xref: /openbmc/qemu/hw/nvme/ctrl.c (revision 51e47cf8)
1 /*
2  * QEMU NVM Express Controller
3  *
4  * Copyright (c) 2012, Intel Corporation
5  *
6  * Written by Keith Busch <keith.busch@intel.com>
7  *
8  * This code is licensed under the GNU GPL v2 or later.
9  */
10 
11 /**
12  * Reference Specs: http://www.nvmexpress.org, 1.4, 1.3, 1.2, 1.1, 1.0e
13  *
14  *  https://nvmexpress.org/developers/nvme-specification/
15  *
16  *
17  * Notes on coding style
18  * ---------------------
19  * While QEMU coding style prefers lowercase hexadecimals in constants, the
20  * NVMe subsystem use thes format from the NVMe specifications in the comments
21  * (i.e. 'h' suffix instead of '0x' prefix).
22  *
23  * Usage
24  * -----
25  * See docs/system/nvme.rst for extensive documentation.
26  *
27  * Add options:
28  *      -drive file=<file>,if=none,id=<drive_id>
29  *      -device nvme-subsys,id=<subsys_id>,nqn=<nqn_id>
30  *      -device nvme,serial=<serial>,id=<bus_name>, \
31  *              cmb_size_mb=<cmb_size_mb[optional]>, \
32  *              [pmrdev=<mem_backend_file_id>,] \
33  *              max_ioqpairs=<N[optional]>, \
34  *              aerl=<N[optional]>,aer_max_queued=<N[optional]>, \
35  *              mdts=<N[optional]>,vsl=<N[optional]>, \
36  *              zoned.zasl=<N[optional]>, \
37  *              zoned.auto_transition=<on|off[optional]>, \
38  *              sriov_max_vfs=<N[optional]> \
39  *              sriov_vq_flexible=<N[optional]> \
40  *              sriov_vi_flexible=<N[optional]> \
41  *              sriov_max_vi_per_vf=<N[optional]> \
42  *              sriov_max_vq_per_vf=<N[optional]> \
43  *              subsys=<subsys_id>
44  *      -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
45  *              zoned=<true|false[optional]>, \
46  *              subsys=<subsys_id>,detached=<true|false[optional]>
47  *
48  * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
49  * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. By default, the
50  * device will use the "v1.4 CMB scheme" - use the `legacy-cmb` parameter to
51  * always enable the CMBLOC and CMBSZ registers (v1.3 behavior).
52  *
53  * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
54  * For example:
55  * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
56  *  size=<size> .... -device nvme,...,pmrdev=<mem_id>
57  *
58  * The PMR will use BAR 4/5 exclusively.
59  *
60  * To place controller(s) and namespace(s) to a subsystem, then provide
61  * nvme-subsys device as above.
62  *
63  * nvme subsystem device parameters
64  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
65  * - `nqn`
66  *   This parameter provides the `<nqn_id>` part of the string
67  *   `nqn.2019-08.org.qemu:<nqn_id>` which will be reported in the SUBNQN field
68  *   of subsystem controllers. Note that `<nqn_id>` should be unique per
69  *   subsystem, but this is not enforced by QEMU. If not specified, it will
70  *   default to the value of the `id` parameter (`<subsys_id>`).
71  *
72  * nvme device parameters
73  * ~~~~~~~~~~~~~~~~~~~~~~
74  * - `subsys`
75  *   Specifying this parameter attaches the controller to the subsystem and
76  *   the SUBNQN field in the controller will report the NQN of the subsystem
77  *   device. This also enables multi controller capability represented in
78  *   Identify Controller data structure in CMIC (Controller Multi-path I/O and
79  *   Namespace Sharing Capabilities).
80  *
81  * - `aerl`
82  *   The Asynchronous Event Request Limit (AERL). Indicates the maximum number
83  *   of concurrently outstanding Asynchronous Event Request commands support
84  *   by the controller. This is a 0's based value.
85  *
86  * - `aer_max_queued`
87  *   This is the maximum number of events that the device will enqueue for
88  *   completion when there are no outstanding AERs. When the maximum number of
89  *   enqueued events are reached, subsequent events will be dropped.
90  *
91  * - `mdts`
92  *   Indicates the maximum data transfer size for a command that transfers data
93  *   between host-accessible memory and the controller. The value is specified
94  *   as a power of two (2^n) and is in units of the minimum memory page size
95  *   (CAP.MPSMIN). The default value is 7 (i.e. 512 KiB).
96  *
97  * - `vsl`
98  *   Indicates the maximum data size limit for the Verify command. Like `mdts`,
99  *   this value is specified as a power of two (2^n) and is in units of the
100  *   minimum memory page size (CAP.MPSMIN). The default value is 7 (i.e. 512
101  *   KiB).
102  *
103  * - `zoned.zasl`
104  *   Indicates the maximum data transfer size for the Zone Append command. Like
105  *   `mdts`, the value is specified as a power of two (2^n) and is in units of
106  *   the minimum memory page size (CAP.MPSMIN). The default value is 0 (i.e.
107  *   defaulting to the value of `mdts`).
108  *
109  * - `zoned.auto_transition`
110  *   Indicates if zones in zone state implicitly opened can be automatically
111  *   transitioned to zone state closed for resource management purposes.
112  *   Defaults to 'on'.
113  *
114  * - `sriov_max_vfs`
115  *   Indicates the maximum number of PCIe virtual functions supported
116  *   by the controller. The default value is 0. Specifying a non-zero value
117  *   enables reporting of both SR-IOV and ARI capabilities by the NVMe device.
118  *   Virtual function controllers will not report SR-IOV capability.
119  *
120  *   NOTE: Single Root I/O Virtualization support is experimental.
121  *   All the related parameters may be subject to change.
122  *
123  * - `sriov_vq_flexible`
124  *   Indicates the total number of flexible queue resources assignable to all
125  *   the secondary controllers. Implicitly sets the number of primary
126  *   controller's private resources to `(max_ioqpairs - sriov_vq_flexible)`.
127  *
128  * - `sriov_vi_flexible`
129  *   Indicates the total number of flexible interrupt resources assignable to
130  *   all the secondary controllers. Implicitly sets the number of primary
131  *   controller's private resources to `(msix_qsize - sriov_vi_flexible)`.
132  *
133  * - `sriov_max_vi_per_vf`
134  *   Indicates the maximum number of virtual interrupt resources assignable
135  *   to a secondary controller. The default 0 resolves to
136  *   `(sriov_vi_flexible / sriov_max_vfs)`.
137  *
138  * - `sriov_max_vq_per_vf`
139  *   Indicates the maximum number of virtual queue resources assignable to
140  *   a secondary controller. The default 0 resolves to
141  *   `(sriov_vq_flexible / sriov_max_vfs)`.
142  *
143  * nvme namespace device parameters
144  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
145  * - `shared`
146  *   When the parent nvme device (as defined explicitly by the 'bus' parameter
147  *   or implicitly by the most recently defined NvmeBus) is linked to an
148  *   nvme-subsys device, the namespace will be attached to all controllers in
149  *   the subsystem. If set to 'off' (the default), the namespace will remain a
150  *   private namespace and may only be attached to a single controller at a
151  *   time.
152  *
153  * - `detached`
154  *   This parameter is only valid together with the `subsys` parameter. If left
155  *   at the default value (`false/off`), the namespace will be attached to all
156  *   controllers in the NVMe subsystem at boot-up. If set to `true/on`, the
157  *   namespace will be available in the subsystem but not attached to any
158  *   controllers.
159  *
160  * Setting `zoned` to true selects Zoned Command Set at the namespace.
161  * In this case, the following namespace properties are available to configure
162  * zoned operation:
163  *     zoned.zone_size=<zone size in bytes, default: 128MiB>
164  *         The number may be followed by K, M, G as in kilo-, mega- or giga-.
165  *
166  *     zoned.zone_capacity=<zone capacity in bytes, default: zone size>
167  *         The value 0 (default) forces zone capacity to be the same as zone
168  *         size. The value of this property may not exceed zone size.
169  *
170  *     zoned.descr_ext_size=<zone descriptor extension size, default 0>
171  *         This value needs to be specified in 64B units. If it is zero,
172  *         namespace(s) will not support zone descriptor extensions.
173  *
174  *     zoned.max_active=<Maximum Active Resources (zones), default: 0>
175  *         The default value means there is no limit to the number of
176  *         concurrently active zones.
177  *
178  *     zoned.max_open=<Maximum Open Resources (zones), default: 0>
179  *         The default value means there is no limit to the number of
180  *         concurrently open zones.
181  *
182  *     zoned.cross_read=<enable RAZB, default: false>
183  *         Setting this property to true enables Read Across Zone Boundaries.
184  */
185 
186 #include "qemu/osdep.h"
187 #include "qemu/cutils.h"
188 #include "qemu/error-report.h"
189 #include "qemu/log.h"
190 #include "qemu/units.h"
191 #include "qemu/range.h"
192 #include "qapi/error.h"
193 #include "qapi/visitor.h"
194 #include "sysemu/sysemu.h"
195 #include "sysemu/block-backend.h"
196 #include "sysemu/hostmem.h"
197 #include "hw/pci/msix.h"
198 #include "hw/pci/pcie_sriov.h"
199 #include "migration/vmstate.h"
200 
201 #include "nvme.h"
202 #include "dif.h"
203 #include "trace.h"
204 
205 #define NVME_MAX_IOQPAIRS 0xffff
206 #define NVME_DB_SIZE  4
207 #define NVME_SPEC_VER 0x00010400
208 #define NVME_CMB_BIR 2
209 #define NVME_PMR_BIR 4
210 #define NVME_TEMPERATURE 0x143
211 #define NVME_TEMPERATURE_WARNING 0x157
212 #define NVME_TEMPERATURE_CRITICAL 0x175
213 #define NVME_NUM_FW_SLOTS 1
214 #define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
215 #define NVME_MAX_VFS 127
216 #define NVME_VF_RES_GRANULARITY 1
217 #define NVME_VF_OFFSET 0x1
218 #define NVME_VF_STRIDE 1
219 
220 #define NVME_GUEST_ERR(trace, fmt, ...) \
221     do { \
222         (trace_##trace)(__VA_ARGS__); \
223         qemu_log_mask(LOG_GUEST_ERROR, #trace \
224             " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
225     } while (0)
226 
227 static const bool nvme_feature_support[NVME_FID_MAX] = {
228     [NVME_ARBITRATION]              = true,
229     [NVME_POWER_MANAGEMENT]         = true,
230     [NVME_TEMPERATURE_THRESHOLD]    = true,
231     [NVME_ERROR_RECOVERY]           = true,
232     [NVME_VOLATILE_WRITE_CACHE]     = true,
233     [NVME_NUMBER_OF_QUEUES]         = true,
234     [NVME_INTERRUPT_COALESCING]     = true,
235     [NVME_INTERRUPT_VECTOR_CONF]    = true,
236     [NVME_WRITE_ATOMICITY]          = true,
237     [NVME_ASYNCHRONOUS_EVENT_CONF]  = true,
238     [NVME_TIMESTAMP]                = true,
239     [NVME_HOST_BEHAVIOR_SUPPORT]    = true,
240     [NVME_COMMAND_SET_PROFILE]      = true,
241     [NVME_FDP_MODE]                 = true,
242     [NVME_FDP_EVENTS]               = true,
243 };
244 
245 static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
246     [NVME_TEMPERATURE_THRESHOLD]    = NVME_FEAT_CAP_CHANGE,
247     [NVME_ERROR_RECOVERY]           = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
248     [NVME_VOLATILE_WRITE_CACHE]     = NVME_FEAT_CAP_CHANGE,
249     [NVME_NUMBER_OF_QUEUES]         = NVME_FEAT_CAP_CHANGE,
250     [NVME_ASYNCHRONOUS_EVENT_CONF]  = NVME_FEAT_CAP_CHANGE,
251     [NVME_TIMESTAMP]                = NVME_FEAT_CAP_CHANGE,
252     [NVME_HOST_BEHAVIOR_SUPPORT]    = NVME_FEAT_CAP_CHANGE,
253     [NVME_COMMAND_SET_PROFILE]      = NVME_FEAT_CAP_CHANGE,
254     [NVME_FDP_MODE]                 = NVME_FEAT_CAP_CHANGE,
255     [NVME_FDP_EVENTS]               = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
256 };
257 
258 static const uint32_t nvme_cse_acs[256] = {
259     [NVME_ADM_CMD_DELETE_SQ]        = NVME_CMD_EFF_CSUPP,
260     [NVME_ADM_CMD_CREATE_SQ]        = NVME_CMD_EFF_CSUPP,
261     [NVME_ADM_CMD_GET_LOG_PAGE]     = NVME_CMD_EFF_CSUPP,
262     [NVME_ADM_CMD_DELETE_CQ]        = NVME_CMD_EFF_CSUPP,
263     [NVME_ADM_CMD_CREATE_CQ]        = NVME_CMD_EFF_CSUPP,
264     [NVME_ADM_CMD_IDENTIFY]         = NVME_CMD_EFF_CSUPP,
265     [NVME_ADM_CMD_ABORT]            = NVME_CMD_EFF_CSUPP,
266     [NVME_ADM_CMD_SET_FEATURES]     = NVME_CMD_EFF_CSUPP,
267     [NVME_ADM_CMD_GET_FEATURES]     = NVME_CMD_EFF_CSUPP,
268     [NVME_ADM_CMD_ASYNC_EV_REQ]     = NVME_CMD_EFF_CSUPP,
269     [NVME_ADM_CMD_NS_ATTACHMENT]    = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_NIC,
270     [NVME_ADM_CMD_VIRT_MNGMT]       = NVME_CMD_EFF_CSUPP,
271     [NVME_ADM_CMD_DBBUF_CONFIG]     = NVME_CMD_EFF_CSUPP,
272     [NVME_ADM_CMD_FORMAT_NVM]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
273     [NVME_ADM_CMD_DIRECTIVE_RECV]   = NVME_CMD_EFF_CSUPP,
274     [NVME_ADM_CMD_DIRECTIVE_SEND]   = NVME_CMD_EFF_CSUPP,
275 };
276 
277 static const uint32_t nvme_cse_iocs_none[256];
278 
279 static const uint32_t nvme_cse_iocs_nvm[256] = {
280     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
281     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
282     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
283     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
284     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
285     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
286     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
287     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
288     [NVME_CMD_IO_MGMT_RECV]         = NVME_CMD_EFF_CSUPP,
289     [NVME_CMD_IO_MGMT_SEND]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
290 };
291 
292 static const uint32_t nvme_cse_iocs_zoned[256] = {
293     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
294     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
295     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
296     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
297     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
298     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
299     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
300     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
301     [NVME_CMD_ZONE_APPEND]          = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
302     [NVME_CMD_ZONE_MGMT_SEND]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
303     [NVME_CMD_ZONE_MGMT_RECV]       = NVME_CMD_EFF_CSUPP,
304 };
305 
306 static void nvme_process_sq(void *opaque);
307 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst);
308 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n);
309 
310 static uint16_t nvme_sqid(NvmeRequest *req)
311 {
312     return le16_to_cpu(req->sq->sqid);
313 }
314 
315 static inline uint16_t nvme_make_pid(NvmeNamespace *ns, uint16_t rg,
316                                      uint16_t ph)
317 {
318     uint16_t rgif = ns->endgrp->fdp.rgif;
319 
320     if (!rgif) {
321         return ph;
322     }
323 
324     return (rg << (16 - rgif)) | ph;
325 }
326 
327 static inline bool nvme_ph_valid(NvmeNamespace *ns, uint16_t ph)
328 {
329     return ph < ns->fdp.nphs;
330 }
331 
332 static inline bool nvme_rg_valid(NvmeEnduranceGroup *endgrp, uint16_t rg)
333 {
334     return rg < endgrp->fdp.nrg;
335 }
336 
337 static inline uint16_t nvme_pid2ph(NvmeNamespace *ns, uint16_t pid)
338 {
339     uint16_t rgif = ns->endgrp->fdp.rgif;
340 
341     if (!rgif) {
342         return pid;
343     }
344 
345     return pid & ((1 << (15 - rgif)) - 1);
346 }
347 
348 static inline uint16_t nvme_pid2rg(NvmeNamespace *ns, uint16_t pid)
349 {
350     uint16_t rgif = ns->endgrp->fdp.rgif;
351 
352     if (!rgif) {
353         return 0;
354     }
355 
356     return pid >> (16 - rgif);
357 }
358 
359 static inline bool nvme_parse_pid(NvmeNamespace *ns, uint16_t pid,
360                                   uint16_t *ph, uint16_t *rg)
361 {
362     *rg = nvme_pid2rg(ns, pid);
363     *ph = nvme_pid2ph(ns, pid);
364 
365     return nvme_ph_valid(ns, *ph) && nvme_rg_valid(ns->endgrp, *rg);
366 }
367 
368 static void nvme_assign_zone_state(NvmeNamespace *ns, NvmeZone *zone,
369                                    NvmeZoneState state)
370 {
371     if (QTAILQ_IN_USE(zone, entry)) {
372         switch (nvme_get_zone_state(zone)) {
373         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
374             QTAILQ_REMOVE(&ns->exp_open_zones, zone, entry);
375             break;
376         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
377             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
378             break;
379         case NVME_ZONE_STATE_CLOSED:
380             QTAILQ_REMOVE(&ns->closed_zones, zone, entry);
381             break;
382         case NVME_ZONE_STATE_FULL:
383             QTAILQ_REMOVE(&ns->full_zones, zone, entry);
384         default:
385             ;
386         }
387     }
388 
389     nvme_set_zone_state(zone, state);
390 
391     switch (state) {
392     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
393         QTAILQ_INSERT_TAIL(&ns->exp_open_zones, zone, entry);
394         break;
395     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
396         QTAILQ_INSERT_TAIL(&ns->imp_open_zones, zone, entry);
397         break;
398     case NVME_ZONE_STATE_CLOSED:
399         QTAILQ_INSERT_TAIL(&ns->closed_zones, zone, entry);
400         break;
401     case NVME_ZONE_STATE_FULL:
402         QTAILQ_INSERT_TAIL(&ns->full_zones, zone, entry);
403     case NVME_ZONE_STATE_READ_ONLY:
404         break;
405     default:
406         zone->d.za = 0;
407     }
408 }
409 
410 static uint16_t nvme_zns_check_resources(NvmeNamespace *ns, uint32_t act,
411                                          uint32_t opn, uint32_t zrwa)
412 {
413     if (ns->params.max_active_zones != 0 &&
414         ns->nr_active_zones + act > ns->params.max_active_zones) {
415         trace_pci_nvme_err_insuff_active_res(ns->params.max_active_zones);
416         return NVME_ZONE_TOO_MANY_ACTIVE | NVME_DNR;
417     }
418 
419     if (ns->params.max_open_zones != 0 &&
420         ns->nr_open_zones + opn > ns->params.max_open_zones) {
421         trace_pci_nvme_err_insuff_open_res(ns->params.max_open_zones);
422         return NVME_ZONE_TOO_MANY_OPEN | NVME_DNR;
423     }
424 
425     if (zrwa > ns->zns.numzrwa) {
426         return NVME_NOZRWA | NVME_DNR;
427     }
428 
429     return NVME_SUCCESS;
430 }
431 
432 /*
433  * Check if we can open a zone without exceeding open/active limits.
434  * AOR stands for "Active and Open Resources" (see TP 4053 section 2.5).
435  */
436 static uint16_t nvme_aor_check(NvmeNamespace *ns, uint32_t act, uint32_t opn)
437 {
438     return nvme_zns_check_resources(ns, act, opn, 0);
439 }
440 
441 static NvmeFdpEvent *nvme_fdp_alloc_event(NvmeCtrl *n, NvmeFdpEventBuffer *ebuf)
442 {
443     NvmeFdpEvent *ret = NULL;
444     bool is_full = ebuf->next == ebuf->start && ebuf->nelems;
445 
446     ret = &ebuf->events[ebuf->next++];
447     if (unlikely(ebuf->next == NVME_FDP_MAX_EVENTS)) {
448         ebuf->next = 0;
449     }
450     if (is_full) {
451         ebuf->start = ebuf->next;
452     } else {
453         ebuf->nelems++;
454     }
455 
456     memset(ret, 0, sizeof(NvmeFdpEvent));
457     ret->timestamp = nvme_get_timestamp(n);
458 
459     return ret;
460 }
461 
462 static inline int log_event(NvmeRuHandle *ruh, uint8_t event_type)
463 {
464     return (ruh->event_filter >> nvme_fdp_evf_shifts[event_type]) & 0x1;
465 }
466 
467 static bool nvme_update_ruh(NvmeCtrl *n, NvmeNamespace *ns, uint16_t pid)
468 {
469     NvmeEnduranceGroup *endgrp = ns->endgrp;
470     NvmeRuHandle *ruh;
471     NvmeReclaimUnit *ru;
472     NvmeFdpEvent *e = NULL;
473     uint16_t ph, rg, ruhid;
474 
475     if (!nvme_parse_pid(ns, pid, &ph, &rg)) {
476         return false;
477     }
478 
479     ruhid = ns->fdp.phs[ph];
480 
481     ruh = &endgrp->fdp.ruhs[ruhid];
482     ru = &ruh->rus[rg];
483 
484     if (ru->ruamw) {
485         if (log_event(ruh, FDP_EVT_RU_NOT_FULLY_WRITTEN)) {
486             e = nvme_fdp_alloc_event(n, &endgrp->fdp.host_events);
487             e->type = FDP_EVT_RU_NOT_FULLY_WRITTEN;
488             e->flags = FDPEF_PIV | FDPEF_NSIDV | FDPEF_LV;
489             e->pid = cpu_to_le16(pid);
490             e->nsid = cpu_to_le32(ns->params.nsid);
491             e->rgid = cpu_to_le16(rg);
492             e->ruhid = cpu_to_le16(ruhid);
493         }
494 
495         /* log (eventual) GC overhead of prematurely swapping the RU */
496         nvme_fdp_stat_inc(&endgrp->fdp.mbmw, nvme_l2b(ns, ru->ruamw));
497     }
498 
499     ru->ruamw = ruh->ruamw;
500 
501     return true;
502 }
503 
504 static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
505 {
506     hwaddr hi, lo;
507 
508     if (!n->cmb.cmse) {
509         return false;
510     }
511 
512     lo = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
513     hi = lo + int128_get64(n->cmb.mem.size);
514 
515     return addr >= lo && addr < hi;
516 }
517 
518 static inline void *nvme_addr_to_cmb(NvmeCtrl *n, hwaddr addr)
519 {
520     hwaddr base = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
521     return &n->cmb.buf[addr - base];
522 }
523 
524 static bool nvme_addr_is_pmr(NvmeCtrl *n, hwaddr addr)
525 {
526     hwaddr hi;
527 
528     if (!n->pmr.cmse) {
529         return false;
530     }
531 
532     hi = n->pmr.cba + int128_get64(n->pmr.dev->mr.size);
533 
534     return addr >= n->pmr.cba && addr < hi;
535 }
536 
537 static inline void *nvme_addr_to_pmr(NvmeCtrl *n, hwaddr addr)
538 {
539     return memory_region_get_ram_ptr(&n->pmr.dev->mr) + (addr - n->pmr.cba);
540 }
541 
542 static inline bool nvme_addr_is_iomem(NvmeCtrl *n, hwaddr addr)
543 {
544     hwaddr hi, lo;
545 
546     /*
547      * The purpose of this check is to guard against invalid "local" access to
548      * the iomem (i.e. controller registers). Thus, we check against the range
549      * covered by the 'bar0' MemoryRegion since that is currently composed of
550      * two subregions (the NVMe "MBAR" and the MSI-X table/pba). Note, however,
551      * that if the device model is ever changed to allow the CMB to be located
552      * in BAR0 as well, then this must be changed.
553      */
554     lo = n->bar0.addr;
555     hi = lo + int128_get64(n->bar0.size);
556 
557     return addr >= lo && addr < hi;
558 }
559 
560 static int nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
561 {
562     hwaddr hi = addr + size - 1;
563     if (hi < addr) {
564         return 1;
565     }
566 
567     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
568         memcpy(buf, nvme_addr_to_cmb(n, addr), size);
569         return 0;
570     }
571 
572     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
573         memcpy(buf, nvme_addr_to_pmr(n, addr), size);
574         return 0;
575     }
576 
577     return pci_dma_read(PCI_DEVICE(n), addr, buf, size);
578 }
579 
580 static int nvme_addr_write(NvmeCtrl *n, hwaddr addr, const void *buf, int size)
581 {
582     hwaddr hi = addr + size - 1;
583     if (hi < addr) {
584         return 1;
585     }
586 
587     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
588         memcpy(nvme_addr_to_cmb(n, addr), buf, size);
589         return 0;
590     }
591 
592     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
593         memcpy(nvme_addr_to_pmr(n, addr), buf, size);
594         return 0;
595     }
596 
597     return pci_dma_write(PCI_DEVICE(n), addr, buf, size);
598 }
599 
600 static bool nvme_nsid_valid(NvmeCtrl *n, uint32_t nsid)
601 {
602     return nsid &&
603         (nsid == NVME_NSID_BROADCAST || nsid <= NVME_MAX_NAMESPACES);
604 }
605 
606 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
607 {
608     return sqid < n->conf_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
609 }
610 
611 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
612 {
613     return cqid < n->conf_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
614 }
615 
616 static void nvme_inc_cq_tail(NvmeCQueue *cq)
617 {
618     cq->tail++;
619     if (cq->tail >= cq->size) {
620         cq->tail = 0;
621         cq->phase = !cq->phase;
622     }
623 }
624 
625 static void nvme_inc_sq_head(NvmeSQueue *sq)
626 {
627     sq->head = (sq->head + 1) % sq->size;
628 }
629 
630 static uint8_t nvme_cq_full(NvmeCQueue *cq)
631 {
632     return (cq->tail + 1) % cq->size == cq->head;
633 }
634 
635 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
636 {
637     return sq->head == sq->tail;
638 }
639 
640 static void nvme_irq_check(NvmeCtrl *n)
641 {
642     PCIDevice *pci = PCI_DEVICE(n);
643     uint32_t intms = ldl_le_p(&n->bar.intms);
644 
645     if (msix_enabled(pci)) {
646         return;
647     }
648     if (~intms & n->irq_status) {
649         pci_irq_assert(pci);
650     } else {
651         pci_irq_deassert(pci);
652     }
653 }
654 
655 static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
656 {
657     PCIDevice *pci = PCI_DEVICE(n);
658 
659     if (cq->irq_enabled) {
660         if (msix_enabled(pci)) {
661             trace_pci_nvme_irq_msix(cq->vector);
662             msix_notify(pci, cq->vector);
663         } else {
664             trace_pci_nvme_irq_pin();
665             assert(cq->vector < 32);
666             n->irq_status |= 1 << cq->vector;
667             nvme_irq_check(n);
668         }
669     } else {
670         trace_pci_nvme_irq_masked();
671     }
672 }
673 
674 static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
675 {
676     if (cq->irq_enabled) {
677         if (msix_enabled(PCI_DEVICE(n))) {
678             return;
679         } else {
680             assert(cq->vector < 32);
681             if (!n->cq_pending) {
682                 n->irq_status &= ~(1 << cq->vector);
683             }
684             nvme_irq_check(n);
685         }
686     }
687 }
688 
689 static void nvme_req_clear(NvmeRequest *req)
690 {
691     req->ns = NULL;
692     req->opaque = NULL;
693     req->aiocb = NULL;
694     memset(&req->cqe, 0x0, sizeof(req->cqe));
695     req->status = NVME_SUCCESS;
696 }
697 
698 static inline void nvme_sg_init(NvmeCtrl *n, NvmeSg *sg, bool dma)
699 {
700     if (dma) {
701         pci_dma_sglist_init(&sg->qsg, PCI_DEVICE(n), 0);
702         sg->flags = NVME_SG_DMA;
703     } else {
704         qemu_iovec_init(&sg->iov, 0);
705     }
706 
707     sg->flags |= NVME_SG_ALLOC;
708 }
709 
710 static inline void nvme_sg_unmap(NvmeSg *sg)
711 {
712     if (!(sg->flags & NVME_SG_ALLOC)) {
713         return;
714     }
715 
716     if (sg->flags & NVME_SG_DMA) {
717         qemu_sglist_destroy(&sg->qsg);
718     } else {
719         qemu_iovec_destroy(&sg->iov);
720     }
721 
722     memset(sg, 0x0, sizeof(*sg));
723 }
724 
725 /*
726  * When metadata is transfered as extended LBAs, the DPTR mapped into `sg`
727  * holds both data and metadata. This function splits the data and metadata
728  * into two separate QSG/IOVs.
729  */
730 static void nvme_sg_split(NvmeSg *sg, NvmeNamespace *ns, NvmeSg *data,
731                           NvmeSg *mdata)
732 {
733     NvmeSg *dst = data;
734     uint32_t trans_len, count = ns->lbasz;
735     uint64_t offset = 0;
736     bool dma = sg->flags & NVME_SG_DMA;
737     size_t sge_len;
738     size_t sg_len = dma ? sg->qsg.size : sg->iov.size;
739     int sg_idx = 0;
740 
741     assert(sg->flags & NVME_SG_ALLOC);
742 
743     while (sg_len) {
744         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
745 
746         trans_len = MIN(sg_len, count);
747         trans_len = MIN(trans_len, sge_len - offset);
748 
749         if (dst) {
750             if (dma) {
751                 qemu_sglist_add(&dst->qsg, sg->qsg.sg[sg_idx].base + offset,
752                                 trans_len);
753             } else {
754                 qemu_iovec_add(&dst->iov,
755                                sg->iov.iov[sg_idx].iov_base + offset,
756                                trans_len);
757             }
758         }
759 
760         sg_len -= trans_len;
761         count -= trans_len;
762         offset += trans_len;
763 
764         if (count == 0) {
765             dst = (dst == data) ? mdata : data;
766             count = (dst == data) ? ns->lbasz : ns->lbaf.ms;
767         }
768 
769         if (sge_len == offset) {
770             offset = 0;
771             sg_idx++;
772         }
773     }
774 }
775 
776 static uint16_t nvme_map_addr_cmb(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
777                                   size_t len)
778 {
779     if (!len) {
780         return NVME_SUCCESS;
781     }
782 
783     trace_pci_nvme_map_addr_cmb(addr, len);
784 
785     if (!nvme_addr_is_cmb(n, addr) || !nvme_addr_is_cmb(n, addr + len - 1)) {
786         return NVME_DATA_TRAS_ERROR;
787     }
788 
789     qemu_iovec_add(iov, nvme_addr_to_cmb(n, addr), len);
790 
791     return NVME_SUCCESS;
792 }
793 
794 static uint16_t nvme_map_addr_pmr(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
795                                   size_t len)
796 {
797     if (!len) {
798         return NVME_SUCCESS;
799     }
800 
801     if (!nvme_addr_is_pmr(n, addr) || !nvme_addr_is_pmr(n, addr + len - 1)) {
802         return NVME_DATA_TRAS_ERROR;
803     }
804 
805     qemu_iovec_add(iov, nvme_addr_to_pmr(n, addr), len);
806 
807     return NVME_SUCCESS;
808 }
809 
810 static uint16_t nvme_map_addr(NvmeCtrl *n, NvmeSg *sg, hwaddr addr, size_t len)
811 {
812     bool cmb = false, pmr = false;
813 
814     if (!len) {
815         return NVME_SUCCESS;
816     }
817 
818     trace_pci_nvme_map_addr(addr, len);
819 
820     if (nvme_addr_is_iomem(n, addr)) {
821         return NVME_DATA_TRAS_ERROR;
822     }
823 
824     if (nvme_addr_is_cmb(n, addr)) {
825         cmb = true;
826     } else if (nvme_addr_is_pmr(n, addr)) {
827         pmr = true;
828     }
829 
830     if (cmb || pmr) {
831         if (sg->flags & NVME_SG_DMA) {
832             return NVME_INVALID_USE_OF_CMB | NVME_DNR;
833         }
834 
835         if (sg->iov.niov + 1 > IOV_MAX) {
836             goto max_mappings_exceeded;
837         }
838 
839         if (cmb) {
840             return nvme_map_addr_cmb(n, &sg->iov, addr, len);
841         } else {
842             return nvme_map_addr_pmr(n, &sg->iov, addr, len);
843         }
844     }
845 
846     if (!(sg->flags & NVME_SG_DMA)) {
847         return NVME_INVALID_USE_OF_CMB | NVME_DNR;
848     }
849 
850     if (sg->qsg.nsg + 1 > IOV_MAX) {
851         goto max_mappings_exceeded;
852     }
853 
854     qemu_sglist_add(&sg->qsg, addr, len);
855 
856     return NVME_SUCCESS;
857 
858 max_mappings_exceeded:
859     NVME_GUEST_ERR(pci_nvme_ub_too_many_mappings,
860                    "number of mappings exceed 1024");
861     return NVME_INTERNAL_DEV_ERROR | NVME_DNR;
862 }
863 
864 static inline bool nvme_addr_is_dma(NvmeCtrl *n, hwaddr addr)
865 {
866     return !(nvme_addr_is_cmb(n, addr) || nvme_addr_is_pmr(n, addr));
867 }
868 
869 static uint16_t nvme_map_prp(NvmeCtrl *n, NvmeSg *sg, uint64_t prp1,
870                              uint64_t prp2, uint32_t len)
871 {
872     hwaddr trans_len = n->page_size - (prp1 % n->page_size);
873     trans_len = MIN(len, trans_len);
874     int num_prps = (len >> n->page_bits) + 1;
875     uint16_t status;
876     int ret;
877 
878     trace_pci_nvme_map_prp(trans_len, len, prp1, prp2, num_prps);
879 
880     nvme_sg_init(n, sg, nvme_addr_is_dma(n, prp1));
881 
882     status = nvme_map_addr(n, sg, prp1, trans_len);
883     if (status) {
884         goto unmap;
885     }
886 
887     len -= trans_len;
888     if (len) {
889         if (len > n->page_size) {
890             uint64_t prp_list[n->max_prp_ents];
891             uint32_t nents, prp_trans;
892             int i = 0;
893 
894             /*
895              * The first PRP list entry, pointed to by PRP2 may contain offset.
896              * Hence, we need to calculate the number of entries in based on
897              * that offset.
898              */
899             nents = (n->page_size - (prp2 & (n->page_size - 1))) >> 3;
900             prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
901             ret = nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
902             if (ret) {
903                 trace_pci_nvme_err_addr_read(prp2);
904                 status = NVME_DATA_TRAS_ERROR;
905                 goto unmap;
906             }
907             while (len != 0) {
908                 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
909 
910                 if (i == nents - 1 && len > n->page_size) {
911                     if (unlikely(prp_ent & (n->page_size - 1))) {
912                         trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
913                         status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
914                         goto unmap;
915                     }
916 
917                     i = 0;
918                     nents = (len + n->page_size - 1) >> n->page_bits;
919                     nents = MIN(nents, n->max_prp_ents);
920                     prp_trans = nents * sizeof(uint64_t);
921                     ret = nvme_addr_read(n, prp_ent, (void *)prp_list,
922                                          prp_trans);
923                     if (ret) {
924                         trace_pci_nvme_err_addr_read(prp_ent);
925                         status = NVME_DATA_TRAS_ERROR;
926                         goto unmap;
927                     }
928                     prp_ent = le64_to_cpu(prp_list[i]);
929                 }
930 
931                 if (unlikely(prp_ent & (n->page_size - 1))) {
932                     trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
933                     status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
934                     goto unmap;
935                 }
936 
937                 trans_len = MIN(len, n->page_size);
938                 status = nvme_map_addr(n, sg, prp_ent, trans_len);
939                 if (status) {
940                     goto unmap;
941                 }
942 
943                 len -= trans_len;
944                 i++;
945             }
946         } else {
947             if (unlikely(prp2 & (n->page_size - 1))) {
948                 trace_pci_nvme_err_invalid_prp2_align(prp2);
949                 status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
950                 goto unmap;
951             }
952             status = nvme_map_addr(n, sg, prp2, len);
953             if (status) {
954                 goto unmap;
955             }
956         }
957     }
958 
959     return NVME_SUCCESS;
960 
961 unmap:
962     nvme_sg_unmap(sg);
963     return status;
964 }
965 
966 /*
967  * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
968  * number of bytes mapped in len.
969  */
970 static uint16_t nvme_map_sgl_data(NvmeCtrl *n, NvmeSg *sg,
971                                   NvmeSglDescriptor *segment, uint64_t nsgld,
972                                   size_t *len, NvmeCmd *cmd)
973 {
974     dma_addr_t addr, trans_len;
975     uint32_t dlen;
976     uint16_t status;
977 
978     for (int i = 0; i < nsgld; i++) {
979         uint8_t type = NVME_SGL_TYPE(segment[i].type);
980 
981         switch (type) {
982         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
983             break;
984         case NVME_SGL_DESCR_TYPE_SEGMENT:
985         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
986             return NVME_INVALID_NUM_SGL_DESCRS | NVME_DNR;
987         default:
988             return NVME_SGL_DESCR_TYPE_INVALID | NVME_DNR;
989         }
990 
991         dlen = le32_to_cpu(segment[i].len);
992 
993         if (!dlen) {
994             continue;
995         }
996 
997         if (*len == 0) {
998             /*
999              * All data has been mapped, but the SGL contains additional
1000              * segments and/or descriptors. The controller might accept
1001              * ignoring the rest of the SGL.
1002              */
1003             uint32_t sgls = le32_to_cpu(n->id_ctrl.sgls);
1004             if (sgls & NVME_CTRL_SGLS_EXCESS_LENGTH) {
1005                 break;
1006             }
1007 
1008             trace_pci_nvme_err_invalid_sgl_excess_length(dlen);
1009             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1010         }
1011 
1012         trans_len = MIN(*len, dlen);
1013 
1014         addr = le64_to_cpu(segment[i].addr);
1015 
1016         if (UINT64_MAX - addr < dlen) {
1017             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1018         }
1019 
1020         status = nvme_map_addr(n, sg, addr, trans_len);
1021         if (status) {
1022             return status;
1023         }
1024 
1025         *len -= trans_len;
1026     }
1027 
1028     return NVME_SUCCESS;
1029 }
1030 
1031 static uint16_t nvme_map_sgl(NvmeCtrl *n, NvmeSg *sg, NvmeSglDescriptor sgl,
1032                              size_t len, NvmeCmd *cmd)
1033 {
1034     /*
1035      * Read the segment in chunks of 256 descriptors (one 4k page) to avoid
1036      * dynamically allocating a potentially huge SGL. The spec allows the SGL
1037      * to be larger (as in number of bytes required to describe the SGL
1038      * descriptors and segment chain) than the command transfer size, so it is
1039      * not bounded by MDTS.
1040      */
1041     const int SEG_CHUNK_SIZE = 256;
1042 
1043     NvmeSglDescriptor segment[SEG_CHUNK_SIZE], *sgld, *last_sgld;
1044     uint64_t nsgld;
1045     uint32_t seg_len;
1046     uint16_t status;
1047     hwaddr addr;
1048     int ret;
1049 
1050     sgld = &sgl;
1051     addr = le64_to_cpu(sgl.addr);
1052 
1053     trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl.type), len);
1054 
1055     nvme_sg_init(n, sg, nvme_addr_is_dma(n, addr));
1056 
1057     /*
1058      * If the entire transfer can be described with a single data block it can
1059      * be mapped directly.
1060      */
1061     if (NVME_SGL_TYPE(sgl.type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1062         status = nvme_map_sgl_data(n, sg, sgld, 1, &len, cmd);
1063         if (status) {
1064             goto unmap;
1065         }
1066 
1067         goto out;
1068     }
1069 
1070     for (;;) {
1071         switch (NVME_SGL_TYPE(sgld->type)) {
1072         case NVME_SGL_DESCR_TYPE_SEGMENT:
1073         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
1074             break;
1075         default:
1076             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1077         }
1078 
1079         seg_len = le32_to_cpu(sgld->len);
1080 
1081         /* check the length of the (Last) Segment descriptor */
1082         if (!seg_len || seg_len & 0xf) {
1083             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1084         }
1085 
1086         if (UINT64_MAX - addr < seg_len) {
1087             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1088         }
1089 
1090         nsgld = seg_len / sizeof(NvmeSglDescriptor);
1091 
1092         while (nsgld > SEG_CHUNK_SIZE) {
1093             if (nvme_addr_read(n, addr, segment, sizeof(segment))) {
1094                 trace_pci_nvme_err_addr_read(addr);
1095                 status = NVME_DATA_TRAS_ERROR;
1096                 goto unmap;
1097             }
1098 
1099             status = nvme_map_sgl_data(n, sg, segment, SEG_CHUNK_SIZE,
1100                                        &len, cmd);
1101             if (status) {
1102                 goto unmap;
1103             }
1104 
1105             nsgld -= SEG_CHUNK_SIZE;
1106             addr += SEG_CHUNK_SIZE * sizeof(NvmeSglDescriptor);
1107         }
1108 
1109         ret = nvme_addr_read(n, addr, segment, nsgld *
1110                              sizeof(NvmeSglDescriptor));
1111         if (ret) {
1112             trace_pci_nvme_err_addr_read(addr);
1113             status = NVME_DATA_TRAS_ERROR;
1114             goto unmap;
1115         }
1116 
1117         last_sgld = &segment[nsgld - 1];
1118 
1119         /*
1120          * If the segment ends with a Data Block, then we are done.
1121          */
1122         if (NVME_SGL_TYPE(last_sgld->type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1123             status = nvme_map_sgl_data(n, sg, segment, nsgld, &len, cmd);
1124             if (status) {
1125                 goto unmap;
1126             }
1127 
1128             goto out;
1129         }
1130 
1131         /*
1132          * If the last descriptor was not a Data Block, then the current
1133          * segment must not be a Last Segment.
1134          */
1135         if (NVME_SGL_TYPE(sgld->type) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT) {
1136             status = NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1137             goto unmap;
1138         }
1139 
1140         sgld = last_sgld;
1141         addr = le64_to_cpu(sgld->addr);
1142 
1143         /*
1144          * Do not map the last descriptor; it will be a Segment or Last Segment
1145          * descriptor and is handled by the next iteration.
1146          */
1147         status = nvme_map_sgl_data(n, sg, segment, nsgld - 1, &len, cmd);
1148         if (status) {
1149             goto unmap;
1150         }
1151     }
1152 
1153 out:
1154     /* if there is any residual left in len, the SGL was too short */
1155     if (len) {
1156         status = NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1157         goto unmap;
1158     }
1159 
1160     return NVME_SUCCESS;
1161 
1162 unmap:
1163     nvme_sg_unmap(sg);
1164     return status;
1165 }
1166 
1167 uint16_t nvme_map_dptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1168                        NvmeCmd *cmd)
1169 {
1170     uint64_t prp1, prp2;
1171 
1172     switch (NVME_CMD_FLAGS_PSDT(cmd->flags)) {
1173     case NVME_PSDT_PRP:
1174         prp1 = le64_to_cpu(cmd->dptr.prp1);
1175         prp2 = le64_to_cpu(cmd->dptr.prp2);
1176 
1177         return nvme_map_prp(n, sg, prp1, prp2, len);
1178     case NVME_PSDT_SGL_MPTR_CONTIGUOUS:
1179     case NVME_PSDT_SGL_MPTR_SGL:
1180         return nvme_map_sgl(n, sg, cmd->dptr.sgl, len, cmd);
1181     default:
1182         return NVME_INVALID_FIELD;
1183     }
1184 }
1185 
1186 static uint16_t nvme_map_mptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1187                               NvmeCmd *cmd)
1188 {
1189     int psdt = NVME_CMD_FLAGS_PSDT(cmd->flags);
1190     hwaddr mptr = le64_to_cpu(cmd->mptr);
1191     uint16_t status;
1192 
1193     if (psdt == NVME_PSDT_SGL_MPTR_SGL) {
1194         NvmeSglDescriptor sgl;
1195 
1196         if (nvme_addr_read(n, mptr, &sgl, sizeof(sgl))) {
1197             return NVME_DATA_TRAS_ERROR;
1198         }
1199 
1200         status = nvme_map_sgl(n, sg, sgl, len, cmd);
1201         if (status && (status & 0x7ff) == NVME_DATA_SGL_LEN_INVALID) {
1202             status = NVME_MD_SGL_LEN_INVALID | NVME_DNR;
1203         }
1204 
1205         return status;
1206     }
1207 
1208     nvme_sg_init(n, sg, nvme_addr_is_dma(n, mptr));
1209     status = nvme_map_addr(n, sg, mptr, len);
1210     if (status) {
1211         nvme_sg_unmap(sg);
1212     }
1213 
1214     return status;
1215 }
1216 
1217 static uint16_t nvme_map_data(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1218 {
1219     NvmeNamespace *ns = req->ns;
1220     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1221     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1222     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1223     size_t len = nvme_l2b(ns, nlb);
1224     uint16_t status;
1225 
1226     if (nvme_ns_ext(ns) &&
1227         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1228         NvmeSg sg;
1229 
1230         len += nvme_m2b(ns, nlb);
1231 
1232         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1233         if (status) {
1234             return status;
1235         }
1236 
1237         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1238         nvme_sg_split(&sg, ns, &req->sg, NULL);
1239         nvme_sg_unmap(&sg);
1240 
1241         return NVME_SUCCESS;
1242     }
1243 
1244     return nvme_map_dptr(n, &req->sg, len, &req->cmd);
1245 }
1246 
1247 static uint16_t nvme_map_mdata(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1248 {
1249     NvmeNamespace *ns = req->ns;
1250     size_t len = nvme_m2b(ns, nlb);
1251     uint16_t status;
1252 
1253     if (nvme_ns_ext(ns)) {
1254         NvmeSg sg;
1255 
1256         len += nvme_l2b(ns, nlb);
1257 
1258         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1259         if (status) {
1260             return status;
1261         }
1262 
1263         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1264         nvme_sg_split(&sg, ns, NULL, &req->sg);
1265         nvme_sg_unmap(&sg);
1266 
1267         return NVME_SUCCESS;
1268     }
1269 
1270     return nvme_map_mptr(n, &req->sg, len, &req->cmd);
1271 }
1272 
1273 static uint16_t nvme_tx_interleaved(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr,
1274                                     uint32_t len, uint32_t bytes,
1275                                     int32_t skip_bytes, int64_t offset,
1276                                     NvmeTxDirection dir)
1277 {
1278     hwaddr addr;
1279     uint32_t trans_len, count = bytes;
1280     bool dma = sg->flags & NVME_SG_DMA;
1281     int64_t sge_len;
1282     int sg_idx = 0;
1283     int ret;
1284 
1285     assert(sg->flags & NVME_SG_ALLOC);
1286 
1287     while (len) {
1288         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
1289 
1290         if (sge_len - offset < 0) {
1291             offset -= sge_len;
1292             sg_idx++;
1293             continue;
1294         }
1295 
1296         if (sge_len == offset) {
1297             offset = 0;
1298             sg_idx++;
1299             continue;
1300         }
1301 
1302         trans_len = MIN(len, count);
1303         trans_len = MIN(trans_len, sge_len - offset);
1304 
1305         if (dma) {
1306             addr = sg->qsg.sg[sg_idx].base + offset;
1307         } else {
1308             addr = (hwaddr)(uintptr_t)sg->iov.iov[sg_idx].iov_base + offset;
1309         }
1310 
1311         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1312             ret = nvme_addr_read(n, addr, ptr, trans_len);
1313         } else {
1314             ret = nvme_addr_write(n, addr, ptr, trans_len);
1315         }
1316 
1317         if (ret) {
1318             return NVME_DATA_TRAS_ERROR;
1319         }
1320 
1321         ptr += trans_len;
1322         len -= trans_len;
1323         count -= trans_len;
1324         offset += trans_len;
1325 
1326         if (count == 0) {
1327             count = bytes;
1328             offset += skip_bytes;
1329         }
1330     }
1331 
1332     return NVME_SUCCESS;
1333 }
1334 
1335 static uint16_t nvme_tx(NvmeCtrl *n, NvmeSg *sg, void *ptr, uint32_t len,
1336                         NvmeTxDirection dir)
1337 {
1338     assert(sg->flags & NVME_SG_ALLOC);
1339 
1340     if (sg->flags & NVME_SG_DMA) {
1341         const MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
1342         dma_addr_t residual;
1343 
1344         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1345             dma_buf_write(ptr, len, &residual, &sg->qsg, attrs);
1346         } else {
1347             dma_buf_read(ptr, len, &residual, &sg->qsg, attrs);
1348         }
1349 
1350         if (unlikely(residual)) {
1351             trace_pci_nvme_err_invalid_dma();
1352             return NVME_INVALID_FIELD | NVME_DNR;
1353         }
1354     } else {
1355         size_t bytes;
1356 
1357         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1358             bytes = qemu_iovec_to_buf(&sg->iov, 0, ptr, len);
1359         } else {
1360             bytes = qemu_iovec_from_buf(&sg->iov, 0, ptr, len);
1361         }
1362 
1363         if (unlikely(bytes != len)) {
1364             trace_pci_nvme_err_invalid_dma();
1365             return NVME_INVALID_FIELD | NVME_DNR;
1366         }
1367     }
1368 
1369     return NVME_SUCCESS;
1370 }
1371 
1372 static inline uint16_t nvme_c2h(NvmeCtrl *n, void *ptr, uint32_t len,
1373                                 NvmeRequest *req)
1374 {
1375     uint16_t status;
1376 
1377     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1378     if (status) {
1379         return status;
1380     }
1381 
1382     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_FROM_DEVICE);
1383 }
1384 
1385 static inline uint16_t nvme_h2c(NvmeCtrl *n, void *ptr, uint32_t len,
1386                                 NvmeRequest *req)
1387 {
1388     uint16_t status;
1389 
1390     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1391     if (status) {
1392         return status;
1393     }
1394 
1395     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_TO_DEVICE);
1396 }
1397 
1398 uint16_t nvme_bounce_data(NvmeCtrl *n, void *ptr, uint32_t len,
1399                           NvmeTxDirection dir, NvmeRequest *req)
1400 {
1401     NvmeNamespace *ns = req->ns;
1402     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1403     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1404     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1405 
1406     if (nvme_ns_ext(ns) &&
1407         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1408         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbasz,
1409                                    ns->lbaf.ms, 0, dir);
1410     }
1411 
1412     return nvme_tx(n, &req->sg, ptr, len, dir);
1413 }
1414 
1415 uint16_t nvme_bounce_mdata(NvmeCtrl *n, void *ptr, uint32_t len,
1416                            NvmeTxDirection dir, NvmeRequest *req)
1417 {
1418     NvmeNamespace *ns = req->ns;
1419     uint16_t status;
1420 
1421     if (nvme_ns_ext(ns)) {
1422         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbaf.ms,
1423                                    ns->lbasz, ns->lbasz, dir);
1424     }
1425 
1426     nvme_sg_unmap(&req->sg);
1427 
1428     status = nvme_map_mptr(n, &req->sg, len, &req->cmd);
1429     if (status) {
1430         return status;
1431     }
1432 
1433     return nvme_tx(n, &req->sg, ptr, len, dir);
1434 }
1435 
1436 static inline void nvme_blk_read(BlockBackend *blk, int64_t offset,
1437                                  uint32_t align, BlockCompletionFunc *cb,
1438                                  NvmeRequest *req)
1439 {
1440     assert(req->sg.flags & NVME_SG_ALLOC);
1441 
1442     if (req->sg.flags & NVME_SG_DMA) {
1443         req->aiocb = dma_blk_read(blk, &req->sg.qsg, offset, align, cb, req);
1444     } else {
1445         req->aiocb = blk_aio_preadv(blk, offset, &req->sg.iov, 0, cb, req);
1446     }
1447 }
1448 
1449 static inline void nvme_blk_write(BlockBackend *blk, int64_t offset,
1450                                   uint32_t align, BlockCompletionFunc *cb,
1451                                   NvmeRequest *req)
1452 {
1453     assert(req->sg.flags & NVME_SG_ALLOC);
1454 
1455     if (req->sg.flags & NVME_SG_DMA) {
1456         req->aiocb = dma_blk_write(blk, &req->sg.qsg, offset, align, cb, req);
1457     } else {
1458         req->aiocb = blk_aio_pwritev(blk, offset, &req->sg.iov, 0, cb, req);
1459     }
1460 }
1461 
1462 static void nvme_update_cq_eventidx(const NvmeCQueue *cq)
1463 {
1464     uint32_t v = cpu_to_le32(cq->head);
1465 
1466     trace_pci_nvme_update_cq_eventidx(cq->cqid, cq->head);
1467 
1468     pci_dma_write(PCI_DEVICE(cq->ctrl), cq->ei_addr, &v, sizeof(v));
1469 }
1470 
1471 static void nvme_update_cq_head(NvmeCQueue *cq)
1472 {
1473     uint32_t v;
1474 
1475     pci_dma_read(PCI_DEVICE(cq->ctrl), cq->db_addr, &v, sizeof(v));
1476 
1477     cq->head = le32_to_cpu(v);
1478 
1479     trace_pci_nvme_update_cq_head(cq->cqid, cq->head);
1480 }
1481 
1482 static void nvme_post_cqes(void *opaque)
1483 {
1484     NvmeCQueue *cq = opaque;
1485     NvmeCtrl *n = cq->ctrl;
1486     NvmeRequest *req, *next;
1487     bool pending = cq->head != cq->tail;
1488     int ret;
1489 
1490     QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
1491         NvmeSQueue *sq;
1492         hwaddr addr;
1493 
1494         if (n->dbbuf_enabled) {
1495             nvme_update_cq_eventidx(cq);
1496             nvme_update_cq_head(cq);
1497         }
1498 
1499         if (nvme_cq_full(cq)) {
1500             break;
1501         }
1502 
1503         sq = req->sq;
1504         req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
1505         req->cqe.sq_id = cpu_to_le16(sq->sqid);
1506         req->cqe.sq_head = cpu_to_le16(sq->head);
1507         addr = cq->dma_addr + cq->tail * n->cqe_size;
1508         ret = pci_dma_write(PCI_DEVICE(n), addr, (void *)&req->cqe,
1509                             sizeof(req->cqe));
1510         if (ret) {
1511             trace_pci_nvme_err_addr_write(addr);
1512             trace_pci_nvme_err_cfs();
1513             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
1514             break;
1515         }
1516         QTAILQ_REMOVE(&cq->req_list, req, entry);
1517         nvme_inc_cq_tail(cq);
1518         nvme_sg_unmap(&req->sg);
1519         QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
1520     }
1521     if (cq->tail != cq->head) {
1522         if (cq->irq_enabled && !pending) {
1523             n->cq_pending++;
1524         }
1525 
1526         nvme_irq_assert(n, cq);
1527     }
1528 }
1529 
1530 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
1531 {
1532     assert(cq->cqid == req->sq->cqid);
1533     trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
1534                                           le32_to_cpu(req->cqe.result),
1535                                           le32_to_cpu(req->cqe.dw1),
1536                                           req->status);
1537 
1538     if (req->status) {
1539         trace_pci_nvme_err_req_status(nvme_cid(req), nvme_nsid(req->ns),
1540                                       req->status, req->cmd.opcode);
1541     }
1542 
1543     QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
1544     QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
1545 
1546     qemu_bh_schedule(cq->bh);
1547 }
1548 
1549 static void nvme_process_aers(void *opaque)
1550 {
1551     NvmeCtrl *n = opaque;
1552     NvmeAsyncEvent *event, *next;
1553 
1554     trace_pci_nvme_process_aers(n->aer_queued);
1555 
1556     QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
1557         NvmeRequest *req;
1558         NvmeAerResult *result;
1559 
1560         /* can't post cqe if there is nothing to complete */
1561         if (!n->outstanding_aers) {
1562             trace_pci_nvme_no_outstanding_aers();
1563             break;
1564         }
1565 
1566         /* ignore if masked (cqe posted, but event not cleared) */
1567         if (n->aer_mask & (1 << event->result.event_type)) {
1568             trace_pci_nvme_aer_masked(event->result.event_type, n->aer_mask);
1569             continue;
1570         }
1571 
1572         QTAILQ_REMOVE(&n->aer_queue, event, entry);
1573         n->aer_queued--;
1574 
1575         n->aer_mask |= 1 << event->result.event_type;
1576         n->outstanding_aers--;
1577 
1578         req = n->aer_reqs[n->outstanding_aers];
1579 
1580         result = (NvmeAerResult *) &req->cqe.result;
1581         result->event_type = event->result.event_type;
1582         result->event_info = event->result.event_info;
1583         result->log_page = event->result.log_page;
1584         g_free(event);
1585 
1586         trace_pci_nvme_aer_post_cqe(result->event_type, result->event_info,
1587                                     result->log_page);
1588 
1589         nvme_enqueue_req_completion(&n->admin_cq, req);
1590     }
1591 }
1592 
1593 static void nvme_enqueue_event(NvmeCtrl *n, uint8_t event_type,
1594                                uint8_t event_info, uint8_t log_page)
1595 {
1596     NvmeAsyncEvent *event;
1597 
1598     trace_pci_nvme_enqueue_event(event_type, event_info, log_page);
1599 
1600     if (n->aer_queued == n->params.aer_max_queued) {
1601         trace_pci_nvme_enqueue_event_noqueue(n->aer_queued);
1602         return;
1603     }
1604 
1605     event = g_new(NvmeAsyncEvent, 1);
1606     event->result = (NvmeAerResult) {
1607         .event_type = event_type,
1608         .event_info = event_info,
1609         .log_page   = log_page,
1610     };
1611 
1612     QTAILQ_INSERT_TAIL(&n->aer_queue, event, entry);
1613     n->aer_queued++;
1614 
1615     nvme_process_aers(n);
1616 }
1617 
1618 static void nvme_smart_event(NvmeCtrl *n, uint8_t event)
1619 {
1620     uint8_t aer_info;
1621 
1622     /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
1623     if (!(NVME_AEC_SMART(n->features.async_config) & event)) {
1624         return;
1625     }
1626 
1627     switch (event) {
1628     case NVME_SMART_SPARE:
1629         aer_info = NVME_AER_INFO_SMART_SPARE_THRESH;
1630         break;
1631     case NVME_SMART_TEMPERATURE:
1632         aer_info = NVME_AER_INFO_SMART_TEMP_THRESH;
1633         break;
1634     case NVME_SMART_RELIABILITY:
1635     case NVME_SMART_MEDIA_READ_ONLY:
1636     case NVME_SMART_FAILED_VOLATILE_MEDIA:
1637     case NVME_SMART_PMR_UNRELIABLE:
1638         aer_info = NVME_AER_INFO_SMART_RELIABILITY;
1639         break;
1640     default:
1641         return;
1642     }
1643 
1644     nvme_enqueue_event(n, NVME_AER_TYPE_SMART, aer_info, NVME_LOG_SMART_INFO);
1645 }
1646 
1647 static void nvme_clear_events(NvmeCtrl *n, uint8_t event_type)
1648 {
1649     n->aer_mask &= ~(1 << event_type);
1650     if (!QTAILQ_EMPTY(&n->aer_queue)) {
1651         nvme_process_aers(n);
1652     }
1653 }
1654 
1655 static inline uint16_t nvme_check_mdts(NvmeCtrl *n, size_t len)
1656 {
1657     uint8_t mdts = n->params.mdts;
1658 
1659     if (mdts && len > n->page_size << mdts) {
1660         trace_pci_nvme_err_mdts(len);
1661         return NVME_INVALID_FIELD | NVME_DNR;
1662     }
1663 
1664     return NVME_SUCCESS;
1665 }
1666 
1667 static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
1668                                          uint32_t nlb)
1669 {
1670     uint64_t nsze = le64_to_cpu(ns->id_ns.nsze);
1671 
1672     if (unlikely(UINT64_MAX - slba < nlb || slba + nlb > nsze)) {
1673         trace_pci_nvme_err_invalid_lba_range(slba, nlb, nsze);
1674         return NVME_LBA_RANGE | NVME_DNR;
1675     }
1676 
1677     return NVME_SUCCESS;
1678 }
1679 
1680 static int nvme_block_status_all(NvmeNamespace *ns, uint64_t slba,
1681                                  uint32_t nlb, int flags)
1682 {
1683     BlockDriverState *bs = blk_bs(ns->blkconf.blk);
1684 
1685     int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
1686     int64_t offset = nvme_l2b(ns, slba);
1687     int ret;
1688 
1689     /*
1690      * `pnum` holds the number of bytes after offset that shares the same
1691      * allocation status as the byte at offset. If `pnum` is different from
1692      * `bytes`, we should check the allocation status of the next range and
1693      * continue this until all bytes have been checked.
1694      */
1695     do {
1696         bytes -= pnum;
1697 
1698         ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
1699         if (ret < 0) {
1700             return ret;
1701         }
1702 
1703 
1704         trace_pci_nvme_block_status(offset, bytes, pnum, ret,
1705                                     !!(ret & BDRV_BLOCK_ZERO));
1706 
1707         if (!(ret & flags)) {
1708             return 1;
1709         }
1710 
1711         offset += pnum;
1712     } while (pnum != bytes);
1713 
1714     return 0;
1715 }
1716 
1717 static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
1718                                  uint32_t nlb)
1719 {
1720     int ret;
1721     Error *err = NULL;
1722 
1723     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_DATA);
1724     if (ret) {
1725         if (ret < 0) {
1726             error_setg_errno(&err, -ret, "unable to get block status");
1727             error_report_err(err);
1728 
1729             return NVME_INTERNAL_DEV_ERROR;
1730         }
1731 
1732         return NVME_DULB;
1733     }
1734 
1735     return NVME_SUCCESS;
1736 }
1737 
1738 static void nvme_aio_err(NvmeRequest *req, int ret)
1739 {
1740     uint16_t status = NVME_SUCCESS;
1741     Error *local_err = NULL;
1742 
1743     switch (req->cmd.opcode) {
1744     case NVME_CMD_READ:
1745         status = NVME_UNRECOVERED_READ;
1746         break;
1747     case NVME_CMD_FLUSH:
1748     case NVME_CMD_WRITE:
1749     case NVME_CMD_WRITE_ZEROES:
1750     case NVME_CMD_ZONE_APPEND:
1751         status = NVME_WRITE_FAULT;
1752         break;
1753     default:
1754         status = NVME_INTERNAL_DEV_ERROR;
1755         break;
1756     }
1757 
1758     trace_pci_nvme_err_aio(nvme_cid(req), strerror(-ret), status);
1759 
1760     error_setg_errno(&local_err, -ret, "aio failed");
1761     error_report_err(local_err);
1762 
1763     /*
1764      * Set the command status code to the first encountered error but allow a
1765      * subsequent Internal Device Error to trump it.
1766      */
1767     if (req->status && status != NVME_INTERNAL_DEV_ERROR) {
1768         return;
1769     }
1770 
1771     req->status = status;
1772 }
1773 
1774 static inline uint32_t nvme_zone_idx(NvmeNamespace *ns, uint64_t slba)
1775 {
1776     return ns->zone_size_log2 > 0 ? slba >> ns->zone_size_log2 :
1777                                     slba / ns->zone_size;
1778 }
1779 
1780 static inline NvmeZone *nvme_get_zone_by_slba(NvmeNamespace *ns, uint64_t slba)
1781 {
1782     uint32_t zone_idx = nvme_zone_idx(ns, slba);
1783 
1784     if (zone_idx >= ns->num_zones) {
1785         return NULL;
1786     }
1787 
1788     return &ns->zone_array[zone_idx];
1789 }
1790 
1791 static uint16_t nvme_check_zone_state_for_write(NvmeZone *zone)
1792 {
1793     uint64_t zslba = zone->d.zslba;
1794 
1795     switch (nvme_get_zone_state(zone)) {
1796     case NVME_ZONE_STATE_EMPTY:
1797     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1798     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1799     case NVME_ZONE_STATE_CLOSED:
1800         return NVME_SUCCESS;
1801     case NVME_ZONE_STATE_FULL:
1802         trace_pci_nvme_err_zone_is_full(zslba);
1803         return NVME_ZONE_FULL;
1804     case NVME_ZONE_STATE_OFFLINE:
1805         trace_pci_nvme_err_zone_is_offline(zslba);
1806         return NVME_ZONE_OFFLINE;
1807     case NVME_ZONE_STATE_READ_ONLY:
1808         trace_pci_nvme_err_zone_is_read_only(zslba);
1809         return NVME_ZONE_READ_ONLY;
1810     default:
1811         assert(false);
1812     }
1813 
1814     return NVME_INTERNAL_DEV_ERROR;
1815 }
1816 
1817 static uint16_t nvme_check_zone_write(NvmeNamespace *ns, NvmeZone *zone,
1818                                       uint64_t slba, uint32_t nlb)
1819 {
1820     uint64_t zcap = nvme_zone_wr_boundary(zone);
1821     uint16_t status;
1822 
1823     status = nvme_check_zone_state_for_write(zone);
1824     if (status) {
1825         return status;
1826     }
1827 
1828     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1829         uint64_t ezrwa = zone->w_ptr + 2 * ns->zns.zrwas;
1830 
1831         if (slba < zone->w_ptr || slba + nlb > ezrwa) {
1832             trace_pci_nvme_err_zone_invalid_write(slba, zone->w_ptr);
1833             return NVME_ZONE_INVALID_WRITE;
1834         }
1835     } else {
1836         if (unlikely(slba != zone->w_ptr)) {
1837             trace_pci_nvme_err_write_not_at_wp(slba, zone->d.zslba,
1838                                                zone->w_ptr);
1839             return NVME_ZONE_INVALID_WRITE;
1840         }
1841     }
1842 
1843     if (unlikely((slba + nlb) > zcap)) {
1844         trace_pci_nvme_err_zone_boundary(slba, nlb, zcap);
1845         return NVME_ZONE_BOUNDARY_ERROR;
1846     }
1847 
1848     return NVME_SUCCESS;
1849 }
1850 
1851 static uint16_t nvme_check_zone_state_for_read(NvmeZone *zone)
1852 {
1853     switch (nvme_get_zone_state(zone)) {
1854     case NVME_ZONE_STATE_EMPTY:
1855     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1856     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1857     case NVME_ZONE_STATE_FULL:
1858     case NVME_ZONE_STATE_CLOSED:
1859     case NVME_ZONE_STATE_READ_ONLY:
1860         return NVME_SUCCESS;
1861     case NVME_ZONE_STATE_OFFLINE:
1862         trace_pci_nvme_err_zone_is_offline(zone->d.zslba);
1863         return NVME_ZONE_OFFLINE;
1864     default:
1865         assert(false);
1866     }
1867 
1868     return NVME_INTERNAL_DEV_ERROR;
1869 }
1870 
1871 static uint16_t nvme_check_zone_read(NvmeNamespace *ns, uint64_t slba,
1872                                      uint32_t nlb)
1873 {
1874     NvmeZone *zone;
1875     uint64_t bndry, end;
1876     uint16_t status;
1877 
1878     zone = nvme_get_zone_by_slba(ns, slba);
1879     assert(zone);
1880 
1881     bndry = nvme_zone_rd_boundary(ns, zone);
1882     end = slba + nlb;
1883 
1884     status = nvme_check_zone_state_for_read(zone);
1885     if (status) {
1886         ;
1887     } else if (unlikely(end > bndry)) {
1888         if (!ns->params.cross_zone_read) {
1889             status = NVME_ZONE_BOUNDARY_ERROR;
1890         } else {
1891             /*
1892              * Read across zone boundary - check that all subsequent
1893              * zones that are being read have an appropriate state.
1894              */
1895             do {
1896                 zone++;
1897                 status = nvme_check_zone_state_for_read(zone);
1898                 if (status) {
1899                     break;
1900                 }
1901             } while (end > nvme_zone_rd_boundary(ns, zone));
1902         }
1903     }
1904 
1905     return status;
1906 }
1907 
1908 static uint16_t nvme_zrm_finish(NvmeNamespace *ns, NvmeZone *zone)
1909 {
1910     switch (nvme_get_zone_state(zone)) {
1911     case NVME_ZONE_STATE_FULL:
1912         return NVME_SUCCESS;
1913 
1914     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1915     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1916         nvme_aor_dec_open(ns);
1917         /* fallthrough */
1918     case NVME_ZONE_STATE_CLOSED:
1919         nvme_aor_dec_active(ns);
1920 
1921         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1922             zone->d.za &= ~NVME_ZA_ZRWA_VALID;
1923             if (ns->params.numzrwa) {
1924                 ns->zns.numzrwa++;
1925             }
1926         }
1927 
1928         /* fallthrough */
1929     case NVME_ZONE_STATE_EMPTY:
1930         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_FULL);
1931         return NVME_SUCCESS;
1932 
1933     default:
1934         return NVME_ZONE_INVAL_TRANSITION;
1935     }
1936 }
1937 
1938 static uint16_t nvme_zrm_close(NvmeNamespace *ns, NvmeZone *zone)
1939 {
1940     switch (nvme_get_zone_state(zone)) {
1941     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1942     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1943         nvme_aor_dec_open(ns);
1944         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
1945         /* fall through */
1946     case NVME_ZONE_STATE_CLOSED:
1947         return NVME_SUCCESS;
1948 
1949     default:
1950         return NVME_ZONE_INVAL_TRANSITION;
1951     }
1952 }
1953 
1954 static uint16_t nvme_zrm_reset(NvmeNamespace *ns, NvmeZone *zone)
1955 {
1956     switch (nvme_get_zone_state(zone)) {
1957     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1958     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1959         nvme_aor_dec_open(ns);
1960         /* fallthrough */
1961     case NVME_ZONE_STATE_CLOSED:
1962         nvme_aor_dec_active(ns);
1963 
1964         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1965             if (ns->params.numzrwa) {
1966                 ns->zns.numzrwa++;
1967             }
1968         }
1969 
1970         /* fallthrough */
1971     case NVME_ZONE_STATE_FULL:
1972         zone->w_ptr = zone->d.zslba;
1973         zone->d.wp = zone->w_ptr;
1974         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
1975         /* fallthrough */
1976     case NVME_ZONE_STATE_EMPTY:
1977         return NVME_SUCCESS;
1978 
1979     default:
1980         return NVME_ZONE_INVAL_TRANSITION;
1981     }
1982 }
1983 
1984 static void nvme_zrm_auto_transition_zone(NvmeNamespace *ns)
1985 {
1986     NvmeZone *zone;
1987 
1988     if (ns->params.max_open_zones &&
1989         ns->nr_open_zones == ns->params.max_open_zones) {
1990         zone = QTAILQ_FIRST(&ns->imp_open_zones);
1991         if (zone) {
1992             /*
1993              * Automatically close this implicitly open zone.
1994              */
1995             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
1996             nvme_zrm_close(ns, zone);
1997         }
1998     }
1999 }
2000 
2001 enum {
2002     NVME_ZRM_AUTO = 1 << 0,
2003     NVME_ZRM_ZRWA = 1 << 1,
2004 };
2005 
2006 static uint16_t nvme_zrm_open_flags(NvmeCtrl *n, NvmeNamespace *ns,
2007                                     NvmeZone *zone, int flags)
2008 {
2009     int act = 0;
2010     uint16_t status;
2011 
2012     switch (nvme_get_zone_state(zone)) {
2013     case NVME_ZONE_STATE_EMPTY:
2014         act = 1;
2015 
2016         /* fallthrough */
2017 
2018     case NVME_ZONE_STATE_CLOSED:
2019         if (n->params.auto_transition_zones) {
2020             nvme_zrm_auto_transition_zone(ns);
2021         }
2022         status = nvme_zns_check_resources(ns, act, 1,
2023                                           (flags & NVME_ZRM_ZRWA) ? 1 : 0);
2024         if (status) {
2025             return status;
2026         }
2027 
2028         if (act) {
2029             nvme_aor_inc_active(ns);
2030         }
2031 
2032         nvme_aor_inc_open(ns);
2033 
2034         if (flags & NVME_ZRM_AUTO) {
2035             nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_IMPLICITLY_OPEN);
2036             return NVME_SUCCESS;
2037         }
2038 
2039         /* fallthrough */
2040 
2041     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
2042         if (flags & NVME_ZRM_AUTO) {
2043             return NVME_SUCCESS;
2044         }
2045 
2046         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EXPLICITLY_OPEN);
2047 
2048         /* fallthrough */
2049 
2050     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
2051         if (flags & NVME_ZRM_ZRWA) {
2052             ns->zns.numzrwa--;
2053 
2054             zone->d.za |= NVME_ZA_ZRWA_VALID;
2055         }
2056 
2057         return NVME_SUCCESS;
2058 
2059     default:
2060         return NVME_ZONE_INVAL_TRANSITION;
2061     }
2062 }
2063 
2064 static inline uint16_t nvme_zrm_auto(NvmeCtrl *n, NvmeNamespace *ns,
2065                                      NvmeZone *zone)
2066 {
2067     return nvme_zrm_open_flags(n, ns, zone, NVME_ZRM_AUTO);
2068 }
2069 
2070 static void nvme_advance_zone_wp(NvmeNamespace *ns, NvmeZone *zone,
2071                                  uint32_t nlb)
2072 {
2073     zone->d.wp += nlb;
2074 
2075     if (zone->d.wp == nvme_zone_wr_boundary(zone)) {
2076         nvme_zrm_finish(ns, zone);
2077     }
2078 }
2079 
2080 static void nvme_zoned_zrwa_implicit_flush(NvmeNamespace *ns, NvmeZone *zone,
2081                                            uint32_t nlbc)
2082 {
2083     uint16_t nzrwafgs = DIV_ROUND_UP(nlbc, ns->zns.zrwafg);
2084 
2085     nlbc = nzrwafgs * ns->zns.zrwafg;
2086 
2087     trace_pci_nvme_zoned_zrwa_implicit_flush(zone->d.zslba, nlbc);
2088 
2089     zone->w_ptr += nlbc;
2090 
2091     nvme_advance_zone_wp(ns, zone, nlbc);
2092 }
2093 
2094 static void nvme_finalize_zoned_write(NvmeNamespace *ns, NvmeRequest *req)
2095 {
2096     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2097     NvmeZone *zone;
2098     uint64_t slba;
2099     uint32_t nlb;
2100 
2101     slba = le64_to_cpu(rw->slba);
2102     nlb = le16_to_cpu(rw->nlb) + 1;
2103     zone = nvme_get_zone_by_slba(ns, slba);
2104     assert(zone);
2105 
2106     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
2107         uint64_t ezrwa = zone->w_ptr + ns->zns.zrwas - 1;
2108         uint64_t elba = slba + nlb - 1;
2109 
2110         if (elba > ezrwa) {
2111             nvme_zoned_zrwa_implicit_flush(ns, zone, elba - ezrwa);
2112         }
2113 
2114         return;
2115     }
2116 
2117     nvme_advance_zone_wp(ns, zone, nlb);
2118 }
2119 
2120 static inline bool nvme_is_write(NvmeRequest *req)
2121 {
2122     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2123 
2124     return rw->opcode == NVME_CMD_WRITE ||
2125            rw->opcode == NVME_CMD_ZONE_APPEND ||
2126            rw->opcode == NVME_CMD_WRITE_ZEROES;
2127 }
2128 
2129 static AioContext *nvme_get_aio_context(BlockAIOCB *acb)
2130 {
2131     return qemu_get_aio_context();
2132 }
2133 
2134 static void nvme_misc_cb(void *opaque, int ret)
2135 {
2136     NvmeRequest *req = opaque;
2137 
2138     trace_pci_nvme_misc_cb(nvme_cid(req));
2139 
2140     if (ret) {
2141         nvme_aio_err(req, ret);
2142     }
2143 
2144     nvme_enqueue_req_completion(nvme_cq(req), req);
2145 }
2146 
2147 void nvme_rw_complete_cb(void *opaque, int ret)
2148 {
2149     NvmeRequest *req = opaque;
2150     NvmeNamespace *ns = req->ns;
2151     BlockBackend *blk = ns->blkconf.blk;
2152     BlockAcctCookie *acct = &req->acct;
2153     BlockAcctStats *stats = blk_get_stats(blk);
2154 
2155     trace_pci_nvme_rw_complete_cb(nvme_cid(req), blk_name(blk));
2156 
2157     if (ret) {
2158         block_acct_failed(stats, acct);
2159         nvme_aio_err(req, ret);
2160     } else {
2161         block_acct_done(stats, acct);
2162     }
2163 
2164     if (ns->params.zoned && nvme_is_write(req)) {
2165         nvme_finalize_zoned_write(ns, req);
2166     }
2167 
2168     nvme_enqueue_req_completion(nvme_cq(req), req);
2169 }
2170 
2171 static void nvme_rw_cb(void *opaque, int ret)
2172 {
2173     NvmeRequest *req = opaque;
2174     NvmeNamespace *ns = req->ns;
2175 
2176     BlockBackend *blk = ns->blkconf.blk;
2177 
2178     trace_pci_nvme_rw_cb(nvme_cid(req), blk_name(blk));
2179 
2180     if (ret) {
2181         goto out;
2182     }
2183 
2184     if (ns->lbaf.ms) {
2185         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2186         uint64_t slba = le64_to_cpu(rw->slba);
2187         uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
2188         uint64_t offset = nvme_moff(ns, slba);
2189 
2190         if (req->cmd.opcode == NVME_CMD_WRITE_ZEROES) {
2191             size_t mlen = nvme_m2b(ns, nlb);
2192 
2193             req->aiocb = blk_aio_pwrite_zeroes(blk, offset, mlen,
2194                                                BDRV_REQ_MAY_UNMAP,
2195                                                nvme_rw_complete_cb, req);
2196             return;
2197         }
2198 
2199         if (nvme_ns_ext(ns) || req->cmd.mptr) {
2200             uint16_t status;
2201 
2202             nvme_sg_unmap(&req->sg);
2203             status = nvme_map_mdata(nvme_ctrl(req), nlb, req);
2204             if (status) {
2205                 ret = -EFAULT;
2206                 goto out;
2207             }
2208 
2209             if (req->cmd.opcode == NVME_CMD_READ) {
2210                 return nvme_blk_read(blk, offset, 1, nvme_rw_complete_cb, req);
2211             }
2212 
2213             return nvme_blk_write(blk, offset, 1, nvme_rw_complete_cb, req);
2214         }
2215     }
2216 
2217 out:
2218     nvme_rw_complete_cb(req, ret);
2219 }
2220 
2221 static void nvme_verify_cb(void *opaque, int ret)
2222 {
2223     NvmeBounceContext *ctx = opaque;
2224     NvmeRequest *req = ctx->req;
2225     NvmeNamespace *ns = req->ns;
2226     BlockBackend *blk = ns->blkconf.blk;
2227     BlockAcctCookie *acct = &req->acct;
2228     BlockAcctStats *stats = blk_get_stats(blk);
2229     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2230     uint64_t slba = le64_to_cpu(rw->slba);
2231     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2232     uint16_t apptag = le16_to_cpu(rw->apptag);
2233     uint16_t appmask = le16_to_cpu(rw->appmask);
2234     uint64_t reftag = le32_to_cpu(rw->reftag);
2235     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2236     uint16_t status;
2237 
2238     reftag |= cdw3 << 32;
2239 
2240     trace_pci_nvme_verify_cb(nvme_cid(req), prinfo, apptag, appmask, reftag);
2241 
2242     if (ret) {
2243         block_acct_failed(stats, acct);
2244         nvme_aio_err(req, ret);
2245         goto out;
2246     }
2247 
2248     block_acct_done(stats, acct);
2249 
2250     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2251         status = nvme_dif_mangle_mdata(ns, ctx->mdata.bounce,
2252                                        ctx->mdata.iov.size, slba);
2253         if (status) {
2254             req->status = status;
2255             goto out;
2256         }
2257 
2258         req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2259                                      ctx->mdata.bounce, ctx->mdata.iov.size,
2260                                      prinfo, slba, apptag, appmask, &reftag);
2261     }
2262 
2263 out:
2264     qemu_iovec_destroy(&ctx->data.iov);
2265     g_free(ctx->data.bounce);
2266 
2267     qemu_iovec_destroy(&ctx->mdata.iov);
2268     g_free(ctx->mdata.bounce);
2269 
2270     g_free(ctx);
2271 
2272     nvme_enqueue_req_completion(nvme_cq(req), req);
2273 }
2274 
2275 
2276 static void nvme_verify_mdata_in_cb(void *opaque, int ret)
2277 {
2278     NvmeBounceContext *ctx = opaque;
2279     NvmeRequest *req = ctx->req;
2280     NvmeNamespace *ns = req->ns;
2281     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2282     uint64_t slba = le64_to_cpu(rw->slba);
2283     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2284     size_t mlen = nvme_m2b(ns, nlb);
2285     uint64_t offset = nvme_moff(ns, slba);
2286     BlockBackend *blk = ns->blkconf.blk;
2287 
2288     trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req), blk_name(blk));
2289 
2290     if (ret) {
2291         goto out;
2292     }
2293 
2294     ctx->mdata.bounce = g_malloc(mlen);
2295 
2296     qemu_iovec_reset(&ctx->mdata.iov);
2297     qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2298 
2299     req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2300                                 nvme_verify_cb, ctx);
2301     return;
2302 
2303 out:
2304     nvme_verify_cb(ctx, ret);
2305 }
2306 
2307 struct nvme_compare_ctx {
2308     struct {
2309         QEMUIOVector iov;
2310         uint8_t *bounce;
2311     } data;
2312 
2313     struct {
2314         QEMUIOVector iov;
2315         uint8_t *bounce;
2316     } mdata;
2317 };
2318 
2319 static void nvme_compare_mdata_cb(void *opaque, int ret)
2320 {
2321     NvmeRequest *req = opaque;
2322     NvmeNamespace *ns = req->ns;
2323     NvmeCtrl *n = nvme_ctrl(req);
2324     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2325     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2326     uint16_t apptag = le16_to_cpu(rw->apptag);
2327     uint16_t appmask = le16_to_cpu(rw->appmask);
2328     uint64_t reftag = le32_to_cpu(rw->reftag);
2329     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2330     struct nvme_compare_ctx *ctx = req->opaque;
2331     g_autofree uint8_t *buf = NULL;
2332     BlockBackend *blk = ns->blkconf.blk;
2333     BlockAcctCookie *acct = &req->acct;
2334     BlockAcctStats *stats = blk_get_stats(blk);
2335     uint16_t status = NVME_SUCCESS;
2336 
2337     reftag |= cdw3 << 32;
2338 
2339     trace_pci_nvme_compare_mdata_cb(nvme_cid(req));
2340 
2341     if (ret) {
2342         block_acct_failed(stats, acct);
2343         nvme_aio_err(req, ret);
2344         goto out;
2345     }
2346 
2347     buf = g_malloc(ctx->mdata.iov.size);
2348 
2349     status = nvme_bounce_mdata(n, buf, ctx->mdata.iov.size,
2350                                NVME_TX_DIRECTION_TO_DEVICE, req);
2351     if (status) {
2352         req->status = status;
2353         goto out;
2354     }
2355 
2356     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2357         uint64_t slba = le64_to_cpu(rw->slba);
2358         uint8_t *bufp;
2359         uint8_t *mbufp = ctx->mdata.bounce;
2360         uint8_t *end = mbufp + ctx->mdata.iov.size;
2361         int16_t pil = 0;
2362 
2363         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2364                                 ctx->mdata.bounce, ctx->mdata.iov.size, prinfo,
2365                                 slba, apptag, appmask, &reftag);
2366         if (status) {
2367             req->status = status;
2368             goto out;
2369         }
2370 
2371         /*
2372          * When formatted with protection information, do not compare the DIF
2373          * tuple.
2374          */
2375         if (!(ns->id_ns.dps & NVME_ID_NS_DPS_FIRST_EIGHT)) {
2376             pil = ns->lbaf.ms - nvme_pi_tuple_size(ns);
2377         }
2378 
2379         for (bufp = buf; mbufp < end; bufp += ns->lbaf.ms, mbufp += ns->lbaf.ms) {
2380             if (memcmp(bufp + pil, mbufp + pil, ns->lbaf.ms - pil)) {
2381                 req->status = NVME_CMP_FAILURE | NVME_DNR;
2382                 goto out;
2383             }
2384         }
2385 
2386         goto out;
2387     }
2388 
2389     if (memcmp(buf, ctx->mdata.bounce, ctx->mdata.iov.size)) {
2390         req->status = NVME_CMP_FAILURE | NVME_DNR;
2391         goto out;
2392     }
2393 
2394     block_acct_done(stats, acct);
2395 
2396 out:
2397     qemu_iovec_destroy(&ctx->data.iov);
2398     g_free(ctx->data.bounce);
2399 
2400     qemu_iovec_destroy(&ctx->mdata.iov);
2401     g_free(ctx->mdata.bounce);
2402 
2403     g_free(ctx);
2404 
2405     nvme_enqueue_req_completion(nvme_cq(req), req);
2406 }
2407 
2408 static void nvme_compare_data_cb(void *opaque, int ret)
2409 {
2410     NvmeRequest *req = opaque;
2411     NvmeCtrl *n = nvme_ctrl(req);
2412     NvmeNamespace *ns = req->ns;
2413     BlockBackend *blk = ns->blkconf.blk;
2414     BlockAcctCookie *acct = &req->acct;
2415     BlockAcctStats *stats = blk_get_stats(blk);
2416 
2417     struct nvme_compare_ctx *ctx = req->opaque;
2418     g_autofree uint8_t *buf = NULL;
2419     uint16_t status;
2420 
2421     trace_pci_nvme_compare_data_cb(nvme_cid(req));
2422 
2423     if (ret) {
2424         block_acct_failed(stats, acct);
2425         nvme_aio_err(req, ret);
2426         goto out;
2427     }
2428 
2429     buf = g_malloc(ctx->data.iov.size);
2430 
2431     status = nvme_bounce_data(n, buf, ctx->data.iov.size,
2432                               NVME_TX_DIRECTION_TO_DEVICE, req);
2433     if (status) {
2434         req->status = status;
2435         goto out;
2436     }
2437 
2438     if (memcmp(buf, ctx->data.bounce, ctx->data.iov.size)) {
2439         req->status = NVME_CMP_FAILURE | NVME_DNR;
2440         goto out;
2441     }
2442 
2443     if (ns->lbaf.ms) {
2444         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2445         uint64_t slba = le64_to_cpu(rw->slba);
2446         uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2447         size_t mlen = nvme_m2b(ns, nlb);
2448         uint64_t offset = nvme_moff(ns, slba);
2449 
2450         ctx->mdata.bounce = g_malloc(mlen);
2451 
2452         qemu_iovec_init(&ctx->mdata.iov, 1);
2453         qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2454 
2455         req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2456                                     nvme_compare_mdata_cb, req);
2457         return;
2458     }
2459 
2460     block_acct_done(stats, acct);
2461 
2462 out:
2463     qemu_iovec_destroy(&ctx->data.iov);
2464     g_free(ctx->data.bounce);
2465     g_free(ctx);
2466 
2467     nvme_enqueue_req_completion(nvme_cq(req), req);
2468 }
2469 
2470 typedef struct NvmeDSMAIOCB {
2471     BlockAIOCB common;
2472     BlockAIOCB *aiocb;
2473     NvmeRequest *req;
2474     int ret;
2475 
2476     NvmeDsmRange *range;
2477     unsigned int nr;
2478     unsigned int idx;
2479 } NvmeDSMAIOCB;
2480 
2481 static void nvme_dsm_cancel(BlockAIOCB *aiocb)
2482 {
2483     NvmeDSMAIOCB *iocb = container_of(aiocb, NvmeDSMAIOCB, common);
2484 
2485     /* break nvme_dsm_cb loop */
2486     iocb->idx = iocb->nr;
2487     iocb->ret = -ECANCELED;
2488 
2489     if (iocb->aiocb) {
2490         blk_aio_cancel_async(iocb->aiocb);
2491         iocb->aiocb = NULL;
2492     } else {
2493         /*
2494          * We only reach this if nvme_dsm_cancel() has already been called or
2495          * the command ran to completion.
2496          */
2497         assert(iocb->idx == iocb->nr);
2498     }
2499 }
2500 
2501 static const AIOCBInfo nvme_dsm_aiocb_info = {
2502     .aiocb_size   = sizeof(NvmeDSMAIOCB),
2503     .cancel_async = nvme_dsm_cancel,
2504 };
2505 
2506 static void nvme_dsm_cb(void *opaque, int ret);
2507 
2508 static void nvme_dsm_md_cb(void *opaque, int ret)
2509 {
2510     NvmeDSMAIOCB *iocb = opaque;
2511     NvmeRequest *req = iocb->req;
2512     NvmeNamespace *ns = req->ns;
2513     NvmeDsmRange *range;
2514     uint64_t slba;
2515     uint32_t nlb;
2516 
2517     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
2518         goto done;
2519     }
2520 
2521     range = &iocb->range[iocb->idx - 1];
2522     slba = le64_to_cpu(range->slba);
2523     nlb = le32_to_cpu(range->nlb);
2524 
2525     /*
2526      * Check that all block were discarded (zeroed); otherwise we do not zero
2527      * the metadata.
2528      */
2529 
2530     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_ZERO);
2531     if (ret) {
2532         if (ret < 0) {
2533             goto done;
2534         }
2535 
2536         nvme_dsm_cb(iocb, 0);
2537         return;
2538     }
2539 
2540     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_moff(ns, slba),
2541                                         nvme_m2b(ns, nlb), BDRV_REQ_MAY_UNMAP,
2542                                         nvme_dsm_cb, iocb);
2543     return;
2544 
2545 done:
2546     nvme_dsm_cb(iocb, ret);
2547 }
2548 
2549 static void nvme_dsm_cb(void *opaque, int ret)
2550 {
2551     NvmeDSMAIOCB *iocb = opaque;
2552     NvmeRequest *req = iocb->req;
2553     NvmeCtrl *n = nvme_ctrl(req);
2554     NvmeNamespace *ns = req->ns;
2555     NvmeDsmRange *range;
2556     uint64_t slba;
2557     uint32_t nlb;
2558 
2559     if (iocb->ret < 0) {
2560         goto done;
2561     } else if (ret < 0) {
2562         iocb->ret = ret;
2563         goto done;
2564     }
2565 
2566 next:
2567     if (iocb->idx == iocb->nr) {
2568         goto done;
2569     }
2570 
2571     range = &iocb->range[iocb->idx++];
2572     slba = le64_to_cpu(range->slba);
2573     nlb = le32_to_cpu(range->nlb);
2574 
2575     trace_pci_nvme_dsm_deallocate(slba, nlb);
2576 
2577     if (nlb > n->dmrsl) {
2578         trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
2579         goto next;
2580     }
2581 
2582     if (nvme_check_bounds(ns, slba, nlb)) {
2583         trace_pci_nvme_err_invalid_lba_range(slba, nlb,
2584                                              ns->id_ns.nsze);
2585         goto next;
2586     }
2587 
2588     iocb->aiocb = blk_aio_pdiscard(ns->blkconf.blk, nvme_l2b(ns, slba),
2589                                    nvme_l2b(ns, nlb),
2590                                    nvme_dsm_md_cb, iocb);
2591     return;
2592 
2593 done:
2594     iocb->aiocb = NULL;
2595     iocb->common.cb(iocb->common.opaque, iocb->ret);
2596     qemu_aio_unref(iocb);
2597 }
2598 
2599 static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
2600 {
2601     NvmeNamespace *ns = req->ns;
2602     NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
2603     uint32_t attr = le32_to_cpu(dsm->attributes);
2604     uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
2605     uint16_t status = NVME_SUCCESS;
2606 
2607     trace_pci_nvme_dsm(nr, attr);
2608 
2609     if (attr & NVME_DSMGMT_AD) {
2610         NvmeDSMAIOCB *iocb = blk_aio_get(&nvme_dsm_aiocb_info, ns->blkconf.blk,
2611                                          nvme_misc_cb, req);
2612 
2613         iocb->req = req;
2614         iocb->ret = 0;
2615         iocb->range = g_new(NvmeDsmRange, nr);
2616         iocb->nr = nr;
2617         iocb->idx = 0;
2618 
2619         status = nvme_h2c(n, (uint8_t *)iocb->range, sizeof(NvmeDsmRange) * nr,
2620                           req);
2621         if (status) {
2622             g_free(iocb->range);
2623             qemu_aio_unref(iocb);
2624 
2625             return status;
2626         }
2627 
2628         req->aiocb = &iocb->common;
2629         nvme_dsm_cb(iocb, 0);
2630 
2631         return NVME_NO_COMPLETE;
2632     }
2633 
2634     return status;
2635 }
2636 
2637 static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
2638 {
2639     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2640     NvmeNamespace *ns = req->ns;
2641     BlockBackend *blk = ns->blkconf.blk;
2642     uint64_t slba = le64_to_cpu(rw->slba);
2643     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2644     size_t len = nvme_l2b(ns, nlb);
2645     int64_t offset = nvme_l2b(ns, slba);
2646     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2647     uint32_t reftag = le32_to_cpu(rw->reftag);
2648     NvmeBounceContext *ctx = NULL;
2649     uint16_t status;
2650 
2651     trace_pci_nvme_verify(nvme_cid(req), nvme_nsid(ns), slba, nlb);
2652 
2653     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2654         status = nvme_check_prinfo(ns, prinfo, slba, reftag);
2655         if (status) {
2656             return status;
2657         }
2658 
2659         if (prinfo & NVME_PRINFO_PRACT) {
2660             return NVME_INVALID_PROT_INFO | NVME_DNR;
2661         }
2662     }
2663 
2664     if (len > n->page_size << n->params.vsl) {
2665         return NVME_INVALID_FIELD | NVME_DNR;
2666     }
2667 
2668     status = nvme_check_bounds(ns, slba, nlb);
2669     if (status) {
2670         return status;
2671     }
2672 
2673     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2674         status = nvme_check_dulbe(ns, slba, nlb);
2675         if (status) {
2676             return status;
2677         }
2678     }
2679 
2680     ctx = g_new0(NvmeBounceContext, 1);
2681     ctx->req = req;
2682 
2683     ctx->data.bounce = g_malloc(len);
2684 
2685     qemu_iovec_init(&ctx->data.iov, 1);
2686     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, len);
2687 
2688     block_acct_start(blk_get_stats(blk), &req->acct, ctx->data.iov.size,
2689                      BLOCK_ACCT_READ);
2690 
2691     req->aiocb = blk_aio_preadv(ns->blkconf.blk, offset, &ctx->data.iov, 0,
2692                                 nvme_verify_mdata_in_cb, ctx);
2693     return NVME_NO_COMPLETE;
2694 }
2695 
2696 typedef struct NvmeCopyAIOCB {
2697     BlockAIOCB common;
2698     BlockAIOCB *aiocb;
2699     NvmeRequest *req;
2700     int ret;
2701 
2702     void *ranges;
2703     unsigned int format;
2704     int nr;
2705     int idx;
2706 
2707     uint8_t *bounce;
2708     QEMUIOVector iov;
2709     struct {
2710         BlockAcctCookie read;
2711         BlockAcctCookie write;
2712     } acct;
2713 
2714     uint64_t reftag;
2715     uint64_t slba;
2716 
2717     NvmeZone *zone;
2718 } NvmeCopyAIOCB;
2719 
2720 static void nvme_copy_cancel(BlockAIOCB *aiocb)
2721 {
2722     NvmeCopyAIOCB *iocb = container_of(aiocb, NvmeCopyAIOCB, common);
2723 
2724     iocb->ret = -ECANCELED;
2725 
2726     if (iocb->aiocb) {
2727         blk_aio_cancel_async(iocb->aiocb);
2728         iocb->aiocb = NULL;
2729     }
2730 }
2731 
2732 static const AIOCBInfo nvme_copy_aiocb_info = {
2733     .aiocb_size   = sizeof(NvmeCopyAIOCB),
2734     .cancel_async = nvme_copy_cancel,
2735 };
2736 
2737 static void nvme_copy_done(NvmeCopyAIOCB *iocb)
2738 {
2739     NvmeRequest *req = iocb->req;
2740     NvmeNamespace *ns = req->ns;
2741     BlockAcctStats *stats = blk_get_stats(ns->blkconf.blk);
2742 
2743     if (iocb->idx != iocb->nr) {
2744         req->cqe.result = cpu_to_le32(iocb->idx);
2745     }
2746 
2747     qemu_iovec_destroy(&iocb->iov);
2748     g_free(iocb->bounce);
2749 
2750     if (iocb->ret < 0) {
2751         block_acct_failed(stats, &iocb->acct.read);
2752         block_acct_failed(stats, &iocb->acct.write);
2753     } else {
2754         block_acct_done(stats, &iocb->acct.read);
2755         block_acct_done(stats, &iocb->acct.write);
2756     }
2757 
2758     iocb->common.cb(iocb->common.opaque, iocb->ret);
2759     qemu_aio_unref(iocb);
2760 }
2761 
2762 static void nvme_do_copy(NvmeCopyAIOCB *iocb);
2763 
2764 static void nvme_copy_source_range_parse_format0(void *ranges, int idx,
2765                                                  uint64_t *slba, uint32_t *nlb,
2766                                                  uint16_t *apptag,
2767                                                  uint16_t *appmask,
2768                                                  uint64_t *reftag)
2769 {
2770     NvmeCopySourceRangeFormat0 *_ranges = ranges;
2771 
2772     if (slba) {
2773         *slba = le64_to_cpu(_ranges[idx].slba);
2774     }
2775 
2776     if (nlb) {
2777         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2778     }
2779 
2780     if (apptag) {
2781         *apptag = le16_to_cpu(_ranges[idx].apptag);
2782     }
2783 
2784     if (appmask) {
2785         *appmask = le16_to_cpu(_ranges[idx].appmask);
2786     }
2787 
2788     if (reftag) {
2789         *reftag = le32_to_cpu(_ranges[idx].reftag);
2790     }
2791 }
2792 
2793 static void nvme_copy_source_range_parse_format1(void *ranges, int idx,
2794                                                  uint64_t *slba, uint32_t *nlb,
2795                                                  uint16_t *apptag,
2796                                                  uint16_t *appmask,
2797                                                  uint64_t *reftag)
2798 {
2799     NvmeCopySourceRangeFormat1 *_ranges = ranges;
2800 
2801     if (slba) {
2802         *slba = le64_to_cpu(_ranges[idx].slba);
2803     }
2804 
2805     if (nlb) {
2806         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2807     }
2808 
2809     if (apptag) {
2810         *apptag = le16_to_cpu(_ranges[idx].apptag);
2811     }
2812 
2813     if (appmask) {
2814         *appmask = le16_to_cpu(_ranges[idx].appmask);
2815     }
2816 
2817     if (reftag) {
2818         *reftag = 0;
2819 
2820         *reftag |= (uint64_t)_ranges[idx].sr[4] << 40;
2821         *reftag |= (uint64_t)_ranges[idx].sr[5] << 32;
2822         *reftag |= (uint64_t)_ranges[idx].sr[6] << 24;
2823         *reftag |= (uint64_t)_ranges[idx].sr[7] << 16;
2824         *reftag |= (uint64_t)_ranges[idx].sr[8] << 8;
2825         *reftag |= (uint64_t)_ranges[idx].sr[9];
2826     }
2827 }
2828 
2829 static void nvme_copy_source_range_parse(void *ranges, int idx, uint8_t format,
2830                                          uint64_t *slba, uint32_t *nlb,
2831                                          uint16_t *apptag, uint16_t *appmask,
2832                                          uint64_t *reftag)
2833 {
2834     switch (format) {
2835     case NVME_COPY_FORMAT_0:
2836         nvme_copy_source_range_parse_format0(ranges, idx, slba, nlb, apptag,
2837                                              appmask, reftag);
2838         break;
2839 
2840     case NVME_COPY_FORMAT_1:
2841         nvme_copy_source_range_parse_format1(ranges, idx, slba, nlb, apptag,
2842                                              appmask, reftag);
2843         break;
2844 
2845     default:
2846         abort();
2847     }
2848 }
2849 
2850 static void nvme_copy_out_completed_cb(void *opaque, int ret)
2851 {
2852     NvmeCopyAIOCB *iocb = opaque;
2853     NvmeRequest *req = iocb->req;
2854     NvmeNamespace *ns = req->ns;
2855     uint32_t nlb;
2856 
2857     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2858                                  &nlb, NULL, NULL, NULL);
2859 
2860     if (ret < 0) {
2861         iocb->ret = ret;
2862         goto out;
2863     } else if (iocb->ret < 0) {
2864         goto out;
2865     }
2866 
2867     if (ns->params.zoned) {
2868         nvme_advance_zone_wp(ns, iocb->zone, nlb);
2869     }
2870 
2871     iocb->idx++;
2872     iocb->slba += nlb;
2873 out:
2874     nvme_do_copy(iocb);
2875 }
2876 
2877 static void nvme_copy_out_cb(void *opaque, int ret)
2878 {
2879     NvmeCopyAIOCB *iocb = opaque;
2880     NvmeRequest *req = iocb->req;
2881     NvmeNamespace *ns = req->ns;
2882     uint32_t nlb;
2883     size_t mlen;
2884     uint8_t *mbounce;
2885 
2886     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
2887         goto out;
2888     }
2889 
2890     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2891                                  &nlb, NULL, NULL, NULL);
2892 
2893     mlen = nvme_m2b(ns, nlb);
2894     mbounce = iocb->bounce + nvme_l2b(ns, nlb);
2895 
2896     qemu_iovec_reset(&iocb->iov);
2897     qemu_iovec_add(&iocb->iov, mbounce, mlen);
2898 
2899     iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_moff(ns, iocb->slba),
2900                                   &iocb->iov, 0, nvme_copy_out_completed_cb,
2901                                   iocb);
2902 
2903     return;
2904 
2905 out:
2906     nvme_copy_out_completed_cb(iocb, ret);
2907 }
2908 
2909 static void nvme_copy_in_completed_cb(void *opaque, int ret)
2910 {
2911     NvmeCopyAIOCB *iocb = opaque;
2912     NvmeRequest *req = iocb->req;
2913     NvmeNamespace *ns = req->ns;
2914     uint32_t nlb;
2915     uint64_t slba;
2916     uint16_t apptag, appmask;
2917     uint64_t reftag;
2918     size_t len;
2919     uint16_t status;
2920 
2921     if (ret < 0) {
2922         iocb->ret = ret;
2923         goto out;
2924     } else if (iocb->ret < 0) {
2925         goto out;
2926     }
2927 
2928     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
2929                                  &nlb, &apptag, &appmask, &reftag);
2930     len = nvme_l2b(ns, nlb);
2931 
2932     trace_pci_nvme_copy_out(iocb->slba, nlb);
2933 
2934     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2935         NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2936 
2937         uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
2938         uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
2939 
2940         size_t mlen = nvme_m2b(ns, nlb);
2941         uint8_t *mbounce = iocb->bounce + nvme_l2b(ns, nlb);
2942 
2943         status = nvme_dif_mangle_mdata(ns, mbounce, mlen, slba);
2944         if (status) {
2945             goto invalid;
2946         }
2947         status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen, prinfor,
2948                                 slba, apptag, appmask, &reftag);
2949         if (status) {
2950             goto invalid;
2951         }
2952 
2953         apptag = le16_to_cpu(copy->apptag);
2954         appmask = le16_to_cpu(copy->appmask);
2955 
2956         if (prinfow & NVME_PRINFO_PRACT) {
2957             status = nvme_check_prinfo(ns, prinfow, iocb->slba, iocb->reftag);
2958             if (status) {
2959                 goto invalid;
2960             }
2961 
2962             nvme_dif_pract_generate_dif(ns, iocb->bounce, len, mbounce, mlen,
2963                                         apptag, &iocb->reftag);
2964         } else {
2965             status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen,
2966                                     prinfow, iocb->slba, apptag, appmask,
2967                                     &iocb->reftag);
2968             if (status) {
2969                 goto invalid;
2970             }
2971         }
2972     }
2973 
2974     status = nvme_check_bounds(ns, iocb->slba, nlb);
2975     if (status) {
2976         goto invalid;
2977     }
2978 
2979     if (ns->params.zoned) {
2980         status = nvme_check_zone_write(ns, iocb->zone, iocb->slba, nlb);
2981         if (status) {
2982             goto invalid;
2983         }
2984 
2985         if (!(iocb->zone->d.za & NVME_ZA_ZRWA_VALID)) {
2986             iocb->zone->w_ptr += nlb;
2987         }
2988     }
2989 
2990     qemu_iovec_reset(&iocb->iov);
2991     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
2992 
2993     iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, iocb->slba),
2994                                   &iocb->iov, 0, nvme_copy_out_cb, iocb);
2995 
2996     return;
2997 
2998 invalid:
2999     req->status = status;
3000     iocb->ret = -1;
3001 out:
3002     nvme_do_copy(iocb);
3003 }
3004 
3005 static void nvme_copy_in_cb(void *opaque, int ret)
3006 {
3007     NvmeCopyAIOCB *iocb = opaque;
3008     NvmeRequest *req = iocb->req;
3009     NvmeNamespace *ns = req->ns;
3010     uint64_t slba;
3011     uint32_t nlb;
3012 
3013     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
3014         goto out;
3015     }
3016 
3017     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
3018                                  &nlb, NULL, NULL, NULL);
3019 
3020     qemu_iovec_reset(&iocb->iov);
3021     qemu_iovec_add(&iocb->iov, iocb->bounce + nvme_l2b(ns, nlb),
3022                    nvme_m2b(ns, nlb));
3023 
3024     iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_moff(ns, slba),
3025                                  &iocb->iov, 0, nvme_copy_in_completed_cb,
3026                                  iocb);
3027     return;
3028 
3029 out:
3030     nvme_copy_in_completed_cb(iocb, ret);
3031 }
3032 
3033 static void nvme_do_copy(NvmeCopyAIOCB *iocb)
3034 {
3035     NvmeRequest *req = iocb->req;
3036     NvmeNamespace *ns = req->ns;
3037     uint64_t slba;
3038     uint32_t nlb;
3039     size_t len;
3040     uint16_t status;
3041 
3042     if (iocb->ret < 0) {
3043         goto done;
3044     }
3045 
3046     if (iocb->idx == iocb->nr) {
3047         goto done;
3048     }
3049 
3050     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
3051                                  &nlb, NULL, NULL, NULL);
3052     len = nvme_l2b(ns, nlb);
3053 
3054     trace_pci_nvme_copy_source_range(slba, nlb);
3055 
3056     if (nlb > le16_to_cpu(ns->id_ns.mssrl)) {
3057         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3058         goto invalid;
3059     }
3060 
3061     status = nvme_check_bounds(ns, slba, nlb);
3062     if (status) {
3063         goto invalid;
3064     }
3065 
3066     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3067         status = nvme_check_dulbe(ns, slba, nlb);
3068         if (status) {
3069             goto invalid;
3070         }
3071     }
3072 
3073     if (ns->params.zoned) {
3074         status = nvme_check_zone_read(ns, slba, nlb);
3075         if (status) {
3076             goto invalid;
3077         }
3078     }
3079 
3080     qemu_iovec_reset(&iocb->iov);
3081     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3082 
3083     iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_l2b(ns, slba),
3084                                  &iocb->iov, 0, nvme_copy_in_cb, iocb);
3085     return;
3086 
3087 invalid:
3088     req->status = status;
3089     iocb->ret = -1;
3090 done:
3091     nvme_copy_done(iocb);
3092 }
3093 
3094 static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
3095 {
3096     NvmeNamespace *ns = req->ns;
3097     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
3098     NvmeCopyAIOCB *iocb = blk_aio_get(&nvme_copy_aiocb_info, ns->blkconf.blk,
3099                                       nvme_misc_cb, req);
3100     uint16_t nr = copy->nr + 1;
3101     uint8_t format = copy->control[0] & 0xf;
3102     uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
3103     uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
3104     size_t len = sizeof(NvmeCopySourceRangeFormat0);
3105 
3106     uint16_t status;
3107 
3108     trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);
3109 
3110     iocb->ranges = NULL;
3111     iocb->zone = NULL;
3112 
3113     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) &&
3114         ((prinfor & NVME_PRINFO_PRACT) != (prinfow & NVME_PRINFO_PRACT))) {
3115         status = NVME_INVALID_FIELD | NVME_DNR;
3116         goto invalid;
3117     }
3118 
3119     if (!(n->id_ctrl.ocfs & (1 << format))) {
3120         trace_pci_nvme_err_copy_invalid_format(format);
3121         status = NVME_INVALID_FIELD | NVME_DNR;
3122         goto invalid;
3123     }
3124 
3125     if (nr > ns->id_ns.msrc + 1) {
3126         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3127         goto invalid;
3128     }
3129 
3130     if ((ns->pif == 0x0 && format != 0x0) ||
3131         (ns->pif != 0x0 && format != 0x1)) {
3132         status = NVME_INVALID_FORMAT | NVME_DNR;
3133         goto invalid;
3134     }
3135 
3136     if (ns->pif) {
3137         len = sizeof(NvmeCopySourceRangeFormat1);
3138     }
3139 
3140     iocb->format = format;
3141     iocb->ranges = g_malloc_n(nr, len);
3142     status = nvme_h2c(n, (uint8_t *)iocb->ranges, len * nr, req);
3143     if (status) {
3144         goto invalid;
3145     }
3146 
3147     iocb->slba = le64_to_cpu(copy->sdlba);
3148 
3149     if (ns->params.zoned) {
3150         iocb->zone = nvme_get_zone_by_slba(ns, iocb->slba);
3151         if (!iocb->zone) {
3152             status = NVME_LBA_RANGE | NVME_DNR;
3153             goto invalid;
3154         }
3155 
3156         status = nvme_zrm_auto(n, ns, iocb->zone);
3157         if (status) {
3158             goto invalid;
3159         }
3160     }
3161 
3162     iocb->req = req;
3163     iocb->ret = 0;
3164     iocb->nr = nr;
3165     iocb->idx = 0;
3166     iocb->reftag = le32_to_cpu(copy->reftag);
3167     iocb->reftag |= (uint64_t)le32_to_cpu(copy->cdw3) << 32;
3168     iocb->bounce = g_malloc_n(le16_to_cpu(ns->id_ns.mssrl),
3169                               ns->lbasz + ns->lbaf.ms);
3170 
3171     qemu_iovec_init(&iocb->iov, 1);
3172 
3173     block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.read, 0,
3174                      BLOCK_ACCT_READ);
3175     block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.write, 0,
3176                      BLOCK_ACCT_WRITE);
3177 
3178     req->aiocb = &iocb->common;
3179     nvme_do_copy(iocb);
3180 
3181     return NVME_NO_COMPLETE;
3182 
3183 invalid:
3184     g_free(iocb->ranges);
3185     qemu_aio_unref(iocb);
3186     return status;
3187 }
3188 
3189 static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req)
3190 {
3191     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3192     NvmeNamespace *ns = req->ns;
3193     BlockBackend *blk = ns->blkconf.blk;
3194     uint64_t slba = le64_to_cpu(rw->slba);
3195     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
3196     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3197     size_t data_len = nvme_l2b(ns, nlb);
3198     size_t len = data_len;
3199     int64_t offset = nvme_l2b(ns, slba);
3200     struct nvme_compare_ctx *ctx = NULL;
3201     uint16_t status;
3202 
3203     trace_pci_nvme_compare(nvme_cid(req), nvme_nsid(ns), slba, nlb);
3204 
3205     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) && (prinfo & NVME_PRINFO_PRACT)) {
3206         return NVME_INVALID_PROT_INFO | NVME_DNR;
3207     }
3208 
3209     if (nvme_ns_ext(ns)) {
3210         len += nvme_m2b(ns, nlb);
3211     }
3212 
3213     status = nvme_check_mdts(n, len);
3214     if (status) {
3215         return status;
3216     }
3217 
3218     status = nvme_check_bounds(ns, slba, nlb);
3219     if (status) {
3220         return status;
3221     }
3222 
3223     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3224         status = nvme_check_dulbe(ns, slba, nlb);
3225         if (status) {
3226             return status;
3227         }
3228     }
3229 
3230     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
3231     if (status) {
3232         return status;
3233     }
3234 
3235     ctx = g_new(struct nvme_compare_ctx, 1);
3236     ctx->data.bounce = g_malloc(data_len);
3237 
3238     req->opaque = ctx;
3239 
3240     qemu_iovec_init(&ctx->data.iov, 1);
3241     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, data_len);
3242 
3243     block_acct_start(blk_get_stats(blk), &req->acct, data_len,
3244                      BLOCK_ACCT_READ);
3245     req->aiocb = blk_aio_preadv(blk, offset, &ctx->data.iov, 0,
3246                                 nvme_compare_data_cb, req);
3247 
3248     return NVME_NO_COMPLETE;
3249 }
3250 
3251 typedef struct NvmeFlushAIOCB {
3252     BlockAIOCB common;
3253     BlockAIOCB *aiocb;
3254     NvmeRequest *req;
3255     int ret;
3256 
3257     NvmeNamespace *ns;
3258     uint32_t nsid;
3259     bool broadcast;
3260 } NvmeFlushAIOCB;
3261 
3262 static void nvme_flush_cancel(BlockAIOCB *acb)
3263 {
3264     NvmeFlushAIOCB *iocb = container_of(acb, NvmeFlushAIOCB, common);
3265 
3266     iocb->ret = -ECANCELED;
3267 
3268     if (iocb->aiocb) {
3269         blk_aio_cancel_async(iocb->aiocb);
3270         iocb->aiocb = NULL;
3271     }
3272 }
3273 
3274 static const AIOCBInfo nvme_flush_aiocb_info = {
3275     .aiocb_size = sizeof(NvmeFlushAIOCB),
3276     .cancel_async = nvme_flush_cancel,
3277     .get_aio_context = nvme_get_aio_context,
3278 };
3279 
3280 static void nvme_do_flush(NvmeFlushAIOCB *iocb);
3281 
3282 static void nvme_flush_ns_cb(void *opaque, int ret)
3283 {
3284     NvmeFlushAIOCB *iocb = opaque;
3285     NvmeNamespace *ns = iocb->ns;
3286 
3287     if (ret < 0) {
3288         iocb->ret = ret;
3289         goto out;
3290     } else if (iocb->ret < 0) {
3291         goto out;
3292     }
3293 
3294     if (ns) {
3295         trace_pci_nvme_flush_ns(iocb->nsid);
3296 
3297         iocb->ns = NULL;
3298         iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);
3299         return;
3300     }
3301 
3302 out:
3303     nvme_do_flush(iocb);
3304 }
3305 
3306 static void nvme_do_flush(NvmeFlushAIOCB *iocb)
3307 {
3308     NvmeRequest *req = iocb->req;
3309     NvmeCtrl *n = nvme_ctrl(req);
3310     int i;
3311 
3312     if (iocb->ret < 0) {
3313         goto done;
3314     }
3315 
3316     if (iocb->broadcast) {
3317         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
3318             iocb->ns = nvme_ns(n, i);
3319             if (iocb->ns) {
3320                 iocb->nsid = i;
3321                 break;
3322             }
3323         }
3324     }
3325 
3326     if (!iocb->ns) {
3327         goto done;
3328     }
3329 
3330     nvme_flush_ns_cb(iocb, 0);
3331     return;
3332 
3333 done:
3334     iocb->common.cb(iocb->common.opaque, iocb->ret);
3335     qemu_aio_unref(iocb);
3336 }
3337 
3338 static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req)
3339 {
3340     NvmeFlushAIOCB *iocb;
3341     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
3342     uint16_t status;
3343 
3344     iocb = qemu_aio_get(&nvme_flush_aiocb_info, NULL, nvme_misc_cb, req);
3345 
3346     iocb->req = req;
3347     iocb->ret = 0;
3348     iocb->ns = NULL;
3349     iocb->nsid = 0;
3350     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
3351 
3352     if (!iocb->broadcast) {
3353         if (!nvme_nsid_valid(n, nsid)) {
3354             status = NVME_INVALID_NSID | NVME_DNR;
3355             goto out;
3356         }
3357 
3358         iocb->ns = nvme_ns(n, nsid);
3359         if (!iocb->ns) {
3360             status = NVME_INVALID_FIELD | NVME_DNR;
3361             goto out;
3362         }
3363 
3364         iocb->nsid = nsid;
3365     }
3366 
3367     req->aiocb = &iocb->common;
3368     nvme_do_flush(iocb);
3369 
3370     return NVME_NO_COMPLETE;
3371 
3372 out:
3373     qemu_aio_unref(iocb);
3374 
3375     return status;
3376 }
3377 
3378 static uint16_t nvme_read(NvmeCtrl *n, NvmeRequest *req)
3379 {
3380     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3381     NvmeNamespace *ns = req->ns;
3382     uint64_t slba = le64_to_cpu(rw->slba);
3383     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3384     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3385     uint64_t data_size = nvme_l2b(ns, nlb);
3386     uint64_t mapped_size = data_size;
3387     uint64_t data_offset;
3388     BlockBackend *blk = ns->blkconf.blk;
3389     uint16_t status;
3390 
3391     if (nvme_ns_ext(ns)) {
3392         mapped_size += nvme_m2b(ns, nlb);
3393 
3394         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3395             bool pract = prinfo & NVME_PRINFO_PRACT;
3396 
3397             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3398                 mapped_size = data_size;
3399             }
3400         }
3401     }
3402 
3403     trace_pci_nvme_read(nvme_cid(req), nvme_nsid(ns), nlb, mapped_size, slba);
3404 
3405     status = nvme_check_mdts(n, mapped_size);
3406     if (status) {
3407         goto invalid;
3408     }
3409 
3410     status = nvme_check_bounds(ns, slba, nlb);
3411     if (status) {
3412         goto invalid;
3413     }
3414 
3415     if (ns->params.zoned) {
3416         status = nvme_check_zone_read(ns, slba, nlb);
3417         if (status) {
3418             trace_pci_nvme_err_zone_read_not_ok(slba, nlb, status);
3419             goto invalid;
3420         }
3421     }
3422 
3423     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3424         status = nvme_check_dulbe(ns, slba, nlb);
3425         if (status) {
3426             goto invalid;
3427         }
3428     }
3429 
3430     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3431         return nvme_dif_rw(n, req);
3432     }
3433 
3434     status = nvme_map_data(n, nlb, req);
3435     if (status) {
3436         goto invalid;
3437     }
3438 
3439     data_offset = nvme_l2b(ns, slba);
3440 
3441     block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3442                      BLOCK_ACCT_READ);
3443     nvme_blk_read(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3444     return NVME_NO_COMPLETE;
3445 
3446 invalid:
3447     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_READ);
3448     return status | NVME_DNR;
3449 }
3450 
3451 static void nvme_do_write_fdp(NvmeCtrl *n, NvmeRequest *req, uint64_t slba,
3452                               uint32_t nlb)
3453 {
3454     NvmeNamespace *ns = req->ns;
3455     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3456     uint64_t data_size = nvme_l2b(ns, nlb);
3457     uint32_t dw12 = le32_to_cpu(req->cmd.cdw12);
3458     uint8_t dtype = (dw12 >> 20) & 0xf;
3459     uint16_t pid = le16_to_cpu(rw->dspec);
3460     uint16_t ph, rg, ruhid;
3461     NvmeReclaimUnit *ru;
3462 
3463     if (dtype != NVME_DIRECTIVE_DATA_PLACEMENT ||
3464         !nvme_parse_pid(ns, pid, &ph, &rg)) {
3465         ph = 0;
3466         rg = 0;
3467     }
3468 
3469     ruhid = ns->fdp.phs[ph];
3470     ru = &ns->endgrp->fdp.ruhs[ruhid].rus[rg];
3471 
3472     nvme_fdp_stat_inc(&ns->endgrp->fdp.hbmw, data_size);
3473     nvme_fdp_stat_inc(&ns->endgrp->fdp.mbmw, data_size);
3474 
3475     while (nlb) {
3476         if (nlb < ru->ruamw) {
3477             ru->ruamw -= nlb;
3478             break;
3479         }
3480 
3481         nlb -= ru->ruamw;
3482         nvme_update_ruh(n, ns, pid);
3483     }
3484 }
3485 
3486 static uint16_t nvme_do_write(NvmeCtrl *n, NvmeRequest *req, bool append,
3487                               bool wrz)
3488 {
3489     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3490     NvmeNamespace *ns = req->ns;
3491     uint64_t slba = le64_to_cpu(rw->slba);
3492     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3493     uint16_t ctrl = le16_to_cpu(rw->control);
3494     uint8_t prinfo = NVME_RW_PRINFO(ctrl);
3495     uint64_t data_size = nvme_l2b(ns, nlb);
3496     uint64_t mapped_size = data_size;
3497     uint64_t data_offset;
3498     NvmeZone *zone;
3499     NvmeZonedResult *res = (NvmeZonedResult *)&req->cqe;
3500     BlockBackend *blk = ns->blkconf.blk;
3501     uint16_t status;
3502 
3503     if (nvme_ns_ext(ns)) {
3504         mapped_size += nvme_m2b(ns, nlb);
3505 
3506         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3507             bool pract = prinfo & NVME_PRINFO_PRACT;
3508 
3509             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3510                 mapped_size -= nvme_m2b(ns, nlb);
3511             }
3512         }
3513     }
3514 
3515     trace_pci_nvme_write(nvme_cid(req), nvme_io_opc_str(rw->opcode),
3516                          nvme_nsid(ns), nlb, mapped_size, slba);
3517 
3518     if (!wrz) {
3519         status = nvme_check_mdts(n, mapped_size);
3520         if (status) {
3521             goto invalid;
3522         }
3523     }
3524 
3525     status = nvme_check_bounds(ns, slba, nlb);
3526     if (status) {
3527         goto invalid;
3528     }
3529 
3530     if (ns->params.zoned) {
3531         zone = nvme_get_zone_by_slba(ns, slba);
3532         assert(zone);
3533 
3534         if (append) {
3535             bool piremap = !!(ctrl & NVME_RW_PIREMAP);
3536 
3537             if (unlikely(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3538                 return NVME_INVALID_ZONE_OP | NVME_DNR;
3539             }
3540 
3541             if (unlikely(slba != zone->d.zslba)) {
3542                 trace_pci_nvme_err_append_not_at_start(slba, zone->d.zslba);
3543                 status = NVME_INVALID_FIELD;
3544                 goto invalid;
3545             }
3546 
3547             if (n->params.zasl &&
3548                 data_size > (uint64_t)n->page_size << n->params.zasl) {
3549                 trace_pci_nvme_err_zasl(data_size);
3550                 return NVME_INVALID_FIELD | NVME_DNR;
3551             }
3552 
3553             slba = zone->w_ptr;
3554             rw->slba = cpu_to_le64(slba);
3555             res->slba = cpu_to_le64(slba);
3556 
3557             switch (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3558             case NVME_ID_NS_DPS_TYPE_1:
3559                 if (!piremap) {
3560                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3561                 }
3562 
3563                 /* fallthrough */
3564 
3565             case NVME_ID_NS_DPS_TYPE_2:
3566                 if (piremap) {
3567                     uint32_t reftag = le32_to_cpu(rw->reftag);
3568                     rw->reftag = cpu_to_le32(reftag + (slba - zone->d.zslba));
3569                 }
3570 
3571                 break;
3572 
3573             case NVME_ID_NS_DPS_TYPE_3:
3574                 if (piremap) {
3575                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3576                 }
3577 
3578                 break;
3579             }
3580         }
3581 
3582         status = nvme_check_zone_write(ns, zone, slba, nlb);
3583         if (status) {
3584             goto invalid;
3585         }
3586 
3587         status = nvme_zrm_auto(n, ns, zone);
3588         if (status) {
3589             goto invalid;
3590         }
3591 
3592         if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3593             zone->w_ptr += nlb;
3594         }
3595     } else if (ns->endgrp && ns->endgrp->fdp.enabled) {
3596         nvme_do_write_fdp(n, req, slba, nlb);
3597     }
3598 
3599     data_offset = nvme_l2b(ns, slba);
3600 
3601     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3602         return nvme_dif_rw(n, req);
3603     }
3604 
3605     if (!wrz) {
3606         status = nvme_map_data(n, nlb, req);
3607         if (status) {
3608             goto invalid;
3609         }
3610 
3611         block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3612                          BLOCK_ACCT_WRITE);
3613         nvme_blk_write(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3614     } else {
3615         req->aiocb = blk_aio_pwrite_zeroes(blk, data_offset, data_size,
3616                                            BDRV_REQ_MAY_UNMAP, nvme_rw_cb,
3617                                            req);
3618     }
3619 
3620     return NVME_NO_COMPLETE;
3621 
3622 invalid:
3623     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_WRITE);
3624     return status | NVME_DNR;
3625 }
3626 
3627 static inline uint16_t nvme_write(NvmeCtrl *n, NvmeRequest *req)
3628 {
3629     return nvme_do_write(n, req, false, false);
3630 }
3631 
3632 static inline uint16_t nvme_write_zeroes(NvmeCtrl *n, NvmeRequest *req)
3633 {
3634     return nvme_do_write(n, req, false, true);
3635 }
3636 
3637 static inline uint16_t nvme_zone_append(NvmeCtrl *n, NvmeRequest *req)
3638 {
3639     return nvme_do_write(n, req, true, false);
3640 }
3641 
3642 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace *ns, NvmeCmd *c,
3643                                             uint64_t *slba, uint32_t *zone_idx)
3644 {
3645     uint32_t dw10 = le32_to_cpu(c->cdw10);
3646     uint32_t dw11 = le32_to_cpu(c->cdw11);
3647 
3648     if (!ns->params.zoned) {
3649         trace_pci_nvme_err_invalid_opc(c->opcode);
3650         return NVME_INVALID_OPCODE | NVME_DNR;
3651     }
3652 
3653     *slba = ((uint64_t)dw11) << 32 | dw10;
3654     if (unlikely(*slba >= ns->id_ns.nsze)) {
3655         trace_pci_nvme_err_invalid_lba_range(*slba, 0, ns->id_ns.nsze);
3656         *slba = 0;
3657         return NVME_LBA_RANGE | NVME_DNR;
3658     }
3659 
3660     *zone_idx = nvme_zone_idx(ns, *slba);
3661     assert(*zone_idx < ns->num_zones);
3662 
3663     return NVME_SUCCESS;
3664 }
3665 
3666 typedef uint16_t (*op_handler_t)(NvmeNamespace *, NvmeZone *, NvmeZoneState,
3667                                  NvmeRequest *);
3668 
3669 enum NvmeZoneProcessingMask {
3670     NVME_PROC_CURRENT_ZONE    = 0,
3671     NVME_PROC_OPENED_ZONES    = 1 << 0,
3672     NVME_PROC_CLOSED_ZONES    = 1 << 1,
3673     NVME_PROC_READ_ONLY_ZONES = 1 << 2,
3674     NVME_PROC_FULL_ZONES      = 1 << 3,
3675 };
3676 
3677 static uint16_t nvme_open_zone(NvmeNamespace *ns, NvmeZone *zone,
3678                                NvmeZoneState state, NvmeRequest *req)
3679 {
3680     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
3681     int flags = 0;
3682 
3683     if (cmd->zsflags & NVME_ZSFLAG_ZRWA_ALLOC) {
3684         uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3685 
3686         if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3687             return NVME_INVALID_ZONE_OP | NVME_DNR;
3688         }
3689 
3690         if (zone->w_ptr % ns->zns.zrwafg) {
3691             return NVME_NOZRWA | NVME_DNR;
3692         }
3693 
3694         flags = NVME_ZRM_ZRWA;
3695     }
3696 
3697     return nvme_zrm_open_flags(nvme_ctrl(req), ns, zone, flags);
3698 }
3699 
3700 static uint16_t nvme_close_zone(NvmeNamespace *ns, NvmeZone *zone,
3701                                 NvmeZoneState state, NvmeRequest *req)
3702 {
3703     return nvme_zrm_close(ns, zone);
3704 }
3705 
3706 static uint16_t nvme_finish_zone(NvmeNamespace *ns, NvmeZone *zone,
3707                                  NvmeZoneState state, NvmeRequest *req)
3708 {
3709     return nvme_zrm_finish(ns, zone);
3710 }
3711 
3712 static uint16_t nvme_offline_zone(NvmeNamespace *ns, NvmeZone *zone,
3713                                   NvmeZoneState state, NvmeRequest *req)
3714 {
3715     switch (state) {
3716     case NVME_ZONE_STATE_READ_ONLY:
3717         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_OFFLINE);
3718         /* fall through */
3719     case NVME_ZONE_STATE_OFFLINE:
3720         return NVME_SUCCESS;
3721     default:
3722         return NVME_ZONE_INVAL_TRANSITION;
3723     }
3724 }
3725 
3726 static uint16_t nvme_set_zd_ext(NvmeNamespace *ns, NvmeZone *zone)
3727 {
3728     uint16_t status;
3729     uint8_t state = nvme_get_zone_state(zone);
3730 
3731     if (state == NVME_ZONE_STATE_EMPTY) {
3732         status = nvme_aor_check(ns, 1, 0);
3733         if (status) {
3734             return status;
3735         }
3736         nvme_aor_inc_active(ns);
3737         zone->d.za |= NVME_ZA_ZD_EXT_VALID;
3738         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
3739         return NVME_SUCCESS;
3740     }
3741 
3742     return NVME_ZONE_INVAL_TRANSITION;
3743 }
3744 
3745 static uint16_t nvme_bulk_proc_zone(NvmeNamespace *ns, NvmeZone *zone,
3746                                     enum NvmeZoneProcessingMask proc_mask,
3747                                     op_handler_t op_hndlr, NvmeRequest *req)
3748 {
3749     uint16_t status = NVME_SUCCESS;
3750     NvmeZoneState zs = nvme_get_zone_state(zone);
3751     bool proc_zone;
3752 
3753     switch (zs) {
3754     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3755     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3756         proc_zone = proc_mask & NVME_PROC_OPENED_ZONES;
3757         break;
3758     case NVME_ZONE_STATE_CLOSED:
3759         proc_zone = proc_mask & NVME_PROC_CLOSED_ZONES;
3760         break;
3761     case NVME_ZONE_STATE_READ_ONLY:
3762         proc_zone = proc_mask & NVME_PROC_READ_ONLY_ZONES;
3763         break;
3764     case NVME_ZONE_STATE_FULL:
3765         proc_zone = proc_mask & NVME_PROC_FULL_ZONES;
3766         break;
3767     default:
3768         proc_zone = false;
3769     }
3770 
3771     if (proc_zone) {
3772         status = op_hndlr(ns, zone, zs, req);
3773     }
3774 
3775     return status;
3776 }
3777 
3778 static uint16_t nvme_do_zone_op(NvmeNamespace *ns, NvmeZone *zone,
3779                                 enum NvmeZoneProcessingMask proc_mask,
3780                                 op_handler_t op_hndlr, NvmeRequest *req)
3781 {
3782     NvmeZone *next;
3783     uint16_t status = NVME_SUCCESS;
3784     int i;
3785 
3786     if (!proc_mask) {
3787         status = op_hndlr(ns, zone, nvme_get_zone_state(zone), req);
3788     } else {
3789         if (proc_mask & NVME_PROC_CLOSED_ZONES) {
3790             QTAILQ_FOREACH_SAFE(zone, &ns->closed_zones, entry, next) {
3791                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3792                                              req);
3793                 if (status && status != NVME_NO_COMPLETE) {
3794                     goto out;
3795                 }
3796             }
3797         }
3798         if (proc_mask & NVME_PROC_OPENED_ZONES) {
3799             QTAILQ_FOREACH_SAFE(zone, &ns->imp_open_zones, entry, next) {
3800                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3801                                              req);
3802                 if (status && status != NVME_NO_COMPLETE) {
3803                     goto out;
3804                 }
3805             }
3806 
3807             QTAILQ_FOREACH_SAFE(zone, &ns->exp_open_zones, entry, next) {
3808                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3809                                              req);
3810                 if (status && status != NVME_NO_COMPLETE) {
3811                     goto out;
3812                 }
3813             }
3814         }
3815         if (proc_mask & NVME_PROC_FULL_ZONES) {
3816             QTAILQ_FOREACH_SAFE(zone, &ns->full_zones, entry, next) {
3817                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3818                                              req);
3819                 if (status && status != NVME_NO_COMPLETE) {
3820                     goto out;
3821                 }
3822             }
3823         }
3824 
3825         if (proc_mask & NVME_PROC_READ_ONLY_ZONES) {
3826             for (i = 0; i < ns->num_zones; i++, zone++) {
3827                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3828                                              req);
3829                 if (status && status != NVME_NO_COMPLETE) {
3830                     goto out;
3831                 }
3832             }
3833         }
3834     }
3835 
3836 out:
3837     return status;
3838 }
3839 
3840 typedef struct NvmeZoneResetAIOCB {
3841     BlockAIOCB common;
3842     BlockAIOCB *aiocb;
3843     NvmeRequest *req;
3844     int ret;
3845 
3846     bool all;
3847     int idx;
3848     NvmeZone *zone;
3849 } NvmeZoneResetAIOCB;
3850 
3851 static void nvme_zone_reset_cancel(BlockAIOCB *aiocb)
3852 {
3853     NvmeZoneResetAIOCB *iocb = container_of(aiocb, NvmeZoneResetAIOCB, common);
3854     NvmeRequest *req = iocb->req;
3855     NvmeNamespace *ns = req->ns;
3856 
3857     iocb->idx = ns->num_zones;
3858 
3859     iocb->ret = -ECANCELED;
3860 
3861     if (iocb->aiocb) {
3862         blk_aio_cancel_async(iocb->aiocb);
3863         iocb->aiocb = NULL;
3864     }
3865 }
3866 
3867 static const AIOCBInfo nvme_zone_reset_aiocb_info = {
3868     .aiocb_size = sizeof(NvmeZoneResetAIOCB),
3869     .cancel_async = nvme_zone_reset_cancel,
3870 };
3871 
3872 static void nvme_zone_reset_cb(void *opaque, int ret);
3873 
3874 static void nvme_zone_reset_epilogue_cb(void *opaque, int ret)
3875 {
3876     NvmeZoneResetAIOCB *iocb = opaque;
3877     NvmeRequest *req = iocb->req;
3878     NvmeNamespace *ns = req->ns;
3879     int64_t moff;
3880     int count;
3881 
3882     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
3883         goto out;
3884     }
3885 
3886     moff = nvme_moff(ns, iocb->zone->d.zslba);
3887     count = nvme_m2b(ns, ns->zone_size);
3888 
3889     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, moff, count,
3890                                         BDRV_REQ_MAY_UNMAP,
3891                                         nvme_zone_reset_cb, iocb);
3892     return;
3893 
3894 out:
3895     nvme_zone_reset_cb(iocb, ret);
3896 }
3897 
3898 static void nvme_zone_reset_cb(void *opaque, int ret)
3899 {
3900     NvmeZoneResetAIOCB *iocb = opaque;
3901     NvmeRequest *req = iocb->req;
3902     NvmeNamespace *ns = req->ns;
3903 
3904     if (iocb->ret < 0) {
3905         goto done;
3906     } else if (ret < 0) {
3907         iocb->ret = ret;
3908         goto done;
3909     }
3910 
3911     if (iocb->zone) {
3912         nvme_zrm_reset(ns, iocb->zone);
3913 
3914         if (!iocb->all) {
3915             goto done;
3916         }
3917     }
3918 
3919     while (iocb->idx < ns->num_zones) {
3920         NvmeZone *zone = &ns->zone_array[iocb->idx++];
3921 
3922         switch (nvme_get_zone_state(zone)) {
3923         case NVME_ZONE_STATE_EMPTY:
3924             if (!iocb->all) {
3925                 goto done;
3926             }
3927 
3928             continue;
3929 
3930         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3931         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3932         case NVME_ZONE_STATE_CLOSED:
3933         case NVME_ZONE_STATE_FULL:
3934             iocb->zone = zone;
3935             break;
3936 
3937         default:
3938             continue;
3939         }
3940 
3941         trace_pci_nvme_zns_zone_reset(zone->d.zslba);
3942 
3943         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk,
3944                                             nvme_l2b(ns, zone->d.zslba),
3945                                             nvme_l2b(ns, ns->zone_size),
3946                                             BDRV_REQ_MAY_UNMAP,
3947                                             nvme_zone_reset_epilogue_cb,
3948                                             iocb);
3949         return;
3950     }
3951 
3952 done:
3953     iocb->aiocb = NULL;
3954 
3955     iocb->common.cb(iocb->common.opaque, iocb->ret);
3956     qemu_aio_unref(iocb);
3957 }
3958 
3959 static uint16_t nvme_zone_mgmt_send_zrwa_flush(NvmeCtrl *n, NvmeZone *zone,
3960                                                uint64_t elba, NvmeRequest *req)
3961 {
3962     NvmeNamespace *ns = req->ns;
3963     uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3964     uint64_t wp = zone->d.wp;
3965     uint32_t nlb = elba - wp + 1;
3966     uint16_t status;
3967 
3968 
3969     if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3970         return NVME_INVALID_ZONE_OP | NVME_DNR;
3971     }
3972 
3973     if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3974         return NVME_INVALID_FIELD | NVME_DNR;
3975     }
3976 
3977     if (elba < wp || elba > wp + ns->zns.zrwas) {
3978         return NVME_ZONE_BOUNDARY_ERROR | NVME_DNR;
3979     }
3980 
3981     if (nlb % ns->zns.zrwafg) {
3982         return NVME_INVALID_FIELD | NVME_DNR;
3983     }
3984 
3985     status = nvme_zrm_auto(n, ns, zone);
3986     if (status) {
3987         return status;
3988     }
3989 
3990     zone->w_ptr += nlb;
3991 
3992     nvme_advance_zone_wp(ns, zone, nlb);
3993 
3994     return NVME_SUCCESS;
3995 }
3996 
3997 static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
3998 {
3999     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
4000     NvmeNamespace *ns = req->ns;
4001     NvmeZone *zone;
4002     NvmeZoneResetAIOCB *iocb;
4003     uint8_t *zd_ext;
4004     uint64_t slba = 0;
4005     uint32_t zone_idx = 0;
4006     uint16_t status;
4007     uint8_t action = cmd->zsa;
4008     bool all;
4009     enum NvmeZoneProcessingMask proc_mask = NVME_PROC_CURRENT_ZONE;
4010 
4011     all = cmd->zsflags & NVME_ZSFLAG_SELECT_ALL;
4012 
4013     req->status = NVME_SUCCESS;
4014 
4015     if (!all) {
4016         status = nvme_get_mgmt_zone_slba_idx(ns, &req->cmd, &slba, &zone_idx);
4017         if (status) {
4018             return status;
4019         }
4020     }
4021 
4022     zone = &ns->zone_array[zone_idx];
4023     if (slba != zone->d.zslba && action != NVME_ZONE_ACTION_ZRWA_FLUSH) {
4024         trace_pci_nvme_err_unaligned_zone_cmd(action, slba, zone->d.zslba);
4025         return NVME_INVALID_FIELD | NVME_DNR;
4026     }
4027 
4028     switch (action) {
4029 
4030     case NVME_ZONE_ACTION_OPEN:
4031         if (all) {
4032             proc_mask = NVME_PROC_CLOSED_ZONES;
4033         }
4034         trace_pci_nvme_open_zone(slba, zone_idx, all);
4035         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_open_zone, req);
4036         break;
4037 
4038     case NVME_ZONE_ACTION_CLOSE:
4039         if (all) {
4040             proc_mask = NVME_PROC_OPENED_ZONES;
4041         }
4042         trace_pci_nvme_close_zone(slba, zone_idx, all);
4043         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_close_zone, req);
4044         break;
4045 
4046     case NVME_ZONE_ACTION_FINISH:
4047         if (all) {
4048             proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES;
4049         }
4050         trace_pci_nvme_finish_zone(slba, zone_idx, all);
4051         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_finish_zone, req);
4052         break;
4053 
4054     case NVME_ZONE_ACTION_RESET:
4055         trace_pci_nvme_reset_zone(slba, zone_idx, all);
4056 
4057         iocb = blk_aio_get(&nvme_zone_reset_aiocb_info, ns->blkconf.blk,
4058                            nvme_misc_cb, req);
4059 
4060         iocb->req = req;
4061         iocb->ret = 0;
4062         iocb->all = all;
4063         iocb->idx = zone_idx;
4064         iocb->zone = NULL;
4065 
4066         req->aiocb = &iocb->common;
4067         nvme_zone_reset_cb(iocb, 0);
4068 
4069         return NVME_NO_COMPLETE;
4070 
4071     case NVME_ZONE_ACTION_OFFLINE:
4072         if (all) {
4073             proc_mask = NVME_PROC_READ_ONLY_ZONES;
4074         }
4075         trace_pci_nvme_offline_zone(slba, zone_idx, all);
4076         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_offline_zone, req);
4077         break;
4078 
4079     case NVME_ZONE_ACTION_SET_ZD_EXT:
4080         trace_pci_nvme_set_descriptor_extension(slba, zone_idx);
4081         if (all || !ns->params.zd_extension_size) {
4082             return NVME_INVALID_FIELD | NVME_DNR;
4083         }
4084         zd_ext = nvme_get_zd_extension(ns, zone_idx);
4085         status = nvme_h2c(n, zd_ext, ns->params.zd_extension_size, req);
4086         if (status) {
4087             trace_pci_nvme_err_zd_extension_map_error(zone_idx);
4088             return status;
4089         }
4090 
4091         status = nvme_set_zd_ext(ns, zone);
4092         if (status == NVME_SUCCESS) {
4093             trace_pci_nvme_zd_extension_set(zone_idx);
4094             return status;
4095         }
4096         break;
4097 
4098     case NVME_ZONE_ACTION_ZRWA_FLUSH:
4099         if (all) {
4100             return NVME_INVALID_FIELD | NVME_DNR;
4101         }
4102 
4103         return nvme_zone_mgmt_send_zrwa_flush(n, zone, slba, req);
4104 
4105     default:
4106         trace_pci_nvme_err_invalid_mgmt_action(action);
4107         status = NVME_INVALID_FIELD;
4108     }
4109 
4110     if (status == NVME_ZONE_INVAL_TRANSITION) {
4111         trace_pci_nvme_err_invalid_zone_state_transition(action, slba,
4112                                                          zone->d.za);
4113     }
4114     if (status) {
4115         status |= NVME_DNR;
4116     }
4117 
4118     return status;
4119 }
4120 
4121 static bool nvme_zone_matches_filter(uint32_t zafs, NvmeZone *zl)
4122 {
4123     NvmeZoneState zs = nvme_get_zone_state(zl);
4124 
4125     switch (zafs) {
4126     case NVME_ZONE_REPORT_ALL:
4127         return true;
4128     case NVME_ZONE_REPORT_EMPTY:
4129         return zs == NVME_ZONE_STATE_EMPTY;
4130     case NVME_ZONE_REPORT_IMPLICITLY_OPEN:
4131         return zs == NVME_ZONE_STATE_IMPLICITLY_OPEN;
4132     case NVME_ZONE_REPORT_EXPLICITLY_OPEN:
4133         return zs == NVME_ZONE_STATE_EXPLICITLY_OPEN;
4134     case NVME_ZONE_REPORT_CLOSED:
4135         return zs == NVME_ZONE_STATE_CLOSED;
4136     case NVME_ZONE_REPORT_FULL:
4137         return zs == NVME_ZONE_STATE_FULL;
4138     case NVME_ZONE_REPORT_READ_ONLY:
4139         return zs == NVME_ZONE_STATE_READ_ONLY;
4140     case NVME_ZONE_REPORT_OFFLINE:
4141         return zs == NVME_ZONE_STATE_OFFLINE;
4142     default:
4143         return false;
4144     }
4145 }
4146 
4147 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4148 {
4149     NvmeCmd *cmd = (NvmeCmd *)&req->cmd;
4150     NvmeNamespace *ns = req->ns;
4151     /* cdw12 is zero-based number of dwords to return. Convert to bytes */
4152     uint32_t data_size = (le32_to_cpu(cmd->cdw12) + 1) << 2;
4153     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
4154     uint32_t zone_idx, zra, zrasf, partial;
4155     uint64_t max_zones, nr_zones = 0;
4156     uint16_t status;
4157     uint64_t slba;
4158     NvmeZoneDescr *z;
4159     NvmeZone *zone;
4160     NvmeZoneReportHeader *header;
4161     void *buf, *buf_p;
4162     size_t zone_entry_sz;
4163     int i;
4164 
4165     req->status = NVME_SUCCESS;
4166 
4167     status = nvme_get_mgmt_zone_slba_idx(ns, cmd, &slba, &zone_idx);
4168     if (status) {
4169         return status;
4170     }
4171 
4172     zra = dw13 & 0xff;
4173     if (zra != NVME_ZONE_REPORT && zra != NVME_ZONE_REPORT_EXTENDED) {
4174         return NVME_INVALID_FIELD | NVME_DNR;
4175     }
4176     if (zra == NVME_ZONE_REPORT_EXTENDED && !ns->params.zd_extension_size) {
4177         return NVME_INVALID_FIELD | NVME_DNR;
4178     }
4179 
4180     zrasf = (dw13 >> 8) & 0xff;
4181     if (zrasf > NVME_ZONE_REPORT_OFFLINE) {
4182         return NVME_INVALID_FIELD | NVME_DNR;
4183     }
4184 
4185     if (data_size < sizeof(NvmeZoneReportHeader)) {
4186         return NVME_INVALID_FIELD | NVME_DNR;
4187     }
4188 
4189     status = nvme_check_mdts(n, data_size);
4190     if (status) {
4191         return status;
4192     }
4193 
4194     partial = (dw13 >> 16) & 0x01;
4195 
4196     zone_entry_sz = sizeof(NvmeZoneDescr);
4197     if (zra == NVME_ZONE_REPORT_EXTENDED) {
4198         zone_entry_sz += ns->params.zd_extension_size;
4199     }
4200 
4201     max_zones = (data_size - sizeof(NvmeZoneReportHeader)) / zone_entry_sz;
4202     buf = g_malloc0(data_size);
4203 
4204     zone = &ns->zone_array[zone_idx];
4205     for (i = zone_idx; i < ns->num_zones; i++) {
4206         if (partial && nr_zones >= max_zones) {
4207             break;
4208         }
4209         if (nvme_zone_matches_filter(zrasf, zone++)) {
4210             nr_zones++;
4211         }
4212     }
4213     header = buf;
4214     header->nr_zones = cpu_to_le64(nr_zones);
4215 
4216     buf_p = buf + sizeof(NvmeZoneReportHeader);
4217     for (; zone_idx < ns->num_zones && max_zones > 0; zone_idx++) {
4218         zone = &ns->zone_array[zone_idx];
4219         if (nvme_zone_matches_filter(zrasf, zone)) {
4220             z = buf_p;
4221             buf_p += sizeof(NvmeZoneDescr);
4222 
4223             z->zt = zone->d.zt;
4224             z->zs = zone->d.zs;
4225             z->zcap = cpu_to_le64(zone->d.zcap);
4226             z->zslba = cpu_to_le64(zone->d.zslba);
4227             z->za = zone->d.za;
4228 
4229             if (nvme_wp_is_valid(zone)) {
4230                 z->wp = cpu_to_le64(zone->d.wp);
4231             } else {
4232                 z->wp = cpu_to_le64(~0ULL);
4233             }
4234 
4235             if (zra == NVME_ZONE_REPORT_EXTENDED) {
4236                 if (zone->d.za & NVME_ZA_ZD_EXT_VALID) {
4237                     memcpy(buf_p, nvme_get_zd_extension(ns, zone_idx),
4238                            ns->params.zd_extension_size);
4239                 }
4240                 buf_p += ns->params.zd_extension_size;
4241             }
4242 
4243             max_zones--;
4244         }
4245     }
4246 
4247     status = nvme_c2h(n, (uint8_t *)buf, data_size, req);
4248 
4249     g_free(buf);
4250 
4251     return status;
4252 }
4253 
4254 static uint16_t nvme_io_mgmt_recv_ruhs(NvmeCtrl *n, NvmeRequest *req,
4255                                        size_t len)
4256 {
4257     NvmeNamespace *ns = req->ns;
4258     NvmeEnduranceGroup *endgrp;
4259     NvmeRuhStatus *hdr;
4260     NvmeRuhStatusDescr *ruhsd;
4261     unsigned int nruhsd;
4262     uint16_t rg, ph, *ruhid;
4263     size_t trans_len;
4264     g_autofree uint8_t *buf = NULL;
4265 
4266     if (!n->subsys) {
4267         return NVME_INVALID_FIELD | NVME_DNR;
4268     }
4269 
4270     if (ns->params.nsid == 0 || ns->params.nsid == 0xffffffff) {
4271         return NVME_INVALID_NSID | NVME_DNR;
4272     }
4273 
4274     if (!n->subsys->endgrp.fdp.enabled) {
4275         return NVME_FDP_DISABLED | NVME_DNR;
4276     }
4277 
4278     endgrp = ns->endgrp;
4279 
4280     nruhsd = ns->fdp.nphs * endgrp->fdp.nrg;
4281     trans_len = sizeof(NvmeRuhStatus) + nruhsd * sizeof(NvmeRuhStatusDescr);
4282     buf = g_malloc(trans_len);
4283 
4284     trans_len = MIN(trans_len, len);
4285 
4286     hdr = (NvmeRuhStatus *)buf;
4287     ruhsd = (NvmeRuhStatusDescr *)(buf + sizeof(NvmeRuhStatus));
4288 
4289     hdr->nruhsd = cpu_to_le16(nruhsd);
4290 
4291     ruhid = ns->fdp.phs;
4292 
4293     for (ph = 0; ph < ns->fdp.nphs; ph++, ruhid++) {
4294         NvmeRuHandle *ruh = &endgrp->fdp.ruhs[*ruhid];
4295 
4296         for (rg = 0; rg < endgrp->fdp.nrg; rg++, ruhsd++) {
4297             uint16_t pid = nvme_make_pid(ns, rg, ph);
4298 
4299             ruhsd->pid = cpu_to_le16(pid);
4300             ruhsd->ruhid = *ruhid;
4301             ruhsd->earutr = 0;
4302             ruhsd->ruamw = cpu_to_le64(ruh->rus[rg].ruamw);
4303         }
4304     }
4305 
4306     return nvme_c2h(n, buf, trans_len, req);
4307 }
4308 
4309 static uint16_t nvme_io_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4310 {
4311     NvmeCmd *cmd = &req->cmd;
4312     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4313     uint32_t numd = le32_to_cpu(cmd->cdw11);
4314     uint8_t mo = (cdw10 & 0xff);
4315     size_t len = (numd + 1) << 2;
4316 
4317     switch (mo) {
4318     case NVME_IOMR_MO_NOP:
4319         return 0;
4320     case NVME_IOMR_MO_RUH_STATUS:
4321         return nvme_io_mgmt_recv_ruhs(n, req, len);
4322     default:
4323         return NVME_INVALID_FIELD | NVME_DNR;
4324     };
4325 }
4326 
4327 static uint16_t nvme_io_mgmt_send_ruh_update(NvmeCtrl *n, NvmeRequest *req)
4328 {
4329     NvmeCmd *cmd = &req->cmd;
4330     NvmeNamespace *ns = req->ns;
4331     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4332     uint16_t ret = NVME_SUCCESS;
4333     uint32_t npid = (cdw10 >> 1) + 1;
4334     unsigned int i = 0;
4335     g_autofree uint16_t *pids = NULL;
4336     uint32_t maxnpid = n->subsys->endgrp.fdp.nrg * n->subsys->endgrp.fdp.nruh;
4337 
4338     if (unlikely(npid >= MIN(NVME_FDP_MAXPIDS, maxnpid))) {
4339         return NVME_INVALID_FIELD | NVME_DNR;
4340     }
4341 
4342     pids = g_new(uint16_t, npid);
4343 
4344     ret = nvme_h2c(n, pids, npid * sizeof(uint16_t), req);
4345     if (ret) {
4346         return ret;
4347     }
4348 
4349     for (; i < npid; i++) {
4350         if (!nvme_update_ruh(n, ns, pids[i])) {
4351             return NVME_INVALID_FIELD | NVME_DNR;
4352         }
4353     }
4354 
4355     return ret;
4356 }
4357 
4358 static uint16_t nvme_io_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4359 {
4360     NvmeCmd *cmd = &req->cmd;
4361     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4362     uint8_t mo = (cdw10 & 0xff);
4363 
4364     switch (mo) {
4365     case NVME_IOMS_MO_NOP:
4366         return 0;
4367     case NVME_IOMS_MO_RUH_UPDATE:
4368         return nvme_io_mgmt_send_ruh_update(n, req);
4369     default:
4370         return NVME_INVALID_FIELD | NVME_DNR;
4371     };
4372 }
4373 
4374 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
4375 {
4376     NvmeNamespace *ns;
4377     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4378 
4379     trace_pci_nvme_io_cmd(nvme_cid(req), nsid, nvme_sqid(req),
4380                           req->cmd.opcode, nvme_io_opc_str(req->cmd.opcode));
4381 
4382     if (!nvme_nsid_valid(n, nsid)) {
4383         return NVME_INVALID_NSID | NVME_DNR;
4384     }
4385 
4386     /*
4387      * In the base NVM command set, Flush may apply to all namespaces
4388      * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4389      * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4390      *
4391      * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4392      * opcode with a specific command since we cannot determine a unique I/O
4393      * command set. Opcode 0h could have any other meaning than something
4394      * equivalent to flushing and say it DOES have completely different
4395      * semantics in some other command set - does an NSID of FFFFFFFFh then
4396      * mean "for all namespaces, apply whatever command set specific command
4397      * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4398      * whatever command that uses the 0h opcode if, and only if, it allows NSID
4399      * to be FFFFFFFFh"?
4400      *
4401      * Anyway (and luckily), for now, we do not care about this since the
4402      * device only supports namespace types that includes the NVM Flush command
4403      * (NVM and Zoned), so always do an NVM Flush.
4404      */
4405     if (req->cmd.opcode == NVME_CMD_FLUSH) {
4406         return nvme_flush(n, req);
4407     }
4408 
4409     ns = nvme_ns(n, nsid);
4410     if (unlikely(!ns)) {
4411         return NVME_INVALID_FIELD | NVME_DNR;
4412     }
4413 
4414     if (!(ns->iocs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
4415         trace_pci_nvme_err_invalid_opc(req->cmd.opcode);
4416         return NVME_INVALID_OPCODE | NVME_DNR;
4417     }
4418 
4419     if (ns->status) {
4420         return ns->status;
4421     }
4422 
4423     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
4424         return NVME_INVALID_FIELD;
4425     }
4426 
4427     req->ns = ns;
4428 
4429     switch (req->cmd.opcode) {
4430     case NVME_CMD_WRITE_ZEROES:
4431         return nvme_write_zeroes(n, req);
4432     case NVME_CMD_ZONE_APPEND:
4433         return nvme_zone_append(n, req);
4434     case NVME_CMD_WRITE:
4435         return nvme_write(n, req);
4436     case NVME_CMD_READ:
4437         return nvme_read(n, req);
4438     case NVME_CMD_COMPARE:
4439         return nvme_compare(n, req);
4440     case NVME_CMD_DSM:
4441         return nvme_dsm(n, req);
4442     case NVME_CMD_VERIFY:
4443         return nvme_verify(n, req);
4444     case NVME_CMD_COPY:
4445         return nvme_copy(n, req);
4446     case NVME_CMD_ZONE_MGMT_SEND:
4447         return nvme_zone_mgmt_send(n, req);
4448     case NVME_CMD_ZONE_MGMT_RECV:
4449         return nvme_zone_mgmt_recv(n, req);
4450     case NVME_CMD_IO_MGMT_RECV:
4451         return nvme_io_mgmt_recv(n, req);
4452     case NVME_CMD_IO_MGMT_SEND:
4453         return nvme_io_mgmt_send(n, req);
4454     default:
4455         assert(false);
4456     }
4457 
4458     return NVME_INVALID_OPCODE | NVME_DNR;
4459 }
4460 
4461 static void nvme_cq_notifier(EventNotifier *e)
4462 {
4463     NvmeCQueue *cq = container_of(e, NvmeCQueue, notifier);
4464     NvmeCtrl *n = cq->ctrl;
4465 
4466     if (!event_notifier_test_and_clear(e)) {
4467         return;
4468     }
4469 
4470     nvme_update_cq_head(cq);
4471 
4472     if (cq->tail == cq->head) {
4473         if (cq->irq_enabled) {
4474             n->cq_pending--;
4475         }
4476 
4477         nvme_irq_deassert(n, cq);
4478     }
4479 
4480     qemu_bh_schedule(cq->bh);
4481 }
4482 
4483 static int nvme_init_cq_ioeventfd(NvmeCQueue *cq)
4484 {
4485     NvmeCtrl *n = cq->ctrl;
4486     uint16_t offset = (cq->cqid << 3) + (1 << 2);
4487     int ret;
4488 
4489     ret = event_notifier_init(&cq->notifier, 0);
4490     if (ret < 0) {
4491         return ret;
4492     }
4493 
4494     event_notifier_set_handler(&cq->notifier, nvme_cq_notifier);
4495     memory_region_add_eventfd(&n->iomem,
4496                               0x1000 + offset, 4, false, 0, &cq->notifier);
4497 
4498     return 0;
4499 }
4500 
4501 static void nvme_sq_notifier(EventNotifier *e)
4502 {
4503     NvmeSQueue *sq = container_of(e, NvmeSQueue, notifier);
4504 
4505     if (!event_notifier_test_and_clear(e)) {
4506         return;
4507     }
4508 
4509     nvme_process_sq(sq);
4510 }
4511 
4512 static int nvme_init_sq_ioeventfd(NvmeSQueue *sq)
4513 {
4514     NvmeCtrl *n = sq->ctrl;
4515     uint16_t offset = sq->sqid << 3;
4516     int ret;
4517 
4518     ret = event_notifier_init(&sq->notifier, 0);
4519     if (ret < 0) {
4520         return ret;
4521     }
4522 
4523     event_notifier_set_handler(&sq->notifier, nvme_sq_notifier);
4524     memory_region_add_eventfd(&n->iomem,
4525                               0x1000 + offset, 4, false, 0, &sq->notifier);
4526 
4527     return 0;
4528 }
4529 
4530 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
4531 {
4532     uint16_t offset = sq->sqid << 3;
4533 
4534     n->sq[sq->sqid] = NULL;
4535     qemu_bh_delete(sq->bh);
4536     if (sq->ioeventfd_enabled) {
4537         memory_region_del_eventfd(&n->iomem,
4538                                   0x1000 + offset, 4, false, 0, &sq->notifier);
4539         event_notifier_set_handler(&sq->notifier, NULL);
4540         event_notifier_cleanup(&sq->notifier);
4541     }
4542     g_free(sq->io_req);
4543     if (sq->sqid) {
4544         g_free(sq);
4545     }
4546 }
4547 
4548 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
4549 {
4550     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4551     NvmeRequest *r, *next;
4552     NvmeSQueue *sq;
4553     NvmeCQueue *cq;
4554     uint16_t qid = le16_to_cpu(c->qid);
4555 
4556     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
4557         trace_pci_nvme_err_invalid_del_sq(qid);
4558         return NVME_INVALID_QID | NVME_DNR;
4559     }
4560 
4561     trace_pci_nvme_del_sq(qid);
4562 
4563     sq = n->sq[qid];
4564     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
4565         r = QTAILQ_FIRST(&sq->out_req_list);
4566         assert(r->aiocb);
4567         blk_aio_cancel(r->aiocb);
4568     }
4569 
4570     assert(QTAILQ_EMPTY(&sq->out_req_list));
4571 
4572     if (!nvme_check_cqid(n, sq->cqid)) {
4573         cq = n->cq[sq->cqid];
4574         QTAILQ_REMOVE(&cq->sq_list, sq, entry);
4575 
4576         nvme_post_cqes(cq);
4577         QTAILQ_FOREACH_SAFE(r, &cq->req_list, entry, next) {
4578             if (r->sq == sq) {
4579                 QTAILQ_REMOVE(&cq->req_list, r, entry);
4580                 QTAILQ_INSERT_TAIL(&sq->req_list, r, entry);
4581             }
4582         }
4583     }
4584 
4585     nvme_free_sq(sq, n);
4586     return NVME_SUCCESS;
4587 }
4588 
4589 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
4590                          uint16_t sqid, uint16_t cqid, uint16_t size)
4591 {
4592     int i;
4593     NvmeCQueue *cq;
4594 
4595     sq->ctrl = n;
4596     sq->dma_addr = dma_addr;
4597     sq->sqid = sqid;
4598     sq->size = size;
4599     sq->cqid = cqid;
4600     sq->head = sq->tail = 0;
4601     sq->io_req = g_new0(NvmeRequest, sq->size);
4602 
4603     QTAILQ_INIT(&sq->req_list);
4604     QTAILQ_INIT(&sq->out_req_list);
4605     for (i = 0; i < sq->size; i++) {
4606         sq->io_req[i].sq = sq;
4607         QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
4608     }
4609 
4610     sq->bh = qemu_bh_new_guarded(nvme_process_sq, sq,
4611                                  &DEVICE(sq->ctrl)->mem_reentrancy_guard);
4612 
4613     if (n->dbbuf_enabled) {
4614         sq->db_addr = n->dbbuf_dbs + (sqid << 3);
4615         sq->ei_addr = n->dbbuf_eis + (sqid << 3);
4616 
4617         if (n->params.ioeventfd && sq->sqid != 0) {
4618             if (!nvme_init_sq_ioeventfd(sq)) {
4619                 sq->ioeventfd_enabled = true;
4620             }
4621         }
4622     }
4623 
4624     assert(n->cq[cqid]);
4625     cq = n->cq[cqid];
4626     QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
4627     n->sq[sqid] = sq;
4628 }
4629 
4630 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeRequest *req)
4631 {
4632     NvmeSQueue *sq;
4633     NvmeCreateSq *c = (NvmeCreateSq *)&req->cmd;
4634 
4635     uint16_t cqid = le16_to_cpu(c->cqid);
4636     uint16_t sqid = le16_to_cpu(c->sqid);
4637     uint16_t qsize = le16_to_cpu(c->qsize);
4638     uint16_t qflags = le16_to_cpu(c->sq_flags);
4639     uint64_t prp1 = le64_to_cpu(c->prp1);
4640 
4641     trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
4642 
4643     if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
4644         trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
4645         return NVME_INVALID_CQID | NVME_DNR;
4646     }
4647     if (unlikely(!sqid || sqid > n->conf_ioqpairs || n->sq[sqid] != NULL)) {
4648         trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
4649         return NVME_INVALID_QID | NVME_DNR;
4650     }
4651     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
4652         trace_pci_nvme_err_invalid_create_sq_size(qsize);
4653         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4654     }
4655     if (unlikely(prp1 & (n->page_size - 1))) {
4656         trace_pci_nvme_err_invalid_create_sq_addr(prp1);
4657         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4658     }
4659     if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
4660         trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
4661         return NVME_INVALID_FIELD | NVME_DNR;
4662     }
4663     sq = g_malloc0(sizeof(*sq));
4664     nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
4665     return NVME_SUCCESS;
4666 }
4667 
4668 struct nvme_stats {
4669     uint64_t units_read;
4670     uint64_t units_written;
4671     uint64_t read_commands;
4672     uint64_t write_commands;
4673 };
4674 
4675 static void nvme_set_blk_stats(NvmeNamespace *ns, struct nvme_stats *stats)
4676 {
4677     BlockAcctStats *s = blk_get_stats(ns->blkconf.blk);
4678 
4679     stats->units_read += s->nr_bytes[BLOCK_ACCT_READ];
4680     stats->units_written += s->nr_bytes[BLOCK_ACCT_WRITE];
4681     stats->read_commands += s->nr_ops[BLOCK_ACCT_READ];
4682     stats->write_commands += s->nr_ops[BLOCK_ACCT_WRITE];
4683 }
4684 
4685 static uint16_t nvme_smart_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4686                                 uint64_t off, NvmeRequest *req)
4687 {
4688     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4689     struct nvme_stats stats = { 0 };
4690     NvmeSmartLog smart = { 0 };
4691     uint32_t trans_len;
4692     NvmeNamespace *ns;
4693     time_t current_ms;
4694     uint64_t u_read, u_written;
4695 
4696     if (off >= sizeof(smart)) {
4697         return NVME_INVALID_FIELD | NVME_DNR;
4698     }
4699 
4700     if (nsid != 0xffffffff) {
4701         ns = nvme_ns(n, nsid);
4702         if (!ns) {
4703             return NVME_INVALID_NSID | NVME_DNR;
4704         }
4705         nvme_set_blk_stats(ns, &stats);
4706     } else {
4707         int i;
4708 
4709         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4710             ns = nvme_ns(n, i);
4711             if (!ns) {
4712                 continue;
4713             }
4714             nvme_set_blk_stats(ns, &stats);
4715         }
4716     }
4717 
4718     trans_len = MIN(sizeof(smart) - off, buf_len);
4719     smart.critical_warning = n->smart_critical_warning;
4720 
4721     u_read = DIV_ROUND_UP(stats.units_read >> BDRV_SECTOR_BITS, 1000);
4722     u_written = DIV_ROUND_UP(stats.units_written >> BDRV_SECTOR_BITS, 1000);
4723 
4724     smart.data_units_read[0] = cpu_to_le64(u_read);
4725     smart.data_units_written[0] = cpu_to_le64(u_written);
4726     smart.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4727     smart.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4728 
4729     smart.temperature = cpu_to_le16(n->temperature);
4730 
4731     if ((n->temperature >= n->features.temp_thresh_hi) ||
4732         (n->temperature <= n->features.temp_thresh_low)) {
4733         smart.critical_warning |= NVME_SMART_TEMPERATURE;
4734     }
4735 
4736     current_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4737     smart.power_on_hours[0] =
4738         cpu_to_le64((((current_ms - n->starttime_ms) / 1000) / 60) / 60);
4739 
4740     if (!rae) {
4741         nvme_clear_events(n, NVME_AER_TYPE_SMART);
4742     }
4743 
4744     return nvme_c2h(n, (uint8_t *) &smart + off, trans_len, req);
4745 }
4746 
4747 static uint16_t nvme_endgrp_info(NvmeCtrl *n,  uint8_t rae, uint32_t buf_len,
4748                                  uint64_t off, NvmeRequest *req)
4749 {
4750     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
4751     uint16_t endgrpid = (dw11 >> 16) & 0xffff;
4752     struct nvme_stats stats = {};
4753     NvmeEndGrpLog info = {};
4754     int i;
4755 
4756     if (!n->subsys || endgrpid != 0x1) {
4757         return NVME_INVALID_FIELD | NVME_DNR;
4758     }
4759 
4760     if (off >= sizeof(info)) {
4761         return NVME_INVALID_FIELD | NVME_DNR;
4762     }
4763 
4764     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4765         NvmeNamespace *ns = nvme_subsys_ns(n->subsys, i);
4766         if (!ns) {
4767             continue;
4768         }
4769 
4770         nvme_set_blk_stats(ns, &stats);
4771     }
4772 
4773     info.data_units_read[0] =
4774         cpu_to_le64(DIV_ROUND_UP(stats.units_read / 1000000000, 1000000000));
4775     info.data_units_written[0] =
4776         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4777     info.media_units_written[0] =
4778         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4779 
4780     info.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4781     info.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4782 
4783     buf_len = MIN(sizeof(info) - off, buf_len);
4784 
4785     return nvme_c2h(n, (uint8_t *)&info + off, buf_len, req);
4786 }
4787 
4788 
4789 static uint16_t nvme_fw_log_info(NvmeCtrl *n, uint32_t buf_len, uint64_t off,
4790                                  NvmeRequest *req)
4791 {
4792     uint32_t trans_len;
4793     NvmeFwSlotInfoLog fw_log = {
4794         .afi = 0x1,
4795     };
4796 
4797     if (off >= sizeof(fw_log)) {
4798         return NVME_INVALID_FIELD | NVME_DNR;
4799     }
4800 
4801     strpadcpy((char *)&fw_log.frs1, sizeof(fw_log.frs1), "1.0", ' ');
4802     trans_len = MIN(sizeof(fw_log) - off, buf_len);
4803 
4804     return nvme_c2h(n, (uint8_t *) &fw_log + off, trans_len, req);
4805 }
4806 
4807 static uint16_t nvme_error_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4808                                 uint64_t off, NvmeRequest *req)
4809 {
4810     uint32_t trans_len;
4811     NvmeErrorLog errlog;
4812 
4813     if (off >= sizeof(errlog)) {
4814         return NVME_INVALID_FIELD | NVME_DNR;
4815     }
4816 
4817     if (!rae) {
4818         nvme_clear_events(n, NVME_AER_TYPE_ERROR);
4819     }
4820 
4821     memset(&errlog, 0x0, sizeof(errlog));
4822     trans_len = MIN(sizeof(errlog) - off, buf_len);
4823 
4824     return nvme_c2h(n, (uint8_t *)&errlog, trans_len, req);
4825 }
4826 
4827 static uint16_t nvme_changed_nslist(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4828                                     uint64_t off, NvmeRequest *req)
4829 {
4830     uint32_t nslist[1024];
4831     uint32_t trans_len;
4832     int i = 0;
4833     uint32_t nsid;
4834 
4835     if (off >= sizeof(nslist)) {
4836         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(nslist));
4837         return NVME_INVALID_FIELD | NVME_DNR;
4838     }
4839 
4840     memset(nslist, 0x0, sizeof(nslist));
4841     trans_len = MIN(sizeof(nslist) - off, buf_len);
4842 
4843     while ((nsid = find_first_bit(n->changed_nsids, NVME_CHANGED_NSID_SIZE)) !=
4844             NVME_CHANGED_NSID_SIZE) {
4845         /*
4846          * If more than 1024 namespaces, the first entry in the log page should
4847          * be set to FFFFFFFFh and the others to 0 as spec.
4848          */
4849         if (i == ARRAY_SIZE(nslist)) {
4850             memset(nslist, 0x0, sizeof(nslist));
4851             nslist[0] = 0xffffffff;
4852             break;
4853         }
4854 
4855         nslist[i++] = nsid;
4856         clear_bit(nsid, n->changed_nsids);
4857     }
4858 
4859     /*
4860      * Remove all the remaining list entries in case returns directly due to
4861      * more than 1024 namespaces.
4862      */
4863     if (nslist[0] == 0xffffffff) {
4864         bitmap_zero(n->changed_nsids, NVME_CHANGED_NSID_SIZE);
4865     }
4866 
4867     if (!rae) {
4868         nvme_clear_events(n, NVME_AER_TYPE_NOTICE);
4869     }
4870 
4871     return nvme_c2h(n, ((uint8_t *)nslist) + off, trans_len, req);
4872 }
4873 
4874 static uint16_t nvme_cmd_effects(NvmeCtrl *n, uint8_t csi, uint32_t buf_len,
4875                                  uint64_t off, NvmeRequest *req)
4876 {
4877     NvmeEffectsLog log = {};
4878     const uint32_t *src_iocs = NULL;
4879     uint32_t trans_len;
4880 
4881     if (off >= sizeof(log)) {
4882         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(log));
4883         return NVME_INVALID_FIELD | NVME_DNR;
4884     }
4885 
4886     switch (NVME_CC_CSS(ldl_le_p(&n->bar.cc))) {
4887     case NVME_CC_CSS_NVM:
4888         src_iocs = nvme_cse_iocs_nvm;
4889         /* fall through */
4890     case NVME_CC_CSS_ADMIN_ONLY:
4891         break;
4892     case NVME_CC_CSS_CSI:
4893         switch (csi) {
4894         case NVME_CSI_NVM:
4895             src_iocs = nvme_cse_iocs_nvm;
4896             break;
4897         case NVME_CSI_ZONED:
4898             src_iocs = nvme_cse_iocs_zoned;
4899             break;
4900         }
4901     }
4902 
4903     memcpy(log.acs, nvme_cse_acs, sizeof(nvme_cse_acs));
4904 
4905     if (src_iocs) {
4906         memcpy(log.iocs, src_iocs, sizeof(log.iocs));
4907     }
4908 
4909     trans_len = MIN(sizeof(log) - off, buf_len);
4910 
4911     return nvme_c2h(n, ((uint8_t *)&log) + off, trans_len, req);
4912 }
4913 
4914 static size_t sizeof_fdp_conf_descr(size_t nruh, size_t vss)
4915 {
4916     size_t entry_siz = sizeof(NvmeFdpDescrHdr) + nruh * sizeof(NvmeRuhDescr)
4917                        + vss;
4918     return ROUND_UP(entry_siz, 8);
4919 }
4920 
4921 static uint16_t nvme_fdp_confs(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
4922                                uint64_t off, NvmeRequest *req)
4923 {
4924     uint32_t log_size, trans_len;
4925     g_autofree uint8_t *buf = NULL;
4926     NvmeFdpDescrHdr *hdr;
4927     NvmeRuhDescr *ruhd;
4928     NvmeEnduranceGroup *endgrp;
4929     NvmeFdpConfsHdr *log;
4930     size_t nruh, fdp_descr_size;
4931     int i;
4932 
4933     if (endgrpid != 1 || !n->subsys) {
4934         return NVME_INVALID_FIELD | NVME_DNR;
4935     }
4936 
4937     endgrp = &n->subsys->endgrp;
4938 
4939     if (endgrp->fdp.enabled) {
4940         nruh = endgrp->fdp.nruh;
4941     } else {
4942         nruh = 1;
4943     }
4944 
4945     fdp_descr_size = sizeof_fdp_conf_descr(nruh, FDPVSS);
4946     log_size = sizeof(NvmeFdpConfsHdr) + fdp_descr_size;
4947 
4948     if (off >= log_size) {
4949         return NVME_INVALID_FIELD | NVME_DNR;
4950     }
4951 
4952     trans_len = MIN(log_size - off, buf_len);
4953 
4954     buf = g_malloc0(log_size);
4955     log = (NvmeFdpConfsHdr *)buf;
4956     hdr = (NvmeFdpDescrHdr *)(log + 1);
4957     ruhd = (NvmeRuhDescr *)(buf + sizeof(*log) + sizeof(*hdr));
4958 
4959     log->num_confs = cpu_to_le16(0);
4960     log->size = cpu_to_le32(log_size);
4961 
4962     hdr->descr_size = cpu_to_le16(fdp_descr_size);
4963     if (endgrp->fdp.enabled) {
4964         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, VALID, 1);
4965         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, RGIF, endgrp->fdp.rgif);
4966         hdr->nrg = cpu_to_le16(endgrp->fdp.nrg);
4967         hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
4968         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
4969         hdr->nnss = cpu_to_le32(NVME_MAX_NAMESPACES);
4970         hdr->runs = cpu_to_le64(endgrp->fdp.runs);
4971 
4972         for (i = 0; i < nruh; i++) {
4973             ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
4974             ruhd++;
4975         }
4976     } else {
4977         /* 1 bit for RUH in PIF -> 2 RUHs max. */
4978         hdr->nrg = cpu_to_le16(1);
4979         hdr->nruh = cpu_to_le16(1);
4980         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
4981         hdr->nnss = cpu_to_le32(1);
4982         hdr->runs = cpu_to_le64(96 * MiB);
4983 
4984         ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
4985     }
4986 
4987     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
4988 }
4989 
4990 static uint16_t nvme_fdp_ruh_usage(NvmeCtrl *n, uint32_t endgrpid,
4991                                    uint32_t dw10, uint32_t dw12,
4992                                    uint32_t buf_len, uint64_t off,
4993                                    NvmeRequest *req)
4994 {
4995     NvmeRuHandle *ruh;
4996     NvmeRuhuLog *hdr;
4997     NvmeRuhuDescr *ruhud;
4998     NvmeEnduranceGroup *endgrp;
4999     g_autofree uint8_t *buf = NULL;
5000     uint32_t log_size, trans_len;
5001     uint16_t i;
5002 
5003     if (endgrpid != 1 || !n->subsys) {
5004         return NVME_INVALID_FIELD | NVME_DNR;
5005     }
5006 
5007     endgrp = &n->subsys->endgrp;
5008 
5009     if (!endgrp->fdp.enabled) {
5010         return NVME_FDP_DISABLED | NVME_DNR;
5011     }
5012 
5013     log_size = sizeof(NvmeRuhuLog) + endgrp->fdp.nruh * sizeof(NvmeRuhuDescr);
5014 
5015     if (off >= log_size) {
5016         return NVME_INVALID_FIELD | NVME_DNR;
5017     }
5018 
5019     trans_len = MIN(log_size - off, buf_len);
5020 
5021     buf = g_malloc0(log_size);
5022     hdr = (NvmeRuhuLog *)buf;
5023     ruhud = (NvmeRuhuDescr *)(hdr + 1);
5024 
5025     ruh = endgrp->fdp.ruhs;
5026     hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5027 
5028     for (i = 0; i < endgrp->fdp.nruh; i++, ruhud++, ruh++) {
5029         ruhud->ruha = ruh->ruha;
5030     }
5031 
5032     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5033 }
5034 
5035 static uint16_t nvme_fdp_stats(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
5036                                uint64_t off, NvmeRequest *req)
5037 {
5038     NvmeEnduranceGroup *endgrp;
5039     NvmeFdpStatsLog log = {};
5040     uint32_t trans_len;
5041 
5042     if (off >= sizeof(NvmeFdpStatsLog)) {
5043         return NVME_INVALID_FIELD | NVME_DNR;
5044     }
5045 
5046     if (endgrpid != 1 || !n->subsys) {
5047         return NVME_INVALID_FIELD | NVME_DNR;
5048     }
5049 
5050     if (!n->subsys->endgrp.fdp.enabled) {
5051         return NVME_FDP_DISABLED | NVME_DNR;
5052     }
5053 
5054     endgrp = &n->subsys->endgrp;
5055 
5056     trans_len = MIN(sizeof(log) - off, buf_len);
5057 
5058     /* spec value is 128 bit, we only use 64 bit */
5059     log.hbmw[0] = cpu_to_le64(endgrp->fdp.hbmw);
5060     log.mbmw[0] = cpu_to_le64(endgrp->fdp.mbmw);
5061     log.mbe[0] = cpu_to_le64(endgrp->fdp.mbe);
5062 
5063     return nvme_c2h(n, (uint8_t *)&log + off, trans_len, req);
5064 }
5065 
5066 static uint16_t nvme_fdp_events(NvmeCtrl *n, uint32_t endgrpid,
5067                                 uint32_t buf_len, uint64_t off,
5068                                 NvmeRequest *req)
5069 {
5070     NvmeEnduranceGroup *endgrp;
5071     NvmeCmd *cmd = &req->cmd;
5072     bool host_events = (cmd->cdw10 >> 8) & 0x1;
5073     uint32_t log_size, trans_len;
5074     NvmeFdpEventBuffer *ebuf;
5075     g_autofree NvmeFdpEventsLog *elog = NULL;
5076     NvmeFdpEvent *event;
5077 
5078     if (endgrpid != 1 || !n->subsys) {
5079         return NVME_INVALID_FIELD | NVME_DNR;
5080     }
5081 
5082     endgrp = &n->subsys->endgrp;
5083 
5084     if (!endgrp->fdp.enabled) {
5085         return NVME_FDP_DISABLED | NVME_DNR;
5086     }
5087 
5088     if (host_events) {
5089         ebuf = &endgrp->fdp.host_events;
5090     } else {
5091         ebuf = &endgrp->fdp.ctrl_events;
5092     }
5093 
5094     log_size = sizeof(NvmeFdpEventsLog) + ebuf->nelems * sizeof(NvmeFdpEvent);
5095     trans_len = MIN(log_size - off, buf_len);
5096     elog = g_malloc0(log_size);
5097     elog->num_events = cpu_to_le32(ebuf->nelems);
5098     event = (NvmeFdpEvent *)(elog + 1);
5099 
5100     if (ebuf->nelems && ebuf->start == ebuf->next) {
5101         unsigned int nelems = (NVME_FDP_MAX_EVENTS - ebuf->start);
5102         /* wrap over, copy [start;NVME_FDP_MAX_EVENTS[ and [0; next[ */
5103         memcpy(event, &ebuf->events[ebuf->start],
5104                sizeof(NvmeFdpEvent) * nelems);
5105         memcpy(event + nelems, ebuf->events,
5106                sizeof(NvmeFdpEvent) * ebuf->next);
5107     } else if (ebuf->start < ebuf->next) {
5108         memcpy(event, &ebuf->events[ebuf->start],
5109                sizeof(NvmeFdpEvent) * (ebuf->next - ebuf->start));
5110     }
5111 
5112     return nvme_c2h(n, (uint8_t *)elog + off, trans_len, req);
5113 }
5114 
5115 static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
5116 {
5117     NvmeCmd *cmd = &req->cmd;
5118 
5119     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5120     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5121     uint32_t dw12 = le32_to_cpu(cmd->cdw12);
5122     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
5123     uint8_t  lid = dw10 & 0xff;
5124     uint8_t  lsp = (dw10 >> 8) & 0xf;
5125     uint8_t  rae = (dw10 >> 15) & 0x1;
5126     uint8_t  csi = le32_to_cpu(cmd->cdw14) >> 24;
5127     uint32_t numdl, numdu, lspi;
5128     uint64_t off, lpol, lpou;
5129     size_t   len;
5130     uint16_t status;
5131 
5132     numdl = (dw10 >> 16);
5133     numdu = (dw11 & 0xffff);
5134     lspi = (dw11 >> 16);
5135     lpol = dw12;
5136     lpou = dw13;
5137 
5138     len = (((numdu << 16) | numdl) + 1) << 2;
5139     off = (lpou << 32ULL) | lpol;
5140 
5141     if (off & 0x3) {
5142         return NVME_INVALID_FIELD | NVME_DNR;
5143     }
5144 
5145     trace_pci_nvme_get_log(nvme_cid(req), lid, lsp, rae, len, off);
5146 
5147     status = nvme_check_mdts(n, len);
5148     if (status) {
5149         return status;
5150     }
5151 
5152     switch (lid) {
5153     case NVME_LOG_ERROR_INFO:
5154         return nvme_error_info(n, rae, len, off, req);
5155     case NVME_LOG_SMART_INFO:
5156         return nvme_smart_info(n, rae, len, off, req);
5157     case NVME_LOG_FW_SLOT_INFO:
5158         return nvme_fw_log_info(n, len, off, req);
5159     case NVME_LOG_CHANGED_NSLIST:
5160         return nvme_changed_nslist(n, rae, len, off, req);
5161     case NVME_LOG_CMD_EFFECTS:
5162         return nvme_cmd_effects(n, csi, len, off, req);
5163     case NVME_LOG_ENDGRP:
5164         return nvme_endgrp_info(n, rae, len, off, req);
5165     case NVME_LOG_FDP_CONFS:
5166         return nvme_fdp_confs(n, lspi, len, off, req);
5167     case NVME_LOG_FDP_RUH_USAGE:
5168         return nvme_fdp_ruh_usage(n, lspi, dw10, dw12, len, off, req);
5169     case NVME_LOG_FDP_STATS:
5170         return nvme_fdp_stats(n, lspi, len, off, req);
5171     case NVME_LOG_FDP_EVENTS:
5172         return nvme_fdp_events(n, lspi, len, off, req);
5173     default:
5174         trace_pci_nvme_err_invalid_log_page(nvme_cid(req), lid);
5175         return NVME_INVALID_FIELD | NVME_DNR;
5176     }
5177 }
5178 
5179 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
5180 {
5181     PCIDevice *pci = PCI_DEVICE(n);
5182     uint16_t offset = (cq->cqid << 3) + (1 << 2);
5183 
5184     n->cq[cq->cqid] = NULL;
5185     qemu_bh_delete(cq->bh);
5186     if (cq->ioeventfd_enabled) {
5187         memory_region_del_eventfd(&n->iomem,
5188                                   0x1000 + offset, 4, false, 0, &cq->notifier);
5189         event_notifier_set_handler(&cq->notifier, NULL);
5190         event_notifier_cleanup(&cq->notifier);
5191     }
5192     if (msix_enabled(pci)) {
5193         msix_vector_unuse(pci, cq->vector);
5194     }
5195     if (cq->cqid) {
5196         g_free(cq);
5197     }
5198 }
5199 
5200 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)
5201 {
5202     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
5203     NvmeCQueue *cq;
5204     uint16_t qid = le16_to_cpu(c->qid);
5205 
5206     if (unlikely(!qid || nvme_check_cqid(n, qid))) {
5207         trace_pci_nvme_err_invalid_del_cq_cqid(qid);
5208         return NVME_INVALID_CQID | NVME_DNR;
5209     }
5210 
5211     cq = n->cq[qid];
5212     if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
5213         trace_pci_nvme_err_invalid_del_cq_notempty(qid);
5214         return NVME_INVALID_QUEUE_DEL;
5215     }
5216 
5217     if (cq->irq_enabled && cq->tail != cq->head) {
5218         n->cq_pending--;
5219     }
5220 
5221     nvme_irq_deassert(n, cq);
5222     trace_pci_nvme_del_cq(qid);
5223     nvme_free_cq(cq, n);
5224     return NVME_SUCCESS;
5225 }
5226 
5227 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
5228                          uint16_t cqid, uint16_t vector, uint16_t size,
5229                          uint16_t irq_enabled)
5230 {
5231     PCIDevice *pci = PCI_DEVICE(n);
5232 
5233     if (msix_enabled(pci)) {
5234         msix_vector_use(pci, vector);
5235     }
5236     cq->ctrl = n;
5237     cq->cqid = cqid;
5238     cq->size = size;
5239     cq->dma_addr = dma_addr;
5240     cq->phase = 1;
5241     cq->irq_enabled = irq_enabled;
5242     cq->vector = vector;
5243     cq->head = cq->tail = 0;
5244     QTAILQ_INIT(&cq->req_list);
5245     QTAILQ_INIT(&cq->sq_list);
5246     if (n->dbbuf_enabled) {
5247         cq->db_addr = n->dbbuf_dbs + (cqid << 3) + (1 << 2);
5248         cq->ei_addr = n->dbbuf_eis + (cqid << 3) + (1 << 2);
5249 
5250         if (n->params.ioeventfd && cqid != 0) {
5251             if (!nvme_init_cq_ioeventfd(cq)) {
5252                 cq->ioeventfd_enabled = true;
5253             }
5254         }
5255     }
5256     n->cq[cqid] = cq;
5257     cq->bh = qemu_bh_new_guarded(nvme_post_cqes, cq,
5258                                  &DEVICE(cq->ctrl)->mem_reentrancy_guard);
5259 }
5260 
5261 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeRequest *req)
5262 {
5263     NvmeCQueue *cq;
5264     NvmeCreateCq *c = (NvmeCreateCq *)&req->cmd;
5265     uint16_t cqid = le16_to_cpu(c->cqid);
5266     uint16_t vector = le16_to_cpu(c->irq_vector);
5267     uint16_t qsize = le16_to_cpu(c->qsize);
5268     uint16_t qflags = le16_to_cpu(c->cq_flags);
5269     uint64_t prp1 = le64_to_cpu(c->prp1);
5270 
5271     trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
5272                              NVME_CQ_FLAGS_IEN(qflags) != 0);
5273 
5274     if (unlikely(!cqid || cqid > n->conf_ioqpairs || n->cq[cqid] != NULL)) {
5275         trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
5276         return NVME_INVALID_QID | NVME_DNR;
5277     }
5278     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
5279         trace_pci_nvme_err_invalid_create_cq_size(qsize);
5280         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5281     }
5282     if (unlikely(prp1 & (n->page_size - 1))) {
5283         trace_pci_nvme_err_invalid_create_cq_addr(prp1);
5284         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
5285     }
5286     if (unlikely(!msix_enabled(PCI_DEVICE(n)) && vector)) {
5287         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5288         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5289     }
5290     if (unlikely(vector >= n->conf_msix_qsize)) {
5291         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5292         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5293     }
5294     if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
5295         trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
5296         return NVME_INVALID_FIELD | NVME_DNR;
5297     }
5298 
5299     cq = g_malloc0(sizeof(*cq));
5300     nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
5301                  NVME_CQ_FLAGS_IEN(qflags));
5302 
5303     /*
5304      * It is only required to set qs_created when creating a completion queue;
5305      * creating a submission queue without a matching completion queue will
5306      * fail.
5307      */
5308     n->qs_created = true;
5309     return NVME_SUCCESS;
5310 }
5311 
5312 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl *n, NvmeRequest *req)
5313 {
5314     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5315 
5316     return nvme_c2h(n, id, sizeof(id), req);
5317 }
5318 
5319 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeRequest *req)
5320 {
5321     trace_pci_nvme_identify_ctrl();
5322 
5323     return nvme_c2h(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl), req);
5324 }
5325 
5326 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl *n, NvmeRequest *req)
5327 {
5328     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5329     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5330     NvmeIdCtrlNvm *id_nvm = (NvmeIdCtrlNvm *)&id;
5331 
5332     trace_pci_nvme_identify_ctrl_csi(c->csi);
5333 
5334     switch (c->csi) {
5335     case NVME_CSI_NVM:
5336         id_nvm->vsl = n->params.vsl;
5337         id_nvm->dmrsl = cpu_to_le32(n->dmrsl);
5338         break;
5339 
5340     case NVME_CSI_ZONED:
5341         ((NvmeIdCtrlZoned *)&id)->zasl = n->params.zasl;
5342         break;
5343 
5344     default:
5345         return NVME_INVALID_FIELD | NVME_DNR;
5346     }
5347 
5348     return nvme_c2h(n, id, sizeof(id), req);
5349 }
5350 
5351 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeRequest *req, bool active)
5352 {
5353     NvmeNamespace *ns;
5354     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5355     uint32_t nsid = le32_to_cpu(c->nsid);
5356 
5357     trace_pci_nvme_identify_ns(nsid);
5358 
5359     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5360         return NVME_INVALID_NSID | NVME_DNR;
5361     }
5362 
5363     ns = nvme_ns(n, nsid);
5364     if (unlikely(!ns)) {
5365         if (!active) {
5366             ns = nvme_subsys_ns(n->subsys, nsid);
5367             if (!ns) {
5368                 return nvme_rpt_empty_id_struct(n, req);
5369             }
5370         } else {
5371             return nvme_rpt_empty_id_struct(n, req);
5372         }
5373     }
5374 
5375     if (active || ns->csi == NVME_CSI_NVM) {
5376         return nvme_c2h(n, (uint8_t *)&ns->id_ns, sizeof(NvmeIdNs), req);
5377     }
5378 
5379     return NVME_INVALID_CMD_SET | NVME_DNR;
5380 }
5381 
5382 static uint16_t nvme_identify_ctrl_list(NvmeCtrl *n, NvmeRequest *req,
5383                                         bool attached)
5384 {
5385     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5386     uint32_t nsid = le32_to_cpu(c->nsid);
5387     uint16_t min_id = le16_to_cpu(c->ctrlid);
5388     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5389     uint16_t *ids = &list[1];
5390     NvmeNamespace *ns;
5391     NvmeCtrl *ctrl;
5392     int cntlid, nr_ids = 0;
5393 
5394     trace_pci_nvme_identify_ctrl_list(c->cns, min_id);
5395 
5396     if (!n->subsys) {
5397         return NVME_INVALID_FIELD | NVME_DNR;
5398     }
5399 
5400     if (attached) {
5401         if (nsid == NVME_NSID_BROADCAST) {
5402             return NVME_INVALID_FIELD | NVME_DNR;
5403         }
5404 
5405         ns = nvme_subsys_ns(n->subsys, nsid);
5406         if (!ns) {
5407             return NVME_INVALID_FIELD | NVME_DNR;
5408         }
5409     }
5410 
5411     for (cntlid = min_id; cntlid < ARRAY_SIZE(n->subsys->ctrls); cntlid++) {
5412         ctrl = nvme_subsys_ctrl(n->subsys, cntlid);
5413         if (!ctrl) {
5414             continue;
5415         }
5416 
5417         if (attached && !nvme_ns(ctrl, nsid)) {
5418             continue;
5419         }
5420 
5421         ids[nr_ids++] = cntlid;
5422     }
5423 
5424     list[0] = nr_ids;
5425 
5426     return nvme_c2h(n, (uint8_t *)list, sizeof(list), req);
5427 }
5428 
5429 static uint16_t nvme_identify_pri_ctrl_cap(NvmeCtrl *n, NvmeRequest *req)
5430 {
5431     trace_pci_nvme_identify_pri_ctrl_cap(le16_to_cpu(n->pri_ctrl_cap.cntlid));
5432 
5433     return nvme_c2h(n, (uint8_t *)&n->pri_ctrl_cap,
5434                     sizeof(NvmePriCtrlCap), req);
5435 }
5436 
5437 static uint16_t nvme_identify_sec_ctrl_list(NvmeCtrl *n, NvmeRequest *req)
5438 {
5439     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5440     uint16_t pri_ctrl_id = le16_to_cpu(n->pri_ctrl_cap.cntlid);
5441     uint16_t min_id = le16_to_cpu(c->ctrlid);
5442     uint8_t num_sec_ctrl = n->sec_ctrl_list.numcntl;
5443     NvmeSecCtrlList list = {0};
5444     uint8_t i;
5445 
5446     for (i = 0; i < num_sec_ctrl; i++) {
5447         if (n->sec_ctrl_list.sec[i].scid >= min_id) {
5448             list.numcntl = num_sec_ctrl - i;
5449             memcpy(&list.sec, n->sec_ctrl_list.sec + i,
5450                    list.numcntl * sizeof(NvmeSecCtrlEntry));
5451             break;
5452         }
5453     }
5454 
5455     trace_pci_nvme_identify_sec_ctrl_list(pri_ctrl_id, list.numcntl);
5456 
5457     return nvme_c2h(n, (uint8_t *)&list, sizeof(list), req);
5458 }
5459 
5460 static uint16_t nvme_identify_ns_csi(NvmeCtrl *n, NvmeRequest *req,
5461                                      bool active)
5462 {
5463     NvmeNamespace *ns;
5464     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5465     uint32_t nsid = le32_to_cpu(c->nsid);
5466 
5467     trace_pci_nvme_identify_ns_csi(nsid, c->csi);
5468 
5469     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5470         return NVME_INVALID_NSID | NVME_DNR;
5471     }
5472 
5473     ns = nvme_ns(n, nsid);
5474     if (unlikely(!ns)) {
5475         if (!active) {
5476             ns = nvme_subsys_ns(n->subsys, nsid);
5477             if (!ns) {
5478                 return nvme_rpt_empty_id_struct(n, req);
5479             }
5480         } else {
5481             return nvme_rpt_empty_id_struct(n, req);
5482         }
5483     }
5484 
5485     if (c->csi == NVME_CSI_NVM) {
5486         return nvme_c2h(n, (uint8_t *)&ns->id_ns_nvm, sizeof(NvmeIdNsNvm),
5487                         req);
5488     } else if (c->csi == NVME_CSI_ZONED && ns->csi == NVME_CSI_ZONED) {
5489         return nvme_c2h(n, (uint8_t *)ns->id_ns_zoned, sizeof(NvmeIdNsZoned),
5490                         req);
5491     }
5492 
5493     return NVME_INVALID_FIELD | NVME_DNR;
5494 }
5495 
5496 static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeRequest *req,
5497                                      bool active)
5498 {
5499     NvmeNamespace *ns;
5500     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5501     uint32_t min_nsid = le32_to_cpu(c->nsid);
5502     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5503     static const int data_len = sizeof(list);
5504     uint32_t *list_ptr = (uint32_t *)list;
5505     int i, j = 0;
5506 
5507     trace_pci_nvme_identify_nslist(min_nsid);
5508 
5509     /*
5510      * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
5511      * since the Active Namespace ID List should return namespaces with ids
5512      * *higher* than the NSID specified in the command. This is also specified
5513      * in the spec (NVM Express v1.3d, Section 5.15.4).
5514      */
5515     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5516         return NVME_INVALID_NSID | NVME_DNR;
5517     }
5518 
5519     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5520         ns = nvme_ns(n, i);
5521         if (!ns) {
5522             if (!active) {
5523                 ns = nvme_subsys_ns(n->subsys, i);
5524                 if (!ns) {
5525                     continue;
5526                 }
5527             } else {
5528                 continue;
5529             }
5530         }
5531         if (ns->params.nsid <= min_nsid) {
5532             continue;
5533         }
5534         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5535         if (j == data_len / sizeof(uint32_t)) {
5536             break;
5537         }
5538     }
5539 
5540     return nvme_c2h(n, list, data_len, req);
5541 }
5542 
5543 static uint16_t nvme_identify_nslist_csi(NvmeCtrl *n, NvmeRequest *req,
5544                                          bool active)
5545 {
5546     NvmeNamespace *ns;
5547     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5548     uint32_t min_nsid = le32_to_cpu(c->nsid);
5549     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5550     static const int data_len = sizeof(list);
5551     uint32_t *list_ptr = (uint32_t *)list;
5552     int i, j = 0;
5553 
5554     trace_pci_nvme_identify_nslist_csi(min_nsid, c->csi);
5555 
5556     /*
5557      * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
5558      */
5559     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5560         return NVME_INVALID_NSID | NVME_DNR;
5561     }
5562 
5563     if (c->csi != NVME_CSI_NVM && c->csi != NVME_CSI_ZONED) {
5564         return NVME_INVALID_FIELD | NVME_DNR;
5565     }
5566 
5567     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5568         ns = nvme_ns(n, i);
5569         if (!ns) {
5570             if (!active) {
5571                 ns = nvme_subsys_ns(n->subsys, i);
5572                 if (!ns) {
5573                     continue;
5574                 }
5575             } else {
5576                 continue;
5577             }
5578         }
5579         if (ns->params.nsid <= min_nsid || c->csi != ns->csi) {
5580             continue;
5581         }
5582         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5583         if (j == data_len / sizeof(uint32_t)) {
5584             break;
5585         }
5586     }
5587 
5588     return nvme_c2h(n, list, data_len, req);
5589 }
5590 
5591 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl *n, NvmeRequest *req)
5592 {
5593     NvmeNamespace *ns;
5594     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5595     uint32_t nsid = le32_to_cpu(c->nsid);
5596     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5597     uint8_t *pos = list;
5598     struct {
5599         NvmeIdNsDescr hdr;
5600         uint8_t v[NVME_NIDL_UUID];
5601     } QEMU_PACKED uuid = {};
5602     struct {
5603         NvmeIdNsDescr hdr;
5604         uint64_t v;
5605     } QEMU_PACKED eui64 = {};
5606     struct {
5607         NvmeIdNsDescr hdr;
5608         uint8_t v;
5609     } QEMU_PACKED csi = {};
5610 
5611     trace_pci_nvme_identify_ns_descr_list(nsid);
5612 
5613     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5614         return NVME_INVALID_NSID | NVME_DNR;
5615     }
5616 
5617     ns = nvme_ns(n, nsid);
5618     if (unlikely(!ns)) {
5619         return NVME_INVALID_FIELD | NVME_DNR;
5620     }
5621 
5622     if (!qemu_uuid_is_null(&ns->params.uuid)) {
5623         uuid.hdr.nidt = NVME_NIDT_UUID;
5624         uuid.hdr.nidl = NVME_NIDL_UUID;
5625         memcpy(uuid.v, ns->params.uuid.data, NVME_NIDL_UUID);
5626         memcpy(pos, &uuid, sizeof(uuid));
5627         pos += sizeof(uuid);
5628     }
5629 
5630     if (ns->params.eui64) {
5631         eui64.hdr.nidt = NVME_NIDT_EUI64;
5632         eui64.hdr.nidl = NVME_NIDL_EUI64;
5633         eui64.v = cpu_to_be64(ns->params.eui64);
5634         memcpy(pos, &eui64, sizeof(eui64));
5635         pos += sizeof(eui64);
5636     }
5637 
5638     csi.hdr.nidt = NVME_NIDT_CSI;
5639     csi.hdr.nidl = NVME_NIDL_CSI;
5640     csi.v = ns->csi;
5641     memcpy(pos, &csi, sizeof(csi));
5642     pos += sizeof(csi);
5643 
5644     return nvme_c2h(n, list, sizeof(list), req);
5645 }
5646 
5647 static uint16_t nvme_identify_cmd_set(NvmeCtrl *n, NvmeRequest *req)
5648 {
5649     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5650     static const int data_len = sizeof(list);
5651 
5652     trace_pci_nvme_identify_cmd_set();
5653 
5654     NVME_SET_CSI(*list, NVME_CSI_NVM);
5655     NVME_SET_CSI(*list, NVME_CSI_ZONED);
5656 
5657     return nvme_c2h(n, list, data_len, req);
5658 }
5659 
5660 static uint16_t nvme_identify(NvmeCtrl *n, NvmeRequest *req)
5661 {
5662     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5663 
5664     trace_pci_nvme_identify(nvme_cid(req), c->cns, le16_to_cpu(c->ctrlid),
5665                             c->csi);
5666 
5667     switch (c->cns) {
5668     case NVME_ID_CNS_NS:
5669         return nvme_identify_ns(n, req, true);
5670     case NVME_ID_CNS_NS_PRESENT:
5671         return nvme_identify_ns(n, req, false);
5672     case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST:
5673         return nvme_identify_ctrl_list(n, req, true);
5674     case NVME_ID_CNS_CTRL_LIST:
5675         return nvme_identify_ctrl_list(n, req, false);
5676     case NVME_ID_CNS_PRIMARY_CTRL_CAP:
5677         return nvme_identify_pri_ctrl_cap(n, req);
5678     case NVME_ID_CNS_SECONDARY_CTRL_LIST:
5679         return nvme_identify_sec_ctrl_list(n, req);
5680     case NVME_ID_CNS_CS_NS:
5681         return nvme_identify_ns_csi(n, req, true);
5682     case NVME_ID_CNS_CS_NS_PRESENT:
5683         return nvme_identify_ns_csi(n, req, false);
5684     case NVME_ID_CNS_CTRL:
5685         return nvme_identify_ctrl(n, req);
5686     case NVME_ID_CNS_CS_CTRL:
5687         return nvme_identify_ctrl_csi(n, req);
5688     case NVME_ID_CNS_NS_ACTIVE_LIST:
5689         return nvme_identify_nslist(n, req, true);
5690     case NVME_ID_CNS_NS_PRESENT_LIST:
5691         return nvme_identify_nslist(n, req, false);
5692     case NVME_ID_CNS_CS_NS_ACTIVE_LIST:
5693         return nvme_identify_nslist_csi(n, req, true);
5694     case NVME_ID_CNS_CS_NS_PRESENT_LIST:
5695         return nvme_identify_nslist_csi(n, req, false);
5696     case NVME_ID_CNS_NS_DESCR_LIST:
5697         return nvme_identify_ns_descr_list(n, req);
5698     case NVME_ID_CNS_IO_COMMAND_SET:
5699         return nvme_identify_cmd_set(n, req);
5700     default:
5701         trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
5702         return NVME_INVALID_FIELD | NVME_DNR;
5703     }
5704 }
5705 
5706 static uint16_t nvme_abort(NvmeCtrl *n, NvmeRequest *req)
5707 {
5708     uint16_t sqid = le32_to_cpu(req->cmd.cdw10) & 0xffff;
5709 
5710     req->cqe.result = 1;
5711     if (nvme_check_sqid(n, sqid)) {
5712         return NVME_INVALID_FIELD | NVME_DNR;
5713     }
5714 
5715     return NVME_SUCCESS;
5716 }
5717 
5718 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
5719 {
5720     trace_pci_nvme_setfeat_timestamp(ts);
5721 
5722     n->host_timestamp = le64_to_cpu(ts);
5723     n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5724 }
5725 
5726 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
5727 {
5728     uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5729     uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
5730 
5731     union nvme_timestamp {
5732         struct {
5733             uint64_t timestamp:48;
5734             uint64_t sync:1;
5735             uint64_t origin:3;
5736             uint64_t rsvd1:12;
5737         };
5738         uint64_t all;
5739     };
5740 
5741     union nvme_timestamp ts;
5742     ts.all = 0;
5743     ts.timestamp = n->host_timestamp + elapsed_time;
5744 
5745     /* If the host timestamp is non-zero, set the timestamp origin */
5746     ts.origin = n->host_timestamp ? 0x01 : 0x00;
5747 
5748     trace_pci_nvme_getfeat_timestamp(ts.all);
5749 
5750     return cpu_to_le64(ts.all);
5751 }
5752 
5753 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
5754 {
5755     uint64_t timestamp = nvme_get_timestamp(n);
5756 
5757     return nvme_c2h(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
5758 }
5759 
5760 static int nvme_get_feature_fdp(NvmeCtrl *n, uint32_t endgrpid,
5761                                 uint32_t *result)
5762 {
5763     *result = 0;
5764 
5765     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
5766         return NVME_INVALID_FIELD | NVME_DNR;
5767     }
5768 
5769     *result = FIELD_DP16(0, FEAT_FDP, FDPE, 1);
5770     *result = FIELD_DP16(*result, FEAT_FDP, CONF_NDX, 0);
5771 
5772     return NVME_SUCCESS;
5773 }
5774 
5775 static uint16_t nvme_get_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
5776                                             NvmeRequest *req, uint32_t *result)
5777 {
5778     NvmeCmd *cmd = &req->cmd;
5779     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
5780     uint16_t ph = cdw11 & 0xffff;
5781     uint8_t noet = (cdw11 >> 16) & 0xff;
5782     uint16_t ruhid, ret;
5783     uint32_t nentries = 0;
5784     uint8_t s_events_ndx = 0;
5785     size_t s_events_siz = sizeof(NvmeFdpEventDescr) * noet;
5786     g_autofree NvmeFdpEventDescr *s_events = g_malloc0(s_events_siz);
5787     NvmeRuHandle *ruh;
5788     NvmeFdpEventDescr *s_event;
5789 
5790     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
5791         return NVME_FDP_DISABLED | NVME_DNR;
5792     }
5793 
5794     if (!nvme_ph_valid(ns, ph)) {
5795         return NVME_INVALID_FIELD | NVME_DNR;
5796     }
5797 
5798     ruhid = ns->fdp.phs[ph];
5799     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
5800 
5801     assert(ruh);
5802 
5803     if (unlikely(noet == 0)) {
5804         return NVME_INVALID_FIELD | NVME_DNR;
5805     }
5806 
5807     for (uint8_t event_type = 0; event_type < FDP_EVT_MAX; event_type++) {
5808         uint8_t shift = nvme_fdp_evf_shifts[event_type];
5809         if (!shift && event_type) {
5810             /*
5811              * only first entry (event_type == 0) has a shift value of 0
5812              * other entries are simply unpopulated.
5813              */
5814             continue;
5815         }
5816 
5817         nentries++;
5818 
5819         s_event = &s_events[s_events_ndx];
5820         s_event->evt = event_type;
5821         s_event->evta = (ruh->event_filter >> shift) & 0x1;
5822 
5823         /* break if all `noet` entries are filled */
5824         if ((++s_events_ndx) == noet) {
5825             break;
5826         }
5827     }
5828 
5829     ret = nvme_c2h(n, s_events, s_events_siz, req);
5830     if (ret) {
5831         return ret;
5832     }
5833 
5834     *result = nentries;
5835     return NVME_SUCCESS;
5836 }
5837 
5838 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeRequest *req)
5839 {
5840     NvmeCmd *cmd = &req->cmd;
5841     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5842     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5843     uint32_t nsid = le32_to_cpu(cmd->nsid);
5844     uint32_t result;
5845     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
5846     NvmeGetFeatureSelect sel = NVME_GETFEAT_SELECT(dw10);
5847     uint16_t iv;
5848     NvmeNamespace *ns;
5849     int i;
5850     uint16_t endgrpid = 0, ret = NVME_SUCCESS;
5851 
5852     static const uint32_t nvme_feature_default[NVME_FID_MAX] = {
5853         [NVME_ARBITRATION] = NVME_ARB_AB_NOLIMIT,
5854     };
5855 
5856     trace_pci_nvme_getfeat(nvme_cid(req), nsid, fid, sel, dw11);
5857 
5858     if (!nvme_feature_support[fid]) {
5859         return NVME_INVALID_FIELD | NVME_DNR;
5860     }
5861 
5862     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
5863         if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5864             /*
5865              * The Reservation Notification Mask and Reservation Persistence
5866              * features require a status code of Invalid Field in Command when
5867              * NSID is FFFFFFFFh. Since the device does not support those
5868              * features we can always return Invalid Namespace or Format as we
5869              * should do for all other features.
5870              */
5871             return NVME_INVALID_NSID | NVME_DNR;
5872         }
5873 
5874         if (!nvme_ns(n, nsid)) {
5875             return NVME_INVALID_FIELD | NVME_DNR;
5876         }
5877     }
5878 
5879     switch (sel) {
5880     case NVME_GETFEAT_SELECT_CURRENT:
5881         break;
5882     case NVME_GETFEAT_SELECT_SAVED:
5883         /* no features are saveable by the controller; fallthrough */
5884     case NVME_GETFEAT_SELECT_DEFAULT:
5885         goto defaults;
5886     case NVME_GETFEAT_SELECT_CAP:
5887         result = nvme_feature_cap[fid];
5888         goto out;
5889     }
5890 
5891     switch (fid) {
5892     case NVME_TEMPERATURE_THRESHOLD:
5893         result = 0;
5894 
5895         /*
5896          * The controller only implements the Composite Temperature sensor, so
5897          * return 0 for all other sensors.
5898          */
5899         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
5900             goto out;
5901         }
5902 
5903         switch (NVME_TEMP_THSEL(dw11)) {
5904         case NVME_TEMP_THSEL_OVER:
5905             result = n->features.temp_thresh_hi;
5906             goto out;
5907         case NVME_TEMP_THSEL_UNDER:
5908             result = n->features.temp_thresh_low;
5909             goto out;
5910         }
5911 
5912         return NVME_INVALID_FIELD | NVME_DNR;
5913     case NVME_ERROR_RECOVERY:
5914         if (!nvme_nsid_valid(n, nsid)) {
5915             return NVME_INVALID_NSID | NVME_DNR;
5916         }
5917 
5918         ns = nvme_ns(n, nsid);
5919         if (unlikely(!ns)) {
5920             return NVME_INVALID_FIELD | NVME_DNR;
5921         }
5922 
5923         result = ns->features.err_rec;
5924         goto out;
5925     case NVME_VOLATILE_WRITE_CACHE:
5926         result = 0;
5927         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5928             ns = nvme_ns(n, i);
5929             if (!ns) {
5930                 continue;
5931             }
5932 
5933             result = blk_enable_write_cache(ns->blkconf.blk);
5934             if (result) {
5935                 break;
5936             }
5937         }
5938         trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
5939         goto out;
5940     case NVME_ASYNCHRONOUS_EVENT_CONF:
5941         result = n->features.async_config;
5942         goto out;
5943     case NVME_TIMESTAMP:
5944         return nvme_get_feature_timestamp(n, req);
5945     case NVME_HOST_BEHAVIOR_SUPPORT:
5946         return nvme_c2h(n, (uint8_t *)&n->features.hbs,
5947                         sizeof(n->features.hbs), req);
5948     case NVME_FDP_MODE:
5949         endgrpid = dw11 & 0xff;
5950 
5951         if (endgrpid != 0x1) {
5952             return NVME_INVALID_FIELD | NVME_DNR;
5953         }
5954 
5955         ret = nvme_get_feature_fdp(n, endgrpid, &result);
5956         if (ret) {
5957             return ret;
5958         }
5959         goto out;
5960     case NVME_FDP_EVENTS:
5961         if (!nvme_nsid_valid(n, nsid)) {
5962             return NVME_INVALID_NSID | NVME_DNR;
5963         }
5964 
5965         ns = nvme_ns(n, nsid);
5966         if (unlikely(!ns)) {
5967             return NVME_INVALID_FIELD | NVME_DNR;
5968         }
5969 
5970         ret = nvme_get_feature_fdp_events(n, ns, req, &result);
5971         if (ret) {
5972             return ret;
5973         }
5974         goto out;
5975     default:
5976         break;
5977     }
5978 
5979 defaults:
5980     switch (fid) {
5981     case NVME_TEMPERATURE_THRESHOLD:
5982         result = 0;
5983 
5984         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
5985             break;
5986         }
5987 
5988         if (NVME_TEMP_THSEL(dw11) == NVME_TEMP_THSEL_OVER) {
5989             result = NVME_TEMPERATURE_WARNING;
5990         }
5991 
5992         break;
5993     case NVME_NUMBER_OF_QUEUES:
5994         result = (n->conf_ioqpairs - 1) | ((n->conf_ioqpairs - 1) << 16);
5995         trace_pci_nvme_getfeat_numq(result);
5996         break;
5997     case NVME_INTERRUPT_VECTOR_CONF:
5998         iv = dw11 & 0xffff;
5999         if (iv >= n->conf_ioqpairs + 1) {
6000             return NVME_INVALID_FIELD | NVME_DNR;
6001         }
6002 
6003         result = iv;
6004         if (iv == n->admin_cq.vector) {
6005             result |= NVME_INTVC_NOCOALESCING;
6006         }
6007         break;
6008     case NVME_FDP_MODE:
6009         endgrpid = dw11 & 0xff;
6010 
6011         if (endgrpid != 0x1) {
6012             return NVME_INVALID_FIELD | NVME_DNR;
6013         }
6014 
6015         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6016         if (ret) {
6017             return ret;
6018         }
6019         goto out;
6020 
6021         break;
6022     default:
6023         result = nvme_feature_default[fid];
6024         break;
6025     }
6026 
6027 out:
6028     req->cqe.result = cpu_to_le32(result);
6029     return ret;
6030 }
6031 
6032 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
6033 {
6034     uint16_t ret;
6035     uint64_t timestamp;
6036 
6037     ret = nvme_h2c(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
6038     if (ret) {
6039         return ret;
6040     }
6041 
6042     nvme_set_timestamp(n, timestamp);
6043 
6044     return NVME_SUCCESS;
6045 }
6046 
6047 static uint16_t nvme_set_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
6048                                             NvmeRequest *req)
6049 {
6050     NvmeCmd *cmd = &req->cmd;
6051     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
6052     uint16_t ph = cdw11 & 0xffff;
6053     uint8_t noet = (cdw11 >> 16) & 0xff;
6054     uint16_t ret, ruhid;
6055     uint8_t enable = le32_to_cpu(cmd->cdw12) & 0x1;
6056     uint8_t event_mask = 0;
6057     unsigned int i;
6058     g_autofree uint8_t *events = g_malloc0(noet);
6059     NvmeRuHandle *ruh = NULL;
6060 
6061     assert(ns);
6062 
6063     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6064         return NVME_FDP_DISABLED | NVME_DNR;
6065     }
6066 
6067     if (!nvme_ph_valid(ns, ph)) {
6068         return NVME_INVALID_FIELD | NVME_DNR;
6069     }
6070 
6071     ruhid = ns->fdp.phs[ph];
6072     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
6073 
6074     ret = nvme_h2c(n, events, noet, req);
6075     if (ret) {
6076         return ret;
6077     }
6078 
6079     for (i = 0; i < noet; i++) {
6080         event_mask |= (1 << nvme_fdp_evf_shifts[events[i]]);
6081     }
6082 
6083     if (enable) {
6084         ruh->event_filter |= event_mask;
6085     } else {
6086         ruh->event_filter = ruh->event_filter & ~event_mask;
6087     }
6088 
6089     return NVME_SUCCESS;
6090 }
6091 
6092 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
6093 {
6094     NvmeNamespace *ns = NULL;
6095 
6096     NvmeCmd *cmd = &req->cmd;
6097     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
6098     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
6099     uint32_t nsid = le32_to_cpu(cmd->nsid);
6100     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
6101     uint8_t save = NVME_SETFEAT_SAVE(dw10);
6102     uint16_t status;
6103     int i;
6104 
6105     trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
6106 
6107     if (save && !(nvme_feature_cap[fid] & NVME_FEAT_CAP_SAVE)) {
6108         return NVME_FID_NOT_SAVEABLE | NVME_DNR;
6109     }
6110 
6111     if (!nvme_feature_support[fid]) {
6112         return NVME_INVALID_FIELD | NVME_DNR;
6113     }
6114 
6115     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
6116         if (nsid != NVME_NSID_BROADCAST) {
6117             if (!nvme_nsid_valid(n, nsid)) {
6118                 return NVME_INVALID_NSID | NVME_DNR;
6119             }
6120 
6121             ns = nvme_ns(n, nsid);
6122             if (unlikely(!ns)) {
6123                 return NVME_INVALID_FIELD | NVME_DNR;
6124             }
6125         }
6126     } else if (nsid && nsid != NVME_NSID_BROADCAST) {
6127         if (!nvme_nsid_valid(n, nsid)) {
6128             return NVME_INVALID_NSID | NVME_DNR;
6129         }
6130 
6131         return NVME_FEAT_NOT_NS_SPEC | NVME_DNR;
6132     }
6133 
6134     if (!(nvme_feature_cap[fid] & NVME_FEAT_CAP_CHANGE)) {
6135         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6136     }
6137 
6138     switch (fid) {
6139     case NVME_TEMPERATURE_THRESHOLD:
6140         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6141             break;
6142         }
6143 
6144         switch (NVME_TEMP_THSEL(dw11)) {
6145         case NVME_TEMP_THSEL_OVER:
6146             n->features.temp_thresh_hi = NVME_TEMP_TMPTH(dw11);
6147             break;
6148         case NVME_TEMP_THSEL_UNDER:
6149             n->features.temp_thresh_low = NVME_TEMP_TMPTH(dw11);
6150             break;
6151         default:
6152             return NVME_INVALID_FIELD | NVME_DNR;
6153         }
6154 
6155         if ((n->temperature >= n->features.temp_thresh_hi) ||
6156             (n->temperature <= n->features.temp_thresh_low)) {
6157             nvme_smart_event(n, NVME_SMART_TEMPERATURE);
6158         }
6159 
6160         break;
6161     case NVME_ERROR_RECOVERY:
6162         if (nsid == NVME_NSID_BROADCAST) {
6163             for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6164                 ns = nvme_ns(n, i);
6165 
6166                 if (!ns) {
6167                     continue;
6168                 }
6169 
6170                 if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat)) {
6171                     ns->features.err_rec = dw11;
6172                 }
6173             }
6174 
6175             break;
6176         }
6177 
6178         assert(ns);
6179         if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat))  {
6180             ns->features.err_rec = dw11;
6181         }
6182         break;
6183     case NVME_VOLATILE_WRITE_CACHE:
6184         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6185             ns = nvme_ns(n, i);
6186             if (!ns) {
6187                 continue;
6188             }
6189 
6190             if (!(dw11 & 0x1) && blk_enable_write_cache(ns->blkconf.blk)) {
6191                 blk_flush(ns->blkconf.blk);
6192             }
6193 
6194             blk_set_enable_write_cache(ns->blkconf.blk, dw11 & 1);
6195         }
6196 
6197         break;
6198 
6199     case NVME_NUMBER_OF_QUEUES:
6200         if (n->qs_created) {
6201             return NVME_CMD_SEQ_ERROR | NVME_DNR;
6202         }
6203 
6204         /*
6205          * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
6206          * and NSQR.
6207          */
6208         if ((dw11 & 0xffff) == 0xffff || ((dw11 >> 16) & 0xffff) == 0xffff) {
6209             return NVME_INVALID_FIELD | NVME_DNR;
6210         }
6211 
6212         trace_pci_nvme_setfeat_numq((dw11 & 0xffff) + 1,
6213                                     ((dw11 >> 16) & 0xffff) + 1,
6214                                     n->conf_ioqpairs,
6215                                     n->conf_ioqpairs);
6216         req->cqe.result = cpu_to_le32((n->conf_ioqpairs - 1) |
6217                                       ((n->conf_ioqpairs - 1) << 16));
6218         break;
6219     case NVME_ASYNCHRONOUS_EVENT_CONF:
6220         n->features.async_config = dw11;
6221         break;
6222     case NVME_TIMESTAMP:
6223         return nvme_set_feature_timestamp(n, req);
6224     case NVME_HOST_BEHAVIOR_SUPPORT:
6225         status = nvme_h2c(n, (uint8_t *)&n->features.hbs,
6226                           sizeof(n->features.hbs), req);
6227         if (status) {
6228             return status;
6229         }
6230 
6231         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6232             ns = nvme_ns(n, i);
6233 
6234             if (!ns) {
6235                 continue;
6236             }
6237 
6238             ns->id_ns.nlbaf = ns->nlbaf - 1;
6239             if (!n->features.hbs.lbafee) {
6240                 ns->id_ns.nlbaf = MIN(ns->id_ns.nlbaf, 15);
6241             }
6242         }
6243 
6244         return status;
6245     case NVME_COMMAND_SET_PROFILE:
6246         if (dw11 & 0x1ff) {
6247             trace_pci_nvme_err_invalid_iocsci(dw11 & 0x1ff);
6248             return NVME_CMD_SET_CMB_REJECTED | NVME_DNR;
6249         }
6250         break;
6251     case NVME_FDP_MODE:
6252         /* spec: abort with cmd seq err if there's one or more NS' in endgrp */
6253         return NVME_CMD_SEQ_ERROR | NVME_DNR;
6254     case NVME_FDP_EVENTS:
6255         return nvme_set_feature_fdp_events(n, ns, req);
6256     default:
6257         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6258     }
6259     return NVME_SUCCESS;
6260 }
6261 
6262 static uint16_t nvme_aer(NvmeCtrl *n, NvmeRequest *req)
6263 {
6264     trace_pci_nvme_aer(nvme_cid(req));
6265 
6266     if (n->outstanding_aers > n->params.aerl) {
6267         trace_pci_nvme_aer_aerl_exceeded();
6268         return NVME_AER_LIMIT_EXCEEDED;
6269     }
6270 
6271     n->aer_reqs[n->outstanding_aers] = req;
6272     n->outstanding_aers++;
6273 
6274     if (!QTAILQ_EMPTY(&n->aer_queue)) {
6275         nvme_process_aers(n);
6276     }
6277 
6278     return NVME_NO_COMPLETE;
6279 }
6280 
6281 static void nvme_update_dmrsl(NvmeCtrl *n)
6282 {
6283     int nsid;
6284 
6285     for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
6286         NvmeNamespace *ns = nvme_ns(n, nsid);
6287         if (!ns) {
6288             continue;
6289         }
6290 
6291         n->dmrsl = MIN_NON_ZERO(n->dmrsl,
6292                                 BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
6293     }
6294 }
6295 
6296 static void nvme_select_iocs_ns(NvmeCtrl *n, NvmeNamespace *ns)
6297 {
6298     uint32_t cc = ldl_le_p(&n->bar.cc);
6299 
6300     ns->iocs = nvme_cse_iocs_none;
6301     switch (ns->csi) {
6302     case NVME_CSI_NVM:
6303         if (NVME_CC_CSS(cc) != NVME_CC_CSS_ADMIN_ONLY) {
6304             ns->iocs = nvme_cse_iocs_nvm;
6305         }
6306         break;
6307     case NVME_CSI_ZONED:
6308         if (NVME_CC_CSS(cc) == NVME_CC_CSS_CSI) {
6309             ns->iocs = nvme_cse_iocs_zoned;
6310         } else if (NVME_CC_CSS(cc) == NVME_CC_CSS_NVM) {
6311             ns->iocs = nvme_cse_iocs_nvm;
6312         }
6313         break;
6314     }
6315 }
6316 
6317 static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
6318 {
6319     NvmeNamespace *ns;
6320     NvmeCtrl *ctrl;
6321     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
6322     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6323     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6324     uint8_t sel = dw10 & 0xf;
6325     uint16_t *nr_ids = &list[0];
6326     uint16_t *ids = &list[1];
6327     uint16_t ret;
6328     int i;
6329 
6330     trace_pci_nvme_ns_attachment(nvme_cid(req), dw10 & 0xf);
6331 
6332     if (!nvme_nsid_valid(n, nsid)) {
6333         return NVME_INVALID_NSID | NVME_DNR;
6334     }
6335 
6336     ns = nvme_subsys_ns(n->subsys, nsid);
6337     if (!ns) {
6338         return NVME_INVALID_FIELD | NVME_DNR;
6339     }
6340 
6341     ret = nvme_h2c(n, (uint8_t *)list, 4096, req);
6342     if (ret) {
6343         return ret;
6344     }
6345 
6346     if (!*nr_ids) {
6347         return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6348     }
6349 
6350     *nr_ids = MIN(*nr_ids, NVME_CONTROLLER_LIST_SIZE - 1);
6351     for (i = 0; i < *nr_ids; i++) {
6352         ctrl = nvme_subsys_ctrl(n->subsys, ids[i]);
6353         if (!ctrl) {
6354             return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6355         }
6356 
6357         switch (sel) {
6358         case NVME_NS_ATTACHMENT_ATTACH:
6359             if (nvme_ns(ctrl, nsid)) {
6360                 return NVME_NS_ALREADY_ATTACHED | NVME_DNR;
6361             }
6362 
6363             if (ns->attached && !ns->params.shared) {
6364                 return NVME_NS_PRIVATE | NVME_DNR;
6365             }
6366 
6367             nvme_attach_ns(ctrl, ns);
6368             nvme_select_iocs_ns(ctrl, ns);
6369 
6370             break;
6371 
6372         case NVME_NS_ATTACHMENT_DETACH:
6373             if (!nvme_ns(ctrl, nsid)) {
6374                 return NVME_NS_NOT_ATTACHED | NVME_DNR;
6375             }
6376 
6377             ctrl->namespaces[nsid] = NULL;
6378             ns->attached--;
6379 
6380             nvme_update_dmrsl(ctrl);
6381 
6382             break;
6383 
6384         default:
6385             return NVME_INVALID_FIELD | NVME_DNR;
6386         }
6387 
6388         /*
6389          * Add namespace id to the changed namespace id list for event clearing
6390          * via Get Log Page command.
6391          */
6392         if (!test_and_set_bit(nsid, ctrl->changed_nsids)) {
6393             nvme_enqueue_event(ctrl, NVME_AER_TYPE_NOTICE,
6394                                NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED,
6395                                NVME_LOG_CHANGED_NSLIST);
6396         }
6397     }
6398 
6399     return NVME_SUCCESS;
6400 }
6401 
6402 typedef struct NvmeFormatAIOCB {
6403     BlockAIOCB common;
6404     BlockAIOCB *aiocb;
6405     NvmeRequest *req;
6406     int ret;
6407 
6408     NvmeNamespace *ns;
6409     uint32_t nsid;
6410     bool broadcast;
6411     int64_t offset;
6412 
6413     uint8_t lbaf;
6414     uint8_t mset;
6415     uint8_t pi;
6416     uint8_t pil;
6417 } NvmeFormatAIOCB;
6418 
6419 static void nvme_format_cancel(BlockAIOCB *aiocb)
6420 {
6421     NvmeFormatAIOCB *iocb = container_of(aiocb, NvmeFormatAIOCB, common);
6422 
6423     iocb->ret = -ECANCELED;
6424 
6425     if (iocb->aiocb) {
6426         blk_aio_cancel_async(iocb->aiocb);
6427         iocb->aiocb = NULL;
6428     }
6429 }
6430 
6431 static const AIOCBInfo nvme_format_aiocb_info = {
6432     .aiocb_size = sizeof(NvmeFormatAIOCB),
6433     .cancel_async = nvme_format_cancel,
6434     .get_aio_context = nvme_get_aio_context,
6435 };
6436 
6437 static void nvme_format_set(NvmeNamespace *ns, uint8_t lbaf, uint8_t mset,
6438                             uint8_t pi, uint8_t pil)
6439 {
6440     uint8_t lbafl = lbaf & 0xf;
6441     uint8_t lbafu = lbaf >> 4;
6442 
6443     trace_pci_nvme_format_set(ns->params.nsid, lbaf, mset, pi, pil);
6444 
6445     ns->id_ns.dps = (pil << 3) | pi;
6446     ns->id_ns.flbas = (lbafu << 5) | (mset << 4) | lbafl;
6447 
6448     nvme_ns_init_format(ns);
6449 }
6450 
6451 static void nvme_do_format(NvmeFormatAIOCB *iocb);
6452 
6453 static void nvme_format_ns_cb(void *opaque, int ret)
6454 {
6455     NvmeFormatAIOCB *iocb = opaque;
6456     NvmeNamespace *ns = iocb->ns;
6457     int bytes;
6458 
6459     if (iocb->ret < 0) {
6460         goto done;
6461     } else if (ret < 0) {
6462         iocb->ret = ret;
6463         goto done;
6464     }
6465 
6466     assert(ns);
6467 
6468     if (iocb->offset < ns->size) {
6469         bytes = MIN(BDRV_REQUEST_MAX_BYTES, ns->size - iocb->offset);
6470 
6471         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, iocb->offset,
6472                                             bytes, BDRV_REQ_MAY_UNMAP,
6473                                             nvme_format_ns_cb, iocb);
6474 
6475         iocb->offset += bytes;
6476         return;
6477     }
6478 
6479     nvme_format_set(ns, iocb->lbaf, iocb->mset, iocb->pi, iocb->pil);
6480     ns->status = 0x0;
6481     iocb->ns = NULL;
6482     iocb->offset = 0;
6483 
6484 done:
6485     nvme_do_format(iocb);
6486 }
6487 
6488 static uint16_t nvme_format_check(NvmeNamespace *ns, uint8_t lbaf, uint8_t pi)
6489 {
6490     if (ns->params.zoned) {
6491         return NVME_INVALID_FORMAT | NVME_DNR;
6492     }
6493 
6494     if (lbaf > ns->id_ns.nlbaf) {
6495         return NVME_INVALID_FORMAT | NVME_DNR;
6496     }
6497 
6498     if (pi && (ns->id_ns.lbaf[lbaf].ms < nvme_pi_tuple_size(ns))) {
6499         return NVME_INVALID_FORMAT | NVME_DNR;
6500     }
6501 
6502     if (pi && pi > NVME_ID_NS_DPS_TYPE_3) {
6503         return NVME_INVALID_FIELD | NVME_DNR;
6504     }
6505 
6506     return NVME_SUCCESS;
6507 }
6508 
6509 static void nvme_do_format(NvmeFormatAIOCB *iocb)
6510 {
6511     NvmeRequest *req = iocb->req;
6512     NvmeCtrl *n = nvme_ctrl(req);
6513     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6514     uint8_t lbaf = dw10 & 0xf;
6515     uint8_t pi = (dw10 >> 5) & 0x7;
6516     uint16_t status;
6517     int i;
6518 
6519     if (iocb->ret < 0) {
6520         goto done;
6521     }
6522 
6523     if (iocb->broadcast) {
6524         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
6525             iocb->ns = nvme_ns(n, i);
6526             if (iocb->ns) {
6527                 iocb->nsid = i;
6528                 break;
6529             }
6530         }
6531     }
6532 
6533     if (!iocb->ns) {
6534         goto done;
6535     }
6536 
6537     status = nvme_format_check(iocb->ns, lbaf, pi);
6538     if (status) {
6539         req->status = status;
6540         goto done;
6541     }
6542 
6543     iocb->ns->status = NVME_FORMAT_IN_PROGRESS;
6544     nvme_format_ns_cb(iocb, 0);
6545     return;
6546 
6547 done:
6548     iocb->common.cb(iocb->common.opaque, iocb->ret);
6549     qemu_aio_unref(iocb);
6550 }
6551 
6552 static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
6553 {
6554     NvmeFormatAIOCB *iocb;
6555     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6556     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6557     uint8_t lbaf = dw10 & 0xf;
6558     uint8_t mset = (dw10 >> 4) & 0x1;
6559     uint8_t pi = (dw10 >> 5) & 0x7;
6560     uint8_t pil = (dw10 >> 8) & 0x1;
6561     uint8_t lbafu = (dw10 >> 12) & 0x3;
6562     uint16_t status;
6563 
6564     iocb = qemu_aio_get(&nvme_format_aiocb_info, NULL, nvme_misc_cb, req);
6565 
6566     iocb->req = req;
6567     iocb->ret = 0;
6568     iocb->ns = NULL;
6569     iocb->nsid = 0;
6570     iocb->lbaf = lbaf;
6571     iocb->mset = mset;
6572     iocb->pi = pi;
6573     iocb->pil = pil;
6574     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
6575     iocb->offset = 0;
6576 
6577     if (n->features.hbs.lbafee) {
6578         iocb->lbaf |= lbafu << 4;
6579     }
6580 
6581     if (!iocb->broadcast) {
6582         if (!nvme_nsid_valid(n, nsid)) {
6583             status = NVME_INVALID_NSID | NVME_DNR;
6584             goto out;
6585         }
6586 
6587         iocb->ns = nvme_ns(n, nsid);
6588         if (!iocb->ns) {
6589             status = NVME_INVALID_FIELD | NVME_DNR;
6590             goto out;
6591         }
6592     }
6593 
6594     req->aiocb = &iocb->common;
6595     nvme_do_format(iocb);
6596 
6597     return NVME_NO_COMPLETE;
6598 
6599 out:
6600     qemu_aio_unref(iocb);
6601 
6602     return status;
6603 }
6604 
6605 static void nvme_get_virt_res_num(NvmeCtrl *n, uint8_t rt, int *num_total,
6606                                   int *num_prim, int *num_sec)
6607 {
6608     *num_total = le32_to_cpu(rt ?
6609                              n->pri_ctrl_cap.vifrt : n->pri_ctrl_cap.vqfrt);
6610     *num_prim = le16_to_cpu(rt ?
6611                             n->pri_ctrl_cap.virfap : n->pri_ctrl_cap.vqrfap);
6612     *num_sec = le16_to_cpu(rt ? n->pri_ctrl_cap.virfa : n->pri_ctrl_cap.vqrfa);
6613 }
6614 
6615 static uint16_t nvme_assign_virt_res_to_prim(NvmeCtrl *n, NvmeRequest *req,
6616                                              uint16_t cntlid, uint8_t rt,
6617                                              int nr)
6618 {
6619     int num_total, num_prim, num_sec;
6620 
6621     if (cntlid != n->cntlid) {
6622         return NVME_INVALID_CTRL_ID | NVME_DNR;
6623     }
6624 
6625     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6626 
6627     if (nr > num_total) {
6628         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6629     }
6630 
6631     if (nr > num_total - num_sec) {
6632         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6633     }
6634 
6635     if (rt) {
6636         n->next_pri_ctrl_cap.virfap = cpu_to_le16(nr);
6637     } else {
6638         n->next_pri_ctrl_cap.vqrfap = cpu_to_le16(nr);
6639     }
6640 
6641     req->cqe.result = cpu_to_le32(nr);
6642     return req->status;
6643 }
6644 
6645 static void nvme_update_virt_res(NvmeCtrl *n, NvmeSecCtrlEntry *sctrl,
6646                                  uint8_t rt, int nr)
6647 {
6648     int prev_nr, prev_total;
6649 
6650     if (rt) {
6651         prev_nr = le16_to_cpu(sctrl->nvi);
6652         prev_total = le32_to_cpu(n->pri_ctrl_cap.virfa);
6653         sctrl->nvi = cpu_to_le16(nr);
6654         n->pri_ctrl_cap.virfa = cpu_to_le32(prev_total + nr - prev_nr);
6655     } else {
6656         prev_nr = le16_to_cpu(sctrl->nvq);
6657         prev_total = le32_to_cpu(n->pri_ctrl_cap.vqrfa);
6658         sctrl->nvq = cpu_to_le16(nr);
6659         n->pri_ctrl_cap.vqrfa = cpu_to_le32(prev_total + nr - prev_nr);
6660     }
6661 }
6662 
6663 static uint16_t nvme_assign_virt_res_to_sec(NvmeCtrl *n, NvmeRequest *req,
6664                                             uint16_t cntlid, uint8_t rt, int nr)
6665 {
6666     int num_total, num_prim, num_sec, num_free, diff, limit;
6667     NvmeSecCtrlEntry *sctrl;
6668 
6669     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6670     if (!sctrl) {
6671         return NVME_INVALID_CTRL_ID | NVME_DNR;
6672     }
6673 
6674     if (sctrl->scs) {
6675         return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6676     }
6677 
6678     limit = le16_to_cpu(rt ? n->pri_ctrl_cap.vifrsm : n->pri_ctrl_cap.vqfrsm);
6679     if (nr > limit) {
6680         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6681     }
6682 
6683     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6684     num_free = num_total - num_prim - num_sec;
6685     diff = nr - le16_to_cpu(rt ? sctrl->nvi : sctrl->nvq);
6686 
6687     if (diff > num_free) {
6688         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6689     }
6690 
6691     nvme_update_virt_res(n, sctrl, rt, nr);
6692     req->cqe.result = cpu_to_le32(nr);
6693 
6694     return req->status;
6695 }
6696 
6697 static uint16_t nvme_virt_set_state(NvmeCtrl *n, uint16_t cntlid, bool online)
6698 {
6699     PCIDevice *pci = PCI_DEVICE(n);
6700     NvmeCtrl *sn = NULL;
6701     NvmeSecCtrlEntry *sctrl;
6702     int vf_index;
6703 
6704     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6705     if (!sctrl) {
6706         return NVME_INVALID_CTRL_ID | NVME_DNR;
6707     }
6708 
6709     if (!pci_is_vf(pci)) {
6710         vf_index = le16_to_cpu(sctrl->vfn) - 1;
6711         sn = NVME(pcie_sriov_get_vf_at_index(pci, vf_index));
6712     }
6713 
6714     if (online) {
6715         if (!sctrl->nvi || (le16_to_cpu(sctrl->nvq) < 2) || !sn) {
6716             return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6717         }
6718 
6719         if (!sctrl->scs) {
6720             sctrl->scs = 0x1;
6721             nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6722         }
6723     } else {
6724         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_INTERRUPT, 0);
6725         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_QUEUE, 0);
6726 
6727         if (sctrl->scs) {
6728             sctrl->scs = 0x0;
6729             if (sn) {
6730                 nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6731             }
6732         }
6733     }
6734 
6735     return NVME_SUCCESS;
6736 }
6737 
6738 static uint16_t nvme_virt_mngmt(NvmeCtrl *n, NvmeRequest *req)
6739 {
6740     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6741     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
6742     uint8_t act = dw10 & 0xf;
6743     uint8_t rt = (dw10 >> 8) & 0x7;
6744     uint16_t cntlid = (dw10 >> 16) & 0xffff;
6745     int nr = dw11 & 0xffff;
6746 
6747     trace_pci_nvme_virt_mngmt(nvme_cid(req), act, cntlid, rt ? "VI" : "VQ", nr);
6748 
6749     if (rt != NVME_VIRT_RES_QUEUE && rt != NVME_VIRT_RES_INTERRUPT) {
6750         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6751     }
6752 
6753     switch (act) {
6754     case NVME_VIRT_MNGMT_ACTION_SEC_ASSIGN:
6755         return nvme_assign_virt_res_to_sec(n, req, cntlid, rt, nr);
6756     case NVME_VIRT_MNGMT_ACTION_PRM_ALLOC:
6757         return nvme_assign_virt_res_to_prim(n, req, cntlid, rt, nr);
6758     case NVME_VIRT_MNGMT_ACTION_SEC_ONLINE:
6759         return nvme_virt_set_state(n, cntlid, true);
6760     case NVME_VIRT_MNGMT_ACTION_SEC_OFFLINE:
6761         return nvme_virt_set_state(n, cntlid, false);
6762     default:
6763         return NVME_INVALID_FIELD | NVME_DNR;
6764     }
6765 }
6766 
6767 static uint16_t nvme_dbbuf_config(NvmeCtrl *n, const NvmeRequest *req)
6768 {
6769     PCIDevice *pci = PCI_DEVICE(n);
6770     uint64_t dbs_addr = le64_to_cpu(req->cmd.dptr.prp1);
6771     uint64_t eis_addr = le64_to_cpu(req->cmd.dptr.prp2);
6772     int i;
6773 
6774     /* Address should be page aligned */
6775     if (dbs_addr & (n->page_size - 1) || eis_addr & (n->page_size - 1)) {
6776         return NVME_INVALID_FIELD | NVME_DNR;
6777     }
6778 
6779     /* Save shadow buffer base addr for use during queue creation */
6780     n->dbbuf_dbs = dbs_addr;
6781     n->dbbuf_eis = eis_addr;
6782     n->dbbuf_enabled = true;
6783 
6784     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
6785         NvmeSQueue *sq = n->sq[i];
6786         NvmeCQueue *cq = n->cq[i];
6787 
6788         if (sq) {
6789             /*
6790              * CAP.DSTRD is 0, so offset of ith sq db_addr is (i<<3)
6791              * nvme_process_db() uses this hard-coded way to calculate
6792              * doorbell offsets. Be consistent with that here.
6793              */
6794             sq->db_addr = dbs_addr + (i << 3);
6795             sq->ei_addr = eis_addr + (i << 3);
6796             pci_dma_write(pci, sq->db_addr, &sq->tail, sizeof(sq->tail));
6797 
6798             if (n->params.ioeventfd && sq->sqid != 0) {
6799                 if (!nvme_init_sq_ioeventfd(sq)) {
6800                     sq->ioeventfd_enabled = true;
6801                 }
6802             }
6803         }
6804 
6805         if (cq) {
6806             /* CAP.DSTRD is 0, so offset of ith cq db_addr is (i<<3)+(1<<2) */
6807             cq->db_addr = dbs_addr + (i << 3) + (1 << 2);
6808             cq->ei_addr = eis_addr + (i << 3) + (1 << 2);
6809             pci_dma_write(pci, cq->db_addr, &cq->head, sizeof(cq->head));
6810 
6811             if (n->params.ioeventfd && cq->cqid != 0) {
6812                 if (!nvme_init_cq_ioeventfd(cq)) {
6813                     cq->ioeventfd_enabled = true;
6814                 }
6815             }
6816         }
6817     }
6818 
6819     trace_pci_nvme_dbbuf_config(dbs_addr, eis_addr);
6820 
6821     return NVME_SUCCESS;
6822 }
6823 
6824 static uint16_t nvme_directive_send(NvmeCtrl *n, NvmeRequest *req)
6825 {
6826     return NVME_INVALID_FIELD | NVME_DNR;
6827 }
6828 
6829 static uint16_t nvme_directive_receive(NvmeCtrl *n, NvmeRequest *req)
6830 {
6831     NvmeNamespace *ns;
6832     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6833     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
6834     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6835     uint8_t doper, dtype;
6836     uint32_t numd, trans_len;
6837     NvmeDirectiveIdentify id = {
6838         .supported = 1 << NVME_DIRECTIVE_IDENTIFY,
6839         .enabled = 1 << NVME_DIRECTIVE_IDENTIFY,
6840     };
6841 
6842     numd = dw10 + 1;
6843     doper = dw11 & 0xff;
6844     dtype = (dw11 >> 8) & 0xff;
6845 
6846     trans_len = MIN(sizeof(NvmeDirectiveIdentify), numd << 2);
6847 
6848     if (nsid == NVME_NSID_BROADCAST || dtype != NVME_DIRECTIVE_IDENTIFY ||
6849         doper != NVME_DIRECTIVE_RETURN_PARAMS) {
6850         return NVME_INVALID_FIELD | NVME_DNR;
6851     }
6852 
6853     ns = nvme_ns(n, nsid);
6854     if (!ns) {
6855         return NVME_INVALID_FIELD | NVME_DNR;
6856     }
6857 
6858     switch (dtype) {
6859     case NVME_DIRECTIVE_IDENTIFY:
6860         switch (doper) {
6861         case NVME_DIRECTIVE_RETURN_PARAMS:
6862             if (ns->endgrp->fdp.enabled) {
6863                 id.supported |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6864                 id.enabled |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6865                 id.persistent |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6866             }
6867 
6868             return nvme_c2h(n, (uint8_t *)&id, trans_len, req);
6869 
6870         default:
6871             return NVME_INVALID_FIELD | NVME_DNR;
6872         }
6873 
6874     default:
6875         return NVME_INVALID_FIELD;
6876     }
6877 }
6878 
6879 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
6880 {
6881     trace_pci_nvme_admin_cmd(nvme_cid(req), nvme_sqid(req), req->cmd.opcode,
6882                              nvme_adm_opc_str(req->cmd.opcode));
6883 
6884     if (!(nvme_cse_acs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
6885         trace_pci_nvme_err_invalid_admin_opc(req->cmd.opcode);
6886         return NVME_INVALID_OPCODE | NVME_DNR;
6887     }
6888 
6889     /* SGLs shall not be used for Admin commands in NVMe over PCIe */
6890     if (NVME_CMD_FLAGS_PSDT(req->cmd.flags) != NVME_PSDT_PRP) {
6891         return NVME_INVALID_FIELD | NVME_DNR;
6892     }
6893 
6894     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
6895         return NVME_INVALID_FIELD;
6896     }
6897 
6898     switch (req->cmd.opcode) {
6899     case NVME_ADM_CMD_DELETE_SQ:
6900         return nvme_del_sq(n, req);
6901     case NVME_ADM_CMD_CREATE_SQ:
6902         return nvme_create_sq(n, req);
6903     case NVME_ADM_CMD_GET_LOG_PAGE:
6904         return nvme_get_log(n, req);
6905     case NVME_ADM_CMD_DELETE_CQ:
6906         return nvme_del_cq(n, req);
6907     case NVME_ADM_CMD_CREATE_CQ:
6908         return nvme_create_cq(n, req);
6909     case NVME_ADM_CMD_IDENTIFY:
6910         return nvme_identify(n, req);
6911     case NVME_ADM_CMD_ABORT:
6912         return nvme_abort(n, req);
6913     case NVME_ADM_CMD_SET_FEATURES:
6914         return nvme_set_feature(n, req);
6915     case NVME_ADM_CMD_GET_FEATURES:
6916         return nvme_get_feature(n, req);
6917     case NVME_ADM_CMD_ASYNC_EV_REQ:
6918         return nvme_aer(n, req);
6919     case NVME_ADM_CMD_NS_ATTACHMENT:
6920         return nvme_ns_attachment(n, req);
6921     case NVME_ADM_CMD_VIRT_MNGMT:
6922         return nvme_virt_mngmt(n, req);
6923     case NVME_ADM_CMD_DBBUF_CONFIG:
6924         return nvme_dbbuf_config(n, req);
6925     case NVME_ADM_CMD_FORMAT_NVM:
6926         return nvme_format(n, req);
6927     case NVME_ADM_CMD_DIRECTIVE_SEND:
6928         return nvme_directive_send(n, req);
6929     case NVME_ADM_CMD_DIRECTIVE_RECV:
6930         return nvme_directive_receive(n, req);
6931     default:
6932         assert(false);
6933     }
6934 
6935     return NVME_INVALID_OPCODE | NVME_DNR;
6936 }
6937 
6938 static void nvme_update_sq_eventidx(const NvmeSQueue *sq)
6939 {
6940     uint32_t v = cpu_to_le32(sq->tail);
6941 
6942     trace_pci_nvme_update_sq_eventidx(sq->sqid, sq->tail);
6943 
6944     pci_dma_write(PCI_DEVICE(sq->ctrl), sq->ei_addr, &v, sizeof(v));
6945 }
6946 
6947 static void nvme_update_sq_tail(NvmeSQueue *sq)
6948 {
6949     uint32_t v;
6950 
6951     pci_dma_read(PCI_DEVICE(sq->ctrl), sq->db_addr, &v, sizeof(v));
6952 
6953     sq->tail = le32_to_cpu(v);
6954 
6955     trace_pci_nvme_update_sq_tail(sq->sqid, sq->tail);
6956 }
6957 
6958 static void nvme_process_sq(void *opaque)
6959 {
6960     NvmeSQueue *sq = opaque;
6961     NvmeCtrl *n = sq->ctrl;
6962     NvmeCQueue *cq = n->cq[sq->cqid];
6963 
6964     uint16_t status;
6965     hwaddr addr;
6966     NvmeCmd cmd;
6967     NvmeRequest *req;
6968 
6969     if (n->dbbuf_enabled) {
6970         nvme_update_sq_tail(sq);
6971     }
6972 
6973     while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
6974         addr = sq->dma_addr + sq->head * n->sqe_size;
6975         if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
6976             trace_pci_nvme_err_addr_read(addr);
6977             trace_pci_nvme_err_cfs();
6978             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
6979             break;
6980         }
6981         nvme_inc_sq_head(sq);
6982 
6983         req = QTAILQ_FIRST(&sq->req_list);
6984         QTAILQ_REMOVE(&sq->req_list, req, entry);
6985         QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
6986         nvme_req_clear(req);
6987         req->cqe.cid = cmd.cid;
6988         memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
6989 
6990         status = sq->sqid ? nvme_io_cmd(n, req) :
6991             nvme_admin_cmd(n, req);
6992         if (status != NVME_NO_COMPLETE) {
6993             req->status = status;
6994             nvme_enqueue_req_completion(cq, req);
6995         }
6996 
6997         if (n->dbbuf_enabled) {
6998             nvme_update_sq_eventidx(sq);
6999             nvme_update_sq_tail(sq);
7000         }
7001     }
7002 }
7003 
7004 static void nvme_update_msixcap_ts(PCIDevice *pci_dev, uint32_t table_size)
7005 {
7006     uint8_t *config;
7007 
7008     if (!msix_present(pci_dev)) {
7009         return;
7010     }
7011 
7012     assert(table_size > 0 && table_size <= pci_dev->msix_entries_nr);
7013 
7014     config = pci_dev->config + pci_dev->msix_cap;
7015     pci_set_word_by_mask(config + PCI_MSIX_FLAGS, PCI_MSIX_FLAGS_QSIZE,
7016                          table_size - 1);
7017 }
7018 
7019 static void nvme_activate_virt_res(NvmeCtrl *n)
7020 {
7021     PCIDevice *pci_dev = PCI_DEVICE(n);
7022     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7023     NvmeSecCtrlEntry *sctrl;
7024 
7025     /* -1 to account for the admin queue */
7026     if (pci_is_vf(pci_dev)) {
7027         sctrl = nvme_sctrl(n);
7028         cap->vqprt = sctrl->nvq;
7029         cap->viprt = sctrl->nvi;
7030         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7031         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7032     } else {
7033         cap->vqrfap = n->next_pri_ctrl_cap.vqrfap;
7034         cap->virfap = n->next_pri_ctrl_cap.virfap;
7035         n->conf_ioqpairs = le16_to_cpu(cap->vqprt) +
7036                            le16_to_cpu(cap->vqrfap) - 1;
7037         n->conf_msix_qsize = le16_to_cpu(cap->viprt) +
7038                              le16_to_cpu(cap->virfap);
7039     }
7040 }
7041 
7042 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst)
7043 {
7044     PCIDevice *pci_dev = PCI_DEVICE(n);
7045     NvmeSecCtrlEntry *sctrl;
7046     NvmeNamespace *ns;
7047     int i;
7048 
7049     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7050         ns = nvme_ns(n, i);
7051         if (!ns) {
7052             continue;
7053         }
7054 
7055         nvme_ns_drain(ns);
7056     }
7057 
7058     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7059         if (n->sq[i] != NULL) {
7060             nvme_free_sq(n->sq[i], n);
7061         }
7062     }
7063     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7064         if (n->cq[i] != NULL) {
7065             nvme_free_cq(n->cq[i], n);
7066         }
7067     }
7068 
7069     while (!QTAILQ_EMPTY(&n->aer_queue)) {
7070         NvmeAsyncEvent *event = QTAILQ_FIRST(&n->aer_queue);
7071         QTAILQ_REMOVE(&n->aer_queue, event, entry);
7072         g_free(event);
7073     }
7074 
7075     if (n->params.sriov_max_vfs) {
7076         if (!pci_is_vf(pci_dev)) {
7077             for (i = 0; i < n->sec_ctrl_list.numcntl; i++) {
7078                 sctrl = &n->sec_ctrl_list.sec[i];
7079                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
7080             }
7081 
7082             if (rst != NVME_RESET_CONTROLLER) {
7083                 pcie_sriov_pf_disable_vfs(pci_dev);
7084             }
7085         }
7086 
7087         if (rst != NVME_RESET_CONTROLLER) {
7088             nvme_activate_virt_res(n);
7089         }
7090     }
7091 
7092     n->aer_queued = 0;
7093     n->aer_mask = 0;
7094     n->outstanding_aers = 0;
7095     n->qs_created = false;
7096 
7097     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
7098 
7099     if (pci_is_vf(pci_dev)) {
7100         sctrl = nvme_sctrl(n);
7101 
7102         stl_le_p(&n->bar.csts, sctrl->scs ? 0 : NVME_CSTS_FAILED);
7103     } else {
7104         stl_le_p(&n->bar.csts, 0);
7105     }
7106 
7107     stl_le_p(&n->bar.intms, 0);
7108     stl_le_p(&n->bar.intmc, 0);
7109     stl_le_p(&n->bar.cc, 0);
7110 
7111     n->dbbuf_dbs = 0;
7112     n->dbbuf_eis = 0;
7113     n->dbbuf_enabled = false;
7114 }
7115 
7116 static void nvme_ctrl_shutdown(NvmeCtrl *n)
7117 {
7118     NvmeNamespace *ns;
7119     int i;
7120 
7121     if (n->pmr.dev) {
7122         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7123     }
7124 
7125     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7126         ns = nvme_ns(n, i);
7127         if (!ns) {
7128             continue;
7129         }
7130 
7131         nvme_ns_shutdown(ns);
7132     }
7133 }
7134 
7135 static void nvme_select_iocs(NvmeCtrl *n)
7136 {
7137     NvmeNamespace *ns;
7138     int i;
7139 
7140     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7141         ns = nvme_ns(n, i);
7142         if (!ns) {
7143             continue;
7144         }
7145 
7146         nvme_select_iocs_ns(n, ns);
7147     }
7148 }
7149 
7150 static int nvme_start_ctrl(NvmeCtrl *n)
7151 {
7152     uint64_t cap = ldq_le_p(&n->bar.cap);
7153     uint32_t cc = ldl_le_p(&n->bar.cc);
7154     uint32_t aqa = ldl_le_p(&n->bar.aqa);
7155     uint64_t asq = ldq_le_p(&n->bar.asq);
7156     uint64_t acq = ldq_le_p(&n->bar.acq);
7157     uint32_t page_bits = NVME_CC_MPS(cc) + 12;
7158     uint32_t page_size = 1 << page_bits;
7159     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
7160 
7161     if (pci_is_vf(PCI_DEVICE(n)) && !sctrl->scs) {
7162         trace_pci_nvme_err_startfail_virt_state(le16_to_cpu(sctrl->nvi),
7163                                                 le16_to_cpu(sctrl->nvq));
7164         return -1;
7165     }
7166     if (unlikely(n->cq[0])) {
7167         trace_pci_nvme_err_startfail_cq();
7168         return -1;
7169     }
7170     if (unlikely(n->sq[0])) {
7171         trace_pci_nvme_err_startfail_sq();
7172         return -1;
7173     }
7174     if (unlikely(asq & (page_size - 1))) {
7175         trace_pci_nvme_err_startfail_asq_misaligned(asq);
7176         return -1;
7177     }
7178     if (unlikely(acq & (page_size - 1))) {
7179         trace_pci_nvme_err_startfail_acq_misaligned(acq);
7180         return -1;
7181     }
7182     if (unlikely(!(NVME_CAP_CSS(cap) & (1 << NVME_CC_CSS(cc))))) {
7183         trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc));
7184         return -1;
7185     }
7186     if (unlikely(NVME_CC_MPS(cc) < NVME_CAP_MPSMIN(cap))) {
7187         trace_pci_nvme_err_startfail_page_too_small(
7188                     NVME_CC_MPS(cc),
7189                     NVME_CAP_MPSMIN(cap));
7190         return -1;
7191     }
7192     if (unlikely(NVME_CC_MPS(cc) >
7193                  NVME_CAP_MPSMAX(cap))) {
7194         trace_pci_nvme_err_startfail_page_too_large(
7195                     NVME_CC_MPS(cc),
7196                     NVME_CAP_MPSMAX(cap));
7197         return -1;
7198     }
7199     if (unlikely(NVME_CC_IOCQES(cc) <
7200                  NVME_CTRL_CQES_MIN(n->id_ctrl.cqes))) {
7201         trace_pci_nvme_err_startfail_cqent_too_small(
7202                     NVME_CC_IOCQES(cc),
7203                     NVME_CTRL_CQES_MIN(cap));
7204         return -1;
7205     }
7206     if (unlikely(NVME_CC_IOCQES(cc) >
7207                  NVME_CTRL_CQES_MAX(n->id_ctrl.cqes))) {
7208         trace_pci_nvme_err_startfail_cqent_too_large(
7209                     NVME_CC_IOCQES(cc),
7210                     NVME_CTRL_CQES_MAX(cap));
7211         return -1;
7212     }
7213     if (unlikely(NVME_CC_IOSQES(cc) <
7214                  NVME_CTRL_SQES_MIN(n->id_ctrl.sqes))) {
7215         trace_pci_nvme_err_startfail_sqent_too_small(
7216                     NVME_CC_IOSQES(cc),
7217                     NVME_CTRL_SQES_MIN(cap));
7218         return -1;
7219     }
7220     if (unlikely(NVME_CC_IOSQES(cc) >
7221                  NVME_CTRL_SQES_MAX(n->id_ctrl.sqes))) {
7222         trace_pci_nvme_err_startfail_sqent_too_large(
7223                     NVME_CC_IOSQES(cc),
7224                     NVME_CTRL_SQES_MAX(cap));
7225         return -1;
7226     }
7227     if (unlikely(!NVME_AQA_ASQS(aqa))) {
7228         trace_pci_nvme_err_startfail_asqent_sz_zero();
7229         return -1;
7230     }
7231     if (unlikely(!NVME_AQA_ACQS(aqa))) {
7232         trace_pci_nvme_err_startfail_acqent_sz_zero();
7233         return -1;
7234     }
7235 
7236     n->page_bits = page_bits;
7237     n->page_size = page_size;
7238     n->max_prp_ents = n->page_size / sizeof(uint64_t);
7239     n->cqe_size = 1 << NVME_CC_IOCQES(cc);
7240     n->sqe_size = 1 << NVME_CC_IOSQES(cc);
7241     nvme_init_cq(&n->admin_cq, n, acq, 0, 0, NVME_AQA_ACQS(aqa) + 1, 1);
7242     nvme_init_sq(&n->admin_sq, n, asq, 0, 0, NVME_AQA_ASQS(aqa) + 1);
7243 
7244     nvme_set_timestamp(n, 0ULL);
7245 
7246     nvme_select_iocs(n);
7247 
7248     return 0;
7249 }
7250 
7251 static void nvme_cmb_enable_regs(NvmeCtrl *n)
7252 {
7253     uint32_t cmbloc = ldl_le_p(&n->bar.cmbloc);
7254     uint32_t cmbsz = ldl_le_p(&n->bar.cmbsz);
7255 
7256     NVME_CMBLOC_SET_CDPCILS(cmbloc, 1);
7257     NVME_CMBLOC_SET_CDPMLS(cmbloc, 1);
7258     NVME_CMBLOC_SET_BIR(cmbloc, NVME_CMB_BIR);
7259     stl_le_p(&n->bar.cmbloc, cmbloc);
7260 
7261     NVME_CMBSZ_SET_SQS(cmbsz, 1);
7262     NVME_CMBSZ_SET_CQS(cmbsz, 0);
7263     NVME_CMBSZ_SET_LISTS(cmbsz, 1);
7264     NVME_CMBSZ_SET_RDS(cmbsz, 1);
7265     NVME_CMBSZ_SET_WDS(cmbsz, 1);
7266     NVME_CMBSZ_SET_SZU(cmbsz, 2); /* MBs */
7267     NVME_CMBSZ_SET_SZ(cmbsz, n->params.cmb_size_mb);
7268     stl_le_p(&n->bar.cmbsz, cmbsz);
7269 }
7270 
7271 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
7272                            unsigned size)
7273 {
7274     PCIDevice *pci = PCI_DEVICE(n);
7275     uint64_t cap = ldq_le_p(&n->bar.cap);
7276     uint32_t cc = ldl_le_p(&n->bar.cc);
7277     uint32_t intms = ldl_le_p(&n->bar.intms);
7278     uint32_t csts = ldl_le_p(&n->bar.csts);
7279     uint32_t pmrsts = ldl_le_p(&n->bar.pmrsts);
7280 
7281     if (unlikely(offset & (sizeof(uint32_t) - 1))) {
7282         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
7283                        "MMIO write not 32-bit aligned,"
7284                        " offset=0x%"PRIx64"", offset);
7285         /* should be ignored, fall through for now */
7286     }
7287 
7288     if (unlikely(size < sizeof(uint32_t))) {
7289         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
7290                        "MMIO write smaller than 32-bits,"
7291                        " offset=0x%"PRIx64", size=%u",
7292                        offset, size);
7293         /* should be ignored, fall through for now */
7294     }
7295 
7296     switch (offset) {
7297     case NVME_REG_INTMS:
7298         if (unlikely(msix_enabled(pci))) {
7299             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7300                            "undefined access to interrupt mask set"
7301                            " when MSI-X is enabled");
7302             /* should be ignored, fall through for now */
7303         }
7304         intms |= data;
7305         stl_le_p(&n->bar.intms, intms);
7306         n->bar.intmc = n->bar.intms;
7307         trace_pci_nvme_mmio_intm_set(data & 0xffffffff, intms);
7308         nvme_irq_check(n);
7309         break;
7310     case NVME_REG_INTMC:
7311         if (unlikely(msix_enabled(pci))) {
7312             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7313                            "undefined access to interrupt mask clr"
7314                            " when MSI-X is enabled");
7315             /* should be ignored, fall through for now */
7316         }
7317         intms &= ~data;
7318         stl_le_p(&n->bar.intms, intms);
7319         n->bar.intmc = n->bar.intms;
7320         trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, intms);
7321         nvme_irq_check(n);
7322         break;
7323     case NVME_REG_CC:
7324         stl_le_p(&n->bar.cc, data);
7325 
7326         trace_pci_nvme_mmio_cfg(data & 0xffffffff);
7327 
7328         if (NVME_CC_SHN(data) && !(NVME_CC_SHN(cc))) {
7329             trace_pci_nvme_mmio_shutdown_set();
7330             nvme_ctrl_shutdown(n);
7331             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7332             csts |= NVME_CSTS_SHST_COMPLETE;
7333         } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(cc)) {
7334             trace_pci_nvme_mmio_shutdown_cleared();
7335             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7336         }
7337 
7338         if (NVME_CC_EN(data) && !NVME_CC_EN(cc)) {
7339             if (unlikely(nvme_start_ctrl(n))) {
7340                 trace_pci_nvme_err_startfail();
7341                 csts = NVME_CSTS_FAILED;
7342             } else {
7343                 trace_pci_nvme_mmio_start_success();
7344                 csts = NVME_CSTS_READY;
7345             }
7346         } else if (!NVME_CC_EN(data) && NVME_CC_EN(cc)) {
7347             trace_pci_nvme_mmio_stopped();
7348             nvme_ctrl_reset(n, NVME_RESET_CONTROLLER);
7349 
7350             break;
7351         }
7352 
7353         stl_le_p(&n->bar.csts, csts);
7354 
7355         break;
7356     case NVME_REG_CSTS:
7357         if (data & (1 << 4)) {
7358             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
7359                            "attempted to W1C CSTS.NSSRO"
7360                            " but CAP.NSSRS is zero (not supported)");
7361         } else if (data != 0) {
7362             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
7363                            "attempted to set a read only bit"
7364                            " of controller status");
7365         }
7366         break;
7367     case NVME_REG_NSSR:
7368         if (data == 0x4e564d65) {
7369             trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
7370         } else {
7371             /* The spec says that writes of other values have no effect */
7372             return;
7373         }
7374         break;
7375     case NVME_REG_AQA:
7376         stl_le_p(&n->bar.aqa, data);
7377         trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
7378         break;
7379     case NVME_REG_ASQ:
7380         stn_le_p(&n->bar.asq, size, data);
7381         trace_pci_nvme_mmio_asqaddr(data);
7382         break;
7383     case NVME_REG_ASQ + 4:
7384         stl_le_p((uint8_t *)&n->bar.asq + 4, data);
7385         trace_pci_nvme_mmio_asqaddr_hi(data, ldq_le_p(&n->bar.asq));
7386         break;
7387     case NVME_REG_ACQ:
7388         trace_pci_nvme_mmio_acqaddr(data);
7389         stn_le_p(&n->bar.acq, size, data);
7390         break;
7391     case NVME_REG_ACQ + 4:
7392         stl_le_p((uint8_t *)&n->bar.acq + 4, data);
7393         trace_pci_nvme_mmio_acqaddr_hi(data, ldq_le_p(&n->bar.acq));
7394         break;
7395     case NVME_REG_CMBLOC:
7396         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
7397                        "invalid write to reserved CMBLOC"
7398                        " when CMBSZ is zero, ignored");
7399         return;
7400     case NVME_REG_CMBSZ:
7401         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
7402                        "invalid write to read only CMBSZ, ignored");
7403         return;
7404     case NVME_REG_CMBMSC:
7405         if (!NVME_CAP_CMBS(cap)) {
7406             return;
7407         }
7408 
7409         stn_le_p(&n->bar.cmbmsc, size, data);
7410         n->cmb.cmse = false;
7411 
7412         if (NVME_CMBMSC_CRE(data)) {
7413             nvme_cmb_enable_regs(n);
7414 
7415             if (NVME_CMBMSC_CMSE(data)) {
7416                 uint64_t cmbmsc = ldq_le_p(&n->bar.cmbmsc);
7417                 hwaddr cba = NVME_CMBMSC_CBA(cmbmsc) << CMBMSC_CBA_SHIFT;
7418                 if (cba + int128_get64(n->cmb.mem.size) < cba) {
7419                     uint32_t cmbsts = ldl_le_p(&n->bar.cmbsts);
7420                     NVME_CMBSTS_SET_CBAI(cmbsts, 1);
7421                     stl_le_p(&n->bar.cmbsts, cmbsts);
7422                     return;
7423                 }
7424 
7425                 n->cmb.cba = cba;
7426                 n->cmb.cmse = true;
7427             }
7428         } else {
7429             n->bar.cmbsz = 0;
7430             n->bar.cmbloc = 0;
7431         }
7432 
7433         return;
7434     case NVME_REG_CMBMSC + 4:
7435         stl_le_p((uint8_t *)&n->bar.cmbmsc + 4, data);
7436         return;
7437 
7438     case NVME_REG_PMRCAP:
7439         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
7440                        "invalid write to PMRCAP register, ignored");
7441         return;
7442     case NVME_REG_PMRCTL:
7443         if (!NVME_CAP_PMRS(cap)) {
7444             return;
7445         }
7446 
7447         stl_le_p(&n->bar.pmrctl, data);
7448         if (NVME_PMRCTL_EN(data)) {
7449             memory_region_set_enabled(&n->pmr.dev->mr, true);
7450             pmrsts = 0;
7451         } else {
7452             memory_region_set_enabled(&n->pmr.dev->mr, false);
7453             NVME_PMRSTS_SET_NRDY(pmrsts, 1);
7454             n->pmr.cmse = false;
7455         }
7456         stl_le_p(&n->bar.pmrsts, pmrsts);
7457         return;
7458     case NVME_REG_PMRSTS:
7459         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
7460                        "invalid write to PMRSTS register, ignored");
7461         return;
7462     case NVME_REG_PMREBS:
7463         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
7464                        "invalid write to PMREBS register, ignored");
7465         return;
7466     case NVME_REG_PMRSWTP:
7467         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
7468                        "invalid write to PMRSWTP register, ignored");
7469         return;
7470     case NVME_REG_PMRMSCL:
7471         if (!NVME_CAP_PMRS(cap)) {
7472             return;
7473         }
7474 
7475         stl_le_p(&n->bar.pmrmscl, data);
7476         n->pmr.cmse = false;
7477 
7478         if (NVME_PMRMSCL_CMSE(data)) {
7479             uint64_t pmrmscu = ldl_le_p(&n->bar.pmrmscu);
7480             hwaddr cba = pmrmscu << 32 |
7481                 (NVME_PMRMSCL_CBA(data) << PMRMSCL_CBA_SHIFT);
7482             if (cba + int128_get64(n->pmr.dev->mr.size) < cba) {
7483                 NVME_PMRSTS_SET_CBAI(pmrsts, 1);
7484                 stl_le_p(&n->bar.pmrsts, pmrsts);
7485                 return;
7486             }
7487 
7488             n->pmr.cmse = true;
7489             n->pmr.cba = cba;
7490         }
7491 
7492         return;
7493     case NVME_REG_PMRMSCU:
7494         if (!NVME_CAP_PMRS(cap)) {
7495             return;
7496         }
7497 
7498         stl_le_p(&n->bar.pmrmscu, data);
7499         return;
7500     default:
7501         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
7502                        "invalid MMIO write,"
7503                        " offset=0x%"PRIx64", data=%"PRIx64"",
7504                        offset, data);
7505         break;
7506     }
7507 }
7508 
7509 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
7510 {
7511     NvmeCtrl *n = (NvmeCtrl *)opaque;
7512     uint8_t *ptr = (uint8_t *)&n->bar;
7513 
7514     trace_pci_nvme_mmio_read(addr, size);
7515 
7516     if (unlikely(addr & (sizeof(uint32_t) - 1))) {
7517         NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
7518                        "MMIO read not 32-bit aligned,"
7519                        " offset=0x%"PRIx64"", addr);
7520         /* should RAZ, fall through for now */
7521     } else if (unlikely(size < sizeof(uint32_t))) {
7522         NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
7523                        "MMIO read smaller than 32-bits,"
7524                        " offset=0x%"PRIx64"", addr);
7525         /* should RAZ, fall through for now */
7526     }
7527 
7528     if (addr > sizeof(n->bar) - size) {
7529         NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
7530                        "MMIO read beyond last register,"
7531                        " offset=0x%"PRIx64", returning 0", addr);
7532 
7533         return 0;
7534     }
7535 
7536     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7537         addr != NVME_REG_CSTS) {
7538         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7539         return 0;
7540     }
7541 
7542     /*
7543      * When PMRWBM bit 1 is set then read from
7544      * from PMRSTS should ensure prior writes
7545      * made it to persistent media
7546      */
7547     if (addr == NVME_REG_PMRSTS &&
7548         (NVME_PMRCAP_PMRWBM(ldl_le_p(&n->bar.pmrcap)) & 0x02)) {
7549         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7550     }
7551 
7552     return ldn_le_p(ptr + addr, size);
7553 }
7554 
7555 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
7556 {
7557     PCIDevice *pci = PCI_DEVICE(n);
7558     uint32_t qid;
7559 
7560     if (unlikely(addr & ((1 << 2) - 1))) {
7561         NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
7562                        "doorbell write not 32-bit aligned,"
7563                        " offset=0x%"PRIx64", ignoring", addr);
7564         return;
7565     }
7566 
7567     if (((addr - 0x1000) >> 2) & 1) {
7568         /* Completion queue doorbell write */
7569 
7570         uint16_t new_head = val & 0xffff;
7571         int start_sqs;
7572         NvmeCQueue *cq;
7573 
7574         qid = (addr - (0x1000 + (1 << 2))) >> 3;
7575         if (unlikely(nvme_check_cqid(n, qid))) {
7576             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
7577                            "completion queue doorbell write"
7578                            " for nonexistent queue,"
7579                            " sqid=%"PRIu32", ignoring", qid);
7580 
7581             /*
7582              * NVM Express v1.3d, Section 4.1 state: "If host software writes
7583              * an invalid value to the Submission Queue Tail Doorbell or
7584              * Completion Queue Head Doorbell regiter and an Asynchronous Event
7585              * Request command is outstanding, then an asynchronous event is
7586              * posted to the Admin Completion Queue with a status code of
7587              * Invalid Doorbell Write Value."
7588              *
7589              * Also note that the spec includes the "Invalid Doorbell Register"
7590              * status code, but nowhere does it specify when to use it.
7591              * However, it seems reasonable to use it here in a similar
7592              * fashion.
7593              */
7594             if (n->outstanding_aers) {
7595                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7596                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7597                                    NVME_LOG_ERROR_INFO);
7598             }
7599 
7600             return;
7601         }
7602 
7603         cq = n->cq[qid];
7604         if (unlikely(new_head >= cq->size)) {
7605             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
7606                            "completion queue doorbell write value"
7607                            " beyond queue size, sqid=%"PRIu32","
7608                            " new_head=%"PRIu16", ignoring",
7609                            qid, new_head);
7610 
7611             if (n->outstanding_aers) {
7612                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7613                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7614                                    NVME_LOG_ERROR_INFO);
7615             }
7616 
7617             return;
7618         }
7619 
7620         trace_pci_nvme_mmio_doorbell_cq(cq->cqid, new_head);
7621 
7622         start_sqs = nvme_cq_full(cq) ? 1 : 0;
7623         cq->head = new_head;
7624         if (!qid && n->dbbuf_enabled) {
7625             pci_dma_write(pci, cq->db_addr, &cq->head, sizeof(cq->head));
7626         }
7627         if (start_sqs) {
7628             NvmeSQueue *sq;
7629             QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
7630                 qemu_bh_schedule(sq->bh);
7631             }
7632             qemu_bh_schedule(cq->bh);
7633         }
7634 
7635         if (cq->tail == cq->head) {
7636             if (cq->irq_enabled) {
7637                 n->cq_pending--;
7638             }
7639 
7640             nvme_irq_deassert(n, cq);
7641         }
7642     } else {
7643         /* Submission queue doorbell write */
7644 
7645         uint16_t new_tail = val & 0xffff;
7646         NvmeSQueue *sq;
7647 
7648         qid = (addr - 0x1000) >> 3;
7649         if (unlikely(nvme_check_sqid(n, qid))) {
7650             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
7651                            "submission queue doorbell write"
7652                            " for nonexistent queue,"
7653                            " sqid=%"PRIu32", ignoring", qid);
7654 
7655             if (n->outstanding_aers) {
7656                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7657                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7658                                    NVME_LOG_ERROR_INFO);
7659             }
7660 
7661             return;
7662         }
7663 
7664         sq = n->sq[qid];
7665         if (unlikely(new_tail >= sq->size)) {
7666             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
7667                            "submission queue doorbell write value"
7668                            " beyond queue size, sqid=%"PRIu32","
7669                            " new_tail=%"PRIu16", ignoring",
7670                            qid, new_tail);
7671 
7672             if (n->outstanding_aers) {
7673                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7674                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7675                                    NVME_LOG_ERROR_INFO);
7676             }
7677 
7678             return;
7679         }
7680 
7681         trace_pci_nvme_mmio_doorbell_sq(sq->sqid, new_tail);
7682 
7683         sq->tail = new_tail;
7684         if (!qid && n->dbbuf_enabled) {
7685             /*
7686              * The spec states "the host shall also update the controller's
7687              * corresponding doorbell property to match the value of that entry
7688              * in the Shadow Doorbell buffer."
7689              *
7690              * Since this context is currently a VM trap, we can safely enforce
7691              * the requirement from the device side in case the host is
7692              * misbehaving.
7693              *
7694              * Note, we shouldn't have to do this, but various drivers
7695              * including ones that run on Linux, are not updating Admin Queues,
7696              * so we can't trust reading it for an appropriate sq tail.
7697              */
7698             pci_dma_write(pci, sq->db_addr, &sq->tail, sizeof(sq->tail));
7699         }
7700 
7701         qemu_bh_schedule(sq->bh);
7702     }
7703 }
7704 
7705 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
7706                             unsigned size)
7707 {
7708     NvmeCtrl *n = (NvmeCtrl *)opaque;
7709 
7710     trace_pci_nvme_mmio_write(addr, data, size);
7711 
7712     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7713         addr != NVME_REG_CSTS) {
7714         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7715         return;
7716     }
7717 
7718     if (addr < sizeof(n->bar)) {
7719         nvme_write_bar(n, addr, data, size);
7720     } else {
7721         nvme_process_db(n, addr, data);
7722     }
7723 }
7724 
7725 static const MemoryRegionOps nvme_mmio_ops = {
7726     .read = nvme_mmio_read,
7727     .write = nvme_mmio_write,
7728     .endianness = DEVICE_LITTLE_ENDIAN,
7729     .impl = {
7730         .min_access_size = 2,
7731         .max_access_size = 8,
7732     },
7733 };
7734 
7735 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
7736                            unsigned size)
7737 {
7738     NvmeCtrl *n = (NvmeCtrl *)opaque;
7739     stn_le_p(&n->cmb.buf[addr], size, data);
7740 }
7741 
7742 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
7743 {
7744     NvmeCtrl *n = (NvmeCtrl *)opaque;
7745     return ldn_le_p(&n->cmb.buf[addr], size);
7746 }
7747 
7748 static const MemoryRegionOps nvme_cmb_ops = {
7749     .read = nvme_cmb_read,
7750     .write = nvme_cmb_write,
7751     .endianness = DEVICE_LITTLE_ENDIAN,
7752     .impl = {
7753         .min_access_size = 1,
7754         .max_access_size = 8,
7755     },
7756 };
7757 
7758 static bool nvme_check_params(NvmeCtrl *n, Error **errp)
7759 {
7760     NvmeParams *params = &n->params;
7761 
7762     if (params->num_queues) {
7763         warn_report("num_queues is deprecated; please use max_ioqpairs "
7764                     "instead");
7765 
7766         params->max_ioqpairs = params->num_queues - 1;
7767     }
7768 
7769     if (n->namespace.blkconf.blk && n->subsys) {
7770         error_setg(errp, "subsystem support is unavailable with legacy "
7771                    "namespace ('drive' property)");
7772         return false;
7773     }
7774 
7775     if (params->max_ioqpairs < 1 ||
7776         params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
7777         error_setg(errp, "max_ioqpairs must be between 1 and %d",
7778                    NVME_MAX_IOQPAIRS);
7779         return false;
7780     }
7781 
7782     if (params->msix_qsize < 1 ||
7783         params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
7784         error_setg(errp, "msix_qsize must be between 1 and %d",
7785                    PCI_MSIX_FLAGS_QSIZE + 1);
7786         return false;
7787     }
7788 
7789     if (!params->serial) {
7790         error_setg(errp, "serial property not set");
7791         return false;
7792     }
7793 
7794     if (n->pmr.dev) {
7795         if (host_memory_backend_is_mapped(n->pmr.dev)) {
7796             error_setg(errp, "can't use already busy memdev: %s",
7797                        object_get_canonical_path_component(OBJECT(n->pmr.dev)));
7798             return false;
7799         }
7800 
7801         if (!is_power_of_2(n->pmr.dev->size)) {
7802             error_setg(errp, "pmr backend size needs to be power of 2 in size");
7803             return false;
7804         }
7805 
7806         host_memory_backend_set_mapped(n->pmr.dev, true);
7807     }
7808 
7809     if (n->params.zasl > n->params.mdts) {
7810         error_setg(errp, "zoned.zasl (Zone Append Size Limit) must be less "
7811                    "than or equal to mdts (Maximum Data Transfer Size)");
7812         return false;
7813     }
7814 
7815     if (!n->params.vsl) {
7816         error_setg(errp, "vsl must be non-zero");
7817         return false;
7818     }
7819 
7820     if (params->sriov_max_vfs) {
7821         if (!n->subsys) {
7822             error_setg(errp, "subsystem is required for the use of SR-IOV");
7823             return false;
7824         }
7825 
7826         if (params->sriov_max_vfs > NVME_MAX_VFS) {
7827             error_setg(errp, "sriov_max_vfs must be between 0 and %d",
7828                        NVME_MAX_VFS);
7829             return false;
7830         }
7831 
7832         if (params->cmb_size_mb) {
7833             error_setg(errp, "CMB is not supported with SR-IOV");
7834             return false;
7835         }
7836 
7837         if (n->pmr.dev) {
7838             error_setg(errp, "PMR is not supported with SR-IOV");
7839             return false;
7840         }
7841 
7842         if (!params->sriov_vq_flexible || !params->sriov_vi_flexible) {
7843             error_setg(errp, "both sriov_vq_flexible and sriov_vi_flexible"
7844                        " must be set for the use of SR-IOV");
7845             return false;
7846         }
7847 
7848         if (params->sriov_vq_flexible < params->sriov_max_vfs * 2) {
7849             error_setg(errp, "sriov_vq_flexible must be greater than or equal"
7850                        " to %d (sriov_max_vfs * 2)", params->sriov_max_vfs * 2);
7851             return false;
7852         }
7853 
7854         if (params->max_ioqpairs < params->sriov_vq_flexible + 2) {
7855             error_setg(errp, "(max_ioqpairs - sriov_vq_flexible) must be"
7856                        " greater than or equal to 2");
7857             return false;
7858         }
7859 
7860         if (params->sriov_vi_flexible < params->sriov_max_vfs) {
7861             error_setg(errp, "sriov_vi_flexible must be greater than or equal"
7862                        " to %d (sriov_max_vfs)", params->sriov_max_vfs);
7863             return false;
7864         }
7865 
7866         if (params->msix_qsize < params->sriov_vi_flexible + 1) {
7867             error_setg(errp, "(msix_qsize - sriov_vi_flexible) must be"
7868                        " greater than or equal to 1");
7869             return false;
7870         }
7871 
7872         if (params->sriov_max_vi_per_vf &&
7873             (params->sriov_max_vi_per_vf - 1) % NVME_VF_RES_GRANULARITY) {
7874             error_setg(errp, "sriov_max_vi_per_vf must meet:"
7875                        " (sriov_max_vi_per_vf - 1) %% %d == 0 and"
7876                        " sriov_max_vi_per_vf >= 1", NVME_VF_RES_GRANULARITY);
7877             return false;
7878         }
7879 
7880         if (params->sriov_max_vq_per_vf &&
7881             (params->sriov_max_vq_per_vf < 2 ||
7882              (params->sriov_max_vq_per_vf - 1) % NVME_VF_RES_GRANULARITY)) {
7883             error_setg(errp, "sriov_max_vq_per_vf must meet:"
7884                        " (sriov_max_vq_per_vf - 1) %% %d == 0 and"
7885                        " sriov_max_vq_per_vf >= 2", NVME_VF_RES_GRANULARITY);
7886             return false;
7887         }
7888     }
7889 
7890     return true;
7891 }
7892 
7893 static void nvme_init_state(NvmeCtrl *n)
7894 {
7895     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7896     NvmeSecCtrlList *list = &n->sec_ctrl_list;
7897     NvmeSecCtrlEntry *sctrl;
7898     PCIDevice *pci = PCI_DEVICE(n);
7899     uint8_t max_vfs;
7900     int i;
7901 
7902     if (pci_is_vf(pci)) {
7903         sctrl = nvme_sctrl(n);
7904         max_vfs = 0;
7905         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7906         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7907     } else {
7908         max_vfs = n->params.sriov_max_vfs;
7909         n->conf_ioqpairs = n->params.max_ioqpairs;
7910         n->conf_msix_qsize = n->params.msix_qsize;
7911     }
7912 
7913     n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
7914     n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
7915     n->temperature = NVME_TEMPERATURE;
7916     n->features.temp_thresh_hi = NVME_TEMPERATURE_WARNING;
7917     n->starttime_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
7918     n->aer_reqs = g_new0(NvmeRequest *, n->params.aerl + 1);
7919     QTAILQ_INIT(&n->aer_queue);
7920 
7921     list->numcntl = cpu_to_le16(max_vfs);
7922     for (i = 0; i < max_vfs; i++) {
7923         sctrl = &list->sec[i];
7924         sctrl->pcid = cpu_to_le16(n->cntlid);
7925         sctrl->vfn = cpu_to_le16(i + 1);
7926     }
7927 
7928     cap->cntlid = cpu_to_le16(n->cntlid);
7929     cap->crt = NVME_CRT_VQ | NVME_CRT_VI;
7930 
7931     if (pci_is_vf(pci)) {
7932         cap->vqprt = cpu_to_le16(1 + n->conf_ioqpairs);
7933     } else {
7934         cap->vqprt = cpu_to_le16(1 + n->params.max_ioqpairs -
7935                                  n->params.sriov_vq_flexible);
7936         cap->vqfrt = cpu_to_le32(n->params.sriov_vq_flexible);
7937         cap->vqrfap = cap->vqfrt;
7938         cap->vqgran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
7939         cap->vqfrsm = n->params.sriov_max_vq_per_vf ?
7940                         cpu_to_le16(n->params.sriov_max_vq_per_vf) :
7941                         cap->vqfrt / MAX(max_vfs, 1);
7942     }
7943 
7944     if (pci_is_vf(pci)) {
7945         cap->viprt = cpu_to_le16(n->conf_msix_qsize);
7946     } else {
7947         cap->viprt = cpu_to_le16(n->params.msix_qsize -
7948                                  n->params.sriov_vi_flexible);
7949         cap->vifrt = cpu_to_le32(n->params.sriov_vi_flexible);
7950         cap->virfap = cap->vifrt;
7951         cap->vigran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
7952         cap->vifrsm = n->params.sriov_max_vi_per_vf ?
7953                         cpu_to_le16(n->params.sriov_max_vi_per_vf) :
7954                         cap->vifrt / MAX(max_vfs, 1);
7955     }
7956 }
7957 
7958 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
7959 {
7960     uint64_t cmb_size = n->params.cmb_size_mb * MiB;
7961     uint64_t cap = ldq_le_p(&n->bar.cap);
7962 
7963     n->cmb.buf = g_malloc0(cmb_size);
7964     memory_region_init_io(&n->cmb.mem, OBJECT(n), &nvme_cmb_ops, n,
7965                           "nvme-cmb", cmb_size);
7966     pci_register_bar(pci_dev, NVME_CMB_BIR,
7967                      PCI_BASE_ADDRESS_SPACE_MEMORY |
7968                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
7969                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->cmb.mem);
7970 
7971     NVME_CAP_SET_CMBS(cap, 1);
7972     stq_le_p(&n->bar.cap, cap);
7973 
7974     if (n->params.legacy_cmb) {
7975         nvme_cmb_enable_regs(n);
7976         n->cmb.cmse = true;
7977     }
7978 }
7979 
7980 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
7981 {
7982     uint32_t pmrcap = ldl_le_p(&n->bar.pmrcap);
7983 
7984     NVME_PMRCAP_SET_RDS(pmrcap, 1);
7985     NVME_PMRCAP_SET_WDS(pmrcap, 1);
7986     NVME_PMRCAP_SET_BIR(pmrcap, NVME_PMR_BIR);
7987     /* Turn on bit 1 support */
7988     NVME_PMRCAP_SET_PMRWBM(pmrcap, 0x02);
7989     NVME_PMRCAP_SET_CMSS(pmrcap, 1);
7990     stl_le_p(&n->bar.pmrcap, pmrcap);
7991 
7992     pci_register_bar(pci_dev, NVME_PMR_BIR,
7993                      PCI_BASE_ADDRESS_SPACE_MEMORY |
7994                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
7995                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmr.dev->mr);
7996 
7997     memory_region_set_enabled(&n->pmr.dev->mr, false);
7998 }
7999 
8000 static uint64_t nvme_bar_size(unsigned total_queues, unsigned total_irqs,
8001                               unsigned *msix_table_offset,
8002                               unsigned *msix_pba_offset)
8003 {
8004     uint64_t bar_size, msix_table_size, msix_pba_size;
8005 
8006     bar_size = sizeof(NvmeBar) + 2 * total_queues * NVME_DB_SIZE;
8007     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8008 
8009     if (msix_table_offset) {
8010         *msix_table_offset = bar_size;
8011     }
8012 
8013     msix_table_size = PCI_MSIX_ENTRY_SIZE * total_irqs;
8014     bar_size += msix_table_size;
8015     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8016 
8017     if (msix_pba_offset) {
8018         *msix_pba_offset = bar_size;
8019     }
8020 
8021     msix_pba_size = QEMU_ALIGN_UP(total_irqs, 64) / 8;
8022     bar_size += msix_pba_size;
8023 
8024     bar_size = pow2ceil(bar_size);
8025     return bar_size;
8026 }
8027 
8028 static void nvme_init_sriov(NvmeCtrl *n, PCIDevice *pci_dev, uint16_t offset)
8029 {
8030     uint16_t vf_dev_id = n->params.use_intel_id ?
8031                          PCI_DEVICE_ID_INTEL_NVME : PCI_DEVICE_ID_REDHAT_NVME;
8032     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
8033     uint64_t bar_size = nvme_bar_size(le16_to_cpu(cap->vqfrsm),
8034                                       le16_to_cpu(cap->vifrsm),
8035                                       NULL, NULL);
8036 
8037     pcie_sriov_pf_init(pci_dev, offset, "nvme", vf_dev_id,
8038                        n->params.sriov_max_vfs, n->params.sriov_max_vfs,
8039                        NVME_VF_OFFSET, NVME_VF_STRIDE);
8040 
8041     pcie_sriov_pf_init_vf_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8042                               PCI_BASE_ADDRESS_MEM_TYPE_64, bar_size);
8043 }
8044 
8045 static int nvme_add_pm_capability(PCIDevice *pci_dev, uint8_t offset)
8046 {
8047     Error *err = NULL;
8048     int ret;
8049 
8050     ret = pci_add_capability(pci_dev, PCI_CAP_ID_PM, offset,
8051                              PCI_PM_SIZEOF, &err);
8052     if (err) {
8053         error_report_err(err);
8054         return ret;
8055     }
8056 
8057     pci_set_word(pci_dev->config + offset + PCI_PM_PMC,
8058                  PCI_PM_CAP_VER_1_2);
8059     pci_set_word(pci_dev->config + offset + PCI_PM_CTRL,
8060                  PCI_PM_CTRL_NO_SOFT_RESET);
8061     pci_set_word(pci_dev->wmask + offset + PCI_PM_CTRL,
8062                  PCI_PM_CTRL_STATE_MASK);
8063 
8064     return 0;
8065 }
8066 
8067 static bool nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
8068 {
8069     ERRP_GUARD();
8070     uint8_t *pci_conf = pci_dev->config;
8071     uint64_t bar_size;
8072     unsigned msix_table_offset, msix_pba_offset;
8073     int ret;
8074 
8075     pci_conf[PCI_INTERRUPT_PIN] = 1;
8076     pci_config_set_prog_interface(pci_conf, 0x2);
8077 
8078     if (n->params.use_intel_id) {
8079         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_INTEL);
8080         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_INTEL_NVME);
8081     } else {
8082         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_REDHAT);
8083         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_REDHAT_NVME);
8084     }
8085 
8086     pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
8087     nvme_add_pm_capability(pci_dev, 0x60);
8088     pcie_endpoint_cap_init(pci_dev, 0x80);
8089     pcie_cap_flr_init(pci_dev);
8090     if (n->params.sriov_max_vfs) {
8091         pcie_ari_init(pci_dev, 0x100, 1);
8092     }
8093 
8094     /* add one to max_ioqpairs to account for the admin queue pair */
8095     bar_size = nvme_bar_size(n->params.max_ioqpairs + 1, n->params.msix_qsize,
8096                              &msix_table_offset, &msix_pba_offset);
8097 
8098     memory_region_init(&n->bar0, OBJECT(n), "nvme-bar0", bar_size);
8099     memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
8100                           msix_table_offset);
8101     memory_region_add_subregion(&n->bar0, 0, &n->iomem);
8102 
8103     if (pci_is_vf(pci_dev)) {
8104         pcie_sriov_vf_register_bar(pci_dev, 0, &n->bar0);
8105     } else {
8106         pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8107                          PCI_BASE_ADDRESS_MEM_TYPE_64, &n->bar0);
8108     }
8109     ret = msix_init(pci_dev, n->params.msix_qsize,
8110                     &n->bar0, 0, msix_table_offset,
8111                     &n->bar0, 0, msix_pba_offset, 0, errp);
8112     if (ret == -ENOTSUP) {
8113         /* report that msix is not supported, but do not error out */
8114         warn_report_err(*errp);
8115         *errp = NULL;
8116     } else if (ret < 0) {
8117         /* propagate error to caller */
8118         return false;
8119     }
8120 
8121     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
8122 
8123     if (n->params.cmb_size_mb) {
8124         nvme_init_cmb(n, pci_dev);
8125     }
8126 
8127     if (n->pmr.dev) {
8128         nvme_init_pmr(n, pci_dev);
8129     }
8130 
8131     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8132         nvme_init_sriov(n, pci_dev, 0x120);
8133     }
8134 
8135     return true;
8136 }
8137 
8138 static void nvme_init_subnqn(NvmeCtrl *n)
8139 {
8140     NvmeSubsystem *subsys = n->subsys;
8141     NvmeIdCtrl *id = &n->id_ctrl;
8142 
8143     if (!subsys) {
8144         snprintf((char *)id->subnqn, sizeof(id->subnqn),
8145                  "nqn.2019-08.org.qemu:%s", n->params.serial);
8146     } else {
8147         pstrcpy((char *)id->subnqn, sizeof(id->subnqn), (char*)subsys->subnqn);
8148     }
8149 }
8150 
8151 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
8152 {
8153     NvmeIdCtrl *id = &n->id_ctrl;
8154     uint8_t *pci_conf = pci_dev->config;
8155     uint64_t cap = ldq_le_p(&n->bar.cap);
8156     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
8157     uint32_t ctratt;
8158 
8159     id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
8160     id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
8161     strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
8162     strpadcpy((char *)id->fr, sizeof(id->fr), QEMU_VERSION, ' ');
8163     strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
8164 
8165     id->cntlid = cpu_to_le16(n->cntlid);
8166 
8167     id->oaes = cpu_to_le32(NVME_OAES_NS_ATTR);
8168     ctratt = NVME_CTRATT_ELBAS;
8169 
8170     id->rab = 6;
8171 
8172     if (n->params.use_intel_id) {
8173         id->ieee[0] = 0xb3;
8174         id->ieee[1] = 0x02;
8175         id->ieee[2] = 0x00;
8176     } else {
8177         id->ieee[0] = 0x00;
8178         id->ieee[1] = 0x54;
8179         id->ieee[2] = 0x52;
8180     }
8181 
8182     id->mdts = n->params.mdts;
8183     id->ver = cpu_to_le32(NVME_SPEC_VER);
8184     id->oacs =
8185         cpu_to_le16(NVME_OACS_NS_MGMT | NVME_OACS_FORMAT | NVME_OACS_DBBUF |
8186                     NVME_OACS_DIRECTIVES);
8187     id->cntrltype = 0x1;
8188 
8189     /*
8190      * Because the controller always completes the Abort command immediately,
8191      * there can never be more than one concurrently executing Abort command,
8192      * so this value is never used for anything. Note that there can easily be
8193      * many Abort commands in the queues, but they are not considered
8194      * "executing" until processed by nvme_abort.
8195      *
8196      * The specification recommends a value of 3 for Abort Command Limit (four
8197      * concurrently outstanding Abort commands), so lets use that though it is
8198      * inconsequential.
8199      */
8200     id->acl = 3;
8201     id->aerl = n->params.aerl;
8202     id->frmw = (NVME_NUM_FW_SLOTS << 1) | NVME_FRMW_SLOT1_RO;
8203     id->lpa = NVME_LPA_NS_SMART | NVME_LPA_CSE | NVME_LPA_EXTENDED;
8204 
8205     /* recommended default value (~70 C) */
8206     id->wctemp = cpu_to_le16(NVME_TEMPERATURE_WARNING);
8207     id->cctemp = cpu_to_le16(NVME_TEMPERATURE_CRITICAL);
8208 
8209     id->sqes = (0x6 << 4) | 0x6;
8210     id->cqes = (0x4 << 4) | 0x4;
8211     id->nn = cpu_to_le32(NVME_MAX_NAMESPACES);
8212     id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP |
8213                            NVME_ONCS_FEATURES | NVME_ONCS_DSM |
8214                            NVME_ONCS_COMPARE | NVME_ONCS_COPY);
8215 
8216     /*
8217      * NOTE: If this device ever supports a command set that does NOT use 0x0
8218      * as a Flush-equivalent operation, support for the broadcast NSID in Flush
8219      * should probably be removed.
8220      *
8221      * See comment in nvme_io_cmd.
8222      */
8223     id->vwc = NVME_VWC_NSID_BROADCAST_SUPPORT | NVME_VWC_PRESENT;
8224 
8225     id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0 | NVME_OCFS_COPY_FORMAT_1);
8226     id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN);
8227 
8228     nvme_init_subnqn(n);
8229 
8230     id->psd[0].mp = cpu_to_le16(0x9c4);
8231     id->psd[0].enlat = cpu_to_le32(0x10);
8232     id->psd[0].exlat = cpu_to_le32(0x4);
8233 
8234     if (n->subsys) {
8235         id->cmic |= NVME_CMIC_MULTI_CTRL;
8236         ctratt |= NVME_CTRATT_ENDGRPS;
8237 
8238         id->endgidmax = cpu_to_le16(0x1);
8239 
8240         if (n->subsys->endgrp.fdp.enabled) {
8241             ctratt |= NVME_CTRATT_FDPS;
8242         }
8243     }
8244 
8245     id->ctratt = cpu_to_le32(ctratt);
8246 
8247     NVME_CAP_SET_MQES(cap, 0x7ff);
8248     NVME_CAP_SET_CQR(cap, 1);
8249     NVME_CAP_SET_TO(cap, 0xf);
8250     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_NVM);
8251     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_CSI_SUPP);
8252     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_ADMIN_ONLY);
8253     NVME_CAP_SET_MPSMAX(cap, 4);
8254     NVME_CAP_SET_CMBS(cap, n->params.cmb_size_mb ? 1 : 0);
8255     NVME_CAP_SET_PMRS(cap, n->pmr.dev ? 1 : 0);
8256     stq_le_p(&n->bar.cap, cap);
8257 
8258     stl_le_p(&n->bar.vs, NVME_SPEC_VER);
8259     n->bar.intmc = n->bar.intms = 0;
8260 
8261     if (pci_is_vf(pci_dev) && !sctrl->scs) {
8262         stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
8263     }
8264 }
8265 
8266 static int nvme_init_subsys(NvmeCtrl *n, Error **errp)
8267 {
8268     int cntlid;
8269 
8270     if (!n->subsys) {
8271         return 0;
8272     }
8273 
8274     cntlid = nvme_subsys_register_ctrl(n, errp);
8275     if (cntlid < 0) {
8276         return -1;
8277     }
8278 
8279     n->cntlid = cntlid;
8280 
8281     return 0;
8282 }
8283 
8284 void nvme_attach_ns(NvmeCtrl *n, NvmeNamespace *ns)
8285 {
8286     uint32_t nsid = ns->params.nsid;
8287     assert(nsid && nsid <= NVME_MAX_NAMESPACES);
8288 
8289     n->namespaces[nsid] = ns;
8290     ns->attached++;
8291 
8292     n->dmrsl = MIN_NON_ZERO(n->dmrsl,
8293                             BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
8294 }
8295 
8296 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
8297 {
8298     NvmeCtrl *n = NVME(pci_dev);
8299     DeviceState *dev = DEVICE(pci_dev);
8300     NvmeNamespace *ns;
8301     NvmeCtrl *pn = NVME(pcie_sriov_get_pf(pci_dev));
8302 
8303     if (pci_is_vf(pci_dev)) {
8304         /*
8305          * VFs derive settings from the parent. PF's lifespan exceeds
8306          * that of VF's, so it's safe to share params.serial.
8307          */
8308         memcpy(&n->params, &pn->params, sizeof(NvmeParams));
8309         n->subsys = pn->subsys;
8310     }
8311 
8312     if (!nvme_check_params(n, errp)) {
8313         return;
8314     }
8315 
8316     qbus_init(&n->bus, sizeof(NvmeBus), TYPE_NVME_BUS, dev, dev->id);
8317 
8318     if (nvme_init_subsys(n, errp)) {
8319         return;
8320     }
8321     nvme_init_state(n);
8322     if (!nvme_init_pci(n, pci_dev, errp)) {
8323         return;
8324     }
8325     nvme_init_ctrl(n, pci_dev);
8326 
8327     /* setup a namespace if the controller drive property was given */
8328     if (n->namespace.blkconf.blk) {
8329         ns = &n->namespace;
8330         ns->params.nsid = 1;
8331 
8332         if (nvme_ns_setup(ns, errp)) {
8333             return;
8334         }
8335 
8336         nvme_attach_ns(n, ns);
8337     }
8338 }
8339 
8340 static void nvme_exit(PCIDevice *pci_dev)
8341 {
8342     NvmeCtrl *n = NVME(pci_dev);
8343     NvmeNamespace *ns;
8344     int i;
8345 
8346     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8347 
8348     if (n->subsys) {
8349         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
8350             ns = nvme_ns(n, i);
8351             if (ns) {
8352                 ns->attached--;
8353             }
8354         }
8355 
8356         nvme_subsys_unregister_ctrl(n->subsys, n);
8357     }
8358 
8359     g_free(n->cq);
8360     g_free(n->sq);
8361     g_free(n->aer_reqs);
8362 
8363     if (n->params.cmb_size_mb) {
8364         g_free(n->cmb.buf);
8365     }
8366 
8367     if (n->pmr.dev) {
8368         host_memory_backend_set_mapped(n->pmr.dev, false);
8369     }
8370 
8371     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8372         pcie_sriov_pf_exit(pci_dev);
8373     }
8374 
8375     msix_uninit(pci_dev, &n->bar0, &n->bar0);
8376     memory_region_del_subregion(&n->bar0, &n->iomem);
8377 }
8378 
8379 static Property nvme_props[] = {
8380     DEFINE_BLOCK_PROPERTIES(NvmeCtrl, namespace.blkconf),
8381     DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmr.dev, TYPE_MEMORY_BACKEND,
8382                      HostMemoryBackend *),
8383     DEFINE_PROP_LINK("subsys", NvmeCtrl, subsys, TYPE_NVME_SUBSYS,
8384                      NvmeSubsystem *),
8385     DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
8386     DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
8387     DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
8388     DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
8389     DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
8390     DEFINE_PROP_UINT8("aerl", NvmeCtrl, params.aerl, 3),
8391     DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl, params.aer_max_queued, 64),
8392     DEFINE_PROP_UINT8("mdts", NvmeCtrl, params.mdts, 7),
8393     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
8394     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
8395     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
8396     DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl, params.ioeventfd, false),
8397     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
8398     DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl,
8399                      params.auto_transition_zones, true),
8400     DEFINE_PROP_UINT8("sriov_max_vfs", NvmeCtrl, params.sriov_max_vfs, 0),
8401     DEFINE_PROP_UINT16("sriov_vq_flexible", NvmeCtrl,
8402                        params.sriov_vq_flexible, 0),
8403     DEFINE_PROP_UINT16("sriov_vi_flexible", NvmeCtrl,
8404                        params.sriov_vi_flexible, 0),
8405     DEFINE_PROP_UINT8("sriov_max_vi_per_vf", NvmeCtrl,
8406                       params.sriov_max_vi_per_vf, 0),
8407     DEFINE_PROP_UINT8("sriov_max_vq_per_vf", NvmeCtrl,
8408                       params.sriov_max_vq_per_vf, 0),
8409     DEFINE_PROP_END_OF_LIST(),
8410 };
8411 
8412 static void nvme_get_smart_warning(Object *obj, Visitor *v, const char *name,
8413                                    void *opaque, Error **errp)
8414 {
8415     NvmeCtrl *n = NVME(obj);
8416     uint8_t value = n->smart_critical_warning;
8417 
8418     visit_type_uint8(v, name, &value, errp);
8419 }
8420 
8421 static void nvme_set_smart_warning(Object *obj, Visitor *v, const char *name,
8422                                    void *opaque, Error **errp)
8423 {
8424     NvmeCtrl *n = NVME(obj);
8425     uint8_t value, old_value, cap = 0, index, event;
8426 
8427     if (!visit_type_uint8(v, name, &value, errp)) {
8428         return;
8429     }
8430 
8431     cap = NVME_SMART_SPARE | NVME_SMART_TEMPERATURE | NVME_SMART_RELIABILITY
8432           | NVME_SMART_MEDIA_READ_ONLY | NVME_SMART_FAILED_VOLATILE_MEDIA;
8433     if (NVME_CAP_PMRS(ldq_le_p(&n->bar.cap))) {
8434         cap |= NVME_SMART_PMR_UNRELIABLE;
8435     }
8436 
8437     if ((value & cap) != value) {
8438         error_setg(errp, "unsupported smart critical warning bits: 0x%x",
8439                    value & ~cap);
8440         return;
8441     }
8442 
8443     old_value = n->smart_critical_warning;
8444     n->smart_critical_warning = value;
8445 
8446     /* only inject new bits of smart critical warning */
8447     for (index = 0; index < NVME_SMART_WARN_MAX; index++) {
8448         event = 1 << index;
8449         if (value & ~old_value & event)
8450             nvme_smart_event(n, event);
8451     }
8452 }
8453 
8454 static void nvme_pci_reset(DeviceState *qdev)
8455 {
8456     PCIDevice *pci_dev = PCI_DEVICE(qdev);
8457     NvmeCtrl *n = NVME(pci_dev);
8458 
8459     trace_pci_nvme_pci_reset();
8460     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8461 }
8462 
8463 static void nvme_sriov_pre_write_ctrl(PCIDevice *dev, uint32_t address,
8464                                       uint32_t val, int len)
8465 {
8466     NvmeCtrl *n = NVME(dev);
8467     NvmeSecCtrlEntry *sctrl;
8468     uint16_t sriov_cap = dev->exp.sriov_cap;
8469     uint32_t off = address - sriov_cap;
8470     int i, num_vfs;
8471 
8472     if (!sriov_cap) {
8473         return;
8474     }
8475 
8476     if (range_covers_byte(off, len, PCI_SRIOV_CTRL)) {
8477         if (!(val & PCI_SRIOV_CTRL_VFE)) {
8478             num_vfs = pci_get_word(dev->config + sriov_cap + PCI_SRIOV_NUM_VF);
8479             for (i = 0; i < num_vfs; i++) {
8480                 sctrl = &n->sec_ctrl_list.sec[i];
8481                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
8482             }
8483         }
8484     }
8485 }
8486 
8487 static void nvme_pci_write_config(PCIDevice *dev, uint32_t address,
8488                                   uint32_t val, int len)
8489 {
8490     nvme_sriov_pre_write_ctrl(dev, address, val, len);
8491     pci_default_write_config(dev, address, val, len);
8492     pcie_cap_flr_write_config(dev, address, val, len);
8493 }
8494 
8495 static const VMStateDescription nvme_vmstate = {
8496     .name = "nvme",
8497     .unmigratable = 1,
8498 };
8499 
8500 static void nvme_class_init(ObjectClass *oc, void *data)
8501 {
8502     DeviceClass *dc = DEVICE_CLASS(oc);
8503     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
8504 
8505     pc->realize = nvme_realize;
8506     pc->config_write = nvme_pci_write_config;
8507     pc->exit = nvme_exit;
8508     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
8509     pc->revision = 2;
8510 
8511     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
8512     dc->desc = "Non-Volatile Memory Express";
8513     device_class_set_props(dc, nvme_props);
8514     dc->vmsd = &nvme_vmstate;
8515     dc->reset = nvme_pci_reset;
8516 }
8517 
8518 static void nvme_instance_init(Object *obj)
8519 {
8520     NvmeCtrl *n = NVME(obj);
8521 
8522     device_add_bootindex_property(obj, &n->namespace.blkconf.bootindex,
8523                                   "bootindex", "/namespace@1,0",
8524                                   DEVICE(obj));
8525 
8526     object_property_add(obj, "smart_critical_warning", "uint8",
8527                         nvme_get_smart_warning,
8528                         nvme_set_smart_warning, NULL, NULL);
8529 }
8530 
8531 static const TypeInfo nvme_info = {
8532     .name          = TYPE_NVME,
8533     .parent        = TYPE_PCI_DEVICE,
8534     .instance_size = sizeof(NvmeCtrl),
8535     .instance_init = nvme_instance_init,
8536     .class_init    = nvme_class_init,
8537     .interfaces = (InterfaceInfo[]) {
8538         { INTERFACE_PCIE_DEVICE },
8539         { }
8540     },
8541 };
8542 
8543 static const TypeInfo nvme_bus_info = {
8544     .name = TYPE_NVME_BUS,
8545     .parent = TYPE_BUS,
8546     .instance_size = sizeof(NvmeBus),
8547 };
8548 
8549 static void nvme_register_types(void)
8550 {
8551     type_register_static(&nvme_info);
8552     type_register_static(&nvme_bus_info);
8553 }
8554 
8555 type_init(nvme_register_types)
8556