xref: /openbmc/qemu/hw/nvme/ctrl.c (revision bad5cfcd)
1 /*
2  * QEMU NVM Express Controller
3  *
4  * Copyright (c) 2012, Intel Corporation
5  *
6  * Written by Keith Busch <keith.busch@intel.com>
7  *
8  * This code is licensed under the GNU GPL v2 or later.
9  */
10 
11 /**
12  * Reference Specs: http://www.nvmexpress.org, 1.4, 1.3, 1.2, 1.1, 1.0e
13  *
14  *  https://nvmexpress.org/developers/nvme-specification/
15  *
16  *
17  * Notes on coding style
18  * ---------------------
19  * While QEMU coding style prefers lowercase hexadecimals in constants, the
20  * NVMe subsystem use thes format from the NVMe specifications in the comments
21  * (i.e. 'h' suffix instead of '0x' prefix).
22  *
23  * Usage
24  * -----
25  * See docs/system/nvme.rst for extensive documentation.
26  *
27  * Add options:
28  *      -drive file=<file>,if=none,id=<drive_id>
29  *      -device nvme-subsys,id=<subsys_id>,nqn=<nqn_id>
30  *      -device nvme,serial=<serial>,id=<bus_name>, \
31  *              cmb_size_mb=<cmb_size_mb[optional]>, \
32  *              [pmrdev=<mem_backend_file_id>,] \
33  *              max_ioqpairs=<N[optional]>, \
34  *              aerl=<N[optional]>,aer_max_queued=<N[optional]>, \
35  *              mdts=<N[optional]>,vsl=<N[optional]>, \
36  *              zoned.zasl=<N[optional]>, \
37  *              zoned.auto_transition=<on|off[optional]>, \
38  *              sriov_max_vfs=<N[optional]> \
39  *              sriov_vq_flexible=<N[optional]> \
40  *              sriov_vi_flexible=<N[optional]> \
41  *              sriov_max_vi_per_vf=<N[optional]> \
42  *              sriov_max_vq_per_vf=<N[optional]> \
43  *              subsys=<subsys_id>
44  *      -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
45  *              zoned=<true|false[optional]>, \
46  *              subsys=<subsys_id>,shared=<true|false[optional]>, \
47  *              detached=<true|false[optional]>, \
48  *              zoned.zone_size=<N[optional]>, \
49  *              zoned.zone_capacity=<N[optional]>, \
50  *              zoned.descr_ext_size=<N[optional]>, \
51  *              zoned.max_active=<N[optional]>, \
52  *              zoned.max_open=<N[optional]>, \
53  *              zoned.cross_read=<true|false[optional]>
54  *
55  * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
56  * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. By default, the
57  * device will use the "v1.4 CMB scheme" - use the `legacy-cmb` parameter to
58  * always enable the CMBLOC and CMBSZ registers (v1.3 behavior).
59  *
60  * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
61  * For example:
62  * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
63  *  size=<size> .... -device nvme,...,pmrdev=<mem_id>
64  *
65  * The PMR will use BAR 4/5 exclusively.
66  *
67  * To place controller(s) and namespace(s) to a subsystem, then provide
68  * nvme-subsys device as above.
69  *
70  * nvme subsystem device parameters
71  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
72  * - `nqn`
73  *   This parameter provides the `<nqn_id>` part of the string
74  *   `nqn.2019-08.org.qemu:<nqn_id>` which will be reported in the SUBNQN field
75  *   of subsystem controllers. Note that `<nqn_id>` should be unique per
76  *   subsystem, but this is not enforced by QEMU. If not specified, it will
77  *   default to the value of the `id` parameter (`<subsys_id>`).
78  *
79  * nvme device parameters
80  * ~~~~~~~~~~~~~~~~~~~~~~
81  * - `subsys`
82  *   Specifying this parameter attaches the controller to the subsystem and
83  *   the SUBNQN field in the controller will report the NQN of the subsystem
84  *   device. This also enables multi controller capability represented in
85  *   Identify Controller data structure in CMIC (Controller Multi-path I/O and
86  *   Namespace Sharing Capabilities).
87  *
88  * - `aerl`
89  *   The Asynchronous Event Request Limit (AERL). Indicates the maximum number
90  *   of concurrently outstanding Asynchronous Event Request commands support
91  *   by the controller. This is a 0's based value.
92  *
93  * - `aer_max_queued`
94  *   This is the maximum number of events that the device will enqueue for
95  *   completion when there are no outstanding AERs. When the maximum number of
96  *   enqueued events are reached, subsequent events will be dropped.
97  *
98  * - `mdts`
99  *   Indicates the maximum data transfer size for a command that transfers data
100  *   between host-accessible memory and the controller. The value is specified
101  *   as a power of two (2^n) and is in units of the minimum memory page size
102  *   (CAP.MPSMIN). The default value is 7 (i.e. 512 KiB).
103  *
104  * - `vsl`
105  *   Indicates the maximum data size limit for the Verify command. Like `mdts`,
106  *   this value is specified as a power of two (2^n) and is in units of the
107  *   minimum memory page size (CAP.MPSMIN). The default value is 7 (i.e. 512
108  *   KiB).
109  *
110  * - `zoned.zasl`
111  *   Indicates the maximum data transfer size for the Zone Append command. Like
112  *   `mdts`, the value is specified as a power of two (2^n) and is in units of
113  *   the minimum memory page size (CAP.MPSMIN). The default value is 0 (i.e.
114  *   defaulting to the value of `mdts`).
115  *
116  * - `zoned.auto_transition`
117  *   Indicates if zones in zone state implicitly opened can be automatically
118  *   transitioned to zone state closed for resource management purposes.
119  *   Defaults to 'on'.
120  *
121  * - `sriov_max_vfs`
122  *   Indicates the maximum number of PCIe virtual functions supported
123  *   by the controller. The default value is 0. Specifying a non-zero value
124  *   enables reporting of both SR-IOV and ARI capabilities by the NVMe device.
125  *   Virtual function controllers will not report SR-IOV capability.
126  *
127  *   NOTE: Single Root I/O Virtualization support is experimental.
128  *   All the related parameters may be subject to change.
129  *
130  * - `sriov_vq_flexible`
131  *   Indicates the total number of flexible queue resources assignable to all
132  *   the secondary controllers. Implicitly sets the number of primary
133  *   controller's private resources to `(max_ioqpairs - sriov_vq_flexible)`.
134  *
135  * - `sriov_vi_flexible`
136  *   Indicates the total number of flexible interrupt resources assignable to
137  *   all the secondary controllers. Implicitly sets the number of primary
138  *   controller's private resources to `(msix_qsize - sriov_vi_flexible)`.
139  *
140  * - `sriov_max_vi_per_vf`
141  *   Indicates the maximum number of virtual interrupt resources assignable
142  *   to a secondary controller. The default 0 resolves to
143  *   `(sriov_vi_flexible / sriov_max_vfs)`.
144  *
145  * - `sriov_max_vq_per_vf`
146  *   Indicates the maximum number of virtual queue resources assignable to
147  *   a secondary controller. The default 0 resolves to
148  *   `(sriov_vq_flexible / sriov_max_vfs)`.
149  *
150  * nvme namespace device parameters
151  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
152  * - `shared`
153  *   When the parent nvme device (as defined explicitly by the 'bus' parameter
154  *   or implicitly by the most recently defined NvmeBus) is linked to an
155  *   nvme-subsys device, the namespace will be attached to all controllers in
156  *   the subsystem. If set to 'off' (the default), the namespace will remain a
157  *   private namespace and may only be attached to a single controller at a
158  *   time.
159  *
160  * - `detached`
161  *   This parameter is only valid together with the `subsys` parameter. If left
162  *   at the default value (`false/off`), the namespace will be attached to all
163  *   controllers in the NVMe subsystem at boot-up. If set to `true/on`, the
164  *   namespace will be available in the subsystem but not attached to any
165  *   controllers.
166  *
167  * Setting `zoned` to true selects Zoned Command Set at the namespace.
168  * In this case, the following namespace properties are available to configure
169  * zoned operation:
170  *     zoned.zone_size=<zone size in bytes, default: 128MiB>
171  *         The number may be followed by K, M, G as in kilo-, mega- or giga-.
172  *
173  *     zoned.zone_capacity=<zone capacity in bytes, default: zone size>
174  *         The value 0 (default) forces zone capacity to be the same as zone
175  *         size. The value of this property may not exceed zone size.
176  *
177  *     zoned.descr_ext_size=<zone descriptor extension size, default 0>
178  *         This value needs to be specified in 64B units. If it is zero,
179  *         namespace(s) will not support zone descriptor extensions.
180  *
181  *     zoned.max_active=<Maximum Active Resources (zones), default: 0>
182  *         The default value means there is no limit to the number of
183  *         concurrently active zones.
184  *
185  *     zoned.max_open=<Maximum Open Resources (zones), default: 0>
186  *         The default value means there is no limit to the number of
187  *         concurrently open zones.
188  *
189  *     zoned.cross_read=<enable RAZB, default: false>
190  *         Setting this property to true enables Read Across Zone Boundaries.
191  */
192 
193 #include "qemu/osdep.h"
194 #include "qemu/cutils.h"
195 #include "qemu/error-report.h"
196 #include "qemu/log.h"
197 #include "qemu/units.h"
198 #include "qemu/range.h"
199 #include "qapi/error.h"
200 #include "qapi/visitor.h"
201 #include "sysemu/sysemu.h"
202 #include "sysemu/block-backend.h"
203 #include "sysemu/hostmem.h"
204 #include "hw/pci/msix.h"
205 #include "hw/pci/pcie_sriov.h"
206 #include "migration/vmstate.h"
207 
208 #include "nvme.h"
209 #include "dif.h"
210 #include "trace.h"
211 
212 #define NVME_MAX_IOQPAIRS 0xffff
213 #define NVME_DB_SIZE  4
214 #define NVME_SPEC_VER 0x00010400
215 #define NVME_CMB_BIR 2
216 #define NVME_PMR_BIR 4
217 #define NVME_TEMPERATURE 0x143
218 #define NVME_TEMPERATURE_WARNING 0x157
219 #define NVME_TEMPERATURE_CRITICAL 0x175
220 #define NVME_NUM_FW_SLOTS 1
221 #define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
222 #define NVME_MAX_VFS 127
223 #define NVME_VF_RES_GRANULARITY 1
224 #define NVME_VF_OFFSET 0x1
225 #define NVME_VF_STRIDE 1
226 
227 #define NVME_GUEST_ERR(trace, fmt, ...) \
228     do { \
229         (trace_##trace)(__VA_ARGS__); \
230         qemu_log_mask(LOG_GUEST_ERROR, #trace \
231             " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
232     } while (0)
233 
234 static const bool nvme_feature_support[NVME_FID_MAX] = {
235     [NVME_ARBITRATION]              = true,
236     [NVME_POWER_MANAGEMENT]         = true,
237     [NVME_TEMPERATURE_THRESHOLD]    = true,
238     [NVME_ERROR_RECOVERY]           = true,
239     [NVME_VOLATILE_WRITE_CACHE]     = true,
240     [NVME_NUMBER_OF_QUEUES]         = true,
241     [NVME_INTERRUPT_COALESCING]     = true,
242     [NVME_INTERRUPT_VECTOR_CONF]    = true,
243     [NVME_WRITE_ATOMICITY]          = true,
244     [NVME_ASYNCHRONOUS_EVENT_CONF]  = true,
245     [NVME_TIMESTAMP]                = true,
246     [NVME_HOST_BEHAVIOR_SUPPORT]    = true,
247     [NVME_COMMAND_SET_PROFILE]      = true,
248     [NVME_FDP_MODE]                 = true,
249     [NVME_FDP_EVENTS]               = true,
250 };
251 
252 static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
253     [NVME_TEMPERATURE_THRESHOLD]    = NVME_FEAT_CAP_CHANGE,
254     [NVME_ERROR_RECOVERY]           = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
255     [NVME_VOLATILE_WRITE_CACHE]     = NVME_FEAT_CAP_CHANGE,
256     [NVME_NUMBER_OF_QUEUES]         = NVME_FEAT_CAP_CHANGE,
257     [NVME_ASYNCHRONOUS_EVENT_CONF]  = NVME_FEAT_CAP_CHANGE,
258     [NVME_TIMESTAMP]                = NVME_FEAT_CAP_CHANGE,
259     [NVME_HOST_BEHAVIOR_SUPPORT]    = NVME_FEAT_CAP_CHANGE,
260     [NVME_COMMAND_SET_PROFILE]      = NVME_FEAT_CAP_CHANGE,
261     [NVME_FDP_MODE]                 = NVME_FEAT_CAP_CHANGE,
262     [NVME_FDP_EVENTS]               = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
263 };
264 
265 static const uint32_t nvme_cse_acs[256] = {
266     [NVME_ADM_CMD_DELETE_SQ]        = NVME_CMD_EFF_CSUPP,
267     [NVME_ADM_CMD_CREATE_SQ]        = NVME_CMD_EFF_CSUPP,
268     [NVME_ADM_CMD_GET_LOG_PAGE]     = NVME_CMD_EFF_CSUPP,
269     [NVME_ADM_CMD_DELETE_CQ]        = NVME_CMD_EFF_CSUPP,
270     [NVME_ADM_CMD_CREATE_CQ]        = NVME_CMD_EFF_CSUPP,
271     [NVME_ADM_CMD_IDENTIFY]         = NVME_CMD_EFF_CSUPP,
272     [NVME_ADM_CMD_ABORT]            = NVME_CMD_EFF_CSUPP,
273     [NVME_ADM_CMD_SET_FEATURES]     = NVME_CMD_EFF_CSUPP,
274     [NVME_ADM_CMD_GET_FEATURES]     = NVME_CMD_EFF_CSUPP,
275     [NVME_ADM_CMD_ASYNC_EV_REQ]     = NVME_CMD_EFF_CSUPP,
276     [NVME_ADM_CMD_NS_ATTACHMENT]    = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_NIC,
277     [NVME_ADM_CMD_VIRT_MNGMT]       = NVME_CMD_EFF_CSUPP,
278     [NVME_ADM_CMD_DBBUF_CONFIG]     = NVME_CMD_EFF_CSUPP,
279     [NVME_ADM_CMD_FORMAT_NVM]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
280     [NVME_ADM_CMD_DIRECTIVE_RECV]   = NVME_CMD_EFF_CSUPP,
281     [NVME_ADM_CMD_DIRECTIVE_SEND]   = NVME_CMD_EFF_CSUPP,
282 };
283 
284 static const uint32_t nvme_cse_iocs_none[256];
285 
286 static const uint32_t nvme_cse_iocs_nvm[256] = {
287     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
288     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
289     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
290     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
291     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
292     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
293     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
294     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
295     [NVME_CMD_IO_MGMT_RECV]         = NVME_CMD_EFF_CSUPP,
296     [NVME_CMD_IO_MGMT_SEND]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
297 };
298 
299 static const uint32_t nvme_cse_iocs_zoned[256] = {
300     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
301     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
302     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
303     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
304     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
305     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
306     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
307     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
308     [NVME_CMD_ZONE_APPEND]          = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
309     [NVME_CMD_ZONE_MGMT_SEND]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
310     [NVME_CMD_ZONE_MGMT_RECV]       = NVME_CMD_EFF_CSUPP,
311 };
312 
313 static void nvme_process_sq(void *opaque);
314 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst);
315 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n);
316 
317 static uint16_t nvme_sqid(NvmeRequest *req)
318 {
319     return le16_to_cpu(req->sq->sqid);
320 }
321 
322 static inline uint16_t nvme_make_pid(NvmeNamespace *ns, uint16_t rg,
323                                      uint16_t ph)
324 {
325     uint16_t rgif = ns->endgrp->fdp.rgif;
326 
327     if (!rgif) {
328         return ph;
329     }
330 
331     return (rg << (16 - rgif)) | ph;
332 }
333 
334 static inline bool nvme_ph_valid(NvmeNamespace *ns, uint16_t ph)
335 {
336     return ph < ns->fdp.nphs;
337 }
338 
339 static inline bool nvme_rg_valid(NvmeEnduranceGroup *endgrp, uint16_t rg)
340 {
341     return rg < endgrp->fdp.nrg;
342 }
343 
344 static inline uint16_t nvme_pid2ph(NvmeNamespace *ns, uint16_t pid)
345 {
346     uint16_t rgif = ns->endgrp->fdp.rgif;
347 
348     if (!rgif) {
349         return pid;
350     }
351 
352     return pid & ((1 << (15 - rgif)) - 1);
353 }
354 
355 static inline uint16_t nvme_pid2rg(NvmeNamespace *ns, uint16_t pid)
356 {
357     uint16_t rgif = ns->endgrp->fdp.rgif;
358 
359     if (!rgif) {
360         return 0;
361     }
362 
363     return pid >> (16 - rgif);
364 }
365 
366 static inline bool nvme_parse_pid(NvmeNamespace *ns, uint16_t pid,
367                                   uint16_t *ph, uint16_t *rg)
368 {
369     *rg = nvme_pid2rg(ns, pid);
370     *ph = nvme_pid2ph(ns, pid);
371 
372     return nvme_ph_valid(ns, *ph) && nvme_rg_valid(ns->endgrp, *rg);
373 }
374 
375 static void nvme_assign_zone_state(NvmeNamespace *ns, NvmeZone *zone,
376                                    NvmeZoneState state)
377 {
378     if (QTAILQ_IN_USE(zone, entry)) {
379         switch (nvme_get_zone_state(zone)) {
380         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
381             QTAILQ_REMOVE(&ns->exp_open_zones, zone, entry);
382             break;
383         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
384             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
385             break;
386         case NVME_ZONE_STATE_CLOSED:
387             QTAILQ_REMOVE(&ns->closed_zones, zone, entry);
388             break;
389         case NVME_ZONE_STATE_FULL:
390             QTAILQ_REMOVE(&ns->full_zones, zone, entry);
391         default:
392             ;
393         }
394     }
395 
396     nvme_set_zone_state(zone, state);
397 
398     switch (state) {
399     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
400         QTAILQ_INSERT_TAIL(&ns->exp_open_zones, zone, entry);
401         break;
402     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
403         QTAILQ_INSERT_TAIL(&ns->imp_open_zones, zone, entry);
404         break;
405     case NVME_ZONE_STATE_CLOSED:
406         QTAILQ_INSERT_TAIL(&ns->closed_zones, zone, entry);
407         break;
408     case NVME_ZONE_STATE_FULL:
409         QTAILQ_INSERT_TAIL(&ns->full_zones, zone, entry);
410     case NVME_ZONE_STATE_READ_ONLY:
411         break;
412     default:
413         zone->d.za = 0;
414     }
415 }
416 
417 static uint16_t nvme_zns_check_resources(NvmeNamespace *ns, uint32_t act,
418                                          uint32_t opn, uint32_t zrwa)
419 {
420     if (ns->params.max_active_zones != 0 &&
421         ns->nr_active_zones + act > ns->params.max_active_zones) {
422         trace_pci_nvme_err_insuff_active_res(ns->params.max_active_zones);
423         return NVME_ZONE_TOO_MANY_ACTIVE | NVME_DNR;
424     }
425 
426     if (ns->params.max_open_zones != 0 &&
427         ns->nr_open_zones + opn > ns->params.max_open_zones) {
428         trace_pci_nvme_err_insuff_open_res(ns->params.max_open_zones);
429         return NVME_ZONE_TOO_MANY_OPEN | NVME_DNR;
430     }
431 
432     if (zrwa > ns->zns.numzrwa) {
433         return NVME_NOZRWA | NVME_DNR;
434     }
435 
436     return NVME_SUCCESS;
437 }
438 
439 /*
440  * Check if we can open a zone without exceeding open/active limits.
441  * AOR stands for "Active and Open Resources" (see TP 4053 section 2.5).
442  */
443 static uint16_t nvme_aor_check(NvmeNamespace *ns, uint32_t act, uint32_t opn)
444 {
445     return nvme_zns_check_resources(ns, act, opn, 0);
446 }
447 
448 static NvmeFdpEvent *nvme_fdp_alloc_event(NvmeCtrl *n, NvmeFdpEventBuffer *ebuf)
449 {
450     NvmeFdpEvent *ret = NULL;
451     bool is_full = ebuf->next == ebuf->start && ebuf->nelems;
452 
453     ret = &ebuf->events[ebuf->next++];
454     if (unlikely(ebuf->next == NVME_FDP_MAX_EVENTS)) {
455         ebuf->next = 0;
456     }
457     if (is_full) {
458         ebuf->start = ebuf->next;
459     } else {
460         ebuf->nelems++;
461     }
462 
463     memset(ret, 0, sizeof(NvmeFdpEvent));
464     ret->timestamp = nvme_get_timestamp(n);
465 
466     return ret;
467 }
468 
469 static inline int log_event(NvmeRuHandle *ruh, uint8_t event_type)
470 {
471     return (ruh->event_filter >> nvme_fdp_evf_shifts[event_type]) & 0x1;
472 }
473 
474 static bool nvme_update_ruh(NvmeCtrl *n, NvmeNamespace *ns, uint16_t pid)
475 {
476     NvmeEnduranceGroup *endgrp = ns->endgrp;
477     NvmeRuHandle *ruh;
478     NvmeReclaimUnit *ru;
479     NvmeFdpEvent *e = NULL;
480     uint16_t ph, rg, ruhid;
481 
482     if (!nvme_parse_pid(ns, pid, &ph, &rg)) {
483         return false;
484     }
485 
486     ruhid = ns->fdp.phs[ph];
487 
488     ruh = &endgrp->fdp.ruhs[ruhid];
489     ru = &ruh->rus[rg];
490 
491     if (ru->ruamw) {
492         if (log_event(ruh, FDP_EVT_RU_NOT_FULLY_WRITTEN)) {
493             e = nvme_fdp_alloc_event(n, &endgrp->fdp.host_events);
494             e->type = FDP_EVT_RU_NOT_FULLY_WRITTEN;
495             e->flags = FDPEF_PIV | FDPEF_NSIDV | FDPEF_LV;
496             e->pid = cpu_to_le16(pid);
497             e->nsid = cpu_to_le32(ns->params.nsid);
498             e->rgid = cpu_to_le16(rg);
499             e->ruhid = cpu_to_le16(ruhid);
500         }
501 
502         /* log (eventual) GC overhead of prematurely swapping the RU */
503         nvme_fdp_stat_inc(&endgrp->fdp.mbmw, nvme_l2b(ns, ru->ruamw));
504     }
505 
506     ru->ruamw = ruh->ruamw;
507 
508     return true;
509 }
510 
511 static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
512 {
513     hwaddr hi, lo;
514 
515     if (!n->cmb.cmse) {
516         return false;
517     }
518 
519     lo = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
520     hi = lo + int128_get64(n->cmb.mem.size);
521 
522     return addr >= lo && addr < hi;
523 }
524 
525 static inline void *nvme_addr_to_cmb(NvmeCtrl *n, hwaddr addr)
526 {
527     hwaddr base = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
528     return &n->cmb.buf[addr - base];
529 }
530 
531 static bool nvme_addr_is_pmr(NvmeCtrl *n, hwaddr addr)
532 {
533     hwaddr hi;
534 
535     if (!n->pmr.cmse) {
536         return false;
537     }
538 
539     hi = n->pmr.cba + int128_get64(n->pmr.dev->mr.size);
540 
541     return addr >= n->pmr.cba && addr < hi;
542 }
543 
544 static inline void *nvme_addr_to_pmr(NvmeCtrl *n, hwaddr addr)
545 {
546     return memory_region_get_ram_ptr(&n->pmr.dev->mr) + (addr - n->pmr.cba);
547 }
548 
549 static inline bool nvme_addr_is_iomem(NvmeCtrl *n, hwaddr addr)
550 {
551     hwaddr hi, lo;
552 
553     /*
554      * The purpose of this check is to guard against invalid "local" access to
555      * the iomem (i.e. controller registers). Thus, we check against the range
556      * covered by the 'bar0' MemoryRegion since that is currently composed of
557      * two subregions (the NVMe "MBAR" and the MSI-X table/pba). Note, however,
558      * that if the device model is ever changed to allow the CMB to be located
559      * in BAR0 as well, then this must be changed.
560      */
561     lo = n->bar0.addr;
562     hi = lo + int128_get64(n->bar0.size);
563 
564     return addr >= lo && addr < hi;
565 }
566 
567 static int nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
568 {
569     hwaddr hi = addr + size - 1;
570     if (hi < addr) {
571         return 1;
572     }
573 
574     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
575         memcpy(buf, nvme_addr_to_cmb(n, addr), size);
576         return 0;
577     }
578 
579     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
580         memcpy(buf, nvme_addr_to_pmr(n, addr), size);
581         return 0;
582     }
583 
584     return pci_dma_read(PCI_DEVICE(n), addr, buf, size);
585 }
586 
587 static int nvme_addr_write(NvmeCtrl *n, hwaddr addr, const void *buf, int size)
588 {
589     hwaddr hi = addr + size - 1;
590     if (hi < addr) {
591         return 1;
592     }
593 
594     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
595         memcpy(nvme_addr_to_cmb(n, addr), buf, size);
596         return 0;
597     }
598 
599     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
600         memcpy(nvme_addr_to_pmr(n, addr), buf, size);
601         return 0;
602     }
603 
604     return pci_dma_write(PCI_DEVICE(n), addr, buf, size);
605 }
606 
607 static bool nvme_nsid_valid(NvmeCtrl *n, uint32_t nsid)
608 {
609     return nsid &&
610         (nsid == NVME_NSID_BROADCAST || nsid <= NVME_MAX_NAMESPACES);
611 }
612 
613 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
614 {
615     return sqid < n->conf_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
616 }
617 
618 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
619 {
620     return cqid < n->conf_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
621 }
622 
623 static void nvme_inc_cq_tail(NvmeCQueue *cq)
624 {
625     cq->tail++;
626     if (cq->tail >= cq->size) {
627         cq->tail = 0;
628         cq->phase = !cq->phase;
629     }
630 }
631 
632 static void nvme_inc_sq_head(NvmeSQueue *sq)
633 {
634     sq->head = (sq->head + 1) % sq->size;
635 }
636 
637 static uint8_t nvme_cq_full(NvmeCQueue *cq)
638 {
639     return (cq->tail + 1) % cq->size == cq->head;
640 }
641 
642 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
643 {
644     return sq->head == sq->tail;
645 }
646 
647 static void nvme_irq_check(NvmeCtrl *n)
648 {
649     PCIDevice *pci = PCI_DEVICE(n);
650     uint32_t intms = ldl_le_p(&n->bar.intms);
651 
652     if (msix_enabled(pci)) {
653         return;
654     }
655     if (~intms & n->irq_status) {
656         pci_irq_assert(pci);
657     } else {
658         pci_irq_deassert(pci);
659     }
660 }
661 
662 static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
663 {
664     PCIDevice *pci = PCI_DEVICE(n);
665 
666     if (cq->irq_enabled) {
667         if (msix_enabled(pci)) {
668             trace_pci_nvme_irq_msix(cq->vector);
669             msix_notify(pci, cq->vector);
670         } else {
671             trace_pci_nvme_irq_pin();
672             assert(cq->vector < 32);
673             n->irq_status |= 1 << cq->vector;
674             nvme_irq_check(n);
675         }
676     } else {
677         trace_pci_nvme_irq_masked();
678     }
679 }
680 
681 static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
682 {
683     if (cq->irq_enabled) {
684         if (msix_enabled(PCI_DEVICE(n))) {
685             return;
686         } else {
687             assert(cq->vector < 32);
688             if (!n->cq_pending) {
689                 n->irq_status &= ~(1 << cq->vector);
690             }
691             nvme_irq_check(n);
692         }
693     }
694 }
695 
696 static void nvme_req_clear(NvmeRequest *req)
697 {
698     req->ns = NULL;
699     req->opaque = NULL;
700     req->aiocb = NULL;
701     memset(&req->cqe, 0x0, sizeof(req->cqe));
702     req->status = NVME_SUCCESS;
703 }
704 
705 static inline void nvme_sg_init(NvmeCtrl *n, NvmeSg *sg, bool dma)
706 {
707     if (dma) {
708         pci_dma_sglist_init(&sg->qsg, PCI_DEVICE(n), 0);
709         sg->flags = NVME_SG_DMA;
710     } else {
711         qemu_iovec_init(&sg->iov, 0);
712     }
713 
714     sg->flags |= NVME_SG_ALLOC;
715 }
716 
717 static inline void nvme_sg_unmap(NvmeSg *sg)
718 {
719     if (!(sg->flags & NVME_SG_ALLOC)) {
720         return;
721     }
722 
723     if (sg->flags & NVME_SG_DMA) {
724         qemu_sglist_destroy(&sg->qsg);
725     } else {
726         qemu_iovec_destroy(&sg->iov);
727     }
728 
729     memset(sg, 0x0, sizeof(*sg));
730 }
731 
732 /*
733  * When metadata is transfered as extended LBAs, the DPTR mapped into `sg`
734  * holds both data and metadata. This function splits the data and metadata
735  * into two separate QSG/IOVs.
736  */
737 static void nvme_sg_split(NvmeSg *sg, NvmeNamespace *ns, NvmeSg *data,
738                           NvmeSg *mdata)
739 {
740     NvmeSg *dst = data;
741     uint32_t trans_len, count = ns->lbasz;
742     uint64_t offset = 0;
743     bool dma = sg->flags & NVME_SG_DMA;
744     size_t sge_len;
745     size_t sg_len = dma ? sg->qsg.size : sg->iov.size;
746     int sg_idx = 0;
747 
748     assert(sg->flags & NVME_SG_ALLOC);
749 
750     while (sg_len) {
751         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
752 
753         trans_len = MIN(sg_len, count);
754         trans_len = MIN(trans_len, sge_len - offset);
755 
756         if (dst) {
757             if (dma) {
758                 qemu_sglist_add(&dst->qsg, sg->qsg.sg[sg_idx].base + offset,
759                                 trans_len);
760             } else {
761                 qemu_iovec_add(&dst->iov,
762                                sg->iov.iov[sg_idx].iov_base + offset,
763                                trans_len);
764             }
765         }
766 
767         sg_len -= trans_len;
768         count -= trans_len;
769         offset += trans_len;
770 
771         if (count == 0) {
772             dst = (dst == data) ? mdata : data;
773             count = (dst == data) ? ns->lbasz : ns->lbaf.ms;
774         }
775 
776         if (sge_len == offset) {
777             offset = 0;
778             sg_idx++;
779         }
780     }
781 }
782 
783 static uint16_t nvme_map_addr_cmb(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
784                                   size_t len)
785 {
786     if (!len) {
787         return NVME_SUCCESS;
788     }
789 
790     trace_pci_nvme_map_addr_cmb(addr, len);
791 
792     if (!nvme_addr_is_cmb(n, addr) || !nvme_addr_is_cmb(n, addr + len - 1)) {
793         return NVME_DATA_TRAS_ERROR;
794     }
795 
796     qemu_iovec_add(iov, nvme_addr_to_cmb(n, addr), len);
797 
798     return NVME_SUCCESS;
799 }
800 
801 static uint16_t nvme_map_addr_pmr(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
802                                   size_t len)
803 {
804     if (!len) {
805         return NVME_SUCCESS;
806     }
807 
808     if (!nvme_addr_is_pmr(n, addr) || !nvme_addr_is_pmr(n, addr + len - 1)) {
809         return NVME_DATA_TRAS_ERROR;
810     }
811 
812     qemu_iovec_add(iov, nvme_addr_to_pmr(n, addr), len);
813 
814     return NVME_SUCCESS;
815 }
816 
817 static uint16_t nvme_map_addr(NvmeCtrl *n, NvmeSg *sg, hwaddr addr, size_t len)
818 {
819     bool cmb = false, pmr = false;
820 
821     if (!len) {
822         return NVME_SUCCESS;
823     }
824 
825     trace_pci_nvme_map_addr(addr, len);
826 
827     if (nvme_addr_is_iomem(n, addr)) {
828         return NVME_DATA_TRAS_ERROR;
829     }
830 
831     if (nvme_addr_is_cmb(n, addr)) {
832         cmb = true;
833     } else if (nvme_addr_is_pmr(n, addr)) {
834         pmr = true;
835     }
836 
837     if (cmb || pmr) {
838         if (sg->flags & NVME_SG_DMA) {
839             return NVME_INVALID_USE_OF_CMB | NVME_DNR;
840         }
841 
842         if (sg->iov.niov + 1 > IOV_MAX) {
843             goto max_mappings_exceeded;
844         }
845 
846         if (cmb) {
847             return nvme_map_addr_cmb(n, &sg->iov, addr, len);
848         } else {
849             return nvme_map_addr_pmr(n, &sg->iov, addr, len);
850         }
851     }
852 
853     if (!(sg->flags & NVME_SG_DMA)) {
854         return NVME_INVALID_USE_OF_CMB | NVME_DNR;
855     }
856 
857     if (sg->qsg.nsg + 1 > IOV_MAX) {
858         goto max_mappings_exceeded;
859     }
860 
861     qemu_sglist_add(&sg->qsg, addr, len);
862 
863     return NVME_SUCCESS;
864 
865 max_mappings_exceeded:
866     NVME_GUEST_ERR(pci_nvme_ub_too_many_mappings,
867                    "number of mappings exceed 1024");
868     return NVME_INTERNAL_DEV_ERROR | NVME_DNR;
869 }
870 
871 static inline bool nvme_addr_is_dma(NvmeCtrl *n, hwaddr addr)
872 {
873     return !(nvme_addr_is_cmb(n, addr) || nvme_addr_is_pmr(n, addr));
874 }
875 
876 static uint16_t nvme_map_prp(NvmeCtrl *n, NvmeSg *sg, uint64_t prp1,
877                              uint64_t prp2, uint32_t len)
878 {
879     hwaddr trans_len = n->page_size - (prp1 % n->page_size);
880     trans_len = MIN(len, trans_len);
881     int num_prps = (len >> n->page_bits) + 1;
882     uint16_t status;
883     int ret;
884 
885     trace_pci_nvme_map_prp(trans_len, len, prp1, prp2, num_prps);
886 
887     nvme_sg_init(n, sg, nvme_addr_is_dma(n, prp1));
888 
889     status = nvme_map_addr(n, sg, prp1, trans_len);
890     if (status) {
891         goto unmap;
892     }
893 
894     len -= trans_len;
895     if (len) {
896         if (len > n->page_size) {
897             g_autofree uint64_t *prp_list = g_new(uint64_t, n->max_prp_ents);
898             uint32_t nents, prp_trans;
899             int i = 0;
900 
901             /*
902              * The first PRP list entry, pointed to by PRP2 may contain offset.
903              * Hence, we need to calculate the number of entries in based on
904              * that offset.
905              */
906             nents = (n->page_size - (prp2 & (n->page_size - 1))) >> 3;
907             prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
908             ret = nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
909             if (ret) {
910                 trace_pci_nvme_err_addr_read(prp2);
911                 status = NVME_DATA_TRAS_ERROR;
912                 goto unmap;
913             }
914             while (len != 0) {
915                 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
916 
917                 if (i == nents - 1 && len > n->page_size) {
918                     if (unlikely(prp_ent & (n->page_size - 1))) {
919                         trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
920                         status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
921                         goto unmap;
922                     }
923 
924                     i = 0;
925                     nents = (len + n->page_size - 1) >> n->page_bits;
926                     nents = MIN(nents, n->max_prp_ents);
927                     prp_trans = nents * sizeof(uint64_t);
928                     ret = nvme_addr_read(n, prp_ent, (void *)prp_list,
929                                          prp_trans);
930                     if (ret) {
931                         trace_pci_nvme_err_addr_read(prp_ent);
932                         status = NVME_DATA_TRAS_ERROR;
933                         goto unmap;
934                     }
935                     prp_ent = le64_to_cpu(prp_list[i]);
936                 }
937 
938                 if (unlikely(prp_ent & (n->page_size - 1))) {
939                     trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
940                     status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
941                     goto unmap;
942                 }
943 
944                 trans_len = MIN(len, n->page_size);
945                 status = nvme_map_addr(n, sg, prp_ent, trans_len);
946                 if (status) {
947                     goto unmap;
948                 }
949 
950                 len -= trans_len;
951                 i++;
952             }
953         } else {
954             if (unlikely(prp2 & (n->page_size - 1))) {
955                 trace_pci_nvme_err_invalid_prp2_align(prp2);
956                 status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
957                 goto unmap;
958             }
959             status = nvme_map_addr(n, sg, prp2, len);
960             if (status) {
961                 goto unmap;
962             }
963         }
964     }
965 
966     return NVME_SUCCESS;
967 
968 unmap:
969     nvme_sg_unmap(sg);
970     return status;
971 }
972 
973 /*
974  * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
975  * number of bytes mapped in len.
976  */
977 static uint16_t nvme_map_sgl_data(NvmeCtrl *n, NvmeSg *sg,
978                                   NvmeSglDescriptor *segment, uint64_t nsgld,
979                                   size_t *len, NvmeCmd *cmd)
980 {
981     dma_addr_t addr, trans_len;
982     uint32_t dlen;
983     uint16_t status;
984 
985     for (int i = 0; i < nsgld; i++) {
986         uint8_t type = NVME_SGL_TYPE(segment[i].type);
987 
988         switch (type) {
989         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
990             break;
991         case NVME_SGL_DESCR_TYPE_SEGMENT:
992         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
993             return NVME_INVALID_NUM_SGL_DESCRS | NVME_DNR;
994         default:
995             return NVME_SGL_DESCR_TYPE_INVALID | NVME_DNR;
996         }
997 
998         dlen = le32_to_cpu(segment[i].len);
999 
1000         if (!dlen) {
1001             continue;
1002         }
1003 
1004         if (*len == 0) {
1005             /*
1006              * All data has been mapped, but the SGL contains additional
1007              * segments and/or descriptors. The controller might accept
1008              * ignoring the rest of the SGL.
1009              */
1010             uint32_t sgls = le32_to_cpu(n->id_ctrl.sgls);
1011             if (sgls & NVME_CTRL_SGLS_EXCESS_LENGTH) {
1012                 break;
1013             }
1014 
1015             trace_pci_nvme_err_invalid_sgl_excess_length(dlen);
1016             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1017         }
1018 
1019         trans_len = MIN(*len, dlen);
1020 
1021         addr = le64_to_cpu(segment[i].addr);
1022 
1023         if (UINT64_MAX - addr < dlen) {
1024             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1025         }
1026 
1027         status = nvme_map_addr(n, sg, addr, trans_len);
1028         if (status) {
1029             return status;
1030         }
1031 
1032         *len -= trans_len;
1033     }
1034 
1035     return NVME_SUCCESS;
1036 }
1037 
1038 static uint16_t nvme_map_sgl(NvmeCtrl *n, NvmeSg *sg, NvmeSglDescriptor sgl,
1039                              size_t len, NvmeCmd *cmd)
1040 {
1041     /*
1042      * Read the segment in chunks of 256 descriptors (one 4k page) to avoid
1043      * dynamically allocating a potentially huge SGL. The spec allows the SGL
1044      * to be larger (as in number of bytes required to describe the SGL
1045      * descriptors and segment chain) than the command transfer size, so it is
1046      * not bounded by MDTS.
1047      */
1048 #define SEG_CHUNK_SIZE 256
1049 
1050     NvmeSglDescriptor segment[SEG_CHUNK_SIZE], *sgld, *last_sgld;
1051     uint64_t nsgld;
1052     uint32_t seg_len;
1053     uint16_t status;
1054     hwaddr addr;
1055     int ret;
1056 
1057     sgld = &sgl;
1058     addr = le64_to_cpu(sgl.addr);
1059 
1060     trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl.type), len);
1061 
1062     nvme_sg_init(n, sg, nvme_addr_is_dma(n, addr));
1063 
1064     /*
1065      * If the entire transfer can be described with a single data block it can
1066      * be mapped directly.
1067      */
1068     if (NVME_SGL_TYPE(sgl.type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1069         status = nvme_map_sgl_data(n, sg, sgld, 1, &len, cmd);
1070         if (status) {
1071             goto unmap;
1072         }
1073 
1074         goto out;
1075     }
1076 
1077     for (;;) {
1078         switch (NVME_SGL_TYPE(sgld->type)) {
1079         case NVME_SGL_DESCR_TYPE_SEGMENT:
1080         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
1081             break;
1082         default:
1083             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1084         }
1085 
1086         seg_len = le32_to_cpu(sgld->len);
1087 
1088         /* check the length of the (Last) Segment descriptor */
1089         if (!seg_len || seg_len & 0xf) {
1090             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1091         }
1092 
1093         if (UINT64_MAX - addr < seg_len) {
1094             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1095         }
1096 
1097         nsgld = seg_len / sizeof(NvmeSglDescriptor);
1098 
1099         while (nsgld > SEG_CHUNK_SIZE) {
1100             if (nvme_addr_read(n, addr, segment, sizeof(segment))) {
1101                 trace_pci_nvme_err_addr_read(addr);
1102                 status = NVME_DATA_TRAS_ERROR;
1103                 goto unmap;
1104             }
1105 
1106             status = nvme_map_sgl_data(n, sg, segment, SEG_CHUNK_SIZE,
1107                                        &len, cmd);
1108             if (status) {
1109                 goto unmap;
1110             }
1111 
1112             nsgld -= SEG_CHUNK_SIZE;
1113             addr += SEG_CHUNK_SIZE * sizeof(NvmeSglDescriptor);
1114         }
1115 
1116         ret = nvme_addr_read(n, addr, segment, nsgld *
1117                              sizeof(NvmeSglDescriptor));
1118         if (ret) {
1119             trace_pci_nvme_err_addr_read(addr);
1120             status = NVME_DATA_TRAS_ERROR;
1121             goto unmap;
1122         }
1123 
1124         last_sgld = &segment[nsgld - 1];
1125 
1126         /*
1127          * If the segment ends with a Data Block, then we are done.
1128          */
1129         if (NVME_SGL_TYPE(last_sgld->type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1130             status = nvme_map_sgl_data(n, sg, segment, nsgld, &len, cmd);
1131             if (status) {
1132                 goto unmap;
1133             }
1134 
1135             goto out;
1136         }
1137 
1138         /*
1139          * If the last descriptor was not a Data Block, then the current
1140          * segment must not be a Last Segment.
1141          */
1142         if (NVME_SGL_TYPE(sgld->type) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT) {
1143             status = NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1144             goto unmap;
1145         }
1146 
1147         sgld = last_sgld;
1148         addr = le64_to_cpu(sgld->addr);
1149 
1150         /*
1151          * Do not map the last descriptor; it will be a Segment or Last Segment
1152          * descriptor and is handled by the next iteration.
1153          */
1154         status = nvme_map_sgl_data(n, sg, segment, nsgld - 1, &len, cmd);
1155         if (status) {
1156             goto unmap;
1157         }
1158     }
1159 
1160 out:
1161     /* if there is any residual left in len, the SGL was too short */
1162     if (len) {
1163         status = NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1164         goto unmap;
1165     }
1166 
1167     return NVME_SUCCESS;
1168 
1169 unmap:
1170     nvme_sg_unmap(sg);
1171     return status;
1172 }
1173 
1174 uint16_t nvme_map_dptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1175                        NvmeCmd *cmd)
1176 {
1177     uint64_t prp1, prp2;
1178 
1179     switch (NVME_CMD_FLAGS_PSDT(cmd->flags)) {
1180     case NVME_PSDT_PRP:
1181         prp1 = le64_to_cpu(cmd->dptr.prp1);
1182         prp2 = le64_to_cpu(cmd->dptr.prp2);
1183 
1184         return nvme_map_prp(n, sg, prp1, prp2, len);
1185     case NVME_PSDT_SGL_MPTR_CONTIGUOUS:
1186     case NVME_PSDT_SGL_MPTR_SGL:
1187         return nvme_map_sgl(n, sg, cmd->dptr.sgl, len, cmd);
1188     default:
1189         return NVME_INVALID_FIELD;
1190     }
1191 }
1192 
1193 static uint16_t nvme_map_mptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1194                               NvmeCmd *cmd)
1195 {
1196     int psdt = NVME_CMD_FLAGS_PSDT(cmd->flags);
1197     hwaddr mptr = le64_to_cpu(cmd->mptr);
1198     uint16_t status;
1199 
1200     if (psdt == NVME_PSDT_SGL_MPTR_SGL) {
1201         NvmeSglDescriptor sgl;
1202 
1203         if (nvme_addr_read(n, mptr, &sgl, sizeof(sgl))) {
1204             return NVME_DATA_TRAS_ERROR;
1205         }
1206 
1207         status = nvme_map_sgl(n, sg, sgl, len, cmd);
1208         if (status && (status & 0x7ff) == NVME_DATA_SGL_LEN_INVALID) {
1209             status = NVME_MD_SGL_LEN_INVALID | NVME_DNR;
1210         }
1211 
1212         return status;
1213     }
1214 
1215     nvme_sg_init(n, sg, nvme_addr_is_dma(n, mptr));
1216     status = nvme_map_addr(n, sg, mptr, len);
1217     if (status) {
1218         nvme_sg_unmap(sg);
1219     }
1220 
1221     return status;
1222 }
1223 
1224 static uint16_t nvme_map_data(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1225 {
1226     NvmeNamespace *ns = req->ns;
1227     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1228     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1229     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1230     size_t len = nvme_l2b(ns, nlb);
1231     uint16_t status;
1232 
1233     if (nvme_ns_ext(ns) &&
1234         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1235         NvmeSg sg;
1236 
1237         len += nvme_m2b(ns, nlb);
1238 
1239         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1240         if (status) {
1241             return status;
1242         }
1243 
1244         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1245         nvme_sg_split(&sg, ns, &req->sg, NULL);
1246         nvme_sg_unmap(&sg);
1247 
1248         return NVME_SUCCESS;
1249     }
1250 
1251     return nvme_map_dptr(n, &req->sg, len, &req->cmd);
1252 }
1253 
1254 static uint16_t nvme_map_mdata(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1255 {
1256     NvmeNamespace *ns = req->ns;
1257     size_t len = nvme_m2b(ns, nlb);
1258     uint16_t status;
1259 
1260     if (nvme_ns_ext(ns)) {
1261         NvmeSg sg;
1262 
1263         len += nvme_l2b(ns, nlb);
1264 
1265         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1266         if (status) {
1267             return status;
1268         }
1269 
1270         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1271         nvme_sg_split(&sg, ns, NULL, &req->sg);
1272         nvme_sg_unmap(&sg);
1273 
1274         return NVME_SUCCESS;
1275     }
1276 
1277     return nvme_map_mptr(n, &req->sg, len, &req->cmd);
1278 }
1279 
1280 static uint16_t nvme_tx_interleaved(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr,
1281                                     uint32_t len, uint32_t bytes,
1282                                     int32_t skip_bytes, int64_t offset,
1283                                     NvmeTxDirection dir)
1284 {
1285     hwaddr addr;
1286     uint32_t trans_len, count = bytes;
1287     bool dma = sg->flags & NVME_SG_DMA;
1288     int64_t sge_len;
1289     int sg_idx = 0;
1290     int ret;
1291 
1292     assert(sg->flags & NVME_SG_ALLOC);
1293 
1294     while (len) {
1295         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
1296 
1297         if (sge_len - offset < 0) {
1298             offset -= sge_len;
1299             sg_idx++;
1300             continue;
1301         }
1302 
1303         if (sge_len == offset) {
1304             offset = 0;
1305             sg_idx++;
1306             continue;
1307         }
1308 
1309         trans_len = MIN(len, count);
1310         trans_len = MIN(trans_len, sge_len - offset);
1311 
1312         if (dma) {
1313             addr = sg->qsg.sg[sg_idx].base + offset;
1314         } else {
1315             addr = (hwaddr)(uintptr_t)sg->iov.iov[sg_idx].iov_base + offset;
1316         }
1317 
1318         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1319             ret = nvme_addr_read(n, addr, ptr, trans_len);
1320         } else {
1321             ret = nvme_addr_write(n, addr, ptr, trans_len);
1322         }
1323 
1324         if (ret) {
1325             return NVME_DATA_TRAS_ERROR;
1326         }
1327 
1328         ptr += trans_len;
1329         len -= trans_len;
1330         count -= trans_len;
1331         offset += trans_len;
1332 
1333         if (count == 0) {
1334             count = bytes;
1335             offset += skip_bytes;
1336         }
1337     }
1338 
1339     return NVME_SUCCESS;
1340 }
1341 
1342 static uint16_t nvme_tx(NvmeCtrl *n, NvmeSg *sg, void *ptr, uint32_t len,
1343                         NvmeTxDirection dir)
1344 {
1345     assert(sg->flags & NVME_SG_ALLOC);
1346 
1347     if (sg->flags & NVME_SG_DMA) {
1348         const MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
1349         dma_addr_t residual;
1350 
1351         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1352             dma_buf_write(ptr, len, &residual, &sg->qsg, attrs);
1353         } else {
1354             dma_buf_read(ptr, len, &residual, &sg->qsg, attrs);
1355         }
1356 
1357         if (unlikely(residual)) {
1358             trace_pci_nvme_err_invalid_dma();
1359             return NVME_INVALID_FIELD | NVME_DNR;
1360         }
1361     } else {
1362         size_t bytes;
1363 
1364         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1365             bytes = qemu_iovec_to_buf(&sg->iov, 0, ptr, len);
1366         } else {
1367             bytes = qemu_iovec_from_buf(&sg->iov, 0, ptr, len);
1368         }
1369 
1370         if (unlikely(bytes != len)) {
1371             trace_pci_nvme_err_invalid_dma();
1372             return NVME_INVALID_FIELD | NVME_DNR;
1373         }
1374     }
1375 
1376     return NVME_SUCCESS;
1377 }
1378 
1379 static inline uint16_t nvme_c2h(NvmeCtrl *n, void *ptr, uint32_t len,
1380                                 NvmeRequest *req)
1381 {
1382     uint16_t status;
1383 
1384     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1385     if (status) {
1386         return status;
1387     }
1388 
1389     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_FROM_DEVICE);
1390 }
1391 
1392 static inline uint16_t nvme_h2c(NvmeCtrl *n, void *ptr, uint32_t len,
1393                                 NvmeRequest *req)
1394 {
1395     uint16_t status;
1396 
1397     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1398     if (status) {
1399         return status;
1400     }
1401 
1402     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_TO_DEVICE);
1403 }
1404 
1405 uint16_t nvme_bounce_data(NvmeCtrl *n, void *ptr, uint32_t len,
1406                           NvmeTxDirection dir, NvmeRequest *req)
1407 {
1408     NvmeNamespace *ns = req->ns;
1409     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1410     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1411     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1412 
1413     if (nvme_ns_ext(ns) &&
1414         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1415         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbasz,
1416                                    ns->lbaf.ms, 0, dir);
1417     }
1418 
1419     return nvme_tx(n, &req->sg, ptr, len, dir);
1420 }
1421 
1422 uint16_t nvme_bounce_mdata(NvmeCtrl *n, void *ptr, uint32_t len,
1423                            NvmeTxDirection dir, NvmeRequest *req)
1424 {
1425     NvmeNamespace *ns = req->ns;
1426     uint16_t status;
1427 
1428     if (nvme_ns_ext(ns)) {
1429         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbaf.ms,
1430                                    ns->lbasz, ns->lbasz, dir);
1431     }
1432 
1433     nvme_sg_unmap(&req->sg);
1434 
1435     status = nvme_map_mptr(n, &req->sg, len, &req->cmd);
1436     if (status) {
1437         return status;
1438     }
1439 
1440     return nvme_tx(n, &req->sg, ptr, len, dir);
1441 }
1442 
1443 static inline void nvme_blk_read(BlockBackend *blk, int64_t offset,
1444                                  uint32_t align, BlockCompletionFunc *cb,
1445                                  NvmeRequest *req)
1446 {
1447     assert(req->sg.flags & NVME_SG_ALLOC);
1448 
1449     if (req->sg.flags & NVME_SG_DMA) {
1450         req->aiocb = dma_blk_read(blk, &req->sg.qsg, offset, align, cb, req);
1451     } else {
1452         req->aiocb = blk_aio_preadv(blk, offset, &req->sg.iov, 0, cb, req);
1453     }
1454 }
1455 
1456 static inline void nvme_blk_write(BlockBackend *blk, int64_t offset,
1457                                   uint32_t align, BlockCompletionFunc *cb,
1458                                   NvmeRequest *req)
1459 {
1460     assert(req->sg.flags & NVME_SG_ALLOC);
1461 
1462     if (req->sg.flags & NVME_SG_DMA) {
1463         req->aiocb = dma_blk_write(blk, &req->sg.qsg, offset, align, cb, req);
1464     } else {
1465         req->aiocb = blk_aio_pwritev(blk, offset, &req->sg.iov, 0, cb, req);
1466     }
1467 }
1468 
1469 static void nvme_update_cq_eventidx(const NvmeCQueue *cq)
1470 {
1471     trace_pci_nvme_update_cq_eventidx(cq->cqid, cq->head);
1472 
1473     stl_le_pci_dma(PCI_DEVICE(cq->ctrl), cq->ei_addr, cq->head,
1474                    MEMTXATTRS_UNSPECIFIED);
1475 }
1476 
1477 static void nvme_update_cq_head(NvmeCQueue *cq)
1478 {
1479     ldl_le_pci_dma(PCI_DEVICE(cq->ctrl), cq->db_addr, &cq->head,
1480                    MEMTXATTRS_UNSPECIFIED);
1481 
1482     trace_pci_nvme_update_cq_head(cq->cqid, cq->head);
1483 }
1484 
1485 static void nvme_post_cqes(void *opaque)
1486 {
1487     NvmeCQueue *cq = opaque;
1488     NvmeCtrl *n = cq->ctrl;
1489     NvmeRequest *req, *next;
1490     bool pending = cq->head != cq->tail;
1491     int ret;
1492 
1493     QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
1494         NvmeSQueue *sq;
1495         hwaddr addr;
1496 
1497         if (n->dbbuf_enabled) {
1498             nvme_update_cq_eventidx(cq);
1499             nvme_update_cq_head(cq);
1500         }
1501 
1502         if (nvme_cq_full(cq)) {
1503             break;
1504         }
1505 
1506         sq = req->sq;
1507         req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
1508         req->cqe.sq_id = cpu_to_le16(sq->sqid);
1509         req->cqe.sq_head = cpu_to_le16(sq->head);
1510         addr = cq->dma_addr + (cq->tail << NVME_CQES);
1511         ret = pci_dma_write(PCI_DEVICE(n), addr, (void *)&req->cqe,
1512                             sizeof(req->cqe));
1513         if (ret) {
1514             trace_pci_nvme_err_addr_write(addr);
1515             trace_pci_nvme_err_cfs();
1516             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
1517             break;
1518         }
1519         QTAILQ_REMOVE(&cq->req_list, req, entry);
1520         nvme_inc_cq_tail(cq);
1521         nvme_sg_unmap(&req->sg);
1522         QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
1523     }
1524     if (cq->tail != cq->head) {
1525         if (cq->irq_enabled && !pending) {
1526             n->cq_pending++;
1527         }
1528 
1529         nvme_irq_assert(n, cq);
1530     }
1531 }
1532 
1533 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
1534 {
1535     assert(cq->cqid == req->sq->cqid);
1536     trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
1537                                           le32_to_cpu(req->cqe.result),
1538                                           le32_to_cpu(req->cqe.dw1),
1539                                           req->status);
1540 
1541     if (req->status) {
1542         trace_pci_nvme_err_req_status(nvme_cid(req), nvme_nsid(req->ns),
1543                                       req->status, req->cmd.opcode);
1544     }
1545 
1546     QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
1547     QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
1548 
1549     qemu_bh_schedule(cq->bh);
1550 }
1551 
1552 static void nvme_process_aers(void *opaque)
1553 {
1554     NvmeCtrl *n = opaque;
1555     NvmeAsyncEvent *event, *next;
1556 
1557     trace_pci_nvme_process_aers(n->aer_queued);
1558 
1559     QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
1560         NvmeRequest *req;
1561         NvmeAerResult *result;
1562 
1563         /* can't post cqe if there is nothing to complete */
1564         if (!n->outstanding_aers) {
1565             trace_pci_nvme_no_outstanding_aers();
1566             break;
1567         }
1568 
1569         /* ignore if masked (cqe posted, but event not cleared) */
1570         if (n->aer_mask & (1 << event->result.event_type)) {
1571             trace_pci_nvme_aer_masked(event->result.event_type, n->aer_mask);
1572             continue;
1573         }
1574 
1575         QTAILQ_REMOVE(&n->aer_queue, event, entry);
1576         n->aer_queued--;
1577 
1578         n->aer_mask |= 1 << event->result.event_type;
1579         n->outstanding_aers--;
1580 
1581         req = n->aer_reqs[n->outstanding_aers];
1582 
1583         result = (NvmeAerResult *) &req->cqe.result;
1584         result->event_type = event->result.event_type;
1585         result->event_info = event->result.event_info;
1586         result->log_page = event->result.log_page;
1587         g_free(event);
1588 
1589         trace_pci_nvme_aer_post_cqe(result->event_type, result->event_info,
1590                                     result->log_page);
1591 
1592         nvme_enqueue_req_completion(&n->admin_cq, req);
1593     }
1594 }
1595 
1596 static void nvme_enqueue_event(NvmeCtrl *n, uint8_t event_type,
1597                                uint8_t event_info, uint8_t log_page)
1598 {
1599     NvmeAsyncEvent *event;
1600 
1601     trace_pci_nvme_enqueue_event(event_type, event_info, log_page);
1602 
1603     if (n->aer_queued == n->params.aer_max_queued) {
1604         trace_pci_nvme_enqueue_event_noqueue(n->aer_queued);
1605         return;
1606     }
1607 
1608     event = g_new(NvmeAsyncEvent, 1);
1609     event->result = (NvmeAerResult) {
1610         .event_type = event_type,
1611         .event_info = event_info,
1612         .log_page   = log_page,
1613     };
1614 
1615     QTAILQ_INSERT_TAIL(&n->aer_queue, event, entry);
1616     n->aer_queued++;
1617 
1618     nvme_process_aers(n);
1619 }
1620 
1621 static void nvme_smart_event(NvmeCtrl *n, uint8_t event)
1622 {
1623     uint8_t aer_info;
1624 
1625     /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
1626     if (!(NVME_AEC_SMART(n->features.async_config) & event)) {
1627         return;
1628     }
1629 
1630     switch (event) {
1631     case NVME_SMART_SPARE:
1632         aer_info = NVME_AER_INFO_SMART_SPARE_THRESH;
1633         break;
1634     case NVME_SMART_TEMPERATURE:
1635         aer_info = NVME_AER_INFO_SMART_TEMP_THRESH;
1636         break;
1637     case NVME_SMART_RELIABILITY:
1638     case NVME_SMART_MEDIA_READ_ONLY:
1639     case NVME_SMART_FAILED_VOLATILE_MEDIA:
1640     case NVME_SMART_PMR_UNRELIABLE:
1641         aer_info = NVME_AER_INFO_SMART_RELIABILITY;
1642         break;
1643     default:
1644         return;
1645     }
1646 
1647     nvme_enqueue_event(n, NVME_AER_TYPE_SMART, aer_info, NVME_LOG_SMART_INFO);
1648 }
1649 
1650 static void nvme_clear_events(NvmeCtrl *n, uint8_t event_type)
1651 {
1652     n->aer_mask &= ~(1 << event_type);
1653     if (!QTAILQ_EMPTY(&n->aer_queue)) {
1654         nvme_process_aers(n);
1655     }
1656 }
1657 
1658 static inline uint16_t nvme_check_mdts(NvmeCtrl *n, size_t len)
1659 {
1660     uint8_t mdts = n->params.mdts;
1661 
1662     if (mdts && len > n->page_size << mdts) {
1663         trace_pci_nvme_err_mdts(len);
1664         return NVME_INVALID_FIELD | NVME_DNR;
1665     }
1666 
1667     return NVME_SUCCESS;
1668 }
1669 
1670 static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
1671                                          uint32_t nlb)
1672 {
1673     uint64_t nsze = le64_to_cpu(ns->id_ns.nsze);
1674 
1675     if (unlikely(UINT64_MAX - slba < nlb || slba + nlb > nsze)) {
1676         trace_pci_nvme_err_invalid_lba_range(slba, nlb, nsze);
1677         return NVME_LBA_RANGE | NVME_DNR;
1678     }
1679 
1680     return NVME_SUCCESS;
1681 }
1682 
1683 static int nvme_block_status_all(NvmeNamespace *ns, uint64_t slba,
1684                                  uint32_t nlb, int flags)
1685 {
1686     BlockDriverState *bs = blk_bs(ns->blkconf.blk);
1687 
1688     int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
1689     int64_t offset = nvme_l2b(ns, slba);
1690     int ret;
1691 
1692     /*
1693      * `pnum` holds the number of bytes after offset that shares the same
1694      * allocation status as the byte at offset. If `pnum` is different from
1695      * `bytes`, we should check the allocation status of the next range and
1696      * continue this until all bytes have been checked.
1697      */
1698     do {
1699         bytes -= pnum;
1700 
1701         ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
1702         if (ret < 0) {
1703             return ret;
1704         }
1705 
1706 
1707         trace_pci_nvme_block_status(offset, bytes, pnum, ret,
1708                                     !!(ret & BDRV_BLOCK_ZERO));
1709 
1710         if (!(ret & flags)) {
1711             return 1;
1712         }
1713 
1714         offset += pnum;
1715     } while (pnum != bytes);
1716 
1717     return 0;
1718 }
1719 
1720 static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
1721                                  uint32_t nlb)
1722 {
1723     int ret;
1724     Error *err = NULL;
1725 
1726     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_DATA);
1727     if (ret) {
1728         if (ret < 0) {
1729             error_setg_errno(&err, -ret, "unable to get block status");
1730             error_report_err(err);
1731 
1732             return NVME_INTERNAL_DEV_ERROR;
1733         }
1734 
1735         return NVME_DULB;
1736     }
1737 
1738     return NVME_SUCCESS;
1739 }
1740 
1741 static void nvme_aio_err(NvmeRequest *req, int ret)
1742 {
1743     uint16_t status = NVME_SUCCESS;
1744     Error *local_err = NULL;
1745 
1746     switch (req->cmd.opcode) {
1747     case NVME_CMD_READ:
1748         status = NVME_UNRECOVERED_READ;
1749         break;
1750     case NVME_CMD_FLUSH:
1751     case NVME_CMD_WRITE:
1752     case NVME_CMD_WRITE_ZEROES:
1753     case NVME_CMD_ZONE_APPEND:
1754     case NVME_CMD_COPY:
1755         status = NVME_WRITE_FAULT;
1756         break;
1757     default:
1758         status = NVME_INTERNAL_DEV_ERROR;
1759         break;
1760     }
1761 
1762     trace_pci_nvme_err_aio(nvme_cid(req), strerror(-ret), status);
1763 
1764     error_setg_errno(&local_err, -ret, "aio failed");
1765     error_report_err(local_err);
1766 
1767     /*
1768      * Set the command status code to the first encountered error but allow a
1769      * subsequent Internal Device Error to trump it.
1770      */
1771     if (req->status && status != NVME_INTERNAL_DEV_ERROR) {
1772         return;
1773     }
1774 
1775     req->status = status;
1776 }
1777 
1778 static inline uint32_t nvme_zone_idx(NvmeNamespace *ns, uint64_t slba)
1779 {
1780     return ns->zone_size_log2 > 0 ? slba >> ns->zone_size_log2 :
1781                                     slba / ns->zone_size;
1782 }
1783 
1784 static inline NvmeZone *nvme_get_zone_by_slba(NvmeNamespace *ns, uint64_t slba)
1785 {
1786     uint32_t zone_idx = nvme_zone_idx(ns, slba);
1787 
1788     if (zone_idx >= ns->num_zones) {
1789         return NULL;
1790     }
1791 
1792     return &ns->zone_array[zone_idx];
1793 }
1794 
1795 static uint16_t nvme_check_zone_state_for_write(NvmeZone *zone)
1796 {
1797     uint64_t zslba = zone->d.zslba;
1798 
1799     switch (nvme_get_zone_state(zone)) {
1800     case NVME_ZONE_STATE_EMPTY:
1801     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1802     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1803     case NVME_ZONE_STATE_CLOSED:
1804         return NVME_SUCCESS;
1805     case NVME_ZONE_STATE_FULL:
1806         trace_pci_nvme_err_zone_is_full(zslba);
1807         return NVME_ZONE_FULL;
1808     case NVME_ZONE_STATE_OFFLINE:
1809         trace_pci_nvme_err_zone_is_offline(zslba);
1810         return NVME_ZONE_OFFLINE;
1811     case NVME_ZONE_STATE_READ_ONLY:
1812         trace_pci_nvme_err_zone_is_read_only(zslba);
1813         return NVME_ZONE_READ_ONLY;
1814     default:
1815         assert(false);
1816     }
1817 
1818     return NVME_INTERNAL_DEV_ERROR;
1819 }
1820 
1821 static uint16_t nvme_check_zone_write(NvmeNamespace *ns, NvmeZone *zone,
1822                                       uint64_t slba, uint32_t nlb)
1823 {
1824     uint64_t zcap = nvme_zone_wr_boundary(zone);
1825     uint16_t status;
1826 
1827     status = nvme_check_zone_state_for_write(zone);
1828     if (status) {
1829         return status;
1830     }
1831 
1832     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1833         uint64_t ezrwa = zone->w_ptr + 2 * ns->zns.zrwas;
1834 
1835         if (slba < zone->w_ptr || slba + nlb > ezrwa) {
1836             trace_pci_nvme_err_zone_invalid_write(slba, zone->w_ptr);
1837             return NVME_ZONE_INVALID_WRITE;
1838         }
1839     } else {
1840         if (unlikely(slba != zone->w_ptr)) {
1841             trace_pci_nvme_err_write_not_at_wp(slba, zone->d.zslba,
1842                                                zone->w_ptr);
1843             return NVME_ZONE_INVALID_WRITE;
1844         }
1845     }
1846 
1847     if (unlikely((slba + nlb) > zcap)) {
1848         trace_pci_nvme_err_zone_boundary(slba, nlb, zcap);
1849         return NVME_ZONE_BOUNDARY_ERROR;
1850     }
1851 
1852     return NVME_SUCCESS;
1853 }
1854 
1855 static uint16_t nvme_check_zone_state_for_read(NvmeZone *zone)
1856 {
1857     switch (nvme_get_zone_state(zone)) {
1858     case NVME_ZONE_STATE_EMPTY:
1859     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1860     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1861     case NVME_ZONE_STATE_FULL:
1862     case NVME_ZONE_STATE_CLOSED:
1863     case NVME_ZONE_STATE_READ_ONLY:
1864         return NVME_SUCCESS;
1865     case NVME_ZONE_STATE_OFFLINE:
1866         trace_pci_nvme_err_zone_is_offline(zone->d.zslba);
1867         return NVME_ZONE_OFFLINE;
1868     default:
1869         assert(false);
1870     }
1871 
1872     return NVME_INTERNAL_DEV_ERROR;
1873 }
1874 
1875 static uint16_t nvme_check_zone_read(NvmeNamespace *ns, uint64_t slba,
1876                                      uint32_t nlb)
1877 {
1878     NvmeZone *zone;
1879     uint64_t bndry, end;
1880     uint16_t status;
1881 
1882     zone = nvme_get_zone_by_slba(ns, slba);
1883     assert(zone);
1884 
1885     bndry = nvme_zone_rd_boundary(ns, zone);
1886     end = slba + nlb;
1887 
1888     status = nvme_check_zone_state_for_read(zone);
1889     if (status) {
1890         ;
1891     } else if (unlikely(end > bndry)) {
1892         if (!ns->params.cross_zone_read) {
1893             status = NVME_ZONE_BOUNDARY_ERROR;
1894         } else {
1895             /*
1896              * Read across zone boundary - check that all subsequent
1897              * zones that are being read have an appropriate state.
1898              */
1899             do {
1900                 zone++;
1901                 status = nvme_check_zone_state_for_read(zone);
1902                 if (status) {
1903                     break;
1904                 }
1905             } while (end > nvme_zone_rd_boundary(ns, zone));
1906         }
1907     }
1908 
1909     return status;
1910 }
1911 
1912 static uint16_t nvme_zrm_finish(NvmeNamespace *ns, NvmeZone *zone)
1913 {
1914     switch (nvme_get_zone_state(zone)) {
1915     case NVME_ZONE_STATE_FULL:
1916         return NVME_SUCCESS;
1917 
1918     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1919     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1920         nvme_aor_dec_open(ns);
1921         /* fallthrough */
1922     case NVME_ZONE_STATE_CLOSED:
1923         nvme_aor_dec_active(ns);
1924 
1925         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1926             zone->d.za &= ~NVME_ZA_ZRWA_VALID;
1927             if (ns->params.numzrwa) {
1928                 ns->zns.numzrwa++;
1929             }
1930         }
1931 
1932         /* fallthrough */
1933     case NVME_ZONE_STATE_EMPTY:
1934         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_FULL);
1935         return NVME_SUCCESS;
1936 
1937     default:
1938         return NVME_ZONE_INVAL_TRANSITION;
1939     }
1940 }
1941 
1942 static uint16_t nvme_zrm_close(NvmeNamespace *ns, NvmeZone *zone)
1943 {
1944     switch (nvme_get_zone_state(zone)) {
1945     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1946     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1947         nvme_aor_dec_open(ns);
1948         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
1949         /* fall through */
1950     case NVME_ZONE_STATE_CLOSED:
1951         return NVME_SUCCESS;
1952 
1953     default:
1954         return NVME_ZONE_INVAL_TRANSITION;
1955     }
1956 }
1957 
1958 static uint16_t nvme_zrm_reset(NvmeNamespace *ns, NvmeZone *zone)
1959 {
1960     switch (nvme_get_zone_state(zone)) {
1961     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
1962     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
1963         nvme_aor_dec_open(ns);
1964         /* fallthrough */
1965     case NVME_ZONE_STATE_CLOSED:
1966         nvme_aor_dec_active(ns);
1967 
1968         if (zone->d.za & NVME_ZA_ZRWA_VALID) {
1969             if (ns->params.numzrwa) {
1970                 ns->zns.numzrwa++;
1971             }
1972         }
1973 
1974         /* fallthrough */
1975     case NVME_ZONE_STATE_FULL:
1976         zone->w_ptr = zone->d.zslba;
1977         zone->d.wp = zone->w_ptr;
1978         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
1979         /* fallthrough */
1980     case NVME_ZONE_STATE_EMPTY:
1981         return NVME_SUCCESS;
1982 
1983     default:
1984         return NVME_ZONE_INVAL_TRANSITION;
1985     }
1986 }
1987 
1988 static void nvme_zrm_auto_transition_zone(NvmeNamespace *ns)
1989 {
1990     NvmeZone *zone;
1991 
1992     if (ns->params.max_open_zones &&
1993         ns->nr_open_zones == ns->params.max_open_zones) {
1994         zone = QTAILQ_FIRST(&ns->imp_open_zones);
1995         if (zone) {
1996             /*
1997              * Automatically close this implicitly open zone.
1998              */
1999             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
2000             nvme_zrm_close(ns, zone);
2001         }
2002     }
2003 }
2004 
2005 enum {
2006     NVME_ZRM_AUTO = 1 << 0,
2007     NVME_ZRM_ZRWA = 1 << 1,
2008 };
2009 
2010 static uint16_t nvme_zrm_open_flags(NvmeCtrl *n, NvmeNamespace *ns,
2011                                     NvmeZone *zone, int flags)
2012 {
2013     int act = 0;
2014     uint16_t status;
2015 
2016     switch (nvme_get_zone_state(zone)) {
2017     case NVME_ZONE_STATE_EMPTY:
2018         act = 1;
2019 
2020         /* fallthrough */
2021 
2022     case NVME_ZONE_STATE_CLOSED:
2023         if (n->params.auto_transition_zones) {
2024             nvme_zrm_auto_transition_zone(ns);
2025         }
2026         status = nvme_zns_check_resources(ns, act, 1,
2027                                           (flags & NVME_ZRM_ZRWA) ? 1 : 0);
2028         if (status) {
2029             return status;
2030         }
2031 
2032         if (act) {
2033             nvme_aor_inc_active(ns);
2034         }
2035 
2036         nvme_aor_inc_open(ns);
2037 
2038         if (flags & NVME_ZRM_AUTO) {
2039             nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_IMPLICITLY_OPEN);
2040             return NVME_SUCCESS;
2041         }
2042 
2043         /* fallthrough */
2044 
2045     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
2046         if (flags & NVME_ZRM_AUTO) {
2047             return NVME_SUCCESS;
2048         }
2049 
2050         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EXPLICITLY_OPEN);
2051 
2052         /* fallthrough */
2053 
2054     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
2055         if (flags & NVME_ZRM_ZRWA) {
2056             ns->zns.numzrwa--;
2057 
2058             zone->d.za |= NVME_ZA_ZRWA_VALID;
2059         }
2060 
2061         return NVME_SUCCESS;
2062 
2063     default:
2064         return NVME_ZONE_INVAL_TRANSITION;
2065     }
2066 }
2067 
2068 static inline uint16_t nvme_zrm_auto(NvmeCtrl *n, NvmeNamespace *ns,
2069                                      NvmeZone *zone)
2070 {
2071     return nvme_zrm_open_flags(n, ns, zone, NVME_ZRM_AUTO);
2072 }
2073 
2074 static void nvme_advance_zone_wp(NvmeNamespace *ns, NvmeZone *zone,
2075                                  uint32_t nlb)
2076 {
2077     zone->d.wp += nlb;
2078 
2079     if (zone->d.wp == nvme_zone_wr_boundary(zone)) {
2080         nvme_zrm_finish(ns, zone);
2081     }
2082 }
2083 
2084 static void nvme_zoned_zrwa_implicit_flush(NvmeNamespace *ns, NvmeZone *zone,
2085                                            uint32_t nlbc)
2086 {
2087     uint16_t nzrwafgs = DIV_ROUND_UP(nlbc, ns->zns.zrwafg);
2088 
2089     nlbc = nzrwafgs * ns->zns.zrwafg;
2090 
2091     trace_pci_nvme_zoned_zrwa_implicit_flush(zone->d.zslba, nlbc);
2092 
2093     zone->w_ptr += nlbc;
2094 
2095     nvme_advance_zone_wp(ns, zone, nlbc);
2096 }
2097 
2098 static void nvme_finalize_zoned_write(NvmeNamespace *ns, NvmeRequest *req)
2099 {
2100     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2101     NvmeZone *zone;
2102     uint64_t slba;
2103     uint32_t nlb;
2104 
2105     slba = le64_to_cpu(rw->slba);
2106     nlb = le16_to_cpu(rw->nlb) + 1;
2107     zone = nvme_get_zone_by_slba(ns, slba);
2108     assert(zone);
2109 
2110     if (zone->d.za & NVME_ZA_ZRWA_VALID) {
2111         uint64_t ezrwa = zone->w_ptr + ns->zns.zrwas - 1;
2112         uint64_t elba = slba + nlb - 1;
2113 
2114         if (elba > ezrwa) {
2115             nvme_zoned_zrwa_implicit_flush(ns, zone, elba - ezrwa);
2116         }
2117 
2118         return;
2119     }
2120 
2121     nvme_advance_zone_wp(ns, zone, nlb);
2122 }
2123 
2124 static inline bool nvme_is_write(NvmeRequest *req)
2125 {
2126     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2127 
2128     return rw->opcode == NVME_CMD_WRITE ||
2129            rw->opcode == NVME_CMD_ZONE_APPEND ||
2130            rw->opcode == NVME_CMD_WRITE_ZEROES;
2131 }
2132 
2133 static AioContext *nvme_get_aio_context(BlockAIOCB *acb)
2134 {
2135     return qemu_get_aio_context();
2136 }
2137 
2138 static void nvme_misc_cb(void *opaque, int ret)
2139 {
2140     NvmeRequest *req = opaque;
2141 
2142     trace_pci_nvme_misc_cb(nvme_cid(req));
2143 
2144     if (ret) {
2145         nvme_aio_err(req, ret);
2146     }
2147 
2148     nvme_enqueue_req_completion(nvme_cq(req), req);
2149 }
2150 
2151 void nvme_rw_complete_cb(void *opaque, int ret)
2152 {
2153     NvmeRequest *req = opaque;
2154     NvmeNamespace *ns = req->ns;
2155     BlockBackend *blk = ns->blkconf.blk;
2156     BlockAcctCookie *acct = &req->acct;
2157     BlockAcctStats *stats = blk_get_stats(blk);
2158 
2159     trace_pci_nvme_rw_complete_cb(nvme_cid(req), blk_name(blk));
2160 
2161     if (ret) {
2162         block_acct_failed(stats, acct);
2163         nvme_aio_err(req, ret);
2164     } else {
2165         block_acct_done(stats, acct);
2166     }
2167 
2168     if (ns->params.zoned && nvme_is_write(req)) {
2169         nvme_finalize_zoned_write(ns, req);
2170     }
2171 
2172     nvme_enqueue_req_completion(nvme_cq(req), req);
2173 }
2174 
2175 static void nvme_rw_cb(void *opaque, int ret)
2176 {
2177     NvmeRequest *req = opaque;
2178     NvmeNamespace *ns = req->ns;
2179 
2180     BlockBackend *blk = ns->blkconf.blk;
2181 
2182     trace_pci_nvme_rw_cb(nvme_cid(req), blk_name(blk));
2183 
2184     if (ret) {
2185         goto out;
2186     }
2187 
2188     if (ns->lbaf.ms) {
2189         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2190         uint64_t slba = le64_to_cpu(rw->slba);
2191         uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
2192         uint64_t offset = nvme_moff(ns, slba);
2193 
2194         if (req->cmd.opcode == NVME_CMD_WRITE_ZEROES) {
2195             size_t mlen = nvme_m2b(ns, nlb);
2196 
2197             req->aiocb = blk_aio_pwrite_zeroes(blk, offset, mlen,
2198                                                BDRV_REQ_MAY_UNMAP,
2199                                                nvme_rw_complete_cb, req);
2200             return;
2201         }
2202 
2203         if (nvme_ns_ext(ns) || req->cmd.mptr) {
2204             uint16_t status;
2205 
2206             nvme_sg_unmap(&req->sg);
2207             status = nvme_map_mdata(nvme_ctrl(req), nlb, req);
2208             if (status) {
2209                 ret = -EFAULT;
2210                 goto out;
2211             }
2212 
2213             if (req->cmd.opcode == NVME_CMD_READ) {
2214                 return nvme_blk_read(blk, offset, 1, nvme_rw_complete_cb, req);
2215             }
2216 
2217             return nvme_blk_write(blk, offset, 1, nvme_rw_complete_cb, req);
2218         }
2219     }
2220 
2221 out:
2222     nvme_rw_complete_cb(req, ret);
2223 }
2224 
2225 static void nvme_verify_cb(void *opaque, int ret)
2226 {
2227     NvmeBounceContext *ctx = opaque;
2228     NvmeRequest *req = ctx->req;
2229     NvmeNamespace *ns = req->ns;
2230     BlockBackend *blk = ns->blkconf.blk;
2231     BlockAcctCookie *acct = &req->acct;
2232     BlockAcctStats *stats = blk_get_stats(blk);
2233     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2234     uint64_t slba = le64_to_cpu(rw->slba);
2235     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2236     uint16_t apptag = le16_to_cpu(rw->apptag);
2237     uint16_t appmask = le16_to_cpu(rw->appmask);
2238     uint64_t reftag = le32_to_cpu(rw->reftag);
2239     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2240     uint16_t status;
2241 
2242     reftag |= cdw3 << 32;
2243 
2244     trace_pci_nvme_verify_cb(nvme_cid(req), prinfo, apptag, appmask, reftag);
2245 
2246     if (ret) {
2247         block_acct_failed(stats, acct);
2248         nvme_aio_err(req, ret);
2249         goto out;
2250     }
2251 
2252     block_acct_done(stats, acct);
2253 
2254     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2255         status = nvme_dif_mangle_mdata(ns, ctx->mdata.bounce,
2256                                        ctx->mdata.iov.size, slba);
2257         if (status) {
2258             req->status = status;
2259             goto out;
2260         }
2261 
2262         req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2263                                      ctx->mdata.bounce, ctx->mdata.iov.size,
2264                                      prinfo, slba, apptag, appmask, &reftag);
2265     }
2266 
2267 out:
2268     qemu_iovec_destroy(&ctx->data.iov);
2269     g_free(ctx->data.bounce);
2270 
2271     qemu_iovec_destroy(&ctx->mdata.iov);
2272     g_free(ctx->mdata.bounce);
2273 
2274     g_free(ctx);
2275 
2276     nvme_enqueue_req_completion(nvme_cq(req), req);
2277 }
2278 
2279 
2280 static void nvme_verify_mdata_in_cb(void *opaque, int ret)
2281 {
2282     NvmeBounceContext *ctx = opaque;
2283     NvmeRequest *req = ctx->req;
2284     NvmeNamespace *ns = req->ns;
2285     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2286     uint64_t slba = le64_to_cpu(rw->slba);
2287     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2288     size_t mlen = nvme_m2b(ns, nlb);
2289     uint64_t offset = nvme_moff(ns, slba);
2290     BlockBackend *blk = ns->blkconf.blk;
2291 
2292     trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req), blk_name(blk));
2293 
2294     if (ret) {
2295         goto out;
2296     }
2297 
2298     ctx->mdata.bounce = g_malloc(mlen);
2299 
2300     qemu_iovec_reset(&ctx->mdata.iov);
2301     qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2302 
2303     req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2304                                 nvme_verify_cb, ctx);
2305     return;
2306 
2307 out:
2308     nvme_verify_cb(ctx, ret);
2309 }
2310 
2311 struct nvme_compare_ctx {
2312     struct {
2313         QEMUIOVector iov;
2314         uint8_t *bounce;
2315     } data;
2316 
2317     struct {
2318         QEMUIOVector iov;
2319         uint8_t *bounce;
2320     } mdata;
2321 };
2322 
2323 static void nvme_compare_mdata_cb(void *opaque, int ret)
2324 {
2325     NvmeRequest *req = opaque;
2326     NvmeNamespace *ns = req->ns;
2327     NvmeCtrl *n = nvme_ctrl(req);
2328     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2329     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2330     uint16_t apptag = le16_to_cpu(rw->apptag);
2331     uint16_t appmask = le16_to_cpu(rw->appmask);
2332     uint64_t reftag = le32_to_cpu(rw->reftag);
2333     uint64_t cdw3 = le32_to_cpu(rw->cdw3);
2334     struct nvme_compare_ctx *ctx = req->opaque;
2335     g_autofree uint8_t *buf = NULL;
2336     BlockBackend *blk = ns->blkconf.blk;
2337     BlockAcctCookie *acct = &req->acct;
2338     BlockAcctStats *stats = blk_get_stats(blk);
2339     uint16_t status = NVME_SUCCESS;
2340 
2341     reftag |= cdw3 << 32;
2342 
2343     trace_pci_nvme_compare_mdata_cb(nvme_cid(req));
2344 
2345     if (ret) {
2346         block_acct_failed(stats, acct);
2347         nvme_aio_err(req, ret);
2348         goto out;
2349     }
2350 
2351     buf = g_malloc(ctx->mdata.iov.size);
2352 
2353     status = nvme_bounce_mdata(n, buf, ctx->mdata.iov.size,
2354                                NVME_TX_DIRECTION_TO_DEVICE, req);
2355     if (status) {
2356         req->status = status;
2357         goto out;
2358     }
2359 
2360     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2361         uint64_t slba = le64_to_cpu(rw->slba);
2362         uint8_t *bufp;
2363         uint8_t *mbufp = ctx->mdata.bounce;
2364         uint8_t *end = mbufp + ctx->mdata.iov.size;
2365         int16_t pil = 0;
2366 
2367         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
2368                                 ctx->mdata.bounce, ctx->mdata.iov.size, prinfo,
2369                                 slba, apptag, appmask, &reftag);
2370         if (status) {
2371             req->status = status;
2372             goto out;
2373         }
2374 
2375         /*
2376          * When formatted with protection information, do not compare the DIF
2377          * tuple.
2378          */
2379         if (!(ns->id_ns.dps & NVME_ID_NS_DPS_FIRST_EIGHT)) {
2380             pil = ns->lbaf.ms - nvme_pi_tuple_size(ns);
2381         }
2382 
2383         for (bufp = buf; mbufp < end; bufp += ns->lbaf.ms, mbufp += ns->lbaf.ms) {
2384             if (memcmp(bufp + pil, mbufp + pil, ns->lbaf.ms - pil)) {
2385                 req->status = NVME_CMP_FAILURE | NVME_DNR;
2386                 goto out;
2387             }
2388         }
2389 
2390         goto out;
2391     }
2392 
2393     if (memcmp(buf, ctx->mdata.bounce, ctx->mdata.iov.size)) {
2394         req->status = NVME_CMP_FAILURE | NVME_DNR;
2395         goto out;
2396     }
2397 
2398     block_acct_done(stats, acct);
2399 
2400 out:
2401     qemu_iovec_destroy(&ctx->data.iov);
2402     g_free(ctx->data.bounce);
2403 
2404     qemu_iovec_destroy(&ctx->mdata.iov);
2405     g_free(ctx->mdata.bounce);
2406 
2407     g_free(ctx);
2408 
2409     nvme_enqueue_req_completion(nvme_cq(req), req);
2410 }
2411 
2412 static void nvme_compare_data_cb(void *opaque, int ret)
2413 {
2414     NvmeRequest *req = opaque;
2415     NvmeCtrl *n = nvme_ctrl(req);
2416     NvmeNamespace *ns = req->ns;
2417     BlockBackend *blk = ns->blkconf.blk;
2418     BlockAcctCookie *acct = &req->acct;
2419     BlockAcctStats *stats = blk_get_stats(blk);
2420 
2421     struct nvme_compare_ctx *ctx = req->opaque;
2422     g_autofree uint8_t *buf = NULL;
2423     uint16_t status;
2424 
2425     trace_pci_nvme_compare_data_cb(nvme_cid(req));
2426 
2427     if (ret) {
2428         block_acct_failed(stats, acct);
2429         nvme_aio_err(req, ret);
2430         goto out;
2431     }
2432 
2433     buf = g_malloc(ctx->data.iov.size);
2434 
2435     status = nvme_bounce_data(n, buf, ctx->data.iov.size,
2436                               NVME_TX_DIRECTION_TO_DEVICE, req);
2437     if (status) {
2438         req->status = status;
2439         goto out;
2440     }
2441 
2442     if (memcmp(buf, ctx->data.bounce, ctx->data.iov.size)) {
2443         req->status = NVME_CMP_FAILURE | NVME_DNR;
2444         goto out;
2445     }
2446 
2447     if (ns->lbaf.ms) {
2448         NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2449         uint64_t slba = le64_to_cpu(rw->slba);
2450         uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2451         size_t mlen = nvme_m2b(ns, nlb);
2452         uint64_t offset = nvme_moff(ns, slba);
2453 
2454         ctx->mdata.bounce = g_malloc(mlen);
2455 
2456         qemu_iovec_init(&ctx->mdata.iov, 1);
2457         qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);
2458 
2459         req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
2460                                     nvme_compare_mdata_cb, req);
2461         return;
2462     }
2463 
2464     block_acct_done(stats, acct);
2465 
2466 out:
2467     qemu_iovec_destroy(&ctx->data.iov);
2468     g_free(ctx->data.bounce);
2469     g_free(ctx);
2470 
2471     nvme_enqueue_req_completion(nvme_cq(req), req);
2472 }
2473 
2474 typedef struct NvmeDSMAIOCB {
2475     BlockAIOCB common;
2476     BlockAIOCB *aiocb;
2477     NvmeRequest *req;
2478     int ret;
2479 
2480     NvmeDsmRange *range;
2481     unsigned int nr;
2482     unsigned int idx;
2483 } NvmeDSMAIOCB;
2484 
2485 static void nvme_dsm_cancel(BlockAIOCB *aiocb)
2486 {
2487     NvmeDSMAIOCB *iocb = container_of(aiocb, NvmeDSMAIOCB, common);
2488 
2489     /* break nvme_dsm_cb loop */
2490     iocb->idx = iocb->nr;
2491     iocb->ret = -ECANCELED;
2492 
2493     if (iocb->aiocb) {
2494         blk_aio_cancel_async(iocb->aiocb);
2495         iocb->aiocb = NULL;
2496     } else {
2497         /*
2498          * We only reach this if nvme_dsm_cancel() has already been called or
2499          * the command ran to completion.
2500          */
2501         assert(iocb->idx == iocb->nr);
2502     }
2503 }
2504 
2505 static const AIOCBInfo nvme_dsm_aiocb_info = {
2506     .aiocb_size   = sizeof(NvmeDSMAIOCB),
2507     .cancel_async = nvme_dsm_cancel,
2508 };
2509 
2510 static void nvme_dsm_cb(void *opaque, int ret);
2511 
2512 static void nvme_dsm_md_cb(void *opaque, int ret)
2513 {
2514     NvmeDSMAIOCB *iocb = opaque;
2515     NvmeRequest *req = iocb->req;
2516     NvmeNamespace *ns = req->ns;
2517     NvmeDsmRange *range;
2518     uint64_t slba;
2519     uint32_t nlb;
2520 
2521     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
2522         goto done;
2523     }
2524 
2525     range = &iocb->range[iocb->idx - 1];
2526     slba = le64_to_cpu(range->slba);
2527     nlb = le32_to_cpu(range->nlb);
2528 
2529     /*
2530      * Check that all block were discarded (zeroed); otherwise we do not zero
2531      * the metadata.
2532      */
2533 
2534     ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_ZERO);
2535     if (ret) {
2536         if (ret < 0) {
2537             goto done;
2538         }
2539 
2540         nvme_dsm_cb(iocb, 0);
2541         return;
2542     }
2543 
2544     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_moff(ns, slba),
2545                                         nvme_m2b(ns, nlb), BDRV_REQ_MAY_UNMAP,
2546                                         nvme_dsm_cb, iocb);
2547     return;
2548 
2549 done:
2550     nvme_dsm_cb(iocb, ret);
2551 }
2552 
2553 static void nvme_dsm_cb(void *opaque, int ret)
2554 {
2555     NvmeDSMAIOCB *iocb = opaque;
2556     NvmeRequest *req = iocb->req;
2557     NvmeCtrl *n = nvme_ctrl(req);
2558     NvmeNamespace *ns = req->ns;
2559     NvmeDsmRange *range;
2560     uint64_t slba;
2561     uint32_t nlb;
2562 
2563     if (iocb->ret < 0) {
2564         goto done;
2565     } else if (ret < 0) {
2566         iocb->ret = ret;
2567         goto done;
2568     }
2569 
2570 next:
2571     if (iocb->idx == iocb->nr) {
2572         goto done;
2573     }
2574 
2575     range = &iocb->range[iocb->idx++];
2576     slba = le64_to_cpu(range->slba);
2577     nlb = le32_to_cpu(range->nlb);
2578 
2579     trace_pci_nvme_dsm_deallocate(slba, nlb);
2580 
2581     if (nlb > n->dmrsl) {
2582         trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
2583         goto next;
2584     }
2585 
2586     if (nvme_check_bounds(ns, slba, nlb)) {
2587         trace_pci_nvme_err_invalid_lba_range(slba, nlb,
2588                                              ns->id_ns.nsze);
2589         goto next;
2590     }
2591 
2592     iocb->aiocb = blk_aio_pdiscard(ns->blkconf.blk, nvme_l2b(ns, slba),
2593                                    nvme_l2b(ns, nlb),
2594                                    nvme_dsm_md_cb, iocb);
2595     return;
2596 
2597 done:
2598     iocb->aiocb = NULL;
2599     iocb->common.cb(iocb->common.opaque, iocb->ret);
2600     qemu_aio_unref(iocb);
2601 }
2602 
2603 static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
2604 {
2605     NvmeNamespace *ns = req->ns;
2606     NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
2607     uint32_t attr = le32_to_cpu(dsm->attributes);
2608     uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
2609     uint16_t status = NVME_SUCCESS;
2610 
2611     trace_pci_nvme_dsm(nr, attr);
2612 
2613     if (attr & NVME_DSMGMT_AD) {
2614         NvmeDSMAIOCB *iocb = blk_aio_get(&nvme_dsm_aiocb_info, ns->blkconf.blk,
2615                                          nvme_misc_cb, req);
2616 
2617         iocb->req = req;
2618         iocb->ret = 0;
2619         iocb->range = g_new(NvmeDsmRange, nr);
2620         iocb->nr = nr;
2621         iocb->idx = 0;
2622 
2623         status = nvme_h2c(n, (uint8_t *)iocb->range, sizeof(NvmeDsmRange) * nr,
2624                           req);
2625         if (status) {
2626             g_free(iocb->range);
2627             qemu_aio_unref(iocb);
2628 
2629             return status;
2630         }
2631 
2632         req->aiocb = &iocb->common;
2633         nvme_dsm_cb(iocb, 0);
2634 
2635         return NVME_NO_COMPLETE;
2636     }
2637 
2638     return status;
2639 }
2640 
2641 static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
2642 {
2643     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
2644     NvmeNamespace *ns = req->ns;
2645     BlockBackend *blk = ns->blkconf.blk;
2646     uint64_t slba = le64_to_cpu(rw->slba);
2647     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
2648     size_t len = nvme_l2b(ns, nlb);
2649     int64_t offset = nvme_l2b(ns, slba);
2650     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
2651     uint32_t reftag = le32_to_cpu(rw->reftag);
2652     NvmeBounceContext *ctx = NULL;
2653     uint16_t status;
2654 
2655     trace_pci_nvme_verify(nvme_cid(req), nvme_nsid(ns), slba, nlb);
2656 
2657     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2658         status = nvme_check_prinfo(ns, prinfo, slba, reftag);
2659         if (status) {
2660             return status;
2661         }
2662 
2663         if (prinfo & NVME_PRINFO_PRACT) {
2664             return NVME_INVALID_PROT_INFO | NVME_DNR;
2665         }
2666     }
2667 
2668     if (len > n->page_size << n->params.vsl) {
2669         return NVME_INVALID_FIELD | NVME_DNR;
2670     }
2671 
2672     status = nvme_check_bounds(ns, slba, nlb);
2673     if (status) {
2674         return status;
2675     }
2676 
2677     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
2678         status = nvme_check_dulbe(ns, slba, nlb);
2679         if (status) {
2680             return status;
2681         }
2682     }
2683 
2684     ctx = g_new0(NvmeBounceContext, 1);
2685     ctx->req = req;
2686 
2687     ctx->data.bounce = g_malloc(len);
2688 
2689     qemu_iovec_init(&ctx->data.iov, 1);
2690     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, len);
2691 
2692     block_acct_start(blk_get_stats(blk), &req->acct, ctx->data.iov.size,
2693                      BLOCK_ACCT_READ);
2694 
2695     req->aiocb = blk_aio_preadv(ns->blkconf.blk, offset, &ctx->data.iov, 0,
2696                                 nvme_verify_mdata_in_cb, ctx);
2697     return NVME_NO_COMPLETE;
2698 }
2699 
2700 typedef struct NvmeCopyAIOCB {
2701     BlockAIOCB common;
2702     BlockAIOCB *aiocb;
2703     NvmeRequest *req;
2704     int ret;
2705 
2706     void *ranges;
2707     unsigned int format;
2708     int nr;
2709     int idx;
2710 
2711     uint8_t *bounce;
2712     QEMUIOVector iov;
2713     struct {
2714         BlockAcctCookie read;
2715         BlockAcctCookie write;
2716     } acct;
2717 
2718     uint64_t reftag;
2719     uint64_t slba;
2720 
2721     NvmeZone *zone;
2722 } NvmeCopyAIOCB;
2723 
2724 static void nvme_copy_cancel(BlockAIOCB *aiocb)
2725 {
2726     NvmeCopyAIOCB *iocb = container_of(aiocb, NvmeCopyAIOCB, common);
2727 
2728     iocb->ret = -ECANCELED;
2729 
2730     if (iocb->aiocb) {
2731         blk_aio_cancel_async(iocb->aiocb);
2732         iocb->aiocb = NULL;
2733     }
2734 }
2735 
2736 static const AIOCBInfo nvme_copy_aiocb_info = {
2737     .aiocb_size   = sizeof(NvmeCopyAIOCB),
2738     .cancel_async = nvme_copy_cancel,
2739 };
2740 
2741 static void nvme_copy_done(NvmeCopyAIOCB *iocb)
2742 {
2743     NvmeRequest *req = iocb->req;
2744     NvmeNamespace *ns = req->ns;
2745     BlockAcctStats *stats = blk_get_stats(ns->blkconf.blk);
2746 
2747     if (iocb->idx != iocb->nr) {
2748         req->cqe.result = cpu_to_le32(iocb->idx);
2749     }
2750 
2751     qemu_iovec_destroy(&iocb->iov);
2752     g_free(iocb->bounce);
2753 
2754     if (iocb->ret < 0) {
2755         block_acct_failed(stats, &iocb->acct.read);
2756         block_acct_failed(stats, &iocb->acct.write);
2757     } else {
2758         block_acct_done(stats, &iocb->acct.read);
2759         block_acct_done(stats, &iocb->acct.write);
2760     }
2761 
2762     iocb->common.cb(iocb->common.opaque, iocb->ret);
2763     qemu_aio_unref(iocb);
2764 }
2765 
2766 static void nvme_do_copy(NvmeCopyAIOCB *iocb);
2767 
2768 static void nvme_copy_source_range_parse_format0(void *ranges, int idx,
2769                                                  uint64_t *slba, uint32_t *nlb,
2770                                                  uint16_t *apptag,
2771                                                  uint16_t *appmask,
2772                                                  uint64_t *reftag)
2773 {
2774     NvmeCopySourceRangeFormat0 *_ranges = ranges;
2775 
2776     if (slba) {
2777         *slba = le64_to_cpu(_ranges[idx].slba);
2778     }
2779 
2780     if (nlb) {
2781         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2782     }
2783 
2784     if (apptag) {
2785         *apptag = le16_to_cpu(_ranges[idx].apptag);
2786     }
2787 
2788     if (appmask) {
2789         *appmask = le16_to_cpu(_ranges[idx].appmask);
2790     }
2791 
2792     if (reftag) {
2793         *reftag = le32_to_cpu(_ranges[idx].reftag);
2794     }
2795 }
2796 
2797 static void nvme_copy_source_range_parse_format1(void *ranges, int idx,
2798                                                  uint64_t *slba, uint32_t *nlb,
2799                                                  uint16_t *apptag,
2800                                                  uint16_t *appmask,
2801                                                  uint64_t *reftag)
2802 {
2803     NvmeCopySourceRangeFormat1 *_ranges = ranges;
2804 
2805     if (slba) {
2806         *slba = le64_to_cpu(_ranges[idx].slba);
2807     }
2808 
2809     if (nlb) {
2810         *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
2811     }
2812 
2813     if (apptag) {
2814         *apptag = le16_to_cpu(_ranges[idx].apptag);
2815     }
2816 
2817     if (appmask) {
2818         *appmask = le16_to_cpu(_ranges[idx].appmask);
2819     }
2820 
2821     if (reftag) {
2822         *reftag = 0;
2823 
2824         *reftag |= (uint64_t)_ranges[idx].sr[4] << 40;
2825         *reftag |= (uint64_t)_ranges[idx].sr[5] << 32;
2826         *reftag |= (uint64_t)_ranges[idx].sr[6] << 24;
2827         *reftag |= (uint64_t)_ranges[idx].sr[7] << 16;
2828         *reftag |= (uint64_t)_ranges[idx].sr[8] << 8;
2829         *reftag |= (uint64_t)_ranges[idx].sr[9];
2830     }
2831 }
2832 
2833 static void nvme_copy_source_range_parse(void *ranges, int idx, uint8_t format,
2834                                          uint64_t *slba, uint32_t *nlb,
2835                                          uint16_t *apptag, uint16_t *appmask,
2836                                          uint64_t *reftag)
2837 {
2838     switch (format) {
2839     case NVME_COPY_FORMAT_0:
2840         nvme_copy_source_range_parse_format0(ranges, idx, slba, nlb, apptag,
2841                                              appmask, reftag);
2842         break;
2843 
2844     case NVME_COPY_FORMAT_1:
2845         nvme_copy_source_range_parse_format1(ranges, idx, slba, nlb, apptag,
2846                                              appmask, reftag);
2847         break;
2848 
2849     default:
2850         abort();
2851     }
2852 }
2853 
2854 static inline uint16_t nvme_check_copy_mcl(NvmeNamespace *ns,
2855                                            NvmeCopyAIOCB *iocb, uint16_t nr)
2856 {
2857     uint32_t copy_len = 0;
2858 
2859     for (int idx = 0; idx < nr; idx++) {
2860         uint32_t nlb;
2861         nvme_copy_source_range_parse(iocb->ranges, idx, iocb->format, NULL,
2862                                      &nlb, NULL, NULL, NULL);
2863         copy_len += nlb + 1;
2864     }
2865 
2866     if (copy_len > ns->id_ns.mcl) {
2867         return NVME_CMD_SIZE_LIMIT | NVME_DNR;
2868     }
2869 
2870     return NVME_SUCCESS;
2871 }
2872 
2873 static void nvme_copy_out_completed_cb(void *opaque, int ret)
2874 {
2875     NvmeCopyAIOCB *iocb = opaque;
2876     NvmeRequest *req = iocb->req;
2877     NvmeNamespace *ns = req->ns;
2878     uint32_t nlb;
2879 
2880     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2881                                  &nlb, NULL, NULL, NULL);
2882 
2883     if (ret < 0) {
2884         iocb->ret = ret;
2885         goto out;
2886     } else if (iocb->ret < 0) {
2887         goto out;
2888     }
2889 
2890     if (ns->params.zoned) {
2891         nvme_advance_zone_wp(ns, iocb->zone, nlb);
2892     }
2893 
2894     iocb->idx++;
2895     iocb->slba += nlb;
2896 out:
2897     nvme_do_copy(iocb);
2898 }
2899 
2900 static void nvme_copy_out_cb(void *opaque, int ret)
2901 {
2902     NvmeCopyAIOCB *iocb = opaque;
2903     NvmeRequest *req = iocb->req;
2904     NvmeNamespace *ns = req->ns;
2905     uint32_t nlb;
2906     size_t mlen;
2907     uint8_t *mbounce;
2908 
2909     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
2910         goto out;
2911     }
2912 
2913     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
2914                                  &nlb, NULL, NULL, NULL);
2915 
2916     mlen = nvme_m2b(ns, nlb);
2917     mbounce = iocb->bounce + nvme_l2b(ns, nlb);
2918 
2919     qemu_iovec_reset(&iocb->iov);
2920     qemu_iovec_add(&iocb->iov, mbounce, mlen);
2921 
2922     iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_moff(ns, iocb->slba),
2923                                   &iocb->iov, 0, nvme_copy_out_completed_cb,
2924                                   iocb);
2925 
2926     return;
2927 
2928 out:
2929     nvme_copy_out_completed_cb(iocb, ret);
2930 }
2931 
2932 static void nvme_copy_in_completed_cb(void *opaque, int ret)
2933 {
2934     NvmeCopyAIOCB *iocb = opaque;
2935     NvmeRequest *req = iocb->req;
2936     NvmeNamespace *ns = req->ns;
2937     uint32_t nlb;
2938     uint64_t slba;
2939     uint16_t apptag, appmask;
2940     uint64_t reftag;
2941     size_t len;
2942     uint16_t status;
2943 
2944     if (ret < 0) {
2945         iocb->ret = ret;
2946         goto out;
2947     } else if (iocb->ret < 0) {
2948         goto out;
2949     }
2950 
2951     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
2952                                  &nlb, &apptag, &appmask, &reftag);
2953     len = nvme_l2b(ns, nlb);
2954 
2955     trace_pci_nvme_copy_out(iocb->slba, nlb);
2956 
2957     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
2958         NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
2959 
2960         uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
2961         uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
2962 
2963         size_t mlen = nvme_m2b(ns, nlb);
2964         uint8_t *mbounce = iocb->bounce + nvme_l2b(ns, nlb);
2965 
2966         status = nvme_dif_mangle_mdata(ns, mbounce, mlen, slba);
2967         if (status) {
2968             goto invalid;
2969         }
2970         status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen, prinfor,
2971                                 slba, apptag, appmask, &reftag);
2972         if (status) {
2973             goto invalid;
2974         }
2975 
2976         apptag = le16_to_cpu(copy->apptag);
2977         appmask = le16_to_cpu(copy->appmask);
2978 
2979         if (prinfow & NVME_PRINFO_PRACT) {
2980             status = nvme_check_prinfo(ns, prinfow, iocb->slba, iocb->reftag);
2981             if (status) {
2982                 goto invalid;
2983             }
2984 
2985             nvme_dif_pract_generate_dif(ns, iocb->bounce, len, mbounce, mlen,
2986                                         apptag, &iocb->reftag);
2987         } else {
2988             status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen,
2989                                     prinfow, iocb->slba, apptag, appmask,
2990                                     &iocb->reftag);
2991             if (status) {
2992                 goto invalid;
2993             }
2994         }
2995     }
2996 
2997     status = nvme_check_bounds(ns, iocb->slba, nlb);
2998     if (status) {
2999         goto invalid;
3000     }
3001 
3002     if (ns->params.zoned) {
3003         status = nvme_check_zone_write(ns, iocb->zone, iocb->slba, nlb);
3004         if (status) {
3005             goto invalid;
3006         }
3007 
3008         if (!(iocb->zone->d.za & NVME_ZA_ZRWA_VALID)) {
3009             iocb->zone->w_ptr += nlb;
3010         }
3011     }
3012 
3013     qemu_iovec_reset(&iocb->iov);
3014     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3015 
3016     iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, iocb->slba),
3017                                   &iocb->iov, 0, nvme_copy_out_cb, iocb);
3018 
3019     return;
3020 
3021 invalid:
3022     req->status = status;
3023     iocb->ret = -1;
3024 out:
3025     nvme_do_copy(iocb);
3026 }
3027 
3028 static void nvme_copy_in_cb(void *opaque, int ret)
3029 {
3030     NvmeCopyAIOCB *iocb = opaque;
3031     NvmeRequest *req = iocb->req;
3032     NvmeNamespace *ns = req->ns;
3033     uint64_t slba;
3034     uint32_t nlb;
3035 
3036     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
3037         goto out;
3038     }
3039 
3040     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
3041                                  &nlb, NULL, NULL, NULL);
3042 
3043     qemu_iovec_reset(&iocb->iov);
3044     qemu_iovec_add(&iocb->iov, iocb->bounce + nvme_l2b(ns, nlb),
3045                    nvme_m2b(ns, nlb));
3046 
3047     iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_moff(ns, slba),
3048                                  &iocb->iov, 0, nvme_copy_in_completed_cb,
3049                                  iocb);
3050     return;
3051 
3052 out:
3053     nvme_copy_in_completed_cb(iocb, ret);
3054 }
3055 
3056 static void nvme_do_copy(NvmeCopyAIOCB *iocb)
3057 {
3058     NvmeRequest *req = iocb->req;
3059     NvmeNamespace *ns = req->ns;
3060     uint64_t slba;
3061     uint32_t nlb;
3062     size_t len;
3063     uint16_t status;
3064 
3065     if (iocb->ret < 0) {
3066         goto done;
3067     }
3068 
3069     if (iocb->idx == iocb->nr) {
3070         goto done;
3071     }
3072 
3073     nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
3074                                  &nlb, NULL, NULL, NULL);
3075     len = nvme_l2b(ns, nlb);
3076 
3077     trace_pci_nvme_copy_source_range(slba, nlb);
3078 
3079     if (nlb > le16_to_cpu(ns->id_ns.mssrl)) {
3080         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3081         goto invalid;
3082     }
3083 
3084     status = nvme_check_bounds(ns, slba, nlb);
3085     if (status) {
3086         goto invalid;
3087     }
3088 
3089     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3090         status = nvme_check_dulbe(ns, slba, nlb);
3091         if (status) {
3092             goto invalid;
3093         }
3094     }
3095 
3096     if (ns->params.zoned) {
3097         status = nvme_check_zone_read(ns, slba, nlb);
3098         if (status) {
3099             goto invalid;
3100         }
3101     }
3102 
3103     qemu_iovec_reset(&iocb->iov);
3104     qemu_iovec_add(&iocb->iov, iocb->bounce, len);
3105 
3106     iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_l2b(ns, slba),
3107                                  &iocb->iov, 0, nvme_copy_in_cb, iocb);
3108     return;
3109 
3110 invalid:
3111     req->status = status;
3112     iocb->ret = -1;
3113 done:
3114     nvme_copy_done(iocb);
3115 }
3116 
3117 static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
3118 {
3119     NvmeNamespace *ns = req->ns;
3120     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
3121     NvmeCopyAIOCB *iocb = blk_aio_get(&nvme_copy_aiocb_info, ns->blkconf.blk,
3122                                       nvme_misc_cb, req);
3123     uint16_t nr = copy->nr + 1;
3124     uint8_t format = copy->control[0] & 0xf;
3125     uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
3126     uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
3127     size_t len = sizeof(NvmeCopySourceRangeFormat0);
3128 
3129     uint16_t status;
3130 
3131     trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);
3132 
3133     iocb->ranges = NULL;
3134     iocb->zone = NULL;
3135 
3136     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) &&
3137         ((prinfor & NVME_PRINFO_PRACT) != (prinfow & NVME_PRINFO_PRACT))) {
3138         status = NVME_INVALID_FIELD | NVME_DNR;
3139         goto invalid;
3140     }
3141 
3142     if (!(n->id_ctrl.ocfs & (1 << format))) {
3143         trace_pci_nvme_err_copy_invalid_format(format);
3144         status = NVME_INVALID_FIELD | NVME_DNR;
3145         goto invalid;
3146     }
3147 
3148     if (nr > ns->id_ns.msrc + 1) {
3149         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
3150         goto invalid;
3151     }
3152 
3153     if ((ns->pif == 0x0 && format != 0x0) ||
3154         (ns->pif != 0x0 && format != 0x1)) {
3155         status = NVME_INVALID_FORMAT | NVME_DNR;
3156         goto invalid;
3157     }
3158 
3159     if (ns->pif) {
3160         len = sizeof(NvmeCopySourceRangeFormat1);
3161     }
3162 
3163     iocb->format = format;
3164     iocb->ranges = g_malloc_n(nr, len);
3165     status = nvme_h2c(n, (uint8_t *)iocb->ranges, len * nr, req);
3166     if (status) {
3167         goto invalid;
3168     }
3169 
3170     iocb->slba = le64_to_cpu(copy->sdlba);
3171 
3172     if (ns->params.zoned) {
3173         iocb->zone = nvme_get_zone_by_slba(ns, iocb->slba);
3174         if (!iocb->zone) {
3175             status = NVME_LBA_RANGE | NVME_DNR;
3176             goto invalid;
3177         }
3178 
3179         status = nvme_zrm_auto(n, ns, iocb->zone);
3180         if (status) {
3181             goto invalid;
3182         }
3183     }
3184 
3185     status = nvme_check_copy_mcl(ns, iocb, nr);
3186     if (status) {
3187         goto invalid;
3188     }
3189 
3190     iocb->req = req;
3191     iocb->ret = 0;
3192     iocb->nr = nr;
3193     iocb->idx = 0;
3194     iocb->reftag = le32_to_cpu(copy->reftag);
3195     iocb->reftag |= (uint64_t)le32_to_cpu(copy->cdw3) << 32;
3196     iocb->bounce = g_malloc_n(le16_to_cpu(ns->id_ns.mssrl),
3197                               ns->lbasz + ns->lbaf.ms);
3198 
3199     qemu_iovec_init(&iocb->iov, 1);
3200 
3201     block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.read, 0,
3202                      BLOCK_ACCT_READ);
3203     block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.write, 0,
3204                      BLOCK_ACCT_WRITE);
3205 
3206     req->aiocb = &iocb->common;
3207     nvme_do_copy(iocb);
3208 
3209     return NVME_NO_COMPLETE;
3210 
3211 invalid:
3212     g_free(iocb->ranges);
3213     qemu_aio_unref(iocb);
3214     return status;
3215 }
3216 
3217 static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req)
3218 {
3219     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3220     NvmeNamespace *ns = req->ns;
3221     BlockBackend *blk = ns->blkconf.blk;
3222     uint64_t slba = le64_to_cpu(rw->slba);
3223     uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
3224     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3225     size_t data_len = nvme_l2b(ns, nlb);
3226     size_t len = data_len;
3227     int64_t offset = nvme_l2b(ns, slba);
3228     struct nvme_compare_ctx *ctx = NULL;
3229     uint16_t status;
3230 
3231     trace_pci_nvme_compare(nvme_cid(req), nvme_nsid(ns), slba, nlb);
3232 
3233     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) && (prinfo & NVME_PRINFO_PRACT)) {
3234         return NVME_INVALID_PROT_INFO | NVME_DNR;
3235     }
3236 
3237     if (nvme_ns_ext(ns)) {
3238         len += nvme_m2b(ns, nlb);
3239     }
3240 
3241     status = nvme_check_mdts(n, len);
3242     if (status) {
3243         return status;
3244     }
3245 
3246     status = nvme_check_bounds(ns, slba, nlb);
3247     if (status) {
3248         return status;
3249     }
3250 
3251     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3252         status = nvme_check_dulbe(ns, slba, nlb);
3253         if (status) {
3254             return status;
3255         }
3256     }
3257 
3258     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
3259     if (status) {
3260         return status;
3261     }
3262 
3263     ctx = g_new(struct nvme_compare_ctx, 1);
3264     ctx->data.bounce = g_malloc(data_len);
3265 
3266     req->opaque = ctx;
3267 
3268     qemu_iovec_init(&ctx->data.iov, 1);
3269     qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, data_len);
3270 
3271     block_acct_start(blk_get_stats(blk), &req->acct, data_len,
3272                      BLOCK_ACCT_READ);
3273     req->aiocb = blk_aio_preadv(blk, offset, &ctx->data.iov, 0,
3274                                 nvme_compare_data_cb, req);
3275 
3276     return NVME_NO_COMPLETE;
3277 }
3278 
3279 typedef struct NvmeFlushAIOCB {
3280     BlockAIOCB common;
3281     BlockAIOCB *aiocb;
3282     NvmeRequest *req;
3283     int ret;
3284 
3285     NvmeNamespace *ns;
3286     uint32_t nsid;
3287     bool broadcast;
3288 } NvmeFlushAIOCB;
3289 
3290 static void nvme_flush_cancel(BlockAIOCB *acb)
3291 {
3292     NvmeFlushAIOCB *iocb = container_of(acb, NvmeFlushAIOCB, common);
3293 
3294     iocb->ret = -ECANCELED;
3295 
3296     if (iocb->aiocb) {
3297         blk_aio_cancel_async(iocb->aiocb);
3298         iocb->aiocb = NULL;
3299     }
3300 }
3301 
3302 static const AIOCBInfo nvme_flush_aiocb_info = {
3303     .aiocb_size = sizeof(NvmeFlushAIOCB),
3304     .cancel_async = nvme_flush_cancel,
3305     .get_aio_context = nvme_get_aio_context,
3306 };
3307 
3308 static void nvme_do_flush(NvmeFlushAIOCB *iocb);
3309 
3310 static void nvme_flush_ns_cb(void *opaque, int ret)
3311 {
3312     NvmeFlushAIOCB *iocb = opaque;
3313     NvmeNamespace *ns = iocb->ns;
3314 
3315     if (ret < 0) {
3316         iocb->ret = ret;
3317         goto out;
3318     } else if (iocb->ret < 0) {
3319         goto out;
3320     }
3321 
3322     if (ns) {
3323         trace_pci_nvme_flush_ns(iocb->nsid);
3324 
3325         iocb->ns = NULL;
3326         iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);
3327         return;
3328     }
3329 
3330 out:
3331     nvme_do_flush(iocb);
3332 }
3333 
3334 static void nvme_do_flush(NvmeFlushAIOCB *iocb)
3335 {
3336     NvmeRequest *req = iocb->req;
3337     NvmeCtrl *n = nvme_ctrl(req);
3338     int i;
3339 
3340     if (iocb->ret < 0) {
3341         goto done;
3342     }
3343 
3344     if (iocb->broadcast) {
3345         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
3346             iocb->ns = nvme_ns(n, i);
3347             if (iocb->ns) {
3348                 iocb->nsid = i;
3349                 break;
3350             }
3351         }
3352     }
3353 
3354     if (!iocb->ns) {
3355         goto done;
3356     }
3357 
3358     nvme_flush_ns_cb(iocb, 0);
3359     return;
3360 
3361 done:
3362     iocb->common.cb(iocb->common.opaque, iocb->ret);
3363     qemu_aio_unref(iocb);
3364 }
3365 
3366 static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req)
3367 {
3368     NvmeFlushAIOCB *iocb;
3369     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
3370     uint16_t status;
3371 
3372     iocb = qemu_aio_get(&nvme_flush_aiocb_info, NULL, nvme_misc_cb, req);
3373 
3374     iocb->req = req;
3375     iocb->ret = 0;
3376     iocb->ns = NULL;
3377     iocb->nsid = 0;
3378     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
3379 
3380     if (!iocb->broadcast) {
3381         if (!nvme_nsid_valid(n, nsid)) {
3382             status = NVME_INVALID_NSID | NVME_DNR;
3383             goto out;
3384         }
3385 
3386         iocb->ns = nvme_ns(n, nsid);
3387         if (!iocb->ns) {
3388             status = NVME_INVALID_FIELD | NVME_DNR;
3389             goto out;
3390         }
3391 
3392         iocb->nsid = nsid;
3393     }
3394 
3395     req->aiocb = &iocb->common;
3396     nvme_do_flush(iocb);
3397 
3398     return NVME_NO_COMPLETE;
3399 
3400 out:
3401     qemu_aio_unref(iocb);
3402 
3403     return status;
3404 }
3405 
3406 static uint16_t nvme_read(NvmeCtrl *n, NvmeRequest *req)
3407 {
3408     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3409     NvmeNamespace *ns = req->ns;
3410     uint64_t slba = le64_to_cpu(rw->slba);
3411     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3412     uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
3413     uint64_t data_size = nvme_l2b(ns, nlb);
3414     uint64_t mapped_size = data_size;
3415     uint64_t data_offset;
3416     BlockBackend *blk = ns->blkconf.blk;
3417     uint16_t status;
3418 
3419     if (nvme_ns_ext(ns)) {
3420         mapped_size += nvme_m2b(ns, nlb);
3421 
3422         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3423             bool pract = prinfo & NVME_PRINFO_PRACT;
3424 
3425             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3426                 mapped_size = data_size;
3427             }
3428         }
3429     }
3430 
3431     trace_pci_nvme_read(nvme_cid(req), nvme_nsid(ns), nlb, mapped_size, slba);
3432 
3433     status = nvme_check_mdts(n, mapped_size);
3434     if (status) {
3435         goto invalid;
3436     }
3437 
3438     status = nvme_check_bounds(ns, slba, nlb);
3439     if (status) {
3440         goto invalid;
3441     }
3442 
3443     if (ns->params.zoned) {
3444         status = nvme_check_zone_read(ns, slba, nlb);
3445         if (status) {
3446             trace_pci_nvme_err_zone_read_not_ok(slba, nlb, status);
3447             goto invalid;
3448         }
3449     }
3450 
3451     if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
3452         status = nvme_check_dulbe(ns, slba, nlb);
3453         if (status) {
3454             goto invalid;
3455         }
3456     }
3457 
3458     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3459         return nvme_dif_rw(n, req);
3460     }
3461 
3462     status = nvme_map_data(n, nlb, req);
3463     if (status) {
3464         goto invalid;
3465     }
3466 
3467     data_offset = nvme_l2b(ns, slba);
3468 
3469     block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3470                      BLOCK_ACCT_READ);
3471     nvme_blk_read(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3472     return NVME_NO_COMPLETE;
3473 
3474 invalid:
3475     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_READ);
3476     return status | NVME_DNR;
3477 }
3478 
3479 static void nvme_do_write_fdp(NvmeCtrl *n, NvmeRequest *req, uint64_t slba,
3480                               uint32_t nlb)
3481 {
3482     NvmeNamespace *ns = req->ns;
3483     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3484     uint64_t data_size = nvme_l2b(ns, nlb);
3485     uint32_t dw12 = le32_to_cpu(req->cmd.cdw12);
3486     uint8_t dtype = (dw12 >> 20) & 0xf;
3487     uint16_t pid = le16_to_cpu(rw->dspec);
3488     uint16_t ph, rg, ruhid;
3489     NvmeReclaimUnit *ru;
3490 
3491     if (dtype != NVME_DIRECTIVE_DATA_PLACEMENT ||
3492         !nvme_parse_pid(ns, pid, &ph, &rg)) {
3493         ph = 0;
3494         rg = 0;
3495     }
3496 
3497     ruhid = ns->fdp.phs[ph];
3498     ru = &ns->endgrp->fdp.ruhs[ruhid].rus[rg];
3499 
3500     nvme_fdp_stat_inc(&ns->endgrp->fdp.hbmw, data_size);
3501     nvme_fdp_stat_inc(&ns->endgrp->fdp.mbmw, data_size);
3502 
3503     while (nlb) {
3504         if (nlb < ru->ruamw) {
3505             ru->ruamw -= nlb;
3506             break;
3507         }
3508 
3509         nlb -= ru->ruamw;
3510         nvme_update_ruh(n, ns, pid);
3511     }
3512 }
3513 
3514 static uint16_t nvme_do_write(NvmeCtrl *n, NvmeRequest *req, bool append,
3515                               bool wrz)
3516 {
3517     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
3518     NvmeNamespace *ns = req->ns;
3519     uint64_t slba = le64_to_cpu(rw->slba);
3520     uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
3521     uint16_t ctrl = le16_to_cpu(rw->control);
3522     uint8_t prinfo = NVME_RW_PRINFO(ctrl);
3523     uint64_t data_size = nvme_l2b(ns, nlb);
3524     uint64_t mapped_size = data_size;
3525     uint64_t data_offset;
3526     NvmeZone *zone;
3527     NvmeZonedResult *res = (NvmeZonedResult *)&req->cqe;
3528     BlockBackend *blk = ns->blkconf.blk;
3529     uint16_t status;
3530 
3531     if (nvme_ns_ext(ns)) {
3532         mapped_size += nvme_m2b(ns, nlb);
3533 
3534         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3535             bool pract = prinfo & NVME_PRINFO_PRACT;
3536 
3537             if (pract && ns->lbaf.ms == nvme_pi_tuple_size(ns)) {
3538                 mapped_size -= nvme_m2b(ns, nlb);
3539             }
3540         }
3541     }
3542 
3543     trace_pci_nvme_write(nvme_cid(req), nvme_io_opc_str(rw->opcode),
3544                          nvme_nsid(ns), nlb, mapped_size, slba);
3545 
3546     if (!wrz) {
3547         status = nvme_check_mdts(n, mapped_size);
3548         if (status) {
3549             goto invalid;
3550         }
3551     }
3552 
3553     status = nvme_check_bounds(ns, slba, nlb);
3554     if (status) {
3555         goto invalid;
3556     }
3557 
3558     if (ns->params.zoned) {
3559         zone = nvme_get_zone_by_slba(ns, slba);
3560         assert(zone);
3561 
3562         if (append) {
3563             bool piremap = !!(ctrl & NVME_RW_PIREMAP);
3564 
3565             if (unlikely(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3566                 return NVME_INVALID_ZONE_OP | NVME_DNR;
3567             }
3568 
3569             if (unlikely(slba != zone->d.zslba)) {
3570                 trace_pci_nvme_err_append_not_at_start(slba, zone->d.zslba);
3571                 status = NVME_INVALID_FIELD;
3572                 goto invalid;
3573             }
3574 
3575             if (n->params.zasl &&
3576                 data_size > (uint64_t)n->page_size << n->params.zasl) {
3577                 trace_pci_nvme_err_zasl(data_size);
3578                 return NVME_INVALID_FIELD | NVME_DNR;
3579             }
3580 
3581             slba = zone->w_ptr;
3582             rw->slba = cpu_to_le64(slba);
3583             res->slba = cpu_to_le64(slba);
3584 
3585             switch (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3586             case NVME_ID_NS_DPS_TYPE_1:
3587                 if (!piremap) {
3588                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3589                 }
3590 
3591                 /* fallthrough */
3592 
3593             case NVME_ID_NS_DPS_TYPE_2:
3594                 if (piremap) {
3595                     uint32_t reftag = le32_to_cpu(rw->reftag);
3596                     rw->reftag = cpu_to_le32(reftag + (slba - zone->d.zslba));
3597                 }
3598 
3599                 break;
3600 
3601             case NVME_ID_NS_DPS_TYPE_3:
3602                 if (piremap) {
3603                     return NVME_INVALID_PROT_INFO | NVME_DNR;
3604                 }
3605 
3606                 break;
3607             }
3608         }
3609 
3610         status = nvme_check_zone_write(ns, zone, slba, nlb);
3611         if (status) {
3612             goto invalid;
3613         }
3614 
3615         status = nvme_zrm_auto(n, ns, zone);
3616         if (status) {
3617             goto invalid;
3618         }
3619 
3620         if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
3621             zone->w_ptr += nlb;
3622         }
3623     } else if (ns->endgrp && ns->endgrp->fdp.enabled) {
3624         nvme_do_write_fdp(n, req, slba, nlb);
3625     }
3626 
3627     data_offset = nvme_l2b(ns, slba);
3628 
3629     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
3630         return nvme_dif_rw(n, req);
3631     }
3632 
3633     if (!wrz) {
3634         status = nvme_map_data(n, nlb, req);
3635         if (status) {
3636             goto invalid;
3637         }
3638 
3639         block_acct_start(blk_get_stats(blk), &req->acct, data_size,
3640                          BLOCK_ACCT_WRITE);
3641         nvme_blk_write(blk, data_offset, BDRV_SECTOR_SIZE, nvme_rw_cb, req);
3642     } else {
3643         req->aiocb = blk_aio_pwrite_zeroes(blk, data_offset, data_size,
3644                                            BDRV_REQ_MAY_UNMAP, nvme_rw_cb,
3645                                            req);
3646     }
3647 
3648     return NVME_NO_COMPLETE;
3649 
3650 invalid:
3651     block_acct_invalid(blk_get_stats(blk), BLOCK_ACCT_WRITE);
3652     return status | NVME_DNR;
3653 }
3654 
3655 static inline uint16_t nvme_write(NvmeCtrl *n, NvmeRequest *req)
3656 {
3657     return nvme_do_write(n, req, false, false);
3658 }
3659 
3660 static inline uint16_t nvme_write_zeroes(NvmeCtrl *n, NvmeRequest *req)
3661 {
3662     return nvme_do_write(n, req, false, true);
3663 }
3664 
3665 static inline uint16_t nvme_zone_append(NvmeCtrl *n, NvmeRequest *req)
3666 {
3667     return nvme_do_write(n, req, true, false);
3668 }
3669 
3670 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace *ns, NvmeCmd *c,
3671                                             uint64_t *slba, uint32_t *zone_idx)
3672 {
3673     uint32_t dw10 = le32_to_cpu(c->cdw10);
3674     uint32_t dw11 = le32_to_cpu(c->cdw11);
3675 
3676     if (!ns->params.zoned) {
3677         trace_pci_nvme_err_invalid_opc(c->opcode);
3678         return NVME_INVALID_OPCODE | NVME_DNR;
3679     }
3680 
3681     *slba = ((uint64_t)dw11) << 32 | dw10;
3682     if (unlikely(*slba >= ns->id_ns.nsze)) {
3683         trace_pci_nvme_err_invalid_lba_range(*slba, 0, ns->id_ns.nsze);
3684         *slba = 0;
3685         return NVME_LBA_RANGE | NVME_DNR;
3686     }
3687 
3688     *zone_idx = nvme_zone_idx(ns, *slba);
3689     assert(*zone_idx < ns->num_zones);
3690 
3691     return NVME_SUCCESS;
3692 }
3693 
3694 typedef uint16_t (*op_handler_t)(NvmeNamespace *, NvmeZone *, NvmeZoneState,
3695                                  NvmeRequest *);
3696 
3697 enum NvmeZoneProcessingMask {
3698     NVME_PROC_CURRENT_ZONE    = 0,
3699     NVME_PROC_OPENED_ZONES    = 1 << 0,
3700     NVME_PROC_CLOSED_ZONES    = 1 << 1,
3701     NVME_PROC_READ_ONLY_ZONES = 1 << 2,
3702     NVME_PROC_FULL_ZONES      = 1 << 3,
3703 };
3704 
3705 static uint16_t nvme_open_zone(NvmeNamespace *ns, NvmeZone *zone,
3706                                NvmeZoneState state, NvmeRequest *req)
3707 {
3708     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
3709     int flags = 0;
3710 
3711     if (cmd->zsflags & NVME_ZSFLAG_ZRWA_ALLOC) {
3712         uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3713 
3714         if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3715             return NVME_INVALID_ZONE_OP | NVME_DNR;
3716         }
3717 
3718         if (zone->w_ptr % ns->zns.zrwafg) {
3719             return NVME_NOZRWA | NVME_DNR;
3720         }
3721 
3722         flags = NVME_ZRM_ZRWA;
3723     }
3724 
3725     return nvme_zrm_open_flags(nvme_ctrl(req), ns, zone, flags);
3726 }
3727 
3728 static uint16_t nvme_close_zone(NvmeNamespace *ns, NvmeZone *zone,
3729                                 NvmeZoneState state, NvmeRequest *req)
3730 {
3731     return nvme_zrm_close(ns, zone);
3732 }
3733 
3734 static uint16_t nvme_finish_zone(NvmeNamespace *ns, NvmeZone *zone,
3735                                  NvmeZoneState state, NvmeRequest *req)
3736 {
3737     return nvme_zrm_finish(ns, zone);
3738 }
3739 
3740 static uint16_t nvme_offline_zone(NvmeNamespace *ns, NvmeZone *zone,
3741                                   NvmeZoneState state, NvmeRequest *req)
3742 {
3743     switch (state) {
3744     case NVME_ZONE_STATE_READ_ONLY:
3745         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_OFFLINE);
3746         /* fall through */
3747     case NVME_ZONE_STATE_OFFLINE:
3748         return NVME_SUCCESS;
3749     default:
3750         return NVME_ZONE_INVAL_TRANSITION;
3751     }
3752 }
3753 
3754 static uint16_t nvme_set_zd_ext(NvmeNamespace *ns, NvmeZone *zone)
3755 {
3756     uint16_t status;
3757     uint8_t state = nvme_get_zone_state(zone);
3758 
3759     if (state == NVME_ZONE_STATE_EMPTY) {
3760         status = nvme_aor_check(ns, 1, 0);
3761         if (status) {
3762             return status;
3763         }
3764         nvme_aor_inc_active(ns);
3765         zone->d.za |= NVME_ZA_ZD_EXT_VALID;
3766         nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
3767         return NVME_SUCCESS;
3768     }
3769 
3770     return NVME_ZONE_INVAL_TRANSITION;
3771 }
3772 
3773 static uint16_t nvme_bulk_proc_zone(NvmeNamespace *ns, NvmeZone *zone,
3774                                     enum NvmeZoneProcessingMask proc_mask,
3775                                     op_handler_t op_hndlr, NvmeRequest *req)
3776 {
3777     uint16_t status = NVME_SUCCESS;
3778     NvmeZoneState zs = nvme_get_zone_state(zone);
3779     bool proc_zone;
3780 
3781     switch (zs) {
3782     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3783     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3784         proc_zone = proc_mask & NVME_PROC_OPENED_ZONES;
3785         break;
3786     case NVME_ZONE_STATE_CLOSED:
3787         proc_zone = proc_mask & NVME_PROC_CLOSED_ZONES;
3788         break;
3789     case NVME_ZONE_STATE_READ_ONLY:
3790         proc_zone = proc_mask & NVME_PROC_READ_ONLY_ZONES;
3791         break;
3792     case NVME_ZONE_STATE_FULL:
3793         proc_zone = proc_mask & NVME_PROC_FULL_ZONES;
3794         break;
3795     default:
3796         proc_zone = false;
3797     }
3798 
3799     if (proc_zone) {
3800         status = op_hndlr(ns, zone, zs, req);
3801     }
3802 
3803     return status;
3804 }
3805 
3806 static uint16_t nvme_do_zone_op(NvmeNamespace *ns, NvmeZone *zone,
3807                                 enum NvmeZoneProcessingMask proc_mask,
3808                                 op_handler_t op_hndlr, NvmeRequest *req)
3809 {
3810     NvmeZone *next;
3811     uint16_t status = NVME_SUCCESS;
3812     int i;
3813 
3814     if (!proc_mask) {
3815         status = op_hndlr(ns, zone, nvme_get_zone_state(zone), req);
3816     } else {
3817         if (proc_mask & NVME_PROC_CLOSED_ZONES) {
3818             QTAILQ_FOREACH_SAFE(zone, &ns->closed_zones, entry, next) {
3819                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3820                                              req);
3821                 if (status && status != NVME_NO_COMPLETE) {
3822                     goto out;
3823                 }
3824             }
3825         }
3826         if (proc_mask & NVME_PROC_OPENED_ZONES) {
3827             QTAILQ_FOREACH_SAFE(zone, &ns->imp_open_zones, entry, next) {
3828                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3829                                              req);
3830                 if (status && status != NVME_NO_COMPLETE) {
3831                     goto out;
3832                 }
3833             }
3834 
3835             QTAILQ_FOREACH_SAFE(zone, &ns->exp_open_zones, entry, next) {
3836                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3837                                              req);
3838                 if (status && status != NVME_NO_COMPLETE) {
3839                     goto out;
3840                 }
3841             }
3842         }
3843         if (proc_mask & NVME_PROC_FULL_ZONES) {
3844             QTAILQ_FOREACH_SAFE(zone, &ns->full_zones, entry, next) {
3845                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3846                                              req);
3847                 if (status && status != NVME_NO_COMPLETE) {
3848                     goto out;
3849                 }
3850             }
3851         }
3852 
3853         if (proc_mask & NVME_PROC_READ_ONLY_ZONES) {
3854             for (i = 0; i < ns->num_zones; i++, zone++) {
3855                 status = nvme_bulk_proc_zone(ns, zone, proc_mask, op_hndlr,
3856                                              req);
3857                 if (status && status != NVME_NO_COMPLETE) {
3858                     goto out;
3859                 }
3860             }
3861         }
3862     }
3863 
3864 out:
3865     return status;
3866 }
3867 
3868 typedef struct NvmeZoneResetAIOCB {
3869     BlockAIOCB common;
3870     BlockAIOCB *aiocb;
3871     NvmeRequest *req;
3872     int ret;
3873 
3874     bool all;
3875     int idx;
3876     NvmeZone *zone;
3877 } NvmeZoneResetAIOCB;
3878 
3879 static void nvme_zone_reset_cancel(BlockAIOCB *aiocb)
3880 {
3881     NvmeZoneResetAIOCB *iocb = container_of(aiocb, NvmeZoneResetAIOCB, common);
3882     NvmeRequest *req = iocb->req;
3883     NvmeNamespace *ns = req->ns;
3884 
3885     iocb->idx = ns->num_zones;
3886 
3887     iocb->ret = -ECANCELED;
3888 
3889     if (iocb->aiocb) {
3890         blk_aio_cancel_async(iocb->aiocb);
3891         iocb->aiocb = NULL;
3892     }
3893 }
3894 
3895 static const AIOCBInfo nvme_zone_reset_aiocb_info = {
3896     .aiocb_size = sizeof(NvmeZoneResetAIOCB),
3897     .cancel_async = nvme_zone_reset_cancel,
3898 };
3899 
3900 static void nvme_zone_reset_cb(void *opaque, int ret);
3901 
3902 static void nvme_zone_reset_epilogue_cb(void *opaque, int ret)
3903 {
3904     NvmeZoneResetAIOCB *iocb = opaque;
3905     NvmeRequest *req = iocb->req;
3906     NvmeNamespace *ns = req->ns;
3907     int64_t moff;
3908     int count;
3909 
3910     if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
3911         goto out;
3912     }
3913 
3914     moff = nvme_moff(ns, iocb->zone->d.zslba);
3915     count = nvme_m2b(ns, ns->zone_size);
3916 
3917     iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, moff, count,
3918                                         BDRV_REQ_MAY_UNMAP,
3919                                         nvme_zone_reset_cb, iocb);
3920     return;
3921 
3922 out:
3923     nvme_zone_reset_cb(iocb, ret);
3924 }
3925 
3926 static void nvme_zone_reset_cb(void *opaque, int ret)
3927 {
3928     NvmeZoneResetAIOCB *iocb = opaque;
3929     NvmeRequest *req = iocb->req;
3930     NvmeNamespace *ns = req->ns;
3931 
3932     if (iocb->ret < 0) {
3933         goto done;
3934     } else if (ret < 0) {
3935         iocb->ret = ret;
3936         goto done;
3937     }
3938 
3939     if (iocb->zone) {
3940         nvme_zrm_reset(ns, iocb->zone);
3941 
3942         if (!iocb->all) {
3943             goto done;
3944         }
3945     }
3946 
3947     while (iocb->idx < ns->num_zones) {
3948         NvmeZone *zone = &ns->zone_array[iocb->idx++];
3949 
3950         switch (nvme_get_zone_state(zone)) {
3951         case NVME_ZONE_STATE_EMPTY:
3952             if (!iocb->all) {
3953                 goto done;
3954             }
3955 
3956             continue;
3957 
3958         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
3959         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
3960         case NVME_ZONE_STATE_CLOSED:
3961         case NVME_ZONE_STATE_FULL:
3962             iocb->zone = zone;
3963             break;
3964 
3965         default:
3966             continue;
3967         }
3968 
3969         trace_pci_nvme_zns_zone_reset(zone->d.zslba);
3970 
3971         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk,
3972                                             nvme_l2b(ns, zone->d.zslba),
3973                                             nvme_l2b(ns, ns->zone_size),
3974                                             BDRV_REQ_MAY_UNMAP,
3975                                             nvme_zone_reset_epilogue_cb,
3976                                             iocb);
3977         return;
3978     }
3979 
3980 done:
3981     iocb->aiocb = NULL;
3982 
3983     iocb->common.cb(iocb->common.opaque, iocb->ret);
3984     qemu_aio_unref(iocb);
3985 }
3986 
3987 static uint16_t nvme_zone_mgmt_send_zrwa_flush(NvmeCtrl *n, NvmeZone *zone,
3988                                                uint64_t elba, NvmeRequest *req)
3989 {
3990     NvmeNamespace *ns = req->ns;
3991     uint16_t ozcs = le16_to_cpu(ns->id_ns_zoned->ozcs);
3992     uint64_t wp = zone->d.wp;
3993     uint32_t nlb = elba - wp + 1;
3994     uint16_t status;
3995 
3996 
3997     if (!(ozcs & NVME_ID_NS_ZONED_OZCS_ZRWASUP)) {
3998         return NVME_INVALID_ZONE_OP | NVME_DNR;
3999     }
4000 
4001     if (!(zone->d.za & NVME_ZA_ZRWA_VALID)) {
4002         return NVME_INVALID_FIELD | NVME_DNR;
4003     }
4004 
4005     if (elba < wp || elba > wp + ns->zns.zrwas) {
4006         return NVME_ZONE_BOUNDARY_ERROR | NVME_DNR;
4007     }
4008 
4009     if (nlb % ns->zns.zrwafg) {
4010         return NVME_INVALID_FIELD | NVME_DNR;
4011     }
4012 
4013     status = nvme_zrm_auto(n, ns, zone);
4014     if (status) {
4015         return status;
4016     }
4017 
4018     zone->w_ptr += nlb;
4019 
4020     nvme_advance_zone_wp(ns, zone, nlb);
4021 
4022     return NVME_SUCCESS;
4023 }
4024 
4025 static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4026 {
4027     NvmeZoneSendCmd *cmd = (NvmeZoneSendCmd *)&req->cmd;
4028     NvmeNamespace *ns = req->ns;
4029     NvmeZone *zone;
4030     NvmeZoneResetAIOCB *iocb;
4031     uint8_t *zd_ext;
4032     uint64_t slba = 0;
4033     uint32_t zone_idx = 0;
4034     uint16_t status;
4035     uint8_t action = cmd->zsa;
4036     bool all;
4037     enum NvmeZoneProcessingMask proc_mask = NVME_PROC_CURRENT_ZONE;
4038 
4039     all = cmd->zsflags & NVME_ZSFLAG_SELECT_ALL;
4040 
4041     req->status = NVME_SUCCESS;
4042 
4043     if (!all) {
4044         status = nvme_get_mgmt_zone_slba_idx(ns, &req->cmd, &slba, &zone_idx);
4045         if (status) {
4046             return status;
4047         }
4048     }
4049 
4050     zone = &ns->zone_array[zone_idx];
4051     if (slba != zone->d.zslba && action != NVME_ZONE_ACTION_ZRWA_FLUSH) {
4052         trace_pci_nvme_err_unaligned_zone_cmd(action, slba, zone->d.zslba);
4053         return NVME_INVALID_FIELD | NVME_DNR;
4054     }
4055 
4056     switch (action) {
4057 
4058     case NVME_ZONE_ACTION_OPEN:
4059         if (all) {
4060             proc_mask = NVME_PROC_CLOSED_ZONES;
4061         }
4062         trace_pci_nvme_open_zone(slba, zone_idx, all);
4063         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_open_zone, req);
4064         break;
4065 
4066     case NVME_ZONE_ACTION_CLOSE:
4067         if (all) {
4068             proc_mask = NVME_PROC_OPENED_ZONES;
4069         }
4070         trace_pci_nvme_close_zone(slba, zone_idx, all);
4071         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_close_zone, req);
4072         break;
4073 
4074     case NVME_ZONE_ACTION_FINISH:
4075         if (all) {
4076             proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES;
4077         }
4078         trace_pci_nvme_finish_zone(slba, zone_idx, all);
4079         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_finish_zone, req);
4080         break;
4081 
4082     case NVME_ZONE_ACTION_RESET:
4083         trace_pci_nvme_reset_zone(slba, zone_idx, all);
4084 
4085         iocb = blk_aio_get(&nvme_zone_reset_aiocb_info, ns->blkconf.blk,
4086                            nvme_misc_cb, req);
4087 
4088         iocb->req = req;
4089         iocb->ret = 0;
4090         iocb->all = all;
4091         iocb->idx = zone_idx;
4092         iocb->zone = NULL;
4093 
4094         req->aiocb = &iocb->common;
4095         nvme_zone_reset_cb(iocb, 0);
4096 
4097         return NVME_NO_COMPLETE;
4098 
4099     case NVME_ZONE_ACTION_OFFLINE:
4100         if (all) {
4101             proc_mask = NVME_PROC_READ_ONLY_ZONES;
4102         }
4103         trace_pci_nvme_offline_zone(slba, zone_idx, all);
4104         status = nvme_do_zone_op(ns, zone, proc_mask, nvme_offline_zone, req);
4105         break;
4106 
4107     case NVME_ZONE_ACTION_SET_ZD_EXT:
4108         trace_pci_nvme_set_descriptor_extension(slba, zone_idx);
4109         if (all || !ns->params.zd_extension_size) {
4110             return NVME_INVALID_FIELD | NVME_DNR;
4111         }
4112         zd_ext = nvme_get_zd_extension(ns, zone_idx);
4113         status = nvme_h2c(n, zd_ext, ns->params.zd_extension_size, req);
4114         if (status) {
4115             trace_pci_nvme_err_zd_extension_map_error(zone_idx);
4116             return status;
4117         }
4118 
4119         status = nvme_set_zd_ext(ns, zone);
4120         if (status == NVME_SUCCESS) {
4121             trace_pci_nvme_zd_extension_set(zone_idx);
4122             return status;
4123         }
4124         break;
4125 
4126     case NVME_ZONE_ACTION_ZRWA_FLUSH:
4127         if (all) {
4128             return NVME_INVALID_FIELD | NVME_DNR;
4129         }
4130 
4131         return nvme_zone_mgmt_send_zrwa_flush(n, zone, slba, req);
4132 
4133     default:
4134         trace_pci_nvme_err_invalid_mgmt_action(action);
4135         status = NVME_INVALID_FIELD;
4136     }
4137 
4138     if (status == NVME_ZONE_INVAL_TRANSITION) {
4139         trace_pci_nvme_err_invalid_zone_state_transition(action, slba,
4140                                                          zone->d.za);
4141     }
4142     if (status) {
4143         status |= NVME_DNR;
4144     }
4145 
4146     return status;
4147 }
4148 
4149 static bool nvme_zone_matches_filter(uint32_t zafs, NvmeZone *zl)
4150 {
4151     NvmeZoneState zs = nvme_get_zone_state(zl);
4152 
4153     switch (zafs) {
4154     case NVME_ZONE_REPORT_ALL:
4155         return true;
4156     case NVME_ZONE_REPORT_EMPTY:
4157         return zs == NVME_ZONE_STATE_EMPTY;
4158     case NVME_ZONE_REPORT_IMPLICITLY_OPEN:
4159         return zs == NVME_ZONE_STATE_IMPLICITLY_OPEN;
4160     case NVME_ZONE_REPORT_EXPLICITLY_OPEN:
4161         return zs == NVME_ZONE_STATE_EXPLICITLY_OPEN;
4162     case NVME_ZONE_REPORT_CLOSED:
4163         return zs == NVME_ZONE_STATE_CLOSED;
4164     case NVME_ZONE_REPORT_FULL:
4165         return zs == NVME_ZONE_STATE_FULL;
4166     case NVME_ZONE_REPORT_READ_ONLY:
4167         return zs == NVME_ZONE_STATE_READ_ONLY;
4168     case NVME_ZONE_REPORT_OFFLINE:
4169         return zs == NVME_ZONE_STATE_OFFLINE;
4170     default:
4171         return false;
4172     }
4173 }
4174 
4175 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4176 {
4177     NvmeCmd *cmd = (NvmeCmd *)&req->cmd;
4178     NvmeNamespace *ns = req->ns;
4179     /* cdw12 is zero-based number of dwords to return. Convert to bytes */
4180     uint32_t data_size = (le32_to_cpu(cmd->cdw12) + 1) << 2;
4181     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
4182     uint32_t zone_idx, zra, zrasf, partial;
4183     uint64_t max_zones, nr_zones = 0;
4184     uint16_t status;
4185     uint64_t slba;
4186     NvmeZoneDescr *z;
4187     NvmeZone *zone;
4188     NvmeZoneReportHeader *header;
4189     void *buf, *buf_p;
4190     size_t zone_entry_sz;
4191     int i;
4192 
4193     req->status = NVME_SUCCESS;
4194 
4195     status = nvme_get_mgmt_zone_slba_idx(ns, cmd, &slba, &zone_idx);
4196     if (status) {
4197         return status;
4198     }
4199 
4200     zra = dw13 & 0xff;
4201     if (zra != NVME_ZONE_REPORT && zra != NVME_ZONE_REPORT_EXTENDED) {
4202         return NVME_INVALID_FIELD | NVME_DNR;
4203     }
4204     if (zra == NVME_ZONE_REPORT_EXTENDED && !ns->params.zd_extension_size) {
4205         return NVME_INVALID_FIELD | NVME_DNR;
4206     }
4207 
4208     zrasf = (dw13 >> 8) & 0xff;
4209     if (zrasf > NVME_ZONE_REPORT_OFFLINE) {
4210         return NVME_INVALID_FIELD | NVME_DNR;
4211     }
4212 
4213     if (data_size < sizeof(NvmeZoneReportHeader)) {
4214         return NVME_INVALID_FIELD | NVME_DNR;
4215     }
4216 
4217     status = nvme_check_mdts(n, data_size);
4218     if (status) {
4219         return status;
4220     }
4221 
4222     partial = (dw13 >> 16) & 0x01;
4223 
4224     zone_entry_sz = sizeof(NvmeZoneDescr);
4225     if (zra == NVME_ZONE_REPORT_EXTENDED) {
4226         zone_entry_sz += ns->params.zd_extension_size;
4227     }
4228 
4229     max_zones = (data_size - sizeof(NvmeZoneReportHeader)) / zone_entry_sz;
4230     buf = g_malloc0(data_size);
4231 
4232     zone = &ns->zone_array[zone_idx];
4233     for (i = zone_idx; i < ns->num_zones; i++) {
4234         if (partial && nr_zones >= max_zones) {
4235             break;
4236         }
4237         if (nvme_zone_matches_filter(zrasf, zone++)) {
4238             nr_zones++;
4239         }
4240     }
4241     header = buf;
4242     header->nr_zones = cpu_to_le64(nr_zones);
4243 
4244     buf_p = buf + sizeof(NvmeZoneReportHeader);
4245     for (; zone_idx < ns->num_zones && max_zones > 0; zone_idx++) {
4246         zone = &ns->zone_array[zone_idx];
4247         if (nvme_zone_matches_filter(zrasf, zone)) {
4248             z = buf_p;
4249             buf_p += sizeof(NvmeZoneDescr);
4250 
4251             z->zt = zone->d.zt;
4252             z->zs = zone->d.zs;
4253             z->zcap = cpu_to_le64(zone->d.zcap);
4254             z->zslba = cpu_to_le64(zone->d.zslba);
4255             z->za = zone->d.za;
4256 
4257             if (nvme_wp_is_valid(zone)) {
4258                 z->wp = cpu_to_le64(zone->d.wp);
4259             } else {
4260                 z->wp = cpu_to_le64(~0ULL);
4261             }
4262 
4263             if (zra == NVME_ZONE_REPORT_EXTENDED) {
4264                 if (zone->d.za & NVME_ZA_ZD_EXT_VALID) {
4265                     memcpy(buf_p, nvme_get_zd_extension(ns, zone_idx),
4266                            ns->params.zd_extension_size);
4267                 }
4268                 buf_p += ns->params.zd_extension_size;
4269             }
4270 
4271             max_zones--;
4272         }
4273     }
4274 
4275     status = nvme_c2h(n, (uint8_t *)buf, data_size, req);
4276 
4277     g_free(buf);
4278 
4279     return status;
4280 }
4281 
4282 static uint16_t nvme_io_mgmt_recv_ruhs(NvmeCtrl *n, NvmeRequest *req,
4283                                        size_t len)
4284 {
4285     NvmeNamespace *ns = req->ns;
4286     NvmeEnduranceGroup *endgrp;
4287     NvmeRuhStatus *hdr;
4288     NvmeRuhStatusDescr *ruhsd;
4289     unsigned int nruhsd;
4290     uint16_t rg, ph, *ruhid;
4291     size_t trans_len;
4292     g_autofree uint8_t *buf = NULL;
4293 
4294     if (!n->subsys) {
4295         return NVME_INVALID_FIELD | NVME_DNR;
4296     }
4297 
4298     if (ns->params.nsid == 0 || ns->params.nsid == 0xffffffff) {
4299         return NVME_INVALID_NSID | NVME_DNR;
4300     }
4301 
4302     if (!n->subsys->endgrp.fdp.enabled) {
4303         return NVME_FDP_DISABLED | NVME_DNR;
4304     }
4305 
4306     endgrp = ns->endgrp;
4307 
4308     nruhsd = ns->fdp.nphs * endgrp->fdp.nrg;
4309     trans_len = sizeof(NvmeRuhStatus) + nruhsd * sizeof(NvmeRuhStatusDescr);
4310     buf = g_malloc(trans_len);
4311 
4312     trans_len = MIN(trans_len, len);
4313 
4314     hdr = (NvmeRuhStatus *)buf;
4315     ruhsd = (NvmeRuhStatusDescr *)(buf + sizeof(NvmeRuhStatus));
4316 
4317     hdr->nruhsd = cpu_to_le16(nruhsd);
4318 
4319     ruhid = ns->fdp.phs;
4320 
4321     for (ph = 0; ph < ns->fdp.nphs; ph++, ruhid++) {
4322         NvmeRuHandle *ruh = &endgrp->fdp.ruhs[*ruhid];
4323 
4324         for (rg = 0; rg < endgrp->fdp.nrg; rg++, ruhsd++) {
4325             uint16_t pid = nvme_make_pid(ns, rg, ph);
4326 
4327             ruhsd->pid = cpu_to_le16(pid);
4328             ruhsd->ruhid = *ruhid;
4329             ruhsd->earutr = 0;
4330             ruhsd->ruamw = cpu_to_le64(ruh->rus[rg].ruamw);
4331         }
4332     }
4333 
4334     return nvme_c2h(n, buf, trans_len, req);
4335 }
4336 
4337 static uint16_t nvme_io_mgmt_recv(NvmeCtrl *n, NvmeRequest *req)
4338 {
4339     NvmeCmd *cmd = &req->cmd;
4340     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4341     uint32_t numd = le32_to_cpu(cmd->cdw11);
4342     uint8_t mo = (cdw10 & 0xff);
4343     size_t len = (numd + 1) << 2;
4344 
4345     switch (mo) {
4346     case NVME_IOMR_MO_NOP:
4347         return 0;
4348     case NVME_IOMR_MO_RUH_STATUS:
4349         return nvme_io_mgmt_recv_ruhs(n, req, len);
4350     default:
4351         return NVME_INVALID_FIELD | NVME_DNR;
4352     };
4353 }
4354 
4355 static uint16_t nvme_io_mgmt_send_ruh_update(NvmeCtrl *n, NvmeRequest *req)
4356 {
4357     NvmeCmd *cmd = &req->cmd;
4358     NvmeNamespace *ns = req->ns;
4359     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4360     uint16_t ret = NVME_SUCCESS;
4361     uint32_t npid = (cdw10 >> 1) + 1;
4362     unsigned int i = 0;
4363     g_autofree uint16_t *pids = NULL;
4364     uint32_t maxnpid;
4365 
4366     if (!ns->endgrp || !ns->endgrp->fdp.enabled) {
4367         return NVME_FDP_DISABLED | NVME_DNR;
4368     }
4369 
4370     maxnpid = n->subsys->endgrp.fdp.nrg * n->subsys->endgrp.fdp.nruh;
4371 
4372     if (unlikely(npid >= MIN(NVME_FDP_MAXPIDS, maxnpid))) {
4373         return NVME_INVALID_FIELD | NVME_DNR;
4374     }
4375 
4376     pids = g_new(uint16_t, npid);
4377 
4378     ret = nvme_h2c(n, pids, npid * sizeof(uint16_t), req);
4379     if (ret) {
4380         return ret;
4381     }
4382 
4383     for (; i < npid; i++) {
4384         if (!nvme_update_ruh(n, ns, pids[i])) {
4385             return NVME_INVALID_FIELD | NVME_DNR;
4386         }
4387     }
4388 
4389     return ret;
4390 }
4391 
4392 static uint16_t nvme_io_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4393 {
4394     NvmeCmd *cmd = &req->cmd;
4395     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4396     uint8_t mo = (cdw10 & 0xff);
4397 
4398     switch (mo) {
4399     case NVME_IOMS_MO_NOP:
4400         return 0;
4401     case NVME_IOMS_MO_RUH_UPDATE:
4402         return nvme_io_mgmt_send_ruh_update(n, req);
4403     default:
4404         return NVME_INVALID_FIELD | NVME_DNR;
4405     };
4406 }
4407 
4408 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
4409 {
4410     NvmeNamespace *ns;
4411     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4412 
4413     trace_pci_nvme_io_cmd(nvme_cid(req), nsid, nvme_sqid(req),
4414                           req->cmd.opcode, nvme_io_opc_str(req->cmd.opcode));
4415 
4416     if (!nvme_nsid_valid(n, nsid)) {
4417         return NVME_INVALID_NSID | NVME_DNR;
4418     }
4419 
4420     /*
4421      * In the base NVM command set, Flush may apply to all namespaces
4422      * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4423      * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4424      *
4425      * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4426      * opcode with a specific command since we cannot determine a unique I/O
4427      * command set. Opcode 0h could have any other meaning than something
4428      * equivalent to flushing and say it DOES have completely different
4429      * semantics in some other command set - does an NSID of FFFFFFFFh then
4430      * mean "for all namespaces, apply whatever command set specific command
4431      * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4432      * whatever command that uses the 0h opcode if, and only if, it allows NSID
4433      * to be FFFFFFFFh"?
4434      *
4435      * Anyway (and luckily), for now, we do not care about this since the
4436      * device only supports namespace types that includes the NVM Flush command
4437      * (NVM and Zoned), so always do an NVM Flush.
4438      */
4439     if (req->cmd.opcode == NVME_CMD_FLUSH) {
4440         return nvme_flush(n, req);
4441     }
4442 
4443     ns = nvme_ns(n, nsid);
4444     if (unlikely(!ns)) {
4445         return NVME_INVALID_FIELD | NVME_DNR;
4446     }
4447 
4448     if (!(ns->iocs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
4449         trace_pci_nvme_err_invalid_opc(req->cmd.opcode);
4450         return NVME_INVALID_OPCODE | NVME_DNR;
4451     }
4452 
4453     if (ns->status) {
4454         return ns->status;
4455     }
4456 
4457     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
4458         return NVME_INVALID_FIELD;
4459     }
4460 
4461     req->ns = ns;
4462 
4463     switch (req->cmd.opcode) {
4464     case NVME_CMD_WRITE_ZEROES:
4465         return nvme_write_zeroes(n, req);
4466     case NVME_CMD_ZONE_APPEND:
4467         return nvme_zone_append(n, req);
4468     case NVME_CMD_WRITE:
4469         return nvme_write(n, req);
4470     case NVME_CMD_READ:
4471         return nvme_read(n, req);
4472     case NVME_CMD_COMPARE:
4473         return nvme_compare(n, req);
4474     case NVME_CMD_DSM:
4475         return nvme_dsm(n, req);
4476     case NVME_CMD_VERIFY:
4477         return nvme_verify(n, req);
4478     case NVME_CMD_COPY:
4479         return nvme_copy(n, req);
4480     case NVME_CMD_ZONE_MGMT_SEND:
4481         return nvme_zone_mgmt_send(n, req);
4482     case NVME_CMD_ZONE_MGMT_RECV:
4483         return nvme_zone_mgmt_recv(n, req);
4484     case NVME_CMD_IO_MGMT_RECV:
4485         return nvme_io_mgmt_recv(n, req);
4486     case NVME_CMD_IO_MGMT_SEND:
4487         return nvme_io_mgmt_send(n, req);
4488     default:
4489         assert(false);
4490     }
4491 
4492     return NVME_INVALID_OPCODE | NVME_DNR;
4493 }
4494 
4495 static void nvme_cq_notifier(EventNotifier *e)
4496 {
4497     NvmeCQueue *cq = container_of(e, NvmeCQueue, notifier);
4498     NvmeCtrl *n = cq->ctrl;
4499 
4500     if (!event_notifier_test_and_clear(e)) {
4501         return;
4502     }
4503 
4504     nvme_update_cq_head(cq);
4505 
4506     if (cq->tail == cq->head) {
4507         if (cq->irq_enabled) {
4508             n->cq_pending--;
4509         }
4510 
4511         nvme_irq_deassert(n, cq);
4512     }
4513 
4514     qemu_bh_schedule(cq->bh);
4515 }
4516 
4517 static int nvme_init_cq_ioeventfd(NvmeCQueue *cq)
4518 {
4519     NvmeCtrl *n = cq->ctrl;
4520     uint16_t offset = (cq->cqid << 3) + (1 << 2);
4521     int ret;
4522 
4523     ret = event_notifier_init(&cq->notifier, 0);
4524     if (ret < 0) {
4525         return ret;
4526     }
4527 
4528     event_notifier_set_handler(&cq->notifier, nvme_cq_notifier);
4529     memory_region_add_eventfd(&n->iomem,
4530                               0x1000 + offset, 4, false, 0, &cq->notifier);
4531 
4532     return 0;
4533 }
4534 
4535 static void nvme_sq_notifier(EventNotifier *e)
4536 {
4537     NvmeSQueue *sq = container_of(e, NvmeSQueue, notifier);
4538 
4539     if (!event_notifier_test_and_clear(e)) {
4540         return;
4541     }
4542 
4543     nvme_process_sq(sq);
4544 }
4545 
4546 static int nvme_init_sq_ioeventfd(NvmeSQueue *sq)
4547 {
4548     NvmeCtrl *n = sq->ctrl;
4549     uint16_t offset = sq->sqid << 3;
4550     int ret;
4551 
4552     ret = event_notifier_init(&sq->notifier, 0);
4553     if (ret < 0) {
4554         return ret;
4555     }
4556 
4557     event_notifier_set_handler(&sq->notifier, nvme_sq_notifier);
4558     memory_region_add_eventfd(&n->iomem,
4559                               0x1000 + offset, 4, false, 0, &sq->notifier);
4560 
4561     return 0;
4562 }
4563 
4564 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
4565 {
4566     uint16_t offset = sq->sqid << 3;
4567 
4568     n->sq[sq->sqid] = NULL;
4569     qemu_bh_delete(sq->bh);
4570     if (sq->ioeventfd_enabled) {
4571         memory_region_del_eventfd(&n->iomem,
4572                                   0x1000 + offset, 4, false, 0, &sq->notifier);
4573         event_notifier_set_handler(&sq->notifier, NULL);
4574         event_notifier_cleanup(&sq->notifier);
4575     }
4576     g_free(sq->io_req);
4577     if (sq->sqid) {
4578         g_free(sq);
4579     }
4580 }
4581 
4582 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
4583 {
4584     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4585     NvmeRequest *r, *next;
4586     NvmeSQueue *sq;
4587     NvmeCQueue *cq;
4588     uint16_t qid = le16_to_cpu(c->qid);
4589 
4590     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
4591         trace_pci_nvme_err_invalid_del_sq(qid);
4592         return NVME_INVALID_QID | NVME_DNR;
4593     }
4594 
4595     trace_pci_nvme_del_sq(qid);
4596 
4597     sq = n->sq[qid];
4598     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
4599         r = QTAILQ_FIRST(&sq->out_req_list);
4600         assert(r->aiocb);
4601         blk_aio_cancel(r->aiocb);
4602     }
4603 
4604     assert(QTAILQ_EMPTY(&sq->out_req_list));
4605 
4606     if (!nvme_check_cqid(n, sq->cqid)) {
4607         cq = n->cq[sq->cqid];
4608         QTAILQ_REMOVE(&cq->sq_list, sq, entry);
4609 
4610         nvme_post_cqes(cq);
4611         QTAILQ_FOREACH_SAFE(r, &cq->req_list, entry, next) {
4612             if (r->sq == sq) {
4613                 QTAILQ_REMOVE(&cq->req_list, r, entry);
4614                 QTAILQ_INSERT_TAIL(&sq->req_list, r, entry);
4615             }
4616         }
4617     }
4618 
4619     nvme_free_sq(sq, n);
4620     return NVME_SUCCESS;
4621 }
4622 
4623 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
4624                          uint16_t sqid, uint16_t cqid, uint16_t size)
4625 {
4626     int i;
4627     NvmeCQueue *cq;
4628 
4629     sq->ctrl = n;
4630     sq->dma_addr = dma_addr;
4631     sq->sqid = sqid;
4632     sq->size = size;
4633     sq->cqid = cqid;
4634     sq->head = sq->tail = 0;
4635     sq->io_req = g_new0(NvmeRequest, sq->size);
4636 
4637     QTAILQ_INIT(&sq->req_list);
4638     QTAILQ_INIT(&sq->out_req_list);
4639     for (i = 0; i < sq->size; i++) {
4640         sq->io_req[i].sq = sq;
4641         QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
4642     }
4643 
4644     sq->bh = qemu_bh_new_guarded(nvme_process_sq, sq,
4645                                  &DEVICE(sq->ctrl)->mem_reentrancy_guard);
4646 
4647     if (n->dbbuf_enabled) {
4648         sq->db_addr = n->dbbuf_dbs + (sqid << 3);
4649         sq->ei_addr = n->dbbuf_eis + (sqid << 3);
4650 
4651         if (n->params.ioeventfd && sq->sqid != 0) {
4652             if (!nvme_init_sq_ioeventfd(sq)) {
4653                 sq->ioeventfd_enabled = true;
4654             }
4655         }
4656     }
4657 
4658     assert(n->cq[cqid]);
4659     cq = n->cq[cqid];
4660     QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
4661     n->sq[sqid] = sq;
4662 }
4663 
4664 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeRequest *req)
4665 {
4666     NvmeSQueue *sq;
4667     NvmeCreateSq *c = (NvmeCreateSq *)&req->cmd;
4668 
4669     uint16_t cqid = le16_to_cpu(c->cqid);
4670     uint16_t sqid = le16_to_cpu(c->sqid);
4671     uint16_t qsize = le16_to_cpu(c->qsize);
4672     uint16_t qflags = le16_to_cpu(c->sq_flags);
4673     uint64_t prp1 = le64_to_cpu(c->prp1);
4674 
4675     trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
4676 
4677     if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
4678         trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
4679         return NVME_INVALID_CQID | NVME_DNR;
4680     }
4681     if (unlikely(!sqid || sqid > n->conf_ioqpairs || n->sq[sqid] != NULL)) {
4682         trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
4683         return NVME_INVALID_QID | NVME_DNR;
4684     }
4685     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
4686         trace_pci_nvme_err_invalid_create_sq_size(qsize);
4687         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4688     }
4689     if (unlikely(prp1 & (n->page_size - 1))) {
4690         trace_pci_nvme_err_invalid_create_sq_addr(prp1);
4691         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4692     }
4693     if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
4694         trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
4695         return NVME_INVALID_FIELD | NVME_DNR;
4696     }
4697     sq = g_malloc0(sizeof(*sq));
4698     nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
4699     return NVME_SUCCESS;
4700 }
4701 
4702 struct nvme_stats {
4703     uint64_t units_read;
4704     uint64_t units_written;
4705     uint64_t read_commands;
4706     uint64_t write_commands;
4707 };
4708 
4709 static void nvme_set_blk_stats(NvmeNamespace *ns, struct nvme_stats *stats)
4710 {
4711     BlockAcctStats *s = blk_get_stats(ns->blkconf.blk);
4712 
4713     stats->units_read += s->nr_bytes[BLOCK_ACCT_READ];
4714     stats->units_written += s->nr_bytes[BLOCK_ACCT_WRITE];
4715     stats->read_commands += s->nr_ops[BLOCK_ACCT_READ];
4716     stats->write_commands += s->nr_ops[BLOCK_ACCT_WRITE];
4717 }
4718 
4719 static uint16_t nvme_smart_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4720                                 uint64_t off, NvmeRequest *req)
4721 {
4722     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4723     struct nvme_stats stats = { 0 };
4724     NvmeSmartLog smart = { 0 };
4725     uint32_t trans_len;
4726     NvmeNamespace *ns;
4727     time_t current_ms;
4728     uint64_t u_read, u_written;
4729 
4730     if (off >= sizeof(smart)) {
4731         return NVME_INVALID_FIELD | NVME_DNR;
4732     }
4733 
4734     if (nsid != 0xffffffff) {
4735         ns = nvme_ns(n, nsid);
4736         if (!ns) {
4737             return NVME_INVALID_NSID | NVME_DNR;
4738         }
4739         nvme_set_blk_stats(ns, &stats);
4740     } else {
4741         int i;
4742 
4743         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4744             ns = nvme_ns(n, i);
4745             if (!ns) {
4746                 continue;
4747             }
4748             nvme_set_blk_stats(ns, &stats);
4749         }
4750     }
4751 
4752     trans_len = MIN(sizeof(smart) - off, buf_len);
4753     smart.critical_warning = n->smart_critical_warning;
4754 
4755     u_read = DIV_ROUND_UP(stats.units_read >> BDRV_SECTOR_BITS, 1000);
4756     u_written = DIV_ROUND_UP(stats.units_written >> BDRV_SECTOR_BITS, 1000);
4757 
4758     smart.data_units_read[0] = cpu_to_le64(u_read);
4759     smart.data_units_written[0] = cpu_to_le64(u_written);
4760     smart.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4761     smart.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4762 
4763     smart.temperature = cpu_to_le16(n->temperature);
4764 
4765     if ((n->temperature >= n->features.temp_thresh_hi) ||
4766         (n->temperature <= n->features.temp_thresh_low)) {
4767         smart.critical_warning |= NVME_SMART_TEMPERATURE;
4768     }
4769 
4770     current_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4771     smart.power_on_hours[0] =
4772         cpu_to_le64((((current_ms - n->starttime_ms) / 1000) / 60) / 60);
4773 
4774     if (!rae) {
4775         nvme_clear_events(n, NVME_AER_TYPE_SMART);
4776     }
4777 
4778     return nvme_c2h(n, (uint8_t *) &smart + off, trans_len, req);
4779 }
4780 
4781 static uint16_t nvme_endgrp_info(NvmeCtrl *n,  uint8_t rae, uint32_t buf_len,
4782                                  uint64_t off, NvmeRequest *req)
4783 {
4784     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
4785     uint16_t endgrpid = (dw11 >> 16) & 0xffff;
4786     struct nvme_stats stats = {};
4787     NvmeEndGrpLog info = {};
4788     int i;
4789 
4790     if (!n->subsys || endgrpid != 0x1) {
4791         return NVME_INVALID_FIELD | NVME_DNR;
4792     }
4793 
4794     if (off >= sizeof(info)) {
4795         return NVME_INVALID_FIELD | NVME_DNR;
4796     }
4797 
4798     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4799         NvmeNamespace *ns = nvme_subsys_ns(n->subsys, i);
4800         if (!ns) {
4801             continue;
4802         }
4803 
4804         nvme_set_blk_stats(ns, &stats);
4805     }
4806 
4807     info.data_units_read[0] =
4808         cpu_to_le64(DIV_ROUND_UP(stats.units_read / 1000000000, 1000000000));
4809     info.data_units_written[0] =
4810         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4811     info.media_units_written[0] =
4812         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4813 
4814     info.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4815     info.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4816 
4817     buf_len = MIN(sizeof(info) - off, buf_len);
4818 
4819     return nvme_c2h(n, (uint8_t *)&info + off, buf_len, req);
4820 }
4821 
4822 
4823 static uint16_t nvme_fw_log_info(NvmeCtrl *n, uint32_t buf_len, uint64_t off,
4824                                  NvmeRequest *req)
4825 {
4826     uint32_t trans_len;
4827     NvmeFwSlotInfoLog fw_log = {
4828         .afi = 0x1,
4829     };
4830 
4831     if (off >= sizeof(fw_log)) {
4832         return NVME_INVALID_FIELD | NVME_DNR;
4833     }
4834 
4835     strpadcpy((char *)&fw_log.frs1, sizeof(fw_log.frs1), "1.0", ' ');
4836     trans_len = MIN(sizeof(fw_log) - off, buf_len);
4837 
4838     return nvme_c2h(n, (uint8_t *) &fw_log + off, trans_len, req);
4839 }
4840 
4841 static uint16_t nvme_error_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4842                                 uint64_t off, NvmeRequest *req)
4843 {
4844     uint32_t trans_len;
4845     NvmeErrorLog errlog;
4846 
4847     if (off >= sizeof(errlog)) {
4848         return NVME_INVALID_FIELD | NVME_DNR;
4849     }
4850 
4851     if (!rae) {
4852         nvme_clear_events(n, NVME_AER_TYPE_ERROR);
4853     }
4854 
4855     memset(&errlog, 0x0, sizeof(errlog));
4856     trans_len = MIN(sizeof(errlog) - off, buf_len);
4857 
4858     return nvme_c2h(n, (uint8_t *)&errlog, trans_len, req);
4859 }
4860 
4861 static uint16_t nvme_changed_nslist(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4862                                     uint64_t off, NvmeRequest *req)
4863 {
4864     uint32_t nslist[1024];
4865     uint32_t trans_len;
4866     int i = 0;
4867     uint32_t nsid;
4868 
4869     if (off >= sizeof(nslist)) {
4870         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(nslist));
4871         return NVME_INVALID_FIELD | NVME_DNR;
4872     }
4873 
4874     memset(nslist, 0x0, sizeof(nslist));
4875     trans_len = MIN(sizeof(nslist) - off, buf_len);
4876 
4877     while ((nsid = find_first_bit(n->changed_nsids, NVME_CHANGED_NSID_SIZE)) !=
4878             NVME_CHANGED_NSID_SIZE) {
4879         /*
4880          * If more than 1024 namespaces, the first entry in the log page should
4881          * be set to FFFFFFFFh and the others to 0 as spec.
4882          */
4883         if (i == ARRAY_SIZE(nslist)) {
4884             memset(nslist, 0x0, sizeof(nslist));
4885             nslist[0] = 0xffffffff;
4886             break;
4887         }
4888 
4889         nslist[i++] = nsid;
4890         clear_bit(nsid, n->changed_nsids);
4891     }
4892 
4893     /*
4894      * Remove all the remaining list entries in case returns directly due to
4895      * more than 1024 namespaces.
4896      */
4897     if (nslist[0] == 0xffffffff) {
4898         bitmap_zero(n->changed_nsids, NVME_CHANGED_NSID_SIZE);
4899     }
4900 
4901     if (!rae) {
4902         nvme_clear_events(n, NVME_AER_TYPE_NOTICE);
4903     }
4904 
4905     return nvme_c2h(n, ((uint8_t *)nslist) + off, trans_len, req);
4906 }
4907 
4908 static uint16_t nvme_cmd_effects(NvmeCtrl *n, uint8_t csi, uint32_t buf_len,
4909                                  uint64_t off, NvmeRequest *req)
4910 {
4911     NvmeEffectsLog log = {};
4912     const uint32_t *src_iocs = NULL;
4913     uint32_t trans_len;
4914 
4915     if (off >= sizeof(log)) {
4916         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(log));
4917         return NVME_INVALID_FIELD | NVME_DNR;
4918     }
4919 
4920     switch (NVME_CC_CSS(ldl_le_p(&n->bar.cc))) {
4921     case NVME_CC_CSS_NVM:
4922         src_iocs = nvme_cse_iocs_nvm;
4923         /* fall through */
4924     case NVME_CC_CSS_ADMIN_ONLY:
4925         break;
4926     case NVME_CC_CSS_CSI:
4927         switch (csi) {
4928         case NVME_CSI_NVM:
4929             src_iocs = nvme_cse_iocs_nvm;
4930             break;
4931         case NVME_CSI_ZONED:
4932             src_iocs = nvme_cse_iocs_zoned;
4933             break;
4934         }
4935     }
4936 
4937     memcpy(log.acs, nvme_cse_acs, sizeof(nvme_cse_acs));
4938 
4939     if (src_iocs) {
4940         memcpy(log.iocs, src_iocs, sizeof(log.iocs));
4941     }
4942 
4943     trans_len = MIN(sizeof(log) - off, buf_len);
4944 
4945     return nvme_c2h(n, ((uint8_t *)&log) + off, trans_len, req);
4946 }
4947 
4948 static size_t sizeof_fdp_conf_descr(size_t nruh, size_t vss)
4949 {
4950     size_t entry_siz = sizeof(NvmeFdpDescrHdr) + nruh * sizeof(NvmeRuhDescr)
4951                        + vss;
4952     return ROUND_UP(entry_siz, 8);
4953 }
4954 
4955 static uint16_t nvme_fdp_confs(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
4956                                uint64_t off, NvmeRequest *req)
4957 {
4958     uint32_t log_size, trans_len;
4959     g_autofree uint8_t *buf = NULL;
4960     NvmeFdpDescrHdr *hdr;
4961     NvmeRuhDescr *ruhd;
4962     NvmeEnduranceGroup *endgrp;
4963     NvmeFdpConfsHdr *log;
4964     size_t nruh, fdp_descr_size;
4965     int i;
4966 
4967     if (endgrpid != 1 || !n->subsys) {
4968         return NVME_INVALID_FIELD | NVME_DNR;
4969     }
4970 
4971     endgrp = &n->subsys->endgrp;
4972 
4973     if (endgrp->fdp.enabled) {
4974         nruh = endgrp->fdp.nruh;
4975     } else {
4976         nruh = 1;
4977     }
4978 
4979     fdp_descr_size = sizeof_fdp_conf_descr(nruh, FDPVSS);
4980     log_size = sizeof(NvmeFdpConfsHdr) + fdp_descr_size;
4981 
4982     if (off >= log_size) {
4983         return NVME_INVALID_FIELD | NVME_DNR;
4984     }
4985 
4986     trans_len = MIN(log_size - off, buf_len);
4987 
4988     buf = g_malloc0(log_size);
4989     log = (NvmeFdpConfsHdr *)buf;
4990     hdr = (NvmeFdpDescrHdr *)(log + 1);
4991     ruhd = (NvmeRuhDescr *)(buf + sizeof(*log) + sizeof(*hdr));
4992 
4993     log->num_confs = cpu_to_le16(0);
4994     log->size = cpu_to_le32(log_size);
4995 
4996     hdr->descr_size = cpu_to_le16(fdp_descr_size);
4997     if (endgrp->fdp.enabled) {
4998         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, VALID, 1);
4999         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, RGIF, endgrp->fdp.rgif);
5000         hdr->nrg = cpu_to_le16(endgrp->fdp.nrg);
5001         hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5002         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5003         hdr->nnss = cpu_to_le32(NVME_MAX_NAMESPACES);
5004         hdr->runs = cpu_to_le64(endgrp->fdp.runs);
5005 
5006         for (i = 0; i < nruh; i++) {
5007             ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5008             ruhd++;
5009         }
5010     } else {
5011         /* 1 bit for RUH in PIF -> 2 RUHs max. */
5012         hdr->nrg = cpu_to_le16(1);
5013         hdr->nruh = cpu_to_le16(1);
5014         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5015         hdr->nnss = cpu_to_le32(1);
5016         hdr->runs = cpu_to_le64(96 * MiB);
5017 
5018         ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5019     }
5020 
5021     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5022 }
5023 
5024 static uint16_t nvme_fdp_ruh_usage(NvmeCtrl *n, uint32_t endgrpid,
5025                                    uint32_t dw10, uint32_t dw12,
5026                                    uint32_t buf_len, uint64_t off,
5027                                    NvmeRequest *req)
5028 {
5029     NvmeRuHandle *ruh;
5030     NvmeRuhuLog *hdr;
5031     NvmeRuhuDescr *ruhud;
5032     NvmeEnduranceGroup *endgrp;
5033     g_autofree uint8_t *buf = NULL;
5034     uint32_t log_size, trans_len;
5035     uint16_t i;
5036 
5037     if (endgrpid != 1 || !n->subsys) {
5038         return NVME_INVALID_FIELD | NVME_DNR;
5039     }
5040 
5041     endgrp = &n->subsys->endgrp;
5042 
5043     if (!endgrp->fdp.enabled) {
5044         return NVME_FDP_DISABLED | NVME_DNR;
5045     }
5046 
5047     log_size = sizeof(NvmeRuhuLog) + endgrp->fdp.nruh * sizeof(NvmeRuhuDescr);
5048 
5049     if (off >= log_size) {
5050         return NVME_INVALID_FIELD | NVME_DNR;
5051     }
5052 
5053     trans_len = MIN(log_size - off, buf_len);
5054 
5055     buf = g_malloc0(log_size);
5056     hdr = (NvmeRuhuLog *)buf;
5057     ruhud = (NvmeRuhuDescr *)(hdr + 1);
5058 
5059     ruh = endgrp->fdp.ruhs;
5060     hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5061 
5062     for (i = 0; i < endgrp->fdp.nruh; i++, ruhud++, ruh++) {
5063         ruhud->ruha = ruh->ruha;
5064     }
5065 
5066     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5067 }
5068 
5069 static uint16_t nvme_fdp_stats(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
5070                                uint64_t off, NvmeRequest *req)
5071 {
5072     NvmeEnduranceGroup *endgrp;
5073     NvmeFdpStatsLog log = {};
5074     uint32_t trans_len;
5075 
5076     if (off >= sizeof(NvmeFdpStatsLog)) {
5077         return NVME_INVALID_FIELD | NVME_DNR;
5078     }
5079 
5080     if (endgrpid != 1 || !n->subsys) {
5081         return NVME_INVALID_FIELD | NVME_DNR;
5082     }
5083 
5084     if (!n->subsys->endgrp.fdp.enabled) {
5085         return NVME_FDP_DISABLED | NVME_DNR;
5086     }
5087 
5088     endgrp = &n->subsys->endgrp;
5089 
5090     trans_len = MIN(sizeof(log) - off, buf_len);
5091 
5092     /* spec value is 128 bit, we only use 64 bit */
5093     log.hbmw[0] = cpu_to_le64(endgrp->fdp.hbmw);
5094     log.mbmw[0] = cpu_to_le64(endgrp->fdp.mbmw);
5095     log.mbe[0] = cpu_to_le64(endgrp->fdp.mbe);
5096 
5097     return nvme_c2h(n, (uint8_t *)&log + off, trans_len, req);
5098 }
5099 
5100 static uint16_t nvme_fdp_events(NvmeCtrl *n, uint32_t endgrpid,
5101                                 uint32_t buf_len, uint64_t off,
5102                                 NvmeRequest *req)
5103 {
5104     NvmeEnduranceGroup *endgrp;
5105     NvmeCmd *cmd = &req->cmd;
5106     bool host_events = (cmd->cdw10 >> 8) & 0x1;
5107     uint32_t log_size, trans_len;
5108     NvmeFdpEventBuffer *ebuf;
5109     g_autofree NvmeFdpEventsLog *elog = NULL;
5110     NvmeFdpEvent *event;
5111 
5112     if (endgrpid != 1 || !n->subsys) {
5113         return NVME_INVALID_FIELD | NVME_DNR;
5114     }
5115 
5116     endgrp = &n->subsys->endgrp;
5117 
5118     if (!endgrp->fdp.enabled) {
5119         return NVME_FDP_DISABLED | NVME_DNR;
5120     }
5121 
5122     if (host_events) {
5123         ebuf = &endgrp->fdp.host_events;
5124     } else {
5125         ebuf = &endgrp->fdp.ctrl_events;
5126     }
5127 
5128     log_size = sizeof(NvmeFdpEventsLog) + ebuf->nelems * sizeof(NvmeFdpEvent);
5129 
5130     if (off >= log_size) {
5131         return NVME_INVALID_FIELD | NVME_DNR;
5132     }
5133 
5134     trans_len = MIN(log_size - off, buf_len);
5135     elog = g_malloc0(log_size);
5136     elog->num_events = cpu_to_le32(ebuf->nelems);
5137     event = (NvmeFdpEvent *)(elog + 1);
5138 
5139     if (ebuf->nelems && ebuf->start == ebuf->next) {
5140         unsigned int nelems = (NVME_FDP_MAX_EVENTS - ebuf->start);
5141         /* wrap over, copy [start;NVME_FDP_MAX_EVENTS[ and [0; next[ */
5142         memcpy(event, &ebuf->events[ebuf->start],
5143                sizeof(NvmeFdpEvent) * nelems);
5144         memcpy(event + nelems, ebuf->events,
5145                sizeof(NvmeFdpEvent) * ebuf->next);
5146     } else if (ebuf->start < ebuf->next) {
5147         memcpy(event, &ebuf->events[ebuf->start],
5148                sizeof(NvmeFdpEvent) * (ebuf->next - ebuf->start));
5149     }
5150 
5151     return nvme_c2h(n, (uint8_t *)elog + off, trans_len, req);
5152 }
5153 
5154 static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
5155 {
5156     NvmeCmd *cmd = &req->cmd;
5157 
5158     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5159     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5160     uint32_t dw12 = le32_to_cpu(cmd->cdw12);
5161     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
5162     uint8_t  lid = dw10 & 0xff;
5163     uint8_t  lsp = (dw10 >> 8) & 0xf;
5164     uint8_t  rae = (dw10 >> 15) & 0x1;
5165     uint8_t  csi = le32_to_cpu(cmd->cdw14) >> 24;
5166     uint32_t numdl, numdu, lspi;
5167     uint64_t off, lpol, lpou;
5168     size_t   len;
5169     uint16_t status;
5170 
5171     numdl = (dw10 >> 16);
5172     numdu = (dw11 & 0xffff);
5173     lspi = (dw11 >> 16);
5174     lpol = dw12;
5175     lpou = dw13;
5176 
5177     len = (((numdu << 16) | numdl) + 1) << 2;
5178     off = (lpou << 32ULL) | lpol;
5179 
5180     if (off & 0x3) {
5181         return NVME_INVALID_FIELD | NVME_DNR;
5182     }
5183 
5184     trace_pci_nvme_get_log(nvme_cid(req), lid, lsp, rae, len, off);
5185 
5186     status = nvme_check_mdts(n, len);
5187     if (status) {
5188         return status;
5189     }
5190 
5191     switch (lid) {
5192     case NVME_LOG_ERROR_INFO:
5193         return nvme_error_info(n, rae, len, off, req);
5194     case NVME_LOG_SMART_INFO:
5195         return nvme_smart_info(n, rae, len, off, req);
5196     case NVME_LOG_FW_SLOT_INFO:
5197         return nvme_fw_log_info(n, len, off, req);
5198     case NVME_LOG_CHANGED_NSLIST:
5199         return nvme_changed_nslist(n, rae, len, off, req);
5200     case NVME_LOG_CMD_EFFECTS:
5201         return nvme_cmd_effects(n, csi, len, off, req);
5202     case NVME_LOG_ENDGRP:
5203         return nvme_endgrp_info(n, rae, len, off, req);
5204     case NVME_LOG_FDP_CONFS:
5205         return nvme_fdp_confs(n, lspi, len, off, req);
5206     case NVME_LOG_FDP_RUH_USAGE:
5207         return nvme_fdp_ruh_usage(n, lspi, dw10, dw12, len, off, req);
5208     case NVME_LOG_FDP_STATS:
5209         return nvme_fdp_stats(n, lspi, len, off, req);
5210     case NVME_LOG_FDP_EVENTS:
5211         return nvme_fdp_events(n, lspi, len, off, req);
5212     default:
5213         trace_pci_nvme_err_invalid_log_page(nvme_cid(req), lid);
5214         return NVME_INVALID_FIELD | NVME_DNR;
5215     }
5216 }
5217 
5218 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
5219 {
5220     PCIDevice *pci = PCI_DEVICE(n);
5221     uint16_t offset = (cq->cqid << 3) + (1 << 2);
5222 
5223     n->cq[cq->cqid] = NULL;
5224     qemu_bh_delete(cq->bh);
5225     if (cq->ioeventfd_enabled) {
5226         memory_region_del_eventfd(&n->iomem,
5227                                   0x1000 + offset, 4, false, 0, &cq->notifier);
5228         event_notifier_set_handler(&cq->notifier, NULL);
5229         event_notifier_cleanup(&cq->notifier);
5230     }
5231     if (msix_enabled(pci)) {
5232         msix_vector_unuse(pci, cq->vector);
5233     }
5234     if (cq->cqid) {
5235         g_free(cq);
5236     }
5237 }
5238 
5239 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)
5240 {
5241     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
5242     NvmeCQueue *cq;
5243     uint16_t qid = le16_to_cpu(c->qid);
5244 
5245     if (unlikely(!qid || nvme_check_cqid(n, qid))) {
5246         trace_pci_nvme_err_invalid_del_cq_cqid(qid);
5247         return NVME_INVALID_CQID | NVME_DNR;
5248     }
5249 
5250     cq = n->cq[qid];
5251     if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
5252         trace_pci_nvme_err_invalid_del_cq_notempty(qid);
5253         return NVME_INVALID_QUEUE_DEL;
5254     }
5255 
5256     if (cq->irq_enabled && cq->tail != cq->head) {
5257         n->cq_pending--;
5258     }
5259 
5260     nvme_irq_deassert(n, cq);
5261     trace_pci_nvme_del_cq(qid);
5262     nvme_free_cq(cq, n);
5263     return NVME_SUCCESS;
5264 }
5265 
5266 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
5267                          uint16_t cqid, uint16_t vector, uint16_t size,
5268                          uint16_t irq_enabled)
5269 {
5270     PCIDevice *pci = PCI_DEVICE(n);
5271 
5272     if (msix_enabled(pci)) {
5273         msix_vector_use(pci, vector);
5274     }
5275     cq->ctrl = n;
5276     cq->cqid = cqid;
5277     cq->size = size;
5278     cq->dma_addr = dma_addr;
5279     cq->phase = 1;
5280     cq->irq_enabled = irq_enabled;
5281     cq->vector = vector;
5282     cq->head = cq->tail = 0;
5283     QTAILQ_INIT(&cq->req_list);
5284     QTAILQ_INIT(&cq->sq_list);
5285     if (n->dbbuf_enabled) {
5286         cq->db_addr = n->dbbuf_dbs + (cqid << 3) + (1 << 2);
5287         cq->ei_addr = n->dbbuf_eis + (cqid << 3) + (1 << 2);
5288 
5289         if (n->params.ioeventfd && cqid != 0) {
5290             if (!nvme_init_cq_ioeventfd(cq)) {
5291                 cq->ioeventfd_enabled = true;
5292             }
5293         }
5294     }
5295     n->cq[cqid] = cq;
5296     cq->bh = qemu_bh_new_guarded(nvme_post_cqes, cq,
5297                                  &DEVICE(cq->ctrl)->mem_reentrancy_guard);
5298 }
5299 
5300 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeRequest *req)
5301 {
5302     NvmeCQueue *cq;
5303     NvmeCreateCq *c = (NvmeCreateCq *)&req->cmd;
5304     uint16_t cqid = le16_to_cpu(c->cqid);
5305     uint16_t vector = le16_to_cpu(c->irq_vector);
5306     uint16_t qsize = le16_to_cpu(c->qsize);
5307     uint16_t qflags = le16_to_cpu(c->cq_flags);
5308     uint64_t prp1 = le64_to_cpu(c->prp1);
5309     uint32_t cc = ldq_le_p(&n->bar.cc);
5310     uint8_t iocqes = NVME_CC_IOCQES(cc);
5311     uint8_t iosqes = NVME_CC_IOSQES(cc);
5312 
5313     trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
5314                              NVME_CQ_FLAGS_IEN(qflags) != 0);
5315 
5316     if (iosqes != NVME_SQES || iocqes != NVME_CQES) {
5317         trace_pci_nvme_err_invalid_create_cq_entry_size(iosqes, iocqes);
5318         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5319     }
5320 
5321     if (unlikely(!cqid || cqid > n->conf_ioqpairs || n->cq[cqid] != NULL)) {
5322         trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
5323         return NVME_INVALID_QID | NVME_DNR;
5324     }
5325     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
5326         trace_pci_nvme_err_invalid_create_cq_size(qsize);
5327         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5328     }
5329     if (unlikely(prp1 & (n->page_size - 1))) {
5330         trace_pci_nvme_err_invalid_create_cq_addr(prp1);
5331         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
5332     }
5333     if (unlikely(!msix_enabled(PCI_DEVICE(n)) && vector)) {
5334         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5335         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5336     }
5337     if (unlikely(vector >= n->conf_msix_qsize)) {
5338         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5339         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5340     }
5341     if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
5342         trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
5343         return NVME_INVALID_FIELD | NVME_DNR;
5344     }
5345 
5346     cq = g_malloc0(sizeof(*cq));
5347     nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
5348                  NVME_CQ_FLAGS_IEN(qflags));
5349 
5350     /*
5351      * It is only required to set qs_created when creating a completion queue;
5352      * creating a submission queue without a matching completion queue will
5353      * fail.
5354      */
5355     n->qs_created = true;
5356     return NVME_SUCCESS;
5357 }
5358 
5359 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl *n, NvmeRequest *req)
5360 {
5361     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5362 
5363     return nvme_c2h(n, id, sizeof(id), req);
5364 }
5365 
5366 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeRequest *req)
5367 {
5368     trace_pci_nvme_identify_ctrl();
5369 
5370     return nvme_c2h(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl), req);
5371 }
5372 
5373 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl *n, NvmeRequest *req)
5374 {
5375     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5376     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5377     NvmeIdCtrlNvm *id_nvm = (NvmeIdCtrlNvm *)&id;
5378 
5379     trace_pci_nvme_identify_ctrl_csi(c->csi);
5380 
5381     switch (c->csi) {
5382     case NVME_CSI_NVM:
5383         id_nvm->vsl = n->params.vsl;
5384         id_nvm->dmrsl = cpu_to_le32(n->dmrsl);
5385         break;
5386 
5387     case NVME_CSI_ZONED:
5388         ((NvmeIdCtrlZoned *)&id)->zasl = n->params.zasl;
5389         break;
5390 
5391     default:
5392         return NVME_INVALID_FIELD | NVME_DNR;
5393     }
5394 
5395     return nvme_c2h(n, id, sizeof(id), req);
5396 }
5397 
5398 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeRequest *req, bool active)
5399 {
5400     NvmeNamespace *ns;
5401     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5402     uint32_t nsid = le32_to_cpu(c->nsid);
5403 
5404     trace_pci_nvme_identify_ns(nsid);
5405 
5406     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5407         return NVME_INVALID_NSID | NVME_DNR;
5408     }
5409 
5410     ns = nvme_ns(n, nsid);
5411     if (unlikely(!ns)) {
5412         if (!active) {
5413             ns = nvme_subsys_ns(n->subsys, nsid);
5414             if (!ns) {
5415                 return nvme_rpt_empty_id_struct(n, req);
5416             }
5417         } else {
5418             return nvme_rpt_empty_id_struct(n, req);
5419         }
5420     }
5421 
5422     if (active || ns->csi == NVME_CSI_NVM) {
5423         return nvme_c2h(n, (uint8_t *)&ns->id_ns, sizeof(NvmeIdNs), req);
5424     }
5425 
5426     return NVME_INVALID_CMD_SET | NVME_DNR;
5427 }
5428 
5429 static uint16_t nvme_identify_ctrl_list(NvmeCtrl *n, NvmeRequest *req,
5430                                         bool attached)
5431 {
5432     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5433     uint32_t nsid = le32_to_cpu(c->nsid);
5434     uint16_t min_id = le16_to_cpu(c->ctrlid);
5435     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5436     uint16_t *ids = &list[1];
5437     NvmeNamespace *ns;
5438     NvmeCtrl *ctrl;
5439     int cntlid, nr_ids = 0;
5440 
5441     trace_pci_nvme_identify_ctrl_list(c->cns, min_id);
5442 
5443     if (!n->subsys) {
5444         return NVME_INVALID_FIELD | NVME_DNR;
5445     }
5446 
5447     if (attached) {
5448         if (nsid == NVME_NSID_BROADCAST) {
5449             return NVME_INVALID_FIELD | NVME_DNR;
5450         }
5451 
5452         ns = nvme_subsys_ns(n->subsys, nsid);
5453         if (!ns) {
5454             return NVME_INVALID_FIELD | NVME_DNR;
5455         }
5456     }
5457 
5458     for (cntlid = min_id; cntlid < ARRAY_SIZE(n->subsys->ctrls); cntlid++) {
5459         ctrl = nvme_subsys_ctrl(n->subsys, cntlid);
5460         if (!ctrl) {
5461             continue;
5462         }
5463 
5464         if (attached && !nvme_ns(ctrl, nsid)) {
5465             continue;
5466         }
5467 
5468         ids[nr_ids++] = cntlid;
5469     }
5470 
5471     list[0] = nr_ids;
5472 
5473     return nvme_c2h(n, (uint8_t *)list, sizeof(list), req);
5474 }
5475 
5476 static uint16_t nvme_identify_pri_ctrl_cap(NvmeCtrl *n, NvmeRequest *req)
5477 {
5478     trace_pci_nvme_identify_pri_ctrl_cap(le16_to_cpu(n->pri_ctrl_cap.cntlid));
5479 
5480     return nvme_c2h(n, (uint8_t *)&n->pri_ctrl_cap,
5481                     sizeof(NvmePriCtrlCap), req);
5482 }
5483 
5484 static uint16_t nvme_identify_sec_ctrl_list(NvmeCtrl *n, NvmeRequest *req)
5485 {
5486     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5487     uint16_t pri_ctrl_id = le16_to_cpu(n->pri_ctrl_cap.cntlid);
5488     uint16_t min_id = le16_to_cpu(c->ctrlid);
5489     uint8_t num_sec_ctrl = n->sec_ctrl_list.numcntl;
5490     NvmeSecCtrlList list = {0};
5491     uint8_t i;
5492 
5493     for (i = 0; i < num_sec_ctrl; i++) {
5494         if (n->sec_ctrl_list.sec[i].scid >= min_id) {
5495             list.numcntl = num_sec_ctrl - i;
5496             memcpy(&list.sec, n->sec_ctrl_list.sec + i,
5497                    list.numcntl * sizeof(NvmeSecCtrlEntry));
5498             break;
5499         }
5500     }
5501 
5502     trace_pci_nvme_identify_sec_ctrl_list(pri_ctrl_id, list.numcntl);
5503 
5504     return nvme_c2h(n, (uint8_t *)&list, sizeof(list), req);
5505 }
5506 
5507 static uint16_t nvme_identify_ns_csi(NvmeCtrl *n, NvmeRequest *req,
5508                                      bool active)
5509 {
5510     NvmeNamespace *ns;
5511     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5512     uint32_t nsid = le32_to_cpu(c->nsid);
5513 
5514     trace_pci_nvme_identify_ns_csi(nsid, c->csi);
5515 
5516     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5517         return NVME_INVALID_NSID | NVME_DNR;
5518     }
5519 
5520     ns = nvme_ns(n, nsid);
5521     if (unlikely(!ns)) {
5522         if (!active) {
5523             ns = nvme_subsys_ns(n->subsys, nsid);
5524             if (!ns) {
5525                 return nvme_rpt_empty_id_struct(n, req);
5526             }
5527         } else {
5528             return nvme_rpt_empty_id_struct(n, req);
5529         }
5530     }
5531 
5532     if (c->csi == NVME_CSI_NVM) {
5533         return nvme_c2h(n, (uint8_t *)&ns->id_ns_nvm, sizeof(NvmeIdNsNvm),
5534                         req);
5535     } else if (c->csi == NVME_CSI_ZONED && ns->csi == NVME_CSI_ZONED) {
5536         return nvme_c2h(n, (uint8_t *)ns->id_ns_zoned, sizeof(NvmeIdNsZoned),
5537                         req);
5538     }
5539 
5540     return NVME_INVALID_FIELD | NVME_DNR;
5541 }
5542 
5543 static uint16_t nvme_identify_nslist(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(min_nsid);
5555 
5556     /*
5557      * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
5558      * since the Active Namespace ID List should return namespaces with ids
5559      * *higher* than the NSID specified in the command. This is also specified
5560      * in the spec (NVM Express v1.3d, Section 5.15.4).
5561      */
5562     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5563         return NVME_INVALID_NSID | NVME_DNR;
5564     }
5565 
5566     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5567         ns = nvme_ns(n, i);
5568         if (!ns) {
5569             if (!active) {
5570                 ns = nvme_subsys_ns(n->subsys, i);
5571                 if (!ns) {
5572                     continue;
5573                 }
5574             } else {
5575                 continue;
5576             }
5577         }
5578         if (ns->params.nsid <= min_nsid) {
5579             continue;
5580         }
5581         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5582         if (j == data_len / sizeof(uint32_t)) {
5583             break;
5584         }
5585     }
5586 
5587     return nvme_c2h(n, list, data_len, req);
5588 }
5589 
5590 static uint16_t nvme_identify_nslist_csi(NvmeCtrl *n, NvmeRequest *req,
5591                                          bool active)
5592 {
5593     NvmeNamespace *ns;
5594     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5595     uint32_t min_nsid = le32_to_cpu(c->nsid);
5596     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5597     static const int data_len = sizeof(list);
5598     uint32_t *list_ptr = (uint32_t *)list;
5599     int i, j = 0;
5600 
5601     trace_pci_nvme_identify_nslist_csi(min_nsid, c->csi);
5602 
5603     /*
5604      * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
5605      */
5606     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5607         return NVME_INVALID_NSID | NVME_DNR;
5608     }
5609 
5610     if (c->csi != NVME_CSI_NVM && c->csi != NVME_CSI_ZONED) {
5611         return NVME_INVALID_FIELD | NVME_DNR;
5612     }
5613 
5614     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5615         ns = nvme_ns(n, i);
5616         if (!ns) {
5617             if (!active) {
5618                 ns = nvme_subsys_ns(n->subsys, i);
5619                 if (!ns) {
5620                     continue;
5621                 }
5622             } else {
5623                 continue;
5624             }
5625         }
5626         if (ns->params.nsid <= min_nsid || c->csi != ns->csi) {
5627             continue;
5628         }
5629         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5630         if (j == data_len / sizeof(uint32_t)) {
5631             break;
5632         }
5633     }
5634 
5635     return nvme_c2h(n, list, data_len, req);
5636 }
5637 
5638 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl *n, NvmeRequest *req)
5639 {
5640     NvmeNamespace *ns;
5641     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5642     uint32_t nsid = le32_to_cpu(c->nsid);
5643     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5644     uint8_t *pos = list;
5645     struct {
5646         NvmeIdNsDescr hdr;
5647         uint8_t v[NVME_NIDL_UUID];
5648     } QEMU_PACKED uuid = {};
5649     struct {
5650         NvmeIdNsDescr hdr;
5651         uint64_t v;
5652     } QEMU_PACKED eui64 = {};
5653     struct {
5654         NvmeIdNsDescr hdr;
5655         uint8_t v;
5656     } QEMU_PACKED csi = {};
5657 
5658     trace_pci_nvme_identify_ns_descr_list(nsid);
5659 
5660     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5661         return NVME_INVALID_NSID | NVME_DNR;
5662     }
5663 
5664     ns = nvme_ns(n, nsid);
5665     if (unlikely(!ns)) {
5666         return NVME_INVALID_FIELD | NVME_DNR;
5667     }
5668 
5669     if (!qemu_uuid_is_null(&ns->params.uuid)) {
5670         uuid.hdr.nidt = NVME_NIDT_UUID;
5671         uuid.hdr.nidl = NVME_NIDL_UUID;
5672         memcpy(uuid.v, ns->params.uuid.data, NVME_NIDL_UUID);
5673         memcpy(pos, &uuid, sizeof(uuid));
5674         pos += sizeof(uuid);
5675     }
5676 
5677     if (ns->params.eui64) {
5678         eui64.hdr.nidt = NVME_NIDT_EUI64;
5679         eui64.hdr.nidl = NVME_NIDL_EUI64;
5680         eui64.v = cpu_to_be64(ns->params.eui64);
5681         memcpy(pos, &eui64, sizeof(eui64));
5682         pos += sizeof(eui64);
5683     }
5684 
5685     csi.hdr.nidt = NVME_NIDT_CSI;
5686     csi.hdr.nidl = NVME_NIDL_CSI;
5687     csi.v = ns->csi;
5688     memcpy(pos, &csi, sizeof(csi));
5689     pos += sizeof(csi);
5690 
5691     return nvme_c2h(n, list, sizeof(list), req);
5692 }
5693 
5694 static uint16_t nvme_identify_cmd_set(NvmeCtrl *n, NvmeRequest *req)
5695 {
5696     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5697     static const int data_len = sizeof(list);
5698 
5699     trace_pci_nvme_identify_cmd_set();
5700 
5701     NVME_SET_CSI(*list, NVME_CSI_NVM);
5702     NVME_SET_CSI(*list, NVME_CSI_ZONED);
5703 
5704     return nvme_c2h(n, list, data_len, req);
5705 }
5706 
5707 static uint16_t nvme_identify(NvmeCtrl *n, NvmeRequest *req)
5708 {
5709     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5710 
5711     trace_pci_nvme_identify(nvme_cid(req), c->cns, le16_to_cpu(c->ctrlid),
5712                             c->csi);
5713 
5714     switch (c->cns) {
5715     case NVME_ID_CNS_NS:
5716         return nvme_identify_ns(n, req, true);
5717     case NVME_ID_CNS_NS_PRESENT:
5718         return nvme_identify_ns(n, req, false);
5719     case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST:
5720         return nvme_identify_ctrl_list(n, req, true);
5721     case NVME_ID_CNS_CTRL_LIST:
5722         return nvme_identify_ctrl_list(n, req, false);
5723     case NVME_ID_CNS_PRIMARY_CTRL_CAP:
5724         return nvme_identify_pri_ctrl_cap(n, req);
5725     case NVME_ID_CNS_SECONDARY_CTRL_LIST:
5726         return nvme_identify_sec_ctrl_list(n, req);
5727     case NVME_ID_CNS_CS_NS:
5728         return nvme_identify_ns_csi(n, req, true);
5729     case NVME_ID_CNS_CS_NS_PRESENT:
5730         return nvme_identify_ns_csi(n, req, false);
5731     case NVME_ID_CNS_CTRL:
5732         return nvme_identify_ctrl(n, req);
5733     case NVME_ID_CNS_CS_CTRL:
5734         return nvme_identify_ctrl_csi(n, req);
5735     case NVME_ID_CNS_NS_ACTIVE_LIST:
5736         return nvme_identify_nslist(n, req, true);
5737     case NVME_ID_CNS_NS_PRESENT_LIST:
5738         return nvme_identify_nslist(n, req, false);
5739     case NVME_ID_CNS_CS_NS_ACTIVE_LIST:
5740         return nvme_identify_nslist_csi(n, req, true);
5741     case NVME_ID_CNS_CS_NS_PRESENT_LIST:
5742         return nvme_identify_nslist_csi(n, req, false);
5743     case NVME_ID_CNS_NS_DESCR_LIST:
5744         return nvme_identify_ns_descr_list(n, req);
5745     case NVME_ID_CNS_IO_COMMAND_SET:
5746         return nvme_identify_cmd_set(n, req);
5747     default:
5748         trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
5749         return NVME_INVALID_FIELD | NVME_DNR;
5750     }
5751 }
5752 
5753 static uint16_t nvme_abort(NvmeCtrl *n, NvmeRequest *req)
5754 {
5755     uint16_t sqid = le32_to_cpu(req->cmd.cdw10) & 0xffff;
5756 
5757     req->cqe.result = 1;
5758     if (nvme_check_sqid(n, sqid)) {
5759         return NVME_INVALID_FIELD | NVME_DNR;
5760     }
5761 
5762     return NVME_SUCCESS;
5763 }
5764 
5765 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
5766 {
5767     trace_pci_nvme_setfeat_timestamp(ts);
5768 
5769     n->host_timestamp = le64_to_cpu(ts);
5770     n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5771 }
5772 
5773 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
5774 {
5775     uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5776     uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
5777 
5778     union nvme_timestamp {
5779         struct {
5780             uint64_t timestamp:48;
5781             uint64_t sync:1;
5782             uint64_t origin:3;
5783             uint64_t rsvd1:12;
5784         };
5785         uint64_t all;
5786     };
5787 
5788     union nvme_timestamp ts;
5789     ts.all = 0;
5790     ts.timestamp = n->host_timestamp + elapsed_time;
5791 
5792     /* If the host timestamp is non-zero, set the timestamp origin */
5793     ts.origin = n->host_timestamp ? 0x01 : 0x00;
5794 
5795     trace_pci_nvme_getfeat_timestamp(ts.all);
5796 
5797     return cpu_to_le64(ts.all);
5798 }
5799 
5800 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
5801 {
5802     uint64_t timestamp = nvme_get_timestamp(n);
5803 
5804     return nvme_c2h(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
5805 }
5806 
5807 static int nvme_get_feature_fdp(NvmeCtrl *n, uint32_t endgrpid,
5808                                 uint32_t *result)
5809 {
5810     *result = 0;
5811 
5812     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
5813         return NVME_INVALID_FIELD | NVME_DNR;
5814     }
5815 
5816     *result = FIELD_DP16(0, FEAT_FDP, FDPE, 1);
5817     *result = FIELD_DP16(*result, FEAT_FDP, CONF_NDX, 0);
5818 
5819     return NVME_SUCCESS;
5820 }
5821 
5822 static uint16_t nvme_get_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
5823                                             NvmeRequest *req, uint32_t *result)
5824 {
5825     NvmeCmd *cmd = &req->cmd;
5826     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
5827     uint16_t ph = cdw11 & 0xffff;
5828     uint8_t noet = (cdw11 >> 16) & 0xff;
5829     uint16_t ruhid, ret;
5830     uint32_t nentries = 0;
5831     uint8_t s_events_ndx = 0;
5832     size_t s_events_siz = sizeof(NvmeFdpEventDescr) * noet;
5833     g_autofree NvmeFdpEventDescr *s_events = g_malloc0(s_events_siz);
5834     NvmeRuHandle *ruh;
5835     NvmeFdpEventDescr *s_event;
5836 
5837     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
5838         return NVME_FDP_DISABLED | NVME_DNR;
5839     }
5840 
5841     if (!nvme_ph_valid(ns, ph)) {
5842         return NVME_INVALID_FIELD | NVME_DNR;
5843     }
5844 
5845     ruhid = ns->fdp.phs[ph];
5846     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
5847 
5848     assert(ruh);
5849 
5850     if (unlikely(noet == 0)) {
5851         return NVME_INVALID_FIELD | NVME_DNR;
5852     }
5853 
5854     for (uint8_t event_type = 0; event_type < FDP_EVT_MAX; event_type++) {
5855         uint8_t shift = nvme_fdp_evf_shifts[event_type];
5856         if (!shift && event_type) {
5857             /*
5858              * only first entry (event_type == 0) has a shift value of 0
5859              * other entries are simply unpopulated.
5860              */
5861             continue;
5862         }
5863 
5864         nentries++;
5865 
5866         s_event = &s_events[s_events_ndx];
5867         s_event->evt = event_type;
5868         s_event->evta = (ruh->event_filter >> shift) & 0x1;
5869 
5870         /* break if all `noet` entries are filled */
5871         if ((++s_events_ndx) == noet) {
5872             break;
5873         }
5874     }
5875 
5876     ret = nvme_c2h(n, s_events, s_events_siz, req);
5877     if (ret) {
5878         return ret;
5879     }
5880 
5881     *result = nentries;
5882     return NVME_SUCCESS;
5883 }
5884 
5885 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeRequest *req)
5886 {
5887     NvmeCmd *cmd = &req->cmd;
5888     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5889     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5890     uint32_t nsid = le32_to_cpu(cmd->nsid);
5891     uint32_t result;
5892     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
5893     NvmeGetFeatureSelect sel = NVME_GETFEAT_SELECT(dw10);
5894     uint16_t iv;
5895     NvmeNamespace *ns;
5896     int i;
5897     uint16_t endgrpid = 0, ret = NVME_SUCCESS;
5898 
5899     static const uint32_t nvme_feature_default[NVME_FID_MAX] = {
5900         [NVME_ARBITRATION] = NVME_ARB_AB_NOLIMIT,
5901     };
5902 
5903     trace_pci_nvme_getfeat(nvme_cid(req), nsid, fid, sel, dw11);
5904 
5905     if (!nvme_feature_support[fid]) {
5906         return NVME_INVALID_FIELD | NVME_DNR;
5907     }
5908 
5909     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
5910         if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5911             /*
5912              * The Reservation Notification Mask and Reservation Persistence
5913              * features require a status code of Invalid Field in Command when
5914              * NSID is FFFFFFFFh. Since the device does not support those
5915              * features we can always return Invalid Namespace or Format as we
5916              * should do for all other features.
5917              */
5918             return NVME_INVALID_NSID | NVME_DNR;
5919         }
5920 
5921         if (!nvme_ns(n, nsid)) {
5922             return NVME_INVALID_FIELD | NVME_DNR;
5923         }
5924     }
5925 
5926     switch (sel) {
5927     case NVME_GETFEAT_SELECT_CURRENT:
5928         break;
5929     case NVME_GETFEAT_SELECT_SAVED:
5930         /* no features are saveable by the controller; fallthrough */
5931     case NVME_GETFEAT_SELECT_DEFAULT:
5932         goto defaults;
5933     case NVME_GETFEAT_SELECT_CAP:
5934         result = nvme_feature_cap[fid];
5935         goto out;
5936     }
5937 
5938     switch (fid) {
5939     case NVME_TEMPERATURE_THRESHOLD:
5940         result = 0;
5941 
5942         /*
5943          * The controller only implements the Composite Temperature sensor, so
5944          * return 0 for all other sensors.
5945          */
5946         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
5947             goto out;
5948         }
5949 
5950         switch (NVME_TEMP_THSEL(dw11)) {
5951         case NVME_TEMP_THSEL_OVER:
5952             result = n->features.temp_thresh_hi;
5953             goto out;
5954         case NVME_TEMP_THSEL_UNDER:
5955             result = n->features.temp_thresh_low;
5956             goto out;
5957         }
5958 
5959         return NVME_INVALID_FIELD | NVME_DNR;
5960     case NVME_ERROR_RECOVERY:
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         result = ns->features.err_rec;
5971         goto out;
5972     case NVME_VOLATILE_WRITE_CACHE:
5973         result = 0;
5974         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5975             ns = nvme_ns(n, i);
5976             if (!ns) {
5977                 continue;
5978             }
5979 
5980             result = blk_enable_write_cache(ns->blkconf.blk);
5981             if (result) {
5982                 break;
5983             }
5984         }
5985         trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
5986         goto out;
5987     case NVME_ASYNCHRONOUS_EVENT_CONF:
5988         result = n->features.async_config;
5989         goto out;
5990     case NVME_TIMESTAMP:
5991         return nvme_get_feature_timestamp(n, req);
5992     case NVME_HOST_BEHAVIOR_SUPPORT:
5993         return nvme_c2h(n, (uint8_t *)&n->features.hbs,
5994                         sizeof(n->features.hbs), req);
5995     case NVME_FDP_MODE:
5996         endgrpid = dw11 & 0xff;
5997 
5998         if (endgrpid != 0x1) {
5999             return NVME_INVALID_FIELD | NVME_DNR;
6000         }
6001 
6002         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6003         if (ret) {
6004             return ret;
6005         }
6006         goto out;
6007     case NVME_FDP_EVENTS:
6008         if (!nvme_nsid_valid(n, nsid)) {
6009             return NVME_INVALID_NSID | NVME_DNR;
6010         }
6011 
6012         ns = nvme_ns(n, nsid);
6013         if (unlikely(!ns)) {
6014             return NVME_INVALID_FIELD | NVME_DNR;
6015         }
6016 
6017         ret = nvme_get_feature_fdp_events(n, ns, req, &result);
6018         if (ret) {
6019             return ret;
6020         }
6021         goto out;
6022     default:
6023         break;
6024     }
6025 
6026 defaults:
6027     switch (fid) {
6028     case NVME_TEMPERATURE_THRESHOLD:
6029         result = 0;
6030 
6031         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6032             break;
6033         }
6034 
6035         if (NVME_TEMP_THSEL(dw11) == NVME_TEMP_THSEL_OVER) {
6036             result = NVME_TEMPERATURE_WARNING;
6037         }
6038 
6039         break;
6040     case NVME_NUMBER_OF_QUEUES:
6041         result = (n->conf_ioqpairs - 1) | ((n->conf_ioqpairs - 1) << 16);
6042         trace_pci_nvme_getfeat_numq(result);
6043         break;
6044     case NVME_INTERRUPT_VECTOR_CONF:
6045         iv = dw11 & 0xffff;
6046         if (iv >= n->conf_ioqpairs + 1) {
6047             return NVME_INVALID_FIELD | NVME_DNR;
6048         }
6049 
6050         result = iv;
6051         if (iv == n->admin_cq.vector) {
6052             result |= NVME_INTVC_NOCOALESCING;
6053         }
6054         break;
6055     case NVME_FDP_MODE:
6056         endgrpid = dw11 & 0xff;
6057 
6058         if (endgrpid != 0x1) {
6059             return NVME_INVALID_FIELD | NVME_DNR;
6060         }
6061 
6062         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6063         if (ret) {
6064             return ret;
6065         }
6066         goto out;
6067 
6068         break;
6069     default:
6070         result = nvme_feature_default[fid];
6071         break;
6072     }
6073 
6074 out:
6075     req->cqe.result = cpu_to_le32(result);
6076     return ret;
6077 }
6078 
6079 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
6080 {
6081     uint16_t ret;
6082     uint64_t timestamp;
6083 
6084     ret = nvme_h2c(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
6085     if (ret) {
6086         return ret;
6087     }
6088 
6089     nvme_set_timestamp(n, timestamp);
6090 
6091     return NVME_SUCCESS;
6092 }
6093 
6094 static uint16_t nvme_set_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
6095                                             NvmeRequest *req)
6096 {
6097     NvmeCmd *cmd = &req->cmd;
6098     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
6099     uint16_t ph = cdw11 & 0xffff;
6100     uint8_t noet = (cdw11 >> 16) & 0xff;
6101     uint16_t ret, ruhid;
6102     uint8_t enable = le32_to_cpu(cmd->cdw12) & 0x1;
6103     uint8_t event_mask = 0;
6104     unsigned int i;
6105     g_autofree uint8_t *events = g_malloc0(noet);
6106     NvmeRuHandle *ruh = NULL;
6107 
6108     assert(ns);
6109 
6110     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6111         return NVME_FDP_DISABLED | NVME_DNR;
6112     }
6113 
6114     if (!nvme_ph_valid(ns, ph)) {
6115         return NVME_INVALID_FIELD | NVME_DNR;
6116     }
6117 
6118     ruhid = ns->fdp.phs[ph];
6119     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
6120 
6121     ret = nvme_h2c(n, events, noet, req);
6122     if (ret) {
6123         return ret;
6124     }
6125 
6126     for (i = 0; i < noet; i++) {
6127         event_mask |= (1 << nvme_fdp_evf_shifts[events[i]]);
6128     }
6129 
6130     if (enable) {
6131         ruh->event_filter |= event_mask;
6132     } else {
6133         ruh->event_filter = ruh->event_filter & ~event_mask;
6134     }
6135 
6136     return NVME_SUCCESS;
6137 }
6138 
6139 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
6140 {
6141     NvmeNamespace *ns = NULL;
6142 
6143     NvmeCmd *cmd = &req->cmd;
6144     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
6145     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
6146     uint32_t nsid = le32_to_cpu(cmd->nsid);
6147     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
6148     uint8_t save = NVME_SETFEAT_SAVE(dw10);
6149     uint16_t status;
6150     int i;
6151 
6152     trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
6153 
6154     if (save && !(nvme_feature_cap[fid] & NVME_FEAT_CAP_SAVE)) {
6155         return NVME_FID_NOT_SAVEABLE | NVME_DNR;
6156     }
6157 
6158     if (!nvme_feature_support[fid]) {
6159         return NVME_INVALID_FIELD | NVME_DNR;
6160     }
6161 
6162     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
6163         if (nsid != NVME_NSID_BROADCAST) {
6164             if (!nvme_nsid_valid(n, nsid)) {
6165                 return NVME_INVALID_NSID | NVME_DNR;
6166             }
6167 
6168             ns = nvme_ns(n, nsid);
6169             if (unlikely(!ns)) {
6170                 return NVME_INVALID_FIELD | NVME_DNR;
6171             }
6172         }
6173     } else if (nsid && nsid != NVME_NSID_BROADCAST) {
6174         if (!nvme_nsid_valid(n, nsid)) {
6175             return NVME_INVALID_NSID | NVME_DNR;
6176         }
6177 
6178         return NVME_FEAT_NOT_NS_SPEC | NVME_DNR;
6179     }
6180 
6181     if (!(nvme_feature_cap[fid] & NVME_FEAT_CAP_CHANGE)) {
6182         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6183     }
6184 
6185     switch (fid) {
6186     case NVME_TEMPERATURE_THRESHOLD:
6187         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6188             break;
6189         }
6190 
6191         switch (NVME_TEMP_THSEL(dw11)) {
6192         case NVME_TEMP_THSEL_OVER:
6193             n->features.temp_thresh_hi = NVME_TEMP_TMPTH(dw11);
6194             break;
6195         case NVME_TEMP_THSEL_UNDER:
6196             n->features.temp_thresh_low = NVME_TEMP_TMPTH(dw11);
6197             break;
6198         default:
6199             return NVME_INVALID_FIELD | NVME_DNR;
6200         }
6201 
6202         if ((n->temperature >= n->features.temp_thresh_hi) ||
6203             (n->temperature <= n->features.temp_thresh_low)) {
6204             nvme_smart_event(n, NVME_SMART_TEMPERATURE);
6205         }
6206 
6207         break;
6208     case NVME_ERROR_RECOVERY:
6209         if (nsid == NVME_NSID_BROADCAST) {
6210             for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6211                 ns = nvme_ns(n, i);
6212 
6213                 if (!ns) {
6214                     continue;
6215                 }
6216 
6217                 if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat)) {
6218                     ns->features.err_rec = dw11;
6219                 }
6220             }
6221 
6222             break;
6223         }
6224 
6225         assert(ns);
6226         if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat))  {
6227             ns->features.err_rec = dw11;
6228         }
6229         break;
6230     case NVME_VOLATILE_WRITE_CACHE:
6231         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6232             ns = nvme_ns(n, i);
6233             if (!ns) {
6234                 continue;
6235             }
6236 
6237             if (!(dw11 & 0x1) && blk_enable_write_cache(ns->blkconf.blk)) {
6238                 blk_flush(ns->blkconf.blk);
6239             }
6240 
6241             blk_set_enable_write_cache(ns->blkconf.blk, dw11 & 1);
6242         }
6243 
6244         break;
6245 
6246     case NVME_NUMBER_OF_QUEUES:
6247         if (n->qs_created) {
6248             return NVME_CMD_SEQ_ERROR | NVME_DNR;
6249         }
6250 
6251         /*
6252          * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
6253          * and NSQR.
6254          */
6255         if ((dw11 & 0xffff) == 0xffff || ((dw11 >> 16) & 0xffff) == 0xffff) {
6256             return NVME_INVALID_FIELD | NVME_DNR;
6257         }
6258 
6259         trace_pci_nvme_setfeat_numq((dw11 & 0xffff) + 1,
6260                                     ((dw11 >> 16) & 0xffff) + 1,
6261                                     n->conf_ioqpairs,
6262                                     n->conf_ioqpairs);
6263         req->cqe.result = cpu_to_le32((n->conf_ioqpairs - 1) |
6264                                       ((n->conf_ioqpairs - 1) << 16));
6265         break;
6266     case NVME_ASYNCHRONOUS_EVENT_CONF:
6267         n->features.async_config = dw11;
6268         break;
6269     case NVME_TIMESTAMP:
6270         return nvme_set_feature_timestamp(n, req);
6271     case NVME_HOST_BEHAVIOR_SUPPORT:
6272         status = nvme_h2c(n, (uint8_t *)&n->features.hbs,
6273                           sizeof(n->features.hbs), req);
6274         if (status) {
6275             return status;
6276         }
6277 
6278         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6279             ns = nvme_ns(n, i);
6280 
6281             if (!ns) {
6282                 continue;
6283             }
6284 
6285             ns->id_ns.nlbaf = ns->nlbaf - 1;
6286             if (!n->features.hbs.lbafee) {
6287                 ns->id_ns.nlbaf = MIN(ns->id_ns.nlbaf, 15);
6288             }
6289         }
6290 
6291         return status;
6292     case NVME_COMMAND_SET_PROFILE:
6293         if (dw11 & 0x1ff) {
6294             trace_pci_nvme_err_invalid_iocsci(dw11 & 0x1ff);
6295             return NVME_CMD_SET_CMB_REJECTED | NVME_DNR;
6296         }
6297         break;
6298     case NVME_FDP_MODE:
6299         /* spec: abort with cmd seq err if there's one or more NS' in endgrp */
6300         return NVME_CMD_SEQ_ERROR | NVME_DNR;
6301     case NVME_FDP_EVENTS:
6302         return nvme_set_feature_fdp_events(n, ns, req);
6303     default:
6304         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6305     }
6306     return NVME_SUCCESS;
6307 }
6308 
6309 static uint16_t nvme_aer(NvmeCtrl *n, NvmeRequest *req)
6310 {
6311     trace_pci_nvme_aer(nvme_cid(req));
6312 
6313     if (n->outstanding_aers > n->params.aerl) {
6314         trace_pci_nvme_aer_aerl_exceeded();
6315         return NVME_AER_LIMIT_EXCEEDED;
6316     }
6317 
6318     n->aer_reqs[n->outstanding_aers] = req;
6319     n->outstanding_aers++;
6320 
6321     if (!QTAILQ_EMPTY(&n->aer_queue)) {
6322         nvme_process_aers(n);
6323     }
6324 
6325     return NVME_NO_COMPLETE;
6326 }
6327 
6328 static void nvme_update_dmrsl(NvmeCtrl *n)
6329 {
6330     int nsid;
6331 
6332     for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
6333         NvmeNamespace *ns = nvme_ns(n, nsid);
6334         if (!ns) {
6335             continue;
6336         }
6337 
6338         n->dmrsl = MIN_NON_ZERO(n->dmrsl,
6339                                 BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
6340     }
6341 }
6342 
6343 static void nvme_select_iocs_ns(NvmeCtrl *n, NvmeNamespace *ns)
6344 {
6345     uint32_t cc = ldl_le_p(&n->bar.cc);
6346 
6347     ns->iocs = nvme_cse_iocs_none;
6348     switch (ns->csi) {
6349     case NVME_CSI_NVM:
6350         if (NVME_CC_CSS(cc) != NVME_CC_CSS_ADMIN_ONLY) {
6351             ns->iocs = nvme_cse_iocs_nvm;
6352         }
6353         break;
6354     case NVME_CSI_ZONED:
6355         if (NVME_CC_CSS(cc) == NVME_CC_CSS_CSI) {
6356             ns->iocs = nvme_cse_iocs_zoned;
6357         } else if (NVME_CC_CSS(cc) == NVME_CC_CSS_NVM) {
6358             ns->iocs = nvme_cse_iocs_nvm;
6359         }
6360         break;
6361     }
6362 }
6363 
6364 static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
6365 {
6366     NvmeNamespace *ns;
6367     NvmeCtrl *ctrl;
6368     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
6369     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6370     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6371     uint8_t sel = dw10 & 0xf;
6372     uint16_t *nr_ids = &list[0];
6373     uint16_t *ids = &list[1];
6374     uint16_t ret;
6375     int i;
6376 
6377     trace_pci_nvme_ns_attachment(nvme_cid(req), dw10 & 0xf);
6378 
6379     if (!nvme_nsid_valid(n, nsid)) {
6380         return NVME_INVALID_NSID | NVME_DNR;
6381     }
6382 
6383     ns = nvme_subsys_ns(n->subsys, nsid);
6384     if (!ns) {
6385         return NVME_INVALID_FIELD | NVME_DNR;
6386     }
6387 
6388     ret = nvme_h2c(n, (uint8_t *)list, 4096, req);
6389     if (ret) {
6390         return ret;
6391     }
6392 
6393     if (!*nr_ids) {
6394         return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6395     }
6396 
6397     *nr_ids = MIN(*nr_ids, NVME_CONTROLLER_LIST_SIZE - 1);
6398     for (i = 0; i < *nr_ids; i++) {
6399         ctrl = nvme_subsys_ctrl(n->subsys, ids[i]);
6400         if (!ctrl) {
6401             return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6402         }
6403 
6404         switch (sel) {
6405         case NVME_NS_ATTACHMENT_ATTACH:
6406             if (nvme_ns(ctrl, nsid)) {
6407                 return NVME_NS_ALREADY_ATTACHED | NVME_DNR;
6408             }
6409 
6410             if (ns->attached && !ns->params.shared) {
6411                 return NVME_NS_PRIVATE | NVME_DNR;
6412             }
6413 
6414             nvme_attach_ns(ctrl, ns);
6415             nvme_select_iocs_ns(ctrl, ns);
6416 
6417             break;
6418 
6419         case NVME_NS_ATTACHMENT_DETACH:
6420             if (!nvme_ns(ctrl, nsid)) {
6421                 return NVME_NS_NOT_ATTACHED | NVME_DNR;
6422             }
6423 
6424             ctrl->namespaces[nsid] = NULL;
6425             ns->attached--;
6426 
6427             nvme_update_dmrsl(ctrl);
6428 
6429             break;
6430 
6431         default:
6432             return NVME_INVALID_FIELD | NVME_DNR;
6433         }
6434 
6435         /*
6436          * Add namespace id to the changed namespace id list for event clearing
6437          * via Get Log Page command.
6438          */
6439         if (!test_and_set_bit(nsid, ctrl->changed_nsids)) {
6440             nvme_enqueue_event(ctrl, NVME_AER_TYPE_NOTICE,
6441                                NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED,
6442                                NVME_LOG_CHANGED_NSLIST);
6443         }
6444     }
6445 
6446     return NVME_SUCCESS;
6447 }
6448 
6449 typedef struct NvmeFormatAIOCB {
6450     BlockAIOCB common;
6451     BlockAIOCB *aiocb;
6452     NvmeRequest *req;
6453     int ret;
6454 
6455     NvmeNamespace *ns;
6456     uint32_t nsid;
6457     bool broadcast;
6458     int64_t offset;
6459 
6460     uint8_t lbaf;
6461     uint8_t mset;
6462     uint8_t pi;
6463     uint8_t pil;
6464 } NvmeFormatAIOCB;
6465 
6466 static void nvme_format_cancel(BlockAIOCB *aiocb)
6467 {
6468     NvmeFormatAIOCB *iocb = container_of(aiocb, NvmeFormatAIOCB, common);
6469 
6470     iocb->ret = -ECANCELED;
6471 
6472     if (iocb->aiocb) {
6473         blk_aio_cancel_async(iocb->aiocb);
6474         iocb->aiocb = NULL;
6475     }
6476 }
6477 
6478 static const AIOCBInfo nvme_format_aiocb_info = {
6479     .aiocb_size = sizeof(NvmeFormatAIOCB),
6480     .cancel_async = nvme_format_cancel,
6481     .get_aio_context = nvme_get_aio_context,
6482 };
6483 
6484 static void nvme_format_set(NvmeNamespace *ns, uint8_t lbaf, uint8_t mset,
6485                             uint8_t pi, uint8_t pil)
6486 {
6487     uint8_t lbafl = lbaf & 0xf;
6488     uint8_t lbafu = lbaf >> 4;
6489 
6490     trace_pci_nvme_format_set(ns->params.nsid, lbaf, mset, pi, pil);
6491 
6492     ns->id_ns.dps = (pil << 3) | pi;
6493     ns->id_ns.flbas = (lbafu << 5) | (mset << 4) | lbafl;
6494 
6495     nvme_ns_init_format(ns);
6496 }
6497 
6498 static void nvme_do_format(NvmeFormatAIOCB *iocb);
6499 
6500 static void nvme_format_ns_cb(void *opaque, int ret)
6501 {
6502     NvmeFormatAIOCB *iocb = opaque;
6503     NvmeNamespace *ns = iocb->ns;
6504     int bytes;
6505 
6506     if (iocb->ret < 0) {
6507         goto done;
6508     } else if (ret < 0) {
6509         iocb->ret = ret;
6510         goto done;
6511     }
6512 
6513     assert(ns);
6514 
6515     if (iocb->offset < ns->size) {
6516         bytes = MIN(BDRV_REQUEST_MAX_BYTES, ns->size - iocb->offset);
6517 
6518         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, iocb->offset,
6519                                             bytes, BDRV_REQ_MAY_UNMAP,
6520                                             nvme_format_ns_cb, iocb);
6521 
6522         iocb->offset += bytes;
6523         return;
6524     }
6525 
6526     nvme_format_set(ns, iocb->lbaf, iocb->mset, iocb->pi, iocb->pil);
6527     ns->status = 0x0;
6528     iocb->ns = NULL;
6529     iocb->offset = 0;
6530 
6531 done:
6532     nvme_do_format(iocb);
6533 }
6534 
6535 static uint16_t nvme_format_check(NvmeNamespace *ns, uint8_t lbaf, uint8_t pi)
6536 {
6537     if (ns->params.zoned) {
6538         return NVME_INVALID_FORMAT | NVME_DNR;
6539     }
6540 
6541     if (lbaf > ns->id_ns.nlbaf) {
6542         return NVME_INVALID_FORMAT | NVME_DNR;
6543     }
6544 
6545     if (pi && (ns->id_ns.lbaf[lbaf].ms < nvme_pi_tuple_size(ns))) {
6546         return NVME_INVALID_FORMAT | NVME_DNR;
6547     }
6548 
6549     if (pi && pi > NVME_ID_NS_DPS_TYPE_3) {
6550         return NVME_INVALID_FIELD | NVME_DNR;
6551     }
6552 
6553     return NVME_SUCCESS;
6554 }
6555 
6556 static void nvme_do_format(NvmeFormatAIOCB *iocb)
6557 {
6558     NvmeRequest *req = iocb->req;
6559     NvmeCtrl *n = nvme_ctrl(req);
6560     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6561     uint8_t lbaf = dw10 & 0xf;
6562     uint8_t pi = (dw10 >> 5) & 0x7;
6563     uint16_t status;
6564     int i;
6565 
6566     if (iocb->ret < 0) {
6567         goto done;
6568     }
6569 
6570     if (iocb->broadcast) {
6571         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
6572             iocb->ns = nvme_ns(n, i);
6573             if (iocb->ns) {
6574                 iocb->nsid = i;
6575                 break;
6576             }
6577         }
6578     }
6579 
6580     if (!iocb->ns) {
6581         goto done;
6582     }
6583 
6584     status = nvme_format_check(iocb->ns, lbaf, pi);
6585     if (status) {
6586         req->status = status;
6587         goto done;
6588     }
6589 
6590     iocb->ns->status = NVME_FORMAT_IN_PROGRESS;
6591     nvme_format_ns_cb(iocb, 0);
6592     return;
6593 
6594 done:
6595     iocb->common.cb(iocb->common.opaque, iocb->ret);
6596     qemu_aio_unref(iocb);
6597 }
6598 
6599 static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
6600 {
6601     NvmeFormatAIOCB *iocb;
6602     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6603     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6604     uint8_t lbaf = dw10 & 0xf;
6605     uint8_t mset = (dw10 >> 4) & 0x1;
6606     uint8_t pi = (dw10 >> 5) & 0x7;
6607     uint8_t pil = (dw10 >> 8) & 0x1;
6608     uint8_t lbafu = (dw10 >> 12) & 0x3;
6609     uint16_t status;
6610 
6611     iocb = qemu_aio_get(&nvme_format_aiocb_info, NULL, nvme_misc_cb, req);
6612 
6613     iocb->req = req;
6614     iocb->ret = 0;
6615     iocb->ns = NULL;
6616     iocb->nsid = 0;
6617     iocb->lbaf = lbaf;
6618     iocb->mset = mset;
6619     iocb->pi = pi;
6620     iocb->pil = pil;
6621     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
6622     iocb->offset = 0;
6623 
6624     if (n->features.hbs.lbafee) {
6625         iocb->lbaf |= lbafu << 4;
6626     }
6627 
6628     if (!iocb->broadcast) {
6629         if (!nvme_nsid_valid(n, nsid)) {
6630             status = NVME_INVALID_NSID | NVME_DNR;
6631             goto out;
6632         }
6633 
6634         iocb->ns = nvme_ns(n, nsid);
6635         if (!iocb->ns) {
6636             status = NVME_INVALID_FIELD | NVME_DNR;
6637             goto out;
6638         }
6639     }
6640 
6641     req->aiocb = &iocb->common;
6642     nvme_do_format(iocb);
6643 
6644     return NVME_NO_COMPLETE;
6645 
6646 out:
6647     qemu_aio_unref(iocb);
6648 
6649     return status;
6650 }
6651 
6652 static void nvme_get_virt_res_num(NvmeCtrl *n, uint8_t rt, int *num_total,
6653                                   int *num_prim, int *num_sec)
6654 {
6655     *num_total = le32_to_cpu(rt ?
6656                              n->pri_ctrl_cap.vifrt : n->pri_ctrl_cap.vqfrt);
6657     *num_prim = le16_to_cpu(rt ?
6658                             n->pri_ctrl_cap.virfap : n->pri_ctrl_cap.vqrfap);
6659     *num_sec = le16_to_cpu(rt ? n->pri_ctrl_cap.virfa : n->pri_ctrl_cap.vqrfa);
6660 }
6661 
6662 static uint16_t nvme_assign_virt_res_to_prim(NvmeCtrl *n, NvmeRequest *req,
6663                                              uint16_t cntlid, uint8_t rt,
6664                                              int nr)
6665 {
6666     int num_total, num_prim, num_sec;
6667 
6668     if (cntlid != n->cntlid) {
6669         return NVME_INVALID_CTRL_ID | NVME_DNR;
6670     }
6671 
6672     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6673 
6674     if (nr > num_total) {
6675         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6676     }
6677 
6678     if (nr > num_total - num_sec) {
6679         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6680     }
6681 
6682     if (rt) {
6683         n->next_pri_ctrl_cap.virfap = cpu_to_le16(nr);
6684     } else {
6685         n->next_pri_ctrl_cap.vqrfap = cpu_to_le16(nr);
6686     }
6687 
6688     req->cqe.result = cpu_to_le32(nr);
6689     return req->status;
6690 }
6691 
6692 static void nvme_update_virt_res(NvmeCtrl *n, NvmeSecCtrlEntry *sctrl,
6693                                  uint8_t rt, int nr)
6694 {
6695     int prev_nr, prev_total;
6696 
6697     if (rt) {
6698         prev_nr = le16_to_cpu(sctrl->nvi);
6699         prev_total = le32_to_cpu(n->pri_ctrl_cap.virfa);
6700         sctrl->nvi = cpu_to_le16(nr);
6701         n->pri_ctrl_cap.virfa = cpu_to_le32(prev_total + nr - prev_nr);
6702     } else {
6703         prev_nr = le16_to_cpu(sctrl->nvq);
6704         prev_total = le32_to_cpu(n->pri_ctrl_cap.vqrfa);
6705         sctrl->nvq = cpu_to_le16(nr);
6706         n->pri_ctrl_cap.vqrfa = cpu_to_le32(prev_total + nr - prev_nr);
6707     }
6708 }
6709 
6710 static uint16_t nvme_assign_virt_res_to_sec(NvmeCtrl *n, NvmeRequest *req,
6711                                             uint16_t cntlid, uint8_t rt, int nr)
6712 {
6713     int num_total, num_prim, num_sec, num_free, diff, limit;
6714     NvmeSecCtrlEntry *sctrl;
6715 
6716     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6717     if (!sctrl) {
6718         return NVME_INVALID_CTRL_ID | NVME_DNR;
6719     }
6720 
6721     if (sctrl->scs) {
6722         return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6723     }
6724 
6725     limit = le16_to_cpu(rt ? n->pri_ctrl_cap.vifrsm : n->pri_ctrl_cap.vqfrsm);
6726     if (nr > limit) {
6727         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6728     }
6729 
6730     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6731     num_free = num_total - num_prim - num_sec;
6732     diff = nr - le16_to_cpu(rt ? sctrl->nvi : sctrl->nvq);
6733 
6734     if (diff > num_free) {
6735         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6736     }
6737 
6738     nvme_update_virt_res(n, sctrl, rt, nr);
6739     req->cqe.result = cpu_to_le32(nr);
6740 
6741     return req->status;
6742 }
6743 
6744 static uint16_t nvme_virt_set_state(NvmeCtrl *n, uint16_t cntlid, bool online)
6745 {
6746     PCIDevice *pci = PCI_DEVICE(n);
6747     NvmeCtrl *sn = NULL;
6748     NvmeSecCtrlEntry *sctrl;
6749     int vf_index;
6750 
6751     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6752     if (!sctrl) {
6753         return NVME_INVALID_CTRL_ID | NVME_DNR;
6754     }
6755 
6756     if (!pci_is_vf(pci)) {
6757         vf_index = le16_to_cpu(sctrl->vfn) - 1;
6758         sn = NVME(pcie_sriov_get_vf_at_index(pci, vf_index));
6759     }
6760 
6761     if (online) {
6762         if (!sctrl->nvi || (le16_to_cpu(sctrl->nvq) < 2) || !sn) {
6763             return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6764         }
6765 
6766         if (!sctrl->scs) {
6767             sctrl->scs = 0x1;
6768             nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6769         }
6770     } else {
6771         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_INTERRUPT, 0);
6772         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_QUEUE, 0);
6773 
6774         if (sctrl->scs) {
6775             sctrl->scs = 0x0;
6776             if (sn) {
6777                 nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6778             }
6779         }
6780     }
6781 
6782     return NVME_SUCCESS;
6783 }
6784 
6785 static uint16_t nvme_virt_mngmt(NvmeCtrl *n, NvmeRequest *req)
6786 {
6787     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6788     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
6789     uint8_t act = dw10 & 0xf;
6790     uint8_t rt = (dw10 >> 8) & 0x7;
6791     uint16_t cntlid = (dw10 >> 16) & 0xffff;
6792     int nr = dw11 & 0xffff;
6793 
6794     trace_pci_nvme_virt_mngmt(nvme_cid(req), act, cntlid, rt ? "VI" : "VQ", nr);
6795 
6796     if (rt != NVME_VIRT_RES_QUEUE && rt != NVME_VIRT_RES_INTERRUPT) {
6797         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6798     }
6799 
6800     switch (act) {
6801     case NVME_VIRT_MNGMT_ACTION_SEC_ASSIGN:
6802         return nvme_assign_virt_res_to_sec(n, req, cntlid, rt, nr);
6803     case NVME_VIRT_MNGMT_ACTION_PRM_ALLOC:
6804         return nvme_assign_virt_res_to_prim(n, req, cntlid, rt, nr);
6805     case NVME_VIRT_MNGMT_ACTION_SEC_ONLINE:
6806         return nvme_virt_set_state(n, cntlid, true);
6807     case NVME_VIRT_MNGMT_ACTION_SEC_OFFLINE:
6808         return nvme_virt_set_state(n, cntlid, false);
6809     default:
6810         return NVME_INVALID_FIELD | NVME_DNR;
6811     }
6812 }
6813 
6814 static uint16_t nvme_dbbuf_config(NvmeCtrl *n, const NvmeRequest *req)
6815 {
6816     PCIDevice *pci = PCI_DEVICE(n);
6817     uint64_t dbs_addr = le64_to_cpu(req->cmd.dptr.prp1);
6818     uint64_t eis_addr = le64_to_cpu(req->cmd.dptr.prp2);
6819     int i;
6820 
6821     /* Address should be page aligned */
6822     if (dbs_addr & (n->page_size - 1) || eis_addr & (n->page_size - 1)) {
6823         return NVME_INVALID_FIELD | NVME_DNR;
6824     }
6825 
6826     /* Save shadow buffer base addr for use during queue creation */
6827     n->dbbuf_dbs = dbs_addr;
6828     n->dbbuf_eis = eis_addr;
6829     n->dbbuf_enabled = true;
6830 
6831     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
6832         NvmeSQueue *sq = n->sq[i];
6833         NvmeCQueue *cq = n->cq[i];
6834 
6835         if (sq) {
6836             /*
6837              * CAP.DSTRD is 0, so offset of ith sq db_addr is (i<<3)
6838              * nvme_process_db() uses this hard-coded way to calculate
6839              * doorbell offsets. Be consistent with that here.
6840              */
6841             sq->db_addr = dbs_addr + (i << 3);
6842             sq->ei_addr = eis_addr + (i << 3);
6843             stl_le_pci_dma(pci, sq->db_addr, sq->tail, MEMTXATTRS_UNSPECIFIED);
6844 
6845             if (n->params.ioeventfd && sq->sqid != 0) {
6846                 if (!nvme_init_sq_ioeventfd(sq)) {
6847                     sq->ioeventfd_enabled = true;
6848                 }
6849             }
6850         }
6851 
6852         if (cq) {
6853             /* CAP.DSTRD is 0, so offset of ith cq db_addr is (i<<3)+(1<<2) */
6854             cq->db_addr = dbs_addr + (i << 3) + (1 << 2);
6855             cq->ei_addr = eis_addr + (i << 3) + (1 << 2);
6856             stl_le_pci_dma(pci, cq->db_addr, cq->head, MEMTXATTRS_UNSPECIFIED);
6857 
6858             if (n->params.ioeventfd && cq->cqid != 0) {
6859                 if (!nvme_init_cq_ioeventfd(cq)) {
6860                     cq->ioeventfd_enabled = true;
6861                 }
6862             }
6863         }
6864     }
6865 
6866     trace_pci_nvme_dbbuf_config(dbs_addr, eis_addr);
6867 
6868     return NVME_SUCCESS;
6869 }
6870 
6871 static uint16_t nvme_directive_send(NvmeCtrl *n, NvmeRequest *req)
6872 {
6873     return NVME_INVALID_FIELD | NVME_DNR;
6874 }
6875 
6876 static uint16_t nvme_directive_receive(NvmeCtrl *n, NvmeRequest *req)
6877 {
6878     NvmeNamespace *ns;
6879     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6880     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
6881     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6882     uint8_t doper, dtype;
6883     uint32_t numd, trans_len;
6884     NvmeDirectiveIdentify id = {
6885         .supported = 1 << NVME_DIRECTIVE_IDENTIFY,
6886         .enabled = 1 << NVME_DIRECTIVE_IDENTIFY,
6887     };
6888 
6889     numd = dw10 + 1;
6890     doper = dw11 & 0xff;
6891     dtype = (dw11 >> 8) & 0xff;
6892 
6893     trans_len = MIN(sizeof(NvmeDirectiveIdentify), numd << 2);
6894 
6895     if (nsid == NVME_NSID_BROADCAST || dtype != NVME_DIRECTIVE_IDENTIFY ||
6896         doper != NVME_DIRECTIVE_RETURN_PARAMS) {
6897         return NVME_INVALID_FIELD | NVME_DNR;
6898     }
6899 
6900     ns = nvme_ns(n, nsid);
6901     if (!ns) {
6902         return NVME_INVALID_FIELD | NVME_DNR;
6903     }
6904 
6905     switch (dtype) {
6906     case NVME_DIRECTIVE_IDENTIFY:
6907         switch (doper) {
6908         case NVME_DIRECTIVE_RETURN_PARAMS:
6909             if (ns->endgrp && ns->endgrp->fdp.enabled) {
6910                 id.supported |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6911                 id.enabled |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6912                 id.persistent |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6913             }
6914 
6915             return nvme_c2h(n, (uint8_t *)&id, trans_len, req);
6916 
6917         default:
6918             return NVME_INVALID_FIELD | NVME_DNR;
6919         }
6920 
6921     default:
6922         return NVME_INVALID_FIELD;
6923     }
6924 }
6925 
6926 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
6927 {
6928     trace_pci_nvme_admin_cmd(nvme_cid(req), nvme_sqid(req), req->cmd.opcode,
6929                              nvme_adm_opc_str(req->cmd.opcode));
6930 
6931     if (!(nvme_cse_acs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
6932         trace_pci_nvme_err_invalid_admin_opc(req->cmd.opcode);
6933         return NVME_INVALID_OPCODE | NVME_DNR;
6934     }
6935 
6936     /* SGLs shall not be used for Admin commands in NVMe over PCIe */
6937     if (NVME_CMD_FLAGS_PSDT(req->cmd.flags) != NVME_PSDT_PRP) {
6938         return NVME_INVALID_FIELD | NVME_DNR;
6939     }
6940 
6941     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
6942         return NVME_INVALID_FIELD;
6943     }
6944 
6945     switch (req->cmd.opcode) {
6946     case NVME_ADM_CMD_DELETE_SQ:
6947         return nvme_del_sq(n, req);
6948     case NVME_ADM_CMD_CREATE_SQ:
6949         return nvme_create_sq(n, req);
6950     case NVME_ADM_CMD_GET_LOG_PAGE:
6951         return nvme_get_log(n, req);
6952     case NVME_ADM_CMD_DELETE_CQ:
6953         return nvme_del_cq(n, req);
6954     case NVME_ADM_CMD_CREATE_CQ:
6955         return nvme_create_cq(n, req);
6956     case NVME_ADM_CMD_IDENTIFY:
6957         return nvme_identify(n, req);
6958     case NVME_ADM_CMD_ABORT:
6959         return nvme_abort(n, req);
6960     case NVME_ADM_CMD_SET_FEATURES:
6961         return nvme_set_feature(n, req);
6962     case NVME_ADM_CMD_GET_FEATURES:
6963         return nvme_get_feature(n, req);
6964     case NVME_ADM_CMD_ASYNC_EV_REQ:
6965         return nvme_aer(n, req);
6966     case NVME_ADM_CMD_NS_ATTACHMENT:
6967         return nvme_ns_attachment(n, req);
6968     case NVME_ADM_CMD_VIRT_MNGMT:
6969         return nvme_virt_mngmt(n, req);
6970     case NVME_ADM_CMD_DBBUF_CONFIG:
6971         return nvme_dbbuf_config(n, req);
6972     case NVME_ADM_CMD_FORMAT_NVM:
6973         return nvme_format(n, req);
6974     case NVME_ADM_CMD_DIRECTIVE_SEND:
6975         return nvme_directive_send(n, req);
6976     case NVME_ADM_CMD_DIRECTIVE_RECV:
6977         return nvme_directive_receive(n, req);
6978     default:
6979         assert(false);
6980     }
6981 
6982     return NVME_INVALID_OPCODE | NVME_DNR;
6983 }
6984 
6985 static void nvme_update_sq_eventidx(const NvmeSQueue *sq)
6986 {
6987     trace_pci_nvme_update_sq_eventidx(sq->sqid, sq->tail);
6988 
6989     stl_le_pci_dma(PCI_DEVICE(sq->ctrl), sq->ei_addr, sq->tail,
6990                    MEMTXATTRS_UNSPECIFIED);
6991 }
6992 
6993 static void nvme_update_sq_tail(NvmeSQueue *sq)
6994 {
6995     ldl_le_pci_dma(PCI_DEVICE(sq->ctrl), sq->db_addr, &sq->tail,
6996                    MEMTXATTRS_UNSPECIFIED);
6997 
6998     trace_pci_nvme_update_sq_tail(sq->sqid, sq->tail);
6999 }
7000 
7001 static void nvme_process_sq(void *opaque)
7002 {
7003     NvmeSQueue *sq = opaque;
7004     NvmeCtrl *n = sq->ctrl;
7005     NvmeCQueue *cq = n->cq[sq->cqid];
7006 
7007     uint16_t status;
7008     hwaddr addr;
7009     NvmeCmd cmd;
7010     NvmeRequest *req;
7011 
7012     if (n->dbbuf_enabled) {
7013         nvme_update_sq_tail(sq);
7014     }
7015 
7016     while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
7017         addr = sq->dma_addr + (sq->head << NVME_SQES);
7018         if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
7019             trace_pci_nvme_err_addr_read(addr);
7020             trace_pci_nvme_err_cfs();
7021             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
7022             break;
7023         }
7024         nvme_inc_sq_head(sq);
7025 
7026         req = QTAILQ_FIRST(&sq->req_list);
7027         QTAILQ_REMOVE(&sq->req_list, req, entry);
7028         QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
7029         nvme_req_clear(req);
7030         req->cqe.cid = cmd.cid;
7031         memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
7032 
7033         status = sq->sqid ? nvme_io_cmd(n, req) :
7034             nvme_admin_cmd(n, req);
7035         if (status != NVME_NO_COMPLETE) {
7036             req->status = status;
7037             nvme_enqueue_req_completion(cq, req);
7038         }
7039 
7040         if (n->dbbuf_enabled) {
7041             nvme_update_sq_eventidx(sq);
7042             nvme_update_sq_tail(sq);
7043         }
7044     }
7045 }
7046 
7047 static void nvme_update_msixcap_ts(PCIDevice *pci_dev, uint32_t table_size)
7048 {
7049     uint8_t *config;
7050 
7051     if (!msix_present(pci_dev)) {
7052         return;
7053     }
7054 
7055     assert(table_size > 0 && table_size <= pci_dev->msix_entries_nr);
7056 
7057     config = pci_dev->config + pci_dev->msix_cap;
7058     pci_set_word_by_mask(config + PCI_MSIX_FLAGS, PCI_MSIX_FLAGS_QSIZE,
7059                          table_size - 1);
7060 }
7061 
7062 static void nvme_activate_virt_res(NvmeCtrl *n)
7063 {
7064     PCIDevice *pci_dev = PCI_DEVICE(n);
7065     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7066     NvmeSecCtrlEntry *sctrl;
7067 
7068     /* -1 to account for the admin queue */
7069     if (pci_is_vf(pci_dev)) {
7070         sctrl = nvme_sctrl(n);
7071         cap->vqprt = sctrl->nvq;
7072         cap->viprt = sctrl->nvi;
7073         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7074         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7075     } else {
7076         cap->vqrfap = n->next_pri_ctrl_cap.vqrfap;
7077         cap->virfap = n->next_pri_ctrl_cap.virfap;
7078         n->conf_ioqpairs = le16_to_cpu(cap->vqprt) +
7079                            le16_to_cpu(cap->vqrfap) - 1;
7080         n->conf_msix_qsize = le16_to_cpu(cap->viprt) +
7081                              le16_to_cpu(cap->virfap);
7082     }
7083 }
7084 
7085 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst)
7086 {
7087     PCIDevice *pci_dev = PCI_DEVICE(n);
7088     NvmeSecCtrlEntry *sctrl;
7089     NvmeNamespace *ns;
7090     int i;
7091 
7092     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7093         ns = nvme_ns(n, i);
7094         if (!ns) {
7095             continue;
7096         }
7097 
7098         nvme_ns_drain(ns);
7099     }
7100 
7101     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7102         if (n->sq[i] != NULL) {
7103             nvme_free_sq(n->sq[i], n);
7104         }
7105     }
7106     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7107         if (n->cq[i] != NULL) {
7108             nvme_free_cq(n->cq[i], n);
7109         }
7110     }
7111 
7112     while (!QTAILQ_EMPTY(&n->aer_queue)) {
7113         NvmeAsyncEvent *event = QTAILQ_FIRST(&n->aer_queue);
7114         QTAILQ_REMOVE(&n->aer_queue, event, entry);
7115         g_free(event);
7116     }
7117 
7118     if (n->params.sriov_max_vfs) {
7119         if (!pci_is_vf(pci_dev)) {
7120             for (i = 0; i < n->sec_ctrl_list.numcntl; i++) {
7121                 sctrl = &n->sec_ctrl_list.sec[i];
7122                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
7123             }
7124 
7125             if (rst != NVME_RESET_CONTROLLER) {
7126                 pcie_sriov_pf_disable_vfs(pci_dev);
7127             }
7128         }
7129 
7130         if (rst != NVME_RESET_CONTROLLER) {
7131             nvme_activate_virt_res(n);
7132         }
7133     }
7134 
7135     n->aer_queued = 0;
7136     n->aer_mask = 0;
7137     n->outstanding_aers = 0;
7138     n->qs_created = false;
7139 
7140     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
7141 
7142     if (pci_is_vf(pci_dev)) {
7143         sctrl = nvme_sctrl(n);
7144 
7145         stl_le_p(&n->bar.csts, sctrl->scs ? 0 : NVME_CSTS_FAILED);
7146     } else {
7147         stl_le_p(&n->bar.csts, 0);
7148     }
7149 
7150     stl_le_p(&n->bar.intms, 0);
7151     stl_le_p(&n->bar.intmc, 0);
7152     stl_le_p(&n->bar.cc, 0);
7153 
7154     n->dbbuf_dbs = 0;
7155     n->dbbuf_eis = 0;
7156     n->dbbuf_enabled = false;
7157 }
7158 
7159 static void nvme_ctrl_shutdown(NvmeCtrl *n)
7160 {
7161     NvmeNamespace *ns;
7162     int i;
7163 
7164     if (n->pmr.dev) {
7165         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7166     }
7167 
7168     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7169         ns = nvme_ns(n, i);
7170         if (!ns) {
7171             continue;
7172         }
7173 
7174         nvme_ns_shutdown(ns);
7175     }
7176 }
7177 
7178 static void nvme_select_iocs(NvmeCtrl *n)
7179 {
7180     NvmeNamespace *ns;
7181     int i;
7182 
7183     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7184         ns = nvme_ns(n, i);
7185         if (!ns) {
7186             continue;
7187         }
7188 
7189         nvme_select_iocs_ns(n, ns);
7190     }
7191 }
7192 
7193 static int nvme_start_ctrl(NvmeCtrl *n)
7194 {
7195     uint64_t cap = ldq_le_p(&n->bar.cap);
7196     uint32_t cc = ldl_le_p(&n->bar.cc);
7197     uint32_t aqa = ldl_le_p(&n->bar.aqa);
7198     uint64_t asq = ldq_le_p(&n->bar.asq);
7199     uint64_t acq = ldq_le_p(&n->bar.acq);
7200     uint32_t page_bits = NVME_CC_MPS(cc) + 12;
7201     uint32_t page_size = 1 << page_bits;
7202     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
7203 
7204     if (pci_is_vf(PCI_DEVICE(n)) && !sctrl->scs) {
7205         trace_pci_nvme_err_startfail_virt_state(le16_to_cpu(sctrl->nvi),
7206                                                 le16_to_cpu(sctrl->nvq));
7207         return -1;
7208     }
7209     if (unlikely(n->cq[0])) {
7210         trace_pci_nvme_err_startfail_cq();
7211         return -1;
7212     }
7213     if (unlikely(n->sq[0])) {
7214         trace_pci_nvme_err_startfail_sq();
7215         return -1;
7216     }
7217     if (unlikely(asq & (page_size - 1))) {
7218         trace_pci_nvme_err_startfail_asq_misaligned(asq);
7219         return -1;
7220     }
7221     if (unlikely(acq & (page_size - 1))) {
7222         trace_pci_nvme_err_startfail_acq_misaligned(acq);
7223         return -1;
7224     }
7225     if (unlikely(!(NVME_CAP_CSS(cap) & (1 << NVME_CC_CSS(cc))))) {
7226         trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc));
7227         return -1;
7228     }
7229     if (unlikely(NVME_CC_MPS(cc) < NVME_CAP_MPSMIN(cap))) {
7230         trace_pci_nvme_err_startfail_page_too_small(
7231                     NVME_CC_MPS(cc),
7232                     NVME_CAP_MPSMIN(cap));
7233         return -1;
7234     }
7235     if (unlikely(NVME_CC_MPS(cc) >
7236                  NVME_CAP_MPSMAX(cap))) {
7237         trace_pci_nvme_err_startfail_page_too_large(
7238                     NVME_CC_MPS(cc),
7239                     NVME_CAP_MPSMAX(cap));
7240         return -1;
7241     }
7242     if (unlikely(!NVME_AQA_ASQS(aqa))) {
7243         trace_pci_nvme_err_startfail_asqent_sz_zero();
7244         return -1;
7245     }
7246     if (unlikely(!NVME_AQA_ACQS(aqa))) {
7247         trace_pci_nvme_err_startfail_acqent_sz_zero();
7248         return -1;
7249     }
7250 
7251     n->page_bits = page_bits;
7252     n->page_size = page_size;
7253     n->max_prp_ents = n->page_size / sizeof(uint64_t);
7254     nvme_init_cq(&n->admin_cq, n, acq, 0, 0, NVME_AQA_ACQS(aqa) + 1, 1);
7255     nvme_init_sq(&n->admin_sq, n, asq, 0, 0, NVME_AQA_ASQS(aqa) + 1);
7256 
7257     nvme_set_timestamp(n, 0ULL);
7258 
7259     nvme_select_iocs(n);
7260 
7261     return 0;
7262 }
7263 
7264 static void nvme_cmb_enable_regs(NvmeCtrl *n)
7265 {
7266     uint32_t cmbloc = ldl_le_p(&n->bar.cmbloc);
7267     uint32_t cmbsz = ldl_le_p(&n->bar.cmbsz);
7268 
7269     NVME_CMBLOC_SET_CDPCILS(cmbloc, 1);
7270     NVME_CMBLOC_SET_CDPMLS(cmbloc, 1);
7271     NVME_CMBLOC_SET_BIR(cmbloc, NVME_CMB_BIR);
7272     stl_le_p(&n->bar.cmbloc, cmbloc);
7273 
7274     NVME_CMBSZ_SET_SQS(cmbsz, 1);
7275     NVME_CMBSZ_SET_CQS(cmbsz, 0);
7276     NVME_CMBSZ_SET_LISTS(cmbsz, 1);
7277     NVME_CMBSZ_SET_RDS(cmbsz, 1);
7278     NVME_CMBSZ_SET_WDS(cmbsz, 1);
7279     NVME_CMBSZ_SET_SZU(cmbsz, 2); /* MBs */
7280     NVME_CMBSZ_SET_SZ(cmbsz, n->params.cmb_size_mb);
7281     stl_le_p(&n->bar.cmbsz, cmbsz);
7282 }
7283 
7284 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
7285                            unsigned size)
7286 {
7287     PCIDevice *pci = PCI_DEVICE(n);
7288     uint64_t cap = ldq_le_p(&n->bar.cap);
7289     uint32_t cc = ldl_le_p(&n->bar.cc);
7290     uint32_t intms = ldl_le_p(&n->bar.intms);
7291     uint32_t csts = ldl_le_p(&n->bar.csts);
7292     uint32_t pmrsts = ldl_le_p(&n->bar.pmrsts);
7293 
7294     if (unlikely(offset & (sizeof(uint32_t) - 1))) {
7295         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
7296                        "MMIO write not 32-bit aligned,"
7297                        " offset=0x%"PRIx64"", offset);
7298         /* should be ignored, fall through for now */
7299     }
7300 
7301     if (unlikely(size < sizeof(uint32_t))) {
7302         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
7303                        "MMIO write smaller than 32-bits,"
7304                        " offset=0x%"PRIx64", size=%u",
7305                        offset, size);
7306         /* should be ignored, fall through for now */
7307     }
7308 
7309     switch (offset) {
7310     case NVME_REG_INTMS:
7311         if (unlikely(msix_enabled(pci))) {
7312             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7313                            "undefined access to interrupt mask set"
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_set(data & 0xffffffff, intms);
7321         nvme_irq_check(n);
7322         break;
7323     case NVME_REG_INTMC:
7324         if (unlikely(msix_enabled(pci))) {
7325             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7326                            "undefined access to interrupt mask clr"
7327                            " when MSI-X is enabled");
7328             /* should be ignored, fall through for now */
7329         }
7330         intms &= ~data;
7331         stl_le_p(&n->bar.intms, intms);
7332         n->bar.intmc = n->bar.intms;
7333         trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, intms);
7334         nvme_irq_check(n);
7335         break;
7336     case NVME_REG_CC:
7337         stl_le_p(&n->bar.cc, data);
7338 
7339         trace_pci_nvme_mmio_cfg(data & 0xffffffff);
7340 
7341         if (NVME_CC_SHN(data) && !(NVME_CC_SHN(cc))) {
7342             trace_pci_nvme_mmio_shutdown_set();
7343             nvme_ctrl_shutdown(n);
7344             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7345             csts |= NVME_CSTS_SHST_COMPLETE;
7346         } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(cc)) {
7347             trace_pci_nvme_mmio_shutdown_cleared();
7348             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7349         }
7350 
7351         if (NVME_CC_EN(data) && !NVME_CC_EN(cc)) {
7352             if (unlikely(nvme_start_ctrl(n))) {
7353                 trace_pci_nvme_err_startfail();
7354                 csts = NVME_CSTS_FAILED;
7355             } else {
7356                 trace_pci_nvme_mmio_start_success();
7357                 csts = NVME_CSTS_READY;
7358             }
7359         } else if (!NVME_CC_EN(data) && NVME_CC_EN(cc)) {
7360             trace_pci_nvme_mmio_stopped();
7361             nvme_ctrl_reset(n, NVME_RESET_CONTROLLER);
7362 
7363             break;
7364         }
7365 
7366         stl_le_p(&n->bar.csts, csts);
7367 
7368         break;
7369     case NVME_REG_CSTS:
7370         if (data & (1 << 4)) {
7371             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
7372                            "attempted to W1C CSTS.NSSRO"
7373                            " but CAP.NSSRS is zero (not supported)");
7374         } else if (data != 0) {
7375             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
7376                            "attempted to set a read only bit"
7377                            " of controller status");
7378         }
7379         break;
7380     case NVME_REG_NSSR:
7381         if (data == 0x4e564d65) {
7382             trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
7383         } else {
7384             /* The spec says that writes of other values have no effect */
7385             return;
7386         }
7387         break;
7388     case NVME_REG_AQA:
7389         stl_le_p(&n->bar.aqa, data);
7390         trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
7391         break;
7392     case NVME_REG_ASQ:
7393         stn_le_p(&n->bar.asq, size, data);
7394         trace_pci_nvme_mmio_asqaddr(data);
7395         break;
7396     case NVME_REG_ASQ + 4:
7397         stl_le_p((uint8_t *)&n->bar.asq + 4, data);
7398         trace_pci_nvme_mmio_asqaddr_hi(data, ldq_le_p(&n->bar.asq));
7399         break;
7400     case NVME_REG_ACQ:
7401         trace_pci_nvme_mmio_acqaddr(data);
7402         stn_le_p(&n->bar.acq, size, data);
7403         break;
7404     case NVME_REG_ACQ + 4:
7405         stl_le_p((uint8_t *)&n->bar.acq + 4, data);
7406         trace_pci_nvme_mmio_acqaddr_hi(data, ldq_le_p(&n->bar.acq));
7407         break;
7408     case NVME_REG_CMBLOC:
7409         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
7410                        "invalid write to reserved CMBLOC"
7411                        " when CMBSZ is zero, ignored");
7412         return;
7413     case NVME_REG_CMBSZ:
7414         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
7415                        "invalid write to read only CMBSZ, ignored");
7416         return;
7417     case NVME_REG_CMBMSC:
7418         if (!NVME_CAP_CMBS(cap)) {
7419             return;
7420         }
7421 
7422         stn_le_p(&n->bar.cmbmsc, size, data);
7423         n->cmb.cmse = false;
7424 
7425         if (NVME_CMBMSC_CRE(data)) {
7426             nvme_cmb_enable_regs(n);
7427 
7428             if (NVME_CMBMSC_CMSE(data)) {
7429                 uint64_t cmbmsc = ldq_le_p(&n->bar.cmbmsc);
7430                 hwaddr cba = NVME_CMBMSC_CBA(cmbmsc) << CMBMSC_CBA_SHIFT;
7431                 if (cba + int128_get64(n->cmb.mem.size) < cba) {
7432                     uint32_t cmbsts = ldl_le_p(&n->bar.cmbsts);
7433                     NVME_CMBSTS_SET_CBAI(cmbsts, 1);
7434                     stl_le_p(&n->bar.cmbsts, cmbsts);
7435                     return;
7436                 }
7437 
7438                 n->cmb.cba = cba;
7439                 n->cmb.cmse = true;
7440             }
7441         } else {
7442             n->bar.cmbsz = 0;
7443             n->bar.cmbloc = 0;
7444         }
7445 
7446         return;
7447     case NVME_REG_CMBMSC + 4:
7448         stl_le_p((uint8_t *)&n->bar.cmbmsc + 4, data);
7449         return;
7450 
7451     case NVME_REG_PMRCAP:
7452         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
7453                        "invalid write to PMRCAP register, ignored");
7454         return;
7455     case NVME_REG_PMRCTL:
7456         if (!NVME_CAP_PMRS(cap)) {
7457             return;
7458         }
7459 
7460         stl_le_p(&n->bar.pmrctl, data);
7461         if (NVME_PMRCTL_EN(data)) {
7462             memory_region_set_enabled(&n->pmr.dev->mr, true);
7463             pmrsts = 0;
7464         } else {
7465             memory_region_set_enabled(&n->pmr.dev->mr, false);
7466             NVME_PMRSTS_SET_NRDY(pmrsts, 1);
7467             n->pmr.cmse = false;
7468         }
7469         stl_le_p(&n->bar.pmrsts, pmrsts);
7470         return;
7471     case NVME_REG_PMRSTS:
7472         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
7473                        "invalid write to PMRSTS register, ignored");
7474         return;
7475     case NVME_REG_PMREBS:
7476         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
7477                        "invalid write to PMREBS register, ignored");
7478         return;
7479     case NVME_REG_PMRSWTP:
7480         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
7481                        "invalid write to PMRSWTP register, ignored");
7482         return;
7483     case NVME_REG_PMRMSCL:
7484         if (!NVME_CAP_PMRS(cap)) {
7485             return;
7486         }
7487 
7488         stl_le_p(&n->bar.pmrmscl, data);
7489         n->pmr.cmse = false;
7490 
7491         if (NVME_PMRMSCL_CMSE(data)) {
7492             uint64_t pmrmscu = ldl_le_p(&n->bar.pmrmscu);
7493             hwaddr cba = pmrmscu << 32 |
7494                 (NVME_PMRMSCL_CBA(data) << PMRMSCL_CBA_SHIFT);
7495             if (cba + int128_get64(n->pmr.dev->mr.size) < cba) {
7496                 NVME_PMRSTS_SET_CBAI(pmrsts, 1);
7497                 stl_le_p(&n->bar.pmrsts, pmrsts);
7498                 return;
7499             }
7500 
7501             n->pmr.cmse = true;
7502             n->pmr.cba = cba;
7503         }
7504 
7505         return;
7506     case NVME_REG_PMRMSCU:
7507         if (!NVME_CAP_PMRS(cap)) {
7508             return;
7509         }
7510 
7511         stl_le_p(&n->bar.pmrmscu, data);
7512         return;
7513     default:
7514         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
7515                        "invalid MMIO write,"
7516                        " offset=0x%"PRIx64", data=%"PRIx64"",
7517                        offset, data);
7518         break;
7519     }
7520 }
7521 
7522 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
7523 {
7524     NvmeCtrl *n = (NvmeCtrl *)opaque;
7525     uint8_t *ptr = (uint8_t *)&n->bar;
7526 
7527     trace_pci_nvme_mmio_read(addr, size);
7528 
7529     if (unlikely(addr & (sizeof(uint32_t) - 1))) {
7530         NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
7531                        "MMIO read not 32-bit aligned,"
7532                        " offset=0x%"PRIx64"", addr);
7533         /* should RAZ, fall through for now */
7534     } else if (unlikely(size < sizeof(uint32_t))) {
7535         NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
7536                        "MMIO read smaller than 32-bits,"
7537                        " offset=0x%"PRIx64"", addr);
7538         /* should RAZ, fall through for now */
7539     }
7540 
7541     if (addr > sizeof(n->bar) - size) {
7542         NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
7543                        "MMIO read beyond last register,"
7544                        " offset=0x%"PRIx64", returning 0", addr);
7545 
7546         return 0;
7547     }
7548 
7549     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7550         addr != NVME_REG_CSTS) {
7551         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7552         return 0;
7553     }
7554 
7555     /*
7556      * When PMRWBM bit 1 is set then read from
7557      * from PMRSTS should ensure prior writes
7558      * made it to persistent media
7559      */
7560     if (addr == NVME_REG_PMRSTS &&
7561         (NVME_PMRCAP_PMRWBM(ldl_le_p(&n->bar.pmrcap)) & 0x02)) {
7562         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7563     }
7564 
7565     return ldn_le_p(ptr + addr, size);
7566 }
7567 
7568 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
7569 {
7570     PCIDevice *pci = PCI_DEVICE(n);
7571     uint32_t qid;
7572 
7573     if (unlikely(addr & ((1 << 2) - 1))) {
7574         NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
7575                        "doorbell write not 32-bit aligned,"
7576                        " offset=0x%"PRIx64", ignoring", addr);
7577         return;
7578     }
7579 
7580     if (((addr - 0x1000) >> 2) & 1) {
7581         /* Completion queue doorbell write */
7582 
7583         uint16_t new_head = val & 0xffff;
7584         int start_sqs;
7585         NvmeCQueue *cq;
7586 
7587         qid = (addr - (0x1000 + (1 << 2))) >> 3;
7588         if (unlikely(nvme_check_cqid(n, qid))) {
7589             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
7590                            "completion queue doorbell write"
7591                            " for nonexistent queue,"
7592                            " sqid=%"PRIu32", ignoring", qid);
7593 
7594             /*
7595              * NVM Express v1.3d, Section 4.1 state: "If host software writes
7596              * an invalid value to the Submission Queue Tail Doorbell or
7597              * Completion Queue Head Doorbell regiter and an Asynchronous Event
7598              * Request command is outstanding, then an asynchronous event is
7599              * posted to the Admin Completion Queue with a status code of
7600              * Invalid Doorbell Write Value."
7601              *
7602              * Also note that the spec includes the "Invalid Doorbell Register"
7603              * status code, but nowhere does it specify when to use it.
7604              * However, it seems reasonable to use it here in a similar
7605              * fashion.
7606              */
7607             if (n->outstanding_aers) {
7608                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7609                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7610                                    NVME_LOG_ERROR_INFO);
7611             }
7612 
7613             return;
7614         }
7615 
7616         cq = n->cq[qid];
7617         if (unlikely(new_head >= cq->size)) {
7618             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
7619                            "completion queue doorbell write value"
7620                            " beyond queue size, sqid=%"PRIu32","
7621                            " new_head=%"PRIu16", ignoring",
7622                            qid, new_head);
7623 
7624             if (n->outstanding_aers) {
7625                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7626                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7627                                    NVME_LOG_ERROR_INFO);
7628             }
7629 
7630             return;
7631         }
7632 
7633         trace_pci_nvme_mmio_doorbell_cq(cq->cqid, new_head);
7634 
7635         start_sqs = nvme_cq_full(cq) ? 1 : 0;
7636         cq->head = new_head;
7637         if (!qid && n->dbbuf_enabled) {
7638             stl_le_pci_dma(pci, cq->db_addr, cq->head, MEMTXATTRS_UNSPECIFIED);
7639         }
7640         if (start_sqs) {
7641             NvmeSQueue *sq;
7642             QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
7643                 qemu_bh_schedule(sq->bh);
7644             }
7645             qemu_bh_schedule(cq->bh);
7646         }
7647 
7648         if (cq->tail == cq->head) {
7649             if (cq->irq_enabled) {
7650                 n->cq_pending--;
7651             }
7652 
7653             nvme_irq_deassert(n, cq);
7654         }
7655     } else {
7656         /* Submission queue doorbell write */
7657 
7658         uint16_t new_tail = val & 0xffff;
7659         NvmeSQueue *sq;
7660 
7661         qid = (addr - 0x1000) >> 3;
7662         if (unlikely(nvme_check_sqid(n, qid))) {
7663             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
7664                            "submission queue doorbell write"
7665                            " for nonexistent queue,"
7666                            " sqid=%"PRIu32", ignoring", qid);
7667 
7668             if (n->outstanding_aers) {
7669                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7670                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7671                                    NVME_LOG_ERROR_INFO);
7672             }
7673 
7674             return;
7675         }
7676 
7677         sq = n->sq[qid];
7678         if (unlikely(new_tail >= sq->size)) {
7679             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
7680                            "submission queue doorbell write value"
7681                            " beyond queue size, sqid=%"PRIu32","
7682                            " new_tail=%"PRIu16", ignoring",
7683                            qid, new_tail);
7684 
7685             if (n->outstanding_aers) {
7686                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7687                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7688                                    NVME_LOG_ERROR_INFO);
7689             }
7690 
7691             return;
7692         }
7693 
7694         trace_pci_nvme_mmio_doorbell_sq(sq->sqid, new_tail);
7695 
7696         sq->tail = new_tail;
7697         if (!qid && n->dbbuf_enabled) {
7698             /*
7699              * The spec states "the host shall also update the controller's
7700              * corresponding doorbell property to match the value of that entry
7701              * in the Shadow Doorbell buffer."
7702              *
7703              * Since this context is currently a VM trap, we can safely enforce
7704              * the requirement from the device side in case the host is
7705              * misbehaving.
7706              *
7707              * Note, we shouldn't have to do this, but various drivers
7708              * including ones that run on Linux, are not updating Admin Queues,
7709              * so we can't trust reading it for an appropriate sq tail.
7710              */
7711             stl_le_pci_dma(pci, sq->db_addr, sq->tail, MEMTXATTRS_UNSPECIFIED);
7712         }
7713 
7714         qemu_bh_schedule(sq->bh);
7715     }
7716 }
7717 
7718 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
7719                             unsigned size)
7720 {
7721     NvmeCtrl *n = (NvmeCtrl *)opaque;
7722 
7723     trace_pci_nvme_mmio_write(addr, data, size);
7724 
7725     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7726         addr != NVME_REG_CSTS) {
7727         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7728         return;
7729     }
7730 
7731     if (addr < sizeof(n->bar)) {
7732         nvme_write_bar(n, addr, data, size);
7733     } else {
7734         nvme_process_db(n, addr, data);
7735     }
7736 }
7737 
7738 static const MemoryRegionOps nvme_mmio_ops = {
7739     .read = nvme_mmio_read,
7740     .write = nvme_mmio_write,
7741     .endianness = DEVICE_LITTLE_ENDIAN,
7742     .impl = {
7743         .min_access_size = 2,
7744         .max_access_size = 8,
7745     },
7746 };
7747 
7748 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
7749                            unsigned size)
7750 {
7751     NvmeCtrl *n = (NvmeCtrl *)opaque;
7752     stn_le_p(&n->cmb.buf[addr], size, data);
7753 }
7754 
7755 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
7756 {
7757     NvmeCtrl *n = (NvmeCtrl *)opaque;
7758     return ldn_le_p(&n->cmb.buf[addr], size);
7759 }
7760 
7761 static const MemoryRegionOps nvme_cmb_ops = {
7762     .read = nvme_cmb_read,
7763     .write = nvme_cmb_write,
7764     .endianness = DEVICE_LITTLE_ENDIAN,
7765     .impl = {
7766         .min_access_size = 1,
7767         .max_access_size = 8,
7768     },
7769 };
7770 
7771 static bool nvme_check_params(NvmeCtrl *n, Error **errp)
7772 {
7773     NvmeParams *params = &n->params;
7774 
7775     if (params->num_queues) {
7776         warn_report("num_queues is deprecated; please use max_ioqpairs "
7777                     "instead");
7778 
7779         params->max_ioqpairs = params->num_queues - 1;
7780     }
7781 
7782     if (n->namespace.blkconf.blk && n->subsys) {
7783         error_setg(errp, "subsystem support is unavailable with legacy "
7784                    "namespace ('drive' property)");
7785         return false;
7786     }
7787 
7788     if (params->max_ioqpairs < 1 ||
7789         params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
7790         error_setg(errp, "max_ioqpairs must be between 1 and %d",
7791                    NVME_MAX_IOQPAIRS);
7792         return false;
7793     }
7794 
7795     if (params->msix_qsize < 1 ||
7796         params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
7797         error_setg(errp, "msix_qsize must be between 1 and %d",
7798                    PCI_MSIX_FLAGS_QSIZE + 1);
7799         return false;
7800     }
7801 
7802     if (!params->serial) {
7803         error_setg(errp, "serial property not set");
7804         return false;
7805     }
7806 
7807     if (n->pmr.dev) {
7808         if (host_memory_backend_is_mapped(n->pmr.dev)) {
7809             error_setg(errp, "can't use already busy memdev: %s",
7810                        object_get_canonical_path_component(OBJECT(n->pmr.dev)));
7811             return false;
7812         }
7813 
7814         if (!is_power_of_2(n->pmr.dev->size)) {
7815             error_setg(errp, "pmr backend size needs to be power of 2 in size");
7816             return false;
7817         }
7818 
7819         host_memory_backend_set_mapped(n->pmr.dev, true);
7820     }
7821 
7822     if (n->params.zasl > n->params.mdts) {
7823         error_setg(errp, "zoned.zasl (Zone Append Size Limit) must be less "
7824                    "than or equal to mdts (Maximum Data Transfer Size)");
7825         return false;
7826     }
7827 
7828     if (!n->params.vsl) {
7829         error_setg(errp, "vsl must be non-zero");
7830         return false;
7831     }
7832 
7833     if (params->sriov_max_vfs) {
7834         if (!n->subsys) {
7835             error_setg(errp, "subsystem is required for the use of SR-IOV");
7836             return false;
7837         }
7838 
7839         if (params->sriov_max_vfs > NVME_MAX_VFS) {
7840             error_setg(errp, "sriov_max_vfs must be between 0 and %d",
7841                        NVME_MAX_VFS);
7842             return false;
7843         }
7844 
7845         if (params->cmb_size_mb) {
7846             error_setg(errp, "CMB is not supported with SR-IOV");
7847             return false;
7848         }
7849 
7850         if (n->pmr.dev) {
7851             error_setg(errp, "PMR is not supported with SR-IOV");
7852             return false;
7853         }
7854 
7855         if (!params->sriov_vq_flexible || !params->sriov_vi_flexible) {
7856             error_setg(errp, "both sriov_vq_flexible and sriov_vi_flexible"
7857                        " must be set for the use of SR-IOV");
7858             return false;
7859         }
7860 
7861         if (params->sriov_vq_flexible < params->sriov_max_vfs * 2) {
7862             error_setg(errp, "sriov_vq_flexible must be greater than or equal"
7863                        " to %d (sriov_max_vfs * 2)", params->sriov_max_vfs * 2);
7864             return false;
7865         }
7866 
7867         if (params->max_ioqpairs < params->sriov_vq_flexible + 2) {
7868             error_setg(errp, "(max_ioqpairs - sriov_vq_flexible) must be"
7869                        " greater than or equal to 2");
7870             return false;
7871         }
7872 
7873         if (params->sriov_vi_flexible < params->sriov_max_vfs) {
7874             error_setg(errp, "sriov_vi_flexible must be greater than or equal"
7875                        " to %d (sriov_max_vfs)", params->sriov_max_vfs);
7876             return false;
7877         }
7878 
7879         if (params->msix_qsize < params->sriov_vi_flexible + 1) {
7880             error_setg(errp, "(msix_qsize - sriov_vi_flexible) must be"
7881                        " greater than or equal to 1");
7882             return false;
7883         }
7884 
7885         if (params->sriov_max_vi_per_vf &&
7886             (params->sriov_max_vi_per_vf - 1) % NVME_VF_RES_GRANULARITY) {
7887             error_setg(errp, "sriov_max_vi_per_vf must meet:"
7888                        " (sriov_max_vi_per_vf - 1) %% %d == 0 and"
7889                        " sriov_max_vi_per_vf >= 1", NVME_VF_RES_GRANULARITY);
7890             return false;
7891         }
7892 
7893         if (params->sriov_max_vq_per_vf &&
7894             (params->sriov_max_vq_per_vf < 2 ||
7895              (params->sriov_max_vq_per_vf - 1) % NVME_VF_RES_GRANULARITY)) {
7896             error_setg(errp, "sriov_max_vq_per_vf must meet:"
7897                        " (sriov_max_vq_per_vf - 1) %% %d == 0 and"
7898                        " sriov_max_vq_per_vf >= 2", NVME_VF_RES_GRANULARITY);
7899             return false;
7900         }
7901     }
7902 
7903     return true;
7904 }
7905 
7906 static void nvme_init_state(NvmeCtrl *n)
7907 {
7908     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7909     NvmeSecCtrlList *list = &n->sec_ctrl_list;
7910     NvmeSecCtrlEntry *sctrl;
7911     PCIDevice *pci = PCI_DEVICE(n);
7912     uint8_t max_vfs;
7913     int i;
7914 
7915     if (pci_is_vf(pci)) {
7916         sctrl = nvme_sctrl(n);
7917         max_vfs = 0;
7918         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7919         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7920     } else {
7921         max_vfs = n->params.sriov_max_vfs;
7922         n->conf_ioqpairs = n->params.max_ioqpairs;
7923         n->conf_msix_qsize = n->params.msix_qsize;
7924     }
7925 
7926     n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
7927     n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
7928     n->temperature = NVME_TEMPERATURE;
7929     n->features.temp_thresh_hi = NVME_TEMPERATURE_WARNING;
7930     n->starttime_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
7931     n->aer_reqs = g_new0(NvmeRequest *, n->params.aerl + 1);
7932     QTAILQ_INIT(&n->aer_queue);
7933 
7934     list->numcntl = cpu_to_le16(max_vfs);
7935     for (i = 0; i < max_vfs; i++) {
7936         sctrl = &list->sec[i];
7937         sctrl->pcid = cpu_to_le16(n->cntlid);
7938         sctrl->vfn = cpu_to_le16(i + 1);
7939     }
7940 
7941     cap->cntlid = cpu_to_le16(n->cntlid);
7942     cap->crt = NVME_CRT_VQ | NVME_CRT_VI;
7943 
7944     if (pci_is_vf(pci)) {
7945         cap->vqprt = cpu_to_le16(1 + n->conf_ioqpairs);
7946     } else {
7947         cap->vqprt = cpu_to_le16(1 + n->params.max_ioqpairs -
7948                                  n->params.sriov_vq_flexible);
7949         cap->vqfrt = cpu_to_le32(n->params.sriov_vq_flexible);
7950         cap->vqrfap = cap->vqfrt;
7951         cap->vqgran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
7952         cap->vqfrsm = n->params.sriov_max_vq_per_vf ?
7953                         cpu_to_le16(n->params.sriov_max_vq_per_vf) :
7954                         cap->vqfrt / MAX(max_vfs, 1);
7955     }
7956 
7957     if (pci_is_vf(pci)) {
7958         cap->viprt = cpu_to_le16(n->conf_msix_qsize);
7959     } else {
7960         cap->viprt = cpu_to_le16(n->params.msix_qsize -
7961                                  n->params.sriov_vi_flexible);
7962         cap->vifrt = cpu_to_le32(n->params.sriov_vi_flexible);
7963         cap->virfap = cap->vifrt;
7964         cap->vigran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
7965         cap->vifrsm = n->params.sriov_max_vi_per_vf ?
7966                         cpu_to_le16(n->params.sriov_max_vi_per_vf) :
7967                         cap->vifrt / MAX(max_vfs, 1);
7968     }
7969 }
7970 
7971 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
7972 {
7973     uint64_t cmb_size = n->params.cmb_size_mb * MiB;
7974     uint64_t cap = ldq_le_p(&n->bar.cap);
7975 
7976     n->cmb.buf = g_malloc0(cmb_size);
7977     memory_region_init_io(&n->cmb.mem, OBJECT(n), &nvme_cmb_ops, n,
7978                           "nvme-cmb", cmb_size);
7979     pci_register_bar(pci_dev, NVME_CMB_BIR,
7980                      PCI_BASE_ADDRESS_SPACE_MEMORY |
7981                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
7982                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->cmb.mem);
7983 
7984     NVME_CAP_SET_CMBS(cap, 1);
7985     stq_le_p(&n->bar.cap, cap);
7986 
7987     if (n->params.legacy_cmb) {
7988         nvme_cmb_enable_regs(n);
7989         n->cmb.cmse = true;
7990     }
7991 }
7992 
7993 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
7994 {
7995     uint32_t pmrcap = ldl_le_p(&n->bar.pmrcap);
7996 
7997     NVME_PMRCAP_SET_RDS(pmrcap, 1);
7998     NVME_PMRCAP_SET_WDS(pmrcap, 1);
7999     NVME_PMRCAP_SET_BIR(pmrcap, NVME_PMR_BIR);
8000     /* Turn on bit 1 support */
8001     NVME_PMRCAP_SET_PMRWBM(pmrcap, 0x02);
8002     NVME_PMRCAP_SET_CMSS(pmrcap, 1);
8003     stl_le_p(&n->bar.pmrcap, pmrcap);
8004 
8005     pci_register_bar(pci_dev, NVME_PMR_BIR,
8006                      PCI_BASE_ADDRESS_SPACE_MEMORY |
8007                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
8008                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmr.dev->mr);
8009 
8010     memory_region_set_enabled(&n->pmr.dev->mr, false);
8011 }
8012 
8013 static uint64_t nvme_bar_size(unsigned total_queues, unsigned total_irqs,
8014                               unsigned *msix_table_offset,
8015                               unsigned *msix_pba_offset)
8016 {
8017     uint64_t bar_size, msix_table_size, msix_pba_size;
8018 
8019     bar_size = sizeof(NvmeBar) + 2 * total_queues * NVME_DB_SIZE;
8020     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8021 
8022     if (msix_table_offset) {
8023         *msix_table_offset = bar_size;
8024     }
8025 
8026     msix_table_size = PCI_MSIX_ENTRY_SIZE * total_irqs;
8027     bar_size += msix_table_size;
8028     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8029 
8030     if (msix_pba_offset) {
8031         *msix_pba_offset = bar_size;
8032     }
8033 
8034     msix_pba_size = QEMU_ALIGN_UP(total_irqs, 64) / 8;
8035     bar_size += msix_pba_size;
8036 
8037     bar_size = pow2ceil(bar_size);
8038     return bar_size;
8039 }
8040 
8041 static void nvme_init_sriov(NvmeCtrl *n, PCIDevice *pci_dev, uint16_t offset)
8042 {
8043     uint16_t vf_dev_id = n->params.use_intel_id ?
8044                          PCI_DEVICE_ID_INTEL_NVME : PCI_DEVICE_ID_REDHAT_NVME;
8045     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
8046     uint64_t bar_size = nvme_bar_size(le16_to_cpu(cap->vqfrsm),
8047                                       le16_to_cpu(cap->vifrsm),
8048                                       NULL, NULL);
8049 
8050     pcie_sriov_pf_init(pci_dev, offset, "nvme", vf_dev_id,
8051                        n->params.sriov_max_vfs, n->params.sriov_max_vfs,
8052                        NVME_VF_OFFSET, NVME_VF_STRIDE);
8053 
8054     pcie_sriov_pf_init_vf_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8055                               PCI_BASE_ADDRESS_MEM_TYPE_64, bar_size);
8056 }
8057 
8058 static int nvme_add_pm_capability(PCIDevice *pci_dev, uint8_t offset)
8059 {
8060     Error *err = NULL;
8061     int ret;
8062 
8063     ret = pci_add_capability(pci_dev, PCI_CAP_ID_PM, offset,
8064                              PCI_PM_SIZEOF, &err);
8065     if (err) {
8066         error_report_err(err);
8067         return ret;
8068     }
8069 
8070     pci_set_word(pci_dev->config + offset + PCI_PM_PMC,
8071                  PCI_PM_CAP_VER_1_2);
8072     pci_set_word(pci_dev->config + offset + PCI_PM_CTRL,
8073                  PCI_PM_CTRL_NO_SOFT_RESET);
8074     pci_set_word(pci_dev->wmask + offset + PCI_PM_CTRL,
8075                  PCI_PM_CTRL_STATE_MASK);
8076 
8077     return 0;
8078 }
8079 
8080 static bool nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
8081 {
8082     ERRP_GUARD();
8083     uint8_t *pci_conf = pci_dev->config;
8084     uint64_t bar_size;
8085     unsigned msix_table_offset, msix_pba_offset;
8086     int ret;
8087 
8088     pci_conf[PCI_INTERRUPT_PIN] = 1;
8089     pci_config_set_prog_interface(pci_conf, 0x2);
8090 
8091     if (n->params.use_intel_id) {
8092         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_INTEL);
8093         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_INTEL_NVME);
8094     } else {
8095         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_REDHAT);
8096         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_REDHAT_NVME);
8097     }
8098 
8099     pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
8100     nvme_add_pm_capability(pci_dev, 0x60);
8101     pcie_endpoint_cap_init(pci_dev, 0x80);
8102     pcie_cap_flr_init(pci_dev);
8103     if (n->params.sriov_max_vfs) {
8104         pcie_ari_init(pci_dev, 0x100);
8105     }
8106 
8107     /* add one to max_ioqpairs to account for the admin queue pair */
8108     bar_size = nvme_bar_size(n->params.max_ioqpairs + 1, n->params.msix_qsize,
8109                              &msix_table_offset, &msix_pba_offset);
8110 
8111     memory_region_init(&n->bar0, OBJECT(n), "nvme-bar0", bar_size);
8112     memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
8113                           msix_table_offset);
8114     memory_region_add_subregion(&n->bar0, 0, &n->iomem);
8115 
8116     if (pci_is_vf(pci_dev)) {
8117         pcie_sriov_vf_register_bar(pci_dev, 0, &n->bar0);
8118     } else {
8119         pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8120                          PCI_BASE_ADDRESS_MEM_TYPE_64, &n->bar0);
8121     }
8122     ret = msix_init(pci_dev, n->params.msix_qsize,
8123                     &n->bar0, 0, msix_table_offset,
8124                     &n->bar0, 0, msix_pba_offset, 0, errp);
8125     if (ret == -ENOTSUP) {
8126         /* report that msix is not supported, but do not error out */
8127         warn_report_err(*errp);
8128         *errp = NULL;
8129     } else if (ret < 0) {
8130         /* propagate error to caller */
8131         return false;
8132     }
8133 
8134     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
8135 
8136     if (n->params.cmb_size_mb) {
8137         nvme_init_cmb(n, pci_dev);
8138     }
8139 
8140     if (n->pmr.dev) {
8141         nvme_init_pmr(n, pci_dev);
8142     }
8143 
8144     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8145         nvme_init_sriov(n, pci_dev, 0x120);
8146     }
8147 
8148     return true;
8149 }
8150 
8151 static void nvme_init_subnqn(NvmeCtrl *n)
8152 {
8153     NvmeSubsystem *subsys = n->subsys;
8154     NvmeIdCtrl *id = &n->id_ctrl;
8155 
8156     if (!subsys) {
8157         snprintf((char *)id->subnqn, sizeof(id->subnqn),
8158                  "nqn.2019-08.org.qemu:%s", n->params.serial);
8159     } else {
8160         pstrcpy((char *)id->subnqn, sizeof(id->subnqn), (char*)subsys->subnqn);
8161     }
8162 }
8163 
8164 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
8165 {
8166     NvmeIdCtrl *id = &n->id_ctrl;
8167     uint8_t *pci_conf = pci_dev->config;
8168     uint64_t cap = ldq_le_p(&n->bar.cap);
8169     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
8170     uint32_t ctratt;
8171 
8172     id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
8173     id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
8174     strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
8175     strpadcpy((char *)id->fr, sizeof(id->fr), QEMU_VERSION, ' ');
8176     strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
8177 
8178     id->cntlid = cpu_to_le16(n->cntlid);
8179 
8180     id->oaes = cpu_to_le32(NVME_OAES_NS_ATTR);
8181     ctratt = NVME_CTRATT_ELBAS;
8182 
8183     id->rab = 6;
8184 
8185     if (n->params.use_intel_id) {
8186         id->ieee[0] = 0xb3;
8187         id->ieee[1] = 0x02;
8188         id->ieee[2] = 0x00;
8189     } else {
8190         id->ieee[0] = 0x00;
8191         id->ieee[1] = 0x54;
8192         id->ieee[2] = 0x52;
8193     }
8194 
8195     id->mdts = n->params.mdts;
8196     id->ver = cpu_to_le32(NVME_SPEC_VER);
8197     id->oacs =
8198         cpu_to_le16(NVME_OACS_NS_MGMT | NVME_OACS_FORMAT | NVME_OACS_DBBUF |
8199                     NVME_OACS_DIRECTIVES);
8200     id->cntrltype = 0x1;
8201 
8202     /*
8203      * Because the controller always completes the Abort command immediately,
8204      * there can never be more than one concurrently executing Abort command,
8205      * so this value is never used for anything. Note that there can easily be
8206      * many Abort commands in the queues, but they are not considered
8207      * "executing" until processed by nvme_abort.
8208      *
8209      * The specification recommends a value of 3 for Abort Command Limit (four
8210      * concurrently outstanding Abort commands), so lets use that though it is
8211      * inconsequential.
8212      */
8213     id->acl = 3;
8214     id->aerl = n->params.aerl;
8215     id->frmw = (NVME_NUM_FW_SLOTS << 1) | NVME_FRMW_SLOT1_RO;
8216     id->lpa = NVME_LPA_NS_SMART | NVME_LPA_CSE | NVME_LPA_EXTENDED;
8217 
8218     /* recommended default value (~70 C) */
8219     id->wctemp = cpu_to_le16(NVME_TEMPERATURE_WARNING);
8220     id->cctemp = cpu_to_le16(NVME_TEMPERATURE_CRITICAL);
8221 
8222     id->sqes = (NVME_SQES << 4) | NVME_SQES;
8223     id->cqes = (NVME_CQES << 4) | NVME_CQES;
8224     id->nn = cpu_to_le32(NVME_MAX_NAMESPACES);
8225     id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP |
8226                            NVME_ONCS_FEATURES | NVME_ONCS_DSM |
8227                            NVME_ONCS_COMPARE | NVME_ONCS_COPY);
8228 
8229     /*
8230      * NOTE: If this device ever supports a command set that does NOT use 0x0
8231      * as a Flush-equivalent operation, support for the broadcast NSID in Flush
8232      * should probably be removed.
8233      *
8234      * See comment in nvme_io_cmd.
8235      */
8236     id->vwc = NVME_VWC_NSID_BROADCAST_SUPPORT | NVME_VWC_PRESENT;
8237 
8238     id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0 | NVME_OCFS_COPY_FORMAT_1);
8239     id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN);
8240 
8241     nvme_init_subnqn(n);
8242 
8243     id->psd[0].mp = cpu_to_le16(0x9c4);
8244     id->psd[0].enlat = cpu_to_le32(0x10);
8245     id->psd[0].exlat = cpu_to_le32(0x4);
8246 
8247     if (n->subsys) {
8248         id->cmic |= NVME_CMIC_MULTI_CTRL;
8249         ctratt |= NVME_CTRATT_ENDGRPS;
8250 
8251         id->endgidmax = cpu_to_le16(0x1);
8252 
8253         if (n->subsys->endgrp.fdp.enabled) {
8254             ctratt |= NVME_CTRATT_FDPS;
8255         }
8256     }
8257 
8258     id->ctratt = cpu_to_le32(ctratt);
8259 
8260     NVME_CAP_SET_MQES(cap, 0x7ff);
8261     NVME_CAP_SET_CQR(cap, 1);
8262     NVME_CAP_SET_TO(cap, 0xf);
8263     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_NVM);
8264     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_CSI_SUPP);
8265     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_ADMIN_ONLY);
8266     NVME_CAP_SET_MPSMAX(cap, 4);
8267     NVME_CAP_SET_CMBS(cap, n->params.cmb_size_mb ? 1 : 0);
8268     NVME_CAP_SET_PMRS(cap, n->pmr.dev ? 1 : 0);
8269     stq_le_p(&n->bar.cap, cap);
8270 
8271     stl_le_p(&n->bar.vs, NVME_SPEC_VER);
8272     n->bar.intmc = n->bar.intms = 0;
8273 
8274     if (pci_is_vf(pci_dev) && !sctrl->scs) {
8275         stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
8276     }
8277 }
8278 
8279 static int nvme_init_subsys(NvmeCtrl *n, Error **errp)
8280 {
8281     int cntlid;
8282 
8283     if (!n->subsys) {
8284         return 0;
8285     }
8286 
8287     cntlid = nvme_subsys_register_ctrl(n, errp);
8288     if (cntlid < 0) {
8289         return -1;
8290     }
8291 
8292     n->cntlid = cntlid;
8293 
8294     return 0;
8295 }
8296 
8297 void nvme_attach_ns(NvmeCtrl *n, NvmeNamespace *ns)
8298 {
8299     uint32_t nsid = ns->params.nsid;
8300     assert(nsid && nsid <= NVME_MAX_NAMESPACES);
8301 
8302     n->namespaces[nsid] = ns;
8303     ns->attached++;
8304 
8305     n->dmrsl = MIN_NON_ZERO(n->dmrsl,
8306                             BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
8307 }
8308 
8309 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
8310 {
8311     NvmeCtrl *n = NVME(pci_dev);
8312     DeviceState *dev = DEVICE(pci_dev);
8313     NvmeNamespace *ns;
8314     NvmeCtrl *pn = NVME(pcie_sriov_get_pf(pci_dev));
8315 
8316     if (pci_is_vf(pci_dev)) {
8317         /*
8318          * VFs derive settings from the parent. PF's lifespan exceeds
8319          * that of VF's, so it's safe to share params.serial.
8320          */
8321         memcpy(&n->params, &pn->params, sizeof(NvmeParams));
8322         n->subsys = pn->subsys;
8323     }
8324 
8325     if (!nvme_check_params(n, errp)) {
8326         return;
8327     }
8328 
8329     qbus_init(&n->bus, sizeof(NvmeBus), TYPE_NVME_BUS, dev, dev->id);
8330 
8331     if (nvme_init_subsys(n, errp)) {
8332         return;
8333     }
8334     nvme_init_state(n);
8335     if (!nvme_init_pci(n, pci_dev, errp)) {
8336         return;
8337     }
8338     nvme_init_ctrl(n, pci_dev);
8339 
8340     /* setup a namespace if the controller drive property was given */
8341     if (n->namespace.blkconf.blk) {
8342         ns = &n->namespace;
8343         ns->params.nsid = 1;
8344 
8345         if (nvme_ns_setup(ns, errp)) {
8346             return;
8347         }
8348 
8349         nvme_attach_ns(n, ns);
8350     }
8351 }
8352 
8353 static void nvme_exit(PCIDevice *pci_dev)
8354 {
8355     NvmeCtrl *n = NVME(pci_dev);
8356     NvmeNamespace *ns;
8357     int i;
8358 
8359     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8360 
8361     if (n->subsys) {
8362         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
8363             ns = nvme_ns(n, i);
8364             if (ns) {
8365                 ns->attached--;
8366             }
8367         }
8368 
8369         nvme_subsys_unregister_ctrl(n->subsys, n);
8370     }
8371 
8372     g_free(n->cq);
8373     g_free(n->sq);
8374     g_free(n->aer_reqs);
8375 
8376     if (n->params.cmb_size_mb) {
8377         g_free(n->cmb.buf);
8378     }
8379 
8380     if (n->pmr.dev) {
8381         host_memory_backend_set_mapped(n->pmr.dev, false);
8382     }
8383 
8384     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8385         pcie_sriov_pf_exit(pci_dev);
8386     }
8387 
8388     msix_uninit(pci_dev, &n->bar0, &n->bar0);
8389     memory_region_del_subregion(&n->bar0, &n->iomem);
8390 }
8391 
8392 static Property nvme_props[] = {
8393     DEFINE_BLOCK_PROPERTIES(NvmeCtrl, namespace.blkconf),
8394     DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmr.dev, TYPE_MEMORY_BACKEND,
8395                      HostMemoryBackend *),
8396     DEFINE_PROP_LINK("subsys", NvmeCtrl, subsys, TYPE_NVME_SUBSYS,
8397                      NvmeSubsystem *),
8398     DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
8399     DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
8400     DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
8401     DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
8402     DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
8403     DEFINE_PROP_UINT8("aerl", NvmeCtrl, params.aerl, 3),
8404     DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl, params.aer_max_queued, 64),
8405     DEFINE_PROP_UINT8("mdts", NvmeCtrl, params.mdts, 7),
8406     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
8407     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
8408     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
8409     DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl, params.ioeventfd, false),
8410     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
8411     DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl,
8412                      params.auto_transition_zones, true),
8413     DEFINE_PROP_UINT8("sriov_max_vfs", NvmeCtrl, params.sriov_max_vfs, 0),
8414     DEFINE_PROP_UINT16("sriov_vq_flexible", NvmeCtrl,
8415                        params.sriov_vq_flexible, 0),
8416     DEFINE_PROP_UINT16("sriov_vi_flexible", NvmeCtrl,
8417                        params.sriov_vi_flexible, 0),
8418     DEFINE_PROP_UINT8("sriov_max_vi_per_vf", NvmeCtrl,
8419                       params.sriov_max_vi_per_vf, 0),
8420     DEFINE_PROP_UINT8("sriov_max_vq_per_vf", NvmeCtrl,
8421                       params.sriov_max_vq_per_vf, 0),
8422     DEFINE_PROP_END_OF_LIST(),
8423 };
8424 
8425 static void nvme_get_smart_warning(Object *obj, Visitor *v, const char *name,
8426                                    void *opaque, Error **errp)
8427 {
8428     NvmeCtrl *n = NVME(obj);
8429     uint8_t value = n->smart_critical_warning;
8430 
8431     visit_type_uint8(v, name, &value, errp);
8432 }
8433 
8434 static void nvme_set_smart_warning(Object *obj, Visitor *v, const char *name,
8435                                    void *opaque, Error **errp)
8436 {
8437     NvmeCtrl *n = NVME(obj);
8438     uint8_t value, old_value, cap = 0, index, event;
8439 
8440     if (!visit_type_uint8(v, name, &value, errp)) {
8441         return;
8442     }
8443 
8444     cap = NVME_SMART_SPARE | NVME_SMART_TEMPERATURE | NVME_SMART_RELIABILITY
8445           | NVME_SMART_MEDIA_READ_ONLY | NVME_SMART_FAILED_VOLATILE_MEDIA;
8446     if (NVME_CAP_PMRS(ldq_le_p(&n->bar.cap))) {
8447         cap |= NVME_SMART_PMR_UNRELIABLE;
8448     }
8449 
8450     if ((value & cap) != value) {
8451         error_setg(errp, "unsupported smart critical warning bits: 0x%x",
8452                    value & ~cap);
8453         return;
8454     }
8455 
8456     old_value = n->smart_critical_warning;
8457     n->smart_critical_warning = value;
8458 
8459     /* only inject new bits of smart critical warning */
8460     for (index = 0; index < NVME_SMART_WARN_MAX; index++) {
8461         event = 1 << index;
8462         if (value & ~old_value & event)
8463             nvme_smart_event(n, event);
8464     }
8465 }
8466 
8467 static void nvme_pci_reset(DeviceState *qdev)
8468 {
8469     PCIDevice *pci_dev = PCI_DEVICE(qdev);
8470     NvmeCtrl *n = NVME(pci_dev);
8471 
8472     trace_pci_nvme_pci_reset();
8473     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8474 }
8475 
8476 static void nvme_sriov_pre_write_ctrl(PCIDevice *dev, uint32_t address,
8477                                       uint32_t val, int len)
8478 {
8479     NvmeCtrl *n = NVME(dev);
8480     NvmeSecCtrlEntry *sctrl;
8481     uint16_t sriov_cap = dev->exp.sriov_cap;
8482     uint32_t off = address - sriov_cap;
8483     int i, num_vfs;
8484 
8485     if (!sriov_cap) {
8486         return;
8487     }
8488 
8489     if (range_covers_byte(off, len, PCI_SRIOV_CTRL)) {
8490         if (!(val & PCI_SRIOV_CTRL_VFE)) {
8491             num_vfs = pci_get_word(dev->config + sriov_cap + PCI_SRIOV_NUM_VF);
8492             for (i = 0; i < num_vfs; i++) {
8493                 sctrl = &n->sec_ctrl_list.sec[i];
8494                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
8495             }
8496         }
8497     }
8498 }
8499 
8500 static void nvme_pci_write_config(PCIDevice *dev, uint32_t address,
8501                                   uint32_t val, int len)
8502 {
8503     nvme_sriov_pre_write_ctrl(dev, address, val, len);
8504     pci_default_write_config(dev, address, val, len);
8505     pcie_cap_flr_write_config(dev, address, val, len);
8506 }
8507 
8508 static const VMStateDescription nvme_vmstate = {
8509     .name = "nvme",
8510     .unmigratable = 1,
8511 };
8512 
8513 static void nvme_class_init(ObjectClass *oc, void *data)
8514 {
8515     DeviceClass *dc = DEVICE_CLASS(oc);
8516     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
8517 
8518     pc->realize = nvme_realize;
8519     pc->config_write = nvme_pci_write_config;
8520     pc->exit = nvme_exit;
8521     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
8522     pc->revision = 2;
8523 
8524     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
8525     dc->desc = "Non-Volatile Memory Express";
8526     device_class_set_props(dc, nvme_props);
8527     dc->vmsd = &nvme_vmstate;
8528     dc->reset = nvme_pci_reset;
8529 }
8530 
8531 static void nvme_instance_init(Object *obj)
8532 {
8533     NvmeCtrl *n = NVME(obj);
8534 
8535     device_add_bootindex_property(obj, &n->namespace.blkconf.bootindex,
8536                                   "bootindex", "/namespace@1,0",
8537                                   DEVICE(obj));
8538 
8539     object_property_add(obj, "smart_critical_warning", "uint8",
8540                         nvme_get_smart_warning,
8541                         nvme_set_smart_warning, NULL, NULL);
8542 }
8543 
8544 static const TypeInfo nvme_info = {
8545     .name          = TYPE_NVME,
8546     .parent        = TYPE_PCI_DEVICE,
8547     .instance_size = sizeof(NvmeCtrl),
8548     .instance_init = nvme_instance_init,
8549     .class_init    = nvme_class_init,
8550     .interfaces = (InterfaceInfo[]) {
8551         { INTERFACE_PCIE_DEVICE },
8552         { }
8553     },
8554 };
8555 
8556 static const TypeInfo nvme_bus_info = {
8557     .name = TYPE_NVME_BUS,
8558     .parent = TYPE_BUS,
8559     .instance_size = sizeof(NvmeBus),
8560 };
8561 
8562 static void nvme_register_types(void)
8563 {
8564     type_register_static(&nvme_info);
8565     type_register_static(&nvme_bus_info);
8566 }
8567 
8568 type_init(nvme_register_types)
8569