890                              sizeof(NvmeSglDescriptor));
891         if (ret) {
892             trace_pci_nvme_err_addr_read(addr);
893             status = NVME_DATA_TRAS_ERROR;
894             goto unmap;
895         }
896 
897         last_sgld = &segment[nsgld - 1];
898 
899         /*
900          * If the segment ends with a Data Block or Bit Bucket Descriptor Type,
901          * then we are done.
902          */
903         switch (NVME_SGL_TYPE(last_sgld->type)) {
904         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
905         case NVME_SGL_DESCR_TYPE_BIT_BUCKET:
906             status = nvme_map_sgl_data(n, sg, segment, nsgld, &len, cmd);
907             if (status) {
908                 goto unmap;
909             }
910 
911             goto out;
912 
913         default:
914             break;
915         }
916 
917         /*
918          * If the last descriptor was not a Data Block or Bit Bucket, then the
919          * current segment must not be a Last Segment.
920          */
921         if (NVME_SGL_TYPE(sgld->type) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT) {
922             status = NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
923             goto unmap;
924         }
925 
926         sgld = last_sgld;
927         addr = le64_to_cpu(sgld->addr);
928 
929         /*
930          * Do not map the last descriptor; it will be a Segment or Last Segment
931          * descriptor and is handled by the next iteration.
932          */
933         status = nvme_map_sgl_data(n, sg, segment, nsgld - 1, &len, cmd);
934         if (status) {
935             goto unmap;
936         }
937     }
938 
939 out:
940     /* if there is any residual left in len, the SGL was too short */
941     if (len) {
942         status = NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
943         goto unmap;
944     }
945 
946     return NVME_SUCCESS;
947 
948 unmap:
949     nvme_sg_unmap(sg);
950     return status;
951 }
952 
953 uint16_t nvme_map_dptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
954                        NvmeCmd *cmd)
955 {
956     uint64_t prp1, prp2;
957 
958     switch (NVME_CMD_FLAGS_PSDT(cmd->flags)) {
959     case NVME_PSDT_PRP:
960         prp1 = le64_to_cpu(cmd->dptr.prp1);
961         prp2 = le64_to_cpu(cmd->dptr.prp2);
962 
963         return nvme_map_prp(n, sg, prp1, prp2, len);
964     case NVME_PSDT_SGL_MPTR_CONTIGUOUS:
965     case NVME_PSDT_SGL_MPTR_SGL:
966         return nvme_map_sgl(n, sg, cmd->dptr.sgl, len, cmd);
967     default:
968         return NVME_INVALID_FIELD;
969     }
970 }
971 
972 static uint16_t nvme_map_mptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
973                               NvmeCmd *cmd)
974 {
975     int psdt = NVME_CMD_FLAGS_PSDT(cmd->flags);
976     hwaddr mptr = le64_to_cpu(cmd->mptr);
977     uint16_t status;
978 
979     if (psdt == NVME_PSDT_SGL_MPTR_SGL) {
980         NvmeSglDescriptor sgl;
981 
982         if (nvme_addr_read(n, mptr, &sgl, sizeof(sgl))) {
983             return NVME_DATA_TRAS_ERROR;
984         }
985 
986         status = nvme_map_sgl(n, sg, sgl, len, cmd);
987         if (status && (status & 0x7ff) == NVME_DATA_SGL_LEN_INVALID) {
988             status = NVME_MD_SGL_LEN_INVALID | NVME_DNR;
989         }
990 
991         return status;
992     }
993 
994     nvme_sg_init(n, sg, nvme_addr_is_dma(n, mptr));
995     status = nvme_map_addr(n, sg, mptr, len);
996     if (status) {
997         nvme_sg_unmap(sg);
998     }
999 
1000     return status;
1001 }
1002 
1003 static uint16_t nvme_map_data(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1004 {
1005     NvmeNamespace *ns = req->ns;
1006     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1007     uint16_t ctrl = le16_to_cpu(rw->control);
1008     size_t len = nvme_l2b(ns, nlb);
1009     uint16_t status;
1010 
1011     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) &&
1012         (ctrl & NVME_RW_PRINFO_PRACT && ns->lbaf.ms == 8)) {
1013         goto out;
1014     }
1015 
1016     if (nvme_ns_ext(ns)) {
1017         NvmeSg sg;
1018 
1019         len += nvme_m2b(ns, nlb);
1020 
1021         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1022         if (status) {
1023             return status;
1024         }
1025 
1026         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1027         nvme_sg_split(&sg, ns, &req->sg, NULL);
1028         nvme_sg_unmap(&sg);
1029 
1030         return NVME_SUCCESS;
1031     }
1032 
1033 out:
1034     return nvme_map_dptr(n, &req->sg, len, &req->cmd);
1035 }
1036 
1037 static uint16_t nvme_map_mdata(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1038 {
1039     NvmeNamespace *ns = req->ns;
1040     size_t len = nvme_m2b(ns, nlb);
1041     uint16_t status;
1042 
1043     if (nvme_ns_ext(ns)) {
1044         NvmeSg sg;
1045 
1046         len += nvme_l2b(ns, nlb);
1047 
1048         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1049         if (status) {
1050             return status;
1051         }
1052 
1053         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1054         nvme_sg_split(&sg, ns, NULL, &req->sg);
1055         nvme_sg_unmap(&sg);
1056 
1057         return NVME_SUCCESS;
1058     }
1059 
1060     return nvme_map_mptr(n, &req->sg, len, &req->cmd);
1061 }
1062 
1063 static uint16_t nvme_tx_interleaved(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr,
1064                                     uint32_t len, uint32_t bytes,
1065                                     int32_t skip_bytes, int64_t offset,
1066                                     NvmeTxDirection dir)
1067 {
1068     hwaddr addr;
1069     uint32_t trans_len, count = bytes;
1070     bool dma = sg->flags & NVME_SG_DMA;
1071     int64_t sge_len;
1072     int sg_idx = 0;
1073     int ret;
1074 
1075     assert(sg->flags & NVME_SG_ALLOC);
1076 
1077     while (len) {
1078         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
1079 
1080         if (sge_len - offset < 0) {
1081             offset -= sge_len;
1082             sg_idx++;
1083             continue;
1084         }
1085 
1086         if (sge_len == offset) {
1087             offset = 0;
1088             sg_idx++;
1089             continue;
1090         }
1091 
1092         trans_len = MIN(len, count);
1093         trans_len = MIN(trans_len, sge_len - offset);
1094 
1095         if (dma) {
1096             addr = sg->qsg.sg[sg_idx].base + offset;
1097         } else {
1098             addr = (hwaddr)(uintptr_t)sg->iov.iov[sg_idx].iov_base + offset;
1099         }
1100 
1101         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1102             ret = nvme_addr_read(n, addr, ptr, trans_len);
1103         } else {
1104             ret = nvme_addr_write(n, addr, ptr, trans_len);
1105         }
1106 
1107         if (ret) {
1108             return NVME_DATA_TRAS_ERROR;
1109         }
1110 
1111         ptr += trans_len;
1112         len -= trans_len;
1113         count -= trans_len;
1114         offset += trans_len;
1115 
1116         if (count == 0) {
1117             count = bytes;
1118             offset += skip_bytes;
1119         }
1120     }
1121 
1122     return NVME_SUCCESS;
1123 }
1124 
1125 static uint16_t nvme_tx(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr, uint32_t len,
1126                         NvmeTxDirection dir)
1127 {
1128     assert(sg->flags & NVME_SG_ALLOC);
1129 
1130     if (sg->flags & NVME_SG_DMA) {
1131         uint64_t residual;
1132 
1133         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1134             residual = dma_buf_write(ptr, len, &sg->qsg);
1135         } else {
1136             residual = dma_buf_read(ptr, len, &sg->qsg);
1137         }
1138 
1139         if (unlikely(residual)) {
1140             trace_pci_nvme_err_invalid_dma();
1141             return NVME_INVALID_FIELD | NVME_DNR;
1142         }
1143     } else {
1144         size_t bytes;
1145 
1146         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1147             bytes = qemu_iovec_to_buf(&sg->iov, 0, ptr, len);
1148         } else {
1149             bytes = qemu_iovec_from_buf(&sg->iov, 0, ptr, len);
1150         }
1151 
1152         if (unlikely(bytes != len)) {
1153             trace_pci_nvme_err_invalid_dma();
1154             return NVME_INVALID_FIELD | NVME_DNR;
1155         }
1156     }
1157 
1158     return NVME_SUCCESS;
1159 }
1160 
1161 static inline uint16_t nvme_c2h(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
1162                                 NvmeRequest *req)
1163 {
1164     uint16_t status;
1165 
1166     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1167     if (status) {
1168         return status;
1169     }
1170 
1171     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_FROM_DEVICE);
1172 }
1173 
1174 static inline uint16_t nvme_h2c(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
1175                                 NvmeRequest *req)
1176 {
1177     uint16_t status;
1178 
1179     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1180     if (status) {
1181         return status;
1182     }
1183 
1184     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_TO_DEVICE);
1185 }
1186 
1187 uint16_t nvme_bounce_data(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
1188                           NvmeTxDirection dir, NvmeRequest *req)
1189 {
1190     NvmeNamespace *ns = req->ns;
1191     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1192     uint16_t ctrl = le16_to_cpu(rw->control);
1193 
1194     if (nvme_ns_ext(ns) &&
1195         !(ctrl & NVME_RW_PRINFO_PRACT && ns->lbaf.ms == 8)) {
1196         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbasz,
1197                                    ns->lbaf.ms, 0, dir);
1198     }
1199 
1200     return nvme_tx(n, &req->sg, ptr, len, dir);
1201 }
1202 
1203 uint16_t nvme_bounce_mdata(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
1204                            NvmeTxDirection dir, NvmeRequest *req)
1205 {
1206     NvmeNamespace *ns = req->ns;
1207     uint16_t status;
1208 
1209     if (nvme_ns_ext(ns)) {
1210         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbaf.ms,
1211                                    ns->lbasz, ns->lbasz, dir);
1212     }
1213 
1214     nvme_sg_unmap(&req->sg);
1215 
1216     status = nvme_map_mptr(n, &req->sg, len, &req->cmd);
1217     if (status) {
1218         return status;
1219     }
1220 
1221     return nvme_tx(n, &req->sg, ptr, len, dir);
1222 }
1223 
1224 static inline void nvme_blk_read(BlockBackend *blk, int64_t offset,
1225                                  BlockCompletionFunc *cb, NvmeRequest *req)
1226 {
1227     assert(req->sg.flags & NVME_SG_ALLOC);
1228 
1229     if (req->sg.flags & NVME_SG_DMA) {
1230         req->aiocb = dma_blk_read(blk, &req->sg.qsg, offset, BDRV_SECTOR_SIZE,
1231                                   cb, req);
1232     } else {
1233         req->aiocb = blk_aio_preadv(blk, offset, &req->sg.iov, 0, cb, req);
1234     }
1235 }
1236 
1237 static inline void nvme_blk_write(BlockBackend *blk, int64_t offset,
1238                                   BlockCompletionFunc *cb, NvmeRequest *req)
1239 {
1240     assert(req->sg.flags & NVME_SG_ALLOC);
1241 
1242     if (req->sg.flags & NVME_SG_DMA) {
1243         req->aiocb = dma_blk_write(blk, &req->sg.qsg, offset, BDRV_SECTOR_SIZE,
1244                                    cb, req);
1245     } else {
1246         req->aiocb = blk_aio_pwritev(blk, offset, &req->sg.iov, 0, cb, req);
1247     }
1248 }
1249 
1250 static void nvme_post_cqes(void *opaque)
1251 {
1252     NvmeCQueue *cq = opaque;
1253     NvmeCtrl *n = cq->ctrl;
1254     NvmeRequest *req, *next;
1255     int ret;
1256 
1257     QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
1258         NvmeSQueue *sq;
1259         hwaddr addr;
1260 
1261         if (nvme_cq_full(cq)) {
1262             break;
1263         }
1264 
1265         sq = req->sq;
1266         req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
1267         req->cqe.sq_id = cpu_to_le16(sq->sqid);
1268         req->cqe.sq_head = cpu_to_le16(sq->head);
1269         addr = cq->dma_addr + cq->tail * n->cqe_size;
1270         ret = pci_dma_write(&n->parent_obj, addr, (void *)&req->cqe,
1271                             sizeof(req->cqe));
1272         if (ret) {
1273             trace_pci_nvme_err_addr_write(addr);
1274             trace_pci_nvme_err_cfs();
1275             n->bar.csts = NVME_CSTS_FAILED;
1276             break;
1277         }
1278         QTAILQ_REMOVE(&cq->req_list, req, entry);
1279         nvme_inc_cq_tail(cq);
1280         nvme_sg_unmap(&req->sg);
1281         QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
1282     }
1283     if (cq->tail != cq->head) {
1284         nvme_irq_assert(n, cq);
1285     }
1286 }
1287 
1288 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
1289 {
1290     assert(cq->cqid == req->sq->cqid);
1291     trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
1292                                           req->status);
1293 
1294     if (req->status) {
1295         trace_pci_nvme_err_req_status(nvme_cid(req), nvme_nsid(req->ns),
1296                                       req->status, req->cmd.opcode);
1297     }
1298 
1299     QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
1300     QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
1301     timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
1302 }
1303 
1304 static void nvme_process_aers(void *opaque)
1305 {
1306     NvmeCtrl *n = opaque;
1307     NvmeAsyncEvent *event, *next;
1308 
1309     trace_pci_nvme_process_aers(n->aer_queued);
1310 
1311     QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
1312         NvmeRequest *req;
1313         NvmeAerResult *result;
1314 
1315         /* can't post cqe if there is nothing to complete */
1316         if (!n->outstanding_aers) {
1317             trace_pci_nvme_no_outstanding_aers();
1318             break;
1319         }
1320 
1321         /* ignore if masked (cqe posted, but event not cleared) */
1322         if (n->aer_mask & (1 << event->result.event_type)) {
1323             trace_pci_nvme_aer_masked(event->result.event_type, n->aer_mask);
1324             continue;
1325         }
1326 
1327         QTAILQ_REMOVE(&n->aer_queue, event, entry);
1328         n->aer_queued--;
1329 
1330         n->aer_mask |= 1 << event->result.event_type;
1331         n->outstanding_aers--;
1332 
1333         req = n->aer_reqs[n->outstanding_aers];
1334 
1335         result = (NvmeAerResult *) &req->cqe.result;
1336         result->event_type = event->result.event_type;
1337         result->event_info = event->result.event_info;
1338         result->log_page = event->result.log_page;
1339         g_free(event);
1340 
1341         trace_pci_nvme_aer_post_cqe(result->event_type, result->event_info,
1342                                     result->log_page);
1343 
1344         nvme_enqueue_req_completion(&n->admin_cq, req);
1345     }
1346 }
1347 
1348 static void nvme_enqueue_event(NvmeCtrl *n, uint8_t event_type,
1349                                uint8_t event_info, uint8_t log_page)
1350 {
1351     NvmeAsyncEvent *event;
1352 
1353     trace_pci_nvme_enqueue_event(event_type, event_info, log_page);
1354 
1355     if (n->aer_queued == n->params.aer_max_queued) {
1356         trace_pci_nvme_enqueue_event_noqueue(n->aer_queued);
1357         return;
1358     }
1359 
1360     event = g_new(NvmeAsyncEvent, 1);
1361     event->result = (NvmeAerResult) {
1362         .event_type = event_type,
1363         .event_info = event_info,
1364         .log_page   = log_page,
1365     };
1366 
1367     QTAILQ_INSERT_TAIL(&n->aer_queue, event, entry);
1368     n->aer_queued++;
1369 
1370     nvme_process_aers(n);
1371 }
1372 
1373 static void nvme_smart_event(NvmeCtrl *n, uint8_t event)
1374 {
1375     uint8_t aer_info;
1376 
1377     /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
1378     if (!(NVME_AEC_SMART(n->features.async_config) & event)) {
1379         return;
1380     }
1381 
1382     switch (event) {
1383     case NVME_SMART_SPARE:
1384         aer_info = NVME_AER_INFO_SMART_SPARE_THRESH;
1385         break;
1386     case NVME_SMART_TEMPERATURE:
1387         aer_info = NVME_AER_INFO_SMART_TEMP_THRESH;
1388         break;
1389     case NVME_SMART_RELIABILITY:
1390     case NVME_SMART_MEDIA_READ_ONLY:
1391     case NVME_SMART_FAILED_VOLATILE_MEDIA:
1392     case NVME_SMART_PMR_UNRELIABLE:
1393         aer_info = NVME_AER_INFO_SMART_RELIABILITY;
1394         break;
1395     default:
1396         return;
1397     }
1398 
1399     nvme_enqueue_event(n, NVME_AER_TYPE_SMART, aer_info, NVME_LOG_SMART_INFO);
1400 }
1401 
1402 static void nvme_clear_events(NvmeCtrl *n, uint8_t event_type)
1403 {
1404     n->aer_mask &= ~(1 << event_type);
1405     if (!QTAILQ_EMPTY(&n->aer_queue)) {
1406         nvme_process_aers(n);
1407     }
1408 }
1409 
1410 static inline uint16_t nvme_check_mdts(NvmeCtrl *n, size_t len)
1411 {
1412     uint8_t mdts = n->params.mdts;
1413 
1414     if (mdts && len > n->page_size << mdts) {
1415         trace_pci_nvme_err_mdts(len);
1416         return NVME_INVALID_FIELD | NVME_DNR;
1417     }
1418 
1419     return NVME_SUCCESS;
1420 }
1421 
1422 static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
1423                                          uint32_t nlb)
1424 {
1425     uint64_t nsze = le64_to_cpu(ns->id_ns.nsze);
1426 
1427     if (unlikely(UINT64_MAX - slba < nlb || slba + nlb > nsze)) {
1428         trace_pci_nvme_err_invalid_lba_range(slba, nlb, nsze);
1429         return NVME_LBA_RANGE | NVME_DNR;
1430     }
1431 
1432     return NVME_SUCCESS;
1433 }
1434 
1435 static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
1436                                  uint32_t nlb)
1437 {
1438     BlockDriverState *bs = blk_bs(ns->blkconf.blk);
1439 
1440     int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
1441     int64_t offset = nvme_l2b(ns, slba);
1442     bool zeroed;
1443     int ret;
1444 
1445     Error *local_err = NULL;
1446 
1447     /*
1448      * `pnum` holds the number of bytes after offset that shares the same
1449      * allocation status as the byte at offset. If `pnum` is different from
1450      * `bytes`, we should check the allocation status of the next range and
1451      * continue this until all bytes have been checked.
1452      */
1453     do {
1454         bytes -= pnum;
1455 
1456         ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
1457         if (ret < 0) {
1458             error_setg_errno(&local_err, -ret, "unable to get block status");
1459             error_report_err(local_err);
1460 
1461             return NVME_INTERNAL_DEV_ERROR;
1462         }
1463 
1464         zeroed = !!(ret & BDRV_BLOCK_ZERO);
1465 
1466         trace_pci_nvme_block_status(offset, bytes, pnum, ret, zeroed);
1467 
1468         if (zeroed) {
1469             return NVME_DULB;
1470         }
1471 
1472         offset += pnum;
1473     } while (pnum != bytes);
1474 
1475     return NVME_SUCCESS;
1476 }
1477 
1478 static void nvme_aio_err(NvmeRequest *req, int ret)
1479 {
1480     uint16_t status = NVME_SUCCESS;
1481     Error *local_err = NULL;
1482 
1483     switch (req->cmd.opcode) {
1484     case NVME_CMD_READ:
1485         status = NVME_UNRECOVERED_READ;
1486         break;
1487     case NVME_CMD_FLUSH:
1488     case NVME_CMD_WRITE:
1489     case NVME_CMD_WRITE_ZEROES:
1490     case NVME_CMD_ZONE_APPEND:
1491         status = NVME_WRITE_FAULT;
1492         break;
1493     default:
1494         status = NVME_INTERNAL_DEV_ERROR;
1495         break;
1496     }
1497 
1498     trace_pci_nvme_err_aio(nvme_cid(req), strerror(-ret), status);
1499 
1500     error_setg_errno(&local_err, -ret, "aio failed");
1501     error_report_err(local_err);
1502 
1503     /*
1504      * Set the command status code to the first encountered error but allow a
1505      * subsequent Internal Device Error to trump it.
1506      */
1507     if (req->status && status != NVME_INTERNAL_DEV_ERROR) {
1508         return;
1509     }
1510 
1511     req->status = status;
1512 }
1513 
1514 static inline uint32_t nvme_zone_idx(NvmeNamespace *ns, uint64_t slba)
1515 {
1516     return ns->zone_size_log2 > 0 ? slba >> ns->zone_size_log2 :
1517                                     slba / ns->zone_size;
1518 }
1519 
1520 static inline NvmeZone *nvme_get_zone_by_slba(NvmeNamespace *ns, uint64_t slba)
1521 {
1522     uint32_t zone_idx = nvme_zone_idx(ns, slba);
1523 
1524     assert(zone_idx < ns->num_zones);
1525     return &ns->zone_array[zone_idx];
1526 }
1527 
1528 static uint16_t nvme_check_zone_state_for_write(NvmeZone *zone)
1529 {
1530     uint64_t zslba = zone->d.zslba;
1531 
1532     switch (nvme_get_zone_state(zone)) {
1533     case NVME_ZONE_STATE_EMPTY:
1534     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1535     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1536     case NVME_ZONE_STATE_CLOSED:
1537         return NVME_SUCCESS;
1538     case NVME_ZONE_STATE_FULL:
1539         trace_pci_nvme_err_zone_is_full(zslba);
1540         return NVME_ZONE_FULL;
1541     case NVME_ZONE_STATE_OFFLINE:
1542         trace_pci_nvme_err_zone_is_offline(zslba);
1543         return NVME_ZONE_OFFLINE;
1544     case NVME_ZONE_STATE_READ_ONLY:
1545         trace_pci_nvme_err_zone_is_read_only(zslba);
1546         return NVME_ZONE_READ_ONLY;
1547     default:
1548         assert(false);
1549     }
1550 
1551     return NVME_INTERNAL_DEV_ERROR;
1552 }
1553 
1554 static uint16_t nvme_check_zone_write(NvmeNamespace *ns, NvmeZone *zone,
1555                                       uint64_t slba, uint32_t nlb)
1556 {
1557     uint64_t zcap = nvme_zone_wr_boundary(zone);
1558     uint16_t status;
1559 
1560     status = nvme_check_zone_state_for_write(zone);
1561     if (status) {
1562         return status;
1563     }
1564 
1565     if (unlikely(slba != zone->w_ptr)) {
1566         trace_pci_nvme_err_write_not_at_wp(slba, zone->d.zslba, zone->w_ptr);
1567         return NVME_ZONE_INVALID_WRITE;
1568     }
1569 
1570     if (unlikely((slba + nlb) > zcap)) {
1571         trace_pci_nvme_err_zone_boundary(slba, nlb, zcap);
1572         return NVME_ZONE_BOUNDARY_ERROR;
1573     }
1574 
1575     return NVME_SUCCESS;
1576 }
1577 
1578 static uint16_t nvme_check_zone_state_for_read(NvmeZone *zone)
1579 {
1580     switch (nvme_get_zone_state(zone)) {
1581     case NVME_ZONE_STATE_EMPTY:
1582     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1583     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1584     case NVME_ZONE_STATE_FULL:
1585     case NVME_ZONE_STATE_CLOSED:
1586     case NVME_ZONE_STATE_READ_ONLY:
1587         return NVME_SUCCESS;
1588     case NVME_ZONE_STATE_OFFLINE:
1589         trace_pci_nvme_err_zone_is_offline(zone->d.zslba);
1590         return NVME_ZONE_OFFLINE;
1591     default:
1592         assert(false);
1593     }
1594 
1595     return NVME_INTERNAL_DEV_ERROR;
1596 }
1597 
1598 static uint16_t nvme_check_zone_read(NvmeNamespace *ns, uint64_t slba,
1599                                      uint32_t nlb)
1600 {
1601     NvmeZone *zone = nvme_get_zone_by_slba(ns, slba);
1602     uint64_t bndry = nvme_zone_rd_boundary(ns, zone);
1603     uint64_t end = slba + nlb;
1604     uint16_t status;
1605 
1606     status = nvme_check_zone_state_for_read(zone);
1607     if (status) {
1608         ;
1609     } else if (unlikely(end > bndry)) {
1610         if (!ns->params.cross_zone_read) {
1611             status = NVME_ZONE_BOUNDARY_ERROR;
1612         } else {
1613             /*
1614              * Read across zone boundary - check that all subsequent
1615              * zones that are being read have an appropriate state.
1616              */
1617             do {
1618                 zone++;
1619                 status = nvme_check_zone_state_for_read(zone);
1620                 if (status) {
1621                     break;
1622                 }
1623             } while (end > nvme_zone_rd_boundary(ns, zone));
1624         }
1625     }
1626 
1627     return status;
1628 }
1629 
1630 static uint16_t nvme_zrm_finish(NvmeNamespace *ns, NvmeZone *zone)
1631 {
1632     switch (nvme_get_zone_state(zone)) {
1633     case NVME_ZONE_STATE_FULL:
1634         return NVME_SUCCESS;
1635 
1636     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1637     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1638         nvme_aor_dec_open(ns);
1639         /* fallthrough */
1640     case NVME_ZONE_STATE_CLOSED:
1641         nvme_aor_dec_active(ns);
1642         /* fallthrough */
1643     case NVME_ZONE_STATE_EMPTY:
1644         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_FULL);
1645         return NVME_SUCCESS;
1646 
1647     default:
1648         return NVME_ZONE_INVAL_TRANSITION;
1649     }
1650 }
1651 
1652 static uint16_t nvme_zrm_close(NvmeNamespace *ns, NvmeZone *zone)
1653 {
1654     switch (nvme_get_zone_state(zone)) {
1655     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1656     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1657         nvme_aor_dec_open(ns);
1658         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
1659         /* fall through */
1660     case NVME_ZONE_STATE_CLOSED:
1661         return NVME_SUCCESS;
1662 
1663     default:
1664         return NVME_ZONE_INVAL_TRANSITION;
1665     }
1666 }
1667 
1668 static void nvme_zrm_auto_transition_zone(NvmeNamespace *ns)
1669 {
1670     NvmeZone *zone;
1671 
1672     if (ns->params.max_open_zones &&
1673         ns->nr_open_zones == ns->params.max_open_zones) {
1674         zone = QTAILQ_FIRST(&ns->imp_open_zones);
1675         if (zone) {
1676             /*
1677              * Automatically close this implicitly open zone.
1678              */
1679             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
1680             nvme_zrm_close(ns, zone);
1681         }
1682     }
1683 }
1684 
1685 enum {
1686     NVME_ZRM_AUTO = 1 << 0,
1687 };
1688 
1689 static uint16_t nvme_zrm_open_flags(NvmeNamespace *ns, NvmeZone *zone,
1690                                     int flags)
1691 {
1692     int act = 0;
1693     uint16_t status;
1694 
1695     switch (nvme_get_zone_state(zone)) {
1696     case NVME_ZONE_STATE_EMPTY:
1697         act = 1;
1698 
1699         /* fallthrough */
1700 
1701     case NVME_ZONE_STATE_CLOSED:
1702         nvme_zrm_auto_transition_zone(ns);
1703         status = nvme_aor_check(ns, act, 1);
1704         if (status) {
1705             return status;
1706         }
1707 
1708         if (act) {
1709             nvme_aor_inc_active(ns);
1710         }
1711 
1712         nvme_aor_inc_open(ns);
1713 
1714         if (flags & NVME_ZRM_AUTO) {
1715             nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_IMPLICITLY_OPEN);
1716             return NVME_SUCCESS;
1717         }
1718 
1719         /* fallthrough */
1720 
1721     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1722         if (flags & NVME_ZRM_AUTO) {
1723             return NVME_SUCCESS;
1724         }
1725 
1726         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EXPLICITLY_OPEN);
1727 
1728         /* fallthrough */
1729 
1730     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1731         return NVME_SUCCESS;
1732 
1733     default:
1734         return NVME_ZONE_INVAL_TRANSITION;
1735     }
1736 }
1737 
1738 static inline uint16_t nvme_zrm_auto(NvmeNamespace *ns, NvmeZone *zone)
1739 {
1740     return nvme_zrm_open_flags(ns, zone, NVME_ZRM_AUTO);
1741 }
1742 
1743 static inline uint16_t nvme_zrm_open(NvmeNamespace *ns, NvmeZone *zone)
1744 {
1745     return nvme_zrm_open_flags(ns, zone, 0);
1746 }
1747 
1748 static void nvme_advance_zone_wp(NvmeNamespace *ns, NvmeZone *zone,
1749                                  uint32_t nlb)
1750 {
1751     zone->d.wp += nlb;
1752 
1753     if (zone->d.wp == nvme_zone_wr_boundary(zone)) {
1754         nvme_zrm_finish(ns, zone);
1755     }
1756 }
1757 
1758 static void nvme_finalize_zoned_write(NvmeNamespace *ns, NvmeRequest *req)
1759 {
1760     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1761     NvmeZone *zone;
1762     uint64_t slba;
1763     uint32_t nlb;
1764 
1765     slba = le64_to_cpu(rw->slba);
1766     nlb = le16_to_cpu(rw->nlb) + 1;
1767     zone = nvme_get_zone_by_slba(ns, slba);
1768 
1769     nvme_advance_zone_wp(ns, zone, nlb);
1770 }
1771 
1772 static inline bool nvme_is_write(NvmeRequest *req)
1773 {
1774     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1775 
1776     return rw->opcode == NVME_CMD_WRITE ||
1777            rw->opcode == NVME_CMD_ZONE_APPEND ||
1778            rw->opcode == NVME_CMD_WRITE_ZEROES;
1779 }
1780 
1781 static void nvme_misc_cb(void *opaque, int ret)
1782 {
1783     NvmeRequest *req = opaque;
1784     NvmeNamespace *ns = req->ns;
1785 
1786     BlockBackend *blk = ns->blkconf.blk;
1787     BlockAcctCookie *acct = &req->acct;
1788     BlockAcctStats *stats = blk_get_stats(blk);
1789 
1790     trace_pci_nvme_misc_cb(nvme_cid(req), blk_name(blk));
1791 
1792     if (ret) {
1793         block_acct_failed(stats, acct);
1794         nvme_aio_err(req, ret);
1795     } else {
1796         block_acct_done(stats, acct);
1797     }
1798 
1799     nvme_enqueue_req_completion(nvme_cq(req), req);
1800 }
1801 
1802 void nvme_rw_complete_cb(void *opaque, int ret)
1803 {
1804     NvmeRequest *req = opaque;
1805     NvmeNamespace *ns = req->ns;
1806     BlockBackend *blk = ns->blkconf.blk;
1807     BlockAcctCookie *acct = &req->acct;
1808     BlockAcctStats *stats = blk_get_stats(blk);
1809 
1810     trace_pci_nvme_rw_complete_cb(nvme_cid(req), blk_name(blk));
1811 
1812     if (ret) {
1813         block_acct_failed(stats, acct);
1814         nvme_aio_err(req, ret);
1815     } else {
1816         block_acct_done(stats, acct);
1817     }
1818 
1819     if (ns->params.zoned && nvme_is_write(req)) {
1820         nvme_finalize_zoned_write(ns, req);
1821     }
1822 
1823     nvme_enqueue_req_completion(nvme_cq(req), req);
1824 }
1825 
1826 static void nvme_rw_cb(void *opaque, int ret)
1827 {
1828     NvmeRequest *req = opaque;
1829     NvmeNamespace *ns = req->ns;
1830 
1831     BlockBackend *blk = ns->blkconf.blk;
1832 
1833     trace_pci_nvme_rw_cb(nvme_cid(req), blk_name(blk));
1834 
1835     if (ret) {
1836         goto out;
1837     }
1838 
1839     if (ns->lbaf.ms) {
1840         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1841         uint64_t slba = le64_to_cpu(rw->slba);
1842         uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
1843         uint64_t offset = nvme_moff(ns, slba);
1844 
1845         if (req->cmd.opcode == NVME_CMD_WRITE_ZEROES) {
1846             size_t mlen = nvme_m2b(ns, nlb);
1847 
1848             req->aiocb = blk_aio_pwrite_zeroes(blk, offset, mlen,
1849                                                BDRV_REQ_MAY_UNMAP,
1850                                                nvme_rw_complete_cb, req);
1851             return;
1852         }
1853 
1854         if (nvme_ns_ext(ns) || req->cmd.mptr) {
1855             uint16_t status;
1856 
1857             nvme_sg_unmap(&req->sg);
1858             status = nvme_map_mdata(nvme_ctrl(req), nlb, req);
1859             if (status) {
1860                 ret = -EFAULT;
1861                 goto out;
1862             }
1863 
1864             if (req->cmd.opcode == NVME_CMD_READ) {
1865                 return nvme_blk_read(blk, offset, nvme_rw_complete_cb, req);
1866             }
1867 
1868             return nvme_blk_write(blk, offset, nvme_rw_complete_cb, req);
1869         }
1870     }
1871 
1872 out:
1873     nvme_rw_complete_cb(req, ret);
1874 }
1875 
1876 struct nvme_aio_format_ctx {
1877     NvmeRequest   *req;
1878     NvmeNamespace *ns;
1879 
1880     /* number of outstanding write zeroes for this namespace */
1881     int *count;
1882 };
1883 
1884 static void nvme_aio_format_cb(void *opaque, int ret)
1885 {
1886     struct nvme_aio_format_ctx *ctx = opaque;
1887     NvmeRequest *req = ctx->req;
1888     NvmeNamespace *ns = ctx->ns;
1889     uintptr_t *num_formats = (uintptr_t *)&req->opaque;
1890     int *count = ctx->count;
1891 
1892     g_free(ctx);
1893 
1894     if (ret) {
1895         nvme_aio_err(req, ret);
1896     }
1897 
1898     if (--(*count)) {
1899         return;
1900     }
1901 
1902     g_free(count);
1903     ns->status = 0x0;
1904 
1905     if (--(*num_formats)) {
1906         return;
1907     }
1908 
1909     nvme_enqueue_req_completion(nvme_cq(req), req);
1910 }
1911 
1912 struct nvme_aio_flush_ctx {
1913     NvmeRequest     *req;
1914     NvmeNamespace   *ns;
1915     BlockAcctCookie acct;
1916 };
1917 
1918 static void nvme_aio_flush_cb(void *opaque, int ret)
1919 {
1920     struct nvme_aio_flush_ctx *ctx = opaque;
1921     NvmeRequest *req = ctx->req;
1922     uintptr_t *num_flushes = (uintptr_t *)&req->opaque;
1923 
1924     BlockBackend *blk = ctx->ns->blkconf.blk;
1925     BlockAcctCookie *acct = &ctx->acct;
1926     BlockAcctStats *stats = blk_get_stats(blk);
1927 
1928     trace_pci_nvme_aio_flush_cb(nvme_cid(req), blk_name(blk));
1929 
1930     if (!ret) {
1931         block_acct_done(stats, acct);
1932     } else {
1933         block_acct_failed(stats, acct);
1934         nvme_aio_err(req, ret);
1935     }
1936 
1937     (*num_flushes)--;
1938     g_free(ctx);
1939 
1940     if (*num_flushes) {
1941         return;
1942     }
1943 
1944     nvme_enqueue_req_completion(nvme_cq(req), req);
1945 }
1946 
1947 static void nvme_verify_cb(void *opaque, int ret)
1948 {
1949     NvmeBounceContext *ctx = opaque;
1950     NvmeRequest *req = ctx->req;
1951     NvmeNamespace *ns = req->ns;
1952     BlockBackend *blk = ns->blkconf.blk;
1953     BlockAcctCookie *acct = &req->acct;
1954     BlockAcctStats *stats = blk_get_stats(blk);
1955     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1956     uint64_t slba = le64_to_cpu(rw->slba);
1957     uint16_t ctrl = le16_to_cpu(rw->control);
1958     uint16_t apptag = le16_to_cpu(rw->apptag);
1959     uint16_t appmask = le16_to_cpu(rw->appmask);
1960     uint32_t reftag = le32_to_cpu(rw->reftag);
1961     uint16_t status;
1962 
1963     trace_pci_nvme_verify_cb(nvme_cid(req), NVME_RW_PRINFO(ctrl), apptag,
1964                              appmask, reftag);
1965 
1966     if (ret) {
1967         block_acct_failed(stats, acct);
1968         nvme_aio_err(req, ret);
1969         goto out;
1970     }
1971 
1972     block_acct_done(stats, acct);
1973 
1974     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
1975         status = nvme_dif_mangle_mdata(ns, ctx->mdata.bounce,
1976                                        ctx->mdata.iov.size, slba);
1977         if (status) {
1978             req->status = status;
1979             goto out;
1980         }
1981 
1982         req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
1983                                      ctx->mdata.bounce, ctx->mdata.iov.size,
1984                                      ctrl, slba, apptag, appmask, reftag);
1985     }
1986 
1987 out:
1988     qemu_iovec_destroy(&ctx->data.iov);
1989     g_free(ctx->data.bounce);
1990 
1991     qemu_iovec_destroy(&ctx->mdata.iov);
1992     g_free(ctx->mdata.bounce);
1993 
1994     g_free(ctx);
1995 
1996     nvme_enqueue_req_completion(nvme_cq(req), req);
1997 }
1998 
1999 
2000 static void nvme_verify_mdata_in_cb(void *opaque, int ret)
2001 {
2002     NvmeBounceContext *ctx = opaque;
2003     NvmeRequest *req = ctx->req;
2004     NvmeNamespace *ns = req->ns;
2005     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2006     uint64_t slba = le64_to_cpu(rw->slba);
2007     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2008     size_t mlen = nvme_m2b(ns, nlb);
2009     uint64_t offset = nvme_moff(ns, slba);
2010     BlockBackend *blk = ns->blkconf.blk;
2011 
2012     trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req), blk_name(blk));
2013 
2014     if (ret) {
2015         goto out;
2016     }
2017 
2018     ctx->mdata.bounce = g_malloc(mlen);
2019 
2020     qemu_iovec_reset(&ctx->mdata.iov);
2021     qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2022 
2023     req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2024                                 nvme_verify_cb, ctx);
2025     return;
2026 
2027 out:
2028     nvme_verify_cb(ctx, ret);
2029 }
2030 
2031 static void nvme_aio_discard_cb(void *opaque, int ret)
2032 {
2033     NvmeRequest *req = opaque;
2034     uintptr_t *discards = (uintptr_t *)&req->opaque;
2035 
2036     trace_pci_nvme_aio_discard_cb(nvme_cid(req));
2037 
2038     if (ret) {
2039         nvme_aio_err(req, ret);
2040     }
2041 
2042     (*discards)--;
2043 
2044     if (*discards) {
2045         return;
2046     }
2047 
2048     nvme_enqueue_req_completion(nvme_cq(req), req);
2049 }
2050 
2051 struct nvme_zone_reset_ctx {
2052     NvmeRequest *req;
2053     NvmeZone    *zone;
2054 };
2055 
2056 static void nvme_aio_zone_reset_complete_cb(void *opaque, int ret)
2057 {
2058     struct nvme_zone_reset_ctx *ctx = opaque;
2059     NvmeRequest *req = ctx->req;
2060     NvmeNamespace *ns = req->ns;
2061     NvmeZone *zone = ctx->zone;
2062     uintptr_t *resets = (uintptr_t *)&req->opaque;
2063 
2064     if (ret) {
2065         nvme_aio_err(req, ret);
2066         goto out;
2067     }
2068 
2069     switch (nvme_get_zone_state(zone)) {
2070     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
2071     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
2072         nvme_aor_dec_open(ns);
2073         /* fall through */
2074     case NVME_ZONE_STATE_CLOSED:
2075         nvme_aor_dec_active(ns);
2076         /* fall through */
2077     case NVME_ZONE_STATE_FULL:
2078         zone->w_ptr = zone->d.zslba;
2079         zone->d.wp = zone->w_ptr;
2080         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
2081         /* fall through */
2082     default:
2083         break;
2084     }
2085 
2086 out:
2087     g_free(ctx);
2088 
2089     (*resets)--;
2090 
2091     if (*resets) {
2092         return;
2093     }
2094 
2095     nvme_enqueue_req_completion(nvme_cq(req), req);
2096 }
2097 
2098 static void nvme_aio_zone_reset_cb(void *opaque, int ret)
2099 {
2100     struct nvme_zone_reset_ctx *ctx = opaque;
2101     NvmeRequest *req = ctx->req;
2102     NvmeNamespace *ns = req->ns;
2103     NvmeZone *zone = ctx->zone;
2104 
2105     trace_pci_nvme_aio_zone_reset_cb(nvme_cid(req), zone->d.zslba);
2106 
2107     if (ret) {
2108         goto out;
2109     }
2110 
2111     if (ns->lbaf.ms) {
2112         int64_t offset = nvme_moff(ns, zone->d.zslba);
2113 
2114         blk_aio_pwrite_zeroes(ns->blkconf.blk, offset,
2115                               nvme_m2b(ns, ns->zone_size), BDRV_REQ_MAY_UNMAP,
2116                               nvme_aio_zone_reset_complete_cb, ctx);
2117         return;
2118     }
2119 
2120 out:
2121     nvme_aio_zone_reset_complete_cb(opaque, ret);
2122 }
2123 
2124 struct nvme_copy_ctx {
2125     int copies;
2126     uint8_t *bounce;
2127     uint8_t *mbounce;
2128     uint32_t nlb;
2129     NvmeCopySourceRange *ranges;
2130 };
2131 
2132 struct nvme_copy_in_ctx {
2133     NvmeRequest *req;
2134     QEMUIOVector iov;
2135     NvmeCopySourceRange *range;
2136 };
2137 
2138 static void nvme_copy_complete_cb(void *opaque, int ret)
2139 {
2140     NvmeRequest *req = opaque;
2141     NvmeNamespace *ns = req->ns;
2142     struct nvme_copy_ctx *ctx = req->opaque;
2143 
2144     if (ret) {
2145         block_acct_failed(blk_get_stats(ns->blkconf.blk), &req->acct);
2146         nvme_aio_err(req, ret);
2147         goto out;
2148     }
2149 
2150     block_acct_done(blk_get_stats(ns->blkconf.blk), &req->acct);
2151 
2152 out:
2153     if (ns->params.zoned) {
2154         NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2155         uint64_t sdlba = le64_to_cpu(copy->sdlba);
2156         NvmeZone *zone = nvme_get_zone_by_slba(ns, sdlba);
2157 
2158         nvme_advance_zone_wp(ns, zone, ctx->nlb);
2159     }
2160 
2161     g_free(ctx->bounce);
2162     g_free(ctx->mbounce);
2163     g_free(ctx);
2164 
2165     nvme_enqueue_req_completion(nvme_cq(req), req);
2166 }
2167 
2168 static void nvme_copy_cb(void *opaque, int ret)
2169 {
2170     NvmeRequest *req = opaque;
2171     NvmeNamespace *ns = req->ns;
2172     struct nvme_copy_ctx *ctx = req->opaque;
2173 
2174     trace_pci_nvme_copy_cb(nvme_cid(req));
2175 
2176     if (ret) {
2177         goto out;
2178     }
2179 
2180     if (ns->lbaf.ms) {
2181         NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2182         uint64_t sdlba = le64_to_cpu(copy->sdlba);
2183         int64_t offset = nvme_moff(ns, sdlba);
2184 
2185         qemu_iovec_reset(&req->sg.iov);
2186         qemu_iovec_add(&req->sg.iov, ctx->mbounce, nvme_m2b(ns, ctx->nlb));
2187 
2188         req->aiocb = blk_aio_pwritev(ns->blkconf.blk, offset, &req->sg.iov, 0,
2189                                      nvme_copy_complete_cb, req);
2190         return;
2191     }
2192 
2193 out:
2194     nvme_copy_complete_cb(opaque, ret);
2195 }
2196 
2197 static void nvme_copy_in_complete(NvmeRequest *req)
2198 {
2199     NvmeNamespace *ns = req->ns;
2200     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2201     struct nvme_copy_ctx *ctx = req->opaque;
2202     uint64_t sdlba = le64_to_cpu(copy->sdlba);
2203     uint16_t status;
2204 
2205     trace_pci_nvme_copy_in_complete(nvme_cid(req));
2206 
2207     block_acct_done(blk_get_stats(ns->blkconf.blk), &req->acct);
2208 
2209     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2210         uint16_t prinfor = (copy->control[0] >> 4) & 0xf;
2211         uint16_t prinfow = (copy->control[2] >> 2) & 0xf;
2212         uint16_t nr = copy->nr + 1;
2213         NvmeCopySourceRange *range;
2214         uint64_t slba;
2215         uint32_t nlb;
2216         uint16_t apptag, appmask;
2217         uint32_t reftag;
2218         uint8_t *buf = ctx->bounce, *mbuf = ctx->mbounce;
2219         size_t len, mlen;
2220         int i;
2221 
2222         /*
2223          * The dif helpers expects prinfo to be similar to the control field of
2224          * the NvmeRwCmd, so shift by 10 to fake it.
2225          */
2226         prinfor = prinfor << 10;
2227         prinfow = prinfow << 10;
2228 
2229         for (i = 0; i < nr; i++) {
2230             range = &ctx->ranges[i];
2231             slba = le64_to_cpu(range->slba);
2232             nlb = le16_to_cpu(range->nlb) + 1;
2233             len = nvme_l2b(ns, nlb);
2234             mlen = nvme_m2b(ns, nlb);
2235             apptag = le16_to_cpu(range->apptag);
2236             appmask = le16_to_cpu(range->appmask);
2237             reftag = le32_to_cpu(range->reftag);
2238 
2239             status = nvme_dif_check(ns, buf, len, mbuf, mlen, prinfor, slba,
2240                                     apptag, appmask, reftag);
2241             if (status) {
2242                 goto invalid;
2243             }
2244 
2245             buf += len;
2246             mbuf += mlen;
2247         }
2248 
2249         apptag = le16_to_cpu(copy->apptag);
2250         appmask = le16_to_cpu(copy->appmask);
2251         reftag = le32_to_cpu(copy->reftag);
2252 
2253         if (prinfow & NVME_RW_PRINFO_PRACT) {
2254             size_t len = nvme_l2b(ns, ctx->nlb);
2255             size_t mlen = nvme_m2b(ns, ctx->nlb);
2256 
2257             status = nvme_check_prinfo(ns, prinfow, sdlba, reftag);
2258             if (status) {
2259                 goto invalid;
2260             }
2261 
2262             nvme_dif_pract_generate_dif(ns, ctx->bounce, len, ctx->mbounce,
2263                                         mlen, apptag, reftag);
2264         } else {
2265             status = nvme_dif_check(ns, ctx->bounce, len, ctx->mbounce, mlen,
2266                                     prinfow, sdlba, apptag, appmask, reftag);
2267             if (status) {
2268                 goto invalid;
2269             }
2270         }
2271     }
2272 
2273     status = nvme_check_bounds(ns, sdlba, ctx->nlb);
2274     if (status) {
2275         goto invalid;
2276     }
2277 
2278     if (ns->params.zoned) {
2279         NvmeZone *zone = nvme_get_zone_by_slba(ns, sdlba);
2280 
2281         status = nvme_check_zone_write(ns, zone, sdlba, ctx->nlb);
2282         if (status) {
2283             goto invalid;
2284         }
2285 
2286         status = nvme_zrm_auto(ns, zone);
2287         if (status) {
2288             goto invalid;
2289         }
2290 
2291         zone->w_ptr += ctx->nlb;
2292     }
2293 
2294     qemu_iovec_init(&req->sg.iov, 1);
2295     qemu_iovec_add(&req->sg.iov, ctx->bounce, nvme_l2b(ns, ctx->nlb));
2296 
2297     block_acct_start(blk_get_stats(ns->blkconf.blk), &req->acct, 0,
2298                      BLOCK_ACCT_WRITE);
2299 
2300     req->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, sdlba),
2301                                  &req->sg.iov, 0, nvme_copy_cb, req);
2302 
2303     return;
2304 
2305 invalid:
2306     req->status = status;
2307 
2308     g_free(ctx->bounce);
2309     g_free(ctx);
2310 
2311     nvme_enqueue_req_completion(nvme_cq(req), req);
2312 }
2313 
2314 static void nvme_aio_copy_in_cb(void *opaque, int ret)
2315 {
2316     struct nvme_copy_in_ctx *in_ctx = opaque;
2317     NvmeRequest *req = in_ctx->req;
2318     NvmeNamespace *ns = req->ns;
2319     struct nvme_copy_ctx *ctx = req->opaque;
2320 
2321     qemu_iovec_destroy(&in_ctx->iov);
2322     g_free(in_ctx);
2323 
2324     trace_pci_nvme_aio_copy_in_cb(nvme_cid(req));
2325 
2326     if (ret) {
2327         nvme_aio_err(req, ret);
2328     }
2329 
2330     ctx->copies--;
2331 
2332     if (ctx->copies) {
2333         return;
2334     }
2335 
2336     if (req->status) {
2337         block_acct_failed(blk_get_stats(ns->blkconf.blk), &req->acct);
2338 
2339         g_free(ctx->bounce);
2340         g_free(ctx->mbounce);
2341         g_free(ctx);
2342 
2343         nvme_enqueue_req_completion(nvme_cq(req), req);
2344 
2345         return;
2346     }
2347 
2348     nvme_copy_in_complete(req);
2349 }
2350 
2351 struct nvme_compare_ctx {
2352     struct {
2353         QEMUIOVector iov;
2354         uint8_t *bounce;
2355     } data;
2356 
2357     struct {
2358         QEMUIOVector iov;
2359         uint8_t *bounce;
2360     } mdata;
2361 };
2362 
2363 static void nvme_compare_mdata_cb(void *opaque, int ret)
2364 {
2365     NvmeRequest *req = opaque;
2366     NvmeNamespace *ns = req->ns;
2367     NvmeCtrl *n = nvme_ctrl(req);
2368     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2369     uint16_t ctrl = le16_to_cpu(rw->control);
2370     uint16_t apptag = le16_to_cpu(rw->apptag);
2371     uint16_t appmask = le16_to_cpu(rw->appmask);
2372     uint32_t reftag = le32_to_cpu(rw->reftag);
2373     struct nvme_compare_ctx *ctx = req->opaque;
2374     g_autofree uint8_t *buf = NULL;
2375     BlockBackend *blk = ns->blkconf.blk;
2376     BlockAcctCookie *acct = &req->acct;
2377     BlockAcctStats *stats = blk_get_stats(blk);
2378     uint16_t status = NVME_SUCCESS;
2379 
2380     trace_pci_nvme_compare_mdata_cb(nvme_cid(req));
2381 
2382     if (ret) {
2383         block_acct_failed(stats, acct);
2384         nvme_aio_err(req, ret);
2385         goto out;
2386     }
2387 
2388     buf = g_malloc(ctx->mdata.iov.size);
2389 
2390     status = nvme_bounce_mdata(n, buf, ctx->mdata.iov.size,
2391                                NVME_TX_DIRECTION_TO_DEVICE, req);
2392     if (status) {
2393         req->status = status;
2394         goto out;
2395     }
2396 
2397     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2398         uint64_t slba = le64_to_cpu(rw->slba);
2399         uint8_t *bufp;
2400         uint8_t *mbufp = ctx->mdata.bounce;
2401         uint8_t *end = mbufp + ctx->mdata.iov.size;
2402         int16_t pil = 0;
2403 
2404         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2405                                 ctx->mdata.bounce, ctx->mdata.iov.size, ctrl,
2406                                 slba, apptag, appmask, reftag);
2407         if (status) {
2408             req->status = status;
2409             goto out;
2410         }
2411 
2412         /*
2413          * When formatted with protection information, do not compare the DIF
2414          * tuple.
2415          */
2416         if (!(ns->id_ns.dps & NVME_ID_NS_DPS_FIRST_EIGHT)) {
2417             pil = ns->lbaf.ms - sizeof(NvmeDifTuple);
2418         }
2419 
2420         for (bufp = buf; mbufp < end; bufp += ns->lbaf.ms, mbufp += ns->lbaf.ms) {
2421             if (memcmp(bufp + pil, mbufp + pil, ns->lbaf.ms - pil)) {
2422                 req->status = NVME_CMP_FAILURE;
2423                 goto out;
2424             }
2425         }
2426 
2427         goto out;
2428     }
2429 
2430     if (memcmp(buf, ctx->mdata.bounce, ctx->mdata.iov.size)) {
2431         req->status = NVME_CMP_FAILURE;
2432         goto out;
2433     }
2434 
2435     block_acct_done(stats, acct);
2436 
2437 out:
2438     qemu_iovec_destroy(&ctx->data.iov);
2439     g_free(ctx->data.bounce);
2440 
2441     qemu_iovec_destroy(&ctx->mdata.iov);
2442     g_free(ctx->mdata.bounce);
2443 
2444     g_free(ctx);
2445 
2446     nvme_enqueue_req_completion(nvme_cq(req), req);
2447 }
2448 
2449 static void nvme_compare_data_cb(void *opaque, int ret)
2450 {
2451     NvmeRequest *req = opaque;
2452     NvmeCtrl *n = nvme_ctrl(req);
2453     NvmeNamespace *ns = req->ns;
2454     BlockBackend *blk = ns->blkconf.blk;
2455     BlockAcctCookie *acct = &req->acct;
2456     BlockAcctStats *stats = blk_get_stats(blk);
2457 
2458     struct nvme_compare_ctx *ctx = req->opaque;
2459     g_autofree uint8_t *buf = NULL;
2460     uint16_t status;
2461 
2462     trace_pci_nvme_compare_data_cb(nvme_cid(req));
2463 
2464     if (ret) {
2465         block_acct_failed(stats, acct);
2466         nvme_aio_err(req, ret);
2467         goto out;
2468     }
2469 
2470     buf = g_malloc(ctx->data.iov.size);
2471 
2472     status = nvme_bounce_data(n, buf, ctx->data.iov.size,
2473                               NVME_TX_DIRECTION_TO_DEVICE, req);
2474     if (status) {
2475         req->status = status;
2476         goto out;
2477     }
2478 
2479     if (memcmp(buf, ctx->data.bounce, ctx->data.iov.size)) {
2480         req->status = NVME_CMP_FAILURE;
2481         goto out;
2482     }
2483 
2484     if (ns->lbaf.ms) {
2485         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2486         uint64_t slba = le64_to_cpu(rw->slba);
2487         uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2488         size_t mlen = nvme_m2b(ns, nlb);
2489         uint64_t offset = nvme_moff(ns, slba);
2490 
2491         ctx->mdata.bounce = g_malloc(mlen);
2492 
2493         qemu_iovec_init(&ctx->mdata.iov, 1);
2494         qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2495 
2496         req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2497                                     nvme_compare_mdata_cb, req);
2498         return;
2499     }
2500 
2501     block_acct_done(stats, acct);
2502 
2503 out:
2504     qemu_iovec_destroy(&ctx->data.iov);
2505     g_free(ctx->data.bounce);
2506     g_free(ctx);
2507 
2508     nvme_enqueue_req_completion(nvme_cq(req), req);
2509 }
2510 
2511 static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
2512 {
2513     NvmeNamespace *ns = req->ns;
2514     NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
2515 
2516     uint32_t attr = le32_to_cpu(dsm->attributes);
2517     uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
2518 
2519     uint16_t status = NVME_SUCCESS;
2520 
2521     trace_pci_nvme_dsm(nvme_cid(req), nvme_nsid(ns), nr, attr);
2522 
2523     if (attr & NVME_DSMGMT_AD) {
2524         int64_t offset;
2525         size_t len;
2526         NvmeDsmRange range[nr];
2527         uintptr_t *discards = (uintptr_t *)&req->opaque;
2528 
2529         status = nvme_h2c(n, (uint8_t *)range, sizeof(range), req);
2530         if (status) {
2531             return status;
2532         }
2533 
2534         /*
2535          * AIO callbacks may be called immediately, so initialize discards to 1
2536          * to make sure the the callback does not complete the request before
2537          * all discards have been issued.
2538          */
2539         *discards = 1;
2540 
2541         for (int i = 0; i < nr; i++) {
2542             uint64_t slba = le64_to_cpu(range[i].slba);
2543             uint32_t nlb = le32_to_cpu(range[i].nlb);
2544 
2545             if (nvme_check_bounds(ns, slba, nlb)) {
2546                 continue;
2547             }
2548 
2549             trace_pci_nvme_dsm_deallocate(nvme_cid(req), nvme_nsid(ns), slba,
2550                                           nlb);
2551 
2552             if (nlb > n->dmrsl) {
2553                 trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
2554             }
2555 
2556             offset = nvme_l2b(ns, slba);
2557             len = nvme_l2b(ns, nlb);
2558 
2559             while (len) {
2560                 size_t bytes = MIN(BDRV_REQUEST_MAX_BYTES, len);
2561 
2562                 (*discards)++;
2563 
2564                 blk_aio_pdiscard(ns->blkconf.blk, offset, bytes,
2565                                  nvme_aio_discard_cb, req);
2566 
2567                 offset += bytes;
2568                 len -= bytes;
2569             }
2570         }
2571 
2572         /* account for the 1-initialization */
2573         (*discards)--;
2574 
2575         if (*discards) {
2576             status = NVME_NO_COMPLETE;
2577         } else {
2578             status = req->status;
2579         }
2580     }
2581 
2582     return status;
2583 }
2584 
2585 static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
2586 {
2587     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2588     NvmeNamespace *ns = req->ns;
2589     BlockBackend *blk = ns->blkconf.blk;
2590     uint64_t slba = le64_to_cpu(rw->slba);
2591     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2592     size_t len = nvme_l2b(ns, nlb);
2593     int64_t offset = nvme_l2b(ns, slba);
2594     uint16_t ctrl = le16_to_cpu(rw->control);
2595     uint32_t reftag = le32_to_cpu(rw->reftag);
2596     NvmeBounceContext *ctx = NULL;
2597     uint16_t status;
2598 
2599     trace_pci_nvme_verify(nvme_cid(req), nvme_nsid(ns), slba, nlb);
2600 
2601     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2602         status = nvme_check_prinfo(ns, ctrl, slba, reftag);
2603         if (status) {
2604             return status;
2605         }
2606 
2607         if (ctrl & NVME_RW_PRINFO_PRACT) {
2608             return NVME_INVALID_PROT_INFO | NVME_DNR;
2609         }
2610     }
2611 
2612     if (len > n->page_size << n->params.vsl) {
2613         return NVME_INVALID_FIELD | NVME_DNR;
2614     }
2615 
2616     status = nvme_check_bounds(ns, slba, nlb);
2617     if (status) {
2618         return status;
2619     }
2620 
2621     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2622         status = nvme_check_dulbe(ns, slba, nlb);
2623         if (status) {
2624             return status;
2625         }
2626     }
2627 
2628     ctx = g_new0(NvmeBounceContext, 1);
2629     ctx->req = req;
2630 
2631     ctx->data.bounce = g_malloc(len);
2632 
2633     qemu_iovec_init(&ctx->data.iov, 1);
2634     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, len);
2635 
2636     block_acct_start(blk_get_stats(blk), &req->acct, ctx->data.iov.size,
2637                      BLOCK_ACCT_READ);
2638 
2639     req->aiocb = blk_aio_preadv(ns->blkconf.blk, offset, &ctx->data.iov, 0,
2640                                 nvme_verify_mdata_in_cb, ctx);
2641     return NVME_NO_COMPLETE;
2642 }
2643 
2644 static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
2645 {
2646     NvmeNamespace *ns = req->ns;
2647     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2648 
2649     uint16_t nr = copy->nr + 1;
2650     uint8_t format = copy->control[0] & 0xf;
2651 
2652     /*
2653      * Shift the PRINFOR/PRINFOW values by 10 to allow reusing the
2654      * NVME_RW_PRINFO constants.
2655      */
2656     uint16_t prinfor = ((copy->control[0] >> 4) & 0xf) << 10;
2657     uint16_t prinfow = ((copy->control[2] >> 2) & 0xf) << 10;
2658 
2659     uint32_t nlb = 0;
2660     uint8_t *bounce = NULL, *bouncep = NULL;
2661     uint8_t *mbounce = NULL, *mbouncep = NULL;
2662     struct nvme_copy_ctx *ctx;
2663     uint16_t status;
2664     int i;
2665 
2666     trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);
2667 
2668     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) &&
2669         ((prinfor & NVME_RW_PRINFO_PRACT) != (prinfow & NVME_RW_PRINFO_PRACT))) {
2670         return NVME_INVALID_FIELD | NVME_DNR;
2671     }
2672 
2673     if (!(n->id_ctrl.ocfs & (1 << format))) {
2674         trace_pci_nvme_err_copy_invalid_format(format);
2675         return NVME_INVALID_FIELD | NVME_DNR;
2676     }
2677 
2678     if (nr > ns->id_ns.msrc + 1) {
2679         return NVME_CMD_SIZE_LIMIT | NVME_DNR;
2680     }
2681 
2682     ctx = g_new(struct nvme_copy_ctx, 1);
2683     ctx->ranges = g_new(NvmeCopySourceRange, nr);
2684 
2685     status = nvme_h2c(n, (uint8_t *)ctx->ranges,
2686                       nr * sizeof(NvmeCopySourceRange), req);
2687     if (status) {
2688         goto out;
2689     }
2690 
2691     for (i = 0; i < nr; i++) {
2692         uint64_t slba = le64_to_cpu(ctx->ranges[i].slba);
2693         uint32_t _nlb = le16_to_cpu(ctx->ranges[i].nlb) + 1;
2694 
2695         if (_nlb > le16_to_cpu(ns->id_ns.mssrl)) {
2696             status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
2697             goto out;
2698         }
2699 
2700         status = nvme_check_bounds(ns, slba, _nlb);
2701         if (status) {
2702             goto out;
2703         }
2704 
2705         if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2706             status = nvme_check_dulbe(ns, slba, _nlb);
2707             if (status) {
2708                 goto out;
2709             }
2710         }
2711 
2712         if (ns->params.zoned) {
2713             status = nvme_check_zone_read(ns, slba, _nlb);
2714             if (status) {
2715                 goto out;
2716             }
2717         }
2718 
2719         nlb += _nlb;
2720     }
2721 
2722     if (nlb > le32_to_cpu(ns->id_ns.mcl)) {
2723         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
2724         goto out;
2725     }
2726 
2727     bounce = bouncep = g_malloc(nvme_l2b(ns, nlb));
2728     if (ns->lbaf.ms) {
2729         mbounce = mbouncep = g_malloc(nvme_m2b(ns, nlb));
2730     }
2731 
2732     block_acct_start(blk_get_stats(ns->blkconf.blk), &req->acct, 0,
2733                      BLOCK_ACCT_READ);
2734 
2735     ctx->bounce = bounce;
2736     ctx->mbounce = mbounce;
2737     ctx->nlb = nlb;
2738     ctx->copies = 1;
2739 
2740     req->opaque = ctx;
2741 
2742     for (i = 0; i < nr; i++) {
2743         uint64_t slba = le64_to_cpu(ctx->ranges[i].slba);
2744         uint32_t nlb = le16_to_cpu(ctx->ranges[i].nlb) + 1;
2745 
2746         size_t len = nvme_l2b(ns, nlb);
2747         int64_t offset = nvme_l2b(ns, slba);
2748 
2749         trace_pci_nvme_copy_source_range(slba, nlb);
2750 
2751         struct nvme_copy_in_ctx *in_ctx = g_new(struct nvme_copy_in_ctx, 1);
2752         in_ctx->req = req;
2753 
2754         qemu_iovec_init(&in_ctx->iov, 1);
2755         qemu_iovec_add(&in_ctx->iov, bouncep, len);
2756 
2757         ctx->copies++;
2758 
2759         blk_aio_preadv(ns->blkconf.blk, offset, &in_ctx->iov, 0,
2760                        nvme_aio_copy_in_cb, in_ctx);
2761 
2762         bouncep += len;
2763 
2764         if (ns->lbaf.ms) {
2765             len = nvme_m2b(ns, nlb);
2766             offset = nvme_moff(ns, slba);
2767 
2768             in_ctx = g_new(struct nvme_copy_in_ctx, 1);
2769             in_ctx->req = req;
2770 
2771             qemu_iovec_init(&in_ctx->iov, 1);
2772             qemu_iovec_add(&in_ctx->iov, mbouncep, len);
2773 
2774             ctx->copies++;
2775 
2776             blk_aio_preadv(ns->blkconf.blk, offset, &in_ctx->iov, 0,
2777                            nvme_aio_copy_in_cb, in_ctx);
2778 
2779             mbouncep += len;
2780         }
2781     }
2782 
2783     /* account for the 1-initialization */
2784     ctx->copies--;
2785 
2786     if (!ctx->copies) {
2787         nvme_copy_in_complete(req);
2788     }
2789 
2790     return NVME_NO_COMPLETE;
2791 
2792 out:
2793     g_free(ctx->ranges);
2794     g_free(ctx);
2795 
2796     return status;
2797 }
2798 
2799 static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req)
2800 {
2801     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2802     NvmeNamespace *ns = req->ns;
2803     BlockBackend *blk = ns->blkconf.blk;
2804     uint64_t slba = le64_to_cpu(rw->slba);
2805     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2806     uint16_t ctrl = le16_to_cpu(rw->control);
2807     size_t data_len = nvme_l2b(ns, nlb);
2808     size_t len = data_len;
2809     int64_t offset = nvme_l2b(ns, slba);
2810     struct nvme_compare_ctx *ctx = NULL;
2811     uint16_t status;
2812 
2813     trace_pci_nvme_compare(nvme_cid(req), nvme_nsid(ns), slba, nlb);
2814 
2815     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) && (ctrl & NVME_RW_PRINFO_PRACT)) {
2816         return NVME_INVALID_PROT_INFO | NVME_DNR;
2817     }
2818 
2819     if (nvme_ns_ext(ns)) {
2820         len += nvme_m2b(ns, nlb);
2821     }
2822 
2823     status = nvme_check_mdts(n, len);
2824     if (status) {
2825         return status;
2826     }
2827 
2828     status = nvme_check_bounds(ns, slba, nlb);
2829     if (status) {
2830         return status;
2831     }
2832 
2833     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2834         status = nvme_check_dulbe(ns, slba, nlb);
2835         if (status) {
2836             return status;
2837         }
2838     }
2839 
2840     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
2841     if (status) {
2842         return status;
2843     }
2844 
2845     ctx = g_new(struct nvme_compare_ctx, 1);
2846     ctx->data.bounce = g_malloc(data_len);
2847 
2848     req->opaque = ctx;
2849 
2850     qemu_iovec_init(&ctx->data.iov, 1);
2851     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, data_len);
2852 
2853     block_acct_start(blk_get_stats(blk), &req->acct, data_len,
2854                      BLOCK_ACCT_READ);
2855     req->aiocb = blk_aio_preadv(blk, offset, &ctx->data.iov, 0,
2856                                 nvme_compare_data_cb, req);
2857 
2858     return NVME_NO_COMPLETE;
2859 }
2860 
2861 static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req)
2862 {
2863     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
2864     uintptr_t *num_flushes = (uintptr_t *)&req->opaque;
2865     uint16_t status;
2866     struct nvme_aio_flush_ctx *ctx;
2867     NvmeNamespace *ns;
2868 
2869     trace_pci_nvme_flush(nvme_cid(req), nsid);
2870 
2871     if (nsid != NVME_NSID_BROADCAST) {
2872         req->ns = nvme_ns(n, nsid);
2873         if (unlikely(!req->ns)) {
2874             return NVME_INVALID_FIELD | NVME_DNR;
2875         }
2876 
2877         block_acct_start(blk_get_stats(req->ns->blkconf.blk), &req->acct, 0,
2878                          BLOCK_ACCT_FLUSH);
2879         req->aiocb = blk_aio_flush(req->ns->blkconf.blk, nvme_misc_cb, req);
2880         return NVME_NO_COMPLETE;
2881     }
2882 
2883     /* 1-initialize; see comment in nvme_dsm */
2884     *num_flushes = 1;
2885 
2886     for (int i = 1; i <= NVME_MAX_NAMESPACES; i++) {
2887         ns = nvme_ns(n, i);
2888         if (!ns) {
2889             continue;
2890         }
2891 
2892         ctx = g_new(struct nvme_aio_flush_ctx, 1);
2893         ctx->req = req;
2894         ctx->ns = ns;
2895 
2896         (*num_flushes)++;
2897 
2898         block_acct_start(blk_get_stats(ns->blkconf.blk), &ctx->acct, 0,
2899                          BLOCK_ACCT_FLUSH);
2900         blk_aio_flush(ns->blkconf.blk, nvme_aio_flush_cb, ctx);
2901     }
2902 
2903     /* account for the 1-initialization */
2904     (*num_flushes)--;
2905 
2906     if (*num_flushes) {
2907         status = NVME_NO_COMPLETE;
2908     } else {
2909         status = req->status;
2910     }
2911 
2912     return status;
2913 }
2914 
2915 static uint16_t nvme_read(NvmeCtrl *n, NvmeRequest *req)
2916 {
2917     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2918     NvmeNamespace *ns = req->ns;
2919     uint64_t slba = le64_to_cpu(rw->slba);
2920     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
2921     uint16_t ctrl = le16_to_cpu(rw->control);
2922     uint64_t data_size = nvme_l2b(ns, nlb);
2923     uint64_t mapped_size = data_size;
2924     uint64_t data_offset;
2925     BlockBackend *blk = ns->blkconf.blk;
2926     uint16_t status;
2927 
2928     if (nvme_ns_ext(ns)) {
2929         mapped_size += nvme_m2b(ns, nlb);
2930 
2931         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2932             bool pract = ctrl & NVME_RW_PRINFO_PRACT;
2933 
2934             if (pract && ns->lbaf.ms == 8) {
2935                 mapped_size = data_size;
2936             }
2937         }
2938     }
2939 
2940     trace_pci_nvme_read(nvme_cid(req), nvme_nsid(ns), nlb, mapped_size, slba);
2941 
2942     status = nvme_check_mdts(n, mapped_size);
2943     if (status) {
2944         goto invalid;
2945     }
2946 
2947     status = nvme_check_bounds(ns, slba, nlb);
2948     if (status) {
2949         goto invalid;
2950     }
2951 
2952     if (ns->params.zoned) {
2953         status = nvme_check_zone_read(ns, slba, nlb);
2954         if (status) {
2955             trace_pci_nvme_err_zone_read_not_ok(slba, nlb, status);
2956             goto invalid;
2957         }
2958     }
2959 
2960     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2961         status = nvme_check_dulbe(ns, slba, nlb);
2962         if (status) {
2963             goto invalid;
2964         }
2965     }
2966 
2967     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2968         return nvme_dif_rw(n, req);
2969     }
2970 
2971     status = nvme_map_data(n, nlb, req);
2972     if (status) {
2973         goto invalid;
2974     }
2975 
2976     data_offset = nvme_l2b(ns, slba);
2977 
2978     block_acct_start(blk_get_stats(blk), &req->acct, data_size,
2979                      BLOCK_ACCT_READ);
2980     nvme_blk_read(blk, data_offset, nvme_rw_cb, req);
2981     return NVME_NO_COMPLETE;
2982 
2983 invalid:
2984     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_READ);
2985     return status | NVME_DNR;
2986 }
2987 
2988 static uint16_t nvme_do_write(NvmeCtrl *n, NvmeRequest *req, bool append,
2989                               bool wrz)
2990 {
2991     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2992     NvmeNamespace *ns = req->ns;
2993     uint64_t slba = le64_to_cpu(rw->slba);
2994     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
2995     uint16_t ctrl = le16_to_cpu(rw->control);
2996     uint64_t data_size = nvme_l2b(ns, nlb);
2997     uint64_t mapped_size = data_size;
2998     uint64_t data_offset;
2999     NvmeZone *zone;
3000     NvmeZonedResult *res = (NvmeZonedResult *)&req->cqe;
3001     BlockBackend *blk = ns->blkconf.blk;
3002     uint16_t status;
3003 
3004     if (nvme_ns_ext(ns)) {
3005         mapped_size += nvme_m2b(ns, nlb);
3006 
3007         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3008             bool pract = ctrl & NVME_RW_PRINFO_PRACT;
3009 
3010             if (pract && ns->lbaf.ms == 8) {
3011                 mapped_size -= nvme_m2b(ns, nlb);
3012             }
3013         }
3014     }
3015 
3016     trace_pci_nvme_write(nvme_cid(req), nvme_io_opc_str(rw->opcode),
3017                          nvme_nsid(ns), nlb, mapped_size, slba);
3018 
3019     if (!wrz) {
3020         status = nvme_check_mdts(n, mapped_size);
3021         if (status) {
3022             goto invalid;
3023         }
3024     }
3025 
3026     status = nvme_check_bounds(ns, slba, nlb);
3027     if (status) {
3028         goto invalid;
3029     }
3030 
3031     if (ns->params.zoned) {
3032         zone = nvme_get_zone_by_slba(ns, slba);
3033 
3034         if (append) {
3035             bool piremap = !!(ctrl & NVME_RW_PIREMAP);
3036 
3037             if (unlikely(slba != zone->d.zslba)) {
3038                 trace_pci_nvme_err_append_not_at_start(slba, zone->d.zslba);
3039                 status = NVME_INVALID_FIELD;
3040                 goto invalid;
3041             }
3042 
3043             if (n->params.zasl &&
3044                 data_size > (uint64_t)n->page_size << n->params.zasl) {
3045                 trace_pci_nvme_err_zasl(data_size);
3046                 return NVME_INVALID_FIELD | NVME_DNR;
3047             }
3048 
3049             slba = zone->w_ptr;
3050             rw->slba = cpu_to_le64(slba);
3051             res->slba = cpu_to_le64(slba);
3052 
3053             switch (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3054             case NVME_ID_NS_DPS_TYPE_1:
3055                 if (!piremap) {
3056                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3057                 }
3058 
3059                 /* fallthrough */
3060 
3061             case NVME_ID_NS_DPS_TYPE_2:
3062                 if (piremap) {
3063                     uint32_t reftag = le32_to_cpu(rw->reftag);
3064                     rw->reftag = cpu_to_le32(reftag + (slba - zone->d.zslba));
3065                 }
3066 
3067                 break;
3068 
3069             case NVME_ID_NS_DPS_TYPE_3:
3070                 if (piremap) {
3071                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3072                 }
3073 
3074                 break;
3075             }
3076         }
3077 
3078         status = nvme_check_zone_write(ns, zone, slba, nlb);
3079         if (status) {
3080             goto invalid;
3081         }
3082 
3083         status = nvme_zrm_auto(ns, zone);
3084         if (status) {
3085             goto invalid;
3086         }
3087 
3088         zone->w_ptr += nlb;
3089     }
3090 
3091     data_offset = nvme_l2b(ns, slba);
3092 
3093     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3094         return nvme_dif_rw(n, req);
3095     }
3096 
3097     if (!wrz) {
3098         status = nvme_map_data(n, nlb, req);
3099         if (status) {
3100             goto invalid;
3101         }
3102 
3103         block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3104                          BLOCK_ACCT_WRITE);
3105         nvme_blk_write(blk, data_offset, nvme_rw_cb, req);
3106     } else {
3107         req->aiocb = blk_aio_pwrite_zeroes(blk, data_offset, data_size,
3108                                            BDRV_REQ_MAY_UNMAP, nvme_rw_cb,
3109                                            req);
3110     }
3111 
3112     return NVME_NO_COMPLETE;
3113 
3114 invalid:
3115     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_WRITE);
3116     return status | NVME_DNR;
3117 }
3118 
3119 static inline uint16_t nvme_write(NvmeCtrl *n, NvmeRequest *req)
3120 {
3121     return nvme_do_write(n, req, false, false);
3122 }
3123 
3124 static inline uint16_t nvme_write_zeroes(NvmeCtrl *n, NvmeRequest *req)
3125 {
3126     return nvme_do_write(n, req, false, true);
3127 }
3128 
3129 static inline uint16_t nvme_zone_append(NvmeCtrl *n, NvmeRequest *req)
3130 {
3131     return nvme_do_write(n, req, true, false);
3132 }
3133 
3134 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace *ns, NvmeCmd *c,
3135                                             uint64_t *slba, uint32_t *zone_idx)
3136 {
3137     uint32_t dw10 = le32_to_cpu(c->cdw10);
3138     uint32_t dw11 = le32_to_cpu(c->cdw11);
3139 
3140     if (!ns->params.zoned) {
3141         trace_pci_nvme_err_invalid_opc(c->opcode);
3142         return NVME_INVALID_OPCODE | NVME_DNR;
3143     }
3144 
3145     *slba = ((uint64_t)dw11) << 32 | dw10;
3146     if (unlikely(*slba >= ns->id_ns.nsze)) {
3147         trace_pci_nvme_err_invalid_lba_range(*slba, 0, ns->id_ns.nsze);
3148         *slba = 0;
3149         return NVME_LBA_RANGE | NVME_DNR;
3150     }
3151 
3152     *zone_idx = nvme_zone_idx(ns, *slba);
3153     assert(*zone_idx < ns->num_zones);
3154 
3155     return NVME_SUCCESS;
3156 }
3157 
3158 typedef uint16_t (*op_handler_t)(NvmeNamespace *, NvmeZone *, NvmeZoneState,
3159                                  NvmeRequest *);
3160 
3161 enum NvmeZoneProcessingMask {
3162     NVME_PROC_CURRENT_ZONE    = 0,
3163     NVME_PROC_OPENED_ZONES    = 1 << 0,
3164     NVME_PROC_CLOSED_ZONES    = 1 << 1,
3165     NVME_PROC_READ_ONLY_ZONES = 1 << 2,
3166     NVME_PROC_FULL_ZONES      = 1 << 3,
3167 };
3168 
3169 static uint16_t nvme_open_zone(NvmeNamespace *ns, NvmeZone *zone,
3170                                NvmeZoneState state, NvmeRequest *req)
3171 {
3172     return nvme_zrm_open(ns, zone);
3173 }
3174 
3175 static uint16_t nvme_close_zone(NvmeNamespace *ns, NvmeZone *zone,
3176                                 NvmeZoneState state, NvmeRequest *req)
3177 {
3178     return nvme_zrm_close(ns, zone);
3179 }
3180 
3181 static uint16_t nvme_finish_zone(NvmeNamespace *ns, NvmeZone *zone,
3182                                  NvmeZoneState state, NvmeRequest *req)
3183 {
3184     return nvme_zrm_finish(ns, zone);
3185 }
3186 
3187 static uint16_t nvme_reset_zone(NvmeNamespace *ns, NvmeZone *zone,
3188                                 NvmeZoneState state, NvmeRequest *req)
3189 {
3190     uintptr_t *resets = (uintptr_t *)&req->opaque;
3191     struct nvme_zone_reset_ctx *ctx;
3192 
3193     switch (state) {
3194     case NVME_ZONE_STATE_EMPTY:
3195         return NVME_SUCCESS;
3196     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3197     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3198     case NVME_ZONE_STATE_CLOSED:
3199     case NVME_ZONE_STATE_FULL:
3200         break;
3201     default:
3202         return NVME_ZONE_INVAL_TRANSITION;
3203     }
3204 
3205     /*
3206      * The zone reset aio callback needs to know the zone that is being reset
3207      * in order to transition the zone on completion.
3208      */
3209     ctx = g_new(struct nvme_zone_reset_ctx, 1);
3210     ctx->req = req;
3211     ctx->zone = zone;
3212 
3213     (*resets)++;
3214 
3215     blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_l2b(ns, zone->d.zslba),
3216                           nvme_l2b(ns, ns->zone_size), BDRV_REQ_MAY_UNMAP,
3217                           nvme_aio_zone_reset_cb, ctx);
3218 
3219     return NVME_NO_COMPLETE;
3220 }
3221 
3222 static uint16_t nvme_offline_zone(NvmeNamespace *ns, NvmeZone *zone,
3223                                   NvmeZoneState state, NvmeRequest *req)
3224 {
3225     switch (state) {
3226     case NVME_ZONE_STATE_READ_ONLY:
3227         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_OFFLINE);
3228         /* fall through */
3229     case NVME_ZONE_STATE_OFFLINE:
3230         return NVME_SUCCESS;
3231     default:
3232         return NVME_ZONE_INVAL_TRANSITION;
3233     }
3234 }
3235 
3236 static uint16_t nvme_set_zd_ext(NvmeNamespace *ns, NvmeZone *zone)
3237 {
3238     uint16_t status;
3239     uint8_t state = nvme_get_zone_state(zone);
3240 
3241     if (state == NVME_ZONE_STATE_EMPTY) {
3242         status = nvme_aor_check(ns, 1, 0);
3243         if (status) {
3244             return status;
3245         }
3246         nvme_aor_inc_active(ns);
3247         zone->d.za |= NVME_ZA_ZD_EXT_VALID;
3248         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
3249         return NVME_SUCCESS;
3250     }
3251 
3252     return NVME_ZONE_INVAL_TRANSITION;
3253 }
3254 
3255 static uint16_t nvme_bulk_proc_zone(NvmeNamespace *ns, NvmeZone *zone,
3256                                     enum NvmeZoneProcessingMask proc_mask,
3257                                     op_handler_t op_hndlr, NvmeRequest *req)
3258 {
3259     uint16_t status = NVME_SUCCESS;
3260     NvmeZoneState zs = nvme_get_zone_state(zone);
3261     bool proc_zone;
3262 
3263     switch (zs) {
3264     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3265     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3266         proc_zone = proc_mask & NVME_PROC_OPENED_ZONES;
3267         break;
3268     case NVME_ZONE_STATE_CLOSED:
3269         proc_zone = proc_mask & NVME_PROC_CLOSED_ZONES;
3270         break;
3271     case NVME_ZONE_STATE_READ_ONLY:
3272         proc_zone = proc_mask & NVME_PROC_READ_ONLY_ZONES;
3273         break;
3274     case NVME_ZONE_STATE_FULL:
3275         proc_zone = proc_mask & NVME_PROC_FULL_ZONES;
3276         break;
3277     default:
3278         proc_zone = false;
3279     }
3280 
3281     if (proc_zone) {
3282         status = op_hndlr(ns, zone, zs, req);
3283     }
3284 
3285     return status;
3286 }
3287 
3288 static uint16_t nvme_do_zone_op(NvmeNamespace *ns, NvmeZone *zone,
3289                                 enum NvmeZoneProcessingMask proc_mask,
3290                                 op_handler_t op_hndlr, NvmeRequest *req)
3291 {
3292     NvmeZone *next;
3293     uint16_t status = NVME_SUCCESS;
3294     int i;
3295 
3296     if (!proc_mask) {
3297         status = op_hndlr(ns, zone, nvme_get_zone_state(zone), req);
3298     } else {
3299         if (proc_mask & NVME_PROC_CLOSED_ZONES) {
3300             QTAILQ_FOREACH_SAFE(zone, &ns->closed_zones, entry, next) {
3301                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3302                                              req);
3303                 if (status && status != NVME_NO_COMPLETE) {
3304                     goto out;
3305                 }
3306             }
3307         }
3308         if (proc_mask & NVME_PROC_OPENED_ZONES) {
3309             QTAILQ_FOREACH_SAFE(zone, &ns->imp_open_zones, entry, next) {
3310                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3311                                              req);
3312                 if (status && status != NVME_NO_COMPLETE) {
3313                     goto out;
3314                 }
3315             }
3316 
3317             QTAILQ_FOREACH_SAFE(zone, &ns->exp_open_zones, entry, next) {
3318                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3319                                              req);
3320                 if (status && status != NVME_NO_COMPLETE) {
3321                     goto out;
3322                 }
3323             }
3324         }
3325         if (proc_mask & NVME_PROC_FULL_ZONES) {
3326             QTAILQ_FOREACH_SAFE(zone, &ns->full_zones, entry, next) {
3327                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3328                                              req);
3329                 if (status && status != NVME_NO_COMPLETE) {
3330                     goto out;
3331                 }
3332             }
3333         }
3334 
3335         if (proc_mask & NVME_PROC_READ_ONLY_ZONES) {
3336             for (i = 0; i < ns->num_zones; i++, zone++) {
3337                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3338                                              req);
3339                 if (status && status != NVME_NO_COMPLETE) {
3340                     goto out;
3341                 }
3342             }
3343         }
3344     }
3345 
3346 out:
3347     return status;
3348 }
3349 
3350 static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
3351 {
3352     NvmeCmd *cmd = (NvmeCmd *)&req->cmd;
3353     NvmeNamespace *ns = req->ns;
3354     NvmeZone *zone;
3355     uintptr_t *resets;
3356     uint8_t *zd_ext;
3357     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
3358     uint64_t slba = 0;
3359     uint32_t zone_idx = 0;
3360     uint16_t status;
3361     uint8_t action;
3362     bool all;
3363     enum NvmeZoneProcessingMask proc_mask = NVME_PROC_CURRENT_ZONE;
3364 
3365     action = dw13 & 0xff;
3366     all = dw13 & 0x100;
3367 
3368     req->status = NVME_SUCCESS;
3369 
3370     if (!all) {
3371         status = nvme_get_mgmt_zone_slba_idx(ns, cmd, &slba, &zone_idx);
3372         if (status) {
3373             return status;
3374         }
3375     }
3376 
3377     zone = &ns->zone_array[zone_idx];
3378     if (slba != zone->d.zslba) {
3379         trace_pci_nvme_err_unaligned_zone_cmd(action, slba, zone->d.zslba);
3380         return NVME_INVALID_FIELD | NVME_DNR;
3381     }
3382 
3383     switch (action) {
3384 
3385     case NVME_ZONE_ACTION_OPEN:
3386         if (all) {
3387             proc_mask = NVME_PROC_CLOSED_ZONES;
3388         }
3389         trace_pci_nvme_open_zone(slba, zone_idx, all);
3390         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_open_zone, req);
3391         break;
3392 
3393     case NVME_ZONE_ACTION_CLOSE:
3394         if (all) {
3395             proc_mask = NVME_PROC_OPENED_ZONES;
3396         }
3397         trace_pci_nvme_close_zone(slba, zone_idx, all);
3398         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_close_zone, req);
3399         break;
3400 
3401     case NVME_ZONE_ACTION_FINISH:
3402         if (all) {
3403             proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES;
3404         }
3405         trace_pci_nvme_finish_zone(slba, zone_idx, all);
3406         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_finish_zone, req);
3407         break;
3408 
3409     case NVME_ZONE_ACTION_RESET:
3410         resets = (uintptr_t *)&req->opaque;
3411 
3412         if (all) {
3413             proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES |
3414                 NVME_PROC_FULL_ZONES;
3415         }
3416         trace_pci_nvme_reset_zone(slba, zone_idx, all);
3417 
3418         *resets = 1;
3419 
3420         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_reset_zone, req);
3421 
3422         (*resets)--;
3423 
3424         return *resets ? NVME_NO_COMPLETE : req->status;
3425 
3426     case NVME_ZONE_ACTION_OFFLINE:
3427         if (all) {
3428             proc_mask = NVME_PROC_READ_ONLY_ZONES;
3429         }
3430         trace_pci_nvme_offline_zone(slba, zone_idx, all);
3431         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_offline_zone, req);
3432         break;
3433 
3434     case NVME_ZONE_ACTION_SET_ZD_EXT:
3435         trace_pci_nvme_set_descriptor_extension(slba, zone_idx);
3436         if (all || !ns->params.zd_extension_size) {
3437             return NVME_INVALID_FIELD | NVME_DNR;
3438         }
3439         zd_ext = nvme_get_zd_extension(ns, zone_idx);
3440         status = nvme_h2c(n, zd_ext, ns->params.zd_extension_size, req);
3441         if (status) {
3442             trace_pci_nvme_err_zd_extension_map_error(zone_idx);
3443             return status;
3444         }
3445 
3446         status = nvme_set_zd_ext(ns, zone);
3447         if (status == NVME_SUCCESS) {
3448             trace_pci_nvme_zd_extension_set(zone_idx);
3449             return status;
3450         }
3451         break;
3452 
3453     default:
3454         trace_pci_nvme_err_invalid_mgmt_action(action);
3455         status = NVME_INVALID_FIELD;
3456     }
3457 
3458     if (status == NVME_ZONE_INVAL_TRANSITION) {
3459         trace_pci_nvme_err_invalid_zone_state_transition(action, slba,
3460                                                          zone->d.za);
3461     }
3462     if (status) {
3463         status |= NVME_DNR;
3464     }
3465 
3466     return status;
3467 }
3468 
3469 static bool nvme_zone_matches_filter(uint32_t zafs, NvmeZone *zl)
3470 {
3471     NvmeZoneState zs = nvme_get_zone_state(zl);
3472 
3473     switch (zafs) {
3474     case NVME_ZONE_REPORT_ALL:
3475         return true;
3476     case NVME_ZONE_REPORT_EMPTY:
3477         return zs == NVME_ZONE_STATE_EMPTY;
3478     case NVME_ZONE_REPORT_IMPLICITLY_OPEN:
3479         return zs == NVME_ZONE_STATE_IMPLICITLY_OPEN;
3480     case NVME_ZONE_REPORT_EXPLICITLY_OPEN:
3481         return zs == NVME_ZONE_STATE_EXPLICITLY_OPEN;
3482     case NVME_ZONE_REPORT_CLOSED:
3483         return zs == NVME_ZONE_STATE_CLOSED;
3484     case NVME_ZONE_REPORT_FULL:
3485         return zs == NVME_ZONE_STATE_FULL;
3486     case NVME_ZONE_REPORT_READ_ONLY:
3487         return zs == NVME_ZONE_STATE_READ_ONLY;
3488     case NVME_ZONE_REPORT_OFFLINE:
3489         return zs == NVME_ZONE_STATE_OFFLINE;
3490     default:
3491         return false;
3492     }
3493 }
3494 
3495 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
3496 {
3497     NvmeCmd *cmd = (NvmeCmd *)&req->cmd;
3498     NvmeNamespace *ns = req->ns;
3499     /* cdw12 is zero-based number of dwords to return. Convert to bytes */
3500     uint32_t data_size = (le32_to_cpu(cmd->cdw12) + 1) << 2;
3501     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
3502     uint32_t zone_idx, zra, zrasf, partial;
3503     uint64_t max_zones, nr_zones = 0;
3504     uint16_t status;
3505     uint64_t slba;
3506     NvmeZoneDescr *z;
3507     NvmeZone *zone;
3508     NvmeZoneReportHeader *header;
3509     void *buf, *buf_p;
3510     size_t zone_entry_sz;
3511     int i;
3512 
3513     req->status = NVME_SUCCESS;
3514 
3515     status = nvme_get_mgmt_zone_slba_idx(ns, cmd, &slba, &zone_idx);
3516     if (status) {
3517         return status;
3518     }
3519 
3520     zra = dw13 & 0xff;
3521     if (zra != NVME_ZONE_REPORT && zra != NVME_ZONE_REPORT_EXTENDED) {
3522         return NVME_INVALID_FIELD | NVME_DNR;
3523     }
3524     if (zra == NVME_ZONE_REPORT_EXTENDED && !ns->params.zd_extension_size) {
3525         return NVME_INVALID_FIELD | NVME_DNR;
3526     }
3527 
3528     zrasf = (dw13 >> 8) & 0xff;
3529     if (zrasf > NVME_ZONE_REPORT_OFFLINE) {
3530         return NVME_INVALID_FIELD | NVME_DNR;
3531     }
3532 
3533     if (data_size < sizeof(NvmeZoneReportHeader)) {
3534         return NVME_INVALID_FIELD | NVME_DNR;
3535     }
3536 
3537     status = nvme_check_mdts(n, data_size);
3538     if (status) {
3539         return status;
3540     }
3541 
3542     partial = (dw13 >> 16) & 0x01;
3543 
3544     zone_entry_sz = sizeof(NvmeZoneDescr);
3545     if (zra == NVME_ZONE_REPORT_EXTENDED) {
3546         zone_entry_sz += ns->params.zd_extension_size;
3547     }
3548 
3549     max_zones = (data_size - sizeof(NvmeZoneReportHeader)) / zone_entry_sz;
3550     buf = g_malloc0(data_size);
3551 
3552     zone = &ns->zone_array[zone_idx];
3553     for (i = zone_idx; i < ns->num_zones; i++) {
3554         if (partial && nr_zones >= max_zones) {
3555             break;
3556         }
3557         if (nvme_zone_matches_filter(zrasf, zone++)) {
3558             nr_zones++;
3559         }
3560     }
3561     header = (NvmeZoneReportHeader *)buf;
3562     header->nr_zones = cpu_to_le64(nr_zones);
3563 
3564     buf_p = buf + sizeof(NvmeZoneReportHeader);
3565     for (; zone_idx < ns->num_zones && max_zones > 0; zone_idx++) {
3566         zone = &ns->zone_array[zone_idx];
3567         if (nvme_zone_matches_filter(zrasf, zone)) {
3568             z = (NvmeZoneDescr *)buf_p;
3569             buf_p += sizeof(NvmeZoneDescr);
3570 
3571             z->zt = zone->d.zt;
3572             z->zs = zone->d.zs;
3573             z->zcap = cpu_to_le64(zone->d.zcap);
3574             z->zslba = cpu_to_le64(zone->d.zslba);
3575             z->za = zone->d.za;
3576 
3577             if (nvme_wp_is_valid(zone)) {
3578                 z->wp = cpu_to_le64(zone->d.wp);
3579             } else {
3580                 z->wp = cpu_to_le64(~0ULL);
3581             }
3582 
3583             if (zra == NVME_ZONE_REPORT_EXTENDED) {
3584                 if (zone->d.za & NVME_ZA_ZD_EXT_VALID) {
3585                     memcpy(buf_p, nvme_get_zd_extension(ns, zone_idx),
3586                            ns->params.zd_extension_size);
3587                 }
3588                 buf_p += ns->params.zd_extension_size;
3589             }
3590 
3591             max_zones--;
3592         }
3593     }
3594 
3595     status = nvme_c2h(n, (uint8_t *)buf, data_size, req);
3596 
3597     g_free(buf);
3598 
3599     return status;
3600 }
3601 
3602 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
3603 {
3604     NvmeNamespace *ns;
3605     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
3606 
3607     trace_pci_nvme_io_cmd(nvme_cid(req), nsid, nvme_sqid(req),
3608                           req->cmd.opcode, nvme_io_opc_str(req->cmd.opcode));
3609 
3610     if (!nvme_nsid_valid(n, nsid)) {
3611         return NVME_INVALID_NSID | NVME_DNR;
3612     }
3613 
3614     /*
3615      * In the base NVM command set, Flush may apply to all namespaces
3616      * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
3617      * along with TP 4056 (Namespace Types), it may be pretty screwed up.
3618      *
3619      * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
3620      * opcode with a specific command since we cannot determine a unique I/O
3621      * command set. Opcode 0h could have any other meaning than something
3622      * equivalent to flushing and say it DOES have completely different
3623      * semantics in some other command set - does an NSID of FFFFFFFFh then
3624      * mean "for all namespaces, apply whatever command set specific command
3625      * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
3626      * whatever command that uses the 0h opcode if, and only if, it allows NSID
3627      * to be FFFFFFFFh"?
3628      *
3629      * Anyway (and luckily), for now, we do not care about this since the
3630      * device only supports namespace types that includes the NVM Flush command
3631      * (NVM and Zoned), so always do an NVM Flush.
3632      */
3633     if (req->cmd.opcode == NVME_CMD_FLUSH) {
3634         return nvme_flush(n, req);
3635     }
3636 
3637     ns = nvme_ns(n, nsid);
3638     if (unlikely(!ns)) {
3639         return NVME_INVALID_FIELD | NVME_DNR;
3640     }
3641 
3642     if (!(ns->iocs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
3643         trace_pci_nvme_err_invalid_opc(req->cmd.opcode);
3644         return NVME_INVALID_OPCODE | NVME_DNR;
3645     }
3646 
3647     if (ns->status) {
3648         return ns->status;
3649     }
3650 
3651     req->ns = ns;
3652 
3653     switch (req->cmd.opcode) {
3654     case NVME_CMD_WRITE_ZEROES:
3655         return nvme_write_zeroes(n, req);
3656     case NVME_CMD_ZONE_APPEND:
3657         return nvme_zone_append(n, req);
3658     case NVME_CMD_WRITE:
3659         return nvme_write(n, req);
3660     case NVME_CMD_READ:
3661         return nvme_read(n, req);
3662     case NVME_CMD_COMPARE:
3663         return nvme_compare(n, req);
3664     case NVME_CMD_DSM:
3665         return nvme_dsm(n, req);
3666     case NVME_CMD_VERIFY:
3667         return nvme_verify(n, req);
3668     case NVME_CMD_COPY:
3669         return nvme_copy(n, req);
3670     case NVME_CMD_ZONE_MGMT_SEND:
3671         return nvme_zone_mgmt_send(n, req);
3672     case NVME_CMD_ZONE_MGMT_RECV:
3673         return nvme_zone_mgmt_recv(n, req);
3674     default:
3675         assert(false);
3676     }
3677 
3678     return NVME_INVALID_OPCODE | NVME_DNR;
3679 }
3680 
3681 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
3682 {
3683     n->sq[sq->sqid] = NULL;
3684     timer_free(sq->timer);
3685     g_free(sq->io_req);
3686     if (sq->sqid) {
3687         g_free(sq);
3688     }
3689 }
3690 
3691 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
3692 {
3693     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
3694     NvmeRequest *r, *next;
3695     NvmeSQueue *sq;
3696     NvmeCQueue *cq;
3697     uint16_t qid = le16_to_cpu(c->qid);
3698     uint32_t nsid;
3699 
3700     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
3701         trace_pci_nvme_err_invalid_del_sq(qid);
3702         return NVME_INVALID_QID | NVME_DNR;
3703     }
3704 
3705     trace_pci_nvme_del_sq(qid);
3706 
3707     sq = n->sq[qid];
3708     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
3709         r = QTAILQ_FIRST(&sq->out_req_list);
3710         if (r->aiocb) {
3711             blk_aio_cancel(r->aiocb);
3712         }
3713     }
3714 
3715     /*
3716      * Drain all namespaces if there are still outstanding requests that we
3717      * could not cancel explicitly.
3718      */
3719     if (!QTAILQ_EMPTY(&sq->out_req_list)) {
3720         for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
3721             NvmeNamespace *ns = nvme_ns(n, nsid);
3722             if (ns) {
3723                 nvme_ns_drain(ns);
3724             }
3725         }
3726     }
3727 
3728     assert(QTAILQ_EMPTY(&sq->out_req_list));
3729 
3730     if (!nvme_check_cqid(n, sq->cqid)) {
3731         cq = n->cq[sq->cqid];
3732         QTAILQ_REMOVE(&cq->sq_list, sq, entry);
3733 
3734         nvme_post_cqes(cq);
3735         QTAILQ_FOREACH_SAFE(r, &cq->req_list, entry, next) {
3736             if (r->sq == sq) {
3737                 QTAILQ_REMOVE(&cq->req_list, r, entry);
3738                 QTAILQ_INSERT_TAIL(&sq->req_list, r, entry);
3739             }
3740         }
3741     }
3742 
3743     nvme_free_sq(sq, n);
3744     return NVME_SUCCESS;
3745 }
3746 
3747 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
3748                          uint16_t sqid, uint16_t cqid, uint16_t size)
3749 {
3750     int i;
3751     NvmeCQueue *cq;
3752 
3753     sq->ctrl = n;
3754     sq->dma_addr = dma_addr;
3755     sq->sqid = sqid;
3756     sq->size = size;
3757     sq->cqid = cqid;
3758     sq->head = sq->tail = 0;
3759     sq->io_req = g_new0(NvmeRequest, sq->size);
3760 
3761     QTAILQ_INIT(&sq->req_list);
3762     QTAILQ_INIT(&sq->out_req_list);
3763     for (i = 0; i < sq->size; i++) {
3764         sq->io_req[i].sq = sq;
3765         QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
3766     }
3767     sq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_process_sq, sq);
3768 
3769     assert(n->cq[cqid]);
3770     cq = n->cq[cqid];
3771     QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
3772     n->sq[sqid] = sq;
3773 }
3774 
3775 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeRequest *req)
3776 {
3777     NvmeSQueue *sq;
3778     NvmeCreateSq *c = (NvmeCreateSq *)&req->cmd;
3779 
3780     uint16_t cqid = le16_to_cpu(c->cqid);
3781     uint16_t sqid = le16_to_cpu(c->sqid);
3782     uint16_t qsize = le16_to_cpu(c->qsize);
3783     uint16_t qflags = le16_to_cpu(c->sq_flags);
3784     uint64_t prp1 = le64_to_cpu(c->prp1);
3785 
3786     trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
3787 
3788     if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
3789         trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
3790         return NVME_INVALID_CQID | NVME_DNR;
3791     }
3792     if (unlikely(!sqid || sqid > n->params.max_ioqpairs ||
3793         n->sq[sqid] != NULL)) {
3794         trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
3795         return NVME_INVALID_QID | NVME_DNR;
3796     }
3797     if (unlikely(!qsize || qsize > NVME_CAP_MQES(n->bar.cap))) {
3798         trace_pci_nvme_err_invalid_create_sq_size(qsize);
3799         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
3800     }
3801     if (unlikely(prp1 & (n->page_size - 1))) {
3802         trace_pci_nvme_err_invalid_create_sq_addr(prp1);
3803         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
3804     }
3805     if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
3806         trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
3807         return NVME_INVALID_FIELD | NVME_DNR;
3808     }
3809     sq = g_malloc0(sizeof(*sq));
3810     nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
3811     return NVME_SUCCESS;
3812 }
3813 
3814 struct nvme_stats {
3815     uint64_t units_read;
3816     uint64_t units_written;
3817     uint64_t read_commands;
3818     uint64_t write_commands;
3819 };
3820 
3821 static void nvme_set_blk_stats(NvmeNamespace *ns, struct nvme_stats *stats)
3822 {
3823     BlockAcctStats *s = blk_get_stats(ns->blkconf.blk);
3824 
3825     stats->units_read += s->nr_bytes[BLOCK_ACCT_READ] >> BDRV_SECTOR_BITS;
3826     stats->units_written += s->nr_bytes[BLOCK_ACCT_WRITE] >> BDRV_SECTOR_BITS;
3827     stats->read_commands += s->nr_ops[BLOCK_ACCT_READ];
3828     stats->write_commands += s->nr_ops[BLOCK_ACCT_WRITE];
3829 }
3830 
3831 static uint16_t nvme_smart_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
3832                                 uint64_t off, NvmeRequest *req)
3833 {
3834     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
3835     struct nvme_stats stats = { 0 };
3836     NvmeSmartLog smart = { 0 };
3837     uint32_t trans_len;
3838     NvmeNamespace *ns;
3839     time_t current_ms;
3840 
3841     if (off >= sizeof(smart)) {
3842         return NVME_INVALID_FIELD | NVME_DNR;
3843     }
3844 
3845     if (nsid != 0xffffffff) {
3846         ns = nvme_ns(n, nsid);
3847         if (!ns) {
3848             return NVME_INVALID_NSID | NVME_DNR;
3849         }
3850         nvme_set_blk_stats(ns, &stats);
3851     } else {
3852         int i;
3853 
3854         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
3855             ns = nvme_ns(n, i);
3856             if (!ns) {
3857                 continue;
3858             }
3859             nvme_set_blk_stats(ns, &stats);
3860         }
3861     }
3862 
3863     trans_len = MIN(sizeof(smart) - off, buf_len);
3864     smart.critical_warning = n->smart_critical_warning;
3865 
3866     smart.data_units_read[0] = cpu_to_le64(DIV_ROUND_UP(stats.units_read,
3867                                                         1000));
3868     smart.data_units_written[0] = cpu_to_le64(DIV_ROUND_UP(stats.units_written,
3869                                                            1000));
3870     smart.host_read_commands[0] = cpu_to_le64(stats.read_commands);
3871     smart.host_write_commands[0] = cpu_to_le64(stats.write_commands);
3872 
3873     smart.temperature = cpu_to_le16(n->temperature);
3874 
3875     if ((n->temperature >= n->features.temp_thresh_hi) ||
3876         (n->temperature <= n->features.temp_thresh_low)) {
3877         smart.critical_warning |= NVME_SMART_TEMPERATURE;
3878     }
3879 
3880     current_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
3881     smart.power_on_hours[0] =
3882         cpu_to_le64((((current_ms - n->starttime_ms) / 1000) / 60) / 60);
3883 
3884     if (!rae) {
3885         nvme_clear_events(n, NVME_AER_TYPE_SMART);
3886     }
3887 
3888     return nvme_c2h(n, (uint8_t *) &smart + off, trans_len, req);
3889 }
3890 
3891 static uint16_t nvme_fw_log_info(NvmeCtrl *n, uint32_t buf_len, uint64_t off,
3892                                  NvmeRequest *req)
3893 {
3894     uint32_t trans_len;
3895     NvmeFwSlotInfoLog fw_log = {
3896         .afi = 0x1,
3897     };
3898 
3899     if (off >= sizeof(fw_log)) {
3900         return NVME_INVALID_FIELD | NVME_DNR;
3901     }
3902 
3903     strpadcpy((char *)&fw_log.frs1, sizeof(fw_log.frs1), "1.0", ' ');
3904     trans_len = MIN(sizeof(fw_log) - off, buf_len);
3905 
3906     return nvme_c2h(n, (uint8_t *) &fw_log + off, trans_len, req);
3907 }
3908 
3909 static uint16_t nvme_error_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
3910                                 uint64_t off, NvmeRequest *req)
3911 {
3912     uint32_t trans_len;
3913     NvmeErrorLog errlog;
3914 
3915     if (off >= sizeof(errlog)) {
3916         return NVME_INVALID_FIELD | NVME_DNR;
3917     }
3918 
3919     if (!rae) {
3920         nvme_clear_events(n, NVME_AER_TYPE_ERROR);
3921     }
3922 
3923     memset(&errlog, 0x0, sizeof(errlog));
3924     trans_len = MIN(sizeof(errlog) - off, buf_len);
3925 
3926     return nvme_c2h(n, (uint8_t *)&errlog, trans_len, req);
3927 }
3928 
3929 static uint16_t nvme_changed_nslist(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
3930                                     uint64_t off, NvmeRequest *req)
3931 {
3932     uint32_t nslist[1024];
3933     uint32_t trans_len;
3934     int i = 0;
3935     uint32_t nsid;
3936 
3937     memset(nslist, 0x0, sizeof(nslist));
3938     trans_len = MIN(sizeof(nslist) - off, buf_len);
3939 
3940     while ((nsid = find_first_bit(n->changed_nsids, NVME_CHANGED_NSID_SIZE)) !=
3941             NVME_CHANGED_NSID_SIZE) {
3942         /*
3943          * If more than 1024 namespaces, the first entry in the log page should
3944          * be set to FFFFFFFFh and the others to 0 as spec.
3945          */
3946         if (i == ARRAY_SIZE(nslist)) {
3947             memset(nslist, 0x0, sizeof(nslist));
3948             nslist[0] = 0xffffffff;
3949             break;
3950         }
3951 
3952         nslist[i++] = nsid;
3953         clear_bit(nsid, n->changed_nsids);
3954     }
3955 
3956     /*
3957      * Remove all the remaining list entries in case returns directly due to
3958      * more than 1024 namespaces.
3959      */
3960     if (nslist[0] == 0xffffffff) {
3961         bitmap_zero(n->changed_nsids, NVME_CHANGED_NSID_SIZE);
3962     }
3963 
3964     if (!rae) {
3965         nvme_clear_events(n, NVME_AER_TYPE_NOTICE);
3966     }
3967 
3968     return nvme_c2h(n, ((uint8_t *)nslist) + off, trans_len, req);
3969 }
3970 
3971 static uint16_t nvme_cmd_effects(NvmeCtrl *n, uint8_t csi, uint32_t buf_len,
3972                                  uint64_t off, NvmeRequest *req)
3973 {
3974     NvmeEffectsLog log = {};
3975     const uint32_t *src_iocs = NULL;
3976     uint32_t trans_len;
3977 
3978     if (off >= sizeof(log)) {
3979         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(log));
3980         return NVME_INVALID_FIELD | NVME_DNR;
3981     }
3982 
3983     switch (NVME_CC_CSS(n->bar.cc)) {
3984     case NVME_CC_CSS_NVM:
3985         src_iocs = nvme_cse_iocs_nvm;
3986         /* fall through */
3987     case NVME_CC_CSS_ADMIN_ONLY:
3988         break;
3989     case NVME_CC_CSS_CSI:
3990         switch (csi) {
3991         case NVME_CSI_NVM:
3992             src_iocs = nvme_cse_iocs_nvm;
3993             break;
3994         case NVME_CSI_ZONED:
3995             src_iocs = nvme_cse_iocs_zoned;
3996             break;
3997         }
3998     }
3999 
4000     memcpy(log.acs, nvme_cse_acs, sizeof(nvme_cse_acs));
4001 
4002     if (src_iocs) {
4003         memcpy(log.iocs, src_iocs, sizeof(log.iocs));
4004     }
4005 
4006     trans_len = MIN(sizeof(log) - off, buf_len);
4007 
4008     return nvme_c2h(n, ((uint8_t *)&log) + off, trans_len, req);
4009 }
4010 
4011 static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
4012 {
4013     NvmeCmd *cmd = &req->cmd;
4014 
4015     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
4016     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
4017     uint32_t dw12 = le32_to_cpu(cmd->cdw12);
4018     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
4019     uint8_t  lid = dw10 & 0xff;
4020     uint8_t  lsp = (dw10 >> 8) & 0xf;
4021     uint8_t  rae = (dw10 >> 15) & 0x1;
4022     uint8_t  csi = le32_to_cpu(cmd->cdw14) >> 24;
4023     uint32_t numdl, numdu;
4024     uint64_t off, lpol, lpou;
4025     size_t   len;
4026     uint16_t status;
4027 
4028     numdl = (dw10 >> 16);
4029     numdu = (dw11 & 0xffff);
4030     lpol = dw12;
4031     lpou = dw13;
4032 
4033     len = (((numdu << 16) | numdl) + 1) << 2;
4034     off = (lpou << 32ULL) | lpol;
4035 
4036     if (off & 0x3) {
4037         return NVME_INVALID_FIELD | NVME_DNR;
4038     }
4039 
4040     trace_pci_nvme_get_log(nvme_cid(req), lid, lsp, rae, len, off);
4041 
4042     status = nvme_check_mdts(n, len);
4043     if (status) {
4044         return status;
4045     }
4046 
4047     switch (lid) {
4048     case NVME_LOG_ERROR_INFO:
4049         return nvme_error_info(n, rae, len, off, req);
4050     case NVME_LOG_SMART_INFO:
4051         return nvme_smart_info(n, rae, len, off, req);
4052     case NVME_LOG_FW_SLOT_INFO:
4053         return nvme_fw_log_info(n, len, off, req);
4054     case NVME_LOG_CHANGED_NSLIST:
4055         return nvme_changed_nslist(n, rae, len, off, req);
4056     case NVME_LOG_CMD_EFFECTS:
4057         return nvme_cmd_effects(n, csi, len, off, req);
4058     default:
4059         trace_pci_nvme_err_invalid_log_page(nvme_cid(req), lid);
4060         return NVME_INVALID_FIELD | NVME_DNR;
4061     }
4062 }
4063 
4064 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
4065 {
4066     n->cq[cq->cqid] = NULL;
4067     timer_free(cq->timer);
4068     if (msix_enabled(&n->parent_obj)) {
4069         msix_vector_unuse(&n->parent_obj, cq->vector);
4070     }
4071     if (cq->cqid) {
4072         g_free(cq);
4073     }
4074 }
4075 
4076 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)
4077 {
4078     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4079     NvmeCQueue *cq;
4080     uint16_t qid = le16_to_cpu(c->qid);
4081 
4082     if (unlikely(!qid || nvme_check_cqid(n, qid))) {
4083         trace_pci_nvme_err_invalid_del_cq_cqid(qid);
4084         return NVME_INVALID_CQID | NVME_DNR;
4085     }
4086 
4087     cq = n->cq[qid];
4088     if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
4089         trace_pci_nvme_err_invalid_del_cq_notempty(qid);
4090         return NVME_INVALID_QUEUE_DEL;
4091     }
4092     nvme_irq_deassert(n, cq);
4093     trace_pci_nvme_del_cq(qid);
4094     nvme_free_cq(cq, n);
4095     return NVME_SUCCESS;
4096 }
4097 
4098 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
4099                          uint16_t cqid, uint16_t vector, uint16_t size,
4100                          uint16_t irq_enabled)
4101 {
4102     int ret;
4103 
4104     if (msix_enabled(&n->parent_obj)) {
4105         ret = msix_vector_use(&n->parent_obj, vector);
4106         assert(ret == 0);
4107     }
4108     cq->ctrl = n;
4109     cq->cqid = cqid;
4110     cq->size = size;
4111     cq->dma_addr = dma_addr;
4112     cq->phase = 1;
4113     cq->irq_enabled = irq_enabled;
4114     cq->vector = vector;
4115     cq->head = cq->tail = 0;
4116     QTAILQ_INIT(&cq->req_list);
4117     QTAILQ_INIT(&cq->sq_list);
4118     n->cq[cqid] = cq;
4119     cq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_post_cqes, cq);
4120 }
4121 
4122 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeRequest *req)
4123 {
4124     NvmeCQueue *cq;
4125     NvmeCreateCq *c = (NvmeCreateCq *)&req->cmd;
4126     uint16_t cqid = le16_to_cpu(c->cqid);
4127     uint16_t vector = le16_to_cpu(c->irq_vector);
4128     uint16_t qsize = le16_to_cpu(c->qsize);
4129     uint16_t qflags = le16_to_cpu(c->cq_flags);
4130     uint64_t prp1 = le64_to_cpu(c->prp1);
4131 
4132     trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
4133                              NVME_CQ_FLAGS_IEN(qflags) != 0);
4134 
4135     if (unlikely(!cqid || cqid > n->params.max_ioqpairs ||
4136         n->cq[cqid] != NULL)) {
4137         trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
4138         return NVME_INVALID_QID | NVME_DNR;
4139     }
4140     if (unlikely(!qsize || qsize > NVME_CAP_MQES(n->bar.cap))) {
4141         trace_pci_nvme_err_invalid_create_cq_size(qsize);
4142         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4143     }
4144     if (unlikely(prp1 & (n->page_size - 1))) {
4145         trace_pci_nvme_err_invalid_create_cq_addr(prp1);
4146         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4147     }
4148     if (unlikely(!msix_enabled(&n->parent_obj) && vector)) {
4149         trace_pci_nvme_err_invalid_create_cq_vector(vector);
4150         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
4151     }
4152     if (unlikely(vector >= n->params.msix_qsize)) {
4153         trace_pci_nvme_err_invalid_create_cq_vector(vector);
4154         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
4155     }
4156     if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
4157         trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
4158         return NVME_INVALID_FIELD | NVME_DNR;
4159     }
4160 
4161     cq = g_malloc0(sizeof(*cq));
4162     nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
4163                  NVME_CQ_FLAGS_IEN(qflags));
4164 
4165     /*
4166      * It is only required to set qs_created when creating a completion queue;
4167      * creating a submission queue without a matching completion queue will
4168      * fail.
4169      */
4170     n->qs_created = true;
4171     return NVME_SUCCESS;
4172 }
4173 
4174 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl *n, NvmeRequest *req)
4175 {
4176     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
4177 
4178     return nvme_c2h(n, id, sizeof(id), req);
4179 }
4180 
4181 static inline bool nvme_csi_has_nvm_support(NvmeNamespace *ns)
4182 {
4183     switch (ns->csi) {
4184     case NVME_CSI_NVM:
4185     case NVME_CSI_ZONED:
4186         return true;
4187     }
4188     return false;
4189 }
4190 
4191 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeRequest *req)
4192 {
4193     trace_pci_nvme_identify_ctrl();
4194 
4195     return nvme_c2h(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl), req);
4196 }
4197 
4198 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl *n, NvmeRequest *req)
4199 {
4200     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4201     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
4202     NvmeIdCtrlNvm *id_nvm = (NvmeIdCtrlNvm *)&id;
4203 
4204     trace_pci_nvme_identify_ctrl_csi(c->csi);
4205 
4206     switch (c->csi) {
4207     case NVME_CSI_NVM:
4208         id_nvm->vsl = n->params.vsl;
4209         id_nvm->dmrsl = cpu_to_le32(n->dmrsl);
4210         break;
4211 
4212     case NVME_CSI_ZONED:
4213         ((NvmeIdCtrlZoned *)&id)->zasl = n->params.zasl;
4214         break;
4215 
4216     default:
4217         return NVME_INVALID_FIELD | NVME_DNR;
4218     }
4219 
4220     return nvme_c2h(n, id, sizeof(id), req);
4221 }
4222 
4223 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeRequest *req, bool active)
4224 {
4225     NvmeNamespace *ns;
4226     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4227     uint32_t nsid = le32_to_cpu(c->nsid);
4228 
4229     trace_pci_nvme_identify_ns(nsid);
4230 
4231     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4232         return NVME_INVALID_NSID | NVME_DNR;
4233     }
4234 
4235     ns = nvme_ns(n, nsid);
4236     if (unlikely(!ns)) {
4237         if (!active) {
4238             ns = nvme_subsys_ns(n->subsys, nsid);
4239             if (!ns) {
4240                 return nvme_rpt_empty_id_struct(n, req);
4241             }
4242         } else {
4243             return nvme_rpt_empty_id_struct(n, req);
4244         }
4245     }
4246 
4247     if (c->csi == NVME_CSI_NVM && nvme_csi_has_nvm_support(ns)) {
4248         return nvme_c2h(n, (uint8_t *)&ns->id_ns, sizeof(NvmeIdNs), req);
4249     }
4250 
4251     return NVME_INVALID_CMD_SET | NVME_DNR;
4252 }
4253 
4254 static uint16_t nvme_identify_ns_attached_list(NvmeCtrl *n, NvmeRequest *req)
4255 {
4256     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4257     uint16_t min_id = le16_to_cpu(c->ctrlid);
4258     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
4259     uint16_t *ids = &list[1];
4260     NvmeNamespace *ns;
4261     NvmeCtrl *ctrl;
4262     int cntlid, nr_ids = 0;
4263 
4264     trace_pci_nvme_identify_ns_attached_list(min_id);
4265 
4266     if (c->nsid == NVME_NSID_BROADCAST) {
4267         return NVME_INVALID_FIELD | NVME_DNR;
4268     }
4269 
4270     ns = nvme_subsys_ns(n->subsys, c->nsid);
4271     if (!ns) {
4272         return NVME_INVALID_FIELD | NVME_DNR;
4273     }
4274 
4275     for (cntlid = min_id; cntlid < ARRAY_SIZE(n->subsys->ctrls); cntlid++) {
4276         ctrl = nvme_subsys_ctrl(n->subsys, cntlid);
4277         if (!ctrl) {
4278             continue;
4279         }
4280 
4281         if (!nvme_ns(ctrl, c->nsid)) {
4282             continue;
4283         }
4284 
4285         ids[nr_ids++] = cntlid;
4286     }
4287 
4288     list[0] = nr_ids;
4289 
4290     return nvme_c2h(n, (uint8_t *)list, sizeof(list), req);
4291 }
4292 
4293 static uint16_t nvme_identify_ns_csi(NvmeCtrl *n, NvmeRequest *req,
4294         bool active)
4295 {
4296     NvmeNamespace *ns;
4297     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4298     uint32_t nsid = le32_to_cpu(c->nsid);
4299 
4300     trace_pci_nvme_identify_ns_csi(nsid, c->csi);
4301 
4302     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4303         return NVME_INVALID_NSID | NVME_DNR;
4304     }
4305 
4306     ns = nvme_ns(n, nsid);
4307     if (unlikely(!ns)) {
4308         if (!active) {
4309             ns = nvme_subsys_ns(n->subsys, nsid);
4310             if (!ns) {
4311                 return nvme_rpt_empty_id_struct(n, req);
4312             }
4313         } else {
4314             return nvme_rpt_empty_id_struct(n, req);
4315         }
4316     }
4317 
4318     if (c->csi == NVME_CSI_NVM && nvme_csi_has_nvm_support(ns)) {
4319         return nvme_rpt_empty_id_struct(n, req);
4320     } else if (c->csi == NVME_CSI_ZONED && ns->csi == NVME_CSI_ZONED) {
4321         return nvme_c2h(n, (uint8_t *)ns->id_ns_zoned, sizeof(NvmeIdNsZoned),
4322                         req);
4323     }
4324 
4325     return NVME_INVALID_FIELD | NVME_DNR;
4326 }
4327 
4328 static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeRequest *req,
4329         bool active)
4330 {
4331     NvmeNamespace *ns;
4332     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4333     uint32_t min_nsid = le32_to_cpu(c->nsid);
4334     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
4335     static const int data_len = sizeof(list);
4336     uint32_t *list_ptr = (uint32_t *)list;
4337     int i, j = 0;
4338 
4339     trace_pci_nvme_identify_nslist(min_nsid);
4340 
4341     /*
4342      * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
4343      * since the Active Namespace ID List should return namespaces with ids
4344      * *higher* than the NSID specified in the command. This is also specified
4345      * in the spec (NVM Express v1.3d, Section 5.15.4).
4346      */
4347     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
4348         return NVME_INVALID_NSID | NVME_DNR;
4349     }
4350 
4351     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4352         ns = nvme_ns(n, i);
4353         if (!ns) {
4354             if (!active) {
4355                 ns = nvme_subsys_ns(n->subsys, i);
4356                 if (!ns) {
4357                     continue;
4358                 }
4359             } else {
4360                 continue;
4361             }
4362         }
4363         if (ns->params.nsid <= min_nsid) {
4364             continue;
4365         }
4366         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
4367         if (j == data_len / sizeof(uint32_t)) {
4368             break;
4369         }
4370     }
4371 
4372     return nvme_c2h(n, list, data_len, req);
4373 }
4374 
4375 static uint16_t nvme_identify_nslist_csi(NvmeCtrl *n, NvmeRequest *req,
4376         bool active)
4377 {
4378     NvmeNamespace *ns;
4379     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4380     uint32_t min_nsid = le32_to_cpu(c->nsid);
4381     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
4382     static const int data_len = sizeof(list);
4383     uint32_t *list_ptr = (uint32_t *)list;
4384     int i, j = 0;
4385 
4386     trace_pci_nvme_identify_nslist_csi(min_nsid, c->csi);
4387 
4388     /*
4389      * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
4390      */
4391     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
4392         return NVME_INVALID_NSID | NVME_DNR;
4393     }
4394 
4395     if (c->csi != NVME_CSI_NVM && c->csi != NVME_CSI_ZONED) {
4396         return NVME_INVALID_FIELD | NVME_DNR;
4397     }
4398 
4399     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4400         ns = nvme_ns(n, i);
4401         if (!ns) {
4402             if (!active) {
4403                 ns = nvme_subsys_ns(n->subsys, i);
4404                 if (!ns) {
4405                     continue;
4406                 }
4407             } else {
4408                 continue;
4409             }
4410         }
4411         if (ns->params.nsid <= min_nsid || c->csi != ns->csi) {
4412             continue;
4413         }
4414         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
4415         if (j == data_len / sizeof(uint32_t)) {
4416             break;
4417         }
4418     }
4419 
4420     return nvme_c2h(n, list, data_len, req);
4421 }
4422 
4423 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl *n, NvmeRequest *req)
4424 {
4425     NvmeNamespace *ns;
4426     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4427     uint32_t nsid = le32_to_cpu(c->nsid);
4428     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
4429 
4430     struct data {
4431         struct {
4432             NvmeIdNsDescr hdr;
4433             uint8_t v[NVME_NIDL_UUID];
4434         } uuid;
4435         struct {
4436             NvmeIdNsDescr hdr;
4437             uint8_t v;
4438         } csi;
4439     };
4440 
4441     struct data *ns_descrs = (struct data *)list;
4442 
4443     trace_pci_nvme_identify_ns_descr_list(nsid);
4444 
4445     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4446         return NVME_INVALID_NSID | NVME_DNR;
4447     }
4448 
4449     ns = nvme_ns(n, nsid);
4450     if (unlikely(!ns)) {
4451         return NVME_INVALID_FIELD | NVME_DNR;
4452     }
4453 
4454     /*
4455      * Because the NGUID and EUI64 fields are 0 in the Identify Namespace data
4456      * structure, a Namespace UUID (nidt = 3h) must be reported in the
4457      * Namespace Identification Descriptor. Add the namespace UUID here.
4458      */
4459     ns_descrs->uuid.hdr.nidt = NVME_NIDT_UUID;
4460     ns_descrs->uuid.hdr.nidl = NVME_NIDL_UUID;
4461     memcpy(&ns_descrs->uuid.v, ns->params.uuid.data, NVME_NIDL_UUID);
4462 
4463     ns_descrs->csi.hdr.nidt = NVME_NIDT_CSI;
4464     ns_descrs->csi.hdr.nidl = NVME_NIDL_CSI;
4465     ns_descrs->csi.v = ns->csi;
4466 
4467     return nvme_c2h(n, list, sizeof(list), req);
4468 }
4469 
4470 static uint16_t nvme_identify_cmd_set(NvmeCtrl *n, NvmeRequest *req)
4471 {
4472     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
4473     static const int data_len = sizeof(list);
4474 
4475     trace_pci_nvme_identify_cmd_set();
4476 
4477     NVME_SET_CSI(*list, NVME_CSI_NVM);
4478     NVME_SET_CSI(*list, NVME_CSI_ZONED);
4479 
4480     return nvme_c2h(n, list, data_len, req);
4481 }
4482 
4483 static uint16_t nvme_identify(NvmeCtrl *n, NvmeRequest *req)
4484 {
4485     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4486 
4487     trace_pci_nvme_identify(nvme_cid(req), c->cns, le16_to_cpu(c->ctrlid),
4488                             c->csi);
4489 
4490     switch (c->cns) {
4491     case NVME_ID_CNS_NS:
4492         return nvme_identify_ns(n, req, true);
4493     case NVME_ID_CNS_NS_PRESENT:
4494         return nvme_identify_ns(n, req, false);
4495     case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST:
4496         return nvme_identify_ns_attached_list(n, req);
4497     case NVME_ID_CNS_CS_NS:
4498         return nvme_identify_ns_csi(n, req, true);
4499     case NVME_ID_CNS_CS_NS_PRESENT:
4500         return nvme_identify_ns_csi(n, req, false);
4501     case NVME_ID_CNS_CTRL:
4502         return nvme_identify_ctrl(n, req);
4503     case NVME_ID_CNS_CS_CTRL:
4504         return nvme_identify_ctrl_csi(n, req);
4505     case NVME_ID_CNS_NS_ACTIVE_LIST:
4506         return nvme_identify_nslist(n, req, true);
4507     case NVME_ID_CNS_NS_PRESENT_LIST:
4508         return nvme_identify_nslist(n, req, false);
4509     case NVME_ID_CNS_CS_NS_ACTIVE_LIST:
4510         return nvme_identify_nslist_csi(n, req, true);
4511     case NVME_ID_CNS_CS_NS_PRESENT_LIST:
4512         return nvme_identify_nslist_csi(n, req, false);
4513     case NVME_ID_CNS_NS_DESCR_LIST:
4514         return nvme_identify_ns_descr_list(n, req);
4515     case NVME_ID_CNS_IO_COMMAND_SET:
4516         return nvme_identify_cmd_set(n, req);
4517     default:
4518         trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
4519         return NVME_INVALID_FIELD | NVME_DNR;
4520     }
4521 }
4522 
4523 static uint16_t nvme_abort(NvmeCtrl *n, NvmeRequest *req)
4524 {
4525     uint16_t sqid = le32_to_cpu(req->cmd.cdw10) & 0xffff;
4526 
4527     req->cqe.result = 1;
4528     if (nvme_check_sqid(n, sqid)) {
4529         return NVME_INVALID_FIELD | NVME_DNR;
4530     }
4531 
4532     return NVME_SUCCESS;
4533 }
4534 
4535 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
4536 {
4537     trace_pci_nvme_setfeat_timestamp(ts);
4538 
4539     n->host_timestamp = le64_to_cpu(ts);
4540     n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4541 }
4542 
4543 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
4544 {
4545     uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4546     uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
4547 
4548     union nvme_timestamp {
4549         struct {
4550             uint64_t timestamp:48;
4551             uint64_t sync:1;
4552             uint64_t origin:3;
4553             uint64_t rsvd1:12;
4554         };
4555         uint64_t all;
4556     };
4557 
4558     union nvme_timestamp ts;
4559     ts.all = 0;
4560     ts.timestamp = n->host_timestamp + elapsed_time;
4561 
4562     /* If the host timestamp is non-zero, set the timestamp origin */
4563     ts.origin = n->host_timestamp ? 0x01 : 0x00;
4564 
4565     trace_pci_nvme_getfeat_timestamp(ts.all);
4566 
4567     return cpu_to_le64(ts.all);
4568 }
4569 
4570 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
4571 {
4572     uint64_t timestamp = nvme_get_timestamp(n);
4573 
4574     return nvme_c2h(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
4575 }
4576 
4577 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeRequest *req)
4578 {
4579     NvmeCmd *cmd = &req->cmd;
4580     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
4581     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
4582     uint32_t nsid = le32_to_cpu(cmd->nsid);
4583     uint32_t result;
4584     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
4585     NvmeGetFeatureSelect sel = NVME_GETFEAT_SELECT(dw10);
4586     uint16_t iv;
4587     NvmeNamespace *ns;
4588     int i;
4589 
4590     static const uint32_t nvme_feature_default[NVME_FID_MAX] = {
4591         [NVME_ARBITRATION] = NVME_ARB_AB_NOLIMIT,
4592     };
4593 
4594     trace_pci_nvme_getfeat(nvme_cid(req), nsid, fid, sel, dw11);
4595 
4596     if (!nvme_feature_support[fid]) {
4597         return NVME_INVALID_FIELD | NVME_DNR;
4598     }
4599 
4600     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
4601         if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4602             /*
4603              * The Reservation Notification Mask and Reservation Persistence
4604              * features require a status code of Invalid Field in Command when
4605              * NSID is FFFFFFFFh. Since the device does not support those
4606              * features we can always return Invalid Namespace or Format as we
4607              * should do for all other features.
4608              */
4609             return NVME_INVALID_NSID | NVME_DNR;
4610         }
4611 
4612         if (!nvme_ns(n, nsid)) {
4613             return NVME_INVALID_FIELD | NVME_DNR;
4614         }
4615     }
4616 
4617     switch (sel) {
4618     case NVME_GETFEAT_SELECT_CURRENT:
4619         break;
4620     case NVME_GETFEAT_SELECT_SAVED:
4621         /* no features are saveable by the controller; fallthrough */
4622     case NVME_GETFEAT_SELECT_DEFAULT:
4623         goto defaults;
4624     case NVME_GETFEAT_SELECT_CAP:
4625         result = nvme_feature_cap[fid];
4626         goto out;
4627     }
4628 
4629     switch (fid) {
4630     case NVME_TEMPERATURE_THRESHOLD:
4631         result = 0;
4632 
4633         /*
4634          * The controller only implements the Composite Temperature sensor, so
4635          * return 0 for all other sensors.
4636          */
4637         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
4638             goto out;
4639         }
4640 
4641         switch (NVME_TEMP_THSEL(dw11)) {
4642         case NVME_TEMP_THSEL_OVER:
4643             result = n->features.temp_thresh_hi;
4644             goto out;
4645         case NVME_TEMP_THSEL_UNDER:
4646             result = n->features.temp_thresh_low;
4647             goto out;
4648         }
4649 
4650         return NVME_INVALID_FIELD | NVME_DNR;
4651     case NVME_ERROR_RECOVERY:
4652         if (!nvme_nsid_valid(n, nsid)) {
4653             return NVME_INVALID_NSID | NVME_DNR;
4654         }
4655 
4656         ns = nvme_ns(n, nsid);
4657         if (unlikely(!ns)) {
4658             return NVME_INVALID_FIELD | NVME_DNR;
4659         }
4660 
4661         result = ns->features.err_rec;
4662         goto out;
4663     case NVME_VOLATILE_WRITE_CACHE:
4664         result = 0;
4665         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4666             ns = nvme_ns(n, i);
4667             if (!ns) {
4668                 continue;
4669             }
4670 
4671             result = blk_enable_write_cache(ns->blkconf.blk);
4672             if (result) {
4673                 break;
4674             }
4675         }
4676         trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
4677         goto out;
4678     case NVME_ASYNCHRONOUS_EVENT_CONF:
4679         result = n->features.async_config;
4680         goto out;
4681     case NVME_TIMESTAMP:
4682         return nvme_get_feature_timestamp(n, req);
4683     default:
4684         break;
4685     }
4686 
4687 defaults:
4688     switch (fid) {
4689     case NVME_TEMPERATURE_THRESHOLD:
4690         result = 0;
4691 
4692         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
4693             break;
4694         }
4695 
4696         if (NVME_TEMP_THSEL(dw11) == NVME_TEMP_THSEL_OVER) {
4697             result = NVME_TEMPERATURE_WARNING;
4698         }
4699 
4700         break;
4701     case NVME_NUMBER_OF_QUEUES:
4702         result = (n->params.max_ioqpairs - 1) |
4703             ((n->params.max_ioqpairs - 1) << 16);
4704         trace_pci_nvme_getfeat_numq(result);
4705         break;
4706     case NVME_INTERRUPT_VECTOR_CONF:
4707         iv = dw11 & 0xffff;
4708         if (iv >= n->params.max_ioqpairs + 1) {
4709             return NVME_INVALID_FIELD | NVME_DNR;
4710         }
4711 
4712         result = iv;
4713         if (iv == n->admin_cq.vector) {
4714             result |= NVME_INTVC_NOCOALESCING;
4715         }
4716         break;
4717     default:
4718         result = nvme_feature_default[fid];
4719         break;
4720     }
4721 
4722 out:
4723     req->cqe.result = cpu_to_le32(result);
4724     return NVME_SUCCESS;
4725 }
4726 
4727 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
4728 {
4729     uint16_t ret;
4730     uint64_t timestamp;
4731 
4732     ret = nvme_h2c(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
4733     if (ret) {
4734         return ret;
4735     }
4736 
4737     nvme_set_timestamp(n, timestamp);
4738 
4739     return NVME_SUCCESS;
4740 }
4741 
4742 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
4743 {
4744     NvmeNamespace *ns = NULL;
4745 
4746     NvmeCmd *cmd = &req->cmd;
4747     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
4748     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
4749     uint32_t nsid = le32_to_cpu(cmd->nsid);
4750     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
4751     uint8_t save = NVME_SETFEAT_SAVE(dw10);
4752     int i;
4753 
4754     trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
4755 
4756     if (save && !(nvme_feature_cap[fid] & NVME_FEAT_CAP_SAVE)) {
4757         return NVME_FID_NOT_SAVEABLE | NVME_DNR;
4758     }
4759 
4760     if (!nvme_feature_support[fid]) {
4761         return NVME_INVALID_FIELD | NVME_DNR;
4762     }
4763 
4764     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
4765         if (nsid != NVME_NSID_BROADCAST) {
4766             if (!nvme_nsid_valid(n, nsid)) {
4767                 return NVME_INVALID_NSID | NVME_DNR;
4768             }
4769 
4770             ns = nvme_ns(n, nsid);
4771             if (unlikely(!ns)) {
4772                 return NVME_INVALID_FIELD | NVME_DNR;
4773             }
4774         }
4775     } else if (nsid && nsid != NVME_NSID_BROADCAST) {
4776         if (!nvme_nsid_valid(n, nsid)) {
4777             return NVME_INVALID_NSID | NVME_DNR;
4778         }
4779 
4780         return NVME_FEAT_NOT_NS_SPEC | NVME_DNR;
4781     }
4782 
4783     if (!(nvme_feature_cap[fid] & NVME_FEAT_CAP_CHANGE)) {
4784         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
4785     }
4786 
4787     switch (fid) {
4788     case NVME_TEMPERATURE_THRESHOLD:
4789         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
4790             break;
4791         }
4792 
4793         switch (NVME_TEMP_THSEL(dw11)) {
4794         case NVME_TEMP_THSEL_OVER:
4795             n->features.temp_thresh_hi = NVME_TEMP_TMPTH(dw11);
4796             break;
4797         case NVME_TEMP_THSEL_UNDER:
4798             n->features.temp_thresh_low = NVME_TEMP_TMPTH(dw11);
4799             break;
4800         default:
4801             return NVME_INVALID_FIELD | NVME_DNR;
4802         }
4803 
4804         if ((n->temperature >= n->features.temp_thresh_hi) ||
4805             (n->temperature <= n->features.temp_thresh_low)) {
4806             nvme_smart_event(n, NVME_AER_INFO_SMART_TEMP_THRESH);
4807         }
4808 
4809         break;
4810     case NVME_ERROR_RECOVERY:
4811         if (nsid == NVME_NSID_BROADCAST) {
4812             for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4813                 ns = nvme_ns(n, i);
4814 
4815                 if (!ns) {
4816                     continue;
4817                 }
4818 
4819                 if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat)) {
4820                     ns->features.err_rec = dw11;
4821                 }
4822             }
4823 
4824             break;
4825         }
4826 
4827         assert(ns);
4828         if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat))  {
4829             ns->features.err_rec = dw11;
4830         }
4831         break;
4832     case NVME_VOLATILE_WRITE_CACHE:
4833         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4834             ns = nvme_ns(n, i);
4835             if (!ns) {
4836                 continue;
4837             }
4838 
4839             if (!(dw11 & 0x1) && blk_enable_write_cache(ns->blkconf.blk)) {
4840                 blk_flush(ns->blkconf.blk);
4841             }
4842 
4843             blk_set_enable_write_cache(ns->blkconf.blk, dw11 & 1);
4844         }
4845 
4846         break;
4847 
4848     case NVME_NUMBER_OF_QUEUES:
4849         if (n->qs_created) {
4850             return NVME_CMD_SEQ_ERROR | NVME_DNR;
4851         }
4852 
4853         /*
4854          * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
4855          * and NSQR.
4856          */
4857         if ((dw11 & 0xffff) == 0xffff || ((dw11 >> 16) & 0xffff) == 0xffff) {
4858             return NVME_INVALID_FIELD | NVME_DNR;
4859         }
4860 
4861         trace_pci_nvme_setfeat_numq((dw11 & 0xffff) + 1,
4862                                     ((dw11 >> 16) & 0xffff) + 1,
4863                                     n->params.max_ioqpairs,
4864                                     n->params.max_ioqpairs);
4865         req->cqe.result = cpu_to_le32((n->params.max_ioqpairs - 1) |
4866                                       ((n->params.max_ioqpairs - 1) << 16));
4867         break;
4868     case NVME_ASYNCHRONOUS_EVENT_CONF:
4869         n->features.async_config = dw11;
4870         break;
4871     case NVME_TIMESTAMP:
4872         return nvme_set_feature_timestamp(n, req);
4873     case NVME_COMMAND_SET_PROFILE:
4874         if (dw11 & 0x1ff) {
4875             trace_pci_nvme_err_invalid_iocsci(dw11 & 0x1ff);
4876             return NVME_CMD_SET_CMB_REJECTED | NVME_DNR;
4877         }
4878         break;
4879     default:
4880         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
4881     }
4882     return NVME_SUCCESS;
4883 }
4884 
4885 static uint16_t nvme_aer(NvmeCtrl *n, NvmeRequest *req)
4886 {
4887     trace_pci_nvme_aer(nvme_cid(req));
4888 
4889     if (n->outstanding_aers > n->params.aerl) {
4890         trace_pci_nvme_aer_aerl_exceeded();
4891         return NVME_AER_LIMIT_EXCEEDED;
4892     }
4893 
4894     n->aer_reqs[n->outstanding_aers] = req;
4895     n->outstanding_aers++;
4896 
4897     if (!QTAILQ_EMPTY(&n->aer_queue)) {
4898         nvme_process_aers(n);
4899     }
4900 
4901     return NVME_NO_COMPLETE;
4902 }
4903 
4904 static void nvme_update_dmrsl(NvmeCtrl *n)
4905 {
4906     int nsid;
4907 
4908     for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
4909         NvmeNamespace *ns = nvme_ns(n, nsid);
4910         if (!ns) {
4911             continue;
4912         }
4913 
4914         n->dmrsl = MIN_NON_ZERO(n->dmrsl,
4915                                 BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
4916     }
4917 }
4918 
4919 static void nvme_select_iocs_ns(NvmeCtrl *n, NvmeNamespace *ns)
4920 {
4921     ns->iocs = nvme_cse_iocs_none;
4922     switch (ns->csi) {
4923     case NVME_CSI_NVM:
4924         if (NVME_CC_CSS(n->bar.cc) != NVME_CC_CSS_ADMIN_ONLY) {
4925             ns->iocs = nvme_cse_iocs_nvm;
4926         }
4927         break;
4928     case NVME_CSI_ZONED:
4929         if (NVME_CC_CSS(n->bar.cc) == NVME_CC_CSS_CSI) {
4930             ns->iocs = nvme_cse_iocs_zoned;
4931         } else if (NVME_CC_CSS(n->bar.cc) == NVME_CC_CSS_NVM) {
4932             ns->iocs = nvme_cse_iocs_nvm;
4933         }
4934         break;
4935     }
4936 }
4937 
4938 static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
4939 {
4940     NvmeNamespace *ns;
4941     NvmeCtrl *ctrl;
4942     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
4943     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4944     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
4945     bool attach = !(dw10 & 0xf);
4946     uint16_t *nr_ids = &list[0];
4947     uint16_t *ids = &list[1];
4948     uint16_t ret;
4949     int i;
4950 
4951     trace_pci_nvme_ns_attachment(nvme_cid(req), dw10 & 0xf);
4952 
4953     if (!nvme_nsid_valid(n, nsid)) {
4954         return NVME_INVALID_NSID | NVME_DNR;
4955     }
4956 
4957     ns = nvme_subsys_ns(n->subsys, nsid);
4958     if (!ns) {
4959         return NVME_INVALID_FIELD | NVME_DNR;
4960     }
4961 
4962     ret = nvme_h2c(n, (uint8_t *)list, 4096, req);
4963     if (ret) {
4964         return ret;
4965     }
4966 
4967     if (!*nr_ids) {
4968         return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
4969     }
4970 
4971     *nr_ids = MIN(*nr_ids, NVME_CONTROLLER_LIST_SIZE - 1);
4972     for (i = 0; i < *nr_ids; i++) {
4973         ctrl = nvme_subsys_ctrl(n->subsys, ids[i]);
4974         if (!ctrl) {
4975             return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
4976         }
4977 
4978         if (attach) {
4979             if (nvme_ns(ctrl, nsid)) {
4980                 return NVME_NS_ALREADY_ATTACHED | NVME_DNR;
4981             }
4982 
4983             if (ns->attached && !ns->params.shared) {
4984                 return NVME_NS_PRIVATE | NVME_DNR;
4985             }
4986 
4987             nvme_attach_ns(ctrl, ns);
4988             nvme_select_iocs_ns(ctrl, ns);
4989         } else {
4990             if (!nvme_ns(ctrl, nsid)) {
4991                 return NVME_NS_NOT_ATTACHED | NVME_DNR;
4992             }
4993 
4994             ctrl->namespaces[nsid] = NULL;
4995             ns->attached--;
4996 
4997             nvme_update_dmrsl(ctrl);
4998         }
4999 
5000         /*
5001          * Add namespace id to the changed namespace id list for event clearing
5002          * via Get Log Page command.
5003          */
5004         if (!test_and_set_bit(nsid, ctrl->changed_nsids)) {
5005             nvme_enqueue_event(ctrl, NVME_AER_TYPE_NOTICE,
5006                                NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED,
5007                                NVME_LOG_CHANGED_NSLIST);
5008         }
5009     }
5010 
5011     return NVME_SUCCESS;
5012 }
5013 
5014 static uint16_t nvme_format_ns(NvmeCtrl *n, NvmeNamespace *ns, uint8_t lbaf,
5015                                uint8_t mset, uint8_t pi, uint8_t pil,
5016                                NvmeRequest *req)
5017 {
5018     int64_t len, offset;
5019     struct nvme_aio_format_ctx *ctx;
5020     BlockBackend *blk = ns->blkconf.blk;
5021     uint16_t ms;
5022     uintptr_t *num_formats = (uintptr_t *)&req->opaque;
5023     int *count;
5024 
5025     if (ns->params.zoned) {
5026         return NVME_INVALID_FORMAT | NVME_DNR;
5027     }
5028 
5029     trace_pci_nvme_format_ns(nvme_cid(req), nvme_nsid(ns), lbaf, mset, pi, pil);
5030 
5031     if (lbaf > ns->id_ns.nlbaf) {
5032         return NVME_INVALID_FORMAT | NVME_DNR;
5033     }
5034 
5035     ms = ns->id_ns.lbaf[lbaf].ms;
5036 
5037     if (pi && (ms < sizeof(NvmeDifTuple))) {
5038         return NVME_INVALID_FORMAT | NVME_DNR;
5039     }
5040 
5041     if (pi && pi > NVME_ID_NS_DPS_TYPE_3) {
5042         return NVME_INVALID_FIELD | NVME_DNR;
5043     }
5044 
5045     nvme_ns_drain(ns);
5046     nvme_ns_shutdown(ns);
5047     nvme_ns_cleanup(ns);
5048 
5049     ns->id_ns.dps = (pil << 3) | pi;
5050     ns->id_ns.flbas = lbaf | (mset << 4);
5051 
5052     nvme_ns_init_format(ns);
5053 
5054     ns->status = NVME_FORMAT_IN_PROGRESS;
5055 
5056     len = ns->size;
5057     offset = 0;
5058 
5059     count = g_new(int, 1);
5060     *count = 1;
5061 
5062     (*num_formats)++;
5063 
5064     while (len) {
5065         ctx = g_new(struct nvme_aio_format_ctx, 1);
5066         ctx->req = req;
5067         ctx->ns = ns;
5068         ctx->count = count;
5069 
5070         size_t bytes = MIN(BDRV_REQUEST_MAX_BYTES, len);
5071 
5072         (*count)++;
5073 
5074         blk_aio_pwrite_zeroes(blk, offset, bytes, BDRV_REQ_MAY_UNMAP,
5075                               nvme_aio_format_cb, ctx);
5076 
5077         offset += bytes;
5078         len -= bytes;
5079 
5080     }
5081 
5082     if (--(*count)) {
5083         return NVME_NO_COMPLETE;
5084     }
5085 
5086     g_free(count);
5087     ns->status = 0x0;
5088     (*num_formats)--;
5089 
5090     return NVME_SUCCESS;
5091 }
5092 
5093 static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
5094 {
5095     NvmeNamespace *ns;
5096     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
5097     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
5098     uint8_t lbaf = dw10 & 0xf;
5099     uint8_t mset = (dw10 >> 4) & 0x1;
5100     uint8_t pi = (dw10 >> 5) & 0x7;
5101     uint8_t pil = (dw10 >> 8) & 0x1;
5102     uintptr_t *num_formats = (uintptr_t *)&req->opaque;
5103     uint16_t status;
5104     int i;
5105 
5106     trace_pci_nvme_format(nvme_cid(req), nsid, lbaf, mset, pi, pil);
5107 
5108     /* 1-initialize; see the comment in nvme_dsm */
5109     *num_formats = 1;
5110 
5111     if (nsid != NVME_NSID_BROADCAST) {
5112         if (!nvme_nsid_valid(n, nsid)) {
5113             return NVME_INVALID_NSID | NVME_DNR;
5114         }
5115 
5116         ns = nvme_ns(n, nsid);
5117         if (!ns) {
5118             return NVME_INVALID_FIELD | NVME_DNR;
5119         }
5120 
5121         status = nvme_format_ns(n, ns, lbaf, mset, pi, pil, req);
5122         if (status && status != NVME_NO_COMPLETE) {
5123             req->status = status;
5124         }
5125     } else {
5126         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5127             ns = nvme_ns(n, i);
5128             if (!ns) {
5129                 continue;
5130             }
5131 
5132             status = nvme_format_ns(n, ns, lbaf, mset, pi, pil, req);
5133             if (status && status != NVME_NO_COMPLETE) {
5134                 req->status = status;
5135                 break;
5136             }
5137         }
5138     }
5139 
5140     /* account for the 1-initialization */
5141     if (--(*num_formats)) {
5142         return NVME_NO_COMPLETE;
5143     }
5144 
5145     return req->status;
5146 }
5147 
5148 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
5149 {
5150     trace_pci_nvme_admin_cmd(nvme_cid(req), nvme_sqid(req), req->cmd.opcode,
5151                              nvme_adm_opc_str(req->cmd.opcode));
5152 
5153     if (!(nvme_cse_acs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
5154         trace_pci_nvme_err_invalid_admin_opc(req->cmd.opcode);
5155         return NVME_INVALID_OPCODE | NVME_DNR;
5156     }
5157 
5158     /* SGLs shall not be used for Admin commands in NVMe over PCIe */
5159     if (NVME_CMD_FLAGS_PSDT(req->cmd.flags) != NVME_PSDT_PRP) {
5160         return NVME_INVALID_FIELD | NVME_DNR;
5161     }
5162 
5163     switch (req->cmd.opcode) {
5164     case NVME_ADM_CMD_DELETE_SQ:
5165         return nvme_del_sq(n, req);
5166     case NVME_ADM_CMD_CREATE_SQ:
5167         return nvme_create_sq(n, req);
5168     case NVME_ADM_CMD_GET_LOG_PAGE:
5169         return nvme_get_log(n, req);
5170     case NVME_ADM_CMD_DELETE_CQ:
5171         return nvme_del_cq(n, req);
5172     case NVME_ADM_CMD_CREATE_CQ:
5173         return nvme_create_cq(n, req);
5174     case NVME_ADM_CMD_IDENTIFY:
5175         return nvme_identify(n, req);
5176     case NVME_ADM_CMD_ABORT:
5177         return nvme_abort(n, req);
5178     case NVME_ADM_CMD_SET_FEATURES:
5179         return nvme_set_feature(n, req);
5180     case NVME_ADM_CMD_GET_FEATURES:
5181         return nvme_get_feature(n, req);
5182     case NVME_ADM_CMD_ASYNC_EV_REQ:
5183         return nvme_aer(n, req);
5184     case NVME_ADM_CMD_NS_ATTACHMENT:
5185         return nvme_ns_attachment(n, req);
5186     case NVME_ADM_CMD_FORMAT_NVM:
5187         return nvme_format(n, req);
5188     default:
5189         assert(false);
5190     }
5191 
5192     return NVME_INVALID_OPCODE | NVME_DNR;
5193 }
5194 
5195 static void nvme_process_sq(void *opaque)
5196 {
5197     NvmeSQueue *sq = opaque;
5198     NvmeCtrl *n = sq->ctrl;
5199     NvmeCQueue *cq = n->cq[sq->cqid];
5200 
5201     uint16_t status;
5202     hwaddr addr;
5203     NvmeCmd cmd;
5204     NvmeRequest *req;
5205 
5206     while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
5207         addr = sq->dma_addr + sq->head * n->sqe_size;
5208         if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
5209             trace_pci_nvme_err_addr_read(addr);
5210             trace_pci_nvme_err_cfs();
5211             n->bar.csts = NVME_CSTS_FAILED;
5212             break;
5213         }
5214         nvme_inc_sq_head(sq);
5215 
5216         req = QTAILQ_FIRST(&sq->req_list);
5217         QTAILQ_REMOVE(&sq->req_list, req, entry);
5218         QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
5219         nvme_req_clear(req);
5220         req->cqe.cid = cmd.cid;
5221         memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
5222 
5223         status = sq->sqid ? nvme_io_cmd(n, req) :
5224             nvme_admin_cmd(n, req);
5225         if (status != NVME_NO_COMPLETE) {
5226             req->status = status;
5227             nvme_enqueue_req_completion(cq, req);
5228         }
5229     }
5230 }
5231 
5232 static void nvme_ctrl_reset(NvmeCtrl *n)
5233 {
5234     NvmeNamespace *ns;
5235     int i;
5236 
5237     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5238         ns = nvme_ns(n, i);
5239         if (!ns) {
5240             continue;
5241         }
5242 
5243         nvme_ns_drain(ns);
5244     }
5245 
5246     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
5247         if (n->sq[i] != NULL) {
5248             nvme_free_sq(n->sq[i], n);
5249         }
5250     }
5251     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
5252         if (n->cq[i] != NULL) {
5253             nvme_free_cq(n->cq[i], n);
5254         }
5255     }
5256 
5257     while (!QTAILQ_EMPTY(&n->aer_queue)) {
5258         NvmeAsyncEvent *event = QTAILQ_FIRST(&n->aer_queue);
5259         QTAILQ_REMOVE(&n->aer_queue, event, entry);
5260         g_free(event);
5261     }
5262 
5263     n->aer_queued = 0;
5264     n->outstanding_aers = 0;
5265     n->qs_created = false;
5266 
5267     n->bar.cc = 0;
5268 }
5269 
5270 static void nvme_ctrl_shutdown(NvmeCtrl *n)
5271 {
5272     NvmeNamespace *ns;
5273     int i;
5274 
5275     if (n->pmr.dev) {
5276         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
5277     }
5278 
5279     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5280         ns = nvme_ns(n, i);
5281         if (!ns) {
5282             continue;
5283         }
5284 
5285         nvme_ns_shutdown(ns);
5286     }
5287 }
5288 
5289 static void nvme_select_iocs(NvmeCtrl *n)
5290 {
5291     NvmeNamespace *ns;
5292     int i;
5293 
5294     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5295         ns = nvme_ns(n, i);
5296         if (!ns) {
5297             continue;
5298         }
5299 
5300         nvme_select_iocs_ns(n, ns);
5301     }
5302 }
5303 
5304 static int nvme_start_ctrl(NvmeCtrl *n)
5305 {
5306     uint32_t page_bits = NVME_CC_MPS(n->bar.cc) + 12;
5307     uint32_t page_size = 1 << page_bits;
5308 
5309     if (unlikely(n->cq[0])) {
5310         trace_pci_nvme_err_startfail_cq();
5311         return -1;
5312     }
5313     if (unlikely(n->sq[0])) {
5314         trace_pci_nvme_err_startfail_sq();
5315         return -1;
5316     }
5317     if (unlikely(!n->bar.asq)) {
5318         trace_pci_nvme_err_startfail_nbarasq();
5319         return -1;
5320     }
5321     if (unlikely(!n->bar.acq)) {
5322         trace_pci_nvme_err_startfail_nbaracq();
5323         return -1;
5324     }
5325     if (unlikely(n->bar.asq & (page_size - 1))) {
5326         trace_pci_nvme_err_startfail_asq_misaligned(n->bar.asq);
5327         return -1;
5328     }
5329     if (unlikely(n->bar.acq & (page_size - 1))) {
5330         trace_pci_nvme_err_startfail_acq_misaligned(n->bar.acq);
5331         return -1;
5332     }
5333     if (unlikely(!(NVME_CAP_CSS(n->bar.cap) & (1 << NVME_CC_CSS(n->bar.cc))))) {
5334         trace_pci_nvme_err_startfail_css(NVME_CC_CSS(n->bar.cc));
5335         return -1;
5336     }
5337     if (unlikely(NVME_CC_MPS(n->bar.cc) <
5338                  NVME_CAP_MPSMIN(n->bar.cap))) {
5339         trace_pci_nvme_err_startfail_page_too_small(
5340                     NVME_CC_MPS(n->bar.cc),
5341                     NVME_CAP_MPSMIN(n->bar.cap));
5342         return -1;
5343     }
5344     if (unlikely(NVME_CC_MPS(n->bar.cc) >
5345                  NVME_CAP_MPSMAX(n->bar.cap))) {
5346         trace_pci_nvme_err_startfail_page_too_large(
5347                     NVME_CC_MPS(n->bar.cc),
5348                     NVME_CAP_MPSMAX(n->bar.cap));
5349         return -1;
5350     }
5351     if (unlikely(NVME_CC_IOCQES(n->bar.cc) <
5352                  NVME_CTRL_CQES_MIN(n->id_ctrl.cqes))) {
5353         trace_pci_nvme_err_startfail_cqent_too_small(
5354                     NVME_CC_IOCQES(n->bar.cc),
5355                     NVME_CTRL_CQES_MIN(n->bar.cap));
5356         return -1;
5357     }
5358     if (unlikely(NVME_CC_IOCQES(n->bar.cc) >
5359                  NVME_CTRL_CQES_MAX(n->id_ctrl.cqes))) {
5360         trace_pci_nvme_err_startfail_cqent_too_large(
5361                     NVME_CC_IOCQES(n->bar.cc),
5362                     NVME_CTRL_CQES_MAX(n->bar.cap));
5363         return -1;
5364     }
5365     if (unlikely(NVME_CC_IOSQES(n->bar.cc) <
5366                  NVME_CTRL_SQES_MIN(n->id_ctrl.sqes))) {
5367         trace_pci_nvme_err_startfail_sqent_too_small(
5368                     NVME_CC_IOSQES(n->bar.cc),
5369                     NVME_CTRL_SQES_MIN(n->bar.cap));
5370         return -1;
5371     }
5372     if (unlikely(NVME_CC_IOSQES(n->bar.cc) >
5373                  NVME_CTRL_SQES_MAX(n->id_ctrl.sqes))) {
5374         trace_pci_nvme_err_startfail_sqent_too_large(
5375                     NVME_CC_IOSQES(n->bar.cc),
5376                     NVME_CTRL_SQES_MAX(n->bar.cap));
5377         return -1;
5378     }
5379     if (unlikely(!NVME_AQA_ASQS(n->bar.aqa))) {
5380         trace_pci_nvme_err_startfail_asqent_sz_zero();
5381         return -1;
5382     }
5383     if (unlikely(!NVME_AQA_ACQS(n->bar.aqa))) {
5384         trace_pci_nvme_err_startfail_acqent_sz_zero();
5385         return -1;
5386     }
5387 
5388     n->page_bits = page_bits;
5389     n->page_size = page_size;
5390     n->max_prp_ents = n->page_size / sizeof(uint64_t);
5391     n->cqe_size = 1 << NVME_CC_IOCQES(n->bar.cc);
5392     n->sqe_size = 1 << NVME_CC_IOSQES(n->bar.cc);
5393     nvme_init_cq(&n->admin_cq, n, n->bar.acq, 0, 0,
5394                  NVME_AQA_ACQS(n->bar.aqa) + 1, 1);
5395     nvme_init_sq(&n->admin_sq, n, n->bar.asq, 0, 0,
5396                  NVME_AQA_ASQS(n->bar.aqa) + 1);
5397 
5398     nvme_set_timestamp(n, 0ULL);
5399 
5400     QTAILQ_INIT(&n->aer_queue);
5401 
5402     nvme_select_iocs(n);
5403 
5404     return 0;
5405 }
5406 
5407 static void nvme_cmb_enable_regs(NvmeCtrl *n)
5408 {
5409     NVME_CMBLOC_SET_CDPCILS(n->bar.cmbloc, 1);
5410     NVME_CMBLOC_SET_CDPMLS(n->bar.cmbloc, 1);
5411     NVME_CMBLOC_SET_BIR(n->bar.cmbloc, NVME_CMB_BIR);
5412 
5413     NVME_CMBSZ_SET_SQS(n->bar.cmbsz, 1);
5414     NVME_CMBSZ_SET_CQS(n->bar.cmbsz, 0);
5415     NVME_CMBSZ_SET_LISTS(n->bar.cmbsz, 1);
5416     NVME_CMBSZ_SET_RDS(n->bar.cmbsz, 1);
5417     NVME_CMBSZ_SET_WDS(n->bar.cmbsz, 1);
5418     NVME_CMBSZ_SET_SZU(n->bar.cmbsz, 2); /* MBs */
5419     NVME_CMBSZ_SET_SZ(n->bar.cmbsz, n->params.cmb_size_mb);
5420 }
5421 
5422 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
5423                            unsigned size)
5424 {
5425     if (unlikely(offset & (sizeof(uint32_t) - 1))) {
5426         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
5427                        "MMIO write not 32-bit aligned,"
5428                        " offset=0x%"PRIx64"", offset);
5429         /* should be ignored, fall through for now */
5430     }
5431 
5432     if (unlikely(size < sizeof(uint32_t))) {
5433         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
5434                        "MMIO write smaller than 32-bits,"
5435                        " offset=0x%"PRIx64", size=%u",
5436                        offset, size);
5437         /* should be ignored, fall through for now */
5438     }
5439 
5440     switch (offset) {
5441     case 0xc:   /* INTMS */
5442         if (unlikely(msix_enabled(&(n->parent_obj)))) {
5443             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
5444                            "undefined access to interrupt mask set"
5445                            " when MSI-X is enabled");
5446             /* should be ignored, fall through for now */
5447         }
5448         n->bar.intms |= data & 0xffffffff;
5449         n->bar.intmc = n->bar.intms;
5450         trace_pci_nvme_mmio_intm_set(data & 0xffffffff, n->bar.intmc);
5451         nvme_irq_check(n);
5452         break;
5453     case 0x10:  /* INTMC */
5454         if (unlikely(msix_enabled(&(n->parent_obj)))) {
5455             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
5456                            "undefined access to interrupt mask clr"
5457                            " when MSI-X is enabled");
5458             /* should be ignored, fall through for now */
5459         }
5460         n->bar.intms &= ~(data & 0xffffffff);
5461         n->bar.intmc = n->bar.intms;
5462         trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, n->bar.intmc);
5463         nvme_irq_check(n);
5464         break;
5465     case 0x14:  /* CC */
5466         trace_pci_nvme_mmio_cfg(data & 0xffffffff);
5467         /* Windows first sends data, then sends enable bit */
5468         if (!NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc) &&
5469             !NVME_CC_SHN(data) && !NVME_CC_SHN(n->bar.cc))
5470         {
5471             n->bar.cc = data;
5472         }
5473 
5474         if (NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc)) {
5475             n->bar.cc = data;
5476             if (unlikely(nvme_start_ctrl(n))) {
5477                 trace_pci_nvme_err_startfail();
5478                 n->bar.csts = NVME_CSTS_FAILED;
5479             } else {
5480                 trace_pci_nvme_mmio_start_success();
5481                 n->bar.csts = NVME_CSTS_READY;
5482             }
5483         } else if (!NVME_CC_EN(data) && NVME_CC_EN(n->bar.cc)) {
5484             trace_pci_nvme_mmio_stopped();
5485             nvme_ctrl_reset(n);
5486             n->bar.csts &= ~NVME_CSTS_READY;
5487         }
5488         if (NVME_CC_SHN(data) && !(NVME_CC_SHN(n->bar.cc))) {
5489             trace_pci_nvme_mmio_shutdown_set();
5490             nvme_ctrl_shutdown(n);
5491             n->bar.cc = data;
5492             n->bar.csts |= NVME_CSTS_SHST_COMPLETE;
5493         } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(n->bar.cc)) {
5494             trace_pci_nvme_mmio_shutdown_cleared();
5495             n->bar.csts &= ~NVME_CSTS_SHST_COMPLETE;
5496             n->bar.cc = data;
5497         }
5498         break;
5499     case 0x1c:  /* CSTS */
5500         if (data & (1 << 4)) {
5501             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
5502                            "attempted to W1C CSTS.NSSRO"
5503                            " but CAP.NSSRS is zero (not supported)");
5504         } else if (data != 0) {
5505             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
5506                            "attempted to set a read only bit"
5507                            " of controller status");
5508         }
5509         break;
5510     case 0x20:  /* NSSR */
5511         if (data == 0x4e564d65) {
5512             trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
5513         } else {
5514             /* The spec says that writes of other values have no effect */
5515             return;
5516         }
5517         break;
5518     case 0x24:  /* AQA */
5519         n->bar.aqa = data & 0xffffffff;
5520         trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
5521         break;
5522     case 0x28:  /* ASQ */
5523         n->bar.asq = size == 8 ? data :
5524             (n->bar.asq & ~0xffffffffULL) | (data & 0xffffffff);
5525         trace_pci_nvme_mmio_asqaddr(data);
5526         break;
5527     case 0x2c:  /* ASQ hi */
5528         n->bar.asq = (n->bar.asq & 0xffffffff) | (data << 32);
5529         trace_pci_nvme_mmio_asqaddr_hi(data, n->bar.asq);
5530         break;
5531     case 0x30:  /* ACQ */
5532         trace_pci_nvme_mmio_acqaddr(data);
5533         n->bar.acq = size == 8 ? data :
5534             (n->bar.acq & ~0xffffffffULL) | (data & 0xffffffff);
5535         break;
5536     case 0x34:  /* ACQ hi */
5537         n->bar.acq = (n->bar.acq & 0xffffffff) | (data << 32);
5538         trace_pci_nvme_mmio_acqaddr_hi(data, n->bar.acq);
5539         break;
5540     case 0x38:  /* CMBLOC */
5541         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
5542                        "invalid write to reserved CMBLOC"
5543                        " when CMBSZ is zero, ignored");
5544         return;
5545     case 0x3C:  /* CMBSZ */
5546         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
5547                        "invalid write to read only CMBSZ, ignored");
5548         return;
5549     case 0x50:  /* CMBMSC */
5550         if (!NVME_CAP_CMBS(n->bar.cap)) {
5551             return;
5552         }
5553 
5554         n->bar.cmbmsc = size == 8 ? data :
5555             (n->bar.cmbmsc & ~0xffffffff) | (data & 0xffffffff);
5556         n->cmb.cmse = false;
5557 
5558         if (NVME_CMBMSC_CRE(data)) {
5559             nvme_cmb_enable_regs(n);
5560 
5561             if (NVME_CMBMSC_CMSE(data)) {
5562                 hwaddr cba = NVME_CMBMSC_CBA(data) << CMBMSC_CBA_SHIFT;
5563                 if (cba + int128_get64(n->cmb.mem.size) < cba) {
5564                     NVME_CMBSTS_SET_CBAI(n->bar.cmbsts, 1);
5565                     return;
5566                 }
5567 
5568                 n->cmb.cba = cba;
5569                 n->cmb.cmse = true;
5570             }
5571         } else {
5572             n->bar.cmbsz = 0;
5573             n->bar.cmbloc = 0;
5574         }
5575 
5576         return;
5577     case 0x54:  /* CMBMSC hi */
5578         n->bar.cmbmsc = (n->bar.cmbmsc & 0xffffffff) | (data << 32);
5579         return;
5580 
5581     case 0xe00: /* PMRCAP */
5582         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
5583                        "invalid write to PMRCAP register, ignored");
5584         return;
5585     case 0xe04: /* PMRCTL */
5586         n->bar.pmrctl = data;
5587         if (NVME_PMRCTL_EN(data)) {
5588             memory_region_set_enabled(&n->pmr.dev->mr, true);
5589             n->bar.pmrsts = 0;
5590         } else {
5591             memory_region_set_enabled(&n->pmr.dev->mr, false);
5592             NVME_PMRSTS_SET_NRDY(n->bar.pmrsts, 1);
5593             n->pmr.cmse = false;
5594         }
5595         return;
5596     case 0xe08: /* PMRSTS */
5597         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
5598                        "invalid write to PMRSTS register, ignored");
5599         return;
5600     case 0xe0C: /* PMREBS */
5601         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
5602                        "invalid write to PMREBS register, ignored");
5603         return;
5604     case 0xe10: /* PMRSWTP */
5605         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
5606                        "invalid write to PMRSWTP register, ignored");
5607         return;
5608     case 0xe14: /* PMRMSCL */
5609         if (!NVME_CAP_PMRS(n->bar.cap)) {
5610             return;
5611         }
5612 
5613         n->bar.pmrmsc = (n->bar.pmrmsc & ~0xffffffff) | (data & 0xffffffff);
5614         n->pmr.cmse = false;
5615 
5616         if (NVME_PMRMSC_CMSE(n->bar.pmrmsc)) {
5617             hwaddr cba = NVME_PMRMSC_CBA(n->bar.pmrmsc) << PMRMSC_CBA_SHIFT;
5618             if (cba + int128_get64(n->pmr.dev->mr.size) < cba) {
5619                 NVME_PMRSTS_SET_CBAI(n->bar.pmrsts, 1);
5620                 return;
5621             }
5622 
5623             n->pmr.cmse = true;
5624             n->pmr.cba = cba;
5625         }
5626 
5627         return;
5628     case 0xe18: /* PMRMSCU */
5629         if (!NVME_CAP_PMRS(n->bar.cap)) {
5630             return;
5631         }
5632 
5633         n->bar.pmrmsc = (n->bar.pmrmsc & 0xffffffff) | (data << 32);
5634         return;
5635     default:
5636         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
5637                        "invalid MMIO write,"
5638                        " offset=0x%"PRIx64", data=%"PRIx64"",
5639                        offset, data);
5640         break;
5641     }
5642 }
5643 
5644 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
5645 {
5646     NvmeCtrl *n = (NvmeCtrl *)opaque;
5647     uint8_t *ptr = (uint8_t *)&n->bar;
5648     uint64_t val = 0;
5649 
5650     trace_pci_nvme_mmio_read(addr, size);
5651 
5652     if (unlikely(addr & (sizeof(uint32_t) - 1))) {
5653         NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
5654                        "MMIO read not 32-bit aligned,"
5655                        " offset=0x%"PRIx64"", addr);
5656         /* should RAZ, fall through for now */
5657     } else if (unlikely(size < sizeof(uint32_t))) {
5658         NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
5659                        "MMIO read smaller than 32-bits,"
5660                        " offset=0x%"PRIx64"", addr);
5661         /* should RAZ, fall through for now */
5662     }
5663 
5664     if (addr < sizeof(n->bar)) {
5665         /*
5666          * When PMRWBM bit 1 is set then read from
5667          * from PMRSTS should ensure prior writes
5668          * made it to persistent media
5669          */
5670         if (addr == 0xe08 &&
5671             (NVME_PMRCAP_PMRWBM(n->bar.pmrcap) & 0x02)) {
5672             memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
5673         }
5674         memcpy(&val, ptr + addr, size);
5675     } else {
5676         NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
5677                        "MMIO read beyond last register,"
5678                        " offset=0x%"PRIx64", returning 0", addr);
5679     }
5680 
5681     return val;
5682 }
5683 
5684 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
5685 {
5686     uint32_t qid;
5687 
5688     if (unlikely(addr & ((1 << 2) - 1))) {
5689         NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
5690                        "doorbell write not 32-bit aligned,"
5691                        " offset=0x%"PRIx64", ignoring", addr);
5692         return;
5693     }
5694 
5695     if (((addr - 0x1000) >> 2) & 1) {
5696         /* Completion queue doorbell write */
5697 
5698         uint16_t new_head = val & 0xffff;
5699         int start_sqs;
5700         NvmeCQueue *cq;
5701 
5702         qid = (addr - (0x1000 + (1 << 2))) >> 3;
5703         if (unlikely(nvme_check_cqid(n, qid))) {
5704             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
5705                            "completion queue doorbell write"
5706                            " for nonexistent queue,"
5707                            " sqid=%"PRIu32", ignoring", qid);
5708 
5709             /*
5710              * NVM Express v1.3d, Section 4.1 state: "If host software writes
5711              * an invalid value to the Submission Queue Tail Doorbell or
5712              * Completion Queue Head Doorbell regiter and an Asynchronous Event
5713              * Request command is outstanding, then an asynchronous event is
5714              * posted to the Admin Completion Queue with a status code of
5715              * Invalid Doorbell Write Value."
5716              *
5717              * Also note that the spec includes the "Invalid Doorbell Register"
5718              * status code, but nowhere does it specify when to use it.
5719              * However, it seems reasonable to use it here in a similar
5720              * fashion.
5721              */
5722             if (n->outstanding_aers) {
5723                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
5724                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
5725                                    NVME_LOG_ERROR_INFO);
5726             }
5727 
5728             return;
5729         }
5730 
5731         cq = n->cq[qid];
5732         if (unlikely(new_head >= cq->size)) {
5733             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
5734                            "completion queue doorbell write value"
5735                            " beyond queue size, sqid=%"PRIu32","
5736                            " new_head=%"PRIu16", ignoring",
5737                            qid, new_head);
5738 
5739             if (n->outstanding_aers) {
5740                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
5741                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
5742                                    NVME_LOG_ERROR_INFO);
5743             }
5744 
5745             return;
5746         }
5747 
5748         trace_pci_nvme_mmio_doorbell_cq(cq->cqid, new_head);
5749 
5750         start_sqs = nvme_cq_full(cq) ? 1 : 0;
5751         cq->head = new_head;
5752         if (start_sqs) {
5753             NvmeSQueue *sq;
5754             QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
5755                 timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
5756             }
5757             timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
5758         }
5759 
5760         if (cq->tail == cq->head) {
5761             nvme_irq_deassert(n, cq);
5762         }
5763     } else {
5764         /* Submission queue doorbell write */
5765 
5766         uint16_t new_tail = val & 0xffff;
5767         NvmeSQueue *sq;
5768 
5769         qid = (addr - 0x1000) >> 3;
5770         if (unlikely(nvme_check_sqid(n, qid))) {
5771             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
5772                            "submission queue doorbell write"
5773                            " for nonexistent queue,"
5774                            " sqid=%"PRIu32", ignoring", qid);
5775 
5776             if (n->outstanding_aers) {
5777                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
5778                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
5779                                    NVME_LOG_ERROR_INFO);
5780             }
5781 
5782             return;
5783         }
5784 
5785         sq = n->sq[qid];
5786         if (unlikely(new_tail >= sq->size)) {
5787             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
5788                            "submission queue doorbell write value"
5789                            " beyond queue size, sqid=%"PRIu32","
5790                            " new_tail=%"PRIu16", ignoring",
5791                            qid, new_tail);
5792 
5793             if (n->outstanding_aers) {
5794                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
5795                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
5796                                    NVME_LOG_ERROR_INFO);
5797             }
5798 
5799             return;
5800         }
5801 
5802         trace_pci_nvme_mmio_doorbell_sq(sq->sqid, new_tail);
5803 
5804         sq->tail = new_tail;
5805         timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
5806     }
5807 }
5808 
5809 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
5810                             unsigned size)
5811 {
5812     NvmeCtrl *n = (NvmeCtrl *)opaque;
5813 
5814     trace_pci_nvme_mmio_write(addr, data, size);
5815 
5816     if (addr < sizeof(n->bar)) {
5817         nvme_write_bar(n, addr, data, size);
5818     } else {
5819         nvme_process_db(n, addr, data);
5820     }
5821 }
5822 
5823 static const MemoryRegionOps nvme_mmio_ops = {
5824     .read = nvme_mmio_read,
5825     .write = nvme_mmio_write,
5826     .endianness = DEVICE_LITTLE_ENDIAN,
5827     .impl = {
5828         .min_access_size = 2,
5829         .max_access_size = 8,
5830     },
5831 };
5832 
5833 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
5834                            unsigned size)
5835 {
5836     NvmeCtrl *n = (NvmeCtrl *)opaque;
5837     stn_le_p(&n->cmb.buf[addr], size, data);
5838 }
5839 
5840 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
5841 {
5842     NvmeCtrl *n = (NvmeCtrl *)opaque;
5843     return ldn_le_p(&n->cmb.buf[addr], size);
5844 }
5845 
5846 static const MemoryRegionOps nvme_cmb_ops = {
5847     .read = nvme_cmb_read,
5848     .write = nvme_cmb_write,
5849     .endianness = DEVICE_LITTLE_ENDIAN,
5850     .impl = {
5851         .min_access_size = 1,
5852         .max_access_size = 8,
5853     },
5854 };
5855 
5856 static void nvme_check_constraints(NvmeCtrl *n, Error **errp)
5857 {
5858     NvmeParams *params = &n->params;
5859 
5860     if (params->num_queues) {
5861         warn_report("num_queues is deprecated; please use max_ioqpairs "
5862                     "instead");
5863 
5864         params->max_ioqpairs = params->num_queues - 1;
5865     }
5866 
5867     if (n->namespace.blkconf.blk && n->subsys) {
5868         error_setg(errp, "subsystem support is unavailable with legacy "
5869                    "namespace ('drive' property)");
5870         return;
5871     }
5872 
5873     if (params->max_ioqpairs < 1 ||
5874         params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
5875         error_setg(errp, "max_ioqpairs must be between 1 and %d",
5876                    NVME_MAX_IOQPAIRS);
5877         return;
5878     }
5879 
5880     if (params->msix_qsize < 1 ||
5881         params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
5882         error_setg(errp, "msix_qsize must be between 1 and %d",
5883                    PCI_MSIX_FLAGS_QSIZE + 1);
5884         return;
5885     }
5886 
5887     if (!params->serial) {
5888         error_setg(errp, "serial property not set");
5889         return;
5890     }
5891 
5892     if (n->pmr.dev) {
5893         if (host_memory_backend_is_mapped(n->pmr.dev)) {
5894             error_setg(errp, "can't use already busy memdev: %s",
5895                        object_get_canonical_path_component(OBJECT(n->pmr.dev)));
5896             return;
5897         }
5898 
5899         if (!is_power_of_2(n->pmr.dev->size)) {
5900             error_setg(errp, "pmr backend size needs to be power of 2 in size");
5901             return;
5902         }
5903 
5904         host_memory_backend_set_mapped(n->pmr.dev, true);
5905     }
5906 
5907     if (n->params.zasl > n->params.mdts) {
5908         error_setg(errp, "zoned.zasl (Zone Append Size Limit) must be less "
5909                    "than or equal to mdts (Maximum Data Transfer Size)");
5910         return;
5911     }
5912 
5913     if (!n->params.vsl) {
5914         error_setg(errp, "vsl must be non-zero");
5915         return;
5916     }
5917 }
5918 
5919 static void nvme_init_state(NvmeCtrl *n)
5920 {
5921     /* add one to max_ioqpairs to account for the admin queue pair */
5922     n->reg_size = pow2ceil(sizeof(NvmeBar) +
5923                            2 * (n->params.max_ioqpairs + 1) * NVME_DB_SIZE);
5924     n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
5925     n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
5926     n->temperature = NVME_TEMPERATURE;
5927     n->features.temp_thresh_hi = NVME_TEMPERATURE_WARNING;
5928     n->starttime_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5929     n->aer_reqs = g_new0(NvmeRequest *, n->params.aerl + 1);
5930 }
5931 
5932 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
5933 {
5934     uint64_t cmb_size = n->params.cmb_size_mb * MiB;
5935 
5936     n->cmb.buf = g_malloc0(cmb_size);
5937     memory_region_init_io(&n->cmb.mem, OBJECT(n), &nvme_cmb_ops, n,
5938                           "nvme-cmb", cmb_size);
5939     pci_register_bar(pci_dev, NVME_CMB_BIR,
5940                      PCI_BASE_ADDRESS_SPACE_MEMORY |
5941                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
5942                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->cmb.mem);
5943 
5944     NVME_CAP_SET_CMBS(n->bar.cap, 1);
5945 
5946     if (n->params.legacy_cmb) {
5947         nvme_cmb_enable_regs(n);
5948         n->cmb.cmse = true;
5949     }
5950 }
5951 
5952 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
5953 {
5954     NVME_PMRCAP_SET_RDS(n->bar.pmrcap, 1);
5955     NVME_PMRCAP_SET_WDS(n->bar.pmrcap, 1);
5956     NVME_PMRCAP_SET_BIR(n->bar.pmrcap, NVME_PMR_BIR);
5957     /* Turn on bit 1 support */
5958     NVME_PMRCAP_SET_PMRWBM(n->bar.pmrcap, 0x02);
5959     NVME_PMRCAP_SET_CMSS(n->bar.pmrcap, 1);
5960 
5961     pci_register_bar(pci_dev, NVME_PMRCAP_BIR(n->bar.pmrcap),
5962                      PCI_BASE_ADDRESS_SPACE_MEMORY |
5963                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
5964                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmr.dev->mr);
5965 
5966     memory_region_set_enabled(&n->pmr.dev->mr, false);
5967 }
5968 
5969 static int nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
5970 {
5971     uint8_t *pci_conf = pci_dev->config;
5972     uint64_t bar_size, msix_table_size, msix_pba_size;
5973     unsigned msix_table_offset, msix_pba_offset;
5974     int ret;
5975 
5976     Error *err = NULL;
5977 
5978     pci_conf[PCI_INTERRUPT_PIN] = 1;
5979     pci_config_set_prog_interface(pci_conf, 0x2);
5980 
5981     if (n->params.use_intel_id) {
5982         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_INTEL);
5983         pci_config_set_device_id(pci_conf, 0x5845);
5984     } else {
5985         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_REDHAT);
5986         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_REDHAT_NVME);
5987     }
5988 
5989     pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
5990     pcie_endpoint_cap_init(pci_dev, 0x80);
5991 
5992     bar_size = QEMU_ALIGN_UP(n->reg_size, 4 * KiB);
5993     msix_table_offset = bar_size;
5994     msix_table_size = PCI_MSIX_ENTRY_SIZE * n->params.msix_qsize;
5995 
5996     bar_size += msix_table_size;
5997     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
5998     msix_pba_offset = bar_size;
5999     msix_pba_size = QEMU_ALIGN_UP(n->params.msix_qsize, 64) / 8;
6000 
6001     bar_size += msix_pba_size;
6002     bar_size = pow2ceil(bar_size);
6003 
6004     memory_region_init(&n->bar0, OBJECT(n), "nvme-bar0", bar_size);
6005     memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
6006                           n->reg_size);
6007     memory_region_add_subregion(&n->bar0, 0, &n->iomem);
6008 
6009     pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
6010                      PCI_BASE_ADDRESS_MEM_TYPE_64, &n->bar0);
6011     ret = msix_init(pci_dev, n->params.msix_qsize,
6012                     &n->bar0, 0, msix_table_offset,
6013                     &n->bar0, 0, msix_pba_offset, 0, &err);
6014     if (ret < 0) {
6015         if (ret == -ENOTSUP) {
6016             warn_report_err(err);
6017         } else {
6018             error_propagate(errp, err);
6019             return ret;
6020         }
6021     }
6022 
6023     if (n->params.cmb_size_mb) {
6024         nvme_init_cmb(n, pci_dev);
6025     }
6026 
6027     if (n->pmr.dev) {
6028         nvme_init_pmr(n, pci_dev);
6029     }
6030 
6031     return 0;
6032 }
6033 
6034 static void nvme_init_subnqn(NvmeCtrl *n)
6035 {
6036     NvmeSubsystem *subsys = n->subsys;
6037     NvmeIdCtrl *id = &n->id_ctrl;
6038 
6039     if (!subsys) {
6040         snprintf((char *)id->subnqn, sizeof(id->subnqn),
6041                  "nqn.2019-08.org.qemu:%s", n->params.serial);
6042     } else {
6043         pstrcpy((char *)id->subnqn, sizeof(id->subnqn), (char*)subsys->subnqn);
6044     }
6045 }
6046 
6047 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
6048 {
6049     NvmeIdCtrl *id = &n->id_ctrl;
6050     uint8_t *pci_conf = pci_dev->config;
6051 
6052     id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
6053     id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
6054     strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
6055     strpadcpy((char *)id->fr, sizeof(id->fr), "1.0", ' ');
6056     strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
6057 
6058     id->cntlid = cpu_to_le16(n->cntlid);
6059 
6060     id->oaes = cpu_to_le32(NVME_OAES_NS_ATTR);
6061 
6062     id->rab = 6;
6063 
6064     if (n->params.use_intel_id) {
6065         id->ieee[0] = 0xb3;
6066         id->ieee[1] = 0x02;
6067         id->ieee[2] = 0x00;
6068     } else {
6069         id->ieee[0] = 0x00;
6070         id->ieee[1] = 0x54;
6071         id->ieee[2] = 0x52;
6072     }
6073 
6074     id->mdts = n->params.mdts;
6075     id->ver = cpu_to_le32(NVME_SPEC_VER);
6076     id->oacs = cpu_to_le16(NVME_OACS_NS_MGMT | NVME_OACS_FORMAT);
6077     id->cntrltype = 0x1;
6078 
6079     /*
6080      * Because the controller always completes the Abort command immediately,
6081      * there can never be more than one concurrently executing Abort command,
6082      * so this value is never used for anything. Note that there can easily be
6083      * many Abort commands in the queues, but they are not considered
6084      * "executing" until processed by nvme_abort.
6085      *
6086      * The specification recommends a value of 3 for Abort Command Limit (four
6087      * concurrently outstanding Abort commands), so lets use that though it is
6088      * inconsequential.
6089      */
6090     id->acl = 3;
6091     id->aerl = n->params.aerl;
6092     id->frmw = (NVME_NUM_FW_SLOTS << 1) | NVME_FRMW_SLOT1_RO;
6093     id->lpa = NVME_LPA_NS_SMART | NVME_LPA_CSE | NVME_LPA_EXTENDED;
6094 
6095     /* recommended default value (~70 C) */
6096     id->wctemp = cpu_to_le16(NVME_TEMPERATURE_WARNING);
6097     id->cctemp = cpu_to_le16(NVME_TEMPERATURE_CRITICAL);
6098 
6099     id->sqes = (0x6 << 4) | 0x6;
6100     id->cqes = (0x4 << 4) | 0x4;
6101     id->nn = cpu_to_le32(NVME_MAX_NAMESPACES);
6102     id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP |
6103                            NVME_ONCS_FEATURES | NVME_ONCS_DSM |
6104                            NVME_ONCS_COMPARE | NVME_ONCS_COPY);
6105 
6106     /*
6107      * NOTE: If this device ever supports a command set that does NOT use 0x0
6108      * as a Flush-equivalent operation, support for the broadcast NSID in Flush
6109      * should probably be removed.
6110      *
6111      * See comment in nvme_io_cmd.
6112      */
6113     id->vwc = NVME_VWC_NSID_BROADCAST_SUPPORT | NVME_VWC_PRESENT;
6114 
6115     id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0);
6116     id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN |
6117                            NVME_CTRL_SGLS_BITBUCKET);
6118 
6119     nvme_init_subnqn(n);
6120 
6121     id->psd[0].mp = cpu_to_le16(0x9c4);
6122     id->psd[0].enlat = cpu_to_le32(0x10);
6123     id->psd[0].exlat = cpu_to_le32(0x4);
6124 
6125     if (n->subsys) {
6126         id->cmic |= NVME_CMIC_MULTI_CTRL;
6127     }
6128 
6129     NVME_CAP_SET_MQES(n->bar.cap, 0x7ff);
6130     NVME_CAP_SET_CQR(n->bar.cap, 1);
6131     NVME_CAP_SET_TO(n->bar.cap, 0xf);
6132     NVME_CAP_SET_CSS(n->bar.cap, NVME_CAP_CSS_NVM);
6133     NVME_CAP_SET_CSS(n->bar.cap, NVME_CAP_CSS_CSI_SUPP);
6134     NVME_CAP_SET_CSS(n->bar.cap, NVME_CAP_CSS_ADMIN_ONLY);
6135     NVME_CAP_SET_MPSMAX(n->bar.cap, 4);
6136     NVME_CAP_SET_CMBS(n->bar.cap, n->params.cmb_size_mb ? 1 : 0);
6137     NVME_CAP_SET_PMRS(n->bar.cap, n->pmr.dev ? 1 : 0);
6138 
6139     n->bar.vs = NVME_SPEC_VER;
6140     n->bar.intmc = n->bar.intms = 0;
6141 }
6142 
6143 static int nvme_init_subsys(NvmeCtrl *n, Error **errp)
6144 {
6145     int cntlid;
6146 
6147     if (!n->subsys) {
6148         return 0;
6149     }
6150 
6151     cntlid = nvme_subsys_register_ctrl(n, errp);
6152     if (cntlid < 0) {
6153         return -1;
6154     }
6155 
6156     n->cntlid = cntlid;
6157 
6158     return 0;
6159 }
6160 
6161 void nvme_attach_ns(NvmeCtrl *n, NvmeNamespace *ns)
6162 {
6163     uint32_t nsid = ns->params.nsid;
6164     assert(nsid && nsid <= NVME_MAX_NAMESPACES);
6165 
6166     n->namespaces[nsid] = ns;
6167     ns->attached++;
6168 
6169     n->dmrsl = MIN_NON_ZERO(n->dmrsl,
6170                             BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
6171 }
6172 
6173 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
6174 {
6175     NvmeCtrl *n = NVME(pci_dev);
6176     NvmeNamespace *ns;
6177     Error *local_err = NULL;
6178 
6179     nvme_check_constraints(n, &local_err);
6180     if (local_err) {
6181         error_propagate(errp, local_err);
6182         return;
6183     }
6184 
6185     qbus_create_inplace(&n->bus, sizeof(NvmeBus), TYPE_NVME_BUS,
6186                         &pci_dev->qdev, n->parent_obj.qdev.id);
6187 
6188     nvme_init_state(n);
6189     if (nvme_init_pci(n, pci_dev, errp)) {
6190         return;
6191     }
6192 
6193     if (nvme_init_subsys(n, errp)) {
6194         error_propagate(errp, local_err);
6195         return;
6196     }
6197     nvme_init_ctrl(n, pci_dev);
6198 
6199     /* setup a namespace if the controller drive property was given */
6200     if (n->namespace.blkconf.blk) {
6201         ns = &n->namespace;
6202         ns->params.nsid = 1;
6203 
6204         if (nvme_ns_setup(n, ns, errp)) {
6205             return;
6206         }
6207 
6208         nvme_attach_ns(n, ns);
6209     }
6210 }
6211 
6212 static void nvme_exit(PCIDevice *pci_dev)
6213 {
6214     NvmeCtrl *n = NVME(pci_dev);
6215     NvmeNamespace *ns;
6216     int i;
6217 
6218     nvme_ctrl_reset(n);
6219 
6220     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6221         ns = nvme_ns(n, i);
6222         if (!ns) {
6223             continue;
6224         }
6225 
6226         nvme_ns_cleanup(ns);
6227     }
6228 
6229     g_free(n->cq);
6230     g_free(n->sq);
6231     g_free(n->aer_reqs);
6232 
6233     if (n->params.cmb_size_mb) {
6234         g_free(n->cmb.buf);
6235     }
6236 
6237     if (n->pmr.dev) {
6238         host_memory_backend_set_mapped(n->pmr.dev, false);
6239     }
6240     msix_uninit(pci_dev, &n->bar0, &n->bar0);
6241     memory_region_del_subregion(&n->bar0, &n->iomem);
6242 }
6243 
6244 static Property nvme_props[] = {
6245     DEFINE_BLOCK_PROPERTIES(NvmeCtrl, namespace.blkconf),
6246     DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmr.dev, TYPE_MEMORY_BACKEND,
6247                      HostMemoryBackend *),
6248     DEFINE_PROP_LINK("subsys", NvmeCtrl, subsys, TYPE_NVME_SUBSYS,
6249                      NvmeSubsystem *),
6250     DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
6251     DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
6252     DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
6253     DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
6254     DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
6255     DEFINE_PROP_UINT8("aerl", NvmeCtrl, params.aerl, 3),
6256     DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl, params.aer_max_queued, 64),
6257     DEFINE_PROP_UINT8("mdts", NvmeCtrl, params.mdts, 7),
6258     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
6259     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
6260     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
6261     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
6262     DEFINE_PROP_END_OF_LIST(),
6263 };
6264 
6265 static void nvme_get_smart_warning(Object *obj, Visitor *v, const char *name,
6266                                    void *opaque, Error **errp)
6267 {
6268     NvmeCtrl *n = NVME(obj);
6269     uint8_t value = n->smart_critical_warning;
6270 
6271     visit_type_uint8(v, name, &value, errp);
6272 }
6273 
6274 static void nvme_set_smart_warning(Object *obj, Visitor *v, const char *name,
6275                                    void *opaque, Error **errp)
6276 {
6277     NvmeCtrl *n = NVME(obj);
6278     uint8_t value, old_value, cap = 0, index, event;
6279 
6280     if (!visit_type_uint8(v, name, &value, errp)) {
6281         return;
6282     }
6283 
6284     cap = NVME_SMART_SPARE | NVME_SMART_TEMPERATURE | NVME_SMART_RELIABILITY
6285           | NVME_SMART_MEDIA_READ_ONLY | NVME_SMART_FAILED_VOLATILE_MEDIA;
6286     if (NVME_CAP_PMRS(n->bar.cap)) {
6287         cap |= NVME_SMART_PMR_UNRELIABLE;
6288     }
6289 
6290     if ((value & cap) != value) {
6291         error_setg(errp, "unsupported smart critical warning bits: 0x%x",
6292                    value & ~cap);
6293         return;
6294     }
6295 
6296     old_value = n->smart_critical_warning;
6297     n->smart_critical_warning = value;
6298 
6299     /* only inject new bits of smart critical warning */
6300     for (index = 0; index < NVME_SMART_WARN_MAX; index++) {
6301         event = 1 << index;
6302         if (value & ~old_value & event)
6303             nvme_smart_event(n, event);
6304     }
6305 }
6306 
6307 static const VMStateDescription nvme_vmstate = {
6308     .name = "nvme",
6309     .unmigratable = 1,
6310 };
6311 
6312 static void nvme_class_init(ObjectClass *oc, void *data)
6313 {
6314     DeviceClass *dc = DEVICE_CLASS(oc);
6315     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
6316 
6317     pc->realize = nvme_realize;
6318     pc->exit = nvme_exit;
6319     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
6320     pc->revision = 2;
6321 
6322     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
6323     dc->desc = "Non-Volatile Memory Express";
6324     device_class_set_props(dc, nvme_props);
6325     dc->vmsd = &nvme_vmstate;
6326 }
6327 
6328 static void nvme_instance_init(Object *obj)
6329 {
6330     NvmeCtrl *n = NVME(obj);
6331 
6332     device_add_bootindex_property(obj, &n->namespace.blkconf.bootindex,
6333                                   "bootindex", "/namespace@1,0",
6334                                   DEVICE(obj));
6335 
6336     object_property_add(obj, "smart_critical_warning", "uint8",
6337                         nvme_get_smart_warning,
6338                         nvme_set_smart_warning, NULL, NULL);
6339 }
6340 
6341 static const TypeInfo nvme_info = {
6342     .name          = TYPE_NVME,
6343     .parent        = TYPE_PCI_DEVICE,
6344     .instance_size = sizeof(NvmeCtrl),
6345     .instance_init = nvme_instance_init,
6346     .class_init    = nvme_class_init,
6347     .interfaces = (InterfaceInfo[]) {
6348         { INTERFACE_PCIE_DEVICE },
6349         { }
6350     },
6351 };
6352 
6353 static const TypeInfo nvme_bus_info = {
6354     .name = TYPE_NVME_BUS,
6355     .parent = TYPE_BUS,
6356     .instance_size = sizeof(NvmeBus),
6357 };
6358 
6359 static void nvme_register_types(void)
6360 {
6361     type_register_static(&nvme_info);
6362     type_register_static(&nvme_bus_info);
6363 }
6364 
6365 type_init(nvme_register_types)
6366