xref: /openbmc/qemu/hw/nvme/ctrl.c (revision b0dd0a3d)
1 /*
2  * QEMU NVM Express Controller
3  *
4  * Copyright (c) 2012, Intel Corporation
5  *
6  * Written by Keith Busch <keith.busch@intel.com>
7  *
8  * This code is licensed under the GNU GPL v2 or later.
9  */
10 
11 /**
12  * Reference Specs: http://www.nvmexpress.org, 1.4, 1.3, 1.2, 1.1, 1.0e
13  *
14  *  https://nvmexpress.org/developers/nvme-specification/
15  *
16  *
17  * Notes on coding style
18  * ---------------------
19  * While QEMU coding style prefers lowercase hexadecimals in constants, the
20  * NVMe subsystem use thes format from the NVMe specifications in the comments
21  * (i.e. 'h' suffix instead of '0x' prefix).
22  *
23  * Usage
24  * -----
25  * See docs/system/nvme.rst for extensive documentation.
26  *
27  * Add options:
28  *      -drive file=<file>,if=none,id=<drive_id>
29  *      -device nvme-subsys,id=<subsys_id>,nqn=<nqn_id>
30  *      -device nvme,serial=<serial>,id=<bus_name>, \
31  *              cmb_size_mb=<cmb_size_mb[optional]>, \
32  *              [pmrdev=<mem_backend_file_id>,] \
33  *              max_ioqpairs=<N[optional]>, \
34  *              aerl=<N[optional]>,aer_max_queued=<N[optional]>, \
35  *              mdts=<N[optional]>,vsl=<N[optional]>, \
36  *              zoned.zasl=<N[optional]>, \
37  *              zoned.auto_transition=<on|off[optional]>, \
38  *              subsys=<subsys_id>
39  *      -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
40  *              zoned=<true|false[optional]>, \
41  *              subsys=<subsys_id>,detached=<true|false[optional]>
42  *
43  * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
44  * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. By default, the
45  * device will use the "v1.4 CMB scheme" - use the `legacy-cmb` parameter to
46  * always enable the CMBLOC and CMBSZ registers (v1.3 behavior).
47  *
48  * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
49  * For example:
50  * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
51  *  size=<size> .... -device nvme,...,pmrdev=<mem_id>
52  *
53  * The PMR will use BAR 4/5 exclusively.
54  *
55  * To place controller(s) and namespace(s) to a subsystem, then provide
56  * nvme-subsys device as above.
57  *
58  * nvme subsystem device parameters
59  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
60  * - `nqn`
61  *   This parameter provides the `<nqn_id>` part of the string
62  *   `nqn.2019-08.org.qemu:<nqn_id>` which will be reported in the SUBNQN field
63  *   of subsystem controllers. Note that `<nqn_id>` should be unique per
64  *   subsystem, but this is not enforced by QEMU. If not specified, it will
65  *   default to the value of the `id` parameter (`<subsys_id>`).
66  *
67  * nvme device parameters
68  * ~~~~~~~~~~~~~~~~~~~~~~
69  * - `subsys`
70  *   Specifying this parameter attaches the controller to the subsystem and
71  *   the SUBNQN field in the controller will report the NQN of the subsystem
72  *   device. This also enables multi controller capability represented in
73  *   Identify Controller data structure in CMIC (Controller Multi-path I/O and
74  *   Namesapce Sharing Capabilities).
75  *
76  * - `aerl`
77  *   The Asynchronous Event Request Limit (AERL). Indicates the maximum number
78  *   of concurrently outstanding Asynchronous Event Request commands support
79  *   by the controller. This is a 0's based value.
80  *
81  * - `aer_max_queued`
82  *   This is the maximum number of events that the device will enqueue for
83  *   completion when there are no outstanding AERs. When the maximum number of
84  *   enqueued events are reached, subsequent events will be dropped.
85  *
86  * - `mdts`
87  *   Indicates the maximum data transfer size for a command that transfers data
88  *   between host-accessible memory and the controller. The value is specified
89  *   as a power of two (2^n) and is in units of the minimum memory page size
90  *   (CAP.MPSMIN). The default value is 7 (i.e. 512 KiB).
91  *
92  * - `vsl`
93  *   Indicates the maximum data size limit for the Verify command. Like `mdts`,
94  *   this value is specified as a power of two (2^n) and is in units of the
95  *   minimum memory page size (CAP.MPSMIN). The default value is 7 (i.e. 512
96  *   KiB).
97  *
98  * - `zoned.zasl`
99  *   Indicates the maximum data transfer size for the Zone Append command. Like
100  *   `mdts`, the value is specified as a power of two (2^n) and is in units of
101  *   the minimum memory page size (CAP.MPSMIN). The default value is 0 (i.e.
102  *   defaulting to the value of `mdts`).
103  *
104  * - `zoned.auto_transition`
105  *   Indicates if zones in zone state implicitly opened can be automatically
106  *   transitioned to zone state closed for resource management purposes.
107  *   Defaults to 'on'.
108  *
109  * nvme namespace device parameters
110  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
111  * - `shared`
112  *   When the parent nvme device (as defined explicitly by the 'bus' parameter
113  *   or implicitly by the most recently defined NvmeBus) is linked to an
114  *   nvme-subsys device, the namespace will be attached to all controllers in
115  *   the subsystem. If set to 'off' (the default), the namespace will remain a
116  *   private namespace and may only be attached to a single controller at a
117  *   time.
118  *
119  * - `detached`
120  *   This parameter is only valid together with the `subsys` parameter. If left
121  *   at the default value (`false/off`), the namespace will be attached to all
122  *   controllers in the NVMe subsystem at boot-up. If set to `true/on`, the
123  *   namespace will be available in the subsystem but not attached to any
124  *   controllers.
125  *
126  * Setting `zoned` to true selects Zoned Command Set at the namespace.
127  * In this case, the following namespace properties are available to configure
128  * zoned operation:
129  *     zoned.zone_size=<zone size in bytes, default: 128MiB>
130  *         The number may be followed by K, M, G as in kilo-, mega- or giga-.
131  *
132  *     zoned.zone_capacity=<zone capacity in bytes, default: zone size>
133  *         The value 0 (default) forces zone capacity to be the same as zone
134  *         size. The value of this property may not exceed zone size.
135  *
136  *     zoned.descr_ext_size=<zone descriptor extension size, default 0>
137  *         This value needs to be specified in 64B units. If it is zero,
138  *         namespace(s) will not support zone descriptor extensions.
139  *
140  *     zoned.max_active=<Maximum Active Resources (zones), default: 0>
141  *         The default value means there is no limit to the number of
142  *         concurrently active zones.
143  *
144  *     zoned.max_open=<Maximum Open Resources (zones), default: 0>
145  *         The default value means there is no limit to the number of
146  *         concurrently open zones.
147  *
148  *     zoned.cross_read=<enable RAZB, default: false>
149  *         Setting this property to true enables Read Across Zone Boundaries.
150  */
151 
152 #include "qemu/osdep.h"
153 #include "qemu/cutils.h"
154 #include "qemu/error-report.h"
155 #include "qemu/log.h"
156 #include "qemu/units.h"
157 #include "qapi/error.h"
158 #include "qapi/visitor.h"
159 #include "sysemu/sysemu.h"
160 #include "sysemu/block-backend.h"
161 #include "sysemu/hostmem.h"
162 #include "hw/pci/msix.h"
163 #include "migration/vmstate.h"
164 
165 #include "nvme.h"
166 #include "dif.h"
167 #include "trace.h"
168 
169 #define NVME_MAX_IOQPAIRS 0xffff
170 #define NVME_DB_SIZE  4
171 #define NVME_SPEC_VER 0x00010400
172 #define NVME_CMB_BIR 2
173 #define NVME_PMR_BIR 4
174 #define NVME_TEMPERATURE 0x143
175 #define NVME_TEMPERATURE_WARNING 0x157
176 #define NVME_TEMPERATURE_CRITICAL 0x175
177 #define NVME_NUM_FW_SLOTS 1
178 #define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
179 
180 #define NVME_GUEST_ERR(trace, fmt, ...) \
181     do { \
182         (trace_##trace)(__VA_ARGS__); \
183         qemu_log_mask(LOG_GUEST_ERROR, #trace \
184             " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
185     } while (0)
186 
187 static const bool nvme_feature_support[NVME_FID_MAX] = {
188     [NVME_ARBITRATION]              = true,
189     [NVME_POWER_MANAGEMENT]         = true,
190     [NVME_TEMPERATURE_THRESHOLD]    = true,
191     [NVME_ERROR_RECOVERY]           = true,
192     [NVME_VOLATILE_WRITE_CACHE]     = true,
193     [NVME_NUMBER_OF_QUEUES]         = true,
194     [NVME_INTERRUPT_COALESCING]     = true,
195     [NVME_INTERRUPT_VECTOR_CONF]    = true,
196     [NVME_WRITE_ATOMICITY]          = true,
197     [NVME_ASYNCHRONOUS_EVENT_CONF]  = true,
198     [NVME_TIMESTAMP]                = true,
199     [NVME_HOST_BEHAVIOR_SUPPORT]    = true,
200     [NVME_COMMAND_SET_PROFILE]      = true,
201 };
202 
203 static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
204     [NVME_TEMPERATURE_THRESHOLD]    = NVME_FEAT_CAP_CHANGE,
205     [NVME_ERROR_RECOVERY]           = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
206     [NVME_VOLATILE_WRITE_CACHE]     = NVME_FEAT_CAP_CHANGE,
207     [NVME_NUMBER_OF_QUEUES]         = NVME_FEAT_CAP_CHANGE,
208     [NVME_ASYNCHRONOUS_EVENT_CONF]  = NVME_FEAT_CAP_CHANGE,
209     [NVME_TIMESTAMP]                = NVME_FEAT_CAP_CHANGE,
210     [NVME_HOST_BEHAVIOR_SUPPORT]    = NVME_FEAT_CAP_CHANGE,
211     [NVME_COMMAND_SET_PROFILE]      = NVME_FEAT_CAP_CHANGE,
212 };
213 
214 static const uint32_t nvme_cse_acs[256] = {
215     [NVME_ADM_CMD_DELETE_SQ]        = NVME_CMD_EFF_CSUPP,
216     [NVME_ADM_CMD_CREATE_SQ]        = NVME_CMD_EFF_CSUPP,
217     [NVME_ADM_CMD_GET_LOG_PAGE]     = NVME_CMD_EFF_CSUPP,
218     [NVME_ADM_CMD_DELETE_CQ]        = NVME_CMD_EFF_CSUPP,
219     [NVME_ADM_CMD_CREATE_CQ]        = NVME_CMD_EFF_CSUPP,
220     [NVME_ADM_CMD_IDENTIFY]         = NVME_CMD_EFF_CSUPP,
221     [NVME_ADM_CMD_ABORT]            = NVME_CMD_EFF_CSUPP,
222     [NVME_ADM_CMD_SET_FEATURES]     = NVME_CMD_EFF_CSUPP,
223     [NVME_ADM_CMD_GET_FEATURES]     = NVME_CMD_EFF_CSUPP,
224     [NVME_ADM_CMD_ASYNC_EV_REQ]     = NVME_CMD_EFF_CSUPP,
225     [NVME_ADM_CMD_NS_ATTACHMENT]    = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_NIC,
226     [NVME_ADM_CMD_FORMAT_NVM]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
227 };
228 
229 static const uint32_t nvme_cse_iocs_none[256];
230 
231 static const uint32_t nvme_cse_iocs_nvm[256] = {
232     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
233     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
234     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
235     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
236     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
237     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
238     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
239     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
240 };
241 
242 static const uint32_t nvme_cse_iocs_zoned[256] = {
243     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
244     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
245     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
246     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
247     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
248     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
249     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
250     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
251     [NVME_CMD_ZONE_APPEND]          = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
252     [NVME_CMD_ZONE_MGMT_SEND]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
253     [NVME_CMD_ZONE_MGMT_RECV]       = NVME_CMD_EFF_CSUPP,
254 };
255 
256 static void nvme_process_sq(void *opaque);
257 
258 static uint16_t nvme_sqid(NvmeRequest *req)
259 {
260     return le16_to_cpu(req->sq->sqid);
261 }
262 
263 static void nvme_assign_zone_state(NvmeNamespace *ns, NvmeZone *zone,
264                                    NvmeZoneState state)
265 {
266     if (QTAILQ_IN_USE(zone, entry)) {
267         switch (nvme_get_zone_state(zone)) {
268         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
269             QTAILQ_REMOVE(&ns->exp_open_zones, zone, entry);
270             break;
271         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
272             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
273             break;
274         case NVME_ZONE_STATE_CLOSED:
275             QTAILQ_REMOVE(&ns->closed_zones, zone, entry);
276             break;
277         case NVME_ZONE_STATE_FULL:
278             QTAILQ_REMOVE(&ns->full_zones, zone, entry);
279         default:
280             ;
281         }
282     }
283 
284     nvme_set_zone_state(zone, state);
285 
286     switch (state) {
287     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
288         QTAILQ_INSERT_TAIL(&ns->exp_open_zones, zone, entry);
289         break;
290     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
291         QTAILQ_INSERT_TAIL(&ns->imp_open_zones, zone, entry);
292         break;
293     case NVME_ZONE_STATE_CLOSED:
294         QTAILQ_INSERT_TAIL(&ns->closed_zones, zone, entry);
295         break;
296     case NVME_ZONE_STATE_FULL:
297         QTAILQ_INSERT_TAIL(&ns->full_zones, zone, entry);
298     case NVME_ZONE_STATE_READ_ONLY:
299         break;
300     default:
301         zone->d.za = 0;
302     }
303 }
304 
305 static uint16_t nvme_zns_check_resources(NvmeNamespace *ns, uint32_t act,
306                                          uint32_t opn, uint32_t zrwa)
307 {
308     if (ns->params.max_active_zones != 0 &&
309         ns->nr_active_zones + act > ns->params.max_active_zones) {
310         trace_pci_nvme_err_insuff_active_res(ns->params.max_active_zones);
311         return NVME_ZONE_TOO_MANY_ACTIVE | NVME_DNR;
312     }
313 
314     if (ns->params.max_open_zones != 0 &&
315         ns->nr_open_zones + opn > ns->params.max_open_zones) {
316         trace_pci_nvme_err_insuff_open_res(ns->params.max_open_zones);
317         return NVME_ZONE_TOO_MANY_OPEN | NVME_DNR;
318     }
319 
320     if (zrwa > ns->zns.numzrwa) {
321         return NVME_NOZRWA | NVME_DNR;
322     }
323 
324     return NVME_SUCCESS;
325 }
326 
327 /*
328  * Check if we can open a zone without exceeding open/active limits.
329  * AOR stands for "Active and Open Resources" (see TP 4053 section 2.5).
330  */
331 static uint16_t nvme_aor_check(NvmeNamespace *ns, uint32_t act, uint32_t opn)
332 {
333     return nvme_zns_check_resources(ns, act, opn, 0);
334 }
335 
336 static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
337 {
338     hwaddr hi, lo;
339 
340     if (!n->cmb.cmse) {
341         return false;
342     }
343 
344     lo = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
345     hi = lo + int128_get64(n->cmb.mem.size);
346 
347     return addr >= lo && addr < hi;
348 }
349 
350 static inline void *nvme_addr_to_cmb(NvmeCtrl *n, hwaddr addr)
351 {
352     hwaddr base = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
353     return &n->cmb.buf[addr - base];
354 }
355 
356 static bool nvme_addr_is_pmr(NvmeCtrl *n, hwaddr addr)
357 {
358     hwaddr hi;
359 
360     if (!n->pmr.cmse) {
361         return false;
362     }
363 
364     hi = n->pmr.cba + int128_get64(n->pmr.dev->mr.size);
365 
366     return addr >= n->pmr.cba && addr < hi;
367 }
368 
369 static inline void *nvme_addr_to_pmr(NvmeCtrl *n, hwaddr addr)
370 {
371     return memory_region_get_ram_ptr(&n->pmr.dev->mr) + (addr - n->pmr.cba);
372 }
373 
374 static inline bool nvme_addr_is_iomem(NvmeCtrl *n, hwaddr addr)
375 {
376     hwaddr hi, lo;
377 
378     /*
379      * The purpose of this check is to guard against invalid "local" access to
380      * the iomem (i.e. controller registers). Thus, we check against the range
381      * covered by the 'bar0' MemoryRegion since that is currently composed of
382      * two subregions (the NVMe "MBAR" and the MSI-X table/pba). Note, however,
383      * that if the device model is ever changed to allow the CMB to be located
384      * in BAR0 as well, then this must be changed.
385      */
386     lo = n->bar0.addr;
387     hi = lo + int128_get64(n->bar0.size);
388 
389     return addr >= lo && addr < hi;
390 }
391 
392 static int nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
393 {
394     hwaddr hi = addr + size - 1;
395     if (hi < addr) {
396         return 1;
397     }
398 
399     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
400         memcpy(buf, nvme_addr_to_cmb(n, addr), size);
401         return 0;
402     }
403 
404     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
405         memcpy(buf, nvme_addr_to_pmr(n, addr), size);
406         return 0;
407     }
408 
409     return pci_dma_read(&n->parent_obj, addr, buf, size);
410 }
411 
412 static int nvme_addr_write(NvmeCtrl *n, hwaddr addr, const void *buf, int size)
413 {
414     hwaddr hi = addr + size - 1;
415     if (hi < addr) {
416         return 1;
417     }
418 
419     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
420         memcpy(nvme_addr_to_cmb(n, addr), buf, size);
421         return 0;
422     }
423 
424     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
425         memcpy(nvme_addr_to_pmr(n, addr), buf, size);
426         return 0;
427     }
428 
429     return pci_dma_write(&n->parent_obj, addr, buf, size);
430 }
431 
432 static bool nvme_nsid_valid(NvmeCtrl *n, uint32_t nsid)
433 {
434     return nsid &&
435         (nsid == NVME_NSID_BROADCAST || nsid <= NVME_MAX_NAMESPACES);
436 }
437 
438 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
439 {
440     return sqid < n->params.max_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
441 }
442 
443 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
444 {
445     return cqid < n->params.max_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
446 }
447 
448 static void nvme_inc_cq_tail(NvmeCQueue *cq)
449 {
450     cq->tail++;
451     if (cq->tail >= cq->size) {
452         cq->tail = 0;
453         cq->phase = !cq->phase;
454     }
455 }
456 
457 static void nvme_inc_sq_head(NvmeSQueue *sq)
458 {
459     sq->head = (sq->head + 1) % sq->size;
460 }
461 
462 static uint8_t nvme_cq_full(NvmeCQueue *cq)
463 {
464     return (cq->tail + 1) % cq->size == cq->head;
465 }
466 
467 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
468 {
469     return sq->head == sq->tail;
470 }
471 
472 static void nvme_irq_check(NvmeCtrl *n)
473 {
474     uint32_t intms = ldl_le_p(&n->bar.intms);
475 
476     if (msix_enabled(&(n->parent_obj))) {
477         return;
478     }
479     if (~intms & n->irq_status) {
480         pci_irq_assert(&n->parent_obj);
481     } else {
482         pci_irq_deassert(&n->parent_obj);
483     }
484 }
485 
486 static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
487 {
488     if (cq->irq_enabled) {
489         if (msix_enabled(&(n->parent_obj))) {
490             trace_pci_nvme_irq_msix(cq->vector);
491             msix_notify(&(n->parent_obj), cq->vector);
492         } else {
493             trace_pci_nvme_irq_pin();
494             assert(cq->vector < 32);
495             n->irq_status |= 1 << cq->vector;
496             nvme_irq_check(n);
497         }
498     } else {
499         trace_pci_nvme_irq_masked();
500     }
501 }
502 
503 static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
504 {
505     if (cq->irq_enabled) {
506         if (msix_enabled(&(n->parent_obj))) {
507             return;
508         } else {
509             assert(cq->vector < 32);
510             if (!n->cq_pending) {
511                 n->irq_status &= ~(1 << cq->vector);
512             }
513             nvme_irq_check(n);
514         }
515     }
516 }
517 
518 static void nvme_req_clear(NvmeRequest *req)
519 {
520     req->ns = NULL;
521     req->opaque = NULL;
522     req->aiocb = NULL;
523     memset(&req->cqe, 0x0, sizeof(req->cqe));
524     req->status = NVME_SUCCESS;
525 }
526 
527 static inline void nvme_sg_init(NvmeCtrl *n, NvmeSg *sg, bool dma)
528 {
529     if (dma) {
530         pci_dma_sglist_init(&sg->qsg, &n->parent_obj, 0);
531         sg->flags = NVME_SG_DMA;
532     } else {
533         qemu_iovec_init(&sg->iov, 0);
534     }
535 
536     sg->flags |= NVME_SG_ALLOC;
537 }
538 
539 static inline void nvme_sg_unmap(NvmeSg *sg)
540 {
541     if (!(sg->flags & NVME_SG_ALLOC)) {
542         return;
543     }
544 
545     if (sg->flags & NVME_SG_DMA) {
546         qemu_sglist_destroy(&sg->qsg);
547     } else {
548         qemu_iovec_destroy(&sg->iov);
549     }
550 
551     memset(sg, 0x0, sizeof(*sg));
552 }
553 
554 /*
555  * When metadata is transfered as extended LBAs, the DPTR mapped into `sg`
556  * holds both data and metadata. This function splits the data and metadata
557  * into two separate QSG/IOVs.
558  */
559 static void nvme_sg_split(NvmeSg *sg, NvmeNamespace *ns, NvmeSg *data,
560                           NvmeSg *mdata)
561 {
562     NvmeSg *dst = data;
563     uint32_t trans_len, count = ns->lbasz;
564     uint64_t offset = 0;
565     bool dma = sg->flags & NVME_SG_DMA;
566     size_t sge_len;
567     size_t sg_len = dma ? sg->qsg.size : sg->iov.size;
568     int sg_idx = 0;
569 
570     assert(sg->flags & NVME_SG_ALLOC);
571 
572     while (sg_len) {
573         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
574 
575         trans_len = MIN(sg_len, count);
576         trans_len = MIN(trans_len, sge_len - offset);
577 
578         if (dst) {
579             if (dma) {
580                 qemu_sglist_add(&dst->qsg, sg->qsg.sg[sg_idx].base + offset,
581                                 trans_len);
582             } else {
583                 qemu_iovec_add(&dst->iov,
584                                sg->iov.iov[sg_idx].iov_base + offset,
585                                trans_len);
586             }
587         }
588 
589         sg_len -= trans_len;
590         count -= trans_len;
591         offset += trans_len;
592 
593         if (count == 0) {
594             dst = (dst == data) ? mdata : data;
595             count = (dst == data) ? ns->lbasz : ns->lbaf.ms;
596         }
597 
598         if (sge_len == offset) {
599             offset = 0;
600             sg_idx++;
601         }
602     }
603 }
604 
605 static uint16_t nvme_map_addr_cmb(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
606                                   size_t len)
607 {
608     if (!len) {
609         return NVME_SUCCESS;
610     }
611 
612     trace_pci_nvme_map_addr_cmb(addr, len);
613 
614     if (!nvme_addr_is_cmb(n, addr) || !nvme_addr_is_cmb(n, addr + len - 1)) {
615         return NVME_DATA_TRAS_ERROR;
616     }
617 
618     qemu_iovec_add(iov, nvme_addr_to_cmb(n, addr), len);
619 
620     return NVME_SUCCESS;
621 }
622 
623 static uint16_t nvme_map_addr_pmr(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
624                                   size_t len)
625 {
626     if (!len) {
627         return NVME_SUCCESS;
628     }
629 
630     if (!nvme_addr_is_pmr(n, addr) || !nvme_addr_is_pmr(n, addr + len - 1)) {
631         return NVME_DATA_TRAS_ERROR;
632     }
633 
634     qemu_iovec_add(iov, nvme_addr_to_pmr(n, addr), len);
635 
636     return NVME_SUCCESS;
637 }
638 
639 static uint16_t nvme_map_addr(NvmeCtrl *n, NvmeSg *sg, hwaddr addr, size_t len)
640 {
641     bool cmb = false, pmr = false;
642 
643     if (!len) {
644         return NVME_SUCCESS;
645     }
646 
647     trace_pci_nvme_map_addr(addr, len);
648 
649     if (nvme_addr_is_iomem(n, addr)) {
650         return NVME_DATA_TRAS_ERROR;
651     }
652 
653     if (nvme_addr_is_cmb(n, addr)) {
654         cmb = true;
655     } else if (nvme_addr_is_pmr(n, addr)) {
656         pmr = true;
657     }
658 
659     if (cmb || pmr) {
660         if (sg->flags & NVME_SG_DMA) {
661             return NVME_INVALID_USE_OF_CMB | NVME_DNR;
662         }
663 
664         if (sg->iov.niov + 1 > IOV_MAX) {
665             goto max_mappings_exceeded;
666         }
667 
668         if (cmb) {
669             return nvme_map_addr_cmb(n, &sg->iov, addr, len);
670         } else {
671             return nvme_map_addr_pmr(n, &sg->iov, addr, len);
672         }
673     }
674 
675     if (!(sg->flags & NVME_SG_DMA)) {
676         return NVME_INVALID_USE_OF_CMB | NVME_DNR;
677     }
678 
679     if (sg->qsg.nsg + 1 > IOV_MAX) {
680         goto max_mappings_exceeded;
681     }
682 
683     qemu_sglist_add(&sg->qsg, addr, len);
684 
685     return NVME_SUCCESS;
686 
687 max_mappings_exceeded:
688     NVME_GUEST_ERR(pci_nvme_ub_too_many_mappings,
689                    "number of mappings exceed 1024");
690     return NVME_INTERNAL_DEV_ERROR | NVME_DNR;
691 }
692 
693 static inline bool nvme_addr_is_dma(NvmeCtrl *n, hwaddr addr)
694 {
695     return !(nvme_addr_is_cmb(n, addr) || nvme_addr_is_pmr(n, addr));
696 }
697 
698 static uint16_t nvme_map_prp(NvmeCtrl *n, NvmeSg *sg, uint64_t prp1,
699                              uint64_t prp2, uint32_t len)
700 {
701     hwaddr trans_len = n->page_size - (prp1 % n->page_size);
702     trans_len = MIN(len, trans_len);
703     int num_prps = (len >> n->page_bits) + 1;
704     uint16_t status;
705     int ret;
706 
707     trace_pci_nvme_map_prp(trans_len, len, prp1, prp2, num_prps);
708 
709     nvme_sg_init(n, sg, nvme_addr_is_dma(n, prp1));
710 
711     status = nvme_map_addr(n, sg, prp1, trans_len);
712     if (status) {
713         goto unmap;
714     }
715 
716     len -= trans_len;
717     if (len) {
718         if (len > n->page_size) {
719             uint64_t prp_list[n->max_prp_ents];
720             uint32_t nents, prp_trans;
721             int i = 0;
722 
723             /*
724              * The first PRP list entry, pointed to by PRP2 may contain offset.
725              * Hence, we need to calculate the number of entries in based on
726              * that offset.
727              */
728             nents = (n->page_size - (prp2 & (n->page_size - 1))) >> 3;
729             prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
730             ret = nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
731             if (ret) {
732                 trace_pci_nvme_err_addr_read(prp2);
733                 status = NVME_DATA_TRAS_ERROR;
734                 goto unmap;
735             }
736             while (len != 0) {
737                 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
738 
739                 if (i == nents - 1 && len > n->page_size) {
740                     if (unlikely(prp_ent & (n->page_size - 1))) {
741                         trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
742                         status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
743                         goto unmap;
744                     }
745 
746                     i = 0;
747                     nents = (len + n->page_size - 1) >> n->page_bits;
748                     nents = MIN(nents, n->max_prp_ents);
749                     prp_trans = nents * sizeof(uint64_t);
750                     ret = nvme_addr_read(n, prp_ent, (void *)prp_list,
751                                          prp_trans);
752                     if (ret) {
753                         trace_pci_nvme_err_addr_read(prp_ent);
754                         status = NVME_DATA_TRAS_ERROR;
755                         goto unmap;
756                     }
757                     prp_ent = le64_to_cpu(prp_list[i]);
758                 }
759 
760                 if (unlikely(prp_ent & (n->page_size - 1))) {
761                     trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
762                     status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
763                     goto unmap;
764                 }
765 
766                 trans_len = MIN(len, n->page_size);
767                 status = nvme_map_addr(n, sg, prp_ent, trans_len);
768                 if (status) {
769                     goto unmap;
770                 }
771 
772                 len -= trans_len;
773                 i++;
774             }
775         } else {
776             if (unlikely(prp2 & (n->page_size - 1))) {
777                 trace_pci_nvme_err_invalid_prp2_align(prp2);
778                 status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
779                 goto unmap;
780             }
781             status = nvme_map_addr(n, sg, prp2, len);
782             if (status) {
783                 goto unmap;
784             }
785         }
786     }
787 
788     return NVME_SUCCESS;
789 
790 unmap:
791     nvme_sg_unmap(sg);
792     return status;
793 }
794 
795 /*
796  * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
797  * number of bytes mapped in len.
798  */
799 static uint16_t nvme_map_sgl_data(NvmeCtrl *n, NvmeSg *sg,
800                                   NvmeSglDescriptor *segment, uint64_t nsgld,
801                                   size_t *len, NvmeCmd *cmd)
802 {
803     dma_addr_t addr, trans_len;
804     uint32_t dlen;
805     uint16_t status;
806 
807     for (int i = 0; i < nsgld; i++) {
808         uint8_t type = NVME_SGL_TYPE(segment[i].type);
809 
810         switch (type) {
811         case NVME_SGL_DESCR_TYPE_BIT_BUCKET:
812             if (cmd->opcode == NVME_CMD_WRITE) {
813                 continue;
814             }
815         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
816             break;
817         case NVME_SGL_DESCR_TYPE_SEGMENT:
818         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
819             return NVME_INVALID_NUM_SGL_DESCRS | NVME_DNR;
820         default:
821             return NVME_SGL_DESCR_TYPE_INVALID | NVME_DNR;
822         }
823 
824         dlen = le32_to_cpu(segment[i].len);
825 
826         if (!dlen) {
827             continue;
828         }
829 
830         if (*len == 0) {
831             /*
832              * All data has been mapped, but the SGL contains additional
833              * segments and/or descriptors. The controller might accept
834              * ignoring the rest of the SGL.
835              */
836             uint32_t sgls = le32_to_cpu(n->id_ctrl.sgls);
837             if (sgls & NVME_CTRL_SGLS_EXCESS_LENGTH) {
838                 break;
839             }
840 
841             trace_pci_nvme_err_invalid_sgl_excess_length(dlen);
842             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
843         }
844 
845         trans_len = MIN(*len, dlen);
846 
847         if (type == NVME_SGL_DESCR_TYPE_BIT_BUCKET) {
848             goto next;
849         }
850 
851         addr = le64_to_cpu(segment[i].addr);
852 
853         if (UINT64_MAX - addr < dlen) {
854             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
855         }
856 
857         status = nvme_map_addr(n, sg, addr, trans_len);
858         if (status) {
859             return status;
860         }
861 
862 next:
863         *len -= trans_len;
864     }
865 
866     return NVME_SUCCESS;
867 }
868 
869 static uint16_t nvme_map_sgl(NvmeCtrl *n, NvmeSg *sg, NvmeSglDescriptor sgl,
870                              size_t len, NvmeCmd *cmd)
871 {
872     /*
873      * Read the segment in chunks of 256 descriptors (one 4k page) to avoid
874      * dynamically allocating a potentially huge SGL. The spec allows the SGL
875      * to be larger (as in number of bytes required to describe the SGL
876      * descriptors and segment chain) than the command transfer size, so it is
877      * not bounded by MDTS.
878      */
879     const int SEG_CHUNK_SIZE = 256;
880 
881     NvmeSglDescriptor segment[SEG_CHUNK_SIZE], *sgld, *last_sgld;
882     uint64_t nsgld;
883     uint32_t seg_len;
884     uint16_t status;
885     hwaddr addr;
886     int ret;
887 
888     sgld = &sgl;
889     addr = le64_to_cpu(sgl.addr);
890 
891     trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl.type), len);
892 
893     nvme_sg_init(n, sg, nvme_addr_is_dma(n, addr));
894 
895     /*
896      * If the entire transfer can be described with a single data block it can
897      * be mapped directly.
898      */
899     if (NVME_SGL_TYPE(sgl.type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
900         status = nvme_map_sgl_data(n, sg, sgld, 1, &len, cmd);
901         if (status) {
902             goto unmap;
903         }
904 
905         goto out;
906     }
907 
908     for (;;) {
909         switch (NVME_SGL_TYPE(sgld->type)) {
910         case NVME_SGL_DESCR_TYPE_SEGMENT:
911         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
912             break;
913         default:
914             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
915         }
916 
917         seg_len = le32_to_cpu(sgld->len);
918 
919         /* check the length of the (Last) Segment descriptor */
920         if ((!seg_len || seg_len & 0xf) &&
921             (NVME_SGL_TYPE(sgld->type) != NVME_SGL_DESCR_TYPE_BIT_BUCKET)) {
922             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
923         }
924 
925         if (UINT64_MAX - addr < seg_len) {
926             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
927         }
928 
929         nsgld = seg_len / sizeof(NvmeSglDescriptor);
930 
931         while (nsgld > SEG_CHUNK_SIZE) {
932             if (nvme_addr_read(n, addr, segment, sizeof(segment))) {
933                 trace_pci_nvme_err_addr_read(addr);
934                 status = NVME_DATA_TRAS_ERROR;
935                 goto unmap;
936             }
937 
938             status = nvme_map_sgl_data(n, sg, segment, SEG_CHUNK_SIZE,
939                                        &len, cmd);
940             if (status) {
941                 goto unmap;
942             }
943 
944             nsgld -= SEG_CHUNK_SIZE;
945             addr += SEG_CHUNK_SIZE * sizeof(NvmeSglDescriptor);
946         }
947 
948         ret = nvme_addr_read(n, addr, segment, nsgld *
949                              sizeof(NvmeSglDescriptor));
950         if (ret) {
951             trace_pci_nvme_err_addr_read(addr);
952             status = NVME_DATA_TRAS_ERROR;
953             goto unmap;
954         }
955 
956         last_sgld = &segment[nsgld - 1];
957 
958         /*
959          * If the segment ends with a Data Block or Bit Bucket Descriptor Type,
960          * then we are done.
961          */
962         switch (NVME_SGL_TYPE(last_sgld->type)) {
963         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
964         case NVME_SGL_DESCR_TYPE_BIT_BUCKET:
965             status = nvme_map_sgl_data(n, sg, segment, nsgld, &len, cmd);
966             if (status) {
967                 goto unmap;
968             }
969 
970             goto out;
971 
972         default:
973             break;
974         }
975 
976         /*
977          * If the last descriptor was not a Data Block or Bit Bucket, then the
978          * current segment must not be a Last Segment.
979          */
980         if (NVME_SGL_TYPE(sgld->type) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT) {
981             status = NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
982             goto unmap;
983         }
984 
985         sgld = last_sgld;
986         addr = le64_to_cpu(sgld->addr);
987 
988         /*
989          * Do not map the last descriptor; it will be a Segment or Last Segment
990          * descriptor and is handled by the next iteration.
991          */
992         status = nvme_map_sgl_data(n, sg, segment, nsgld - 1, &len, cmd);
993         if (status) {
994             goto unmap;
995         }
996     }
997 
998 out:
999     /* if there is any residual left in len, the SGL was too short */
1000     if (len) {
1001         status = NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1002         goto unmap;
1003     }
1004 
1005     return NVME_SUCCESS;
1006 
1007 unmap:
1008     nvme_sg_unmap(sg);
1009     return status;
1010 }
1011 
1012 uint16_t nvme_map_dptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1013                        NvmeCmd *cmd)
1014 {
1015     uint64_t prp1, prp2;
1016 
1017     switch (NVME_CMD_FLAGS_PSDT(cmd->flags)) {
1018     case NVME_PSDT_PRP:
1019         prp1 = le64_to_cpu(cmd->dptr.prp1);
1020         prp2 = le64_to_cpu(cmd->dptr.prp2);
1021 
1022         return nvme_map_prp(n, sg, prp1, prp2, len);
1023     case NVME_PSDT_SGL_MPTR_CONTIGUOUS:
1024     case NVME_PSDT_SGL_MPTR_SGL:
1025         return nvme_map_sgl(n, sg, cmd->dptr.sgl, len, cmd);
1026     default:
1027         return NVME_INVALID_FIELD;
1028     }
1029 }
1030 
1031 static uint16_t nvme_map_mptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1032                               NvmeCmd *cmd)
1033 {
1034     int psdt = NVME_CMD_FLAGS_PSDT(cmd->flags);
1035     hwaddr mptr = le64_to_cpu(cmd->mptr);
1036     uint16_t status;
1037 
1038     if (psdt == NVME_PSDT_SGL_MPTR_SGL) {
1039         NvmeSglDescriptor sgl;
1040 
1041         if (nvme_addr_read(n, mptr, &sgl, sizeof(sgl))) {
1042             return NVME_DATA_TRAS_ERROR;
1043         }
1044 
1045         status = nvme_map_sgl(n, sg, sgl, len, cmd);
1046         if (status && (status & 0x7ff) == NVME_DATA_SGL_LEN_INVALID) {
1047             status = NVME_MD_SGL_LEN_INVALID | NVME_DNR;
1048         }
1049 
1050         return status;
1051     }
1052 
1053     nvme_sg_init(n, sg, nvme_addr_is_dma(n, mptr));
1054     status = nvme_map_addr(n, sg, mptr, len);
1055     if (status) {
1056         nvme_sg_unmap(sg);
1057     }
1058 
1059     return status;
1060 }
1061 
1062 static uint16_t nvme_map_data(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1063 {
1064     NvmeNamespace *ns = req->ns;
1065     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1066     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1067     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1068     size_t len = nvme_l2b(ns, nlb);
1069     uint16_t status;
1070 
1071     if (nvme_ns_ext(ns) &&
1072         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1073         NvmeSg sg;
1074 
1075         len += nvme_m2b(ns, nlb);
1076 
1077         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1078         if (status) {
1079             return status;
1080         }
1081 
1082         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1083         nvme_sg_split(&sg, ns, &req->sg, NULL);
1084         nvme_sg_unmap(&sg);
1085 
1086         return NVME_SUCCESS;
1087     }
1088 
1089     return nvme_map_dptr(n, &req->sg, len, &req->cmd);
1090 }
1091 
1092 static uint16_t nvme_map_mdata(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1093 {
1094     NvmeNamespace *ns = req->ns;
1095     size_t len = nvme_m2b(ns, nlb);
1096     uint16_t status;
1097 
1098     if (nvme_ns_ext(ns)) {
1099         NvmeSg sg;
1100 
1101         len += nvme_l2b(ns, nlb);
1102 
1103         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1104         if (status) {
1105             return status;
1106         }
1107 
1108         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1109         nvme_sg_split(&sg, ns, NULL, &req->sg);
1110         nvme_sg_unmap(&sg);
1111 
1112         return NVME_SUCCESS;
1113     }
1114 
1115     return nvme_map_mptr(n, &req->sg, len, &req->cmd);
1116 }
1117 
1118 static uint16_t nvme_tx_interleaved(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr,
1119                                     uint32_t len, uint32_t bytes,
1120                                     int32_t skip_bytes, int64_t offset,
1121                                     NvmeTxDirection dir)
1122 {
1123     hwaddr addr;
1124     uint32_t trans_len, count = bytes;
1125     bool dma = sg->flags & NVME_SG_DMA;
1126     int64_t sge_len;
1127     int sg_idx = 0;
1128     int ret;
1129 
1130     assert(sg->flags & NVME_SG_ALLOC);
1131 
1132     while (len) {
1133         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
1134 
1135         if (sge_len - offset < 0) {
1136             offset -= sge_len;
1137             sg_idx++;
1138             continue;
1139         }
1140 
1141         if (sge_len == offset) {
1142             offset = 0;
1143             sg_idx++;
1144             continue;
1145         }
1146 
1147         trans_len = MIN(len, count);
1148         trans_len = MIN(trans_len, sge_len - offset);
1149 
1150         if (dma) {
1151             addr = sg->qsg.sg[sg_idx].base + offset;
1152         } else {
1153             addr = (hwaddr)(uintptr_t)sg->iov.iov[sg_idx].iov_base + offset;
1154         }
1155 
1156         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1157             ret = nvme_addr_read(n, addr, ptr, trans_len);
1158         } else {
1159             ret = nvme_addr_write(n, addr, ptr, trans_len);
1160         }
1161 
1162         if (ret) {
1163             return NVME_DATA_TRAS_ERROR;
1164         }
1165 
1166         ptr += trans_len;
1167         len -= trans_len;
1168         count -= trans_len;
1169         offset += trans_len;
1170 
1171         if (count == 0) {
1172             count = bytes;
1173             offset += skip_bytes;
1174         }
1175     }
1176 
1177     return NVME_SUCCESS;
1178 }
1179 
1180 static uint16_t nvme_tx(NvmeCtrl *n, NvmeSg *sg, void *ptr, uint32_t len,
1181                         NvmeTxDirection dir)
1182 {
1183     assert(sg->flags & NVME_SG_ALLOC);
1184 
1185     if (sg->flags & NVME_SG_DMA) {
1186         const MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
1187         dma_addr_t residual;
1188 
1189         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1190             dma_buf_write(ptr, len, &residual, &sg->qsg, attrs);
1191         } else {
1192             dma_buf_read(ptr, len, &residual, &sg->qsg, attrs);
1193         }
1194 
1195         if (unlikely(residual)) {
1196             trace_pci_nvme_err_invalid_dma();
1197             return NVME_INVALID_FIELD | NVME_DNR;
1198         }
1199     } else {
1200         size_t bytes;
1201 
1202         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1203             bytes = qemu_iovec_to_buf(&sg->iov, 0, ptr, len);
1204         } else {
1205             bytes = qemu_iovec_from_buf(&sg->iov, 0, ptr, len);
1206         }
1207 
1208         if (unlikely(bytes != len)) {
1209             trace_pci_nvme_err_invalid_dma();
1210             return NVME_INVALID_FIELD | NVME_DNR;
1211         }
1212     }
1213 
1214     return NVME_SUCCESS;
1215 }
1216 
1217 static inline uint16_t nvme_c2h(NvmeCtrl *n, void *ptr, uint32_t len,
1218                                 NvmeRequest *req)
1219 {
1220     uint16_t status;
1221 
1222     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1223     if (status) {
1224         return status;
1225     }
1226 
1227     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_FROM_DEVICE);
1228 }
1229 
1230 static inline uint16_t nvme_h2c(NvmeCtrl *n, void *ptr, uint32_t len,
1231                                 NvmeRequest *req)
1232 {
1233     uint16_t status;
1234 
1235     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1236     if (status) {
1237         return status;
1238     }
1239 
1240     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_TO_DEVICE);
1241 }
1242 
1243 uint16_t nvme_bounce_data(NvmeCtrl *n, void *ptr, uint32_t len,
1244                           NvmeTxDirection dir, NvmeRequest *req)
1245 {
1246     NvmeNamespace *ns = req->ns;
1247     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1248     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1249     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1250 
1251     if (nvme_ns_ext(ns) &&
1252         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1253         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbasz,
1254                                    ns->lbaf.ms, 0, dir);
1255     }
1256 
1257     return nvme_tx(n, &req->sg, ptr, len, dir);
1258 }
1259 
1260 uint16_t nvme_bounce_mdata(NvmeCtrl *n, void *ptr, uint32_t len,
1261                            NvmeTxDirection dir, NvmeRequest *req)
1262 {
1263     NvmeNamespace *ns = req->ns;
1264     uint16_t status;
1265 
1266     if (nvme_ns_ext(ns)) {
1267         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbaf.ms,
1268                                    ns->lbasz, ns->lbasz, dir);
1269     }
1270 
1271     nvme_sg_unmap(&req->sg);
1272 
1273     status = nvme_map_mptr(n, &req->sg, len, &req->cmd);
1274     if (status) {
1275         return status;
1276     }
1277 
1278     return nvme_tx(n, &req->sg, ptr, len, dir);
1279 }
1280 
1281 static inline void nvme_blk_read(BlockBackend *blk, int64_t offset,
1282                                  BlockCompletionFunc *cb, NvmeRequest *req)
1283 {
1284     assert(req->sg.flags & NVME_SG_ALLOC);
1285 
1286     if (req->sg.flags & NVME_SG_DMA) {
1287         req->aiocb = dma_blk_read(blk, &req->sg.qsg, offset, BDRV_SECTOR_SIZE,
1288                                   cb, req);
1289     } else {
1290         req->aiocb = blk_aio_preadv(blk, offset, &req->sg.iov, 0, cb, req);
1291     }
1292 }
1293 
1294 static inline void nvme_blk_write(BlockBackend *blk, int64_t offset,
1295                                   BlockCompletionFunc *cb, NvmeRequest *req)
1296 {
1297     assert(req->sg.flags & NVME_SG_ALLOC);
1298 
1299     if (req->sg.flags & NVME_SG_DMA) {
1300         req->aiocb = dma_blk_write(blk, &req->sg.qsg, offset, BDRV_SECTOR_SIZE,
1301                                    cb, req);
1302     } else {
1303         req->aiocb = blk_aio_pwritev(blk, offset, &req->sg.iov, 0, cb, req);
1304     }
1305 }
1306 
1307 static void nvme_post_cqes(void *opaque)
1308 {
1309     NvmeCQueue *cq = opaque;
1310     NvmeCtrl *n = cq->ctrl;
1311     NvmeRequest *req, *next;
1312     bool pending = cq->head != cq->tail;
1313     int ret;
1314 
1315     QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
1316         NvmeSQueue *sq;
1317         hwaddr addr;
1318 
1319         if (nvme_cq_full(cq)) {
1320             break;
1321         }
1322 
1323         sq = req->sq;
1324         req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
1325         req->cqe.sq_id = cpu_to_le16(sq->sqid);
1326         req->cqe.sq_head = cpu_to_le16(sq->head);
1327         addr = cq->dma_addr + cq->tail * n->cqe_size;
1328         ret = pci_dma_write(&n->parent_obj, addr, (void *)&req->cqe,
1329                             sizeof(req->cqe));
1330         if (ret) {
1331             trace_pci_nvme_err_addr_write(addr);
1332             trace_pci_nvme_err_cfs();
1333             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
1334             break;
1335         }
1336         QTAILQ_REMOVE(&cq->req_list, req, entry);
1337         nvme_inc_cq_tail(cq);
1338         nvme_sg_unmap(&req->sg);
1339         QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
1340     }
1341     if (cq->tail != cq->head) {
1342         if (cq->irq_enabled && !pending) {
1343             n->cq_pending++;
1344         }
1345 
1346         nvme_irq_assert(n, cq);
1347     }
1348 }
1349 
1350 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
1351 {
1352     assert(cq->cqid == req->sq->cqid);
1353     trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
1354                                           le32_to_cpu(req->cqe.result),
1355                                           le32_to_cpu(req->cqe.dw1),
1356                                           req->status);
1357 
1358     if (req->status) {
1359         trace_pci_nvme_err_req_status(nvme_cid(req), nvme_nsid(req->ns),
1360                                       req->status, req->cmd.opcode);
1361     }
1362 
1363     QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
1364     QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
1365     timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
1366 }
1367 
1368 static void nvme_process_aers(void *opaque)
1369 {
1370     NvmeCtrl *n = opaque;
1371     NvmeAsyncEvent *event, *next;
1372 
1373     trace_pci_nvme_process_aers(n->aer_queued);
1374 
1375     QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
1376         NvmeRequest *req;
1377         NvmeAerResult *result;
1378 
1379         /* can't post cqe if there is nothing to complete */
1380         if (!n->outstanding_aers) {
1381             trace_pci_nvme_no_outstanding_aers();
1382             break;
1383         }
1384 
1385         /* ignore if masked (cqe posted, but event not cleared) */
1386         if (n->aer_mask & (1 << event->result.event_type)) {
1387             trace_pci_nvme_aer_masked(event->result.event_type, n->aer_mask);
1388             continue;
1389         }
1390 
1391         QTAILQ_REMOVE(&n->aer_queue, event, entry);
1392         n->aer_queued--;
1393 
1394         n->aer_mask |= 1 << event->result.event_type;
1395         n->outstanding_aers--;
1396 
1397         req = n->aer_reqs[n->outstanding_aers];
1398 
1399         result = (NvmeAerResult *) &req->cqe.result;
1400         result->event_type = event->result.event_type;
1401         result->event_info = event->result.event_info;
1402         result->log_page = event->result.log_page;
1403         g_free(event);
1404 
1405         trace_pci_nvme_aer_post_cqe(result->event_type, result->event_info,
1406                                     result->log_page);
1407 
1408         nvme_enqueue_req_completion(&n->admin_cq, req);
1409     }
1410 }
1411 
1412 static void nvme_enqueue_event(NvmeCtrl *n, uint8_t event_type,
1413                                uint8_t event_info, uint8_t log_page)
1414 {
1415     NvmeAsyncEvent *event;
1416 
1417     trace_pci_nvme_enqueue_event(event_type, event_info, log_page);
1418 
1419     if (n->aer_queued == n->params.aer_max_queued) {
1420         trace_pci_nvme_enqueue_event_noqueue(n->aer_queued);
1421         return;
1422     }
1423 
1424     event = g_new(NvmeAsyncEvent, 1);
1425     event->result = (NvmeAerResult) {
1426         .event_type = event_type,
1427         .event_info = event_info,
1428         .log_page   = log_page,
1429     };
1430 
1431     QTAILQ_INSERT_TAIL(&n->aer_queue, event, entry);
1432     n->aer_queued++;
1433 
1434     nvme_process_aers(n);
1435 }
1436 
1437 static void nvme_smart_event(NvmeCtrl *n, uint8_t event)
1438 {
1439     uint8_t aer_info;
1440 
1441     /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
1442     if (!(NVME_AEC_SMART(n->features.async_config) & event)) {
1443         return;
1444     }
1445 
1446     switch (event) {
1447     case NVME_SMART_SPARE:
1448         aer_info = NVME_AER_INFO_SMART_SPARE_THRESH;
1449         break;
1450     case NVME_SMART_TEMPERATURE:
1451         aer_info = NVME_AER_INFO_SMART_TEMP_THRESH;
1452         break;
1453     case NVME_SMART_RELIABILITY:
1454     case NVME_SMART_MEDIA_READ_ONLY:
1455     case NVME_SMART_FAILED_VOLATILE_MEDIA:
1456     case NVME_SMART_PMR_UNRELIABLE:
1457         aer_info = NVME_AER_INFO_SMART_RELIABILITY;
1458         break;
1459     default:
1460         return;
1461     }
1462 
1463     nvme_enqueue_event(n, NVME_AER_TYPE_SMART, aer_info, NVME_LOG_SMART_INFO);
1464 }
1465 
1466 static void nvme_clear_events(NvmeCtrl *n, uint8_t event_type)
1467 {
1468     n->aer_mask &= ~(1 << event_type);
1469     if (!QTAILQ_EMPTY(&n->aer_queue)) {
1470         nvme_process_aers(n);
1471     }
1472 }
1473 
1474 static inline uint16_t nvme_check_mdts(NvmeCtrl *n, size_t len)
1475 {
1476     uint8_t mdts = n->params.mdts;
1477 
1478     if (mdts && len > n->page_size << mdts) {
1479         trace_pci_nvme_err_mdts(len);
1480         return NVME_INVALID_FIELD | NVME_DNR;
1481     }
1482 
1483     return NVME_SUCCESS;
1484 }
1485 
1486 static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
1487                                          uint32_t nlb)
1488 {
1489     uint64_t nsze = le64_to_cpu(ns->id_ns.nsze);
1490 
1491     if (unlikely(UINT64_MAX - slba < nlb || slba + nlb > nsze)) {
1492         trace_pci_nvme_err_invalid_lba_range(slba, nlb, nsze);
1493         return NVME_LBA_RANGE | NVME_DNR;
1494     }
1495 
1496     return NVME_SUCCESS;
1497 }
1498 
1499 static int nvme_block_status_all(NvmeNamespace *ns, uint64_t slba,
1500                                  uint32_t nlb, int flags)
1501 {
1502     BlockDriverState *bs = blk_bs(ns->blkconf.blk);
1503 
1504     int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
1505     int64_t offset = nvme_l2b(ns, slba);
1506     int ret;
1507 
1508     /*
1509      * `pnum` holds the number of bytes after offset that shares the same
1510      * allocation status as the byte at offset. If `pnum` is different from
1511      * `bytes`, we should check the allocation status of the next range and
1512      * continue this until all bytes have been checked.
1513      */
1514     do {
1515         bytes -= pnum;
1516 
1517         ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
1518         if (ret < 0) {
1519             return ret;
1520         }
1521 
1522 
1523         trace_pci_nvme_block_status(offset, bytes, pnum, ret,
1524                                     !!(ret & BDRV_BLOCK_ZERO));
1525 
1526         if (!(ret & flags)) {
1527             return 1;
1528         }
1529 
1530         offset += pnum;
1531     } while (pnum != bytes);
1532 
1533     return 0;
1534 }
1535 
1536 static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
1537                                  uint32_t nlb)
1538 {
1539     int ret;
1540     Error *err = NULL;
1541 
1542     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_DATA);
1543     if (ret) {
1544         if (ret < 0) {
1545             error_setg_errno(&err, -ret, "unable to get block status");
1546             error_report_err(err);
1547 
1548             return NVME_INTERNAL_DEV_ERROR;
1549         }
1550 
1551         return NVME_DULB;
1552     }
1553 
1554     return NVME_SUCCESS;
1555 }
1556 
1557 static void nvme_aio_err(NvmeRequest *req, int ret)
1558 {
1559     uint16_t status = NVME_SUCCESS;
1560     Error *local_err = NULL;
1561 
1562     switch (req->cmd.opcode) {
1563     case NVME_CMD_READ:
1564         status = NVME_UNRECOVERED_READ;
1565         break;
1566     case NVME_CMD_FLUSH:
1567     case NVME_CMD_WRITE:
1568     case NVME_CMD_WRITE_ZEROES:
1569     case NVME_CMD_ZONE_APPEND:
1570         status = NVME_WRITE_FAULT;
1571         break;
1572     default:
1573         status = NVME_INTERNAL_DEV_ERROR;
1574         break;
1575     }
1576 
1577     trace_pci_nvme_err_aio(nvme_cid(req), strerror(-ret), status);
1578 
1579     error_setg_errno(&local_err, -ret, "aio failed");
1580     error_report_err(local_err);
1581 
1582     /*
1583      * Set the command status code to the first encountered error but allow a
1584      * subsequent Internal Device Error to trump it.
1585      */
1586     if (req->status && status != NVME_INTERNAL_DEV_ERROR) {
1587         return;
1588     }
1589 
1590     req->status = status;
1591 }
1592 
1593 static inline uint32_t nvme_zone_idx(NvmeNamespace *ns, uint64_t slba)
1594 {
1595     return ns->zone_size_log2 > 0 ? slba >> ns->zone_size_log2 :
1596                                     slba / ns->zone_size;
1597 }
1598 
1599 static inline NvmeZone *nvme_get_zone_by_slba(NvmeNamespace *ns, uint64_t slba)
1600 {
1601     uint32_t zone_idx = nvme_zone_idx(ns, slba);
1602 
1603     if (zone_idx >= ns->num_zones) {
1604         return NULL;
1605     }
1606 
1607     return &ns->zone_array[zone_idx];
1608 }
1609 
1610 static uint16_t nvme_check_zone_state_for_write(NvmeZone *zone)
1611 {
1612     uint64_t zslba = zone->d.zslba;
1613 
1614     switch (nvme_get_zone_state(zone)) {
1615     case NVME_ZONE_STATE_EMPTY:
1616     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1617     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1618     case NVME_ZONE_STATE_CLOSED:
1619         return NVME_SUCCESS;
1620     case NVME_ZONE_STATE_FULL:
1621         trace_pci_nvme_err_zone_is_full(zslba);
1622         return NVME_ZONE_FULL;
1623     case NVME_ZONE_STATE_OFFLINE:
1624         trace_pci_nvme_err_zone_is_offline(zslba);
1625         return NVME_ZONE_OFFLINE;
1626     case NVME_ZONE_STATE_READ_ONLY:
1627         trace_pci_nvme_err_zone_is_read_only(zslba);
1628         return NVME_ZONE_READ_ONLY;
1629     default:
1630         assert(false);
1631     }
1632 
1633     return NVME_INTERNAL_DEV_ERROR;
1634 }
1635 
1636 static uint16_t nvme_check_zone_write(NvmeNamespace *ns, NvmeZone *zone,
1637                                       uint64_t slba, uint32_t nlb)
1638 {
1639     uint64_t zcap = nvme_zone_wr_boundary(zone);
1640     uint16_t status;
1641 
1642     status = nvme_check_zone_state_for_write(zone);
1643     if (status) {
1644         return status;
1645     }
1646 
1647     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1648         uint64_t ezrwa = zone->w_ptr + 2 * ns->zns.zrwas;
1649 
1650         if (slba < zone->w_ptr || slba + nlb > ezrwa) {
1651             trace_pci_nvme_err_zone_invalid_write(slba, zone->w_ptr);
1652             return NVME_ZONE_INVALID_WRITE;
1653         }
1654     } else {
1655         if (unlikely(slba != zone->w_ptr)) {
1656             trace_pci_nvme_err_write_not_at_wp(slba, zone->d.zslba,
1657                                                zone->w_ptr);
1658             return NVME_ZONE_INVALID_WRITE;
1659         }
1660     }
1661 
1662     if (unlikely((slba + nlb) > zcap)) {
1663         trace_pci_nvme_err_zone_boundary(slba, nlb, zcap);
1664         return NVME_ZONE_BOUNDARY_ERROR;
1665     }
1666 
1667     return NVME_SUCCESS;
1668 }
1669 
1670 static uint16_t nvme_check_zone_state_for_read(NvmeZone *zone)
1671 {
1672     switch (nvme_get_zone_state(zone)) {
1673     case NVME_ZONE_STATE_EMPTY:
1674     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1675     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1676     case NVME_ZONE_STATE_FULL:
1677     case NVME_ZONE_STATE_CLOSED:
1678     case NVME_ZONE_STATE_READ_ONLY:
1679         return NVME_SUCCESS;
1680     case NVME_ZONE_STATE_OFFLINE:
1681         trace_pci_nvme_err_zone_is_offline(zone->d.zslba);
1682         return NVME_ZONE_OFFLINE;
1683     default:
1684         assert(false);
1685     }
1686 
1687     return NVME_INTERNAL_DEV_ERROR;
1688 }
1689 
1690 static uint16_t nvme_check_zone_read(NvmeNamespace *ns, uint64_t slba,
1691                                      uint32_t nlb)
1692 {
1693     NvmeZone *zone;
1694     uint64_t bndry, end;
1695     uint16_t status;
1696 
1697     zone = nvme_get_zone_by_slba(ns, slba);
1698     assert(zone);
1699 
1700     bndry = nvme_zone_rd_boundary(ns, zone);
1701     end = slba + nlb;
1702 
1703     status = nvme_check_zone_state_for_read(zone);
1704     if (status) {
1705         ;
1706     } else if (unlikely(end > bndry)) {
1707         if (!ns->params.cross_zone_read) {
1708             status = NVME_ZONE_BOUNDARY_ERROR;
1709         } else {
1710             /*
1711              * Read across zone boundary - check that all subsequent
1712              * zones that are being read have an appropriate state.
1713              */
1714             do {
1715                 zone++;
1716                 status = nvme_check_zone_state_for_read(zone);
1717                 if (status) {
1718                     break;
1719                 }
1720             } while (end > nvme_zone_rd_boundary(ns, zone));
1721         }
1722     }
1723 
1724     return status;
1725 }
1726 
1727 static uint16_t nvme_zrm_finish(NvmeNamespace *ns, NvmeZone *zone)
1728 {
1729     switch (nvme_get_zone_state(zone)) {
1730     case NVME_ZONE_STATE_FULL:
1731         return NVME_SUCCESS;
1732 
1733     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1734     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1735         nvme_aor_dec_open(ns);
1736         /* fallthrough */
1737     case NVME_ZONE_STATE_CLOSED:
1738         nvme_aor_dec_active(ns);
1739 
1740         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1741             zone->d.za &= ~NVME_ZA_ZRWA_VALID;
1742             if (ns->params.numzrwa) {
1743                 ns->zns.numzrwa++;
1744             }
1745         }
1746 
1747         /* fallthrough */
1748     case NVME_ZONE_STATE_EMPTY:
1749         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_FULL);
1750         return NVME_SUCCESS;
1751 
1752     default:
1753         return NVME_ZONE_INVAL_TRANSITION;
1754     }
1755 }
1756 
1757 static uint16_t nvme_zrm_close(NvmeNamespace *ns, NvmeZone *zone)
1758 {
1759     switch (nvme_get_zone_state(zone)) {
1760     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1761     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1762         nvme_aor_dec_open(ns);
1763         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
1764         /* fall through */
1765     case NVME_ZONE_STATE_CLOSED:
1766         return NVME_SUCCESS;
1767 
1768     default:
1769         return NVME_ZONE_INVAL_TRANSITION;
1770     }
1771 }
1772 
1773 static uint16_t nvme_zrm_reset(NvmeNamespace *ns, NvmeZone *zone)
1774 {
1775     switch (nvme_get_zone_state(zone)) {
1776     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1777     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1778         nvme_aor_dec_open(ns);
1779         /* fallthrough */
1780     case NVME_ZONE_STATE_CLOSED:
1781         nvme_aor_dec_active(ns);
1782 
1783         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1784             if (ns->params.numzrwa) {
1785                 ns->zns.numzrwa++;
1786             }
1787         }
1788 
1789         /* fallthrough */
1790     case NVME_ZONE_STATE_FULL:
1791         zone->w_ptr = zone->d.zslba;
1792         zone->d.wp = zone->w_ptr;
1793         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
1794         /* fallthrough */
1795     case NVME_ZONE_STATE_EMPTY:
1796         return NVME_SUCCESS;
1797 
1798     default:
1799         return NVME_ZONE_INVAL_TRANSITION;
1800     }
1801 }
1802 
1803 static void nvme_zrm_auto_transition_zone(NvmeNamespace *ns)
1804 {
1805     NvmeZone *zone;
1806 
1807     if (ns->params.max_open_zones &&
1808         ns->nr_open_zones == ns->params.max_open_zones) {
1809         zone = QTAILQ_FIRST(&ns->imp_open_zones);
1810         if (zone) {
1811             /*
1812              * Automatically close this implicitly open zone.
1813              */
1814             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
1815             nvme_zrm_close(ns, zone);
1816         }
1817     }
1818 }
1819 
1820 enum {
1821     NVME_ZRM_AUTO = 1 << 0,
1822     NVME_ZRM_ZRWA = 1 << 1,
1823 };
1824 
1825 static uint16_t nvme_zrm_open_flags(NvmeCtrl *n, NvmeNamespace *ns,
1826                                     NvmeZone *zone, int flags)
1827 {
1828     int act = 0;
1829     uint16_t status;
1830 
1831     switch (nvme_get_zone_state(zone)) {
1832     case NVME_ZONE_STATE_EMPTY:
1833         act = 1;
1834 
1835         /* fallthrough */
1836 
1837     case NVME_ZONE_STATE_CLOSED:
1838         if (n->params.auto_transition_zones) {
1839             nvme_zrm_auto_transition_zone(ns);
1840         }
1841         status = nvme_zns_check_resources(ns, act, 1,
1842                                           (flags & NVME_ZRM_ZRWA) ? 1 : 0);
1843         if (status) {
1844             return status;
1845         }
1846 
1847         if (act) {
1848             nvme_aor_inc_active(ns);
1849         }
1850 
1851         nvme_aor_inc_open(ns);
1852 
1853         if (flags & NVME_ZRM_AUTO) {
1854             nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_IMPLICITLY_OPEN);
1855             return NVME_SUCCESS;
1856         }
1857 
1858         /* fallthrough */
1859 
1860     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1861         if (flags & NVME_ZRM_AUTO) {
1862             return NVME_SUCCESS;
1863         }
1864 
1865         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EXPLICITLY_OPEN);
1866 
1867         /* fallthrough */
1868 
1869     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1870         if (flags & NVME_ZRM_ZRWA) {
1871             ns->zns.numzrwa--;
1872 
1873             zone->d.za |= NVME_ZA_ZRWA_VALID;
1874         }
1875 
1876         return NVME_SUCCESS;
1877 
1878     default:
1879         return NVME_ZONE_INVAL_TRANSITION;
1880     }
1881 }
1882 
1883 static inline uint16_t nvme_zrm_auto(NvmeCtrl *n, NvmeNamespace *ns,
1884                                      NvmeZone *zone)
1885 {
1886     return nvme_zrm_open_flags(n, ns, zone, NVME_ZRM_AUTO);
1887 }
1888 
1889 static void nvme_advance_zone_wp(NvmeNamespace *ns, NvmeZone *zone,
1890                                  uint32_t nlb)
1891 {
1892     zone->d.wp += nlb;
1893 
1894     if (zone->d.wp == nvme_zone_wr_boundary(zone)) {
1895         nvme_zrm_finish(ns, zone);
1896     }
1897 }
1898 
1899 static void nvme_zoned_zrwa_implicit_flush(NvmeNamespace *ns, NvmeZone *zone,
1900                                            uint32_t nlbc)
1901 {
1902     uint16_t nzrwafgs = DIV_ROUND_UP(nlbc, ns->zns.zrwafg);
1903 
1904     nlbc = nzrwafgs * ns->zns.zrwafg;
1905 
1906     trace_pci_nvme_zoned_zrwa_implicit_flush(zone->d.zslba, nlbc);
1907 
1908     zone->w_ptr += nlbc;
1909 
1910     nvme_advance_zone_wp(ns, zone, nlbc);
1911 }
1912 
1913 static void nvme_finalize_zoned_write(NvmeNamespace *ns, NvmeRequest *req)
1914 {
1915     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1916     NvmeZone *zone;
1917     uint64_t slba;
1918     uint32_t nlb;
1919 
1920     slba = le64_to_cpu(rw->slba);
1921     nlb = le16_to_cpu(rw->nlb) + 1;
1922     zone = nvme_get_zone_by_slba(ns, slba);
1923     assert(zone);
1924 
1925     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1926         uint64_t ezrwa = zone->w_ptr + ns->zns.zrwas - 1;
1927         uint64_t elba = slba + nlb - 1;
1928 
1929         if (elba > ezrwa) {
1930             nvme_zoned_zrwa_implicit_flush(ns, zone, elba - ezrwa);
1931         }
1932 
1933         return;
1934     }
1935 
1936     nvme_advance_zone_wp(ns, zone, nlb);
1937 }
1938 
1939 static inline bool nvme_is_write(NvmeRequest *req)
1940 {
1941     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1942 
1943     return rw->opcode == NVME_CMD_WRITE ||
1944            rw->opcode == NVME_CMD_ZONE_APPEND ||
1945            rw->opcode == NVME_CMD_WRITE_ZEROES;
1946 }
1947 
1948 static AioContext *nvme_get_aio_context(BlockAIOCB *acb)
1949 {
1950     return qemu_get_aio_context();
1951 }
1952 
1953 static void nvme_misc_cb(void *opaque, int ret)
1954 {
1955     NvmeRequest *req = opaque;
1956 
1957     trace_pci_nvme_misc_cb(nvme_cid(req));
1958 
1959     if (ret) {
1960         nvme_aio_err(req, ret);
1961     }
1962 
1963     nvme_enqueue_req_completion(nvme_cq(req), req);
1964 }
1965 
1966 void nvme_rw_complete_cb(void *opaque, int ret)
1967 {
1968     NvmeRequest *req = opaque;
1969     NvmeNamespace *ns = req->ns;
1970     BlockBackend *blk = ns->blkconf.blk;
1971     BlockAcctCookie *acct = &req->acct;
1972     BlockAcctStats *stats = blk_get_stats(blk);
1973 
1974     trace_pci_nvme_rw_complete_cb(nvme_cid(req), blk_name(blk));
1975 
1976     if (ret) {
1977         block_acct_failed(stats, acct);
1978         nvme_aio_err(req, ret);
1979     } else {
1980         block_acct_done(stats, acct);
1981     }
1982 
1983     if (ns->params.zoned && nvme_is_write(req)) {
1984         nvme_finalize_zoned_write(ns, req);
1985     }
1986 
1987     nvme_enqueue_req_completion(nvme_cq(req), req);
1988 }
1989 
1990 static void nvme_rw_cb(void *opaque, int ret)
1991 {
1992     NvmeRequest *req = opaque;
1993     NvmeNamespace *ns = req->ns;
1994 
1995     BlockBackend *blk = ns->blkconf.blk;
1996 
1997     trace_pci_nvme_rw_cb(nvme_cid(req), blk_name(blk));
1998 
1999     if (ret) {
2000         goto out;
2001     }
2002 
2003     if (ns->lbaf.ms) {
2004         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2005         uint64_t slba = le64_to_cpu(rw->slba);
2006         uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
2007         uint64_t offset = nvme_moff(ns, slba);
2008 
2009         if (req->cmd.opcode == NVME_CMD_WRITE_ZEROES) {
2010             size_t mlen = nvme_m2b(ns, nlb);
2011 
2012             req->aiocb = blk_aio_pwrite_zeroes(blk, offset, mlen,
2013                                                BDRV_REQ_MAY_UNMAP,
2014                                                nvme_rw_complete_cb, req);
2015             return;
2016         }
2017 
2018         if (nvme_ns_ext(ns) || req->cmd.mptr) {
2019             uint16_t status;
2020 
2021             nvme_sg_unmap(&req->sg);
2022             status = nvme_map_mdata(nvme_ctrl(req), nlb, req);
2023             if (status) {
2024                 ret = -EFAULT;
2025                 goto out;
2026             }
2027 
2028             if (req->cmd.opcode == NVME_CMD_READ) {
2029                 return nvme_blk_read(blk, offset, nvme_rw_complete_cb, req);
2030             }
2031 
2032             return nvme_blk_write(blk, offset, nvme_rw_complete_cb, req);
2033         }
2034     }
2035 
2036 out:
2037     nvme_rw_complete_cb(req, ret);
2038 }
2039 
2040 static void nvme_verify_cb(void *opaque, int ret)
2041 {
2042     NvmeBounceContext *ctx = opaque;
2043     NvmeRequest *req = ctx->req;
2044     NvmeNamespace *ns = req->ns;
2045     BlockBackend *blk = ns->blkconf.blk;
2046     BlockAcctCookie *acct = &req->acct;
2047     BlockAcctStats *stats = blk_get_stats(blk);
2048     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2049     uint64_t slba = le64_to_cpu(rw->slba);
2050     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2051     uint16_t apptag = le16_to_cpu(rw->apptag);
2052     uint16_t appmask = le16_to_cpu(rw->appmask);
2053     uint64_t reftag = le32_to_cpu(rw->reftag);
2054     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2055     uint16_t status;
2056 
2057     reftag |= cdw3 << 32;
2058 
2059     trace_pci_nvme_verify_cb(nvme_cid(req), prinfo, apptag, appmask, reftag);
2060 
2061     if (ret) {
2062         block_acct_failed(stats, acct);
2063         nvme_aio_err(req, ret);
2064         goto out;
2065     }
2066 
2067     block_acct_done(stats, acct);
2068 
2069     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2070         status = nvme_dif_mangle_mdata(ns, ctx->mdata.bounce,
2071                                        ctx->mdata.iov.size, slba);
2072         if (status) {
2073             req->status = status;
2074             goto out;
2075         }
2076 
2077         req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2078                                      ctx->mdata.bounce, ctx->mdata.iov.size,
2079                                      prinfo, slba, apptag, appmask, &reftag);
2080     }
2081 
2082 out:
2083     qemu_iovec_destroy(&ctx->data.iov);
2084     g_free(ctx->data.bounce);
2085 
2086     qemu_iovec_destroy(&ctx->mdata.iov);
2087     g_free(ctx->mdata.bounce);
2088 
2089     g_free(ctx);
2090 
2091     nvme_enqueue_req_completion(nvme_cq(req), req);
2092 }
2093 
2094 
2095 static void nvme_verify_mdata_in_cb(void *opaque, int ret)
2096 {
2097     NvmeBounceContext *ctx = opaque;
2098     NvmeRequest *req = ctx->req;
2099     NvmeNamespace *ns = req->ns;
2100     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2101     uint64_t slba = le64_to_cpu(rw->slba);
2102     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2103     size_t mlen = nvme_m2b(ns, nlb);
2104     uint64_t offset = nvme_moff(ns, slba);
2105     BlockBackend *blk = ns->blkconf.blk;
2106 
2107     trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req), blk_name(blk));
2108 
2109     if (ret) {
2110         goto out;
2111     }
2112 
2113     ctx->mdata.bounce = g_malloc(mlen);
2114 
2115     qemu_iovec_reset(&ctx->mdata.iov);
2116     qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2117 
2118     req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2119                                 nvme_verify_cb, ctx);
2120     return;
2121 
2122 out:
2123     nvme_verify_cb(ctx, ret);
2124 }
2125 
2126 struct nvme_compare_ctx {
2127     struct {
2128         QEMUIOVector iov;
2129         uint8_t *bounce;
2130     } data;
2131 
2132     struct {
2133         QEMUIOVector iov;
2134         uint8_t *bounce;
2135     } mdata;
2136 };
2137 
2138 static void nvme_compare_mdata_cb(void *opaque, int ret)
2139 {
2140     NvmeRequest *req = opaque;
2141     NvmeNamespace *ns = req->ns;
2142     NvmeCtrl *n = nvme_ctrl(req);
2143     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2144     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2145     uint16_t apptag = le16_to_cpu(rw->apptag);
2146     uint16_t appmask = le16_to_cpu(rw->appmask);
2147     uint64_t reftag = le32_to_cpu(rw->reftag);
2148     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2149     struct nvme_compare_ctx *ctx = req->opaque;
2150     g_autofree uint8_t *buf = NULL;
2151     BlockBackend *blk = ns->blkconf.blk;
2152     BlockAcctCookie *acct = &req->acct;
2153     BlockAcctStats *stats = blk_get_stats(blk);
2154     uint16_t status = NVME_SUCCESS;
2155 
2156     reftag |= cdw3 << 32;
2157 
2158     trace_pci_nvme_compare_mdata_cb(nvme_cid(req));
2159 
2160     if (ret) {
2161         block_acct_failed(stats, acct);
2162         nvme_aio_err(req, ret);
2163         goto out;
2164     }
2165 
2166     buf = g_malloc(ctx->mdata.iov.size);
2167 
2168     status = nvme_bounce_mdata(n, buf, ctx->mdata.iov.size,
2169                                NVME_TX_DIRECTION_TO_DEVICE, req);
2170     if (status) {
2171         req->status = status;
2172         goto out;
2173     }
2174 
2175     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2176         uint64_t slba = le64_to_cpu(rw->slba);
2177         uint8_t *bufp;
2178         uint8_t *mbufp = ctx->mdata.bounce;
2179         uint8_t *end = mbufp + ctx->mdata.iov.size;
2180         int16_t pil = 0;
2181 
2182         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2183                                 ctx->mdata.bounce, ctx->mdata.iov.size, prinfo,
2184                                 slba, apptag, appmask, &reftag);
2185         if (status) {
2186             req->status = status;
2187             goto out;
2188         }
2189 
2190         /*
2191          * When formatted with protection information, do not compare the DIF
2192          * tuple.
2193          */
2194         if (!(ns->id_ns.dps & NVME_ID_NS_DPS_FIRST_EIGHT)) {
2195             pil = ns->lbaf.ms - nvme_pi_tuple_size(ns);
2196         }
2197 
2198         for (bufp = buf; mbufp < end; bufp += ns->lbaf.ms, mbufp += ns->lbaf.ms) {
2199             if (memcmp(bufp + pil, mbufp + pil, ns->lbaf.ms - pil)) {
2200                 req->status = NVME_CMP_FAILURE;
2201                 goto out;
2202             }
2203         }
2204 
2205         goto out;
2206     }
2207 
2208     if (memcmp(buf, ctx->mdata.bounce, ctx->mdata.iov.size)) {
2209         req->status = NVME_CMP_FAILURE;
2210         goto out;
2211     }
2212 
2213     block_acct_done(stats, acct);
2214 
2215 out:
2216     qemu_iovec_destroy(&ctx->data.iov);
2217     g_free(ctx->data.bounce);
2218 
2219     qemu_iovec_destroy(&ctx->mdata.iov);
2220     g_free(ctx->mdata.bounce);
2221 
2222     g_free(ctx);
2223 
2224     nvme_enqueue_req_completion(nvme_cq(req), req);
2225 }
2226 
2227 static void nvme_compare_data_cb(void *opaque, int ret)
2228 {
2229     NvmeRequest *req = opaque;
2230     NvmeCtrl *n = nvme_ctrl(req);
2231     NvmeNamespace *ns = req->ns;
2232     BlockBackend *blk = ns->blkconf.blk;
2233     BlockAcctCookie *acct = &req->acct;
2234     BlockAcctStats *stats = blk_get_stats(blk);
2235 
2236     struct nvme_compare_ctx *ctx = req->opaque;
2237     g_autofree uint8_t *buf = NULL;
2238     uint16_t status;
2239 
2240     trace_pci_nvme_compare_data_cb(nvme_cid(req));
2241 
2242     if (ret) {
2243         block_acct_failed(stats, acct);
2244         nvme_aio_err(req, ret);
2245         goto out;
2246     }
2247 
2248     buf = g_malloc(ctx->data.iov.size);
2249 
2250     status = nvme_bounce_data(n, buf, ctx->data.iov.size,
2251                               NVME_TX_DIRECTION_TO_DEVICE, req);
2252     if (status) {
2253         req->status = status;
2254         goto out;
2255     }
2256 
2257     if (memcmp(buf, ctx->data.bounce, ctx->data.iov.size)) {
2258         req->status = NVME_CMP_FAILURE;
2259         goto out;
2260     }
2261 
2262     if (ns->lbaf.ms) {
2263         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2264         uint64_t slba = le64_to_cpu(rw->slba);
2265         uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2266         size_t mlen = nvme_m2b(ns, nlb);
2267         uint64_t offset = nvme_moff(ns, slba);
2268 
2269         ctx->mdata.bounce = g_malloc(mlen);
2270 
2271         qemu_iovec_init(&ctx->mdata.iov, 1);
2272         qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2273 
2274         req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2275                                     nvme_compare_mdata_cb, req);
2276         return;
2277     }
2278 
2279     block_acct_done(stats, acct);
2280 
2281 out:
2282     qemu_iovec_destroy(&ctx->data.iov);
2283     g_free(ctx->data.bounce);
2284     g_free(ctx);
2285 
2286     nvme_enqueue_req_completion(nvme_cq(req), req);
2287 }
2288 
2289 typedef struct NvmeDSMAIOCB {
2290     BlockAIOCB common;
2291     BlockAIOCB *aiocb;
2292     NvmeRequest *req;
2293     QEMUBH *bh;
2294     int ret;
2295 
2296     NvmeDsmRange *range;
2297     unsigned int nr;
2298     unsigned int idx;
2299 } NvmeDSMAIOCB;
2300 
2301 static void nvme_dsm_cancel(BlockAIOCB *aiocb)
2302 {
2303     NvmeDSMAIOCB *iocb = container_of(aiocb, NvmeDSMAIOCB, common);
2304 
2305     /* break nvme_dsm_cb loop */
2306     iocb->idx = iocb->nr;
2307     iocb->ret = -ECANCELED;
2308 
2309     if (iocb->aiocb) {
2310         blk_aio_cancel_async(iocb->aiocb);
2311         iocb->aiocb = NULL;
2312     } else {
2313         /*
2314          * We only reach this if nvme_dsm_cancel() has already been called or
2315          * the command ran to completion and nvme_dsm_bh is scheduled to run.
2316          */
2317         assert(iocb->idx == iocb->nr);
2318     }
2319 }
2320 
2321 static const AIOCBInfo nvme_dsm_aiocb_info = {
2322     .aiocb_size   = sizeof(NvmeDSMAIOCB),
2323     .cancel_async = nvme_dsm_cancel,
2324 };
2325 
2326 static void nvme_dsm_bh(void *opaque)
2327 {
2328     NvmeDSMAIOCB *iocb = opaque;
2329 
2330     iocb->common.cb(iocb->common.opaque, iocb->ret);
2331 
2332     qemu_bh_delete(iocb->bh);
2333     iocb->bh = NULL;
2334     qemu_aio_unref(iocb);
2335 }
2336 
2337 static void nvme_dsm_cb(void *opaque, int ret);
2338 
2339 static void nvme_dsm_md_cb(void *opaque, int ret)
2340 {
2341     NvmeDSMAIOCB *iocb = opaque;
2342     NvmeRequest *req = iocb->req;
2343     NvmeNamespace *ns = req->ns;
2344     NvmeDsmRange *range;
2345     uint64_t slba;
2346     uint32_t nlb;
2347 
2348     if (ret < 0) {
2349         iocb->ret = ret;
2350         goto done;
2351     }
2352 
2353     if (!ns->lbaf.ms) {
2354         nvme_dsm_cb(iocb, 0);
2355         return;
2356     }
2357 
2358     range = &iocb->range[iocb->idx - 1];
2359     slba = le64_to_cpu(range->slba);
2360     nlb = le32_to_cpu(range->nlb);
2361 
2362     /*
2363      * Check that all block were discarded (zeroed); otherwise we do not zero
2364      * the metadata.
2365      */
2366 
2367     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_ZERO);
2368     if (ret) {
2369         if (ret < 0) {
2370             iocb->ret = ret;
2371             goto done;
2372         }
2373 
2374         nvme_dsm_cb(iocb, 0);
2375     }
2376 
2377     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_moff(ns, slba),
2378                                         nvme_m2b(ns, nlb), BDRV_REQ_MAY_UNMAP,
2379                                         nvme_dsm_cb, iocb);
2380     return;
2381 
2382 done:
2383     iocb->aiocb = NULL;
2384     qemu_bh_schedule(iocb->bh);
2385 }
2386 
2387 static void nvme_dsm_cb(void *opaque, int ret)
2388 {
2389     NvmeDSMAIOCB *iocb = opaque;
2390     NvmeRequest *req = iocb->req;
2391     NvmeCtrl *n = nvme_ctrl(req);
2392     NvmeNamespace *ns = req->ns;
2393     NvmeDsmRange *range;
2394     uint64_t slba;
2395     uint32_t nlb;
2396 
2397     if (ret < 0) {
2398         iocb->ret = ret;
2399         goto done;
2400     }
2401 
2402 next:
2403     if (iocb->idx == iocb->nr) {
2404         goto done;
2405     }
2406 
2407     range = &iocb->range[iocb->idx++];
2408     slba = le64_to_cpu(range->slba);
2409     nlb = le32_to_cpu(range->nlb);
2410 
2411     trace_pci_nvme_dsm_deallocate(slba, nlb);
2412 
2413     if (nlb > n->dmrsl) {
2414         trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
2415         goto next;
2416     }
2417 
2418     if (nvme_check_bounds(ns, slba, nlb)) {
2419         trace_pci_nvme_err_invalid_lba_range(slba, nlb,
2420                                              ns->id_ns.nsze);
2421         goto next;
2422     }
2423 
2424     iocb->aiocb = blk_aio_pdiscard(ns->blkconf.blk, nvme_l2b(ns, slba),
2425                                    nvme_l2b(ns, nlb),
2426                                    nvme_dsm_md_cb, iocb);
2427     return;
2428 
2429 done:
2430     iocb->aiocb = NULL;
2431     qemu_bh_schedule(iocb->bh);
2432 }
2433 
2434 static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
2435 {
2436     NvmeNamespace *ns = req->ns;
2437     NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
2438     uint32_t attr = le32_to_cpu(dsm->attributes);
2439     uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
2440     uint16_t status = NVME_SUCCESS;
2441 
2442     trace_pci_nvme_dsm(nr, attr);
2443 
2444     if (attr & NVME_DSMGMT_AD) {
2445         NvmeDSMAIOCB *iocb = blk_aio_get(&nvme_dsm_aiocb_info, ns->blkconf.blk,
2446                                          nvme_misc_cb, req);
2447 
2448         iocb->req = req;
2449         iocb->bh = qemu_bh_new(nvme_dsm_bh, iocb);
2450         iocb->ret = 0;
2451         iocb->range = g_new(NvmeDsmRange, nr);
2452         iocb->nr = nr;
2453         iocb->idx = 0;
2454 
2455         status = nvme_h2c(n, (uint8_t *)iocb->range, sizeof(NvmeDsmRange) * nr,
2456                           req);
2457         if (status) {
2458             return status;
2459         }
2460 
2461         req->aiocb = &iocb->common;
2462         nvme_dsm_cb(iocb, 0);
2463 
2464         return NVME_NO_COMPLETE;
2465     }
2466 
2467     return status;
2468 }
2469 
2470 static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
2471 {
2472     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2473     NvmeNamespace *ns = req->ns;
2474     BlockBackend *blk = ns->blkconf.blk;
2475     uint64_t slba = le64_to_cpu(rw->slba);
2476     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2477     size_t len = nvme_l2b(ns, nlb);
2478     int64_t offset = nvme_l2b(ns, slba);
2479     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2480     uint32_t reftag = le32_to_cpu(rw->reftag);
2481     NvmeBounceContext *ctx = NULL;
2482     uint16_t status;
2483 
2484     trace_pci_nvme_verify(nvme_cid(req), nvme_nsid(ns), slba, nlb);
2485 
2486     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2487         status = nvme_check_prinfo(ns, prinfo, slba, reftag);
2488         if (status) {
2489             return status;
2490         }
2491 
2492         if (prinfo & NVME_PRINFO_PRACT) {
2493             return NVME_INVALID_PROT_INFO | NVME_DNR;
2494         }
2495     }
2496 
2497     if (len > n->page_size << n->params.vsl) {
2498         return NVME_INVALID_FIELD | NVME_DNR;
2499     }
2500 
2501     status = nvme_check_bounds(ns, slba, nlb);
2502     if (status) {
2503         return status;
2504     }
2505 
2506     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2507         status = nvme_check_dulbe(ns, slba, nlb);
2508         if (status) {
2509             return status;
2510         }
2511     }
2512 
2513     ctx = g_new0(NvmeBounceContext, 1);
2514     ctx->req = req;
2515 
2516     ctx->data.bounce = g_malloc(len);
2517 
2518     qemu_iovec_init(&ctx->data.iov, 1);
2519     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, len);
2520 
2521     block_acct_start(blk_get_stats(blk), &req->acct, ctx->data.iov.size,
2522                      BLOCK_ACCT_READ);
2523 
2524     req->aiocb = blk_aio_preadv(ns->blkconf.blk, offset, &ctx->data.iov, 0,
2525                                 nvme_verify_mdata_in_cb, ctx);
2526     return NVME_NO_COMPLETE;
2527 }
2528 
2529 typedef struct NvmeCopyAIOCB {
2530     BlockAIOCB common;
2531     BlockAIOCB *aiocb;
2532     NvmeRequest *req;
2533     QEMUBH *bh;
2534     int ret;
2535 
2536     void *ranges;
2537     unsigned int format;
2538     int nr;
2539     int idx;
2540 
2541     uint8_t *bounce;
2542     QEMUIOVector iov;
2543     struct {
2544         BlockAcctCookie read;
2545         BlockAcctCookie write;
2546     } acct;
2547 
2548     uint64_t reftag;
2549     uint64_t slba;
2550 
2551     NvmeZone *zone;
2552 } NvmeCopyAIOCB;
2553 
2554 static void nvme_copy_cancel(BlockAIOCB *aiocb)
2555 {
2556     NvmeCopyAIOCB *iocb = container_of(aiocb, NvmeCopyAIOCB, common);
2557 
2558     iocb->ret = -ECANCELED;
2559 
2560     if (iocb->aiocb) {
2561         blk_aio_cancel_async(iocb->aiocb);
2562         iocb->aiocb = NULL;
2563     }
2564 }
2565 
2566 static const AIOCBInfo nvme_copy_aiocb_info = {
2567     .aiocb_size   = sizeof(NvmeCopyAIOCB),
2568     .cancel_async = nvme_copy_cancel,
2569 };
2570 
2571 static void nvme_copy_bh(void *opaque)
2572 {
2573     NvmeCopyAIOCB *iocb = opaque;
2574     NvmeRequest *req = iocb->req;
2575     NvmeNamespace *ns = req->ns;
2576     BlockAcctStats *stats = blk_get_stats(ns->blkconf.blk);
2577 
2578     if (iocb->idx != iocb->nr) {
2579         req->cqe.result = cpu_to_le32(iocb->idx);
2580     }
2581 
2582     qemu_iovec_destroy(&iocb->iov);
2583     g_free(iocb->bounce);
2584 
2585     qemu_bh_delete(iocb->bh);
2586     iocb->bh = NULL;
2587 
2588     if (iocb->ret < 0) {
2589         block_acct_failed(stats, &iocb->acct.read);
2590         block_acct_failed(stats, &iocb->acct.write);
2591     } else {
2592         block_acct_done(stats, &iocb->acct.read);
2593         block_acct_done(stats, &iocb->acct.write);
2594     }
2595 
2596     iocb->common.cb(iocb->common.opaque, iocb->ret);
2597     qemu_aio_unref(iocb);
2598 }
2599 
2600 static void nvme_copy_cb(void *opaque, int ret);
2601 
2602 static void nvme_copy_source_range_parse_format0(void *ranges, int idx,
2603                                                  uint64_t *slba, uint32_t *nlb,
2604                                                  uint16_t *apptag,
2605                                                  uint16_t *appmask,
2606                                                  uint64_t *reftag)
2607 {
2608     NvmeCopySourceRangeFormat0 *_ranges = ranges;
2609 
2610     if (slba) {
2611         *slba = le64_to_cpu(_ranges[idx].slba);
2612     }
2613 
2614     if (nlb) {
2615         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2616     }
2617 
2618     if (apptag) {
2619         *apptag = le16_to_cpu(_ranges[idx].apptag);
2620     }
2621 
2622     if (appmask) {
2623         *appmask = le16_to_cpu(_ranges[idx].appmask);
2624     }
2625 
2626     if (reftag) {
2627         *reftag = le32_to_cpu(_ranges[idx].reftag);
2628     }
2629 }
2630 
2631 static void nvme_copy_source_range_parse_format1(void *ranges, int idx,
2632                                                  uint64_t *slba, uint32_t *nlb,
2633                                                  uint16_t *apptag,
2634                                                  uint16_t *appmask,
2635                                                  uint64_t *reftag)
2636 {
2637     NvmeCopySourceRangeFormat1 *_ranges = ranges;
2638 
2639     if (slba) {
2640         *slba = le64_to_cpu(_ranges[idx].slba);
2641     }
2642 
2643     if (nlb) {
2644         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2645     }
2646 
2647     if (apptag) {
2648         *apptag = le16_to_cpu(_ranges[idx].apptag);
2649     }
2650 
2651     if (appmask) {
2652         *appmask = le16_to_cpu(_ranges[idx].appmask);
2653     }
2654 
2655     if (reftag) {
2656         *reftag = 0;
2657 
2658         *reftag |= (uint64_t)_ranges[idx].sr[4] << 40;
2659         *reftag |= (uint64_t)_ranges[idx].sr[5] << 32;
2660         *reftag |= (uint64_t)_ranges[idx].sr[6] << 24;
2661         *reftag |= (uint64_t)_ranges[idx].sr[7] << 16;
2662         *reftag |= (uint64_t)_ranges[idx].sr[8] << 8;
2663         *reftag |= (uint64_t)_ranges[idx].sr[9];
2664     }
2665 }
2666 
2667 static void nvme_copy_source_range_parse(void *ranges, int idx, uint8_t format,
2668                                          uint64_t *slba, uint32_t *nlb,
2669                                          uint16_t *apptag, uint16_t *appmask,
2670                                          uint64_t *reftag)
2671 {
2672     switch (format) {
2673     case NVME_COPY_FORMAT_0:
2674         nvme_copy_source_range_parse_format0(ranges, idx, slba, nlb, apptag,
2675                                              appmask, reftag);
2676         break;
2677 
2678     case NVME_COPY_FORMAT_1:
2679         nvme_copy_source_range_parse_format1(ranges, idx, slba, nlb, apptag,
2680                                              appmask, reftag);
2681         break;
2682 
2683     default:
2684         abort();
2685     }
2686 }
2687 
2688 static void nvme_copy_out_completed_cb(void *opaque, int ret)
2689 {
2690     NvmeCopyAIOCB *iocb = opaque;
2691     NvmeRequest *req = iocb->req;
2692     NvmeNamespace *ns = req->ns;
2693     uint32_t nlb;
2694 
2695     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2696                                  &nlb, NULL, NULL, NULL);
2697 
2698     if (ret < 0) {
2699         iocb->ret = ret;
2700         goto out;
2701     } else if (iocb->ret < 0) {
2702         goto out;
2703     }
2704 
2705     if (ns->params.zoned) {
2706         nvme_advance_zone_wp(ns, iocb->zone, nlb);
2707     }
2708 
2709     iocb->idx++;
2710     iocb->slba += nlb;
2711 out:
2712     nvme_copy_cb(iocb, iocb->ret);
2713 }
2714 
2715 static void nvme_copy_out_cb(void *opaque, int ret)
2716 {
2717     NvmeCopyAIOCB *iocb = opaque;
2718     NvmeRequest *req = iocb->req;
2719     NvmeNamespace *ns = req->ns;
2720     uint32_t nlb;
2721     size_t mlen;
2722     uint8_t *mbounce;
2723 
2724     if (ret < 0) {
2725         iocb->ret = ret;
2726         goto out;
2727     } else if (iocb->ret < 0) {
2728         goto out;
2729     }
2730 
2731     if (!ns->lbaf.ms) {
2732         nvme_copy_out_completed_cb(iocb, 0);
2733         return;
2734     }
2735 
2736     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2737                                  &nlb, NULL, NULL, NULL);
2738 
2739     mlen = nvme_m2b(ns, nlb);
2740     mbounce = iocb->bounce + nvme_l2b(ns, nlb);
2741 
2742     qemu_iovec_reset(&iocb->iov);
2743     qemu_iovec_add(&iocb->iov, mbounce, mlen);
2744 
2745     iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_moff(ns, iocb->slba),
2746                                   &iocb->iov, 0, nvme_copy_out_completed_cb,
2747                                   iocb);
2748 
2749     return;
2750 
2751 out:
2752     nvme_copy_cb(iocb, ret);
2753 }
2754 
2755 static void nvme_copy_in_completed_cb(void *opaque, int ret)
2756 {
2757     NvmeCopyAIOCB *iocb = opaque;
2758     NvmeRequest *req = iocb->req;
2759     NvmeNamespace *ns = req->ns;
2760     uint32_t nlb;
2761     uint64_t slba;
2762     uint16_t apptag, appmask;
2763     uint64_t reftag;
2764     size_t len;
2765     uint16_t status;
2766 
2767     if (ret < 0) {
2768         iocb->ret = ret;
2769         goto out;
2770     } else if (iocb->ret < 0) {
2771         goto out;
2772     }
2773 
2774     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
2775                                  &nlb, &apptag, &appmask, &reftag);
2776     len = nvme_l2b(ns, nlb);
2777 
2778     trace_pci_nvme_copy_out(iocb->slba, nlb);
2779 
2780     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2781         NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2782 
2783         uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
2784         uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
2785 
2786         size_t mlen = nvme_m2b(ns, nlb);
2787         uint8_t *mbounce = iocb->bounce + nvme_l2b(ns, nlb);
2788 
2789         status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen, prinfor,
2790                                 slba, apptag, appmask, &reftag);
2791         if (status) {
2792             goto invalid;
2793         }
2794 
2795         apptag = le16_to_cpu(copy->apptag);
2796         appmask = le16_to_cpu(copy->appmask);
2797 
2798         if (prinfow & NVME_PRINFO_PRACT) {
2799             status = nvme_check_prinfo(ns, prinfow, iocb->slba, iocb->reftag);
2800             if (status) {
2801                 goto invalid;
2802             }
2803 
2804             nvme_dif_pract_generate_dif(ns, iocb->bounce, len, mbounce, mlen,
2805                                         apptag, &iocb->reftag);
2806         } else {
2807             status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen,
2808                                     prinfow, iocb->slba, apptag, appmask,
2809                                     &iocb->reftag);
2810             if (status) {
2811                 goto invalid;
2812             }
2813         }
2814     }
2815 
2816     status = nvme_check_bounds(ns, iocb->slba, nlb);
2817     if (status) {
2818         goto invalid;
2819     }
2820 
2821     if (ns->params.zoned) {
2822         status = nvme_check_zone_write(ns, iocb->zone, iocb->slba, nlb);
2823         if (status) {
2824             goto invalid;
2825         }
2826 
2827         if (!(iocb->zone->d.za & NVME_ZA_ZRWA_VALID)) {
2828             iocb->zone->w_ptr += nlb;
2829         }
2830     }
2831 
2832     qemu_iovec_reset(&iocb->iov);
2833     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
2834 
2835     iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, iocb->slba),
2836                                   &iocb->iov, 0, nvme_copy_out_cb, iocb);
2837 
2838     return;
2839 
2840 invalid:
2841     req->status = status;
2842     iocb->aiocb = NULL;
2843     if (iocb->bh) {
2844         qemu_bh_schedule(iocb->bh);
2845     }
2846 
2847     return;
2848 
2849 out:
2850     nvme_copy_cb(iocb, ret);
2851 }
2852 
2853 static void nvme_copy_in_cb(void *opaque, int ret)
2854 {
2855     NvmeCopyAIOCB *iocb = opaque;
2856     NvmeRequest *req = iocb->req;
2857     NvmeNamespace *ns = req->ns;
2858     uint64_t slba;
2859     uint32_t nlb;
2860 
2861     if (ret < 0) {
2862         iocb->ret = ret;
2863         goto out;
2864     } else if (iocb->ret < 0) {
2865         goto out;
2866     }
2867 
2868     if (!ns->lbaf.ms) {
2869         nvme_copy_in_completed_cb(iocb, 0);
2870         return;
2871     }
2872 
2873     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
2874                                  &nlb, NULL, NULL, NULL);
2875 
2876     qemu_iovec_reset(&iocb->iov);
2877     qemu_iovec_add(&iocb->iov, iocb->bounce + nvme_l2b(ns, nlb),
2878                    nvme_m2b(ns, nlb));
2879 
2880     iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_moff(ns, slba),
2881                                  &iocb->iov, 0, nvme_copy_in_completed_cb,
2882                                  iocb);
2883     return;
2884 
2885 out:
2886     nvme_copy_cb(iocb, iocb->ret);
2887 }
2888 
2889 static void nvme_copy_cb(void *opaque, int ret)
2890 {
2891     NvmeCopyAIOCB *iocb = opaque;
2892     NvmeRequest *req = iocb->req;
2893     NvmeNamespace *ns = req->ns;
2894     uint64_t slba;
2895     uint32_t nlb;
2896     size_t len;
2897     uint16_t status;
2898 
2899     if (ret < 0) {
2900         iocb->ret = ret;
2901         goto done;
2902     } else if (iocb->ret < 0) {
2903         goto done;
2904     }
2905 
2906     if (iocb->idx == iocb->nr) {
2907         goto done;
2908     }
2909 
2910     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
2911                                  &nlb, NULL, NULL, NULL);
2912     len = nvme_l2b(ns, nlb);
2913 
2914     trace_pci_nvme_copy_source_range(slba, nlb);
2915 
2916     if (nlb > le16_to_cpu(ns->id_ns.mssrl)) {
2917         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
2918         goto invalid;
2919     }
2920 
2921     status = nvme_check_bounds(ns, slba, nlb);
2922     if (status) {
2923         goto invalid;
2924     }
2925 
2926     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2927         status = nvme_check_dulbe(ns, slba, nlb);
2928         if (status) {
2929             goto invalid;
2930         }
2931     }
2932 
2933     if (ns->params.zoned) {
2934         status = nvme_check_zone_read(ns, slba, nlb);
2935         if (status) {
2936             goto invalid;
2937         }
2938     }
2939 
2940     qemu_iovec_reset(&iocb->iov);
2941     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
2942 
2943     iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_l2b(ns, slba),
2944                                  &iocb->iov, 0, nvme_copy_in_cb, iocb);
2945     return;
2946 
2947 invalid:
2948     req->status = status;
2949 done:
2950     iocb->aiocb = NULL;
2951     if (iocb->bh) {
2952         qemu_bh_schedule(iocb->bh);
2953     }
2954 }
2955 
2956 
2957 static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
2958 {
2959     NvmeNamespace *ns = req->ns;
2960     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2961     NvmeCopyAIOCB *iocb = blk_aio_get(&nvme_copy_aiocb_info, ns->blkconf.blk,
2962                                       nvme_misc_cb, req);
2963     uint16_t nr = copy->nr + 1;
2964     uint8_t format = copy->control[0] & 0xf;
2965     uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
2966     uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
2967     size_t len = sizeof(NvmeCopySourceRangeFormat0);
2968 
2969     uint16_t status;
2970 
2971     trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);
2972 
2973     iocb->ranges = NULL;
2974     iocb->zone = NULL;
2975 
2976     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) &&
2977         ((prinfor & NVME_PRINFO_PRACT) != (prinfow & NVME_PRINFO_PRACT))) {
2978         status = NVME_INVALID_FIELD | NVME_DNR;
2979         goto invalid;
2980     }
2981 
2982     if (!(n->id_ctrl.ocfs & (1 << format))) {
2983         trace_pci_nvme_err_copy_invalid_format(format);
2984         status = NVME_INVALID_FIELD | NVME_DNR;
2985         goto invalid;
2986     }
2987 
2988     if (nr > ns->id_ns.msrc + 1) {
2989         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
2990         goto invalid;
2991     }
2992 
2993     if (ns->pif && format != 0x1) {
2994         status = NVME_INVALID_FORMAT | NVME_DNR;
2995         goto invalid;
2996     }
2997 
2998     if (ns->pif) {
2999         len = sizeof(NvmeCopySourceRangeFormat1);
3000     }
3001 
3002     iocb->format = format;
3003     iocb->ranges = g_malloc_n(nr, len);
3004     status = nvme_h2c(n, (uint8_t *)iocb->ranges, len * nr, req);
3005     if (status) {
3006         goto invalid;
3007     }
3008 
3009     iocb->slba = le64_to_cpu(copy->sdlba);
3010 
3011     if (ns->params.zoned) {
3012         iocb->zone = nvme_get_zone_by_slba(ns, iocb->slba);
3013         if (!iocb->zone) {
3014             status = NVME_LBA_RANGE | NVME_DNR;
3015             goto invalid;
3016         }
3017 
3018         status = nvme_zrm_auto(n, ns, iocb->zone);
3019         if (status) {
3020             goto invalid;
3021         }
3022     }
3023 
3024     iocb->req = req;
3025     iocb->bh = qemu_bh_new(nvme_copy_bh, iocb);
3026     iocb->ret = 0;
3027     iocb->nr = nr;
3028     iocb->idx = 0;
3029     iocb->reftag = le32_to_cpu(copy->reftag);
3030     iocb->reftag |= (uint64_t)le32_to_cpu(copy->cdw3) << 32;
3031     iocb->bounce = g_malloc_n(le16_to_cpu(ns->id_ns.mssrl),
3032                               ns->lbasz + ns->lbaf.ms);
3033 
3034     qemu_iovec_init(&iocb->iov, 1);
3035 
3036     block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.read, 0,
3037                      BLOCK_ACCT_READ);
3038     block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.write, 0,
3039                      BLOCK_ACCT_WRITE);
3040 
3041     req->aiocb = &iocb->common;
3042     nvme_copy_cb(iocb, 0);
3043 
3044     return NVME_NO_COMPLETE;
3045 
3046 invalid:
3047     g_free(iocb->ranges);
3048     qemu_aio_unref(iocb);
3049     return status;
3050 }
3051 
3052 static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req)
3053 {
3054     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3055     NvmeNamespace *ns = req->ns;
3056     BlockBackend *blk = ns->blkconf.blk;
3057     uint64_t slba = le64_to_cpu(rw->slba);
3058     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
3059     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3060     size_t data_len = nvme_l2b(ns, nlb);
3061     size_t len = data_len;
3062     int64_t offset = nvme_l2b(ns, slba);
3063     struct nvme_compare_ctx *ctx = NULL;
3064     uint16_t status;
3065 
3066     trace_pci_nvme_compare(nvme_cid(req), nvme_nsid(ns), slba, nlb);
3067 
3068     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) && (prinfo & NVME_PRINFO_PRACT)) {
3069         return NVME_INVALID_PROT_INFO | NVME_DNR;
3070     }
3071 
3072     if (nvme_ns_ext(ns)) {
3073         len += nvme_m2b(ns, nlb);
3074     }
3075 
3076     status = nvme_check_mdts(n, len);
3077     if (status) {
3078         return status;
3079     }
3080 
3081     status = nvme_check_bounds(ns, slba, nlb);
3082     if (status) {
3083         return status;
3084     }
3085 
3086     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3087         status = nvme_check_dulbe(ns, slba, nlb);
3088         if (status) {
3089             return status;
3090         }
3091     }
3092 
3093     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
3094     if (status) {
3095         return status;
3096     }
3097 
3098     ctx = g_new(struct nvme_compare_ctx, 1);
3099     ctx->data.bounce = g_malloc(data_len);
3100 
3101     req->opaque = ctx;
3102 
3103     qemu_iovec_init(&ctx->data.iov, 1);
3104     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, data_len);
3105 
3106     block_acct_start(blk_get_stats(blk), &req->acct, data_len,
3107                      BLOCK_ACCT_READ);
3108     req->aiocb = blk_aio_preadv(blk, offset, &ctx->data.iov, 0,
3109                                 nvme_compare_data_cb, req);
3110 
3111     return NVME_NO_COMPLETE;
3112 }
3113 
3114 typedef struct NvmeFlushAIOCB {
3115     BlockAIOCB common;
3116     BlockAIOCB *aiocb;
3117     NvmeRequest *req;
3118     QEMUBH *bh;
3119     int ret;
3120 
3121     NvmeNamespace *ns;
3122     uint32_t nsid;
3123     bool broadcast;
3124 } NvmeFlushAIOCB;
3125 
3126 static void nvme_flush_cancel(BlockAIOCB *acb)
3127 {
3128     NvmeFlushAIOCB *iocb = container_of(acb, NvmeFlushAIOCB, common);
3129 
3130     iocb->ret = -ECANCELED;
3131 
3132     if (iocb->aiocb) {
3133         blk_aio_cancel_async(iocb->aiocb);
3134     }
3135 }
3136 
3137 static const AIOCBInfo nvme_flush_aiocb_info = {
3138     .aiocb_size = sizeof(NvmeFlushAIOCB),
3139     .cancel_async = nvme_flush_cancel,
3140     .get_aio_context = nvme_get_aio_context,
3141 };
3142 
3143 static void nvme_flush_ns_cb(void *opaque, int ret)
3144 {
3145     NvmeFlushAIOCB *iocb = opaque;
3146     NvmeNamespace *ns = iocb->ns;
3147 
3148     if (ret < 0) {
3149         iocb->ret = ret;
3150         goto out;
3151     } else if (iocb->ret < 0) {
3152         goto out;
3153     }
3154 
3155     if (ns) {
3156         trace_pci_nvme_flush_ns(iocb->nsid);
3157 
3158         iocb->ns = NULL;
3159         iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);
3160         return;
3161     }
3162 
3163 out:
3164     iocb->aiocb = NULL;
3165     qemu_bh_schedule(iocb->bh);
3166 }
3167 
3168 static void nvme_flush_bh(void *opaque)
3169 {
3170     NvmeFlushAIOCB *iocb = opaque;
3171     NvmeRequest *req = iocb->req;
3172     NvmeCtrl *n = nvme_ctrl(req);
3173     int i;
3174 
3175     if (iocb->ret < 0) {
3176         goto done;
3177     }
3178 
3179     if (iocb->broadcast) {
3180         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
3181             iocb->ns = nvme_ns(n, i);
3182             if (iocb->ns) {
3183                 iocb->nsid = i;
3184                 break;
3185             }
3186         }
3187     }
3188 
3189     if (!iocb->ns) {
3190         goto done;
3191     }
3192 
3193     nvme_flush_ns_cb(iocb, 0);
3194     return;
3195 
3196 done:
3197     qemu_bh_delete(iocb->bh);
3198     iocb->bh = NULL;
3199 
3200     iocb->common.cb(iocb->common.opaque, iocb->ret);
3201 
3202     qemu_aio_unref(iocb);
3203 
3204     return;
3205 }
3206 
3207 static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req)
3208 {
3209     NvmeFlushAIOCB *iocb;
3210     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
3211     uint16_t status;
3212 
3213     iocb = qemu_aio_get(&nvme_flush_aiocb_info, NULL, nvme_misc_cb, req);
3214 
3215     iocb->req = req;
3216     iocb->bh = qemu_bh_new(nvme_flush_bh, iocb);
3217     iocb->ret = 0;
3218     iocb->ns = NULL;
3219     iocb->nsid = 0;
3220     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
3221 
3222     if (!iocb->broadcast) {
3223         if (!nvme_nsid_valid(n, nsid)) {
3224             status = NVME_INVALID_NSID | NVME_DNR;
3225             goto out;
3226         }
3227 
3228         iocb->ns = nvme_ns(n, nsid);
3229         if (!iocb->ns) {
3230             status = NVME_INVALID_FIELD | NVME_DNR;
3231             goto out;
3232         }
3233 
3234         iocb->nsid = nsid;
3235     }
3236 
3237     req->aiocb = &iocb->common;
3238     qemu_bh_schedule(iocb->bh);
3239 
3240     return NVME_NO_COMPLETE;
3241 
3242 out:
3243     qemu_bh_delete(iocb->bh);
3244     iocb->bh = NULL;
3245     qemu_aio_unref(iocb);
3246 
3247     return status;
3248 }
3249 
3250 static uint16_t nvme_read(NvmeCtrl *n, NvmeRequest *req)
3251 {
3252     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3253     NvmeNamespace *ns = req->ns;
3254     uint64_t slba = le64_to_cpu(rw->slba);
3255     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3256     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3257     uint64_t data_size = nvme_l2b(ns, nlb);
3258     uint64_t mapped_size = data_size;
3259     uint64_t data_offset;
3260     BlockBackend *blk = ns->blkconf.blk;
3261     uint16_t status;
3262 
3263     if (nvme_ns_ext(ns)) {
3264         mapped_size += nvme_m2b(ns, nlb);
3265 
3266         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3267             bool pract = prinfo & NVME_PRINFO_PRACT;
3268 
3269             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3270                 mapped_size = data_size;
3271             }
3272         }
3273     }
3274 
3275     trace_pci_nvme_read(nvme_cid(req), nvme_nsid(ns), nlb, mapped_size, slba);
3276 
3277     status = nvme_check_mdts(n, mapped_size);
3278     if (status) {
3279         goto invalid;
3280     }
3281 
3282     status = nvme_check_bounds(ns, slba, nlb);
3283     if (status) {
3284         goto invalid;
3285     }
3286 
3287     if (ns->params.zoned) {
3288         status = nvme_check_zone_read(ns, slba, nlb);
3289         if (status) {
3290             trace_pci_nvme_err_zone_read_not_ok(slba, nlb, status);
3291             goto invalid;
3292         }
3293     }
3294 
3295     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3296         status = nvme_check_dulbe(ns, slba, nlb);
3297         if (status) {
3298             goto invalid;
3299         }
3300     }
3301 
3302     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3303         return nvme_dif_rw(n, req);
3304     }
3305 
3306     status = nvme_map_data(n, nlb, req);
3307     if (status) {
3308         goto invalid;
3309     }
3310 
3311     data_offset = nvme_l2b(ns, slba);
3312 
3313     block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3314                      BLOCK_ACCT_READ);
3315     nvme_blk_read(blk, data_offset, nvme_rw_cb, req);
3316     return NVME_NO_COMPLETE;
3317 
3318 invalid:
3319     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_READ);
3320     return status | NVME_DNR;
3321 }
3322 
3323 static uint16_t nvme_do_write(NvmeCtrl *n, NvmeRequest *req, bool append,
3324                               bool wrz)
3325 {
3326     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3327     NvmeNamespace *ns = req->ns;
3328     uint64_t slba = le64_to_cpu(rw->slba);
3329     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3330     uint16_t ctrl = le16_to_cpu(rw->control);
3331     uint8_t prinfo = NVME_RW_PRINFO(ctrl);
3332     uint64_t data_size = nvme_l2b(ns, nlb);
3333     uint64_t mapped_size = data_size;
3334     uint64_t data_offset;
3335     NvmeZone *zone;
3336     NvmeZonedResult *res = (NvmeZonedResult *)&req->cqe;
3337     BlockBackend *blk = ns->blkconf.blk;
3338     uint16_t status;
3339 
3340     if (nvme_ns_ext(ns)) {
3341         mapped_size += nvme_m2b(ns, nlb);
3342 
3343         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3344             bool pract = prinfo & NVME_PRINFO_PRACT;
3345 
3346             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3347                 mapped_size -= nvme_m2b(ns, nlb);
3348             }
3349         }
3350     }
3351 
3352     trace_pci_nvme_write(nvme_cid(req), nvme_io_opc_str(rw->opcode),
3353                          nvme_nsid(ns), nlb, mapped_size, slba);
3354 
3355     if (!wrz) {
3356         status = nvme_check_mdts(n, mapped_size);
3357         if (status) {
3358             goto invalid;
3359         }
3360     }
3361 
3362     status = nvme_check_bounds(ns, slba, nlb);
3363     if (status) {
3364         goto invalid;
3365     }
3366 
3367     if (ns->params.zoned) {
3368         zone = nvme_get_zone_by_slba(ns, slba);
3369         assert(zone);
3370 
3371         if (append) {
3372             bool piremap = !!(ctrl & NVME_RW_PIREMAP);
3373 
3374             if (unlikely(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3375                 return NVME_INVALID_ZONE_OP | NVME_DNR;
3376             }
3377 
3378             if (unlikely(slba != zone->d.zslba)) {
3379                 trace_pci_nvme_err_append_not_at_start(slba, zone->d.zslba);
3380                 status = NVME_INVALID_FIELD;
3381                 goto invalid;
3382             }
3383 
3384             if (n->params.zasl &&
3385                 data_size > (uint64_t)n->page_size << n->params.zasl) {
3386                 trace_pci_nvme_err_zasl(data_size);
3387                 return NVME_INVALID_FIELD | NVME_DNR;
3388             }
3389 
3390             slba = zone->w_ptr;
3391             rw->slba = cpu_to_le64(slba);
3392             res->slba = cpu_to_le64(slba);
3393 
3394             switch (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3395             case NVME_ID_NS_DPS_TYPE_1:
3396                 if (!piremap) {
3397                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3398                 }
3399 
3400                 /* fallthrough */
3401 
3402             case NVME_ID_NS_DPS_TYPE_2:
3403                 if (piremap) {
3404                     uint32_t reftag = le32_to_cpu(rw->reftag);
3405                     rw->reftag = cpu_to_le32(reftag + (slba - zone->d.zslba));
3406                 }
3407 
3408                 break;
3409 
3410             case NVME_ID_NS_DPS_TYPE_3:
3411                 if (piremap) {
3412                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3413                 }
3414 
3415                 break;
3416             }
3417         }
3418 
3419         status = nvme_check_zone_write(ns, zone, slba, nlb);
3420         if (status) {
3421             goto invalid;
3422         }
3423 
3424         status = nvme_zrm_auto(n, ns, zone);
3425         if (status) {
3426             goto invalid;
3427         }
3428 
3429         if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3430             zone->w_ptr += nlb;
3431         }
3432     }
3433 
3434     data_offset = nvme_l2b(ns, slba);
3435 
3436     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3437         return nvme_dif_rw(n, req);
3438     }
3439 
3440     if (!wrz) {
3441         status = nvme_map_data(n, nlb, req);
3442         if (status) {
3443             goto invalid;
3444         }
3445 
3446         block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3447                          BLOCK_ACCT_WRITE);
3448         nvme_blk_write(blk, data_offset, nvme_rw_cb, req);
3449     } else {
3450         req->aiocb = blk_aio_pwrite_zeroes(blk, data_offset, data_size,
3451                                            BDRV_REQ_MAY_UNMAP, nvme_rw_cb,
3452                                            req);
3453     }
3454 
3455     return NVME_NO_COMPLETE;
3456 
3457 invalid:
3458     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_WRITE);
3459     return status | NVME_DNR;
3460 }
3461 
3462 static inline uint16_t nvme_write(NvmeCtrl *n, NvmeRequest *req)
3463 {
3464     return nvme_do_write(n, req, false, false);
3465 }
3466 
3467 static inline uint16_t nvme_write_zeroes(NvmeCtrl *n, NvmeRequest *req)
3468 {
3469     return nvme_do_write(n, req, false, true);
3470 }
3471 
3472 static inline uint16_t nvme_zone_append(NvmeCtrl *n, NvmeRequest *req)
3473 {
3474     return nvme_do_write(n, req, true, false);
3475 }
3476 
3477 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace *ns, NvmeCmd *c,
3478                                             uint64_t *slba, uint32_t *zone_idx)
3479 {
3480     uint32_t dw10 = le32_to_cpu(c->cdw10);
3481     uint32_t dw11 = le32_to_cpu(c->cdw11);
3482 
3483     if (!ns->params.zoned) {
3484         trace_pci_nvme_err_invalid_opc(c->opcode);
3485         return NVME_INVALID_OPCODE | NVME_DNR;
3486     }
3487 
3488     *slba = ((uint64_t)dw11) << 32 | dw10;
3489     if (unlikely(*slba >= ns->id_ns.nsze)) {
3490         trace_pci_nvme_err_invalid_lba_range(*slba, 0, ns->id_ns.nsze);
3491         *slba = 0;
3492         return NVME_LBA_RANGE | NVME_DNR;
3493     }
3494 
3495     *zone_idx = nvme_zone_idx(ns, *slba);
3496     assert(*zone_idx < ns->num_zones);
3497 
3498     return NVME_SUCCESS;
3499 }
3500 
3501 typedef uint16_t (*op_handler_t)(NvmeNamespace *, NvmeZone *, NvmeZoneState,
3502                                  NvmeRequest *);
3503 
3504 enum NvmeZoneProcessingMask {
3505     NVME_PROC_CURRENT_ZONE    = 0,
3506     NVME_PROC_OPENED_ZONES    = 1 << 0,
3507     NVME_PROC_CLOSED_ZONES    = 1 << 1,
3508     NVME_PROC_READ_ONLY_ZONES = 1 << 2,
3509     NVME_PROC_FULL_ZONES      = 1 << 3,
3510 };
3511 
3512 static uint16_t nvme_open_zone(NvmeNamespace *ns, NvmeZone *zone,
3513                                NvmeZoneState state, NvmeRequest *req)
3514 {
3515     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
3516     int flags = 0;
3517 
3518     if (cmd->zsflags & NVME_ZSFLAG_ZRWA_ALLOC) {
3519         uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3520 
3521         if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3522             return NVME_INVALID_ZONE_OP | NVME_DNR;
3523         }
3524 
3525         if (zone->w_ptr % ns->zns.zrwafg) {
3526             return NVME_NOZRWA | NVME_DNR;
3527         }
3528 
3529         flags = NVME_ZRM_ZRWA;
3530     }
3531 
3532     return nvme_zrm_open_flags(nvme_ctrl(req), ns, zone, flags);
3533 }
3534 
3535 static uint16_t nvme_close_zone(NvmeNamespace *ns, NvmeZone *zone,
3536                                 NvmeZoneState state, NvmeRequest *req)
3537 {
3538     return nvme_zrm_close(ns, zone);
3539 }
3540 
3541 static uint16_t nvme_finish_zone(NvmeNamespace *ns, NvmeZone *zone,
3542                                  NvmeZoneState state, NvmeRequest *req)
3543 {
3544     return nvme_zrm_finish(ns, zone);
3545 }
3546 
3547 static uint16_t nvme_offline_zone(NvmeNamespace *ns, NvmeZone *zone,
3548                                   NvmeZoneState state, NvmeRequest *req)
3549 {
3550     switch (state) {
3551     case NVME_ZONE_STATE_READ_ONLY:
3552         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_OFFLINE);
3553         /* fall through */
3554     case NVME_ZONE_STATE_OFFLINE:
3555         return NVME_SUCCESS;
3556     default:
3557         return NVME_ZONE_INVAL_TRANSITION;
3558     }
3559 }
3560 
3561 static uint16_t nvme_set_zd_ext(NvmeNamespace *ns, NvmeZone *zone)
3562 {
3563     uint16_t status;
3564     uint8_t state = nvme_get_zone_state(zone);
3565 
3566     if (state == NVME_ZONE_STATE_EMPTY) {
3567         status = nvme_aor_check(ns, 1, 0);
3568         if (status) {
3569             return status;
3570         }
3571         nvme_aor_inc_active(ns);
3572         zone->d.za |= NVME_ZA_ZD_EXT_VALID;
3573         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
3574         return NVME_SUCCESS;
3575     }
3576 
3577     return NVME_ZONE_INVAL_TRANSITION;
3578 }
3579 
3580 static uint16_t nvme_bulk_proc_zone(NvmeNamespace *ns, NvmeZone *zone,
3581                                     enum NvmeZoneProcessingMask proc_mask,
3582                                     op_handler_t op_hndlr, NvmeRequest *req)
3583 {
3584     uint16_t status = NVME_SUCCESS;
3585     NvmeZoneState zs = nvme_get_zone_state(zone);
3586     bool proc_zone;
3587 
3588     switch (zs) {
3589     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3590     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3591         proc_zone = proc_mask & NVME_PROC_OPENED_ZONES;
3592         break;
3593     case NVME_ZONE_STATE_CLOSED:
3594         proc_zone = proc_mask & NVME_PROC_CLOSED_ZONES;
3595         break;
3596     case NVME_ZONE_STATE_READ_ONLY:
3597         proc_zone = proc_mask & NVME_PROC_READ_ONLY_ZONES;
3598         break;
3599     case NVME_ZONE_STATE_FULL:
3600         proc_zone = proc_mask & NVME_PROC_FULL_ZONES;
3601         break;
3602     default:
3603         proc_zone = false;
3604     }
3605 
3606     if (proc_zone) {
3607         status = op_hndlr(ns, zone, zs, req);
3608     }
3609 
3610     return status;
3611 }
3612 
3613 static uint16_t nvme_do_zone_op(NvmeNamespace *ns, NvmeZone *zone,
3614                                 enum NvmeZoneProcessingMask proc_mask,
3615                                 op_handler_t op_hndlr, NvmeRequest *req)
3616 {
3617     NvmeZone *next;
3618     uint16_t status = NVME_SUCCESS;
3619     int i;
3620 
3621     if (!proc_mask) {
3622         status = op_hndlr(ns, zone, nvme_get_zone_state(zone), req);
3623     } else {
3624         if (proc_mask & NVME_PROC_CLOSED_ZONES) {
3625             QTAILQ_FOREACH_SAFE(zone, &ns->closed_zones, entry, next) {
3626                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3627                                              req);
3628                 if (status && status != NVME_NO_COMPLETE) {
3629                     goto out;
3630                 }
3631             }
3632         }
3633         if (proc_mask & NVME_PROC_OPENED_ZONES) {
3634             QTAILQ_FOREACH_SAFE(zone, &ns->imp_open_zones, entry, next) {
3635                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3636                                              req);
3637                 if (status && status != NVME_NO_COMPLETE) {
3638                     goto out;
3639                 }
3640             }
3641 
3642             QTAILQ_FOREACH_SAFE(zone, &ns->exp_open_zones, entry, next) {
3643                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3644                                              req);
3645                 if (status && status != NVME_NO_COMPLETE) {
3646                     goto out;
3647                 }
3648             }
3649         }
3650         if (proc_mask & NVME_PROC_FULL_ZONES) {
3651             QTAILQ_FOREACH_SAFE(zone, &ns->full_zones, entry, next) {
3652                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3653                                              req);
3654                 if (status && status != NVME_NO_COMPLETE) {
3655                     goto out;
3656                 }
3657             }
3658         }
3659 
3660         if (proc_mask & NVME_PROC_READ_ONLY_ZONES) {
3661             for (i = 0; i < ns->num_zones; i++, zone++) {
3662                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3663                                              req);
3664                 if (status && status != NVME_NO_COMPLETE) {
3665                     goto out;
3666                 }
3667             }
3668         }
3669     }
3670 
3671 out:
3672     return status;
3673 }
3674 
3675 typedef struct NvmeZoneResetAIOCB {
3676     BlockAIOCB common;
3677     BlockAIOCB *aiocb;
3678     NvmeRequest *req;
3679     QEMUBH *bh;
3680     int ret;
3681 
3682     bool all;
3683     int idx;
3684     NvmeZone *zone;
3685 } NvmeZoneResetAIOCB;
3686 
3687 static void nvme_zone_reset_cancel(BlockAIOCB *aiocb)
3688 {
3689     NvmeZoneResetAIOCB *iocb = container_of(aiocb, NvmeZoneResetAIOCB, common);
3690     NvmeRequest *req = iocb->req;
3691     NvmeNamespace *ns = req->ns;
3692 
3693     iocb->idx = ns->num_zones;
3694 
3695     iocb->ret = -ECANCELED;
3696 
3697     if (iocb->aiocb) {
3698         blk_aio_cancel_async(iocb->aiocb);
3699         iocb->aiocb = NULL;
3700     }
3701 }
3702 
3703 static const AIOCBInfo nvme_zone_reset_aiocb_info = {
3704     .aiocb_size = sizeof(NvmeZoneResetAIOCB),
3705     .cancel_async = nvme_zone_reset_cancel,
3706 };
3707 
3708 static void nvme_zone_reset_bh(void *opaque)
3709 {
3710     NvmeZoneResetAIOCB *iocb = opaque;
3711 
3712     iocb->common.cb(iocb->common.opaque, iocb->ret);
3713 
3714     qemu_bh_delete(iocb->bh);
3715     iocb->bh = NULL;
3716     qemu_aio_unref(iocb);
3717 }
3718 
3719 static void nvme_zone_reset_cb(void *opaque, int ret);
3720 
3721 static void nvme_zone_reset_epilogue_cb(void *opaque, int ret)
3722 {
3723     NvmeZoneResetAIOCB *iocb = opaque;
3724     NvmeRequest *req = iocb->req;
3725     NvmeNamespace *ns = req->ns;
3726     int64_t moff;
3727     int count;
3728 
3729     if (ret < 0) {
3730         nvme_zone_reset_cb(iocb, ret);
3731         return;
3732     }
3733 
3734     if (!ns->lbaf.ms) {
3735         nvme_zone_reset_cb(iocb, 0);
3736         return;
3737     }
3738 
3739     moff = nvme_moff(ns, iocb->zone->d.zslba);
3740     count = nvme_m2b(ns, ns->zone_size);
3741 
3742     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, moff, count,
3743                                         BDRV_REQ_MAY_UNMAP,
3744                                         nvme_zone_reset_cb, iocb);
3745     return;
3746 }
3747 
3748 static void nvme_zone_reset_cb(void *opaque, int ret)
3749 {
3750     NvmeZoneResetAIOCB *iocb = opaque;
3751     NvmeRequest *req = iocb->req;
3752     NvmeNamespace *ns = req->ns;
3753 
3754     if (ret < 0) {
3755         iocb->ret = ret;
3756         goto done;
3757     }
3758 
3759     if (iocb->zone) {
3760         nvme_zrm_reset(ns, iocb->zone);
3761 
3762         if (!iocb->all) {
3763             goto done;
3764         }
3765     }
3766 
3767     while (iocb->idx < ns->num_zones) {
3768         NvmeZone *zone = &ns->zone_array[iocb->idx++];
3769 
3770         switch (nvme_get_zone_state(zone)) {
3771         case NVME_ZONE_STATE_EMPTY:
3772             if (!iocb->all) {
3773                 goto done;
3774             }
3775 
3776             continue;
3777 
3778         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3779         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3780         case NVME_ZONE_STATE_CLOSED:
3781         case NVME_ZONE_STATE_FULL:
3782             iocb->zone = zone;
3783             break;
3784 
3785         default:
3786             continue;
3787         }
3788 
3789         trace_pci_nvme_zns_zone_reset(zone->d.zslba);
3790 
3791         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk,
3792                                             nvme_l2b(ns, zone->d.zslba),
3793                                             nvme_l2b(ns, ns->zone_size),
3794                                             BDRV_REQ_MAY_UNMAP,
3795                                             nvme_zone_reset_epilogue_cb,
3796                                             iocb);
3797         return;
3798     }
3799 
3800 done:
3801     iocb->aiocb = NULL;
3802     if (iocb->bh) {
3803         qemu_bh_schedule(iocb->bh);
3804     }
3805 }
3806 
3807 static uint16_t nvme_zone_mgmt_send_zrwa_flush(NvmeCtrl *n, NvmeZone *zone,
3808                                                uint64_t elba, NvmeRequest *req)
3809 {
3810     NvmeNamespace *ns = req->ns;
3811     uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3812     uint64_t wp = zone->d.wp;
3813     uint32_t nlb = elba - wp + 1;
3814     uint16_t status;
3815 
3816 
3817     if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3818         return NVME_INVALID_ZONE_OP | NVME_DNR;
3819     }
3820 
3821     if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3822         return NVME_INVALID_FIELD | NVME_DNR;
3823     }
3824 
3825     if (elba < wp || elba > wp + ns->zns.zrwas) {
3826         return NVME_ZONE_BOUNDARY_ERROR | NVME_DNR;
3827     }
3828 
3829     if (nlb % ns->zns.zrwafg) {
3830         return NVME_INVALID_FIELD | NVME_DNR;
3831     }
3832 
3833     status = nvme_zrm_auto(n, ns, zone);
3834     if (status) {
3835         return status;
3836     }
3837 
3838     zone->w_ptr += nlb;
3839 
3840     nvme_advance_zone_wp(ns, zone, nlb);
3841 
3842     return NVME_SUCCESS;
3843 }
3844 
3845 static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
3846 {
3847     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
3848     NvmeNamespace *ns = req->ns;
3849     NvmeZone *zone;
3850     NvmeZoneResetAIOCB *iocb;
3851     uint8_t *zd_ext;
3852     uint64_t slba = 0;
3853     uint32_t zone_idx = 0;
3854     uint16_t status;
3855     uint8_t action = cmd->zsa;
3856     bool all;
3857     enum NvmeZoneProcessingMask proc_mask = NVME_PROC_CURRENT_ZONE;
3858 
3859     all = cmd->zsflags & NVME_ZSFLAG_SELECT_ALL;
3860 
3861     req->status = NVME_SUCCESS;
3862 
3863     if (!all) {
3864         status = nvme_get_mgmt_zone_slba_idx(ns, &req->cmd, &slba, &zone_idx);
3865         if (status) {
3866             return status;
3867         }
3868     }
3869 
3870     zone = &ns->zone_array[zone_idx];
3871     if (slba != zone->d.zslba && action != NVME_ZONE_ACTION_ZRWA_FLUSH) {
3872         trace_pci_nvme_err_unaligned_zone_cmd(action, slba, zone->d.zslba);
3873         return NVME_INVALID_FIELD | NVME_DNR;
3874     }
3875 
3876     switch (action) {
3877 
3878     case NVME_ZONE_ACTION_OPEN:
3879         if (all) {
3880             proc_mask = NVME_PROC_CLOSED_ZONES;
3881         }
3882         trace_pci_nvme_open_zone(slba, zone_idx, all);
3883         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_open_zone, req);
3884         break;
3885 
3886     case NVME_ZONE_ACTION_CLOSE:
3887         if (all) {
3888             proc_mask = NVME_PROC_OPENED_ZONES;
3889         }
3890         trace_pci_nvme_close_zone(slba, zone_idx, all);
3891         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_close_zone, req);
3892         break;
3893 
3894     case NVME_ZONE_ACTION_FINISH:
3895         if (all) {
3896             proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES;
3897         }
3898         trace_pci_nvme_finish_zone(slba, zone_idx, all);
3899         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_finish_zone, req);
3900         break;
3901 
3902     case NVME_ZONE_ACTION_RESET:
3903         trace_pci_nvme_reset_zone(slba, zone_idx, all);
3904 
3905         iocb = blk_aio_get(&nvme_zone_reset_aiocb_info, ns->blkconf.blk,
3906                            nvme_misc_cb, req);
3907 
3908         iocb->req = req;
3909         iocb->bh = qemu_bh_new(nvme_zone_reset_bh, iocb);
3910         iocb->ret = 0;
3911         iocb->all = all;
3912         iocb->idx = zone_idx;
3913         iocb->zone = NULL;
3914 
3915         req->aiocb = &iocb->common;
3916         nvme_zone_reset_cb(iocb, 0);
3917 
3918         return NVME_NO_COMPLETE;
3919 
3920     case NVME_ZONE_ACTION_OFFLINE:
3921         if (all) {
3922             proc_mask = NVME_PROC_READ_ONLY_ZONES;
3923         }
3924         trace_pci_nvme_offline_zone(slba, zone_idx, all);
3925         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_offline_zone, req);
3926         break;
3927 
3928     case NVME_ZONE_ACTION_SET_ZD_EXT:
3929         trace_pci_nvme_set_descriptor_extension(slba, zone_idx);
3930         if (all || !ns->params.zd_extension_size) {
3931             return NVME_INVALID_FIELD | NVME_DNR;
3932         }
3933         zd_ext = nvme_get_zd_extension(ns, zone_idx);
3934         status = nvme_h2c(n, zd_ext, ns->params.zd_extension_size, req);
3935         if (status) {
3936             trace_pci_nvme_err_zd_extension_map_error(zone_idx);
3937             return status;
3938         }
3939 
3940         status = nvme_set_zd_ext(ns, zone);
3941         if (status == NVME_SUCCESS) {
3942             trace_pci_nvme_zd_extension_set(zone_idx);
3943             return status;
3944         }
3945         break;
3946 
3947     case NVME_ZONE_ACTION_ZRWA_FLUSH:
3948         if (all) {
3949             return NVME_INVALID_FIELD | NVME_DNR;
3950         }
3951 
3952         return nvme_zone_mgmt_send_zrwa_flush(n, zone, slba, req);
3953 
3954     default:
3955         trace_pci_nvme_err_invalid_mgmt_action(action);
3956         status = NVME_INVALID_FIELD;
3957     }
3958 
3959     if (status == NVME_ZONE_INVAL_TRANSITION) {
3960         trace_pci_nvme_err_invalid_zone_state_transition(action, slba,
3961                                                          zone->d.za);
3962     }
3963     if (status) {
3964         status |= NVME_DNR;
3965     }
3966 
3967     return status;
3968 }
3969 
3970 static bool nvme_zone_matches_filter(uint32_t zafs, NvmeZone *zl)
3971 {
3972     NvmeZoneState zs = nvme_get_zone_state(zl);
3973 
3974     switch (zafs) {
3975     case NVME_ZONE_REPORT_ALL:
3976         return true;
3977     case NVME_ZONE_REPORT_EMPTY:
3978         return zs == NVME_ZONE_STATE_EMPTY;
3979     case NVME_ZONE_REPORT_IMPLICITLY_OPEN:
3980         return zs == NVME_ZONE_STATE_IMPLICITLY_OPEN;
3981     case NVME_ZONE_REPORT_EXPLICITLY_OPEN:
3982         return zs == NVME_ZONE_STATE_EXPLICITLY_OPEN;
3983     case NVME_ZONE_REPORT_CLOSED:
3984         return zs == NVME_ZONE_STATE_CLOSED;
3985     case NVME_ZONE_REPORT_FULL:
3986         return zs == NVME_ZONE_STATE_FULL;
3987     case NVME_ZONE_REPORT_READ_ONLY:
3988         return zs == NVME_ZONE_STATE_READ_ONLY;
3989     case NVME_ZONE_REPORT_OFFLINE:
3990         return zs == NVME_ZONE_STATE_OFFLINE;
3991     default:
3992         return false;
3993     }
3994 }
3995 
3996 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
3997 {
3998     NvmeCmd *cmd = (NvmeCmd *)&req->cmd;
3999     NvmeNamespace *ns = req->ns;
4000     /* cdw12 is zero-based number of dwords to return. Convert to bytes */
4001     uint32_t data_size = (le32_to_cpu(cmd->cdw12) + 1) << 2;
4002     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
4003     uint32_t zone_idx, zra, zrasf, partial;
4004     uint64_t max_zones, nr_zones = 0;
4005     uint16_t status;
4006     uint64_t slba;
4007     NvmeZoneDescr *z;
4008     NvmeZone *zone;
4009     NvmeZoneReportHeader *header;
4010     void *buf, *buf_p;
4011     size_t zone_entry_sz;
4012     int i;
4013 
4014     req->status = NVME_SUCCESS;
4015 
4016     status = nvme_get_mgmt_zone_slba_idx(ns, cmd, &slba, &zone_idx);
4017     if (status) {
4018         return status;
4019     }
4020 
4021     zra = dw13 & 0xff;
4022     if (zra != NVME_ZONE_REPORT && zra != NVME_ZONE_REPORT_EXTENDED) {
4023         return NVME_INVALID_FIELD | NVME_DNR;
4024     }
4025     if (zra == NVME_ZONE_REPORT_EXTENDED && !ns->params.zd_extension_size) {
4026         return NVME_INVALID_FIELD | NVME_DNR;
4027     }
4028 
4029     zrasf = (dw13 >> 8) & 0xff;
4030     if (zrasf > NVME_ZONE_REPORT_OFFLINE) {
4031         return NVME_INVALID_FIELD | NVME_DNR;
4032     }
4033 
4034     if (data_size < sizeof(NvmeZoneReportHeader)) {
4035         return NVME_INVALID_FIELD | NVME_DNR;
4036     }
4037 
4038     status = nvme_check_mdts(n, data_size);
4039     if (status) {
4040         return status;
4041     }
4042 
4043     partial = (dw13 >> 16) & 0x01;
4044 
4045     zone_entry_sz = sizeof(NvmeZoneDescr);
4046     if (zra == NVME_ZONE_REPORT_EXTENDED) {
4047         zone_entry_sz += ns->params.zd_extension_size;
4048     }
4049 
4050     max_zones = (data_size - sizeof(NvmeZoneReportHeader)) / zone_entry_sz;
4051     buf = g_malloc0(data_size);
4052 
4053     zone = &ns->zone_array[zone_idx];
4054     for (i = zone_idx; i < ns->num_zones; i++) {
4055         if (partial && nr_zones >= max_zones) {
4056             break;
4057         }
4058         if (nvme_zone_matches_filter(zrasf, zone++)) {
4059             nr_zones++;
4060         }
4061     }
4062     header = (NvmeZoneReportHeader *)buf;
4063     header->nr_zones = cpu_to_le64(nr_zones);
4064 
4065     buf_p = buf + sizeof(NvmeZoneReportHeader);
4066     for (; zone_idx < ns->num_zones && max_zones > 0; zone_idx++) {
4067         zone = &ns->zone_array[zone_idx];
4068         if (nvme_zone_matches_filter(zrasf, zone)) {
4069             z = (NvmeZoneDescr *)buf_p;
4070             buf_p += sizeof(NvmeZoneDescr);
4071 
4072             z->zt = zone->d.zt;
4073             z->zs = zone->d.zs;
4074             z->zcap = cpu_to_le64(zone->d.zcap);
4075             z->zslba = cpu_to_le64(zone->d.zslba);
4076             z->za = zone->d.za;
4077 
4078             if (nvme_wp_is_valid(zone)) {
4079                 z->wp = cpu_to_le64(zone->d.wp);
4080             } else {
4081                 z->wp = cpu_to_le64(~0ULL);
4082             }
4083 
4084             if (zra == NVME_ZONE_REPORT_EXTENDED) {
4085                 if (zone->d.za & NVME_ZA_ZD_EXT_VALID) {
4086                     memcpy(buf_p, nvme_get_zd_extension(ns, zone_idx),
4087                            ns->params.zd_extension_size);
4088                 }
4089                 buf_p += ns->params.zd_extension_size;
4090             }
4091 
4092             max_zones--;
4093         }
4094     }
4095 
4096     status = nvme_c2h(n, (uint8_t *)buf, data_size, req);
4097 
4098     g_free(buf);
4099 
4100     return status;
4101 }
4102 
4103 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
4104 {
4105     NvmeNamespace *ns;
4106     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4107 
4108     trace_pci_nvme_io_cmd(nvme_cid(req), nsid, nvme_sqid(req),
4109                           req->cmd.opcode, nvme_io_opc_str(req->cmd.opcode));
4110 
4111     if (!nvme_nsid_valid(n, nsid)) {
4112         return NVME_INVALID_NSID | NVME_DNR;
4113     }
4114 
4115     /*
4116      * In the base NVM command set, Flush may apply to all namespaces
4117      * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4118      * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4119      *
4120      * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4121      * opcode with a specific command since we cannot determine a unique I/O
4122      * command set. Opcode 0h could have any other meaning than something
4123      * equivalent to flushing and say it DOES have completely different
4124      * semantics in some other command set - does an NSID of FFFFFFFFh then
4125      * mean "for all namespaces, apply whatever command set specific command
4126      * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4127      * whatever command that uses the 0h opcode if, and only if, it allows NSID
4128      * to be FFFFFFFFh"?
4129      *
4130      * Anyway (and luckily), for now, we do not care about this since the
4131      * device only supports namespace types that includes the NVM Flush command
4132      * (NVM and Zoned), so always do an NVM Flush.
4133      */
4134     if (req->cmd.opcode == NVME_CMD_FLUSH) {
4135         return nvme_flush(n, req);
4136     }
4137 
4138     ns = nvme_ns(n, nsid);
4139     if (unlikely(!ns)) {
4140         return NVME_INVALID_FIELD | NVME_DNR;
4141     }
4142 
4143     if (!(ns->iocs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
4144         trace_pci_nvme_err_invalid_opc(req->cmd.opcode);
4145         return NVME_INVALID_OPCODE | NVME_DNR;
4146     }
4147 
4148     if (ns->status) {
4149         return ns->status;
4150     }
4151 
4152     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
4153         return NVME_INVALID_FIELD;
4154     }
4155 
4156     req->ns = ns;
4157 
4158     switch (req->cmd.opcode) {
4159     case NVME_CMD_WRITE_ZEROES:
4160         return nvme_write_zeroes(n, req);
4161     case NVME_CMD_ZONE_APPEND:
4162         return nvme_zone_append(n, req);
4163     case NVME_CMD_WRITE:
4164         return nvme_write(n, req);
4165     case NVME_CMD_READ:
4166         return nvme_read(n, req);
4167     case NVME_CMD_COMPARE:
4168         return nvme_compare(n, req);
4169     case NVME_CMD_DSM:
4170         return nvme_dsm(n, req);
4171     case NVME_CMD_VERIFY:
4172         return nvme_verify(n, req);
4173     case NVME_CMD_COPY:
4174         return nvme_copy(n, req);
4175     case NVME_CMD_ZONE_MGMT_SEND:
4176         return nvme_zone_mgmt_send(n, req);
4177     case NVME_CMD_ZONE_MGMT_RECV:
4178         return nvme_zone_mgmt_recv(n, req);
4179     default:
4180         assert(false);
4181     }
4182 
4183     return NVME_INVALID_OPCODE | NVME_DNR;
4184 }
4185 
4186 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
4187 {
4188     n->sq[sq->sqid] = NULL;
4189     timer_free(sq->timer);
4190     g_free(sq->io_req);
4191     if (sq->sqid) {
4192         g_free(sq);
4193     }
4194 }
4195 
4196 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
4197 {
4198     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4199     NvmeRequest *r, *next;
4200     NvmeSQueue *sq;
4201     NvmeCQueue *cq;
4202     uint16_t qid = le16_to_cpu(c->qid);
4203 
4204     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
4205         trace_pci_nvme_err_invalid_del_sq(qid);
4206         return NVME_INVALID_QID | NVME_DNR;
4207     }
4208 
4209     trace_pci_nvme_del_sq(qid);
4210 
4211     sq = n->sq[qid];
4212     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
4213         r = QTAILQ_FIRST(&sq->out_req_list);
4214         assert(r->aiocb);
4215         blk_aio_cancel(r->aiocb);
4216     }
4217 
4218     assert(QTAILQ_EMPTY(&sq->out_req_list));
4219 
4220     if (!nvme_check_cqid(n, sq->cqid)) {
4221         cq = n->cq[sq->cqid];
4222         QTAILQ_REMOVE(&cq->sq_list, sq, entry);
4223 
4224         nvme_post_cqes(cq);
4225         QTAILQ_FOREACH_SAFE(r, &cq->req_list, entry, next) {
4226             if (r->sq == sq) {
4227                 QTAILQ_REMOVE(&cq->req_list, r, entry);
4228                 QTAILQ_INSERT_TAIL(&sq->req_list, r, entry);
4229             }
4230         }
4231     }
4232 
4233     nvme_free_sq(sq, n);
4234     return NVME_SUCCESS;
4235 }
4236 
4237 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
4238                          uint16_t sqid, uint16_t cqid, uint16_t size)
4239 {
4240     int i;
4241     NvmeCQueue *cq;
4242 
4243     sq->ctrl = n;
4244     sq->dma_addr = dma_addr;
4245     sq->sqid = sqid;
4246     sq->size = size;
4247     sq->cqid = cqid;
4248     sq->head = sq->tail = 0;
4249     sq->io_req = g_new0(NvmeRequest, sq->size);
4250 
4251     QTAILQ_INIT(&sq->req_list);
4252     QTAILQ_INIT(&sq->out_req_list);
4253     for (i = 0; i < sq->size; i++) {
4254         sq->io_req[i].sq = sq;
4255         QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
4256     }
4257     sq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_process_sq, sq);
4258 
4259     assert(n->cq[cqid]);
4260     cq = n->cq[cqid];
4261     QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
4262     n->sq[sqid] = sq;
4263 }
4264 
4265 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeRequest *req)
4266 {
4267     NvmeSQueue *sq;
4268     NvmeCreateSq *c = (NvmeCreateSq *)&req->cmd;
4269 
4270     uint16_t cqid = le16_to_cpu(c->cqid);
4271     uint16_t sqid = le16_to_cpu(c->sqid);
4272     uint16_t qsize = le16_to_cpu(c->qsize);
4273     uint16_t qflags = le16_to_cpu(c->sq_flags);
4274     uint64_t prp1 = le64_to_cpu(c->prp1);
4275 
4276     trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
4277 
4278     if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
4279         trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
4280         return NVME_INVALID_CQID | NVME_DNR;
4281     }
4282     if (unlikely(!sqid || sqid > n->params.max_ioqpairs ||
4283         n->sq[sqid] != NULL)) {
4284         trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
4285         return NVME_INVALID_QID | NVME_DNR;
4286     }
4287     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
4288         trace_pci_nvme_err_invalid_create_sq_size(qsize);
4289         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4290     }
4291     if (unlikely(prp1 & (n->page_size - 1))) {
4292         trace_pci_nvme_err_invalid_create_sq_addr(prp1);
4293         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4294     }
4295     if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
4296         trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
4297         return NVME_INVALID_FIELD | NVME_DNR;
4298     }
4299     sq = g_malloc0(sizeof(*sq));
4300     nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
4301     return NVME_SUCCESS;
4302 }
4303 
4304 struct nvme_stats {
4305     uint64_t units_read;
4306     uint64_t units_written;
4307     uint64_t read_commands;
4308     uint64_t write_commands;
4309 };
4310 
4311 static void nvme_set_blk_stats(NvmeNamespace *ns, struct nvme_stats *stats)
4312 {
4313     BlockAcctStats *s = blk_get_stats(ns->blkconf.blk);
4314 
4315     stats->units_read += s->nr_bytes[BLOCK_ACCT_READ] >> BDRV_SECTOR_BITS;
4316     stats->units_written += s->nr_bytes[BLOCK_ACCT_WRITE] >> BDRV_SECTOR_BITS;
4317     stats->read_commands += s->nr_ops[BLOCK_ACCT_READ];
4318     stats->write_commands += s->nr_ops[BLOCK_ACCT_WRITE];
4319 }
4320 
4321 static uint16_t nvme_smart_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4322                                 uint64_t off, NvmeRequest *req)
4323 {
4324     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4325     struct nvme_stats stats = { 0 };
4326     NvmeSmartLog smart = { 0 };
4327     uint32_t trans_len;
4328     NvmeNamespace *ns;
4329     time_t current_ms;
4330 
4331     if (off >= sizeof(smart)) {
4332         return NVME_INVALID_FIELD | NVME_DNR;
4333     }
4334 
4335     if (nsid != 0xffffffff) {
4336         ns = nvme_ns(n, nsid);
4337         if (!ns) {
4338             return NVME_INVALID_NSID | NVME_DNR;
4339         }
4340         nvme_set_blk_stats(ns, &stats);
4341     } else {
4342         int i;
4343 
4344         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4345             ns = nvme_ns(n, i);
4346             if (!ns) {
4347                 continue;
4348             }
4349             nvme_set_blk_stats(ns, &stats);
4350         }
4351     }
4352 
4353     trans_len = MIN(sizeof(smart) - off, buf_len);
4354     smart.critical_warning = n->smart_critical_warning;
4355 
4356     smart.data_units_read[0] = cpu_to_le64(DIV_ROUND_UP(stats.units_read,
4357                                                         1000));
4358     smart.data_units_written[0] = cpu_to_le64(DIV_ROUND_UP(stats.units_written,
4359                                                            1000));
4360     smart.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4361     smart.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4362 
4363     smart.temperature = cpu_to_le16(n->temperature);
4364 
4365     if ((n->temperature >= n->features.temp_thresh_hi) ||
4366         (n->temperature <= n->features.temp_thresh_low)) {
4367         smart.critical_warning |= NVME_SMART_TEMPERATURE;
4368     }
4369 
4370     current_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4371     smart.power_on_hours[0] =
4372         cpu_to_le64((((current_ms - n->starttime_ms) / 1000) / 60) / 60);
4373 
4374     if (!rae) {
4375         nvme_clear_events(n, NVME_AER_TYPE_SMART);
4376     }
4377 
4378     return nvme_c2h(n, (uint8_t *) &smart + off, trans_len, req);
4379 }
4380 
4381 static uint16_t nvme_fw_log_info(NvmeCtrl *n, uint32_t buf_len, uint64_t off,
4382                                  NvmeRequest *req)
4383 {
4384     uint32_t trans_len;
4385     NvmeFwSlotInfoLog fw_log = {
4386         .afi = 0x1,
4387     };
4388 
4389     if (off >= sizeof(fw_log)) {
4390         return NVME_INVALID_FIELD | NVME_DNR;
4391     }
4392 
4393     strpadcpy((char *)&fw_log.frs1, sizeof(fw_log.frs1), "1.0", ' ');
4394     trans_len = MIN(sizeof(fw_log) - off, buf_len);
4395 
4396     return nvme_c2h(n, (uint8_t *) &fw_log + off, trans_len, req);
4397 }
4398 
4399 static uint16_t nvme_error_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4400                                 uint64_t off, NvmeRequest *req)
4401 {
4402     uint32_t trans_len;
4403     NvmeErrorLog errlog;
4404 
4405     if (off >= sizeof(errlog)) {
4406         return NVME_INVALID_FIELD | NVME_DNR;
4407     }
4408 
4409     if (!rae) {
4410         nvme_clear_events(n, NVME_AER_TYPE_ERROR);
4411     }
4412 
4413     memset(&errlog, 0x0, sizeof(errlog));
4414     trans_len = MIN(sizeof(errlog) - off, buf_len);
4415 
4416     return nvme_c2h(n, (uint8_t *)&errlog, trans_len, req);
4417 }
4418 
4419 static uint16_t nvme_changed_nslist(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4420                                     uint64_t off, NvmeRequest *req)
4421 {
4422     uint32_t nslist[1024];
4423     uint32_t trans_len;
4424     int i = 0;
4425     uint32_t nsid;
4426 
4427     if (off >= sizeof(nslist)) {
4428         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(nslist));
4429         return NVME_INVALID_FIELD | NVME_DNR;
4430     }
4431 
4432     memset(nslist, 0x0, sizeof(nslist));
4433     trans_len = MIN(sizeof(nslist) - off, buf_len);
4434 
4435     while ((nsid = find_first_bit(n->changed_nsids, NVME_CHANGED_NSID_SIZE)) !=
4436             NVME_CHANGED_NSID_SIZE) {
4437         /*
4438          * If more than 1024 namespaces, the first entry in the log page should
4439          * be set to FFFFFFFFh and the others to 0 as spec.
4440          */
4441         if (i == ARRAY_SIZE(nslist)) {
4442             memset(nslist, 0x0, sizeof(nslist));
4443             nslist[0] = 0xffffffff;
4444             break;
4445         }
4446 
4447         nslist[i++] = nsid;
4448         clear_bit(nsid, n->changed_nsids);
4449     }
4450 
4451     /*
4452      * Remove all the remaining list entries in case returns directly due to
4453      * more than 1024 namespaces.
4454      */
4455     if (nslist[0] == 0xffffffff) {
4456         bitmap_zero(n->changed_nsids, NVME_CHANGED_NSID_SIZE);
4457     }
4458 
4459     if (!rae) {
4460         nvme_clear_events(n, NVME_AER_TYPE_NOTICE);
4461     }
4462 
4463     return nvme_c2h(n, ((uint8_t *)nslist) + off, trans_len, req);
4464 }
4465 
4466 static uint16_t nvme_cmd_effects(NvmeCtrl *n, uint8_t csi, uint32_t buf_len,
4467                                  uint64_t off, NvmeRequest *req)
4468 {
4469     NvmeEffectsLog log = {};
4470     const uint32_t *src_iocs = NULL;
4471     uint32_t trans_len;
4472 
4473     if (off >= sizeof(log)) {
4474         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(log));
4475         return NVME_INVALID_FIELD | NVME_DNR;
4476     }
4477 
4478     switch (NVME_CC_CSS(ldl_le_p(&n->bar.cc))) {
4479     case NVME_CC_CSS_NVM:
4480         src_iocs = nvme_cse_iocs_nvm;
4481         /* fall through */
4482     case NVME_CC_CSS_ADMIN_ONLY:
4483         break;
4484     case NVME_CC_CSS_CSI:
4485         switch (csi) {
4486         case NVME_CSI_NVM:
4487             src_iocs = nvme_cse_iocs_nvm;
4488             break;
4489         case NVME_CSI_ZONED:
4490             src_iocs = nvme_cse_iocs_zoned;
4491             break;
4492         }
4493     }
4494 
4495     memcpy(log.acs, nvme_cse_acs, sizeof(nvme_cse_acs));
4496 
4497     if (src_iocs) {
4498         memcpy(log.iocs, src_iocs, sizeof(log.iocs));
4499     }
4500 
4501     trans_len = MIN(sizeof(log) - off, buf_len);
4502 
4503     return nvme_c2h(n, ((uint8_t *)&log) + off, trans_len, req);
4504 }
4505 
4506 static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
4507 {
4508     NvmeCmd *cmd = &req->cmd;
4509 
4510     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
4511     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
4512     uint32_t dw12 = le32_to_cpu(cmd->cdw12);
4513     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
4514     uint8_t  lid = dw10 & 0xff;
4515     uint8_t  lsp = (dw10 >> 8) & 0xf;
4516     uint8_t  rae = (dw10 >> 15) & 0x1;
4517     uint8_t  csi = le32_to_cpu(cmd->cdw14) >> 24;
4518     uint32_t numdl, numdu;
4519     uint64_t off, lpol, lpou;
4520     size_t   len;
4521     uint16_t status;
4522 
4523     numdl = (dw10 >> 16);
4524     numdu = (dw11 & 0xffff);
4525     lpol = dw12;
4526     lpou = dw13;
4527 
4528     len = (((numdu << 16) | numdl) + 1) << 2;
4529     off = (lpou << 32ULL) | lpol;
4530 
4531     if (off & 0x3) {
4532         return NVME_INVALID_FIELD | NVME_DNR;
4533     }
4534 
4535     trace_pci_nvme_get_log(nvme_cid(req), lid, lsp, rae, len, off);
4536 
4537     status = nvme_check_mdts(n, len);
4538     if (status) {
4539         return status;
4540     }
4541 
4542     switch (lid) {
4543     case NVME_LOG_ERROR_INFO:
4544         return nvme_error_info(n, rae, len, off, req);
4545     case NVME_LOG_SMART_INFO:
4546         return nvme_smart_info(n, rae, len, off, req);
4547     case NVME_LOG_FW_SLOT_INFO:
4548         return nvme_fw_log_info(n, len, off, req);
4549     case NVME_LOG_CHANGED_NSLIST:
4550         return nvme_changed_nslist(n, rae, len, off, req);
4551     case NVME_LOG_CMD_EFFECTS:
4552         return nvme_cmd_effects(n, csi, len, off, req);
4553     default:
4554         trace_pci_nvme_err_invalid_log_page(nvme_cid(req), lid);
4555         return NVME_INVALID_FIELD | NVME_DNR;
4556     }
4557 }
4558 
4559 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
4560 {
4561     n->cq[cq->cqid] = NULL;
4562     timer_free(cq->timer);
4563     if (msix_enabled(&n->parent_obj)) {
4564         msix_vector_unuse(&n->parent_obj, cq->vector);
4565     }
4566     if (cq->cqid) {
4567         g_free(cq);
4568     }
4569 }
4570 
4571 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)
4572 {
4573     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4574     NvmeCQueue *cq;
4575     uint16_t qid = le16_to_cpu(c->qid);
4576 
4577     if (unlikely(!qid || nvme_check_cqid(n, qid))) {
4578         trace_pci_nvme_err_invalid_del_cq_cqid(qid);
4579         return NVME_INVALID_CQID | NVME_DNR;
4580     }
4581 
4582     cq = n->cq[qid];
4583     if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
4584         trace_pci_nvme_err_invalid_del_cq_notempty(qid);
4585         return NVME_INVALID_QUEUE_DEL;
4586     }
4587 
4588     if (cq->irq_enabled && cq->tail != cq->head) {
4589         n->cq_pending--;
4590     }
4591 
4592     nvme_irq_deassert(n, cq);
4593     trace_pci_nvme_del_cq(qid);
4594     nvme_free_cq(cq, n);
4595     return NVME_SUCCESS;
4596 }
4597 
4598 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
4599                          uint16_t cqid, uint16_t vector, uint16_t size,
4600                          uint16_t irq_enabled)
4601 {
4602     int ret;
4603 
4604     if (msix_enabled(&n->parent_obj)) {
4605         ret = msix_vector_use(&n->parent_obj, vector);
4606         assert(ret == 0);
4607     }
4608     cq->ctrl = n;
4609     cq->cqid = cqid;
4610     cq->size = size;
4611     cq->dma_addr = dma_addr;
4612     cq->phase = 1;
4613     cq->irq_enabled = irq_enabled;
4614     cq->vector = vector;
4615     cq->head = cq->tail = 0;
4616     QTAILQ_INIT(&cq->req_list);
4617     QTAILQ_INIT(&cq->sq_list);
4618     n->cq[cqid] = cq;
4619     cq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_post_cqes, cq);
4620 }
4621 
4622 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeRequest *req)
4623 {
4624     NvmeCQueue *cq;
4625     NvmeCreateCq *c = (NvmeCreateCq *)&req->cmd;
4626     uint16_t cqid = le16_to_cpu(c->cqid);
4627     uint16_t vector = le16_to_cpu(c->irq_vector);
4628     uint16_t qsize = le16_to_cpu(c->qsize);
4629     uint16_t qflags = le16_to_cpu(c->cq_flags);
4630     uint64_t prp1 = le64_to_cpu(c->prp1);
4631 
4632     trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
4633                              NVME_CQ_FLAGS_IEN(qflags) != 0);
4634 
4635     if (unlikely(!cqid || cqid > n->params.max_ioqpairs ||
4636         n->cq[cqid] != NULL)) {
4637         trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
4638         return NVME_INVALID_QID | NVME_DNR;
4639     }
4640     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
4641         trace_pci_nvme_err_invalid_create_cq_size(qsize);
4642         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4643     }
4644     if (unlikely(prp1 & (n->page_size - 1))) {
4645         trace_pci_nvme_err_invalid_create_cq_addr(prp1);
4646         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4647     }
4648     if (unlikely(!msix_enabled(&n->parent_obj) && vector)) {
4649         trace_pci_nvme_err_invalid_create_cq_vector(vector);
4650         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
4651     }
4652     if (unlikely(vector >= n->params.msix_qsize)) {
4653         trace_pci_nvme_err_invalid_create_cq_vector(vector);
4654         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
4655     }
4656     if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
4657         trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
4658         return NVME_INVALID_FIELD | NVME_DNR;
4659     }
4660 
4661     cq = g_malloc0(sizeof(*cq));
4662     nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
4663                  NVME_CQ_FLAGS_IEN(qflags));
4664 
4665     /*
4666      * It is only required to set qs_created when creating a completion queue;
4667      * creating a submission queue without a matching completion queue will
4668      * fail.
4669      */
4670     n->qs_created = true;
4671     return NVME_SUCCESS;
4672 }
4673 
4674 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl *n, NvmeRequest *req)
4675 {
4676     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
4677 
4678     return nvme_c2h(n, id, sizeof(id), req);
4679 }
4680 
4681 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeRequest *req)
4682 {
4683     trace_pci_nvme_identify_ctrl();
4684 
4685     return nvme_c2h(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl), req);
4686 }
4687 
4688 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl *n, NvmeRequest *req)
4689 {
4690     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4691     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
4692     NvmeIdCtrlNvm *id_nvm = (NvmeIdCtrlNvm *)&id;
4693 
4694     trace_pci_nvme_identify_ctrl_csi(c->csi);
4695 
4696     switch (c->csi) {
4697     case NVME_CSI_NVM:
4698         id_nvm->vsl = n->params.vsl;
4699         id_nvm->dmrsl = cpu_to_le32(n->dmrsl);
4700         break;
4701 
4702     case NVME_CSI_ZONED:
4703         ((NvmeIdCtrlZoned *)&id)->zasl = n->params.zasl;
4704         break;
4705 
4706     default:
4707         return NVME_INVALID_FIELD | NVME_DNR;
4708     }
4709 
4710     return nvme_c2h(n, id, sizeof(id), req);
4711 }
4712 
4713 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeRequest *req, bool active)
4714 {
4715     NvmeNamespace *ns;
4716     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4717     uint32_t nsid = le32_to_cpu(c->nsid);
4718 
4719     trace_pci_nvme_identify_ns(nsid);
4720 
4721     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4722         return NVME_INVALID_NSID | NVME_DNR;
4723     }
4724 
4725     ns = nvme_ns(n, nsid);
4726     if (unlikely(!ns)) {
4727         if (!active) {
4728             ns = nvme_subsys_ns(n->subsys, nsid);
4729             if (!ns) {
4730                 return nvme_rpt_empty_id_struct(n, req);
4731             }
4732         } else {
4733             return nvme_rpt_empty_id_struct(n, req);
4734         }
4735     }
4736 
4737     if (active || ns->csi == NVME_CSI_NVM) {
4738         return nvme_c2h(n, (uint8_t *)&ns->id_ns, sizeof(NvmeIdNs), req);
4739     }
4740 
4741     return NVME_INVALID_CMD_SET | NVME_DNR;
4742 }
4743 
4744 static uint16_t nvme_identify_ctrl_list(NvmeCtrl *n, NvmeRequest *req,
4745                                         bool attached)
4746 {
4747     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4748     uint32_t nsid = le32_to_cpu(c->nsid);
4749     uint16_t min_id = le16_to_cpu(c->ctrlid);
4750     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
4751     uint16_t *ids = &list[1];
4752     NvmeNamespace *ns;
4753     NvmeCtrl *ctrl;
4754     int cntlid, nr_ids = 0;
4755 
4756     trace_pci_nvme_identify_ctrl_list(c->cns, min_id);
4757 
4758     if (!n->subsys) {
4759         return NVME_INVALID_FIELD | NVME_DNR;
4760     }
4761 
4762     if (attached) {
4763         if (nsid == NVME_NSID_BROADCAST) {
4764             return NVME_INVALID_FIELD | NVME_DNR;
4765         }
4766 
4767         ns = nvme_subsys_ns(n->subsys, nsid);
4768         if (!ns) {
4769             return NVME_INVALID_FIELD | NVME_DNR;
4770         }
4771     }
4772 
4773     for (cntlid = min_id; cntlid < ARRAY_SIZE(n->subsys->ctrls); cntlid++) {
4774         ctrl = nvme_subsys_ctrl(n->subsys, cntlid);
4775         if (!ctrl) {
4776             continue;
4777         }
4778 
4779         if (attached && !nvme_ns(ctrl, nsid)) {
4780             continue;
4781         }
4782 
4783         ids[nr_ids++] = cntlid;
4784     }
4785 
4786     list[0] = nr_ids;
4787 
4788     return nvme_c2h(n, (uint8_t *)list, sizeof(list), req);
4789 }
4790 
4791 static uint16_t nvme_identify_ns_csi(NvmeCtrl *n, NvmeRequest *req,
4792                                      bool active)
4793 {
4794     NvmeNamespace *ns;
4795     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4796     uint32_t nsid = le32_to_cpu(c->nsid);
4797 
4798     trace_pci_nvme_identify_ns_csi(nsid, c->csi);
4799 
4800     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4801         return NVME_INVALID_NSID | NVME_DNR;
4802     }
4803 
4804     ns = nvme_ns(n, nsid);
4805     if (unlikely(!ns)) {
4806         if (!active) {
4807             ns = nvme_subsys_ns(n->subsys, nsid);
4808             if (!ns) {
4809                 return nvme_rpt_empty_id_struct(n, req);
4810             }
4811         } else {
4812             return nvme_rpt_empty_id_struct(n, req);
4813         }
4814     }
4815 
4816     if (c->csi == NVME_CSI_NVM) {
4817         return nvme_c2h(n, (uint8_t *)&ns->id_ns_nvm, sizeof(NvmeIdNsNvm),
4818                         req);
4819     } else if (c->csi == NVME_CSI_ZONED && ns->csi == NVME_CSI_ZONED) {
4820         return nvme_c2h(n, (uint8_t *)ns->id_ns_zoned, sizeof(NvmeIdNsZoned),
4821                         req);
4822     }
4823 
4824     return NVME_INVALID_FIELD | NVME_DNR;
4825 }
4826 
4827 static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeRequest *req,
4828                                      bool active)
4829 {
4830     NvmeNamespace *ns;
4831     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4832     uint32_t min_nsid = le32_to_cpu(c->nsid);
4833     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
4834     static const int data_len = sizeof(list);
4835     uint32_t *list_ptr = (uint32_t *)list;
4836     int i, j = 0;
4837 
4838     trace_pci_nvme_identify_nslist(min_nsid);
4839 
4840     /*
4841      * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
4842      * since the Active Namespace ID List should return namespaces with ids
4843      * *higher* than the NSID specified in the command. This is also specified
4844      * in the spec (NVM Express v1.3d, Section 5.15.4).
4845      */
4846     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
4847         return NVME_INVALID_NSID | NVME_DNR;
4848     }
4849 
4850     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4851         ns = nvme_ns(n, i);
4852         if (!ns) {
4853             if (!active) {
4854                 ns = nvme_subsys_ns(n->subsys, i);
4855                 if (!ns) {
4856                     continue;
4857                 }
4858             } else {
4859                 continue;
4860             }
4861         }
4862         if (ns->params.nsid <= min_nsid) {
4863             continue;
4864         }
4865         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
4866         if (j == data_len / sizeof(uint32_t)) {
4867             break;
4868         }
4869     }
4870 
4871     return nvme_c2h(n, list, data_len, req);
4872 }
4873 
4874 static uint16_t nvme_identify_nslist_csi(NvmeCtrl *n, NvmeRequest *req,
4875                                          bool active)
4876 {
4877     NvmeNamespace *ns;
4878     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4879     uint32_t min_nsid = le32_to_cpu(c->nsid);
4880     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
4881     static const int data_len = sizeof(list);
4882     uint32_t *list_ptr = (uint32_t *)list;
4883     int i, j = 0;
4884 
4885     trace_pci_nvme_identify_nslist_csi(min_nsid, c->csi);
4886 
4887     /*
4888      * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
4889      */
4890     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
4891         return NVME_INVALID_NSID | NVME_DNR;
4892     }
4893 
4894     if (c->csi != NVME_CSI_NVM && c->csi != NVME_CSI_ZONED) {
4895         return NVME_INVALID_FIELD | NVME_DNR;
4896     }
4897 
4898     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4899         ns = nvme_ns(n, i);
4900         if (!ns) {
4901             if (!active) {
4902                 ns = nvme_subsys_ns(n->subsys, i);
4903                 if (!ns) {
4904                     continue;
4905                 }
4906             } else {
4907                 continue;
4908             }
4909         }
4910         if (ns->params.nsid <= min_nsid || c->csi != ns->csi) {
4911             continue;
4912         }
4913         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
4914         if (j == data_len / sizeof(uint32_t)) {
4915             break;
4916         }
4917     }
4918 
4919     return nvme_c2h(n, list, data_len, req);
4920 }
4921 
4922 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl *n, NvmeRequest *req)
4923 {
4924     NvmeNamespace *ns;
4925     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4926     uint32_t nsid = le32_to_cpu(c->nsid);
4927     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
4928     uint8_t *pos = list;
4929     struct {
4930         NvmeIdNsDescr hdr;
4931         uint8_t v[NVME_NIDL_UUID];
4932     } QEMU_PACKED uuid = {};
4933     struct {
4934         NvmeIdNsDescr hdr;
4935         uint64_t v;
4936     } QEMU_PACKED eui64 = {};
4937     struct {
4938         NvmeIdNsDescr hdr;
4939         uint8_t v;
4940     } QEMU_PACKED csi = {};
4941 
4942     trace_pci_nvme_identify_ns_descr_list(nsid);
4943 
4944     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
4945         return NVME_INVALID_NSID | NVME_DNR;
4946     }
4947 
4948     ns = nvme_ns(n, nsid);
4949     if (unlikely(!ns)) {
4950         return NVME_INVALID_FIELD | NVME_DNR;
4951     }
4952 
4953     /*
4954      * If the EUI-64 field is 0 and the NGUID field is 0, the namespace must
4955      * provide a valid Namespace UUID in the Namespace Identification Descriptor
4956      * data structure. QEMU does not yet support setting NGUID.
4957      */
4958     uuid.hdr.nidt = NVME_NIDT_UUID;
4959     uuid.hdr.nidl = NVME_NIDL_UUID;
4960     memcpy(uuid.v, ns->params.uuid.data, NVME_NIDL_UUID);
4961     memcpy(pos, &uuid, sizeof(uuid));
4962     pos += sizeof(uuid);
4963 
4964     if (ns->params.eui64) {
4965         eui64.hdr.nidt = NVME_NIDT_EUI64;
4966         eui64.hdr.nidl = NVME_NIDL_EUI64;
4967         eui64.v = cpu_to_be64(ns->params.eui64);
4968         memcpy(pos, &eui64, sizeof(eui64));
4969         pos += sizeof(eui64);
4970     }
4971 
4972     csi.hdr.nidt = NVME_NIDT_CSI;
4973     csi.hdr.nidl = NVME_NIDL_CSI;
4974     csi.v = ns->csi;
4975     memcpy(pos, &csi, sizeof(csi));
4976     pos += sizeof(csi);
4977 
4978     return nvme_c2h(n, list, sizeof(list), req);
4979 }
4980 
4981 static uint16_t nvme_identify_cmd_set(NvmeCtrl *n, NvmeRequest *req)
4982 {
4983     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
4984     static const int data_len = sizeof(list);
4985 
4986     trace_pci_nvme_identify_cmd_set();
4987 
4988     NVME_SET_CSI(*list, NVME_CSI_NVM);
4989     NVME_SET_CSI(*list, NVME_CSI_ZONED);
4990 
4991     return nvme_c2h(n, list, data_len, req);
4992 }
4993 
4994 static uint16_t nvme_identify(NvmeCtrl *n, NvmeRequest *req)
4995 {
4996     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
4997 
4998     trace_pci_nvme_identify(nvme_cid(req), c->cns, le16_to_cpu(c->ctrlid),
4999                             c->csi);
5000 
5001     switch (c->cns) {
5002     case NVME_ID_CNS_NS:
5003         return nvme_identify_ns(n, req, true);
5004     case NVME_ID_CNS_NS_PRESENT:
5005         return nvme_identify_ns(n, req, false);
5006     case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST:
5007         return nvme_identify_ctrl_list(n, req, true);
5008     case NVME_ID_CNS_CTRL_LIST:
5009         return nvme_identify_ctrl_list(n, req, false);
5010     case NVME_ID_CNS_CS_NS:
5011         return nvme_identify_ns_csi(n, req, true);
5012     case NVME_ID_CNS_CS_NS_PRESENT:
5013         return nvme_identify_ns_csi(n, req, false);
5014     case NVME_ID_CNS_CTRL:
5015         return nvme_identify_ctrl(n, req);
5016     case NVME_ID_CNS_CS_CTRL:
5017         return nvme_identify_ctrl_csi(n, req);
5018     case NVME_ID_CNS_NS_ACTIVE_LIST:
5019         return nvme_identify_nslist(n, req, true);
5020     case NVME_ID_CNS_NS_PRESENT_LIST:
5021         return nvme_identify_nslist(n, req, false);
5022     case NVME_ID_CNS_CS_NS_ACTIVE_LIST:
5023         return nvme_identify_nslist_csi(n, req, true);
5024     case NVME_ID_CNS_CS_NS_PRESENT_LIST:
5025         return nvme_identify_nslist_csi(n, req, false);
5026     case NVME_ID_CNS_NS_DESCR_LIST:
5027         return nvme_identify_ns_descr_list(n, req);
5028     case NVME_ID_CNS_IO_COMMAND_SET:
5029         return nvme_identify_cmd_set(n, req);
5030     default:
5031         trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
5032         return NVME_INVALID_FIELD | NVME_DNR;
5033     }
5034 }
5035 
5036 static uint16_t nvme_abort(NvmeCtrl *n, NvmeRequest *req)
5037 {
5038     uint16_t sqid = le32_to_cpu(req->cmd.cdw10) & 0xffff;
5039 
5040     req->cqe.result = 1;
5041     if (nvme_check_sqid(n, sqid)) {
5042         return NVME_INVALID_FIELD | NVME_DNR;
5043     }
5044 
5045     return NVME_SUCCESS;
5046 }
5047 
5048 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
5049 {
5050     trace_pci_nvme_setfeat_timestamp(ts);
5051 
5052     n->host_timestamp = le64_to_cpu(ts);
5053     n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5054 }
5055 
5056 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
5057 {
5058     uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5059     uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
5060 
5061     union nvme_timestamp {
5062         struct {
5063             uint64_t timestamp:48;
5064             uint64_t sync:1;
5065             uint64_t origin:3;
5066             uint64_t rsvd1:12;
5067         };
5068         uint64_t all;
5069     };
5070 
5071     union nvme_timestamp ts;
5072     ts.all = 0;
5073     ts.timestamp = n->host_timestamp + elapsed_time;
5074 
5075     /* If the host timestamp is non-zero, set the timestamp origin */
5076     ts.origin = n->host_timestamp ? 0x01 : 0x00;
5077 
5078     trace_pci_nvme_getfeat_timestamp(ts.all);
5079 
5080     return cpu_to_le64(ts.all);
5081 }
5082 
5083 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
5084 {
5085     uint64_t timestamp = nvme_get_timestamp(n);
5086 
5087     return nvme_c2h(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
5088 }
5089 
5090 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeRequest *req)
5091 {
5092     NvmeCmd *cmd = &req->cmd;
5093     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5094     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5095     uint32_t nsid = le32_to_cpu(cmd->nsid);
5096     uint32_t result;
5097     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
5098     NvmeGetFeatureSelect sel = NVME_GETFEAT_SELECT(dw10);
5099     uint16_t iv;
5100     NvmeNamespace *ns;
5101     int i;
5102 
5103     static const uint32_t nvme_feature_default[NVME_FID_MAX] = {
5104         [NVME_ARBITRATION] = NVME_ARB_AB_NOLIMIT,
5105     };
5106 
5107     trace_pci_nvme_getfeat(nvme_cid(req), nsid, fid, sel, dw11);
5108 
5109     if (!nvme_feature_support[fid]) {
5110         return NVME_INVALID_FIELD | NVME_DNR;
5111     }
5112 
5113     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
5114         if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5115             /*
5116              * The Reservation Notification Mask and Reservation Persistence
5117              * features require a status code of Invalid Field in Command when
5118              * NSID is FFFFFFFFh. Since the device does not support those
5119              * features we can always return Invalid Namespace or Format as we
5120              * should do for all other features.
5121              */
5122             return NVME_INVALID_NSID | NVME_DNR;
5123         }
5124 
5125         if (!nvme_ns(n, nsid)) {
5126             return NVME_INVALID_FIELD | NVME_DNR;
5127         }
5128     }
5129 
5130     switch (sel) {
5131     case NVME_GETFEAT_SELECT_CURRENT:
5132         break;
5133     case NVME_GETFEAT_SELECT_SAVED:
5134         /* no features are saveable by the controller; fallthrough */
5135     case NVME_GETFEAT_SELECT_DEFAULT:
5136         goto defaults;
5137     case NVME_GETFEAT_SELECT_CAP:
5138         result = nvme_feature_cap[fid];
5139         goto out;
5140     }
5141 
5142     switch (fid) {
5143     case NVME_TEMPERATURE_THRESHOLD:
5144         result = 0;
5145 
5146         /*
5147          * The controller only implements the Composite Temperature sensor, so
5148          * return 0 for all other sensors.
5149          */
5150         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
5151             goto out;
5152         }
5153 
5154         switch (NVME_TEMP_THSEL(dw11)) {
5155         case NVME_TEMP_THSEL_OVER:
5156             result = n->features.temp_thresh_hi;
5157             goto out;
5158         case NVME_TEMP_THSEL_UNDER:
5159             result = n->features.temp_thresh_low;
5160             goto out;
5161         }
5162 
5163         return NVME_INVALID_FIELD | NVME_DNR;
5164     case NVME_ERROR_RECOVERY:
5165         if (!nvme_nsid_valid(n, nsid)) {
5166             return NVME_INVALID_NSID | NVME_DNR;
5167         }
5168 
5169         ns = nvme_ns(n, nsid);
5170         if (unlikely(!ns)) {
5171             return NVME_INVALID_FIELD | NVME_DNR;
5172         }
5173 
5174         result = ns->features.err_rec;
5175         goto out;
5176     case NVME_VOLATILE_WRITE_CACHE:
5177         result = 0;
5178         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5179             ns = nvme_ns(n, i);
5180             if (!ns) {
5181                 continue;
5182             }
5183 
5184             result = blk_enable_write_cache(ns->blkconf.blk);
5185             if (result) {
5186                 break;
5187             }
5188         }
5189         trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
5190         goto out;
5191     case NVME_ASYNCHRONOUS_EVENT_CONF:
5192         result = n->features.async_config;
5193         goto out;
5194     case NVME_TIMESTAMP:
5195         return nvme_get_feature_timestamp(n, req);
5196     case NVME_HOST_BEHAVIOR_SUPPORT:
5197         return nvme_c2h(n, (uint8_t *)&n->features.hbs,
5198                         sizeof(n->features.hbs), req);
5199     default:
5200         break;
5201     }
5202 
5203 defaults:
5204     switch (fid) {
5205     case NVME_TEMPERATURE_THRESHOLD:
5206         result = 0;
5207 
5208         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
5209             break;
5210         }
5211 
5212         if (NVME_TEMP_THSEL(dw11) == NVME_TEMP_THSEL_OVER) {
5213             result = NVME_TEMPERATURE_WARNING;
5214         }
5215 
5216         break;
5217     case NVME_NUMBER_OF_QUEUES:
5218         result = (n->params.max_ioqpairs - 1) |
5219             ((n->params.max_ioqpairs - 1) << 16);
5220         trace_pci_nvme_getfeat_numq(result);
5221         break;
5222     case NVME_INTERRUPT_VECTOR_CONF:
5223         iv = dw11 & 0xffff;
5224         if (iv >= n->params.max_ioqpairs + 1) {
5225             return NVME_INVALID_FIELD | NVME_DNR;
5226         }
5227 
5228         result = iv;
5229         if (iv == n->admin_cq.vector) {
5230             result |= NVME_INTVC_NOCOALESCING;
5231         }
5232         break;
5233     default:
5234         result = nvme_feature_default[fid];
5235         break;
5236     }
5237 
5238 out:
5239     req->cqe.result = cpu_to_le32(result);
5240     return NVME_SUCCESS;
5241 }
5242 
5243 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
5244 {
5245     uint16_t ret;
5246     uint64_t timestamp;
5247 
5248     ret = nvme_h2c(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
5249     if (ret) {
5250         return ret;
5251     }
5252 
5253     nvme_set_timestamp(n, timestamp);
5254 
5255     return NVME_SUCCESS;
5256 }
5257 
5258 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
5259 {
5260     NvmeNamespace *ns = NULL;
5261 
5262     NvmeCmd *cmd = &req->cmd;
5263     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5264     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5265     uint32_t nsid = le32_to_cpu(cmd->nsid);
5266     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
5267     uint8_t save = NVME_SETFEAT_SAVE(dw10);
5268     uint16_t status;
5269     int i;
5270 
5271     trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
5272 
5273     if (save && !(nvme_feature_cap[fid] & NVME_FEAT_CAP_SAVE)) {
5274         return NVME_FID_NOT_SAVEABLE | NVME_DNR;
5275     }
5276 
5277     if (!nvme_feature_support[fid]) {
5278         return NVME_INVALID_FIELD | NVME_DNR;
5279     }
5280 
5281     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
5282         if (nsid != NVME_NSID_BROADCAST) {
5283             if (!nvme_nsid_valid(n, nsid)) {
5284                 return NVME_INVALID_NSID | NVME_DNR;
5285             }
5286 
5287             ns = nvme_ns(n, nsid);
5288             if (unlikely(!ns)) {
5289                 return NVME_INVALID_FIELD | NVME_DNR;
5290             }
5291         }
5292     } else if (nsid && nsid != NVME_NSID_BROADCAST) {
5293         if (!nvme_nsid_valid(n, nsid)) {
5294             return NVME_INVALID_NSID | NVME_DNR;
5295         }
5296 
5297         return NVME_FEAT_NOT_NS_SPEC | NVME_DNR;
5298     }
5299 
5300     if (!(nvme_feature_cap[fid] & NVME_FEAT_CAP_CHANGE)) {
5301         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
5302     }
5303 
5304     switch (fid) {
5305     case NVME_TEMPERATURE_THRESHOLD:
5306         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
5307             break;
5308         }
5309 
5310         switch (NVME_TEMP_THSEL(dw11)) {
5311         case NVME_TEMP_THSEL_OVER:
5312             n->features.temp_thresh_hi = NVME_TEMP_TMPTH(dw11);
5313             break;
5314         case NVME_TEMP_THSEL_UNDER:
5315             n->features.temp_thresh_low = NVME_TEMP_TMPTH(dw11);
5316             break;
5317         default:
5318             return NVME_INVALID_FIELD | NVME_DNR;
5319         }
5320 
5321         if ((n->temperature >= n->features.temp_thresh_hi) ||
5322             (n->temperature <= n->features.temp_thresh_low)) {
5323             nvme_smart_event(n, NVME_AER_INFO_SMART_TEMP_THRESH);
5324         }
5325 
5326         break;
5327     case NVME_ERROR_RECOVERY:
5328         if (nsid == NVME_NSID_BROADCAST) {
5329             for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5330                 ns = nvme_ns(n, i);
5331 
5332                 if (!ns) {
5333                     continue;
5334                 }
5335 
5336                 if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat)) {
5337                     ns->features.err_rec = dw11;
5338                 }
5339             }
5340 
5341             break;
5342         }
5343 
5344         assert(ns);
5345         if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat))  {
5346             ns->features.err_rec = dw11;
5347         }
5348         break;
5349     case NVME_VOLATILE_WRITE_CACHE:
5350         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5351             ns = nvme_ns(n, i);
5352             if (!ns) {
5353                 continue;
5354             }
5355 
5356             if (!(dw11 & 0x1) && blk_enable_write_cache(ns->blkconf.blk)) {
5357                 blk_flush(ns->blkconf.blk);
5358             }
5359 
5360             blk_set_enable_write_cache(ns->blkconf.blk, dw11 & 1);
5361         }
5362 
5363         break;
5364 
5365     case NVME_NUMBER_OF_QUEUES:
5366         if (n->qs_created) {
5367             return NVME_CMD_SEQ_ERROR | NVME_DNR;
5368         }
5369 
5370         /*
5371          * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
5372          * and NSQR.
5373          */
5374         if ((dw11 & 0xffff) == 0xffff || ((dw11 >> 16) & 0xffff) == 0xffff) {
5375             return NVME_INVALID_FIELD | NVME_DNR;
5376         }
5377 
5378         trace_pci_nvme_setfeat_numq((dw11 & 0xffff) + 1,
5379                                     ((dw11 >> 16) & 0xffff) + 1,
5380                                     n->params.max_ioqpairs,
5381                                     n->params.max_ioqpairs);
5382         req->cqe.result = cpu_to_le32((n->params.max_ioqpairs - 1) |
5383                                       ((n->params.max_ioqpairs - 1) << 16));
5384         break;
5385     case NVME_ASYNCHRONOUS_EVENT_CONF:
5386         n->features.async_config = dw11;
5387         break;
5388     case NVME_TIMESTAMP:
5389         return nvme_set_feature_timestamp(n, req);
5390     case NVME_HOST_BEHAVIOR_SUPPORT:
5391         status = nvme_h2c(n, (uint8_t *)&n->features.hbs,
5392                           sizeof(n->features.hbs), req);
5393         if (status) {
5394             return status;
5395         }
5396 
5397         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5398             ns = nvme_ns(n, i);
5399 
5400             if (!ns) {
5401                 continue;
5402             }
5403 
5404             ns->id_ns.nlbaf = ns->nlbaf - 1;
5405             if (!n->features.hbs.lbafee) {
5406                 ns->id_ns.nlbaf = MIN(ns->id_ns.nlbaf, 15);
5407             }
5408         }
5409 
5410         return status;
5411     case NVME_COMMAND_SET_PROFILE:
5412         if (dw11 & 0x1ff) {
5413             trace_pci_nvme_err_invalid_iocsci(dw11 & 0x1ff);
5414             return NVME_CMD_SET_CMB_REJECTED | NVME_DNR;
5415         }
5416         break;
5417     default:
5418         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
5419     }
5420     return NVME_SUCCESS;
5421 }
5422 
5423 static uint16_t nvme_aer(NvmeCtrl *n, NvmeRequest *req)
5424 {
5425     trace_pci_nvme_aer(nvme_cid(req));
5426 
5427     if (n->outstanding_aers > n->params.aerl) {
5428         trace_pci_nvme_aer_aerl_exceeded();
5429         return NVME_AER_LIMIT_EXCEEDED;
5430     }
5431 
5432     n->aer_reqs[n->outstanding_aers] = req;
5433     n->outstanding_aers++;
5434 
5435     if (!QTAILQ_EMPTY(&n->aer_queue)) {
5436         nvme_process_aers(n);
5437     }
5438 
5439     return NVME_NO_COMPLETE;
5440 }
5441 
5442 static void nvme_update_dmrsl(NvmeCtrl *n)
5443 {
5444     int nsid;
5445 
5446     for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
5447         NvmeNamespace *ns = nvme_ns(n, nsid);
5448         if (!ns) {
5449             continue;
5450         }
5451 
5452         n->dmrsl = MIN_NON_ZERO(n->dmrsl,
5453                                 BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
5454     }
5455 }
5456 
5457 static void nvme_select_iocs_ns(NvmeCtrl *n, NvmeNamespace *ns)
5458 {
5459     uint32_t cc = ldl_le_p(&n->bar.cc);
5460 
5461     ns->iocs = nvme_cse_iocs_none;
5462     switch (ns->csi) {
5463     case NVME_CSI_NVM:
5464         if (NVME_CC_CSS(cc) != NVME_CC_CSS_ADMIN_ONLY) {
5465             ns->iocs = nvme_cse_iocs_nvm;
5466         }
5467         break;
5468     case NVME_CSI_ZONED:
5469         if (NVME_CC_CSS(cc) == NVME_CC_CSS_CSI) {
5470             ns->iocs = nvme_cse_iocs_zoned;
5471         } else if (NVME_CC_CSS(cc) == NVME_CC_CSS_NVM) {
5472             ns->iocs = nvme_cse_iocs_nvm;
5473         }
5474         break;
5475     }
5476 }
5477 
5478 static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
5479 {
5480     NvmeNamespace *ns;
5481     NvmeCtrl *ctrl;
5482     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5483     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
5484     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
5485     uint8_t sel = dw10 & 0xf;
5486     uint16_t *nr_ids = &list[0];
5487     uint16_t *ids = &list[1];
5488     uint16_t ret;
5489     int i;
5490 
5491     trace_pci_nvme_ns_attachment(nvme_cid(req), dw10 & 0xf);
5492 
5493     if (!nvme_nsid_valid(n, nsid)) {
5494         return NVME_INVALID_NSID | NVME_DNR;
5495     }
5496 
5497     ns = nvme_subsys_ns(n->subsys, nsid);
5498     if (!ns) {
5499         return NVME_INVALID_FIELD | NVME_DNR;
5500     }
5501 
5502     ret = nvme_h2c(n, (uint8_t *)list, 4096, req);
5503     if (ret) {
5504         return ret;
5505     }
5506 
5507     if (!*nr_ids) {
5508         return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
5509     }
5510 
5511     *nr_ids = MIN(*nr_ids, NVME_CONTROLLER_LIST_SIZE - 1);
5512     for (i = 0; i < *nr_ids; i++) {
5513         ctrl = nvme_subsys_ctrl(n->subsys, ids[i]);
5514         if (!ctrl) {
5515             return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
5516         }
5517 
5518         switch (sel) {
5519         case NVME_NS_ATTACHMENT_ATTACH:
5520             if (nvme_ns(ctrl, nsid)) {
5521                 return NVME_NS_ALREADY_ATTACHED | NVME_DNR;
5522             }
5523 
5524             if (ns->attached && !ns->params.shared) {
5525                 return NVME_NS_PRIVATE | NVME_DNR;
5526             }
5527 
5528             nvme_attach_ns(ctrl, ns);
5529             nvme_select_iocs_ns(ctrl, ns);
5530 
5531             break;
5532 
5533         case NVME_NS_ATTACHMENT_DETACH:
5534             if (!nvme_ns(ctrl, nsid)) {
5535                 return NVME_NS_NOT_ATTACHED | NVME_DNR;
5536             }
5537 
5538             ctrl->namespaces[nsid] = NULL;
5539             ns->attached--;
5540 
5541             nvme_update_dmrsl(ctrl);
5542 
5543             break;
5544 
5545         default:
5546             return NVME_INVALID_FIELD | NVME_DNR;
5547         }
5548 
5549         /*
5550          * Add namespace id to the changed namespace id list for event clearing
5551          * via Get Log Page command.
5552          */
5553         if (!test_and_set_bit(nsid, ctrl->changed_nsids)) {
5554             nvme_enqueue_event(ctrl, NVME_AER_TYPE_NOTICE,
5555                                NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED,
5556                                NVME_LOG_CHANGED_NSLIST);
5557         }
5558     }
5559 
5560     return NVME_SUCCESS;
5561 }
5562 
5563 typedef struct NvmeFormatAIOCB {
5564     BlockAIOCB common;
5565     BlockAIOCB *aiocb;
5566     QEMUBH *bh;
5567     NvmeRequest *req;
5568     int ret;
5569 
5570     NvmeNamespace *ns;
5571     uint32_t nsid;
5572     bool broadcast;
5573     int64_t offset;
5574 
5575     uint8_t lbaf;
5576     uint8_t mset;
5577     uint8_t pi;
5578     uint8_t pil;
5579 } NvmeFormatAIOCB;
5580 
5581 static void nvme_format_bh(void *opaque);
5582 
5583 static void nvme_format_cancel(BlockAIOCB *aiocb)
5584 {
5585     NvmeFormatAIOCB *iocb = container_of(aiocb, NvmeFormatAIOCB, common);
5586 
5587     if (iocb->aiocb) {
5588         blk_aio_cancel_async(iocb->aiocb);
5589     }
5590 }
5591 
5592 static const AIOCBInfo nvme_format_aiocb_info = {
5593     .aiocb_size = sizeof(NvmeFormatAIOCB),
5594     .cancel_async = nvme_format_cancel,
5595     .get_aio_context = nvme_get_aio_context,
5596 };
5597 
5598 static void nvme_format_set(NvmeNamespace *ns, uint8_t lbaf, uint8_t mset,
5599                             uint8_t pi, uint8_t pil)
5600 {
5601     uint8_t lbafl = lbaf & 0xf;
5602     uint8_t lbafu = lbaf >> 4;
5603 
5604     trace_pci_nvme_format_set(ns->params.nsid, lbaf, mset, pi, pil);
5605 
5606     ns->id_ns.dps = (pil << 3) | pi;
5607     ns->id_ns.flbas = (lbafu << 5) | (mset << 4) | lbafl;
5608 
5609     nvme_ns_init_format(ns);
5610 }
5611 
5612 static void nvme_format_ns_cb(void *opaque, int ret)
5613 {
5614     NvmeFormatAIOCB *iocb = opaque;
5615     NvmeNamespace *ns = iocb->ns;
5616     int bytes;
5617 
5618     if (ret < 0) {
5619         iocb->ret = ret;
5620         goto done;
5621     }
5622 
5623     assert(ns);
5624 
5625     if (iocb->offset < ns->size) {
5626         bytes = MIN(BDRV_REQUEST_MAX_BYTES, ns->size - iocb->offset);
5627 
5628         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, iocb->offset,
5629                                             bytes, BDRV_REQ_MAY_UNMAP,
5630                                             nvme_format_ns_cb, iocb);
5631 
5632         iocb->offset += bytes;
5633         return;
5634     }
5635 
5636     nvme_format_set(ns, iocb->lbaf, iocb->mset, iocb->pi, iocb->pil);
5637     ns->status = 0x0;
5638     iocb->ns = NULL;
5639     iocb->offset = 0;
5640 
5641 done:
5642     iocb->aiocb = NULL;
5643     qemu_bh_schedule(iocb->bh);
5644 }
5645 
5646 static uint16_t nvme_format_check(NvmeNamespace *ns, uint8_t lbaf, uint8_t pi)
5647 {
5648     if (ns->params.zoned) {
5649         return NVME_INVALID_FORMAT | NVME_DNR;
5650     }
5651 
5652     if (lbaf > ns->id_ns.nlbaf) {
5653         return NVME_INVALID_FORMAT | NVME_DNR;
5654     }
5655 
5656     if (pi && (ns->id_ns.lbaf[lbaf].ms < nvme_pi_tuple_size(ns))) {
5657         return NVME_INVALID_FORMAT | NVME_DNR;
5658     }
5659 
5660     if (pi && pi > NVME_ID_NS_DPS_TYPE_3) {
5661         return NVME_INVALID_FIELD | NVME_DNR;
5662     }
5663 
5664     return NVME_SUCCESS;
5665 }
5666 
5667 static void nvme_format_bh(void *opaque)
5668 {
5669     NvmeFormatAIOCB *iocb = opaque;
5670     NvmeRequest *req = iocb->req;
5671     NvmeCtrl *n = nvme_ctrl(req);
5672     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
5673     uint8_t lbaf = dw10 & 0xf;
5674     uint8_t pi = (dw10 >> 5) & 0x7;
5675     uint16_t status;
5676     int i;
5677 
5678     if (iocb->ret < 0) {
5679         goto done;
5680     }
5681 
5682     if (iocb->broadcast) {
5683         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
5684             iocb->ns = nvme_ns(n, i);
5685             if (iocb->ns) {
5686                 iocb->nsid = i;
5687                 break;
5688             }
5689         }
5690     }
5691 
5692     if (!iocb->ns) {
5693         goto done;
5694     }
5695 
5696     status = nvme_format_check(iocb->ns, lbaf, pi);
5697     if (status) {
5698         req->status = status;
5699         goto done;
5700     }
5701 
5702     iocb->ns->status = NVME_FORMAT_IN_PROGRESS;
5703     nvme_format_ns_cb(iocb, 0);
5704     return;
5705 
5706 done:
5707     qemu_bh_delete(iocb->bh);
5708     iocb->bh = NULL;
5709 
5710     iocb->common.cb(iocb->common.opaque, iocb->ret);
5711 
5712     qemu_aio_unref(iocb);
5713 }
5714 
5715 static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
5716 {
5717     NvmeFormatAIOCB *iocb;
5718     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
5719     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
5720     uint8_t lbaf = dw10 & 0xf;
5721     uint8_t mset = (dw10 >> 4) & 0x1;
5722     uint8_t pi = (dw10 >> 5) & 0x7;
5723     uint8_t pil = (dw10 >> 8) & 0x1;
5724     uint8_t lbafu = (dw10 >> 12) & 0x3;
5725     uint16_t status;
5726 
5727     iocb = qemu_aio_get(&nvme_format_aiocb_info, NULL, nvme_misc_cb, req);
5728 
5729     iocb->req = req;
5730     iocb->bh = qemu_bh_new(nvme_format_bh, iocb);
5731     iocb->ret = 0;
5732     iocb->ns = NULL;
5733     iocb->nsid = 0;
5734     iocb->lbaf = lbaf;
5735     iocb->mset = mset;
5736     iocb->pi = pi;
5737     iocb->pil = pil;
5738     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
5739     iocb->offset = 0;
5740 
5741     if (n->features.hbs.lbafee) {
5742         iocb->lbaf |= lbafu << 4;
5743     }
5744 
5745     if (!iocb->broadcast) {
5746         if (!nvme_nsid_valid(n, nsid)) {
5747             status = NVME_INVALID_NSID | NVME_DNR;
5748             goto out;
5749         }
5750 
5751         iocb->ns = nvme_ns(n, nsid);
5752         if (!iocb->ns) {
5753             status = NVME_INVALID_FIELD | NVME_DNR;
5754             goto out;
5755         }
5756     }
5757 
5758     req->aiocb = &iocb->common;
5759     qemu_bh_schedule(iocb->bh);
5760 
5761     return NVME_NO_COMPLETE;
5762 
5763 out:
5764     qemu_bh_delete(iocb->bh);
5765     iocb->bh = NULL;
5766     qemu_aio_unref(iocb);
5767     return status;
5768 }
5769 
5770 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
5771 {
5772     trace_pci_nvme_admin_cmd(nvme_cid(req), nvme_sqid(req), req->cmd.opcode,
5773                              nvme_adm_opc_str(req->cmd.opcode));
5774 
5775     if (!(nvme_cse_acs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
5776         trace_pci_nvme_err_invalid_admin_opc(req->cmd.opcode);
5777         return NVME_INVALID_OPCODE | NVME_DNR;
5778     }
5779 
5780     /* SGLs shall not be used for Admin commands in NVMe over PCIe */
5781     if (NVME_CMD_FLAGS_PSDT(req->cmd.flags) != NVME_PSDT_PRP) {
5782         return NVME_INVALID_FIELD | NVME_DNR;
5783     }
5784 
5785     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
5786         return NVME_INVALID_FIELD;
5787     }
5788 
5789     switch (req->cmd.opcode) {
5790     case NVME_ADM_CMD_DELETE_SQ:
5791         return nvme_del_sq(n, req);
5792     case NVME_ADM_CMD_CREATE_SQ:
5793         return nvme_create_sq(n, req);
5794     case NVME_ADM_CMD_GET_LOG_PAGE:
5795         return nvme_get_log(n, req);
5796     case NVME_ADM_CMD_DELETE_CQ:
5797         return nvme_del_cq(n, req);
5798     case NVME_ADM_CMD_CREATE_CQ:
5799         return nvme_create_cq(n, req);
5800     case NVME_ADM_CMD_IDENTIFY:
5801         return nvme_identify(n, req);
5802     case NVME_ADM_CMD_ABORT:
5803         return nvme_abort(n, req);
5804     case NVME_ADM_CMD_SET_FEATURES:
5805         return nvme_set_feature(n, req);
5806     case NVME_ADM_CMD_GET_FEATURES:
5807         return nvme_get_feature(n, req);
5808     case NVME_ADM_CMD_ASYNC_EV_REQ:
5809         return nvme_aer(n, req);
5810     case NVME_ADM_CMD_NS_ATTACHMENT:
5811         return nvme_ns_attachment(n, req);
5812     case NVME_ADM_CMD_FORMAT_NVM:
5813         return nvme_format(n, req);
5814     default:
5815         assert(false);
5816     }
5817 
5818     return NVME_INVALID_OPCODE | NVME_DNR;
5819 }
5820 
5821 static void nvme_process_sq(void *opaque)
5822 {
5823     NvmeSQueue *sq = opaque;
5824     NvmeCtrl *n = sq->ctrl;
5825     NvmeCQueue *cq = n->cq[sq->cqid];
5826 
5827     uint16_t status;
5828     hwaddr addr;
5829     NvmeCmd cmd;
5830     NvmeRequest *req;
5831 
5832     while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
5833         addr = sq->dma_addr + sq->head * n->sqe_size;
5834         if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
5835             trace_pci_nvme_err_addr_read(addr);
5836             trace_pci_nvme_err_cfs();
5837             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
5838             break;
5839         }
5840         nvme_inc_sq_head(sq);
5841 
5842         req = QTAILQ_FIRST(&sq->req_list);
5843         QTAILQ_REMOVE(&sq->req_list, req, entry);
5844         QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
5845         nvme_req_clear(req);
5846         req->cqe.cid = cmd.cid;
5847         memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
5848 
5849         status = sq->sqid ? nvme_io_cmd(n, req) :
5850             nvme_admin_cmd(n, req);
5851         if (status != NVME_NO_COMPLETE) {
5852             req->status = status;
5853             nvme_enqueue_req_completion(cq, req);
5854         }
5855     }
5856 }
5857 
5858 static void nvme_ctrl_reset(NvmeCtrl *n)
5859 {
5860     NvmeNamespace *ns;
5861     int i;
5862 
5863     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5864         ns = nvme_ns(n, i);
5865         if (!ns) {
5866             continue;
5867         }
5868 
5869         nvme_ns_drain(ns);
5870     }
5871 
5872     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
5873         if (n->sq[i] != NULL) {
5874             nvme_free_sq(n->sq[i], n);
5875         }
5876     }
5877     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
5878         if (n->cq[i] != NULL) {
5879             nvme_free_cq(n->cq[i], n);
5880         }
5881     }
5882 
5883     while (!QTAILQ_EMPTY(&n->aer_queue)) {
5884         NvmeAsyncEvent *event = QTAILQ_FIRST(&n->aer_queue);
5885         QTAILQ_REMOVE(&n->aer_queue, event, entry);
5886         g_free(event);
5887     }
5888 
5889     n->aer_queued = 0;
5890     n->outstanding_aers = 0;
5891     n->qs_created = false;
5892 }
5893 
5894 static void nvme_ctrl_shutdown(NvmeCtrl *n)
5895 {
5896     NvmeNamespace *ns;
5897     int i;
5898 
5899     if (n->pmr.dev) {
5900         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
5901     }
5902 
5903     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5904         ns = nvme_ns(n, i);
5905         if (!ns) {
5906             continue;
5907         }
5908 
5909         nvme_ns_shutdown(ns);
5910     }
5911 }
5912 
5913 static void nvme_select_iocs(NvmeCtrl *n)
5914 {
5915     NvmeNamespace *ns;
5916     int i;
5917 
5918     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5919         ns = nvme_ns(n, i);
5920         if (!ns) {
5921             continue;
5922         }
5923 
5924         nvme_select_iocs_ns(n, ns);
5925     }
5926 }
5927 
5928 static int nvme_start_ctrl(NvmeCtrl *n)
5929 {
5930     uint64_t cap = ldq_le_p(&n->bar.cap);
5931     uint32_t cc = ldl_le_p(&n->bar.cc);
5932     uint32_t aqa = ldl_le_p(&n->bar.aqa);
5933     uint64_t asq = ldq_le_p(&n->bar.asq);
5934     uint64_t acq = ldq_le_p(&n->bar.acq);
5935     uint32_t page_bits = NVME_CC_MPS(cc) + 12;
5936     uint32_t page_size = 1 << page_bits;
5937 
5938     if (unlikely(n->cq[0])) {
5939         trace_pci_nvme_err_startfail_cq();
5940         return -1;
5941     }
5942     if (unlikely(n->sq[0])) {
5943         trace_pci_nvme_err_startfail_sq();
5944         return -1;
5945     }
5946     if (unlikely(asq & (page_size - 1))) {
5947         trace_pci_nvme_err_startfail_asq_misaligned(asq);
5948         return -1;
5949     }
5950     if (unlikely(acq & (page_size - 1))) {
5951         trace_pci_nvme_err_startfail_acq_misaligned(acq);
5952         return -1;
5953     }
5954     if (unlikely(!(NVME_CAP_CSS(cap) & (1 << NVME_CC_CSS(cc))))) {
5955         trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc));
5956         return -1;
5957     }
5958     if (unlikely(NVME_CC_MPS(cc) < NVME_CAP_MPSMIN(cap))) {
5959         trace_pci_nvme_err_startfail_page_too_small(
5960                     NVME_CC_MPS(cc),
5961                     NVME_CAP_MPSMIN(cap));
5962         return -1;
5963     }
5964     if (unlikely(NVME_CC_MPS(cc) >
5965                  NVME_CAP_MPSMAX(cap))) {
5966         trace_pci_nvme_err_startfail_page_too_large(
5967                     NVME_CC_MPS(cc),
5968                     NVME_CAP_MPSMAX(cap));
5969         return -1;
5970     }
5971     if (unlikely(NVME_CC_IOCQES(cc) <
5972                  NVME_CTRL_CQES_MIN(n->id_ctrl.cqes))) {
5973         trace_pci_nvme_err_startfail_cqent_too_small(
5974                     NVME_CC_IOCQES(cc),
5975                     NVME_CTRL_CQES_MIN(cap));
5976         return -1;
5977     }
5978     if (unlikely(NVME_CC_IOCQES(cc) >
5979                  NVME_CTRL_CQES_MAX(n->id_ctrl.cqes))) {
5980         trace_pci_nvme_err_startfail_cqent_too_large(
5981                     NVME_CC_IOCQES(cc),
5982                     NVME_CTRL_CQES_MAX(cap));
5983         return -1;
5984     }
5985     if (unlikely(NVME_CC_IOSQES(cc) <
5986                  NVME_CTRL_SQES_MIN(n->id_ctrl.sqes))) {
5987         trace_pci_nvme_err_startfail_sqent_too_small(
5988                     NVME_CC_IOSQES(cc),
5989                     NVME_CTRL_SQES_MIN(cap));
5990         return -1;
5991     }
5992     if (unlikely(NVME_CC_IOSQES(cc) >
5993                  NVME_CTRL_SQES_MAX(n->id_ctrl.sqes))) {
5994         trace_pci_nvme_err_startfail_sqent_too_large(
5995                     NVME_CC_IOSQES(cc),
5996                     NVME_CTRL_SQES_MAX(cap));
5997         return -1;
5998     }
5999     if (unlikely(!NVME_AQA_ASQS(aqa))) {
6000         trace_pci_nvme_err_startfail_asqent_sz_zero();
6001         return -1;
6002     }
6003     if (unlikely(!NVME_AQA_ACQS(aqa))) {
6004         trace_pci_nvme_err_startfail_acqent_sz_zero();
6005         return -1;
6006     }
6007 
6008     n->page_bits = page_bits;
6009     n->page_size = page_size;
6010     n->max_prp_ents = n->page_size / sizeof(uint64_t);
6011     n->cqe_size = 1 << NVME_CC_IOCQES(cc);
6012     n->sqe_size = 1 << NVME_CC_IOSQES(cc);
6013     nvme_init_cq(&n->admin_cq, n, acq, 0, 0, NVME_AQA_ACQS(aqa) + 1, 1);
6014     nvme_init_sq(&n->admin_sq, n, asq, 0, 0, NVME_AQA_ASQS(aqa) + 1);
6015 
6016     nvme_set_timestamp(n, 0ULL);
6017 
6018     QTAILQ_INIT(&n->aer_queue);
6019 
6020     nvme_select_iocs(n);
6021 
6022     return 0;
6023 }
6024 
6025 static void nvme_cmb_enable_regs(NvmeCtrl *n)
6026 {
6027     uint32_t cmbloc = ldl_le_p(&n->bar.cmbloc);
6028     uint32_t cmbsz = ldl_le_p(&n->bar.cmbsz);
6029 
6030     NVME_CMBLOC_SET_CDPCILS(cmbloc, 1);
6031     NVME_CMBLOC_SET_CDPMLS(cmbloc, 1);
6032     NVME_CMBLOC_SET_BIR(cmbloc, NVME_CMB_BIR);
6033     stl_le_p(&n->bar.cmbloc, cmbloc);
6034 
6035     NVME_CMBSZ_SET_SQS(cmbsz, 1);
6036     NVME_CMBSZ_SET_CQS(cmbsz, 0);
6037     NVME_CMBSZ_SET_LISTS(cmbsz, 1);
6038     NVME_CMBSZ_SET_RDS(cmbsz, 1);
6039     NVME_CMBSZ_SET_WDS(cmbsz, 1);
6040     NVME_CMBSZ_SET_SZU(cmbsz, 2); /* MBs */
6041     NVME_CMBSZ_SET_SZ(cmbsz, n->params.cmb_size_mb);
6042     stl_le_p(&n->bar.cmbsz, cmbsz);
6043 }
6044 
6045 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
6046                            unsigned size)
6047 {
6048     uint64_t cap = ldq_le_p(&n->bar.cap);
6049     uint32_t cc = ldl_le_p(&n->bar.cc);
6050     uint32_t intms = ldl_le_p(&n->bar.intms);
6051     uint32_t csts = ldl_le_p(&n->bar.csts);
6052     uint32_t pmrsts = ldl_le_p(&n->bar.pmrsts);
6053 
6054     if (unlikely(offset & (sizeof(uint32_t) - 1))) {
6055         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
6056                        "MMIO write not 32-bit aligned,"
6057                        " offset=0x%"PRIx64"", offset);
6058         /* should be ignored, fall through for now */
6059     }
6060 
6061     if (unlikely(size < sizeof(uint32_t))) {
6062         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
6063                        "MMIO write smaller than 32-bits,"
6064                        " offset=0x%"PRIx64", size=%u",
6065                        offset, size);
6066         /* should be ignored, fall through for now */
6067     }
6068 
6069     switch (offset) {
6070     case NVME_REG_INTMS:
6071         if (unlikely(msix_enabled(&(n->parent_obj)))) {
6072             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
6073                            "undefined access to interrupt mask set"
6074                            " when MSI-X is enabled");
6075             /* should be ignored, fall through for now */
6076         }
6077         intms |= data;
6078         stl_le_p(&n->bar.intms, intms);
6079         n->bar.intmc = n->bar.intms;
6080         trace_pci_nvme_mmio_intm_set(data & 0xffffffff, intms);
6081         nvme_irq_check(n);
6082         break;
6083     case NVME_REG_INTMC:
6084         if (unlikely(msix_enabled(&(n->parent_obj)))) {
6085             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
6086                            "undefined access to interrupt mask clr"
6087                            " when MSI-X is enabled");
6088             /* should be ignored, fall through for now */
6089         }
6090         intms &= ~data;
6091         stl_le_p(&n->bar.intms, intms);
6092         n->bar.intmc = n->bar.intms;
6093         trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, intms);
6094         nvme_irq_check(n);
6095         break;
6096     case NVME_REG_CC:
6097         trace_pci_nvme_mmio_cfg(data & 0xffffffff);
6098 
6099         /* Windows first sends data, then sends enable bit */
6100         if (!NVME_CC_EN(data) && !NVME_CC_EN(cc) &&
6101             !NVME_CC_SHN(data) && !NVME_CC_SHN(cc))
6102         {
6103             cc = data;
6104         }
6105 
6106         if (NVME_CC_EN(data) && !NVME_CC_EN(cc)) {
6107             cc = data;
6108 
6109             /* flush CC since nvme_start_ctrl() needs the value */
6110             stl_le_p(&n->bar.cc, cc);
6111             if (unlikely(nvme_start_ctrl(n))) {
6112                 trace_pci_nvme_err_startfail();
6113                 csts = NVME_CSTS_FAILED;
6114             } else {
6115                 trace_pci_nvme_mmio_start_success();
6116                 csts = NVME_CSTS_READY;
6117             }
6118         } else if (!NVME_CC_EN(data) && NVME_CC_EN(cc)) {
6119             trace_pci_nvme_mmio_stopped();
6120             nvme_ctrl_reset(n);
6121             cc = 0;
6122             csts &= ~NVME_CSTS_READY;
6123         }
6124 
6125         if (NVME_CC_SHN(data) && !(NVME_CC_SHN(cc))) {
6126             trace_pci_nvme_mmio_shutdown_set();
6127             nvme_ctrl_shutdown(n);
6128             cc = data;
6129             csts |= NVME_CSTS_SHST_COMPLETE;
6130         } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(cc)) {
6131             trace_pci_nvme_mmio_shutdown_cleared();
6132             csts &= ~NVME_CSTS_SHST_COMPLETE;
6133             cc = data;
6134         }
6135 
6136         stl_le_p(&n->bar.cc, cc);
6137         stl_le_p(&n->bar.csts, csts);
6138 
6139         break;
6140     case NVME_REG_CSTS:
6141         if (data & (1 << 4)) {
6142             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
6143                            "attempted to W1C CSTS.NSSRO"
6144                            " but CAP.NSSRS is zero (not supported)");
6145         } else if (data != 0) {
6146             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
6147                            "attempted to set a read only bit"
6148                            " of controller status");
6149         }
6150         break;
6151     case NVME_REG_NSSR:
6152         if (data == 0x4e564d65) {
6153             trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
6154         } else {
6155             /* The spec says that writes of other values have no effect */
6156             return;
6157         }
6158         break;
6159     case NVME_REG_AQA:
6160         stl_le_p(&n->bar.aqa, data);
6161         trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
6162         break;
6163     case NVME_REG_ASQ:
6164         stn_le_p(&n->bar.asq, size, data);
6165         trace_pci_nvme_mmio_asqaddr(data);
6166         break;
6167     case NVME_REG_ASQ + 4:
6168         stl_le_p((uint8_t *)&n->bar.asq + 4, data);
6169         trace_pci_nvme_mmio_asqaddr_hi(data, ldq_le_p(&n->bar.asq));
6170         break;
6171     case NVME_REG_ACQ:
6172         trace_pci_nvme_mmio_acqaddr(data);
6173         stn_le_p(&n->bar.acq, size, data);
6174         break;
6175     case NVME_REG_ACQ + 4:
6176         stl_le_p((uint8_t *)&n->bar.acq + 4, data);
6177         trace_pci_nvme_mmio_acqaddr_hi(data, ldq_le_p(&n->bar.acq));
6178         break;
6179     case NVME_REG_CMBLOC:
6180         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
6181                        "invalid write to reserved CMBLOC"
6182                        " when CMBSZ is zero, ignored");
6183         return;
6184     case NVME_REG_CMBSZ:
6185         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
6186                        "invalid write to read only CMBSZ, ignored");
6187         return;
6188     case NVME_REG_CMBMSC:
6189         if (!NVME_CAP_CMBS(cap)) {
6190             return;
6191         }
6192 
6193         stn_le_p(&n->bar.cmbmsc, size, data);
6194         n->cmb.cmse = false;
6195 
6196         if (NVME_CMBMSC_CRE(data)) {
6197             nvme_cmb_enable_regs(n);
6198 
6199             if (NVME_CMBMSC_CMSE(data)) {
6200                 uint64_t cmbmsc = ldq_le_p(&n->bar.cmbmsc);
6201                 hwaddr cba = NVME_CMBMSC_CBA(cmbmsc) << CMBMSC_CBA_SHIFT;
6202                 if (cba + int128_get64(n->cmb.mem.size) < cba) {
6203                     uint32_t cmbsts = ldl_le_p(&n->bar.cmbsts);
6204                     NVME_CMBSTS_SET_CBAI(cmbsts, 1);
6205                     stl_le_p(&n->bar.cmbsts, cmbsts);
6206                     return;
6207                 }
6208 
6209                 n->cmb.cba = cba;
6210                 n->cmb.cmse = true;
6211             }
6212         } else {
6213             n->bar.cmbsz = 0;
6214             n->bar.cmbloc = 0;
6215         }
6216 
6217         return;
6218     case NVME_REG_CMBMSC + 4:
6219         stl_le_p((uint8_t *)&n->bar.cmbmsc + 4, data);
6220         return;
6221 
6222     case NVME_REG_PMRCAP:
6223         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
6224                        "invalid write to PMRCAP register, ignored");
6225         return;
6226     case NVME_REG_PMRCTL:
6227         if (!NVME_CAP_PMRS(cap)) {
6228             return;
6229         }
6230 
6231         stl_le_p(&n->bar.pmrctl, data);
6232         if (NVME_PMRCTL_EN(data)) {
6233             memory_region_set_enabled(&n->pmr.dev->mr, true);
6234             pmrsts = 0;
6235         } else {
6236             memory_region_set_enabled(&n->pmr.dev->mr, false);
6237             NVME_PMRSTS_SET_NRDY(pmrsts, 1);
6238             n->pmr.cmse = false;
6239         }
6240         stl_le_p(&n->bar.pmrsts, pmrsts);
6241         return;
6242     case NVME_REG_PMRSTS:
6243         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
6244                        "invalid write to PMRSTS register, ignored");
6245         return;
6246     case NVME_REG_PMREBS:
6247         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
6248                        "invalid write to PMREBS register, ignored");
6249         return;
6250     case NVME_REG_PMRSWTP:
6251         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
6252                        "invalid write to PMRSWTP register, ignored");
6253         return;
6254     case NVME_REG_PMRMSCL:
6255         if (!NVME_CAP_PMRS(cap)) {
6256             return;
6257         }
6258 
6259         stl_le_p(&n->bar.pmrmscl, data);
6260         n->pmr.cmse = false;
6261 
6262         if (NVME_PMRMSCL_CMSE(data)) {
6263             uint64_t pmrmscu = ldl_le_p(&n->bar.pmrmscu);
6264             hwaddr cba = pmrmscu << 32 |
6265                 (NVME_PMRMSCL_CBA(data) << PMRMSCL_CBA_SHIFT);
6266             if (cba + int128_get64(n->pmr.dev->mr.size) < cba) {
6267                 NVME_PMRSTS_SET_CBAI(pmrsts, 1);
6268                 stl_le_p(&n->bar.pmrsts, pmrsts);
6269                 return;
6270             }
6271 
6272             n->pmr.cmse = true;
6273             n->pmr.cba = cba;
6274         }
6275 
6276         return;
6277     case NVME_REG_PMRMSCU:
6278         if (!NVME_CAP_PMRS(cap)) {
6279             return;
6280         }
6281 
6282         stl_le_p(&n->bar.pmrmscu, data);
6283         return;
6284     default:
6285         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
6286                        "invalid MMIO write,"
6287                        " offset=0x%"PRIx64", data=%"PRIx64"",
6288                        offset, data);
6289         break;
6290     }
6291 }
6292 
6293 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
6294 {
6295     NvmeCtrl *n = (NvmeCtrl *)opaque;
6296     uint8_t *ptr = (uint8_t *)&n->bar;
6297 
6298     trace_pci_nvme_mmio_read(addr, size);
6299 
6300     if (unlikely(addr & (sizeof(uint32_t) - 1))) {
6301         NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
6302                        "MMIO read not 32-bit aligned,"
6303                        " offset=0x%"PRIx64"", addr);
6304         /* should RAZ, fall through for now */
6305     } else if (unlikely(size < sizeof(uint32_t))) {
6306         NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
6307                        "MMIO read smaller than 32-bits,"
6308                        " offset=0x%"PRIx64"", addr);
6309         /* should RAZ, fall through for now */
6310     }
6311 
6312     if (addr > sizeof(n->bar) - size) {
6313         NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
6314                        "MMIO read beyond last register,"
6315                        " offset=0x%"PRIx64", returning 0", addr);
6316 
6317         return 0;
6318     }
6319 
6320     /*
6321      * When PMRWBM bit 1 is set then read from
6322      * from PMRSTS should ensure prior writes
6323      * made it to persistent media
6324      */
6325     if (addr == NVME_REG_PMRSTS &&
6326         (NVME_PMRCAP_PMRWBM(ldl_le_p(&n->bar.pmrcap)) & 0x02)) {
6327         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
6328     }
6329 
6330     return ldn_le_p(ptr + addr, size);
6331 }
6332 
6333 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
6334 {
6335     uint32_t qid;
6336 
6337     if (unlikely(addr & ((1 << 2) - 1))) {
6338         NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
6339                        "doorbell write not 32-bit aligned,"
6340                        " offset=0x%"PRIx64", ignoring", addr);
6341         return;
6342     }
6343 
6344     if (((addr - 0x1000) >> 2) & 1) {
6345         /* Completion queue doorbell write */
6346 
6347         uint16_t new_head = val & 0xffff;
6348         int start_sqs;
6349         NvmeCQueue *cq;
6350 
6351         qid = (addr - (0x1000 + (1 << 2))) >> 3;
6352         if (unlikely(nvme_check_cqid(n, qid))) {
6353             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
6354                            "completion queue doorbell write"
6355                            " for nonexistent queue,"
6356                            " sqid=%"PRIu32", ignoring", qid);
6357 
6358             /*
6359              * NVM Express v1.3d, Section 4.1 state: "If host software writes
6360              * an invalid value to the Submission Queue Tail Doorbell or
6361              * Completion Queue Head Doorbell regiter and an Asynchronous Event
6362              * Request command is outstanding, then an asynchronous event is
6363              * posted to the Admin Completion Queue with a status code of
6364              * Invalid Doorbell Write Value."
6365              *
6366              * Also note that the spec includes the "Invalid Doorbell Register"
6367              * status code, but nowhere does it specify when to use it.
6368              * However, it seems reasonable to use it here in a similar
6369              * fashion.
6370              */
6371             if (n->outstanding_aers) {
6372                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
6373                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
6374                                    NVME_LOG_ERROR_INFO);
6375             }
6376 
6377             return;
6378         }
6379 
6380         cq = n->cq[qid];
6381         if (unlikely(new_head >= cq->size)) {
6382             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
6383                            "completion queue doorbell write value"
6384                            " beyond queue size, sqid=%"PRIu32","
6385                            " new_head=%"PRIu16", ignoring",
6386                            qid, new_head);
6387 
6388             if (n->outstanding_aers) {
6389                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
6390                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
6391                                    NVME_LOG_ERROR_INFO);
6392             }
6393 
6394             return;
6395         }
6396 
6397         trace_pci_nvme_mmio_doorbell_cq(cq->cqid, new_head);
6398 
6399         start_sqs = nvme_cq_full(cq) ? 1 : 0;
6400         cq->head = new_head;
6401         if (start_sqs) {
6402             NvmeSQueue *sq;
6403             QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
6404                 timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
6405             }
6406             timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
6407         }
6408 
6409         if (cq->tail == cq->head) {
6410             if (cq->irq_enabled) {
6411                 n->cq_pending--;
6412             }
6413 
6414             nvme_irq_deassert(n, cq);
6415         }
6416     } else {
6417         /* Submission queue doorbell write */
6418 
6419         uint16_t new_tail = val & 0xffff;
6420         NvmeSQueue *sq;
6421 
6422         qid = (addr - 0x1000) >> 3;
6423         if (unlikely(nvme_check_sqid(n, qid))) {
6424             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
6425                            "submission queue doorbell write"
6426                            " for nonexistent queue,"
6427                            " sqid=%"PRIu32", ignoring", qid);
6428 
6429             if (n->outstanding_aers) {
6430                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
6431                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
6432                                    NVME_LOG_ERROR_INFO);
6433             }
6434 
6435             return;
6436         }
6437 
6438         sq = n->sq[qid];
6439         if (unlikely(new_tail >= sq->size)) {
6440             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
6441                            "submission queue doorbell write value"
6442                            " beyond queue size, sqid=%"PRIu32","
6443                            " new_tail=%"PRIu16", ignoring",
6444                            qid, new_tail);
6445 
6446             if (n->outstanding_aers) {
6447                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
6448                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
6449                                    NVME_LOG_ERROR_INFO);
6450             }
6451 
6452             return;
6453         }
6454 
6455         trace_pci_nvme_mmio_doorbell_sq(sq->sqid, new_tail);
6456 
6457         sq->tail = new_tail;
6458         timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
6459     }
6460 }
6461 
6462 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
6463                             unsigned size)
6464 {
6465     NvmeCtrl *n = (NvmeCtrl *)opaque;
6466 
6467     trace_pci_nvme_mmio_write(addr, data, size);
6468 
6469     if (addr < sizeof(n->bar)) {
6470         nvme_write_bar(n, addr, data, size);
6471     } else {
6472         nvme_process_db(n, addr, data);
6473     }
6474 }
6475 
6476 static const MemoryRegionOps nvme_mmio_ops = {
6477     .read = nvme_mmio_read,
6478     .write = nvme_mmio_write,
6479     .endianness = DEVICE_LITTLE_ENDIAN,
6480     .impl = {
6481         .min_access_size = 2,
6482         .max_access_size = 8,
6483     },
6484 };
6485 
6486 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
6487                            unsigned size)
6488 {
6489     NvmeCtrl *n = (NvmeCtrl *)opaque;
6490     stn_le_p(&n->cmb.buf[addr], size, data);
6491 }
6492 
6493 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
6494 {
6495     NvmeCtrl *n = (NvmeCtrl *)opaque;
6496     return ldn_le_p(&n->cmb.buf[addr], size);
6497 }
6498 
6499 static const MemoryRegionOps nvme_cmb_ops = {
6500     .read = nvme_cmb_read,
6501     .write = nvme_cmb_write,
6502     .endianness = DEVICE_LITTLE_ENDIAN,
6503     .impl = {
6504         .min_access_size = 1,
6505         .max_access_size = 8,
6506     },
6507 };
6508 
6509 static void nvme_check_constraints(NvmeCtrl *n, Error **errp)
6510 {
6511     NvmeParams *params = &n->params;
6512 
6513     if (params->num_queues) {
6514         warn_report("num_queues is deprecated; please use max_ioqpairs "
6515                     "instead");
6516 
6517         params->max_ioqpairs = params->num_queues - 1;
6518     }
6519 
6520     if (n->namespace.blkconf.blk && n->subsys) {
6521         error_setg(errp, "subsystem support is unavailable with legacy "
6522                    "namespace ('drive' property)");
6523         return;
6524     }
6525 
6526     if (params->max_ioqpairs < 1 ||
6527         params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
6528         error_setg(errp, "max_ioqpairs must be between 1 and %d",
6529                    NVME_MAX_IOQPAIRS);
6530         return;
6531     }
6532 
6533     if (params->msix_qsize < 1 ||
6534         params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
6535         error_setg(errp, "msix_qsize must be between 1 and %d",
6536                    PCI_MSIX_FLAGS_QSIZE + 1);
6537         return;
6538     }
6539 
6540     if (!params->serial) {
6541         error_setg(errp, "serial property not set");
6542         return;
6543     }
6544 
6545     if (n->pmr.dev) {
6546         if (host_memory_backend_is_mapped(n->pmr.dev)) {
6547             error_setg(errp, "can't use already busy memdev: %s",
6548                        object_get_canonical_path_component(OBJECT(n->pmr.dev)));
6549             return;
6550         }
6551 
6552         if (!is_power_of_2(n->pmr.dev->size)) {
6553             error_setg(errp, "pmr backend size needs to be power of 2 in size");
6554             return;
6555         }
6556 
6557         host_memory_backend_set_mapped(n->pmr.dev, true);
6558     }
6559 
6560     if (n->params.zasl > n->params.mdts) {
6561         error_setg(errp, "zoned.zasl (Zone Append Size Limit) must be less "
6562                    "than or equal to mdts (Maximum Data Transfer Size)");
6563         return;
6564     }
6565 
6566     if (!n->params.vsl) {
6567         error_setg(errp, "vsl must be non-zero");
6568         return;
6569     }
6570 }
6571 
6572 static void nvme_init_state(NvmeCtrl *n)
6573 {
6574     /* add one to max_ioqpairs to account for the admin queue pair */
6575     n->reg_size = pow2ceil(sizeof(NvmeBar) +
6576                            2 * (n->params.max_ioqpairs + 1) * NVME_DB_SIZE);
6577     n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
6578     n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
6579     n->temperature = NVME_TEMPERATURE;
6580     n->features.temp_thresh_hi = NVME_TEMPERATURE_WARNING;
6581     n->starttime_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
6582     n->aer_reqs = g_new0(NvmeRequest *, n->params.aerl + 1);
6583 }
6584 
6585 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
6586 {
6587     uint64_t cmb_size = n->params.cmb_size_mb * MiB;
6588     uint64_t cap = ldq_le_p(&n->bar.cap);
6589 
6590     n->cmb.buf = g_malloc0(cmb_size);
6591     memory_region_init_io(&n->cmb.mem, OBJECT(n), &nvme_cmb_ops, n,
6592                           "nvme-cmb", cmb_size);
6593     pci_register_bar(pci_dev, NVME_CMB_BIR,
6594                      PCI_BASE_ADDRESS_SPACE_MEMORY |
6595                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
6596                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->cmb.mem);
6597 
6598     NVME_CAP_SET_CMBS(cap, 1);
6599     stq_le_p(&n->bar.cap, cap);
6600 
6601     if (n->params.legacy_cmb) {
6602         nvme_cmb_enable_regs(n);
6603         n->cmb.cmse = true;
6604     }
6605 }
6606 
6607 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
6608 {
6609     uint32_t pmrcap = ldl_le_p(&n->bar.pmrcap);
6610 
6611     NVME_PMRCAP_SET_RDS(pmrcap, 1);
6612     NVME_PMRCAP_SET_WDS(pmrcap, 1);
6613     NVME_PMRCAP_SET_BIR(pmrcap, NVME_PMR_BIR);
6614     /* Turn on bit 1 support */
6615     NVME_PMRCAP_SET_PMRWBM(pmrcap, 0x02);
6616     NVME_PMRCAP_SET_CMSS(pmrcap, 1);
6617     stl_le_p(&n->bar.pmrcap, pmrcap);
6618 
6619     pci_register_bar(pci_dev, NVME_PMR_BIR,
6620                      PCI_BASE_ADDRESS_SPACE_MEMORY |
6621                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
6622                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmr.dev->mr);
6623 
6624     memory_region_set_enabled(&n->pmr.dev->mr, false);
6625 }
6626 
6627 static int nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
6628 {
6629     uint8_t *pci_conf = pci_dev->config;
6630     uint64_t bar_size, msix_table_size, msix_pba_size;
6631     unsigned msix_table_offset, msix_pba_offset;
6632     int ret;
6633 
6634     Error *err = NULL;
6635 
6636     pci_conf[PCI_INTERRUPT_PIN] = 1;
6637     pci_config_set_prog_interface(pci_conf, 0x2);
6638 
6639     if (n->params.use_intel_id) {
6640         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_INTEL);
6641         pci_config_set_device_id(pci_conf, 0x5845);
6642     } else {
6643         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_REDHAT);
6644         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_REDHAT_NVME);
6645     }
6646 
6647     pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
6648     pcie_endpoint_cap_init(pci_dev, 0x80);
6649 
6650     bar_size = QEMU_ALIGN_UP(n->reg_size, 4 * KiB);
6651     msix_table_offset = bar_size;
6652     msix_table_size = PCI_MSIX_ENTRY_SIZE * n->params.msix_qsize;
6653 
6654     bar_size += msix_table_size;
6655     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
6656     msix_pba_offset = bar_size;
6657     msix_pba_size = QEMU_ALIGN_UP(n->params.msix_qsize, 64) / 8;
6658 
6659     bar_size += msix_pba_size;
6660     bar_size = pow2ceil(bar_size);
6661 
6662     memory_region_init(&n->bar0, OBJECT(n), "nvme-bar0", bar_size);
6663     memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
6664                           n->reg_size);
6665     memory_region_add_subregion(&n->bar0, 0, &n->iomem);
6666 
6667     pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
6668                      PCI_BASE_ADDRESS_MEM_TYPE_64, &n->bar0);
6669     ret = msix_init(pci_dev, n->params.msix_qsize,
6670                     &n->bar0, 0, msix_table_offset,
6671                     &n->bar0, 0, msix_pba_offset, 0, &err);
6672     if (ret < 0) {
6673         if (ret == -ENOTSUP) {
6674             warn_report_err(err);
6675         } else {
6676             error_propagate(errp, err);
6677             return ret;
6678         }
6679     }
6680 
6681     if (n->params.cmb_size_mb) {
6682         nvme_init_cmb(n, pci_dev);
6683     }
6684 
6685     if (n->pmr.dev) {
6686         nvme_init_pmr(n, pci_dev);
6687     }
6688 
6689     return 0;
6690 }
6691 
6692 static void nvme_init_subnqn(NvmeCtrl *n)
6693 {
6694     NvmeSubsystem *subsys = n->subsys;
6695     NvmeIdCtrl *id = &n->id_ctrl;
6696 
6697     if (!subsys) {
6698         snprintf((char *)id->subnqn, sizeof(id->subnqn),
6699                  "nqn.2019-08.org.qemu:%s", n->params.serial);
6700     } else {
6701         pstrcpy((char *)id->subnqn, sizeof(id->subnqn), (char*)subsys->subnqn);
6702     }
6703 }
6704 
6705 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
6706 {
6707     NvmeIdCtrl *id = &n->id_ctrl;
6708     uint8_t *pci_conf = pci_dev->config;
6709     uint64_t cap = ldq_le_p(&n->bar.cap);
6710 
6711     id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
6712     id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
6713     strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
6714     strpadcpy((char *)id->fr, sizeof(id->fr), "1.0", ' ');
6715     strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
6716 
6717     id->cntlid = cpu_to_le16(n->cntlid);
6718 
6719     id->oaes = cpu_to_le32(NVME_OAES_NS_ATTR);
6720     id->ctratt |= cpu_to_le32(NVME_CTRATT_ELBAS);
6721 
6722     id->rab = 6;
6723 
6724     if (n->params.use_intel_id) {
6725         id->ieee[0] = 0xb3;
6726         id->ieee[1] = 0x02;
6727         id->ieee[2] = 0x00;
6728     } else {
6729         id->ieee[0] = 0x00;
6730         id->ieee[1] = 0x54;
6731         id->ieee[2] = 0x52;
6732     }
6733 
6734     id->mdts = n->params.mdts;
6735     id->ver = cpu_to_le32(NVME_SPEC_VER);
6736     id->oacs = cpu_to_le16(NVME_OACS_NS_MGMT | NVME_OACS_FORMAT);
6737     id->cntrltype = 0x1;
6738 
6739     /*
6740      * Because the controller always completes the Abort command immediately,
6741      * there can never be more than one concurrently executing Abort command,
6742      * so this value is never used for anything. Note that there can easily be
6743      * many Abort commands in the queues, but they are not considered
6744      * "executing" until processed by nvme_abort.
6745      *
6746      * The specification recommends a value of 3 for Abort Command Limit (four
6747      * concurrently outstanding Abort commands), so lets use that though it is
6748      * inconsequential.
6749      */
6750     id->acl = 3;
6751     id->aerl = n->params.aerl;
6752     id->frmw = (NVME_NUM_FW_SLOTS << 1) | NVME_FRMW_SLOT1_RO;
6753     id->lpa = NVME_LPA_NS_SMART | NVME_LPA_CSE | NVME_LPA_EXTENDED;
6754 
6755     /* recommended default value (~70 C) */
6756     id->wctemp = cpu_to_le16(NVME_TEMPERATURE_WARNING);
6757     id->cctemp = cpu_to_le16(NVME_TEMPERATURE_CRITICAL);
6758 
6759     id->sqes = (0x6 << 4) | 0x6;
6760     id->cqes = (0x4 << 4) | 0x4;
6761     id->nn = cpu_to_le32(NVME_MAX_NAMESPACES);
6762     id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP |
6763                            NVME_ONCS_FEATURES | NVME_ONCS_DSM |
6764                            NVME_ONCS_COMPARE | NVME_ONCS_COPY);
6765 
6766     /*
6767      * NOTE: If this device ever supports a command set that does NOT use 0x0
6768      * as a Flush-equivalent operation, support for the broadcast NSID in Flush
6769      * should probably be removed.
6770      *
6771      * See comment in nvme_io_cmd.
6772      */
6773     id->vwc = NVME_VWC_NSID_BROADCAST_SUPPORT | NVME_VWC_PRESENT;
6774 
6775     id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0 | NVME_OCFS_COPY_FORMAT_1);
6776     id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN |
6777                            NVME_CTRL_SGLS_BITBUCKET);
6778 
6779     nvme_init_subnqn(n);
6780 
6781     id->psd[0].mp = cpu_to_le16(0x9c4);
6782     id->psd[0].enlat = cpu_to_le32(0x10);
6783     id->psd[0].exlat = cpu_to_le32(0x4);
6784 
6785     if (n->subsys) {
6786         id->cmic |= NVME_CMIC_MULTI_CTRL;
6787     }
6788 
6789     NVME_CAP_SET_MQES(cap, 0x7ff);
6790     NVME_CAP_SET_CQR(cap, 1);
6791     NVME_CAP_SET_TO(cap, 0xf);
6792     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_NVM);
6793     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_CSI_SUPP);
6794     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_ADMIN_ONLY);
6795     NVME_CAP_SET_MPSMAX(cap, 4);
6796     NVME_CAP_SET_CMBS(cap, n->params.cmb_size_mb ? 1 : 0);
6797     NVME_CAP_SET_PMRS(cap, n->pmr.dev ? 1 : 0);
6798     stq_le_p(&n->bar.cap, cap);
6799 
6800     stl_le_p(&n->bar.vs, NVME_SPEC_VER);
6801     n->bar.intmc = n->bar.intms = 0;
6802 }
6803 
6804 static int nvme_init_subsys(NvmeCtrl *n, Error **errp)
6805 {
6806     int cntlid;
6807 
6808     if (!n->subsys) {
6809         return 0;
6810     }
6811 
6812     cntlid = nvme_subsys_register_ctrl(n, errp);
6813     if (cntlid < 0) {
6814         return -1;
6815     }
6816 
6817     n->cntlid = cntlid;
6818 
6819     return 0;
6820 }
6821 
6822 void nvme_attach_ns(NvmeCtrl *n, NvmeNamespace *ns)
6823 {
6824     uint32_t nsid = ns->params.nsid;
6825     assert(nsid && nsid <= NVME_MAX_NAMESPACES);
6826 
6827     n->namespaces[nsid] = ns;
6828     ns->attached++;
6829 
6830     n->dmrsl = MIN_NON_ZERO(n->dmrsl,
6831                             BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
6832 }
6833 
6834 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
6835 {
6836     NvmeCtrl *n = NVME(pci_dev);
6837     NvmeNamespace *ns;
6838     Error *local_err = NULL;
6839 
6840     nvme_check_constraints(n, &local_err);
6841     if (local_err) {
6842         error_propagate(errp, local_err);
6843         return;
6844     }
6845 
6846     qbus_init(&n->bus, sizeof(NvmeBus), TYPE_NVME_BUS,
6847               &pci_dev->qdev, n->parent_obj.qdev.id);
6848 
6849     nvme_init_state(n);
6850     if (nvme_init_pci(n, pci_dev, errp)) {
6851         return;
6852     }
6853 
6854     if (nvme_init_subsys(n, errp)) {
6855         error_propagate(errp, local_err);
6856         return;
6857     }
6858     nvme_init_ctrl(n, pci_dev);
6859 
6860     /* setup a namespace if the controller drive property was given */
6861     if (n->namespace.blkconf.blk) {
6862         ns = &n->namespace;
6863         ns->params.nsid = 1;
6864 
6865         if (nvme_ns_setup(ns, errp)) {
6866             return;
6867         }
6868 
6869         nvme_attach_ns(n, ns);
6870     }
6871 }
6872 
6873 static void nvme_exit(PCIDevice *pci_dev)
6874 {
6875     NvmeCtrl *n = NVME(pci_dev);
6876     NvmeNamespace *ns;
6877     int i;
6878 
6879     nvme_ctrl_reset(n);
6880 
6881     if (n->subsys) {
6882         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6883             ns = nvme_ns(n, i);
6884             if (ns) {
6885                 ns->attached--;
6886             }
6887         }
6888 
6889         nvme_subsys_unregister_ctrl(n->subsys, n);
6890     }
6891 
6892     g_free(n->cq);
6893     g_free(n->sq);
6894     g_free(n->aer_reqs);
6895 
6896     if (n->params.cmb_size_mb) {
6897         g_free(n->cmb.buf);
6898     }
6899 
6900     if (n->pmr.dev) {
6901         host_memory_backend_set_mapped(n->pmr.dev, false);
6902     }
6903     msix_uninit(pci_dev, &n->bar0, &n->bar0);
6904     memory_region_del_subregion(&n->bar0, &n->iomem);
6905 }
6906 
6907 static Property nvme_props[] = {
6908     DEFINE_BLOCK_PROPERTIES(NvmeCtrl, namespace.blkconf),
6909     DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmr.dev, TYPE_MEMORY_BACKEND,
6910                      HostMemoryBackend *),
6911     DEFINE_PROP_LINK("subsys", NvmeCtrl, subsys, TYPE_NVME_SUBSYS,
6912                      NvmeSubsystem *),
6913     DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
6914     DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
6915     DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
6916     DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
6917     DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
6918     DEFINE_PROP_UINT8("aerl", NvmeCtrl, params.aerl, 3),
6919     DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl, params.aer_max_queued, 64),
6920     DEFINE_PROP_UINT8("mdts", NvmeCtrl, params.mdts, 7),
6921     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
6922     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
6923     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
6924     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
6925     DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl,
6926                      params.auto_transition_zones, true),
6927     DEFINE_PROP_END_OF_LIST(),
6928 };
6929 
6930 static void nvme_get_smart_warning(Object *obj, Visitor *v, const char *name,
6931                                    void *opaque, Error **errp)
6932 {
6933     NvmeCtrl *n = NVME(obj);
6934     uint8_t value = n->smart_critical_warning;
6935 
6936     visit_type_uint8(v, name, &value, errp);
6937 }
6938 
6939 static void nvme_set_smart_warning(Object *obj, Visitor *v, const char *name,
6940                                    void *opaque, Error **errp)
6941 {
6942     NvmeCtrl *n = NVME(obj);
6943     uint8_t value, old_value, cap = 0, index, event;
6944 
6945     if (!visit_type_uint8(v, name, &value, errp)) {
6946         return;
6947     }
6948 
6949     cap = NVME_SMART_SPARE | NVME_SMART_TEMPERATURE | NVME_SMART_RELIABILITY
6950           | NVME_SMART_MEDIA_READ_ONLY | NVME_SMART_FAILED_VOLATILE_MEDIA;
6951     if (NVME_CAP_PMRS(ldq_le_p(&n->bar.cap))) {
6952         cap |= NVME_SMART_PMR_UNRELIABLE;
6953     }
6954 
6955     if ((value & cap) != value) {
6956         error_setg(errp, "unsupported smart critical warning bits: 0x%x",
6957                    value & ~cap);
6958         return;
6959     }
6960 
6961     old_value = n->smart_critical_warning;
6962     n->smart_critical_warning = value;
6963 
6964     /* only inject new bits of smart critical warning */
6965     for (index = 0; index < NVME_SMART_WARN_MAX; index++) {
6966         event = 1 << index;
6967         if (value & ~old_value & event)
6968             nvme_smart_event(n, event);
6969     }
6970 }
6971 
6972 static const VMStateDescription nvme_vmstate = {
6973     .name = "nvme",
6974     .unmigratable = 1,
6975 };
6976 
6977 static void nvme_class_init(ObjectClass *oc, void *data)
6978 {
6979     DeviceClass *dc = DEVICE_CLASS(oc);
6980     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
6981 
6982     pc->realize = nvme_realize;
6983     pc->exit = nvme_exit;
6984     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
6985     pc->revision = 2;
6986 
6987     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
6988     dc->desc = "Non-Volatile Memory Express";
6989     device_class_set_props(dc, nvme_props);
6990     dc->vmsd = &nvme_vmstate;
6991 }
6992 
6993 static void nvme_instance_init(Object *obj)
6994 {
6995     NvmeCtrl *n = NVME(obj);
6996 
6997     device_add_bootindex_property(obj, &n->namespace.blkconf.bootindex,
6998                                   "bootindex", "/namespace@1,0",
6999                                   DEVICE(obj));
7000 
7001     object_property_add(obj, "smart_critical_warning", "uint8",
7002                         nvme_get_smart_warning,
7003                         nvme_set_smart_warning, NULL, NULL);
7004 }
7005 
7006 static const TypeInfo nvme_info = {
7007     .name          = TYPE_NVME,
7008     .parent        = TYPE_PCI_DEVICE,
7009     .instance_size = sizeof(NvmeCtrl),
7010     .instance_init = nvme_instance_init,
7011     .class_init    = nvme_class_init,
7012     .interfaces = (InterfaceInfo[]) {
7013         { INTERFACE_PCIE_DEVICE },
7014         { }
7015     },
7016 };
7017 
7018 static const TypeInfo nvme_bus_info = {
7019     .name = TYPE_NVME_BUS,
7020     .parent = TYPE_BUS,
7021     .instance_size = sizeof(NvmeBus),
7022 };
7023 
7024 static void nvme_register_types(void)
7025 {
7026     type_register_static(&nvme_info);
7027     type_register_static(&nvme_bus_info);
7028 }
7029 
7030 type_init(nvme_register_types)
7031