xref: /openbmc/qemu/hw/nvme/ctrl.c (revision d9ab1f1f)
1 /*
2  * QEMU NVM Express Controller
3  *
4  * Copyright (c) 2012, Intel Corporation
5  *
6  * Written by Keith Busch <keith.busch@intel.com>
7  *
8  * This code is licensed under the GNU GPL v2 or later.
9  */
10 
11 /**
12  * Reference Specs: http://www.nvmexpress.org, 1.4, 1.3, 1.2, 1.1, 1.0e
13  *
14  *  https://nvmexpress.org/developers/nvme-specification/
15  *
16  *
17  * Notes on coding style
18  * ---------------------
19  * While QEMU coding style prefers lowercase hexadecimals in constants, the
20  * NVMe subsystem use thes format from the NVMe specifications in the comments
21  * (i.e. 'h' suffix instead of '0x' prefix).
22  *
23  * Usage
24  * -----
25  * See docs/system/nvme.rst for extensive documentation.
26  *
27  * Add options:
28  *      -drive file=<file>,if=none,id=<drive_id>
29  *      -device nvme-subsys,id=<subsys_id>,nqn=<nqn_id>
30  *      -device nvme,serial=<serial>,id=<bus_name>, \
31  *              cmb_size_mb=<cmb_size_mb[optional]>, \
32  *              [pmrdev=<mem_backend_file_id>,] \
33  *              max_ioqpairs=<N[optional]>, \
34  *              aerl=<N[optional]>,aer_max_queued=<N[optional]>, \
35  *              mdts=<N[optional]>,vsl=<N[optional]>, \
36  *              zoned.zasl=<N[optional]>, \
37  *              zoned.auto_transition=<on|off[optional]>, \
38  *              sriov_max_vfs=<N[optional]> \
39  *              sriov_vq_flexible=<N[optional]> \
40  *              sriov_vi_flexible=<N[optional]> \
41  *              sriov_max_vi_per_vf=<N[optional]> \
42  *              sriov_max_vq_per_vf=<N[optional]> \
43  *              subsys=<subsys_id>
44  *      -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
45  *              zoned=<true|false[optional]>, \
46  *              subsys=<subsys_id>,shared=<true|false[optional]>, \
47  *              detached=<true|false[optional]>, \
48  *              zoned.zone_size=<N[optional]>, \
49  *              zoned.zone_capacity=<N[optional]>, \
50  *              zoned.descr_ext_size=<N[optional]>, \
51  *              zoned.max_active=<N[optional]>, \
52  *              zoned.max_open=<N[optional]>, \
53  *              zoned.cross_read=<true|false[optional]>
54  *
55  * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
56  * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. By default, the
57  * device will use the "v1.4 CMB scheme" - use the `legacy-cmb` parameter to
58  * always enable the CMBLOC and CMBSZ registers (v1.3 behavior).
59  *
60  * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
61  * For example:
62  * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
63  *  size=<size> .... -device nvme,...,pmrdev=<mem_id>
64  *
65  * The PMR will use BAR 4/5 exclusively.
66  *
67  * To place controller(s) and namespace(s) to a subsystem, then provide
68  * nvme-subsys device as above.
69  *
70  * nvme subsystem device parameters
71  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
72  * - `nqn`
73  *   This parameter provides the `<nqn_id>` part of the string
74  *   `nqn.2019-08.org.qemu:<nqn_id>` which will be reported in the SUBNQN field
75  *   of subsystem controllers. Note that `<nqn_id>` should be unique per
76  *   subsystem, but this is not enforced by QEMU. If not specified, it will
77  *   default to the value of the `id` parameter (`<subsys_id>`).
78  *
79  * nvme device parameters
80  * ~~~~~~~~~~~~~~~~~~~~~~
81  * - `subsys`
82  *   Specifying this parameter attaches the controller to the subsystem and
83  *   the SUBNQN field in the controller will report the NQN of the subsystem
84  *   device. This also enables multi controller capability represented in
85  *   Identify Controller data structure in CMIC (Controller Multi-path I/O and
86  *   Namespace Sharing Capabilities).
87  *
88  * - `aerl`
89  *   The Asynchronous Event Request Limit (AERL). Indicates the maximum number
90  *   of concurrently outstanding Asynchronous Event Request commands support
91  *   by the controller. This is a 0's based value.
92  *
93  * - `aer_max_queued`
94  *   This is the maximum number of events that the device will enqueue for
95  *   completion when there are no outstanding AERs. When the maximum number of
96  *   enqueued events are reached, subsequent events will be dropped.
97  *
98  * - `mdts`
99  *   Indicates the maximum data transfer size for a command that transfers data
100  *   between host-accessible memory and the controller. The value is specified
101  *   as a power of two (2^n) and is in units of the minimum memory page size
102  *   (CAP.MPSMIN). The default value is 7 (i.e. 512 KiB).
103  *
104  * - `vsl`
105  *   Indicates the maximum data size limit for the Verify command. Like `mdts`,
106  *   this value is specified as a power of two (2^n) and is in units of the
107  *   minimum memory page size (CAP.MPSMIN). The default value is 7 (i.e. 512
108  *   KiB).
109  *
110  * - `zoned.zasl`
111  *   Indicates the maximum data transfer size for the Zone Append command. Like
112  *   `mdts`, the value is specified as a power of two (2^n) and is in units of
113  *   the minimum memory page size (CAP.MPSMIN). The default value is 0 (i.e.
114  *   defaulting to the value of `mdts`).
115  *
116  * - `zoned.auto_transition`
117  *   Indicates if zones in zone state implicitly opened can be automatically
118  *   transitioned to zone state closed for resource management purposes.
119  *   Defaults to 'on'.
120  *
121  * - `sriov_max_vfs`
122  *   Indicates the maximum number of PCIe virtual functions supported
123  *   by the controller. The default value is 0. Specifying a non-zero value
124  *   enables reporting of both SR-IOV and ARI capabilities by the NVMe device.
125  *   Virtual function controllers will not report SR-IOV capability.
126  *
127  *   NOTE: Single Root I/O Virtualization support is experimental.
128  *   All the related parameters may be subject to change.
129  *
130  * - `sriov_vq_flexible`
131  *   Indicates the total number of flexible queue resources assignable to all
132  *   the secondary controllers. Implicitly sets the number of primary
133  *   controller's private resources to `(max_ioqpairs - sriov_vq_flexible)`.
134  *
135  * - `sriov_vi_flexible`
136  *   Indicates the total number of flexible interrupt resources assignable to
137  *   all the secondary controllers. Implicitly sets the number of primary
138  *   controller's private resources to `(msix_qsize - sriov_vi_flexible)`.
139  *
140  * - `sriov_max_vi_per_vf`
141  *   Indicates the maximum number of virtual interrupt resources assignable
142  *   to a secondary controller. The default 0 resolves to
143  *   `(sriov_vi_flexible / sriov_max_vfs)`.
144  *
145  * - `sriov_max_vq_per_vf`
146  *   Indicates the maximum number of virtual queue resources assignable to
147  *   a secondary controller. The default 0 resolves to
148  *   `(sriov_vq_flexible / sriov_max_vfs)`.
149  *
150  * nvme namespace device parameters
151  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
152  * - `shared`
153  *   When the parent nvme device (as defined explicitly by the 'bus' parameter
154  *   or implicitly by the most recently defined NvmeBus) is linked to an
155  *   nvme-subsys device, the namespace will be attached to all controllers in
156  *   the subsystem. If set to 'off' (the default), the namespace will remain a
157  *   private namespace and may only be attached to a single controller at a
158  *   time.
159  *
160  * - `detached`
161  *   This parameter is only valid together with the `subsys` parameter. If left
162  *   at the default value (`false/off`), the namespace will be attached to all
163  *   controllers in the NVMe subsystem at boot-up. If set to `true/on`, the
164  *   namespace will be available in the subsystem but not attached to any
165  *   controllers.
166  *
167  * Setting `zoned` to true selects Zoned Command Set at the namespace.
168  * In this case, the following namespace properties are available to configure
169  * zoned operation:
170  *     zoned.zone_size=<zone size in bytes, default: 128MiB>
171  *         The number may be followed by K, M, G as in kilo-, mega- or giga-.
172  *
173  *     zoned.zone_capacity=<zone capacity in bytes, default: zone size>
174  *         The value 0 (default) forces zone capacity to be the same as zone
175  *         size. The value of this property may not exceed zone size.
176  *
177  *     zoned.descr_ext_size=<zone descriptor extension size, default 0>
178  *         This value needs to be specified in 64B units. If it is zero,
179  *         namespace(s) will not support zone descriptor extensions.
180  *
181  *     zoned.max_active=<Maximum Active Resources (zones), default: 0>
182  *         The default value means there is no limit to the number of
183  *         concurrently active zones.
184  *
185  *     zoned.max_open=<Maximum Open Resources (zones), default: 0>
186  *         The default value means there is no limit to the number of
187  *         concurrently open zones.
188  *
189  *     zoned.cross_read=<enable RAZB, default: false>
190  *         Setting this property to true enables Read Across Zone Boundaries.
191  */
192 
193 #include "qemu/osdep.h"
194 #include "qemu/cutils.h"
195 #include "qemu/error-report.h"
196 #include "qemu/log.h"
197 #include "qemu/units.h"
198 #include "qemu/range.h"
199 #include "qapi/error.h"
200 #include "qapi/visitor.h"
201 #include "sysemu/sysemu.h"
202 #include "sysemu/block-backend.h"
203 #include "sysemu/hostmem.h"
204 #include "hw/pci/msix.h"
205 #include "hw/pci/pcie_sriov.h"
206 #include "migration/vmstate.h"
207 
208 #include "nvme.h"
209 #include "dif.h"
210 #include "trace.h"
211 
212 #define NVME_MAX_IOQPAIRS 0xffff
213 #define NVME_DB_SIZE  4
214 #define NVME_SPEC_VER 0x00010400
215 #define NVME_CMB_BIR 2
216 #define NVME_PMR_BIR 4
217 #define NVME_TEMPERATURE 0x143
218 #define NVME_TEMPERATURE_WARNING 0x157
219 #define NVME_TEMPERATURE_CRITICAL 0x175
220 #define NVME_NUM_FW_SLOTS 1
221 #define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
222 #define NVME_MAX_VFS 127
223 #define NVME_VF_RES_GRANULARITY 1
224 #define NVME_VF_OFFSET 0x1
225 #define NVME_VF_STRIDE 1
226 
227 #define NVME_GUEST_ERR(trace, fmt, ...) \
228     do { \
229         (trace_##trace)(__VA_ARGS__); \
230         qemu_log_mask(LOG_GUEST_ERROR, #trace \
231             " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
232     } while (0)
233 
234 static const bool nvme_feature_support[NVME_FID_MAX] = {
235     [NVME_ARBITRATION]              = true,
236     [NVME_POWER_MANAGEMENT]         = true,
237     [NVME_TEMPERATURE_THRESHOLD]    = true,
238     [NVME_ERROR_RECOVERY]           = true,
239     [NVME_VOLATILE_WRITE_CACHE]     = true,
240     [NVME_NUMBER_OF_QUEUES]         = true,
241     [NVME_INTERRUPT_COALESCING]     = true,
242     [NVME_INTERRUPT_VECTOR_CONF]    = true,
243     [NVME_WRITE_ATOMICITY]          = true,
244     [NVME_ASYNCHRONOUS_EVENT_CONF]  = true,
245     [NVME_TIMESTAMP]                = true,
246     [NVME_HOST_BEHAVIOR_SUPPORT]    = true,
247     [NVME_COMMAND_SET_PROFILE]      = true,
248     [NVME_FDP_MODE]                 = true,
249     [NVME_FDP_EVENTS]               = true,
250 };
251 
252 static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
253     [NVME_TEMPERATURE_THRESHOLD]    = NVME_FEAT_CAP_CHANGE,
254     [NVME_ERROR_RECOVERY]           = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
255     [NVME_VOLATILE_WRITE_CACHE]     = NVME_FEAT_CAP_CHANGE,
256     [NVME_NUMBER_OF_QUEUES]         = NVME_FEAT_CAP_CHANGE,
257     [NVME_ASYNCHRONOUS_EVENT_CONF]  = NVME_FEAT_CAP_CHANGE,
258     [NVME_TIMESTAMP]                = NVME_FEAT_CAP_CHANGE,
259     [NVME_HOST_BEHAVIOR_SUPPORT]    = NVME_FEAT_CAP_CHANGE,
260     [NVME_COMMAND_SET_PROFILE]      = NVME_FEAT_CAP_CHANGE,
261     [NVME_FDP_MODE]                 = NVME_FEAT_CAP_CHANGE,
262     [NVME_FDP_EVENTS]               = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
263 };
264 
265 static const uint32_t nvme_cse_acs[256] = {
266     [NVME_ADM_CMD_DELETE_SQ]        = NVME_CMD_EFF_CSUPP,
267     [NVME_ADM_CMD_CREATE_SQ]        = NVME_CMD_EFF_CSUPP,
268     [NVME_ADM_CMD_GET_LOG_PAGE]     = NVME_CMD_EFF_CSUPP,
269     [NVME_ADM_CMD_DELETE_CQ]        = NVME_CMD_EFF_CSUPP,
270     [NVME_ADM_CMD_CREATE_CQ]        = NVME_CMD_EFF_CSUPP,
271     [NVME_ADM_CMD_IDENTIFY]         = NVME_CMD_EFF_CSUPP,
272     [NVME_ADM_CMD_ABORT]            = NVME_CMD_EFF_CSUPP,
273     [NVME_ADM_CMD_SET_FEATURES]     = NVME_CMD_EFF_CSUPP,
274     [NVME_ADM_CMD_GET_FEATURES]     = NVME_CMD_EFF_CSUPP,
275     [NVME_ADM_CMD_ASYNC_EV_REQ]     = NVME_CMD_EFF_CSUPP,
276     [NVME_ADM_CMD_NS_ATTACHMENT]    = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_NIC,
277     [NVME_ADM_CMD_VIRT_MNGMT]       = NVME_CMD_EFF_CSUPP,
278     [NVME_ADM_CMD_DBBUF_CONFIG]     = NVME_CMD_EFF_CSUPP,
279     [NVME_ADM_CMD_FORMAT_NVM]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
280     [NVME_ADM_CMD_DIRECTIVE_RECV]   = NVME_CMD_EFF_CSUPP,
281     [NVME_ADM_CMD_DIRECTIVE_SEND]   = NVME_CMD_EFF_CSUPP,
282 };
283 
284 static const uint32_t nvme_cse_iocs_none[256];
285 
286 static const uint32_t nvme_cse_iocs_nvm[256] = {
287     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
288     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
289     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
290     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
291     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
292     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
293     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
294     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
295     [NVME_CMD_IO_MGMT_RECV]         = NVME_CMD_EFF_CSUPP,
296     [NVME_CMD_IO_MGMT_SEND]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
297 };
298 
299 static const uint32_t nvme_cse_iocs_zoned[256] = {
300     [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
301     [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
302     [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
303     [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
304     [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
305     [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
306     [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
307     [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
308     [NVME_CMD_ZONE_APPEND]          = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
309     [NVME_CMD_ZONE_MGMT_SEND]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
310     [NVME_CMD_ZONE_MGMT_RECV]       = NVME_CMD_EFF_CSUPP,
311 };
312 
313 static void nvme_process_sq(void *opaque);
314 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst);
315 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n);
316 
317 static uint16_t nvme_sqid(NvmeRequest *req)
318 {
319     return le16_to_cpu(req->sq->sqid);
320 }
321 
322 static inline uint16_t nvme_make_pid(NvmeNamespace *ns, uint16_t rg,
323                                      uint16_t ph)
324 {
325     uint16_t rgif = ns->endgrp->fdp.rgif;
326 
327     if (!rgif) {
328         return ph;
329     }
330 
331     return (rg << (16 - rgif)) | ph;
332 }
333 
334 static inline bool nvme_ph_valid(NvmeNamespace *ns, uint16_t ph)
335 {
336     return ph < ns->fdp.nphs;
337 }
338 
339 static inline bool nvme_rg_valid(NvmeEnduranceGroup *endgrp, uint16_t rg)
340 {
341     return rg < endgrp->fdp.nrg;
342 }
343 
344 static inline uint16_t nvme_pid2ph(NvmeNamespace *ns, uint16_t pid)
345 {
346     uint16_t rgif = ns->endgrp->fdp.rgif;
347 
348     if (!rgif) {
349         return pid;
350     }
351 
352     return pid & ((1 << (15 - rgif)) - 1);
353 }
354 
355 static inline uint16_t nvme_pid2rg(NvmeNamespace *ns, uint16_t pid)
356 {
357     uint16_t rgif = ns->endgrp->fdp.rgif;
358 
359     if (!rgif) {
360         return 0;
361     }
362 
363     return pid >> (16 - rgif);
364 }
365 
366 static inline bool nvme_parse_pid(NvmeNamespace *ns, uint16_t pid,
367                                   uint16_t *ph, uint16_t *rg)
368 {
369     *rg = nvme_pid2rg(ns, pid);
370     *ph = nvme_pid2ph(ns, pid);
371 
372     return nvme_ph_valid(ns, *ph) && nvme_rg_valid(ns->endgrp, *rg);
373 }
374 
375 static void nvme_assign_zone_state(NvmeNamespace *ns, NvmeZone *zone,
376                                    NvmeZoneState state)
377 {
378     if (QTAILQ_IN_USE(zone, entry)) {
379         switch (nvme_get_zone_state(zone)) {
380         case NVME_ZONE_STATE_EXPLICITLY_OPEN:
381             QTAILQ_REMOVE(&ns->exp_open_zones, zone, entry);
382             break;
383         case NVME_ZONE_STATE_IMPLICITLY_OPEN:
384             QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
385             break;
386         case NVME_ZONE_STATE_CLOSED:
387             QTAILQ_REMOVE(&ns->closed_zones, zone, entry);
388             break;
389         case NVME_ZONE_STATE_FULL:
390             QTAILQ_REMOVE(&ns->full_zones, zone, entry);
391         default:
392             ;
393         }
394     }
395 
396     nvme_set_zone_state(zone, state);
397 
398     switch (state) {
399     case NVME_ZONE_STATE_EXPLICITLY_OPEN:
400         QTAILQ_INSERT_TAIL(&ns->exp_open_zones, zone, entry);
401         break;
402     case NVME_ZONE_STATE_IMPLICITLY_OPEN:
403         QTAILQ_INSERT_TAIL(&ns->imp_open_zones, zone, entry);
404         break;
405     case NVME_ZONE_STATE_CLOSED:
406         QTAILQ_INSERT_TAIL(&ns->closed_zones, zone, entry);
407         break;
408     case NVME_ZONE_STATE_FULL:
409         QTAILQ_INSERT_TAIL(&ns->full_zones, zone, entry);
410     case NVME_ZONE_STATE_READ_ONLY:
411         break;
412     default:
413         zone->d.za = 0;
414     }
415 }
416 
417 static uint16_t nvme_zns_check_resources(NvmeNamespace *ns, uint32_t act,
418                                          uint32_t opn, uint32_t zrwa)
419 {
420     if (ns->params.max_active_zones != 0 &&
421         ns->nr_active_zones + act > ns->params.max_active_zones) {
422         trace_pci_nvme_err_insuff_active_res(ns->params.max_active_zones);
423         return NVME_ZONE_TOO_MANY_ACTIVE | NVME_DNR;
424     }
425 
426     if (ns->params.max_open_zones != 0 &&
427         ns->nr_open_zones + opn > ns->params.max_open_zones) {
428         trace_pci_nvme_err_insuff_open_res(ns->params.max_open_zones);
429         return NVME_ZONE_TOO_MANY_OPEN | NVME_DNR;
430     }
431 
432     if (zrwa > ns->zns.numzrwa) {
433         return NVME_NOZRWA | NVME_DNR;
434     }
435 
436     return NVME_SUCCESS;
437 }
438 
439 /*
440  * Check if we can open a zone without exceeding open/active limits.
441  * AOR stands for "Active and Open Resources" (see TP 4053 section 2.5).
442  */
443 static uint16_t nvme_aor_check(NvmeNamespace *ns, uint32_t act, uint32_t opn)
444 {
445     return nvme_zns_check_resources(ns, act, opn, 0);
446 }
447 
448 static NvmeFdpEvent *nvme_fdp_alloc_event(NvmeCtrl *n, NvmeFdpEventBuffer *ebuf)
449 {
450     NvmeFdpEvent *ret = NULL;
451     bool is_full = ebuf->next == ebuf->start && ebuf->nelems;
452 
453     ret = &ebuf->events[ebuf->next++];
454     if (unlikely(ebuf->next == NVME_FDP_MAX_EVENTS)) {
455         ebuf->next = 0;
456     }
457     if (is_full) {
458         ebuf->start = ebuf->next;
459     } else {
460         ebuf->nelems++;
461     }
462 
463     memset(ret, 0, sizeof(NvmeFdpEvent));
464     ret->timestamp = nvme_get_timestamp(n);
465 
466     return ret;
467 }
468 
469 static inline int log_event(NvmeRuHandle *ruh, uint8_t event_type)
470 {
471     return (ruh->event_filter >> nvme_fdp_evf_shifts[event_type]) & 0x1;
472 }
473 
474 static bool nvme_update_ruh(NvmeCtrl *n, NvmeNamespace *ns, uint16_t pid)
475 {
476     NvmeEnduranceGroup *endgrp = ns->endgrp;
477     NvmeRuHandle *ruh;
478     NvmeReclaimUnit *ru;
479     NvmeFdpEvent *e = NULL;
480     uint16_t ph, rg, ruhid;
481 
482     if (!nvme_parse_pid(ns, pid, &ph, &rg)) {
483         return false;
484     }
485 
486     ruhid = ns->fdp.phs[ph];
487 
488     ruh = &endgrp->fdp.ruhs[ruhid];
489     ru = &ruh->rus[rg];
490 
491     if (ru->ruamw) {
492         if (log_event(ruh, FDP_EVT_RU_NOT_FULLY_WRITTEN)) {
493             e = nvme_fdp_alloc_event(n, &endgrp->fdp.host_events);
494             e->type = FDP_EVT_RU_NOT_FULLY_WRITTEN;
495             e->flags = FDPEF_PIV | FDPEF_NSIDV | FDPEF_LV;
496             e->pid = cpu_to_le16(pid);
497             e->nsid = cpu_to_le32(ns->params.nsid);
498             e->rgid = cpu_to_le16(rg);
499             e->ruhid = cpu_to_le16(ruhid);
500         }
501 
502         /* log (eventual) GC overhead of prematurely swapping the RU */
503         nvme_fdp_stat_inc(&endgrp->fdp.mbmw, nvme_l2b(ns, ru->ruamw));
504     }
505 
506     ru->ruamw = ruh->ruamw;
507 
508     return true;
509 }
510 
511 static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
512 {
513     hwaddr hi, lo;
514 
515     if (!n->cmb.cmse) {
516         return false;
517     }
518 
519     lo = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
520     hi = lo + int128_get64(n->cmb.mem.size);
521 
522     return addr >= lo && addr < hi;
523 }
524 
525 static inline void *nvme_addr_to_cmb(NvmeCtrl *n, hwaddr addr)
526 {
527     hwaddr base = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
528     return &n->cmb.buf[addr - base];
529 }
530 
531 static bool nvme_addr_is_pmr(NvmeCtrl *n, hwaddr addr)
532 {
533     hwaddr hi;
534 
535     if (!n->pmr.cmse) {
536         return false;
537     }
538 
539     hi = n->pmr.cba + int128_get64(n->pmr.dev->mr.size);
540 
541     return addr >= n->pmr.cba && addr < hi;
542 }
543 
544 static inline void *nvme_addr_to_pmr(NvmeCtrl *n, hwaddr addr)
545 {
546     return memory_region_get_ram_ptr(&n->pmr.dev->mr) + (addr - n->pmr.cba);
547 }
548 
549 static inline bool nvme_addr_is_iomem(NvmeCtrl *n, hwaddr addr)
550 {
551     hwaddr hi, lo;
552 
553     /*
554      * The purpose of this check is to guard against invalid "local" access to
555      * the iomem (i.e. controller registers). Thus, we check against the range
556      * covered by the 'bar0' MemoryRegion since that is currently composed of
557      * two subregions (the NVMe "MBAR" and the MSI-X table/pba). Note, however,
558      * that if the device model is ever changed to allow the CMB to be located
559      * in BAR0 as well, then this must be changed.
560      */
561     lo = n->bar0.addr;
562     hi = lo + int128_get64(n->bar0.size);
563 
564     return addr >= lo && addr < hi;
565 }
566 
567 static int nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
568 {
569     hwaddr hi = addr + size - 1;
570     if (hi < addr) {
571         return 1;
572     }
573 
574     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
575         memcpy(buf, nvme_addr_to_cmb(n, addr), size);
576         return 0;
577     }
578 
579     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
580         memcpy(buf, nvme_addr_to_pmr(n, addr), size);
581         return 0;
582     }
583 
584     return pci_dma_read(PCI_DEVICE(n), addr, buf, size);
585 }
586 
587 static int nvme_addr_write(NvmeCtrl *n, hwaddr addr, const void *buf, int size)
588 {
589     hwaddr hi = addr + size - 1;
590     if (hi < addr) {
591         return 1;
592     }
593 
594     if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
595         memcpy(nvme_addr_to_cmb(n, addr), buf, size);
596         return 0;
597     }
598 
599     if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
600         memcpy(nvme_addr_to_pmr(n, addr), buf, size);
601         return 0;
602     }
603 
604     return pci_dma_write(PCI_DEVICE(n), addr, buf, size);
605 }
606 
607 static bool nvme_nsid_valid(NvmeCtrl *n, uint32_t nsid)
608 {
609     return nsid &&
610         (nsid == NVME_NSID_BROADCAST || nsid <= NVME_MAX_NAMESPACES);
611 }
612 
613 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
614 {
615     return sqid < n->conf_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
616 }
617 
618 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
619 {
620     return cqid < n->conf_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
621 }
622 
623 static void nvme_inc_cq_tail(NvmeCQueue *cq)
624 {
625     cq->tail++;
626     if (cq->tail >= cq->size) {
627         cq->tail = 0;
628         cq->phase = !cq->phase;
629     }
630 }
631 
632 static void nvme_inc_sq_head(NvmeSQueue *sq)
633 {
634     sq->head = (sq->head + 1) % sq->size;
635 }
636 
637 static uint8_t nvme_cq_full(NvmeCQueue *cq)
638 {
639     return (cq->tail + 1) % cq->size == cq->head;
640 }
641 
642 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
643 {
644     return sq->head == sq->tail;
645 }
646 
647 static void nvme_irq_check(NvmeCtrl *n)
648 {
649     PCIDevice *pci = PCI_DEVICE(n);
650     uint32_t intms = ldl_le_p(&n->bar.intms);
651 
652     if (msix_enabled(pci)) {
653         return;
654     }
655     if (~intms & n->irq_status) {
656         pci_irq_assert(pci);
657     } else {
658         pci_irq_deassert(pci);
659     }
660 }
661 
662 static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
663 {
664     PCIDevice *pci = PCI_DEVICE(n);
665 
666     if (cq->irq_enabled) {
667         if (msix_enabled(pci)) {
668             trace_pci_nvme_irq_msix(cq->vector);
669             msix_notify(pci, cq->vector);
670         } else {
671             trace_pci_nvme_irq_pin();
672             assert(cq->vector < 32);
673             n->irq_status |= 1 << cq->vector;
674             nvme_irq_check(n);
675         }
676     } else {
677         trace_pci_nvme_irq_masked();
678     }
679 }
680 
681 static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
682 {
683     if (cq->irq_enabled) {
684         if (msix_enabled(PCI_DEVICE(n))) {
685             return;
686         } else {
687             assert(cq->vector < 32);
688             if (!n->cq_pending) {
689                 n->irq_status &= ~(1 << cq->vector);
690             }
691             nvme_irq_check(n);
692         }
693     }
694 }
695 
696 static void nvme_req_clear(NvmeRequest *req)
697 {
698     req->ns = NULL;
699     req->opaque = NULL;
700     req->aiocb = NULL;
701     memset(&req->cqe, 0x0, sizeof(req->cqe));
702     req->status = NVME_SUCCESS;
703 }
704 
705 static inline void nvme_sg_init(NvmeCtrl *n, NvmeSg *sg, bool dma)
706 {
707     if (dma) {
708         pci_dma_sglist_init(&sg->qsg, PCI_DEVICE(n), 0);
709         sg->flags = NVME_SG_DMA;
710     } else {
711         qemu_iovec_init(&sg->iov, 0);
712     }
713 
714     sg->flags |= NVME_SG_ALLOC;
715 }
716 
717 static inline void nvme_sg_unmap(NvmeSg *sg)
718 {
719     if (!(sg->flags & NVME_SG_ALLOC)) {
720         return;
721     }
722 
723     if (sg->flags & NVME_SG_DMA) {
724         qemu_sglist_destroy(&sg->qsg);
725     } else {
726         qemu_iovec_destroy(&sg->iov);
727     }
728 
729     memset(sg, 0x0, sizeof(*sg));
730 }
731 
732 /*
733  * When metadata is transfered as extended LBAs, the DPTR mapped into `sg`
734  * holds both data and metadata. This function splits the data and metadata
735  * into two separate QSG/IOVs.
736  */
737 static void nvme_sg_split(NvmeSg *sg, NvmeNamespace *ns, NvmeSg *data,
738                           NvmeSg *mdata)
739 {
740     NvmeSg *dst = data;
741     uint32_t trans_len, count = ns->lbasz;
742     uint64_t offset = 0;
743     bool dma = sg->flags & NVME_SG_DMA;
744     size_t sge_len;
745     size_t sg_len = dma ? sg->qsg.size : sg->iov.size;
746     int sg_idx = 0;
747 
748     assert(sg->flags & NVME_SG_ALLOC);
749 
750     while (sg_len) {
751         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
752 
753         trans_len = MIN(sg_len, count);
754         trans_len = MIN(trans_len, sge_len - offset);
755 
756         if (dst) {
757             if (dma) {
758                 qemu_sglist_add(&dst->qsg, sg->qsg.sg[sg_idx].base + offset,
759                                 trans_len);
760             } else {
761                 qemu_iovec_add(&dst->iov,
762                                sg->iov.iov[sg_idx].iov_base + offset,
763                                trans_len);
764             }
765         }
766 
767         sg_len -= trans_len;
768         count -= trans_len;
769         offset += trans_len;
770 
771         if (count == 0) {
772             dst = (dst == data) ? mdata : data;
773             count = (dst == data) ? ns->lbasz : ns->lbaf.ms;
774         }
775 
776         if (sge_len == offset) {
777             offset = 0;
778             sg_idx++;
779         }
780     }
781 }
782 
783 static uint16_t nvme_map_addr_cmb(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
784                                   size_t len)
785 {
786     if (!len) {
787         return NVME_SUCCESS;
788     }
789 
790     trace_pci_nvme_map_addr_cmb(addr, len);
791 
792     if (!nvme_addr_is_cmb(n, addr) || !nvme_addr_is_cmb(n, addr + len - 1)) {
793         return NVME_DATA_TRAS_ERROR;
794     }
795 
796     qemu_iovec_add(iov, nvme_addr_to_cmb(n, addr), len);
797 
798     return NVME_SUCCESS;
799 }
800 
801 static uint16_t nvme_map_addr_pmr(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
802                                   size_t len)
803 {
804     if (!len) {
805         return NVME_SUCCESS;
806     }
807 
808     if (!nvme_addr_is_pmr(n, addr) || !nvme_addr_is_pmr(n, addr + len - 1)) {
809         return NVME_DATA_TRAS_ERROR;
810     }
811 
812     qemu_iovec_add(iov, nvme_addr_to_pmr(n, addr), len);
813 
814     return NVME_SUCCESS;
815 }
816 
817 static uint16_t nvme_map_addr(NvmeCtrl *n, NvmeSg *sg, hwaddr addr, size_t len)
818 {
819     bool cmb = false, pmr = false;
820 
821     if (!len) {
822         return NVME_SUCCESS;
823     }
824 
825     trace_pci_nvme_map_addr(addr, len);
826 
827     if (nvme_addr_is_iomem(n, addr)) {
828         return NVME_DATA_TRAS_ERROR;
829     }
830 
831     if (nvme_addr_is_cmb(n, addr)) {
832         cmb = true;
833     } else if (nvme_addr_is_pmr(n, addr)) {
834         pmr = true;
835     }
836 
837     if (cmb || pmr) {
838         if (sg->flags & NVME_SG_DMA) {
839             return NVME_INVALID_USE_OF_CMB | NVME_DNR;
840         }
841 
842         if (sg->iov.niov + 1 > IOV_MAX) {
843             goto max_mappings_exceeded;
844         }
845 
846         if (cmb) {
847             return nvme_map_addr_cmb(n, &sg->iov, addr, len);
848         } else {
849             return nvme_map_addr_pmr(n, &sg->iov, addr, len);
850         }
851     }
852 
853     if (!(sg->flags & NVME_SG_DMA)) {
854         return NVME_INVALID_USE_OF_CMB | NVME_DNR;
855     }
856 
857     if (sg->qsg.nsg + 1 > IOV_MAX) {
858         goto max_mappings_exceeded;
859     }
860 
861     qemu_sglist_add(&sg->qsg, addr, len);
862 
863     return NVME_SUCCESS;
864 
865 max_mappings_exceeded:
866     NVME_GUEST_ERR(pci_nvme_ub_too_many_mappings,
867                    "number of mappings exceed 1024");
868     return NVME_INTERNAL_DEV_ERROR | NVME_DNR;
869 }
870 
871 static inline bool nvme_addr_is_dma(NvmeCtrl *n, hwaddr addr)
872 {
873     return !(nvme_addr_is_cmb(n, addr) || nvme_addr_is_pmr(n, addr));
874 }
875 
876 static uint16_t nvme_map_prp(NvmeCtrl *n, NvmeSg *sg, uint64_t prp1,
877                              uint64_t prp2, uint32_t len)
878 {
879     hwaddr trans_len = n->page_size - (prp1 % n->page_size);
880     trans_len = MIN(len, trans_len);
881     int num_prps = (len >> n->page_bits) + 1;
882     uint16_t status;
883     int ret;
884 
885     trace_pci_nvme_map_prp(trans_len, len, prp1, prp2, num_prps);
886 
887     nvme_sg_init(n, sg, nvme_addr_is_dma(n, prp1));
888 
889     status = nvme_map_addr(n, sg, prp1, trans_len);
890     if (status) {
891         goto unmap;
892     }
893 
894     len -= trans_len;
895     if (len) {
896         if (len > n->page_size) {
897             uint64_t prp_list[n->max_prp_ents];
898             uint32_t nents, prp_trans;
899             int i = 0;
900 
901             /*
902              * The first PRP list entry, pointed to by PRP2 may contain offset.
903              * Hence, we need to calculate the number of entries in based on
904              * that offset.
905              */
906             nents = (n->page_size - (prp2 & (n->page_size - 1))) >> 3;
907             prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
908             ret = nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
909             if (ret) {
910                 trace_pci_nvme_err_addr_read(prp2);
911                 status = NVME_DATA_TRAS_ERROR;
912                 goto unmap;
913             }
914             while (len != 0) {
915                 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
916 
917                 if (i == nents - 1 && len > n->page_size) {
918                     if (unlikely(prp_ent & (n->page_size - 1))) {
919                         trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
920                         status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
921                         goto unmap;
922                     }
923 
924                     i = 0;
925                     nents = (len + n->page_size - 1) >> n->page_bits;
926                     nents = MIN(nents, n->max_prp_ents);
927                     prp_trans = nents * sizeof(uint64_t);
928                     ret = nvme_addr_read(n, prp_ent, (void *)prp_list,
929                                          prp_trans);
930                     if (ret) {
931                         trace_pci_nvme_err_addr_read(prp_ent);
932                         status = NVME_DATA_TRAS_ERROR;
933                         goto unmap;
934                     }
935                     prp_ent = le64_to_cpu(prp_list[i]);
936                 }
937 
938                 if (unlikely(prp_ent & (n->page_size - 1))) {
939                     trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
940                     status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
941                     goto unmap;
942                 }
943 
944                 trans_len = MIN(len, n->page_size);
945                 status = nvme_map_addr(n, sg, prp_ent, trans_len);
946                 if (status) {
947                     goto unmap;
948                 }
949 
950                 len -= trans_len;
951                 i++;
952             }
953         } else {
954             if (unlikely(prp2 & (n->page_size - 1))) {
955                 trace_pci_nvme_err_invalid_prp2_align(prp2);
956                 status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
957                 goto unmap;
958             }
959             status = nvme_map_addr(n, sg, prp2, len);
960             if (status) {
961                 goto unmap;
962             }
963         }
964     }
965 
966     return NVME_SUCCESS;
967 
968 unmap:
969     nvme_sg_unmap(sg);
970     return status;
971 }
972 
973 /*
974  * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
975  * number of bytes mapped in len.
976  */
977 static uint16_t nvme_map_sgl_data(NvmeCtrl *n, NvmeSg *sg,
978                                   NvmeSglDescriptor *segment, uint64_t nsgld,
979                                   size_t *len, NvmeCmd *cmd)
980 {
981     dma_addr_t addr, trans_len;
982     uint32_t dlen;
983     uint16_t status;
984 
985     for (int i = 0; i < nsgld; i++) {
986         uint8_t type = NVME_SGL_TYPE(segment[i].type);
987 
988         switch (type) {
989         case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
990             break;
991         case NVME_SGL_DESCR_TYPE_SEGMENT:
992         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
993             return NVME_INVALID_NUM_SGL_DESCRS | NVME_DNR;
994         default:
995             return NVME_SGL_DESCR_TYPE_INVALID | NVME_DNR;
996         }
997 
998         dlen = le32_to_cpu(segment[i].len);
999 
1000         if (!dlen) {
1001             continue;
1002         }
1003 
1004         if (*len == 0) {
1005             /*
1006              * All data has been mapped, but the SGL contains additional
1007              * segments and/or descriptors. The controller might accept
1008              * ignoring the rest of the SGL.
1009              */
1010             uint32_t sgls = le32_to_cpu(n->id_ctrl.sgls);
1011             if (sgls & NVME_CTRL_SGLS_EXCESS_LENGTH) {
1012                 break;
1013             }
1014 
1015             trace_pci_nvme_err_invalid_sgl_excess_length(dlen);
1016             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1017         }
1018 
1019         trans_len = MIN(*len, dlen);
1020 
1021         addr = le64_to_cpu(segment[i].addr);
1022 
1023         if (UINT64_MAX - addr < dlen) {
1024             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1025         }
1026 
1027         status = nvme_map_addr(n, sg, addr, trans_len);
1028         if (status) {
1029             return status;
1030         }
1031 
1032         *len -= trans_len;
1033     }
1034 
1035     return NVME_SUCCESS;
1036 }
1037 
1038 static uint16_t nvme_map_sgl(NvmeCtrl *n, NvmeSg *sg, NvmeSglDescriptor sgl,
1039                              size_t len, NvmeCmd *cmd)
1040 {
1041     /*
1042      * Read the segment in chunks of 256 descriptors (one 4k page) to avoid
1043      * dynamically allocating a potentially huge SGL. The spec allows the SGL
1044      * to be larger (as in number of bytes required to describe the SGL
1045      * descriptors and segment chain) than the command transfer size, so it is
1046      * not bounded by MDTS.
1047      */
1048     const int SEG_CHUNK_SIZE = 256;
1049 
1050     NvmeSglDescriptor segment[SEG_CHUNK_SIZE], *sgld, *last_sgld;
1051     uint64_t nsgld;
1052     uint32_t seg_len;
1053     uint16_t status;
1054     hwaddr addr;
1055     int ret;
1056 
1057     sgld = &sgl;
1058     addr = le64_to_cpu(sgl.addr);
1059 
1060     trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl.type), len);
1061 
1062     nvme_sg_init(n, sg, nvme_addr_is_dma(n, addr));
1063 
1064     /*
1065      * If the entire transfer can be described with a single data block it can
1066      * be mapped directly.
1067      */
1068     if (NVME_SGL_TYPE(sgl.type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1069         status = nvme_map_sgl_data(n, sg, sgld, 1, &len, cmd);
1070         if (status) {
1071             goto unmap;
1072         }
1073 
1074         goto out;
1075     }
1076 
1077     for (;;) {
1078         switch (NVME_SGL_TYPE(sgld->type)) {
1079         case NVME_SGL_DESCR_TYPE_SEGMENT:
1080         case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
1081             break;
1082         default:
1083             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1084         }
1085 
1086         seg_len = le32_to_cpu(sgld->len);
1087 
1088         /* check the length of the (Last) Segment descriptor */
1089         if (!seg_len || seg_len & 0xf) {
1090             return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1091         }
1092 
1093         if (UINT64_MAX - addr < seg_len) {
1094             return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1095         }
1096 
1097         nsgld = seg_len / sizeof(NvmeSglDescriptor);
1098 
1099         while (nsgld > SEG_CHUNK_SIZE) {
1100             if (nvme_addr_read(n, addr, segment, sizeof(segment))) {
1101                 trace_pci_nvme_err_addr_read(addr);
1102                 status = NVME_DATA_TRAS_ERROR;
1103                 goto unmap;
1104             }
1105 
1106             status = nvme_map_sgl_data(n, sg, segment, SEG_CHUNK_SIZE,
1107                                        &len, cmd);
1108             if (status) {
1109                 goto unmap;
1110             }
1111 
1112             nsgld -= SEG_CHUNK_SIZE;
1113             addr += SEG_CHUNK_SIZE * sizeof(NvmeSglDescriptor);
1114         }
1115 
1116         ret = nvme_addr_read(n, addr, segment, nsgld *
1117                              sizeof(NvmeSglDescriptor));
1118         if (ret) {
1119             trace_pci_nvme_err_addr_read(addr);
1120             status = NVME_DATA_TRAS_ERROR;
1121             goto unmap;
1122         }
1123 
1124         last_sgld = &segment[nsgld - 1];
1125 
1126         /*
1127          * If the segment ends with a Data Block, then we are done.
1128          */
1129         if (NVME_SGL_TYPE(last_sgld->type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
1130             status = nvme_map_sgl_data(n, sg, segment, nsgld, &len, cmd);
1131             if (status) {
1132                 goto unmap;
1133             }
1134 
1135             goto out;
1136         }
1137 
1138         /*
1139          * If the last descriptor was not a Data Block, then the current
1140          * segment must not be a Last Segment.
1141          */
1142         if (NVME_SGL_TYPE(sgld->type) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT) {
1143             status = NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
1144             goto unmap;
1145         }
1146 
1147         sgld = last_sgld;
1148         addr = le64_to_cpu(sgld->addr);
1149 
1150         /*
1151          * Do not map the last descriptor; it will be a Segment or Last Segment
1152          * descriptor and is handled by the next iteration.
1153          */
1154         status = nvme_map_sgl_data(n, sg, segment, nsgld - 1, &len, cmd);
1155         if (status) {
1156             goto unmap;
1157         }
1158     }
1159 
1160 out:
1161     /* if there is any residual left in len, the SGL was too short */
1162     if (len) {
1163         status = NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
1164         goto unmap;
1165     }
1166 
1167     return NVME_SUCCESS;
1168 
1169 unmap:
1170     nvme_sg_unmap(sg);
1171     return status;
1172 }
1173 
1174 uint16_t nvme_map_dptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1175                        NvmeCmd *cmd)
1176 {
1177     uint64_t prp1, prp2;
1178 
1179     switch (NVME_CMD_FLAGS_PSDT(cmd->flags)) {
1180     case NVME_PSDT_PRP:
1181         prp1 = le64_to_cpu(cmd->dptr.prp1);
1182         prp2 = le64_to_cpu(cmd->dptr.prp2);
1183 
1184         return nvme_map_prp(n, sg, prp1, prp2, len);
1185     case NVME_PSDT_SGL_MPTR_CONTIGUOUS:
1186     case NVME_PSDT_SGL_MPTR_SGL:
1187         return nvme_map_sgl(n, sg, cmd->dptr.sgl, len, cmd);
1188     default:
1189         return NVME_INVALID_FIELD;
1190     }
1191 }
1192 
1193 static uint16_t nvme_map_mptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
1194                               NvmeCmd *cmd)
1195 {
1196     int psdt = NVME_CMD_FLAGS_PSDT(cmd->flags);
1197     hwaddr mptr = le64_to_cpu(cmd->mptr);
1198     uint16_t status;
1199 
1200     if (psdt == NVME_PSDT_SGL_MPTR_SGL) {
1201         NvmeSglDescriptor sgl;
1202 
1203         if (nvme_addr_read(n, mptr, &sgl, sizeof(sgl))) {
1204             return NVME_DATA_TRAS_ERROR;
1205         }
1206 
1207         status = nvme_map_sgl(n, sg, sgl, len, cmd);
1208         if (status && (status & 0x7ff) == NVME_DATA_SGL_LEN_INVALID) {
1209             status = NVME_MD_SGL_LEN_INVALID | NVME_DNR;
1210         }
1211 
1212         return status;
1213     }
1214 
1215     nvme_sg_init(n, sg, nvme_addr_is_dma(n, mptr));
1216     status = nvme_map_addr(n, sg, mptr, len);
1217     if (status) {
1218         nvme_sg_unmap(sg);
1219     }
1220 
1221     return status;
1222 }
1223 
1224 static uint16_t nvme_map_data(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1225 {
1226     NvmeNamespace *ns = req->ns;
1227     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1228     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1229     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1230     size_t len = nvme_l2b(ns, nlb);
1231     uint16_t status;
1232 
1233     if (nvme_ns_ext(ns) &&
1234         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1235         NvmeSg sg;
1236 
1237         len += nvme_m2b(ns, nlb);
1238 
1239         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1240         if (status) {
1241             return status;
1242         }
1243 
1244         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1245         nvme_sg_split(&sg, ns, &req->sg, NULL);
1246         nvme_sg_unmap(&sg);
1247 
1248         return NVME_SUCCESS;
1249     }
1250 
1251     return nvme_map_dptr(n, &req->sg, len, &req->cmd);
1252 }
1253 
1254 static uint16_t nvme_map_mdata(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
1255 {
1256     NvmeNamespace *ns = req->ns;
1257     size_t len = nvme_m2b(ns, nlb);
1258     uint16_t status;
1259 
1260     if (nvme_ns_ext(ns)) {
1261         NvmeSg sg;
1262 
1263         len += nvme_l2b(ns, nlb);
1264 
1265         status = nvme_map_dptr(n, &sg, len, &req->cmd);
1266         if (status) {
1267             return status;
1268         }
1269 
1270         nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
1271         nvme_sg_split(&sg, ns, NULL, &req->sg);
1272         nvme_sg_unmap(&sg);
1273 
1274         return NVME_SUCCESS;
1275     }
1276 
1277     return nvme_map_mptr(n, &req->sg, len, &req->cmd);
1278 }
1279 
1280 static uint16_t nvme_tx_interleaved(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr,
1281                                     uint32_t len, uint32_t bytes,
1282                                     int32_t skip_bytes, int64_t offset,
1283                                     NvmeTxDirection dir)
1284 {
1285     hwaddr addr;
1286     uint32_t trans_len, count = bytes;
1287     bool dma = sg->flags & NVME_SG_DMA;
1288     int64_t sge_len;
1289     int sg_idx = 0;
1290     int ret;
1291 
1292     assert(sg->flags & NVME_SG_ALLOC);
1293 
1294     while (len) {
1295         sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;
1296 
1297         if (sge_len - offset < 0) {
1298             offset -= sge_len;
1299             sg_idx++;
1300             continue;
1301         }
1302 
1303         if (sge_len == offset) {
1304             offset = 0;
1305             sg_idx++;
1306             continue;
1307         }
1308 
1309         trans_len = MIN(len, count);
1310         trans_len = MIN(trans_len, sge_len - offset);
1311 
1312         if (dma) {
1313             addr = sg->qsg.sg[sg_idx].base + offset;
1314         } else {
1315             addr = (hwaddr)(uintptr_t)sg->iov.iov[sg_idx].iov_base + offset;
1316         }
1317 
1318         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1319             ret = nvme_addr_read(n, addr, ptr, trans_len);
1320         } else {
1321             ret = nvme_addr_write(n, addr, ptr, trans_len);
1322         }
1323 
1324         if (ret) {
1325             return NVME_DATA_TRAS_ERROR;
1326         }
1327 
1328         ptr += trans_len;
1329         len -= trans_len;
1330         count -= trans_len;
1331         offset += trans_len;
1332 
1333         if (count == 0) {
1334             count = bytes;
1335             offset += skip_bytes;
1336         }
1337     }
1338 
1339     return NVME_SUCCESS;
1340 }
1341 
1342 static uint16_t nvme_tx(NvmeCtrl *n, NvmeSg *sg, void *ptr, uint32_t len,
1343                         NvmeTxDirection dir)
1344 {
1345     assert(sg->flags & NVME_SG_ALLOC);
1346 
1347     if (sg->flags & NVME_SG_DMA) {
1348         const MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
1349         dma_addr_t residual;
1350 
1351         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1352             dma_buf_write(ptr, len, &residual, &sg->qsg, attrs);
1353         } else {
1354             dma_buf_read(ptr, len, &residual, &sg->qsg, attrs);
1355         }
1356 
1357         if (unlikely(residual)) {
1358             trace_pci_nvme_err_invalid_dma();
1359             return NVME_INVALID_FIELD | NVME_DNR;
1360         }
1361     } else {
1362         size_t bytes;
1363 
1364         if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
1365             bytes = qemu_iovec_to_buf(&sg->iov, 0, ptr, len);
1366         } else {
1367             bytes = qemu_iovec_from_buf(&sg->iov, 0, ptr, len);
1368         }
1369 
1370         if (unlikely(bytes != len)) {
1371             trace_pci_nvme_err_invalid_dma();
1372             return NVME_INVALID_FIELD | NVME_DNR;
1373         }
1374     }
1375 
1376     return NVME_SUCCESS;
1377 }
1378 
1379 static inline uint16_t nvme_c2h(NvmeCtrl *n, void *ptr, uint32_t len,
1380                                 NvmeRequest *req)
1381 {
1382     uint16_t status;
1383 
1384     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1385     if (status) {
1386         return status;
1387     }
1388 
1389     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_FROM_DEVICE);
1390 }
1391 
1392 static inline uint16_t nvme_h2c(NvmeCtrl *n, void *ptr, uint32_t len,
1393                                 NvmeRequest *req)
1394 {
1395     uint16_t status;
1396 
1397     status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
1398     if (status) {
1399         return status;
1400     }
1401 
1402     return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_TO_DEVICE);
1403 }
1404 
1405 uint16_t nvme_bounce_data(NvmeCtrl *n, void *ptr, uint32_t len,
1406                           NvmeTxDirection dir, NvmeRequest *req)
1407 {
1408     NvmeNamespace *ns = req->ns;
1409     NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
1410     bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
1411     bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
1412 
1413     if (nvme_ns_ext(ns) &&
1414         !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
1415         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbasz,
1416                                    ns->lbaf.ms, 0, dir);
1417     }
1418 
1419     return nvme_tx(n, &req->sg, ptr, len, dir);
1420 }
1421 
1422 uint16_t nvme_bounce_mdata(NvmeCtrl *n, void *ptr, uint32_t len,
1423                            NvmeTxDirection dir, NvmeRequest *req)
1424 {
1425     NvmeNamespace *ns = req->ns;
1426     uint16_t status;
1427 
1428     if (nvme_ns_ext(ns)) {
1429         return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbaf.ms,
1430                                    ns->lbasz, ns->lbasz, dir);
1431     }
1432 
1433     nvme_sg_unmap(&req->sg);
1434 
1435     status = nvme_map_mptr(n, &req->sg, len, &req->cmd);
1436     if (status) {
1437         return status;
1438     }
1439 
1440     return nvme_tx(n, &req->sg, ptr, len, dir);
1441 }
1442 
1443 static inline void nvme_blk_read(BlockBackend *blk, int64_t offset,
1444                                  uint32_t align, BlockCompletionFunc *cb,
1445                                  NvmeRequest *req)
1446 {
1447     assert(req->sg.flags & NVME_SG_ALLOC);
1448 
1449     if (req->sg.flags & NVME_SG_DMA) {
1450         req->aiocb = dma_blk_read(blk, &req->sg.qsg, offset, align, cb, req);
1451     } else {
1452         req->aiocb = blk_aio_preadv(blk, offset, &req->sg.iov, 0, cb, req);
1453     }
1454 }
1455 
1456 static inline void nvme_blk_write(BlockBackend *blk, int64_t offset,
1457                                   uint32_t align, BlockCompletionFunc *cb,
1458                                   NvmeRequest *req)
1459 {
1460     assert(req->sg.flags & NVME_SG_ALLOC);
1461 
1462     if (req->sg.flags & NVME_SG_DMA) {
1463         req->aiocb = dma_blk_write(blk, &req->sg.qsg, offset, align, cb, req);
1464     } else {
1465         req->aiocb = blk_aio_pwritev(blk, offset, &req->sg.iov, 0, cb, req);
1466     }
1467 }
1468 
1469 static void nvme_update_cq_eventidx(const NvmeCQueue *cq)
1470 {
1471     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 * n->cqe_size;
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 = n->subsys->endgrp.fdp.nrg * n->subsys->endgrp.fdp.nruh;
4365 
4366     if (unlikely(npid >= MIN(NVME_FDP_MAXPIDS, maxnpid))) {
4367         return NVME_INVALID_FIELD | NVME_DNR;
4368     }
4369 
4370     pids = g_new(uint16_t, npid);
4371 
4372     ret = nvme_h2c(n, pids, npid * sizeof(uint16_t), req);
4373     if (ret) {
4374         return ret;
4375     }
4376 
4377     for (; i < npid; i++) {
4378         if (!nvme_update_ruh(n, ns, pids[i])) {
4379             return NVME_INVALID_FIELD | NVME_DNR;
4380         }
4381     }
4382 
4383     return ret;
4384 }
4385 
4386 static uint16_t nvme_io_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
4387 {
4388     NvmeCmd *cmd = &req->cmd;
4389     uint32_t cdw10 = le32_to_cpu(cmd->cdw10);
4390     uint8_t mo = (cdw10 & 0xff);
4391 
4392     switch (mo) {
4393     case NVME_IOMS_MO_NOP:
4394         return 0;
4395     case NVME_IOMS_MO_RUH_UPDATE:
4396         return nvme_io_mgmt_send_ruh_update(n, req);
4397     default:
4398         return NVME_INVALID_FIELD | NVME_DNR;
4399     };
4400 }
4401 
4402 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
4403 {
4404     NvmeNamespace *ns;
4405     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4406 
4407     trace_pci_nvme_io_cmd(nvme_cid(req), nsid, nvme_sqid(req),
4408                           req->cmd.opcode, nvme_io_opc_str(req->cmd.opcode));
4409 
4410     if (!nvme_nsid_valid(n, nsid)) {
4411         return NVME_INVALID_NSID | NVME_DNR;
4412     }
4413 
4414     /*
4415      * In the base NVM command set, Flush may apply to all namespaces
4416      * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4417      * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4418      *
4419      * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4420      * opcode with a specific command since we cannot determine a unique I/O
4421      * command set. Opcode 0h could have any other meaning than something
4422      * equivalent to flushing and say it DOES have completely different
4423      * semantics in some other command set - does an NSID of FFFFFFFFh then
4424      * mean "for all namespaces, apply whatever command set specific command
4425      * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4426      * whatever command that uses the 0h opcode if, and only if, it allows NSID
4427      * to be FFFFFFFFh"?
4428      *
4429      * Anyway (and luckily), for now, we do not care about this since the
4430      * device only supports namespace types that includes the NVM Flush command
4431      * (NVM and Zoned), so always do an NVM Flush.
4432      */
4433     if (req->cmd.opcode == NVME_CMD_FLUSH) {
4434         return nvme_flush(n, req);
4435     }
4436 
4437     ns = nvme_ns(n, nsid);
4438     if (unlikely(!ns)) {
4439         return NVME_INVALID_FIELD | NVME_DNR;
4440     }
4441 
4442     if (!(ns->iocs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
4443         trace_pci_nvme_err_invalid_opc(req->cmd.opcode);
4444         return NVME_INVALID_OPCODE | NVME_DNR;
4445     }
4446 
4447     if (ns->status) {
4448         return ns->status;
4449     }
4450 
4451     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
4452         return NVME_INVALID_FIELD;
4453     }
4454 
4455     req->ns = ns;
4456 
4457     switch (req->cmd.opcode) {
4458     case NVME_CMD_WRITE_ZEROES:
4459         return nvme_write_zeroes(n, req);
4460     case NVME_CMD_ZONE_APPEND:
4461         return nvme_zone_append(n, req);
4462     case NVME_CMD_WRITE:
4463         return nvme_write(n, req);
4464     case NVME_CMD_READ:
4465         return nvme_read(n, req);
4466     case NVME_CMD_COMPARE:
4467         return nvme_compare(n, req);
4468     case NVME_CMD_DSM:
4469         return nvme_dsm(n, req);
4470     case NVME_CMD_VERIFY:
4471         return nvme_verify(n, req);
4472     case NVME_CMD_COPY:
4473         return nvme_copy(n, req);
4474     case NVME_CMD_ZONE_MGMT_SEND:
4475         return nvme_zone_mgmt_send(n, req);
4476     case NVME_CMD_ZONE_MGMT_RECV:
4477         return nvme_zone_mgmt_recv(n, req);
4478     case NVME_CMD_IO_MGMT_RECV:
4479         return nvme_io_mgmt_recv(n, req);
4480     case NVME_CMD_IO_MGMT_SEND:
4481         return nvme_io_mgmt_send(n, req);
4482     default:
4483         assert(false);
4484     }
4485 
4486     return NVME_INVALID_OPCODE | NVME_DNR;
4487 }
4488 
4489 static void nvme_cq_notifier(EventNotifier *e)
4490 {
4491     NvmeCQueue *cq = container_of(e, NvmeCQueue, notifier);
4492     NvmeCtrl *n = cq->ctrl;
4493 
4494     if (!event_notifier_test_and_clear(e)) {
4495         return;
4496     }
4497 
4498     nvme_update_cq_head(cq);
4499 
4500     if (cq->tail == cq->head) {
4501         if (cq->irq_enabled) {
4502             n->cq_pending--;
4503         }
4504 
4505         nvme_irq_deassert(n, cq);
4506     }
4507 
4508     qemu_bh_schedule(cq->bh);
4509 }
4510 
4511 static int nvme_init_cq_ioeventfd(NvmeCQueue *cq)
4512 {
4513     NvmeCtrl *n = cq->ctrl;
4514     uint16_t offset = (cq->cqid << 3) + (1 << 2);
4515     int ret;
4516 
4517     ret = event_notifier_init(&cq->notifier, 0);
4518     if (ret < 0) {
4519         return ret;
4520     }
4521 
4522     event_notifier_set_handler(&cq->notifier, nvme_cq_notifier);
4523     memory_region_add_eventfd(&n->iomem,
4524                               0x1000 + offset, 4, false, 0, &cq->notifier);
4525 
4526     return 0;
4527 }
4528 
4529 static void nvme_sq_notifier(EventNotifier *e)
4530 {
4531     NvmeSQueue *sq = container_of(e, NvmeSQueue, notifier);
4532 
4533     if (!event_notifier_test_and_clear(e)) {
4534         return;
4535     }
4536 
4537     nvme_process_sq(sq);
4538 }
4539 
4540 static int nvme_init_sq_ioeventfd(NvmeSQueue *sq)
4541 {
4542     NvmeCtrl *n = sq->ctrl;
4543     uint16_t offset = sq->sqid << 3;
4544     int ret;
4545 
4546     ret = event_notifier_init(&sq->notifier, 0);
4547     if (ret < 0) {
4548         return ret;
4549     }
4550 
4551     event_notifier_set_handler(&sq->notifier, nvme_sq_notifier);
4552     memory_region_add_eventfd(&n->iomem,
4553                               0x1000 + offset, 4, false, 0, &sq->notifier);
4554 
4555     return 0;
4556 }
4557 
4558 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
4559 {
4560     uint16_t offset = sq->sqid << 3;
4561 
4562     n->sq[sq->sqid] = NULL;
4563     qemu_bh_delete(sq->bh);
4564     if (sq->ioeventfd_enabled) {
4565         memory_region_del_eventfd(&n->iomem,
4566                                   0x1000 + offset, 4, false, 0, &sq->notifier);
4567         event_notifier_set_handler(&sq->notifier, NULL);
4568         event_notifier_cleanup(&sq->notifier);
4569     }
4570     g_free(sq->io_req);
4571     if (sq->sqid) {
4572         g_free(sq);
4573     }
4574 }
4575 
4576 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
4577 {
4578     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
4579     NvmeRequest *r, *next;
4580     NvmeSQueue *sq;
4581     NvmeCQueue *cq;
4582     uint16_t qid = le16_to_cpu(c->qid);
4583 
4584     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
4585         trace_pci_nvme_err_invalid_del_sq(qid);
4586         return NVME_INVALID_QID | NVME_DNR;
4587     }
4588 
4589     trace_pci_nvme_del_sq(qid);
4590 
4591     sq = n->sq[qid];
4592     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
4593         r = QTAILQ_FIRST(&sq->out_req_list);
4594         assert(r->aiocb);
4595         blk_aio_cancel(r->aiocb);
4596     }
4597 
4598     assert(QTAILQ_EMPTY(&sq->out_req_list));
4599 
4600     if (!nvme_check_cqid(n, sq->cqid)) {
4601         cq = n->cq[sq->cqid];
4602         QTAILQ_REMOVE(&cq->sq_list, sq, entry);
4603 
4604         nvme_post_cqes(cq);
4605         QTAILQ_FOREACH_SAFE(r, &cq->req_list, entry, next) {
4606             if (r->sq == sq) {
4607                 QTAILQ_REMOVE(&cq->req_list, r, entry);
4608                 QTAILQ_INSERT_TAIL(&sq->req_list, r, entry);
4609             }
4610         }
4611     }
4612 
4613     nvme_free_sq(sq, n);
4614     return NVME_SUCCESS;
4615 }
4616 
4617 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
4618                          uint16_t sqid, uint16_t cqid, uint16_t size)
4619 {
4620     int i;
4621     NvmeCQueue *cq;
4622 
4623     sq->ctrl = n;
4624     sq->dma_addr = dma_addr;
4625     sq->sqid = sqid;
4626     sq->size = size;
4627     sq->cqid = cqid;
4628     sq->head = sq->tail = 0;
4629     sq->io_req = g_new0(NvmeRequest, sq->size);
4630 
4631     QTAILQ_INIT(&sq->req_list);
4632     QTAILQ_INIT(&sq->out_req_list);
4633     for (i = 0; i < sq->size; i++) {
4634         sq->io_req[i].sq = sq;
4635         QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
4636     }
4637 
4638     sq->bh = qemu_bh_new_guarded(nvme_process_sq, sq,
4639                                  &DEVICE(sq->ctrl)->mem_reentrancy_guard);
4640 
4641     if (n->dbbuf_enabled) {
4642         sq->db_addr = n->dbbuf_dbs + (sqid << 3);
4643         sq->ei_addr = n->dbbuf_eis + (sqid << 3);
4644 
4645         if (n->params.ioeventfd && sq->sqid != 0) {
4646             if (!nvme_init_sq_ioeventfd(sq)) {
4647                 sq->ioeventfd_enabled = true;
4648             }
4649         }
4650     }
4651 
4652     assert(n->cq[cqid]);
4653     cq = n->cq[cqid];
4654     QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
4655     n->sq[sqid] = sq;
4656 }
4657 
4658 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeRequest *req)
4659 {
4660     NvmeSQueue *sq;
4661     NvmeCreateSq *c = (NvmeCreateSq *)&req->cmd;
4662 
4663     uint16_t cqid = le16_to_cpu(c->cqid);
4664     uint16_t sqid = le16_to_cpu(c->sqid);
4665     uint16_t qsize = le16_to_cpu(c->qsize);
4666     uint16_t qflags = le16_to_cpu(c->sq_flags);
4667     uint64_t prp1 = le64_to_cpu(c->prp1);
4668 
4669     trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
4670 
4671     if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
4672         trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
4673         return NVME_INVALID_CQID | NVME_DNR;
4674     }
4675     if (unlikely(!sqid || sqid > n->conf_ioqpairs || n->sq[sqid] != NULL)) {
4676         trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
4677         return NVME_INVALID_QID | NVME_DNR;
4678     }
4679     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
4680         trace_pci_nvme_err_invalid_create_sq_size(qsize);
4681         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
4682     }
4683     if (unlikely(prp1 & (n->page_size - 1))) {
4684         trace_pci_nvme_err_invalid_create_sq_addr(prp1);
4685         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
4686     }
4687     if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
4688         trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
4689         return NVME_INVALID_FIELD | NVME_DNR;
4690     }
4691     sq = g_malloc0(sizeof(*sq));
4692     nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
4693     return NVME_SUCCESS;
4694 }
4695 
4696 struct nvme_stats {
4697     uint64_t units_read;
4698     uint64_t units_written;
4699     uint64_t read_commands;
4700     uint64_t write_commands;
4701 };
4702 
4703 static void nvme_set_blk_stats(NvmeNamespace *ns, struct nvme_stats *stats)
4704 {
4705     BlockAcctStats *s = blk_get_stats(ns->blkconf.blk);
4706 
4707     stats->units_read += s->nr_bytes[BLOCK_ACCT_READ];
4708     stats->units_written += s->nr_bytes[BLOCK_ACCT_WRITE];
4709     stats->read_commands += s->nr_ops[BLOCK_ACCT_READ];
4710     stats->write_commands += s->nr_ops[BLOCK_ACCT_WRITE];
4711 }
4712 
4713 static uint16_t nvme_smart_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4714                                 uint64_t off, NvmeRequest *req)
4715 {
4716     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
4717     struct nvme_stats stats = { 0 };
4718     NvmeSmartLog smart = { 0 };
4719     uint32_t trans_len;
4720     NvmeNamespace *ns;
4721     time_t current_ms;
4722     uint64_t u_read, u_written;
4723 
4724     if (off >= sizeof(smart)) {
4725         return NVME_INVALID_FIELD | NVME_DNR;
4726     }
4727 
4728     if (nsid != 0xffffffff) {
4729         ns = nvme_ns(n, nsid);
4730         if (!ns) {
4731             return NVME_INVALID_NSID | NVME_DNR;
4732         }
4733         nvme_set_blk_stats(ns, &stats);
4734     } else {
4735         int i;
4736 
4737         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4738             ns = nvme_ns(n, i);
4739             if (!ns) {
4740                 continue;
4741             }
4742             nvme_set_blk_stats(ns, &stats);
4743         }
4744     }
4745 
4746     trans_len = MIN(sizeof(smart) - off, buf_len);
4747     smart.critical_warning = n->smart_critical_warning;
4748 
4749     u_read = DIV_ROUND_UP(stats.units_read >> BDRV_SECTOR_BITS, 1000);
4750     u_written = DIV_ROUND_UP(stats.units_written >> BDRV_SECTOR_BITS, 1000);
4751 
4752     smart.data_units_read[0] = cpu_to_le64(u_read);
4753     smart.data_units_written[0] = cpu_to_le64(u_written);
4754     smart.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4755     smart.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4756 
4757     smart.temperature = cpu_to_le16(n->temperature);
4758 
4759     if ((n->temperature >= n->features.temp_thresh_hi) ||
4760         (n->temperature <= n->features.temp_thresh_low)) {
4761         smart.critical_warning |= NVME_SMART_TEMPERATURE;
4762     }
4763 
4764     current_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
4765     smart.power_on_hours[0] =
4766         cpu_to_le64((((current_ms - n->starttime_ms) / 1000) / 60) / 60);
4767 
4768     if (!rae) {
4769         nvme_clear_events(n, NVME_AER_TYPE_SMART);
4770     }
4771 
4772     return nvme_c2h(n, (uint8_t *) &smart + off, trans_len, req);
4773 }
4774 
4775 static uint16_t nvme_endgrp_info(NvmeCtrl *n,  uint8_t rae, uint32_t buf_len,
4776                                  uint64_t off, NvmeRequest *req)
4777 {
4778     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
4779     uint16_t endgrpid = (dw11 >> 16) & 0xffff;
4780     struct nvme_stats stats = {};
4781     NvmeEndGrpLog info = {};
4782     int i;
4783 
4784     if (!n->subsys || endgrpid != 0x1) {
4785         return NVME_INVALID_FIELD | NVME_DNR;
4786     }
4787 
4788     if (off >= sizeof(info)) {
4789         return NVME_INVALID_FIELD | NVME_DNR;
4790     }
4791 
4792     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
4793         NvmeNamespace *ns = nvme_subsys_ns(n->subsys, i);
4794         if (!ns) {
4795             continue;
4796         }
4797 
4798         nvme_set_blk_stats(ns, &stats);
4799     }
4800 
4801     info.data_units_read[0] =
4802         cpu_to_le64(DIV_ROUND_UP(stats.units_read / 1000000000, 1000000000));
4803     info.data_units_written[0] =
4804         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4805     info.media_units_written[0] =
4806         cpu_to_le64(DIV_ROUND_UP(stats.units_written / 1000000000, 1000000000));
4807 
4808     info.host_read_commands[0] = cpu_to_le64(stats.read_commands);
4809     info.host_write_commands[0] = cpu_to_le64(stats.write_commands);
4810 
4811     buf_len = MIN(sizeof(info) - off, buf_len);
4812 
4813     return nvme_c2h(n, (uint8_t *)&info + off, buf_len, req);
4814 }
4815 
4816 
4817 static uint16_t nvme_fw_log_info(NvmeCtrl *n, uint32_t buf_len, uint64_t off,
4818                                  NvmeRequest *req)
4819 {
4820     uint32_t trans_len;
4821     NvmeFwSlotInfoLog fw_log = {
4822         .afi = 0x1,
4823     };
4824 
4825     if (off >= sizeof(fw_log)) {
4826         return NVME_INVALID_FIELD | NVME_DNR;
4827     }
4828 
4829     strpadcpy((char *)&fw_log.frs1, sizeof(fw_log.frs1), "1.0", ' ');
4830     trans_len = MIN(sizeof(fw_log) - off, buf_len);
4831 
4832     return nvme_c2h(n, (uint8_t *) &fw_log + off, trans_len, req);
4833 }
4834 
4835 static uint16_t nvme_error_info(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4836                                 uint64_t off, NvmeRequest *req)
4837 {
4838     uint32_t trans_len;
4839     NvmeErrorLog errlog;
4840 
4841     if (off >= sizeof(errlog)) {
4842         return NVME_INVALID_FIELD | NVME_DNR;
4843     }
4844 
4845     if (!rae) {
4846         nvme_clear_events(n, NVME_AER_TYPE_ERROR);
4847     }
4848 
4849     memset(&errlog, 0x0, sizeof(errlog));
4850     trans_len = MIN(sizeof(errlog) - off, buf_len);
4851 
4852     return nvme_c2h(n, (uint8_t *)&errlog, trans_len, req);
4853 }
4854 
4855 static uint16_t nvme_changed_nslist(NvmeCtrl *n, uint8_t rae, uint32_t buf_len,
4856                                     uint64_t off, NvmeRequest *req)
4857 {
4858     uint32_t nslist[1024];
4859     uint32_t trans_len;
4860     int i = 0;
4861     uint32_t nsid;
4862 
4863     if (off >= sizeof(nslist)) {
4864         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(nslist));
4865         return NVME_INVALID_FIELD | NVME_DNR;
4866     }
4867 
4868     memset(nslist, 0x0, sizeof(nslist));
4869     trans_len = MIN(sizeof(nslist) - off, buf_len);
4870 
4871     while ((nsid = find_first_bit(n->changed_nsids, NVME_CHANGED_NSID_SIZE)) !=
4872             NVME_CHANGED_NSID_SIZE) {
4873         /*
4874          * If more than 1024 namespaces, the first entry in the log page should
4875          * be set to FFFFFFFFh and the others to 0 as spec.
4876          */
4877         if (i == ARRAY_SIZE(nslist)) {
4878             memset(nslist, 0x0, sizeof(nslist));
4879             nslist[0] = 0xffffffff;
4880             break;
4881         }
4882 
4883         nslist[i++] = nsid;
4884         clear_bit(nsid, n->changed_nsids);
4885     }
4886 
4887     /*
4888      * Remove all the remaining list entries in case returns directly due to
4889      * more than 1024 namespaces.
4890      */
4891     if (nslist[0] == 0xffffffff) {
4892         bitmap_zero(n->changed_nsids, NVME_CHANGED_NSID_SIZE);
4893     }
4894 
4895     if (!rae) {
4896         nvme_clear_events(n, NVME_AER_TYPE_NOTICE);
4897     }
4898 
4899     return nvme_c2h(n, ((uint8_t *)nslist) + off, trans_len, req);
4900 }
4901 
4902 static uint16_t nvme_cmd_effects(NvmeCtrl *n, uint8_t csi, uint32_t buf_len,
4903                                  uint64_t off, NvmeRequest *req)
4904 {
4905     NvmeEffectsLog log = {};
4906     const uint32_t *src_iocs = NULL;
4907     uint32_t trans_len;
4908 
4909     if (off >= sizeof(log)) {
4910         trace_pci_nvme_err_invalid_log_page_offset(off, sizeof(log));
4911         return NVME_INVALID_FIELD | NVME_DNR;
4912     }
4913 
4914     switch (NVME_CC_CSS(ldl_le_p(&n->bar.cc))) {
4915     case NVME_CC_CSS_NVM:
4916         src_iocs = nvme_cse_iocs_nvm;
4917         /* fall through */
4918     case NVME_CC_CSS_ADMIN_ONLY:
4919         break;
4920     case NVME_CC_CSS_CSI:
4921         switch (csi) {
4922         case NVME_CSI_NVM:
4923             src_iocs = nvme_cse_iocs_nvm;
4924             break;
4925         case NVME_CSI_ZONED:
4926             src_iocs = nvme_cse_iocs_zoned;
4927             break;
4928         }
4929     }
4930 
4931     memcpy(log.acs, nvme_cse_acs, sizeof(nvme_cse_acs));
4932 
4933     if (src_iocs) {
4934         memcpy(log.iocs, src_iocs, sizeof(log.iocs));
4935     }
4936 
4937     trans_len = MIN(sizeof(log) - off, buf_len);
4938 
4939     return nvme_c2h(n, ((uint8_t *)&log) + off, trans_len, req);
4940 }
4941 
4942 static size_t sizeof_fdp_conf_descr(size_t nruh, size_t vss)
4943 {
4944     size_t entry_siz = sizeof(NvmeFdpDescrHdr) + nruh * sizeof(NvmeRuhDescr)
4945                        + vss;
4946     return ROUND_UP(entry_siz, 8);
4947 }
4948 
4949 static uint16_t nvme_fdp_confs(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
4950                                uint64_t off, NvmeRequest *req)
4951 {
4952     uint32_t log_size, trans_len;
4953     g_autofree uint8_t *buf = NULL;
4954     NvmeFdpDescrHdr *hdr;
4955     NvmeRuhDescr *ruhd;
4956     NvmeEnduranceGroup *endgrp;
4957     NvmeFdpConfsHdr *log;
4958     size_t nruh, fdp_descr_size;
4959     int i;
4960 
4961     if (endgrpid != 1 || !n->subsys) {
4962         return NVME_INVALID_FIELD | NVME_DNR;
4963     }
4964 
4965     endgrp = &n->subsys->endgrp;
4966 
4967     if (endgrp->fdp.enabled) {
4968         nruh = endgrp->fdp.nruh;
4969     } else {
4970         nruh = 1;
4971     }
4972 
4973     fdp_descr_size = sizeof_fdp_conf_descr(nruh, FDPVSS);
4974     log_size = sizeof(NvmeFdpConfsHdr) + fdp_descr_size;
4975 
4976     if (off >= log_size) {
4977         return NVME_INVALID_FIELD | NVME_DNR;
4978     }
4979 
4980     trans_len = MIN(log_size - off, buf_len);
4981 
4982     buf = g_malloc0(log_size);
4983     log = (NvmeFdpConfsHdr *)buf;
4984     hdr = (NvmeFdpDescrHdr *)(log + 1);
4985     ruhd = (NvmeRuhDescr *)(buf + sizeof(*log) + sizeof(*hdr));
4986 
4987     log->num_confs = cpu_to_le16(0);
4988     log->size = cpu_to_le32(log_size);
4989 
4990     hdr->descr_size = cpu_to_le16(fdp_descr_size);
4991     if (endgrp->fdp.enabled) {
4992         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, VALID, 1);
4993         hdr->fdpa = FIELD_DP8(hdr->fdpa, FDPA, RGIF, endgrp->fdp.rgif);
4994         hdr->nrg = cpu_to_le16(endgrp->fdp.nrg);
4995         hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
4996         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
4997         hdr->nnss = cpu_to_le32(NVME_MAX_NAMESPACES);
4998         hdr->runs = cpu_to_le64(endgrp->fdp.runs);
4999 
5000         for (i = 0; i < nruh; i++) {
5001             ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5002             ruhd++;
5003         }
5004     } else {
5005         /* 1 bit for RUH in PIF -> 2 RUHs max. */
5006         hdr->nrg = cpu_to_le16(1);
5007         hdr->nruh = cpu_to_le16(1);
5008         hdr->maxpids = cpu_to_le16(NVME_FDP_MAXPIDS - 1);
5009         hdr->nnss = cpu_to_le32(1);
5010         hdr->runs = cpu_to_le64(96 * MiB);
5011 
5012         ruhd->ruht = NVME_RUHT_INITIALLY_ISOLATED;
5013     }
5014 
5015     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5016 }
5017 
5018 static uint16_t nvme_fdp_ruh_usage(NvmeCtrl *n, uint32_t endgrpid,
5019                                    uint32_t dw10, uint32_t dw12,
5020                                    uint32_t buf_len, uint64_t off,
5021                                    NvmeRequest *req)
5022 {
5023     NvmeRuHandle *ruh;
5024     NvmeRuhuLog *hdr;
5025     NvmeRuhuDescr *ruhud;
5026     NvmeEnduranceGroup *endgrp;
5027     g_autofree uint8_t *buf = NULL;
5028     uint32_t log_size, trans_len;
5029     uint16_t i;
5030 
5031     if (endgrpid != 1 || !n->subsys) {
5032         return NVME_INVALID_FIELD | NVME_DNR;
5033     }
5034 
5035     endgrp = &n->subsys->endgrp;
5036 
5037     if (!endgrp->fdp.enabled) {
5038         return NVME_FDP_DISABLED | NVME_DNR;
5039     }
5040 
5041     log_size = sizeof(NvmeRuhuLog) + endgrp->fdp.nruh * sizeof(NvmeRuhuDescr);
5042 
5043     if (off >= log_size) {
5044         return NVME_INVALID_FIELD | NVME_DNR;
5045     }
5046 
5047     trans_len = MIN(log_size - off, buf_len);
5048 
5049     buf = g_malloc0(log_size);
5050     hdr = (NvmeRuhuLog *)buf;
5051     ruhud = (NvmeRuhuDescr *)(hdr + 1);
5052 
5053     ruh = endgrp->fdp.ruhs;
5054     hdr->nruh = cpu_to_le16(endgrp->fdp.nruh);
5055 
5056     for (i = 0; i < endgrp->fdp.nruh; i++, ruhud++, ruh++) {
5057         ruhud->ruha = ruh->ruha;
5058     }
5059 
5060     return nvme_c2h(n, (uint8_t *)buf + off, trans_len, req);
5061 }
5062 
5063 static uint16_t nvme_fdp_stats(NvmeCtrl *n, uint32_t endgrpid, uint32_t buf_len,
5064                                uint64_t off, NvmeRequest *req)
5065 {
5066     NvmeEnduranceGroup *endgrp;
5067     NvmeFdpStatsLog log = {};
5068     uint32_t trans_len;
5069 
5070     if (off >= sizeof(NvmeFdpStatsLog)) {
5071         return NVME_INVALID_FIELD | NVME_DNR;
5072     }
5073 
5074     if (endgrpid != 1 || !n->subsys) {
5075         return NVME_INVALID_FIELD | NVME_DNR;
5076     }
5077 
5078     if (!n->subsys->endgrp.fdp.enabled) {
5079         return NVME_FDP_DISABLED | NVME_DNR;
5080     }
5081 
5082     endgrp = &n->subsys->endgrp;
5083 
5084     trans_len = MIN(sizeof(log) - off, buf_len);
5085 
5086     /* spec value is 128 bit, we only use 64 bit */
5087     log.hbmw[0] = cpu_to_le64(endgrp->fdp.hbmw);
5088     log.mbmw[0] = cpu_to_le64(endgrp->fdp.mbmw);
5089     log.mbe[0] = cpu_to_le64(endgrp->fdp.mbe);
5090 
5091     return nvme_c2h(n, (uint8_t *)&log + off, trans_len, req);
5092 }
5093 
5094 static uint16_t nvme_fdp_events(NvmeCtrl *n, uint32_t endgrpid,
5095                                 uint32_t buf_len, uint64_t off,
5096                                 NvmeRequest *req)
5097 {
5098     NvmeEnduranceGroup *endgrp;
5099     NvmeCmd *cmd = &req->cmd;
5100     bool host_events = (cmd->cdw10 >> 8) & 0x1;
5101     uint32_t log_size, trans_len;
5102     NvmeFdpEventBuffer *ebuf;
5103     g_autofree NvmeFdpEventsLog *elog = NULL;
5104     NvmeFdpEvent *event;
5105 
5106     if (endgrpid != 1 || !n->subsys) {
5107         return NVME_INVALID_FIELD | NVME_DNR;
5108     }
5109 
5110     endgrp = &n->subsys->endgrp;
5111 
5112     if (!endgrp->fdp.enabled) {
5113         return NVME_FDP_DISABLED | NVME_DNR;
5114     }
5115 
5116     if (host_events) {
5117         ebuf = &endgrp->fdp.host_events;
5118     } else {
5119         ebuf = &endgrp->fdp.ctrl_events;
5120     }
5121 
5122     log_size = sizeof(NvmeFdpEventsLog) + ebuf->nelems * sizeof(NvmeFdpEvent);
5123     trans_len = MIN(log_size - off, buf_len);
5124     elog = g_malloc0(log_size);
5125     elog->num_events = cpu_to_le32(ebuf->nelems);
5126     event = (NvmeFdpEvent *)(elog + 1);
5127 
5128     if (ebuf->nelems && ebuf->start == ebuf->next) {
5129         unsigned int nelems = (NVME_FDP_MAX_EVENTS - ebuf->start);
5130         /* wrap over, copy [start;NVME_FDP_MAX_EVENTS[ and [0; next[ */
5131         memcpy(event, &ebuf->events[ebuf->start],
5132                sizeof(NvmeFdpEvent) * nelems);
5133         memcpy(event + nelems, ebuf->events,
5134                sizeof(NvmeFdpEvent) * ebuf->next);
5135     } else if (ebuf->start < ebuf->next) {
5136         memcpy(event, &ebuf->events[ebuf->start],
5137                sizeof(NvmeFdpEvent) * (ebuf->next - ebuf->start));
5138     }
5139 
5140     return nvme_c2h(n, (uint8_t *)elog + off, trans_len, req);
5141 }
5142 
5143 static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
5144 {
5145     NvmeCmd *cmd = &req->cmd;
5146 
5147     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5148     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5149     uint32_t dw12 = le32_to_cpu(cmd->cdw12);
5150     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
5151     uint8_t  lid = dw10 & 0xff;
5152     uint8_t  lsp = (dw10 >> 8) & 0xf;
5153     uint8_t  rae = (dw10 >> 15) & 0x1;
5154     uint8_t  csi = le32_to_cpu(cmd->cdw14) >> 24;
5155     uint32_t numdl, numdu, lspi;
5156     uint64_t off, lpol, lpou;
5157     size_t   len;
5158     uint16_t status;
5159 
5160     numdl = (dw10 >> 16);
5161     numdu = (dw11 & 0xffff);
5162     lspi = (dw11 >> 16);
5163     lpol = dw12;
5164     lpou = dw13;
5165 
5166     len = (((numdu << 16) | numdl) + 1) << 2;
5167     off = (lpou << 32ULL) | lpol;
5168 
5169     if (off & 0x3) {
5170         return NVME_INVALID_FIELD | NVME_DNR;
5171     }
5172 
5173     trace_pci_nvme_get_log(nvme_cid(req), lid, lsp, rae, len, off);
5174 
5175     status = nvme_check_mdts(n, len);
5176     if (status) {
5177         return status;
5178     }
5179 
5180     switch (lid) {
5181     case NVME_LOG_ERROR_INFO:
5182         return nvme_error_info(n, rae, len, off, req);
5183     case NVME_LOG_SMART_INFO:
5184         return nvme_smart_info(n, rae, len, off, req);
5185     case NVME_LOG_FW_SLOT_INFO:
5186         return nvme_fw_log_info(n, len, off, req);
5187     case NVME_LOG_CHANGED_NSLIST:
5188         return nvme_changed_nslist(n, rae, len, off, req);
5189     case NVME_LOG_CMD_EFFECTS:
5190         return nvme_cmd_effects(n, csi, len, off, req);
5191     case NVME_LOG_ENDGRP:
5192         return nvme_endgrp_info(n, rae, len, off, req);
5193     case NVME_LOG_FDP_CONFS:
5194         return nvme_fdp_confs(n, lspi, len, off, req);
5195     case NVME_LOG_FDP_RUH_USAGE:
5196         return nvme_fdp_ruh_usage(n, lspi, dw10, dw12, len, off, req);
5197     case NVME_LOG_FDP_STATS:
5198         return nvme_fdp_stats(n, lspi, len, off, req);
5199     case NVME_LOG_FDP_EVENTS:
5200         return nvme_fdp_events(n, lspi, len, off, req);
5201     default:
5202         trace_pci_nvme_err_invalid_log_page(nvme_cid(req), lid);
5203         return NVME_INVALID_FIELD | NVME_DNR;
5204     }
5205 }
5206 
5207 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
5208 {
5209     PCIDevice *pci = PCI_DEVICE(n);
5210     uint16_t offset = (cq->cqid << 3) + (1 << 2);
5211 
5212     n->cq[cq->cqid] = NULL;
5213     qemu_bh_delete(cq->bh);
5214     if (cq->ioeventfd_enabled) {
5215         memory_region_del_eventfd(&n->iomem,
5216                                   0x1000 + offset, 4, false, 0, &cq->notifier);
5217         event_notifier_set_handler(&cq->notifier, NULL);
5218         event_notifier_cleanup(&cq->notifier);
5219     }
5220     if (msix_enabled(pci)) {
5221         msix_vector_unuse(pci, cq->vector);
5222     }
5223     if (cq->cqid) {
5224         g_free(cq);
5225     }
5226 }
5227 
5228 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)
5229 {
5230     NvmeDeleteQ *c = (NvmeDeleteQ *)&req->cmd;
5231     NvmeCQueue *cq;
5232     uint16_t qid = le16_to_cpu(c->qid);
5233 
5234     if (unlikely(!qid || nvme_check_cqid(n, qid))) {
5235         trace_pci_nvme_err_invalid_del_cq_cqid(qid);
5236         return NVME_INVALID_CQID | NVME_DNR;
5237     }
5238 
5239     cq = n->cq[qid];
5240     if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
5241         trace_pci_nvme_err_invalid_del_cq_notempty(qid);
5242         return NVME_INVALID_QUEUE_DEL;
5243     }
5244 
5245     if (cq->irq_enabled && cq->tail != cq->head) {
5246         n->cq_pending--;
5247     }
5248 
5249     nvme_irq_deassert(n, cq);
5250     trace_pci_nvme_del_cq(qid);
5251     nvme_free_cq(cq, n);
5252     return NVME_SUCCESS;
5253 }
5254 
5255 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
5256                          uint16_t cqid, uint16_t vector, uint16_t size,
5257                          uint16_t irq_enabled)
5258 {
5259     PCIDevice *pci = PCI_DEVICE(n);
5260 
5261     if (msix_enabled(pci)) {
5262         msix_vector_use(pci, vector);
5263     }
5264     cq->ctrl = n;
5265     cq->cqid = cqid;
5266     cq->size = size;
5267     cq->dma_addr = dma_addr;
5268     cq->phase = 1;
5269     cq->irq_enabled = irq_enabled;
5270     cq->vector = vector;
5271     cq->head = cq->tail = 0;
5272     QTAILQ_INIT(&cq->req_list);
5273     QTAILQ_INIT(&cq->sq_list);
5274     if (n->dbbuf_enabled) {
5275         cq->db_addr = n->dbbuf_dbs + (cqid << 3) + (1 << 2);
5276         cq->ei_addr = n->dbbuf_eis + (cqid << 3) + (1 << 2);
5277 
5278         if (n->params.ioeventfd && cqid != 0) {
5279             if (!nvme_init_cq_ioeventfd(cq)) {
5280                 cq->ioeventfd_enabled = true;
5281             }
5282         }
5283     }
5284     n->cq[cqid] = cq;
5285     cq->bh = qemu_bh_new_guarded(nvme_post_cqes, cq,
5286                                  &DEVICE(cq->ctrl)->mem_reentrancy_guard);
5287 }
5288 
5289 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeRequest *req)
5290 {
5291     NvmeCQueue *cq;
5292     NvmeCreateCq *c = (NvmeCreateCq *)&req->cmd;
5293     uint16_t cqid = le16_to_cpu(c->cqid);
5294     uint16_t vector = le16_to_cpu(c->irq_vector);
5295     uint16_t qsize = le16_to_cpu(c->qsize);
5296     uint16_t qflags = le16_to_cpu(c->cq_flags);
5297     uint64_t prp1 = le64_to_cpu(c->prp1);
5298 
5299     trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
5300                              NVME_CQ_FLAGS_IEN(qflags) != 0);
5301 
5302     if (unlikely(!cqid || cqid > n->conf_ioqpairs || n->cq[cqid] != NULL)) {
5303         trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
5304         return NVME_INVALID_QID | NVME_DNR;
5305     }
5306     if (unlikely(!qsize || qsize > NVME_CAP_MQES(ldq_le_p(&n->bar.cap)))) {
5307         trace_pci_nvme_err_invalid_create_cq_size(qsize);
5308         return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
5309     }
5310     if (unlikely(prp1 & (n->page_size - 1))) {
5311         trace_pci_nvme_err_invalid_create_cq_addr(prp1);
5312         return NVME_INVALID_PRP_OFFSET | NVME_DNR;
5313     }
5314     if (unlikely(!msix_enabled(PCI_DEVICE(n)) && vector)) {
5315         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5316         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5317     }
5318     if (unlikely(vector >= n->conf_msix_qsize)) {
5319         trace_pci_nvme_err_invalid_create_cq_vector(vector);
5320         return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
5321     }
5322     if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
5323         trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
5324         return NVME_INVALID_FIELD | NVME_DNR;
5325     }
5326 
5327     cq = g_malloc0(sizeof(*cq));
5328     nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
5329                  NVME_CQ_FLAGS_IEN(qflags));
5330 
5331     /*
5332      * It is only required to set qs_created when creating a completion queue;
5333      * creating a submission queue without a matching completion queue will
5334      * fail.
5335      */
5336     n->qs_created = true;
5337     return NVME_SUCCESS;
5338 }
5339 
5340 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl *n, NvmeRequest *req)
5341 {
5342     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5343 
5344     return nvme_c2h(n, id, sizeof(id), req);
5345 }
5346 
5347 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeRequest *req)
5348 {
5349     trace_pci_nvme_identify_ctrl();
5350 
5351     return nvme_c2h(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl), req);
5352 }
5353 
5354 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl *n, NvmeRequest *req)
5355 {
5356     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5357     uint8_t id[NVME_IDENTIFY_DATA_SIZE] = {};
5358     NvmeIdCtrlNvm *id_nvm = (NvmeIdCtrlNvm *)&id;
5359 
5360     trace_pci_nvme_identify_ctrl_csi(c->csi);
5361 
5362     switch (c->csi) {
5363     case NVME_CSI_NVM:
5364         id_nvm->vsl = n->params.vsl;
5365         id_nvm->dmrsl = cpu_to_le32(n->dmrsl);
5366         break;
5367 
5368     case NVME_CSI_ZONED:
5369         ((NvmeIdCtrlZoned *)&id)->zasl = n->params.zasl;
5370         break;
5371 
5372     default:
5373         return NVME_INVALID_FIELD | NVME_DNR;
5374     }
5375 
5376     return nvme_c2h(n, id, sizeof(id), req);
5377 }
5378 
5379 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeRequest *req, bool active)
5380 {
5381     NvmeNamespace *ns;
5382     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5383     uint32_t nsid = le32_to_cpu(c->nsid);
5384 
5385     trace_pci_nvme_identify_ns(nsid);
5386 
5387     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5388         return NVME_INVALID_NSID | NVME_DNR;
5389     }
5390 
5391     ns = nvme_ns(n, nsid);
5392     if (unlikely(!ns)) {
5393         if (!active) {
5394             ns = nvme_subsys_ns(n->subsys, nsid);
5395             if (!ns) {
5396                 return nvme_rpt_empty_id_struct(n, req);
5397             }
5398         } else {
5399             return nvme_rpt_empty_id_struct(n, req);
5400         }
5401     }
5402 
5403     if (active || ns->csi == NVME_CSI_NVM) {
5404         return nvme_c2h(n, (uint8_t *)&ns->id_ns, sizeof(NvmeIdNs), req);
5405     }
5406 
5407     return NVME_INVALID_CMD_SET | NVME_DNR;
5408 }
5409 
5410 static uint16_t nvme_identify_ctrl_list(NvmeCtrl *n, NvmeRequest *req,
5411                                         bool attached)
5412 {
5413     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5414     uint32_t nsid = le32_to_cpu(c->nsid);
5415     uint16_t min_id = le16_to_cpu(c->ctrlid);
5416     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
5417     uint16_t *ids = &list[1];
5418     NvmeNamespace *ns;
5419     NvmeCtrl *ctrl;
5420     int cntlid, nr_ids = 0;
5421 
5422     trace_pci_nvme_identify_ctrl_list(c->cns, min_id);
5423 
5424     if (!n->subsys) {
5425         return NVME_INVALID_FIELD | NVME_DNR;
5426     }
5427 
5428     if (attached) {
5429         if (nsid == NVME_NSID_BROADCAST) {
5430             return NVME_INVALID_FIELD | NVME_DNR;
5431         }
5432 
5433         ns = nvme_subsys_ns(n->subsys, nsid);
5434         if (!ns) {
5435             return NVME_INVALID_FIELD | NVME_DNR;
5436         }
5437     }
5438 
5439     for (cntlid = min_id; cntlid < ARRAY_SIZE(n->subsys->ctrls); cntlid++) {
5440         ctrl = nvme_subsys_ctrl(n->subsys, cntlid);
5441         if (!ctrl) {
5442             continue;
5443         }
5444 
5445         if (attached && !nvme_ns(ctrl, nsid)) {
5446             continue;
5447         }
5448 
5449         ids[nr_ids++] = cntlid;
5450     }
5451 
5452     list[0] = nr_ids;
5453 
5454     return nvme_c2h(n, (uint8_t *)list, sizeof(list), req);
5455 }
5456 
5457 static uint16_t nvme_identify_pri_ctrl_cap(NvmeCtrl *n, NvmeRequest *req)
5458 {
5459     trace_pci_nvme_identify_pri_ctrl_cap(le16_to_cpu(n->pri_ctrl_cap.cntlid));
5460 
5461     return nvme_c2h(n, (uint8_t *)&n->pri_ctrl_cap,
5462                     sizeof(NvmePriCtrlCap), req);
5463 }
5464 
5465 static uint16_t nvme_identify_sec_ctrl_list(NvmeCtrl *n, NvmeRequest *req)
5466 {
5467     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5468     uint16_t pri_ctrl_id = le16_to_cpu(n->pri_ctrl_cap.cntlid);
5469     uint16_t min_id = le16_to_cpu(c->ctrlid);
5470     uint8_t num_sec_ctrl = n->sec_ctrl_list.numcntl;
5471     NvmeSecCtrlList list = {0};
5472     uint8_t i;
5473 
5474     for (i = 0; i < num_sec_ctrl; i++) {
5475         if (n->sec_ctrl_list.sec[i].scid >= min_id) {
5476             list.numcntl = num_sec_ctrl - i;
5477             memcpy(&list.sec, n->sec_ctrl_list.sec + i,
5478                    list.numcntl * sizeof(NvmeSecCtrlEntry));
5479             break;
5480         }
5481     }
5482 
5483     trace_pci_nvme_identify_sec_ctrl_list(pri_ctrl_id, list.numcntl);
5484 
5485     return nvme_c2h(n, (uint8_t *)&list, sizeof(list), req);
5486 }
5487 
5488 static uint16_t nvme_identify_ns_csi(NvmeCtrl *n, NvmeRequest *req,
5489                                      bool active)
5490 {
5491     NvmeNamespace *ns;
5492     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5493     uint32_t nsid = le32_to_cpu(c->nsid);
5494 
5495     trace_pci_nvme_identify_ns_csi(nsid, c->csi);
5496 
5497     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5498         return NVME_INVALID_NSID | NVME_DNR;
5499     }
5500 
5501     ns = nvme_ns(n, nsid);
5502     if (unlikely(!ns)) {
5503         if (!active) {
5504             ns = nvme_subsys_ns(n->subsys, nsid);
5505             if (!ns) {
5506                 return nvme_rpt_empty_id_struct(n, req);
5507             }
5508         } else {
5509             return nvme_rpt_empty_id_struct(n, req);
5510         }
5511     }
5512 
5513     if (c->csi == NVME_CSI_NVM) {
5514         return nvme_c2h(n, (uint8_t *)&ns->id_ns_nvm, sizeof(NvmeIdNsNvm),
5515                         req);
5516     } else if (c->csi == NVME_CSI_ZONED && ns->csi == NVME_CSI_ZONED) {
5517         return nvme_c2h(n, (uint8_t *)ns->id_ns_zoned, sizeof(NvmeIdNsZoned),
5518                         req);
5519     }
5520 
5521     return NVME_INVALID_FIELD | NVME_DNR;
5522 }
5523 
5524 static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeRequest *req,
5525                                      bool active)
5526 {
5527     NvmeNamespace *ns;
5528     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5529     uint32_t min_nsid = le32_to_cpu(c->nsid);
5530     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5531     static const int data_len = sizeof(list);
5532     uint32_t *list_ptr = (uint32_t *)list;
5533     int i, j = 0;
5534 
5535     trace_pci_nvme_identify_nslist(min_nsid);
5536 
5537     /*
5538      * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
5539      * since the Active Namespace ID List should return namespaces with ids
5540      * *higher* than the NSID specified in the command. This is also specified
5541      * in the spec (NVM Express v1.3d, Section 5.15.4).
5542      */
5543     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5544         return NVME_INVALID_NSID | NVME_DNR;
5545     }
5546 
5547     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5548         ns = nvme_ns(n, i);
5549         if (!ns) {
5550             if (!active) {
5551                 ns = nvme_subsys_ns(n->subsys, i);
5552                 if (!ns) {
5553                     continue;
5554                 }
5555             } else {
5556                 continue;
5557             }
5558         }
5559         if (ns->params.nsid <= min_nsid) {
5560             continue;
5561         }
5562         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5563         if (j == data_len / sizeof(uint32_t)) {
5564             break;
5565         }
5566     }
5567 
5568     return nvme_c2h(n, list, data_len, req);
5569 }
5570 
5571 static uint16_t nvme_identify_nslist_csi(NvmeCtrl *n, NvmeRequest *req,
5572                                          bool active)
5573 {
5574     NvmeNamespace *ns;
5575     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5576     uint32_t min_nsid = le32_to_cpu(c->nsid);
5577     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5578     static const int data_len = sizeof(list);
5579     uint32_t *list_ptr = (uint32_t *)list;
5580     int i, j = 0;
5581 
5582     trace_pci_nvme_identify_nslist_csi(min_nsid, c->csi);
5583 
5584     /*
5585      * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
5586      */
5587     if (min_nsid >= NVME_NSID_BROADCAST - 1) {
5588         return NVME_INVALID_NSID | NVME_DNR;
5589     }
5590 
5591     if (c->csi != NVME_CSI_NVM && c->csi != NVME_CSI_ZONED) {
5592         return NVME_INVALID_FIELD | NVME_DNR;
5593     }
5594 
5595     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5596         ns = nvme_ns(n, i);
5597         if (!ns) {
5598             if (!active) {
5599                 ns = nvme_subsys_ns(n->subsys, i);
5600                 if (!ns) {
5601                     continue;
5602                 }
5603             } else {
5604                 continue;
5605             }
5606         }
5607         if (ns->params.nsid <= min_nsid || c->csi != ns->csi) {
5608             continue;
5609         }
5610         list_ptr[j++] = cpu_to_le32(ns->params.nsid);
5611         if (j == data_len / sizeof(uint32_t)) {
5612             break;
5613         }
5614     }
5615 
5616     return nvme_c2h(n, list, data_len, req);
5617 }
5618 
5619 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl *n, NvmeRequest *req)
5620 {
5621     NvmeNamespace *ns;
5622     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5623     uint32_t nsid = le32_to_cpu(c->nsid);
5624     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5625     uint8_t *pos = list;
5626     struct {
5627         NvmeIdNsDescr hdr;
5628         uint8_t v[NVME_NIDL_UUID];
5629     } QEMU_PACKED uuid = {};
5630     struct {
5631         NvmeIdNsDescr hdr;
5632         uint64_t v;
5633     } QEMU_PACKED eui64 = {};
5634     struct {
5635         NvmeIdNsDescr hdr;
5636         uint8_t v;
5637     } QEMU_PACKED csi = {};
5638 
5639     trace_pci_nvme_identify_ns_descr_list(nsid);
5640 
5641     if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5642         return NVME_INVALID_NSID | NVME_DNR;
5643     }
5644 
5645     ns = nvme_ns(n, nsid);
5646     if (unlikely(!ns)) {
5647         return NVME_INVALID_FIELD | NVME_DNR;
5648     }
5649 
5650     if (!qemu_uuid_is_null(&ns->params.uuid)) {
5651         uuid.hdr.nidt = NVME_NIDT_UUID;
5652         uuid.hdr.nidl = NVME_NIDL_UUID;
5653         memcpy(uuid.v, ns->params.uuid.data, NVME_NIDL_UUID);
5654         memcpy(pos, &uuid, sizeof(uuid));
5655         pos += sizeof(uuid);
5656     }
5657 
5658     if (ns->params.eui64) {
5659         eui64.hdr.nidt = NVME_NIDT_EUI64;
5660         eui64.hdr.nidl = NVME_NIDL_EUI64;
5661         eui64.v = cpu_to_be64(ns->params.eui64);
5662         memcpy(pos, &eui64, sizeof(eui64));
5663         pos += sizeof(eui64);
5664     }
5665 
5666     csi.hdr.nidt = NVME_NIDT_CSI;
5667     csi.hdr.nidl = NVME_NIDL_CSI;
5668     csi.v = ns->csi;
5669     memcpy(pos, &csi, sizeof(csi));
5670     pos += sizeof(csi);
5671 
5672     return nvme_c2h(n, list, sizeof(list), req);
5673 }
5674 
5675 static uint16_t nvme_identify_cmd_set(NvmeCtrl *n, NvmeRequest *req)
5676 {
5677     uint8_t list[NVME_IDENTIFY_DATA_SIZE] = {};
5678     static const int data_len = sizeof(list);
5679 
5680     trace_pci_nvme_identify_cmd_set();
5681 
5682     NVME_SET_CSI(*list, NVME_CSI_NVM);
5683     NVME_SET_CSI(*list, NVME_CSI_ZONED);
5684 
5685     return nvme_c2h(n, list, data_len, req);
5686 }
5687 
5688 static uint16_t nvme_identify(NvmeCtrl *n, NvmeRequest *req)
5689 {
5690     NvmeIdentify *c = (NvmeIdentify *)&req->cmd;
5691 
5692     trace_pci_nvme_identify(nvme_cid(req), c->cns, le16_to_cpu(c->ctrlid),
5693                             c->csi);
5694 
5695     switch (c->cns) {
5696     case NVME_ID_CNS_NS:
5697         return nvme_identify_ns(n, req, true);
5698     case NVME_ID_CNS_NS_PRESENT:
5699         return nvme_identify_ns(n, req, false);
5700     case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST:
5701         return nvme_identify_ctrl_list(n, req, true);
5702     case NVME_ID_CNS_CTRL_LIST:
5703         return nvme_identify_ctrl_list(n, req, false);
5704     case NVME_ID_CNS_PRIMARY_CTRL_CAP:
5705         return nvme_identify_pri_ctrl_cap(n, req);
5706     case NVME_ID_CNS_SECONDARY_CTRL_LIST:
5707         return nvme_identify_sec_ctrl_list(n, req);
5708     case NVME_ID_CNS_CS_NS:
5709         return nvme_identify_ns_csi(n, req, true);
5710     case NVME_ID_CNS_CS_NS_PRESENT:
5711         return nvme_identify_ns_csi(n, req, false);
5712     case NVME_ID_CNS_CTRL:
5713         return nvme_identify_ctrl(n, req);
5714     case NVME_ID_CNS_CS_CTRL:
5715         return nvme_identify_ctrl_csi(n, req);
5716     case NVME_ID_CNS_NS_ACTIVE_LIST:
5717         return nvme_identify_nslist(n, req, true);
5718     case NVME_ID_CNS_NS_PRESENT_LIST:
5719         return nvme_identify_nslist(n, req, false);
5720     case NVME_ID_CNS_CS_NS_ACTIVE_LIST:
5721         return nvme_identify_nslist_csi(n, req, true);
5722     case NVME_ID_CNS_CS_NS_PRESENT_LIST:
5723         return nvme_identify_nslist_csi(n, req, false);
5724     case NVME_ID_CNS_NS_DESCR_LIST:
5725         return nvme_identify_ns_descr_list(n, req);
5726     case NVME_ID_CNS_IO_COMMAND_SET:
5727         return nvme_identify_cmd_set(n, req);
5728     default:
5729         trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
5730         return NVME_INVALID_FIELD | NVME_DNR;
5731     }
5732 }
5733 
5734 static uint16_t nvme_abort(NvmeCtrl *n, NvmeRequest *req)
5735 {
5736     uint16_t sqid = le32_to_cpu(req->cmd.cdw10) & 0xffff;
5737 
5738     req->cqe.result = 1;
5739     if (nvme_check_sqid(n, sqid)) {
5740         return NVME_INVALID_FIELD | NVME_DNR;
5741     }
5742 
5743     return NVME_SUCCESS;
5744 }
5745 
5746 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
5747 {
5748     trace_pci_nvme_setfeat_timestamp(ts);
5749 
5750     n->host_timestamp = le64_to_cpu(ts);
5751     n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5752 }
5753 
5754 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
5755 {
5756     uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
5757     uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
5758 
5759     union nvme_timestamp {
5760         struct {
5761             uint64_t timestamp:48;
5762             uint64_t sync:1;
5763             uint64_t origin:3;
5764             uint64_t rsvd1:12;
5765         };
5766         uint64_t all;
5767     };
5768 
5769     union nvme_timestamp ts;
5770     ts.all = 0;
5771     ts.timestamp = n->host_timestamp + elapsed_time;
5772 
5773     /* If the host timestamp is non-zero, set the timestamp origin */
5774     ts.origin = n->host_timestamp ? 0x01 : 0x00;
5775 
5776     trace_pci_nvme_getfeat_timestamp(ts.all);
5777 
5778     return cpu_to_le64(ts.all);
5779 }
5780 
5781 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
5782 {
5783     uint64_t timestamp = nvme_get_timestamp(n);
5784 
5785     return nvme_c2h(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
5786 }
5787 
5788 static int nvme_get_feature_fdp(NvmeCtrl *n, uint32_t endgrpid,
5789                                 uint32_t *result)
5790 {
5791     *result = 0;
5792 
5793     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
5794         return NVME_INVALID_FIELD | NVME_DNR;
5795     }
5796 
5797     *result = FIELD_DP16(0, FEAT_FDP, FDPE, 1);
5798     *result = FIELD_DP16(*result, FEAT_FDP, CONF_NDX, 0);
5799 
5800     return NVME_SUCCESS;
5801 }
5802 
5803 static uint16_t nvme_get_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
5804                                             NvmeRequest *req, uint32_t *result)
5805 {
5806     NvmeCmd *cmd = &req->cmd;
5807     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
5808     uint16_t ph = cdw11 & 0xffff;
5809     uint8_t noet = (cdw11 >> 16) & 0xff;
5810     uint16_t ruhid, ret;
5811     uint32_t nentries = 0;
5812     uint8_t s_events_ndx = 0;
5813     size_t s_events_siz = sizeof(NvmeFdpEventDescr) * noet;
5814     g_autofree NvmeFdpEventDescr *s_events = g_malloc0(s_events_siz);
5815     NvmeRuHandle *ruh;
5816     NvmeFdpEventDescr *s_event;
5817 
5818     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
5819         return NVME_FDP_DISABLED | NVME_DNR;
5820     }
5821 
5822     if (!nvme_ph_valid(ns, ph)) {
5823         return NVME_INVALID_FIELD | NVME_DNR;
5824     }
5825 
5826     ruhid = ns->fdp.phs[ph];
5827     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
5828 
5829     assert(ruh);
5830 
5831     if (unlikely(noet == 0)) {
5832         return NVME_INVALID_FIELD | NVME_DNR;
5833     }
5834 
5835     for (uint8_t event_type = 0; event_type < FDP_EVT_MAX; event_type++) {
5836         uint8_t shift = nvme_fdp_evf_shifts[event_type];
5837         if (!shift && event_type) {
5838             /*
5839              * only first entry (event_type == 0) has a shift value of 0
5840              * other entries are simply unpopulated.
5841              */
5842             continue;
5843         }
5844 
5845         nentries++;
5846 
5847         s_event = &s_events[s_events_ndx];
5848         s_event->evt = event_type;
5849         s_event->evta = (ruh->event_filter >> shift) & 0x1;
5850 
5851         /* break if all `noet` entries are filled */
5852         if ((++s_events_ndx) == noet) {
5853             break;
5854         }
5855     }
5856 
5857     ret = nvme_c2h(n, s_events, s_events_siz, req);
5858     if (ret) {
5859         return ret;
5860     }
5861 
5862     *result = nentries;
5863     return NVME_SUCCESS;
5864 }
5865 
5866 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeRequest *req)
5867 {
5868     NvmeCmd *cmd = &req->cmd;
5869     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
5870     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
5871     uint32_t nsid = le32_to_cpu(cmd->nsid);
5872     uint32_t result;
5873     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
5874     NvmeGetFeatureSelect sel = NVME_GETFEAT_SELECT(dw10);
5875     uint16_t iv;
5876     NvmeNamespace *ns;
5877     int i;
5878     uint16_t endgrpid = 0, ret = NVME_SUCCESS;
5879 
5880     static const uint32_t nvme_feature_default[NVME_FID_MAX] = {
5881         [NVME_ARBITRATION] = NVME_ARB_AB_NOLIMIT,
5882     };
5883 
5884     trace_pci_nvme_getfeat(nvme_cid(req), nsid, fid, sel, dw11);
5885 
5886     if (!nvme_feature_support[fid]) {
5887         return NVME_INVALID_FIELD | NVME_DNR;
5888     }
5889 
5890     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
5891         if (!nvme_nsid_valid(n, nsid) || nsid == NVME_NSID_BROADCAST) {
5892             /*
5893              * The Reservation Notification Mask and Reservation Persistence
5894              * features require a status code of Invalid Field in Command when
5895              * NSID is FFFFFFFFh. Since the device does not support those
5896              * features we can always return Invalid Namespace or Format as we
5897              * should do for all other features.
5898              */
5899             return NVME_INVALID_NSID | NVME_DNR;
5900         }
5901 
5902         if (!nvme_ns(n, nsid)) {
5903             return NVME_INVALID_FIELD | NVME_DNR;
5904         }
5905     }
5906 
5907     switch (sel) {
5908     case NVME_GETFEAT_SELECT_CURRENT:
5909         break;
5910     case NVME_GETFEAT_SELECT_SAVED:
5911         /* no features are saveable by the controller; fallthrough */
5912     case NVME_GETFEAT_SELECT_DEFAULT:
5913         goto defaults;
5914     case NVME_GETFEAT_SELECT_CAP:
5915         result = nvme_feature_cap[fid];
5916         goto out;
5917     }
5918 
5919     switch (fid) {
5920     case NVME_TEMPERATURE_THRESHOLD:
5921         result = 0;
5922 
5923         /*
5924          * The controller only implements the Composite Temperature sensor, so
5925          * return 0 for all other sensors.
5926          */
5927         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
5928             goto out;
5929         }
5930 
5931         switch (NVME_TEMP_THSEL(dw11)) {
5932         case NVME_TEMP_THSEL_OVER:
5933             result = n->features.temp_thresh_hi;
5934             goto out;
5935         case NVME_TEMP_THSEL_UNDER:
5936             result = n->features.temp_thresh_low;
5937             goto out;
5938         }
5939 
5940         return NVME_INVALID_FIELD | NVME_DNR;
5941     case NVME_ERROR_RECOVERY:
5942         if (!nvme_nsid_valid(n, nsid)) {
5943             return NVME_INVALID_NSID | NVME_DNR;
5944         }
5945 
5946         ns = nvme_ns(n, nsid);
5947         if (unlikely(!ns)) {
5948             return NVME_INVALID_FIELD | NVME_DNR;
5949         }
5950 
5951         result = ns->features.err_rec;
5952         goto out;
5953     case NVME_VOLATILE_WRITE_CACHE:
5954         result = 0;
5955         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
5956             ns = nvme_ns(n, i);
5957             if (!ns) {
5958                 continue;
5959             }
5960 
5961             result = blk_enable_write_cache(ns->blkconf.blk);
5962             if (result) {
5963                 break;
5964             }
5965         }
5966         trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
5967         goto out;
5968     case NVME_ASYNCHRONOUS_EVENT_CONF:
5969         result = n->features.async_config;
5970         goto out;
5971     case NVME_TIMESTAMP:
5972         return nvme_get_feature_timestamp(n, req);
5973     case NVME_HOST_BEHAVIOR_SUPPORT:
5974         return nvme_c2h(n, (uint8_t *)&n->features.hbs,
5975                         sizeof(n->features.hbs), req);
5976     case NVME_FDP_MODE:
5977         endgrpid = dw11 & 0xff;
5978 
5979         if (endgrpid != 0x1) {
5980             return NVME_INVALID_FIELD | NVME_DNR;
5981         }
5982 
5983         ret = nvme_get_feature_fdp(n, endgrpid, &result);
5984         if (ret) {
5985             return ret;
5986         }
5987         goto out;
5988     case NVME_FDP_EVENTS:
5989         if (!nvme_nsid_valid(n, nsid)) {
5990             return NVME_INVALID_NSID | NVME_DNR;
5991         }
5992 
5993         ns = nvme_ns(n, nsid);
5994         if (unlikely(!ns)) {
5995             return NVME_INVALID_FIELD | NVME_DNR;
5996         }
5997 
5998         ret = nvme_get_feature_fdp_events(n, ns, req, &result);
5999         if (ret) {
6000             return ret;
6001         }
6002         goto out;
6003     default:
6004         break;
6005     }
6006 
6007 defaults:
6008     switch (fid) {
6009     case NVME_TEMPERATURE_THRESHOLD:
6010         result = 0;
6011 
6012         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6013             break;
6014         }
6015 
6016         if (NVME_TEMP_THSEL(dw11) == NVME_TEMP_THSEL_OVER) {
6017             result = NVME_TEMPERATURE_WARNING;
6018         }
6019 
6020         break;
6021     case NVME_NUMBER_OF_QUEUES:
6022         result = (n->conf_ioqpairs - 1) | ((n->conf_ioqpairs - 1) << 16);
6023         trace_pci_nvme_getfeat_numq(result);
6024         break;
6025     case NVME_INTERRUPT_VECTOR_CONF:
6026         iv = dw11 & 0xffff;
6027         if (iv >= n->conf_ioqpairs + 1) {
6028             return NVME_INVALID_FIELD | NVME_DNR;
6029         }
6030 
6031         result = iv;
6032         if (iv == n->admin_cq.vector) {
6033             result |= NVME_INTVC_NOCOALESCING;
6034         }
6035         break;
6036     case NVME_FDP_MODE:
6037         endgrpid = dw11 & 0xff;
6038 
6039         if (endgrpid != 0x1) {
6040             return NVME_INVALID_FIELD | NVME_DNR;
6041         }
6042 
6043         ret = nvme_get_feature_fdp(n, endgrpid, &result);
6044         if (ret) {
6045             return ret;
6046         }
6047         goto out;
6048 
6049         break;
6050     default:
6051         result = nvme_feature_default[fid];
6052         break;
6053     }
6054 
6055 out:
6056     req->cqe.result = cpu_to_le32(result);
6057     return ret;
6058 }
6059 
6060 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeRequest *req)
6061 {
6062     uint16_t ret;
6063     uint64_t timestamp;
6064 
6065     ret = nvme_h2c(n, (uint8_t *)&timestamp, sizeof(timestamp), req);
6066     if (ret) {
6067         return ret;
6068     }
6069 
6070     nvme_set_timestamp(n, timestamp);
6071 
6072     return NVME_SUCCESS;
6073 }
6074 
6075 static uint16_t nvme_set_feature_fdp_events(NvmeCtrl *n, NvmeNamespace *ns,
6076                                             NvmeRequest *req)
6077 {
6078     NvmeCmd *cmd = &req->cmd;
6079     uint32_t cdw11 = le32_to_cpu(cmd->cdw11);
6080     uint16_t ph = cdw11 & 0xffff;
6081     uint8_t noet = (cdw11 >> 16) & 0xff;
6082     uint16_t ret, ruhid;
6083     uint8_t enable = le32_to_cpu(cmd->cdw12) & 0x1;
6084     uint8_t event_mask = 0;
6085     unsigned int i;
6086     g_autofree uint8_t *events = g_malloc0(noet);
6087     NvmeRuHandle *ruh = NULL;
6088 
6089     assert(ns);
6090 
6091     if (!n->subsys || !n->subsys->endgrp.fdp.enabled) {
6092         return NVME_FDP_DISABLED | NVME_DNR;
6093     }
6094 
6095     if (!nvme_ph_valid(ns, ph)) {
6096         return NVME_INVALID_FIELD | NVME_DNR;
6097     }
6098 
6099     ruhid = ns->fdp.phs[ph];
6100     ruh = &n->subsys->endgrp.fdp.ruhs[ruhid];
6101 
6102     ret = nvme_h2c(n, events, noet, req);
6103     if (ret) {
6104         return ret;
6105     }
6106 
6107     for (i = 0; i < noet; i++) {
6108         event_mask |= (1 << nvme_fdp_evf_shifts[events[i]]);
6109     }
6110 
6111     if (enable) {
6112         ruh->event_filter |= event_mask;
6113     } else {
6114         ruh->event_filter = ruh->event_filter & ~event_mask;
6115     }
6116 
6117     return NVME_SUCCESS;
6118 }
6119 
6120 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
6121 {
6122     NvmeNamespace *ns = NULL;
6123 
6124     NvmeCmd *cmd = &req->cmd;
6125     uint32_t dw10 = le32_to_cpu(cmd->cdw10);
6126     uint32_t dw11 = le32_to_cpu(cmd->cdw11);
6127     uint32_t nsid = le32_to_cpu(cmd->nsid);
6128     uint8_t fid = NVME_GETSETFEAT_FID(dw10);
6129     uint8_t save = NVME_SETFEAT_SAVE(dw10);
6130     uint16_t status;
6131     int i;
6132 
6133     trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
6134 
6135     if (save && !(nvme_feature_cap[fid] & NVME_FEAT_CAP_SAVE)) {
6136         return NVME_FID_NOT_SAVEABLE | NVME_DNR;
6137     }
6138 
6139     if (!nvme_feature_support[fid]) {
6140         return NVME_INVALID_FIELD | NVME_DNR;
6141     }
6142 
6143     if (nvme_feature_cap[fid] & NVME_FEAT_CAP_NS) {
6144         if (nsid != NVME_NSID_BROADCAST) {
6145             if (!nvme_nsid_valid(n, nsid)) {
6146                 return NVME_INVALID_NSID | NVME_DNR;
6147             }
6148 
6149             ns = nvme_ns(n, nsid);
6150             if (unlikely(!ns)) {
6151                 return NVME_INVALID_FIELD | NVME_DNR;
6152             }
6153         }
6154     } else if (nsid && nsid != NVME_NSID_BROADCAST) {
6155         if (!nvme_nsid_valid(n, nsid)) {
6156             return NVME_INVALID_NSID | NVME_DNR;
6157         }
6158 
6159         return NVME_FEAT_NOT_NS_SPEC | NVME_DNR;
6160     }
6161 
6162     if (!(nvme_feature_cap[fid] & NVME_FEAT_CAP_CHANGE)) {
6163         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6164     }
6165 
6166     switch (fid) {
6167     case NVME_TEMPERATURE_THRESHOLD:
6168         if (NVME_TEMP_TMPSEL(dw11) != NVME_TEMP_TMPSEL_COMPOSITE) {
6169             break;
6170         }
6171 
6172         switch (NVME_TEMP_THSEL(dw11)) {
6173         case NVME_TEMP_THSEL_OVER:
6174             n->features.temp_thresh_hi = NVME_TEMP_TMPTH(dw11);
6175             break;
6176         case NVME_TEMP_THSEL_UNDER:
6177             n->features.temp_thresh_low = NVME_TEMP_TMPTH(dw11);
6178             break;
6179         default:
6180             return NVME_INVALID_FIELD | NVME_DNR;
6181         }
6182 
6183         if ((n->temperature >= n->features.temp_thresh_hi) ||
6184             (n->temperature <= n->features.temp_thresh_low)) {
6185             nvme_smart_event(n, NVME_SMART_TEMPERATURE);
6186         }
6187 
6188         break;
6189     case NVME_ERROR_RECOVERY:
6190         if (nsid == NVME_NSID_BROADCAST) {
6191             for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6192                 ns = nvme_ns(n, i);
6193 
6194                 if (!ns) {
6195                     continue;
6196                 }
6197 
6198                 if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat)) {
6199                     ns->features.err_rec = dw11;
6200                 }
6201             }
6202 
6203             break;
6204         }
6205 
6206         assert(ns);
6207         if (NVME_ID_NS_NSFEAT_DULBE(ns->id_ns.nsfeat))  {
6208             ns->features.err_rec = dw11;
6209         }
6210         break;
6211     case NVME_VOLATILE_WRITE_CACHE:
6212         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6213             ns = nvme_ns(n, i);
6214             if (!ns) {
6215                 continue;
6216             }
6217 
6218             if (!(dw11 & 0x1) && blk_enable_write_cache(ns->blkconf.blk)) {
6219                 blk_flush(ns->blkconf.blk);
6220             }
6221 
6222             blk_set_enable_write_cache(ns->blkconf.blk, dw11 & 1);
6223         }
6224 
6225         break;
6226 
6227     case NVME_NUMBER_OF_QUEUES:
6228         if (n->qs_created) {
6229             return NVME_CMD_SEQ_ERROR | NVME_DNR;
6230         }
6231 
6232         /*
6233          * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
6234          * and NSQR.
6235          */
6236         if ((dw11 & 0xffff) == 0xffff || ((dw11 >> 16) & 0xffff) == 0xffff) {
6237             return NVME_INVALID_FIELD | NVME_DNR;
6238         }
6239 
6240         trace_pci_nvme_setfeat_numq((dw11 & 0xffff) + 1,
6241                                     ((dw11 >> 16) & 0xffff) + 1,
6242                                     n->conf_ioqpairs,
6243                                     n->conf_ioqpairs);
6244         req->cqe.result = cpu_to_le32((n->conf_ioqpairs - 1) |
6245                                       ((n->conf_ioqpairs - 1) << 16));
6246         break;
6247     case NVME_ASYNCHRONOUS_EVENT_CONF:
6248         n->features.async_config = dw11;
6249         break;
6250     case NVME_TIMESTAMP:
6251         return nvme_set_feature_timestamp(n, req);
6252     case NVME_HOST_BEHAVIOR_SUPPORT:
6253         status = nvme_h2c(n, (uint8_t *)&n->features.hbs,
6254                           sizeof(n->features.hbs), req);
6255         if (status) {
6256             return status;
6257         }
6258 
6259         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
6260             ns = nvme_ns(n, i);
6261 
6262             if (!ns) {
6263                 continue;
6264             }
6265 
6266             ns->id_ns.nlbaf = ns->nlbaf - 1;
6267             if (!n->features.hbs.lbafee) {
6268                 ns->id_ns.nlbaf = MIN(ns->id_ns.nlbaf, 15);
6269             }
6270         }
6271 
6272         return status;
6273     case NVME_COMMAND_SET_PROFILE:
6274         if (dw11 & 0x1ff) {
6275             trace_pci_nvme_err_invalid_iocsci(dw11 & 0x1ff);
6276             return NVME_CMD_SET_CMB_REJECTED | NVME_DNR;
6277         }
6278         break;
6279     case NVME_FDP_MODE:
6280         /* spec: abort with cmd seq err if there's one or more NS' in endgrp */
6281         return NVME_CMD_SEQ_ERROR | NVME_DNR;
6282     case NVME_FDP_EVENTS:
6283         return nvme_set_feature_fdp_events(n, ns, req);
6284     default:
6285         return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
6286     }
6287     return NVME_SUCCESS;
6288 }
6289 
6290 static uint16_t nvme_aer(NvmeCtrl *n, NvmeRequest *req)
6291 {
6292     trace_pci_nvme_aer(nvme_cid(req));
6293 
6294     if (n->outstanding_aers > n->params.aerl) {
6295         trace_pci_nvme_aer_aerl_exceeded();
6296         return NVME_AER_LIMIT_EXCEEDED;
6297     }
6298 
6299     n->aer_reqs[n->outstanding_aers] = req;
6300     n->outstanding_aers++;
6301 
6302     if (!QTAILQ_EMPTY(&n->aer_queue)) {
6303         nvme_process_aers(n);
6304     }
6305 
6306     return NVME_NO_COMPLETE;
6307 }
6308 
6309 static void nvme_update_dmrsl(NvmeCtrl *n)
6310 {
6311     int nsid;
6312 
6313     for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
6314         NvmeNamespace *ns = nvme_ns(n, nsid);
6315         if (!ns) {
6316             continue;
6317         }
6318 
6319         n->dmrsl = MIN_NON_ZERO(n->dmrsl,
6320                                 BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
6321     }
6322 }
6323 
6324 static void nvme_select_iocs_ns(NvmeCtrl *n, NvmeNamespace *ns)
6325 {
6326     uint32_t cc = ldl_le_p(&n->bar.cc);
6327 
6328     ns->iocs = nvme_cse_iocs_none;
6329     switch (ns->csi) {
6330     case NVME_CSI_NVM:
6331         if (NVME_CC_CSS(cc) != NVME_CC_CSS_ADMIN_ONLY) {
6332             ns->iocs = nvme_cse_iocs_nvm;
6333         }
6334         break;
6335     case NVME_CSI_ZONED:
6336         if (NVME_CC_CSS(cc) == NVME_CC_CSS_CSI) {
6337             ns->iocs = nvme_cse_iocs_zoned;
6338         } else if (NVME_CC_CSS(cc) == NVME_CC_CSS_NVM) {
6339             ns->iocs = nvme_cse_iocs_nvm;
6340         }
6341         break;
6342     }
6343 }
6344 
6345 static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
6346 {
6347     NvmeNamespace *ns;
6348     NvmeCtrl *ctrl;
6349     uint16_t list[NVME_CONTROLLER_LIST_SIZE] = {};
6350     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6351     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6352     uint8_t sel = dw10 & 0xf;
6353     uint16_t *nr_ids = &list[0];
6354     uint16_t *ids = &list[1];
6355     uint16_t ret;
6356     int i;
6357 
6358     trace_pci_nvme_ns_attachment(nvme_cid(req), dw10 & 0xf);
6359 
6360     if (!nvme_nsid_valid(n, nsid)) {
6361         return NVME_INVALID_NSID | NVME_DNR;
6362     }
6363 
6364     ns = nvme_subsys_ns(n->subsys, nsid);
6365     if (!ns) {
6366         return NVME_INVALID_FIELD | NVME_DNR;
6367     }
6368 
6369     ret = nvme_h2c(n, (uint8_t *)list, 4096, req);
6370     if (ret) {
6371         return ret;
6372     }
6373 
6374     if (!*nr_ids) {
6375         return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6376     }
6377 
6378     *nr_ids = MIN(*nr_ids, NVME_CONTROLLER_LIST_SIZE - 1);
6379     for (i = 0; i < *nr_ids; i++) {
6380         ctrl = nvme_subsys_ctrl(n->subsys, ids[i]);
6381         if (!ctrl) {
6382             return NVME_NS_CTRL_LIST_INVALID | NVME_DNR;
6383         }
6384 
6385         switch (sel) {
6386         case NVME_NS_ATTACHMENT_ATTACH:
6387             if (nvme_ns(ctrl, nsid)) {
6388                 return NVME_NS_ALREADY_ATTACHED | NVME_DNR;
6389             }
6390 
6391             if (ns->attached && !ns->params.shared) {
6392                 return NVME_NS_PRIVATE | NVME_DNR;
6393             }
6394 
6395             nvme_attach_ns(ctrl, ns);
6396             nvme_select_iocs_ns(ctrl, ns);
6397 
6398             break;
6399 
6400         case NVME_NS_ATTACHMENT_DETACH:
6401             if (!nvme_ns(ctrl, nsid)) {
6402                 return NVME_NS_NOT_ATTACHED | NVME_DNR;
6403             }
6404 
6405             ctrl->namespaces[nsid] = NULL;
6406             ns->attached--;
6407 
6408             nvme_update_dmrsl(ctrl);
6409 
6410             break;
6411 
6412         default:
6413             return NVME_INVALID_FIELD | NVME_DNR;
6414         }
6415 
6416         /*
6417          * Add namespace id to the changed namespace id list for event clearing
6418          * via Get Log Page command.
6419          */
6420         if (!test_and_set_bit(nsid, ctrl->changed_nsids)) {
6421             nvme_enqueue_event(ctrl, NVME_AER_TYPE_NOTICE,
6422                                NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED,
6423                                NVME_LOG_CHANGED_NSLIST);
6424         }
6425     }
6426 
6427     return NVME_SUCCESS;
6428 }
6429 
6430 typedef struct NvmeFormatAIOCB {
6431     BlockAIOCB common;
6432     BlockAIOCB *aiocb;
6433     NvmeRequest *req;
6434     int ret;
6435 
6436     NvmeNamespace *ns;
6437     uint32_t nsid;
6438     bool broadcast;
6439     int64_t offset;
6440 
6441     uint8_t lbaf;
6442     uint8_t mset;
6443     uint8_t pi;
6444     uint8_t pil;
6445 } NvmeFormatAIOCB;
6446 
6447 static void nvme_format_cancel(BlockAIOCB *aiocb)
6448 {
6449     NvmeFormatAIOCB *iocb = container_of(aiocb, NvmeFormatAIOCB, common);
6450 
6451     iocb->ret = -ECANCELED;
6452 
6453     if (iocb->aiocb) {
6454         blk_aio_cancel_async(iocb->aiocb);
6455         iocb->aiocb = NULL;
6456     }
6457 }
6458 
6459 static const AIOCBInfo nvme_format_aiocb_info = {
6460     .aiocb_size = sizeof(NvmeFormatAIOCB),
6461     .cancel_async = nvme_format_cancel,
6462     .get_aio_context = nvme_get_aio_context,
6463 };
6464 
6465 static void nvme_format_set(NvmeNamespace *ns, uint8_t lbaf, uint8_t mset,
6466                             uint8_t pi, uint8_t pil)
6467 {
6468     uint8_t lbafl = lbaf & 0xf;
6469     uint8_t lbafu = lbaf >> 4;
6470 
6471     trace_pci_nvme_format_set(ns->params.nsid, lbaf, mset, pi, pil);
6472 
6473     ns->id_ns.dps = (pil << 3) | pi;
6474     ns->id_ns.flbas = (lbafu << 5) | (mset << 4) | lbafl;
6475 
6476     nvme_ns_init_format(ns);
6477 }
6478 
6479 static void nvme_do_format(NvmeFormatAIOCB *iocb);
6480 
6481 static void nvme_format_ns_cb(void *opaque, int ret)
6482 {
6483     NvmeFormatAIOCB *iocb = opaque;
6484     NvmeNamespace *ns = iocb->ns;
6485     int bytes;
6486 
6487     if (iocb->ret < 0) {
6488         goto done;
6489     } else if (ret < 0) {
6490         iocb->ret = ret;
6491         goto done;
6492     }
6493 
6494     assert(ns);
6495 
6496     if (iocb->offset < ns->size) {
6497         bytes = MIN(BDRV_REQUEST_MAX_BYTES, ns->size - iocb->offset);
6498 
6499         iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, iocb->offset,
6500                                             bytes, BDRV_REQ_MAY_UNMAP,
6501                                             nvme_format_ns_cb, iocb);
6502 
6503         iocb->offset += bytes;
6504         return;
6505     }
6506 
6507     nvme_format_set(ns, iocb->lbaf, iocb->mset, iocb->pi, iocb->pil);
6508     ns->status = 0x0;
6509     iocb->ns = NULL;
6510     iocb->offset = 0;
6511 
6512 done:
6513     nvme_do_format(iocb);
6514 }
6515 
6516 static uint16_t nvme_format_check(NvmeNamespace *ns, uint8_t lbaf, uint8_t pi)
6517 {
6518     if (ns->params.zoned) {
6519         return NVME_INVALID_FORMAT | NVME_DNR;
6520     }
6521 
6522     if (lbaf > ns->id_ns.nlbaf) {
6523         return NVME_INVALID_FORMAT | NVME_DNR;
6524     }
6525 
6526     if (pi && (ns->id_ns.lbaf[lbaf].ms < nvme_pi_tuple_size(ns))) {
6527         return NVME_INVALID_FORMAT | NVME_DNR;
6528     }
6529 
6530     if (pi && pi > NVME_ID_NS_DPS_TYPE_3) {
6531         return NVME_INVALID_FIELD | NVME_DNR;
6532     }
6533 
6534     return NVME_SUCCESS;
6535 }
6536 
6537 static void nvme_do_format(NvmeFormatAIOCB *iocb)
6538 {
6539     NvmeRequest *req = iocb->req;
6540     NvmeCtrl *n = nvme_ctrl(req);
6541     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6542     uint8_t lbaf = dw10 & 0xf;
6543     uint8_t pi = (dw10 >> 5) & 0x7;
6544     uint16_t status;
6545     int i;
6546 
6547     if (iocb->ret < 0) {
6548         goto done;
6549     }
6550 
6551     if (iocb->broadcast) {
6552         for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
6553             iocb->ns = nvme_ns(n, i);
6554             if (iocb->ns) {
6555                 iocb->nsid = i;
6556                 break;
6557             }
6558         }
6559     }
6560 
6561     if (!iocb->ns) {
6562         goto done;
6563     }
6564 
6565     status = nvme_format_check(iocb->ns, lbaf, pi);
6566     if (status) {
6567         req->status = status;
6568         goto done;
6569     }
6570 
6571     iocb->ns->status = NVME_FORMAT_IN_PROGRESS;
6572     nvme_format_ns_cb(iocb, 0);
6573     return;
6574 
6575 done:
6576     iocb->common.cb(iocb->common.opaque, iocb->ret);
6577     qemu_aio_unref(iocb);
6578 }
6579 
6580 static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
6581 {
6582     NvmeFormatAIOCB *iocb;
6583     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6584     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6585     uint8_t lbaf = dw10 & 0xf;
6586     uint8_t mset = (dw10 >> 4) & 0x1;
6587     uint8_t pi = (dw10 >> 5) & 0x7;
6588     uint8_t pil = (dw10 >> 8) & 0x1;
6589     uint8_t lbafu = (dw10 >> 12) & 0x3;
6590     uint16_t status;
6591 
6592     iocb = qemu_aio_get(&nvme_format_aiocb_info, NULL, nvme_misc_cb, req);
6593 
6594     iocb->req = req;
6595     iocb->ret = 0;
6596     iocb->ns = NULL;
6597     iocb->nsid = 0;
6598     iocb->lbaf = lbaf;
6599     iocb->mset = mset;
6600     iocb->pi = pi;
6601     iocb->pil = pil;
6602     iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
6603     iocb->offset = 0;
6604 
6605     if (n->features.hbs.lbafee) {
6606         iocb->lbaf |= lbafu << 4;
6607     }
6608 
6609     if (!iocb->broadcast) {
6610         if (!nvme_nsid_valid(n, nsid)) {
6611             status = NVME_INVALID_NSID | NVME_DNR;
6612             goto out;
6613         }
6614 
6615         iocb->ns = nvme_ns(n, nsid);
6616         if (!iocb->ns) {
6617             status = NVME_INVALID_FIELD | NVME_DNR;
6618             goto out;
6619         }
6620     }
6621 
6622     req->aiocb = &iocb->common;
6623     nvme_do_format(iocb);
6624 
6625     return NVME_NO_COMPLETE;
6626 
6627 out:
6628     qemu_aio_unref(iocb);
6629 
6630     return status;
6631 }
6632 
6633 static void nvme_get_virt_res_num(NvmeCtrl *n, uint8_t rt, int *num_total,
6634                                   int *num_prim, int *num_sec)
6635 {
6636     *num_total = le32_to_cpu(rt ?
6637                              n->pri_ctrl_cap.vifrt : n->pri_ctrl_cap.vqfrt);
6638     *num_prim = le16_to_cpu(rt ?
6639                             n->pri_ctrl_cap.virfap : n->pri_ctrl_cap.vqrfap);
6640     *num_sec = le16_to_cpu(rt ? n->pri_ctrl_cap.virfa : n->pri_ctrl_cap.vqrfa);
6641 }
6642 
6643 static uint16_t nvme_assign_virt_res_to_prim(NvmeCtrl *n, NvmeRequest *req,
6644                                              uint16_t cntlid, uint8_t rt,
6645                                              int nr)
6646 {
6647     int num_total, num_prim, num_sec;
6648 
6649     if (cntlid != n->cntlid) {
6650         return NVME_INVALID_CTRL_ID | NVME_DNR;
6651     }
6652 
6653     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6654 
6655     if (nr > num_total) {
6656         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6657     }
6658 
6659     if (nr > num_total - num_sec) {
6660         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6661     }
6662 
6663     if (rt) {
6664         n->next_pri_ctrl_cap.virfap = cpu_to_le16(nr);
6665     } else {
6666         n->next_pri_ctrl_cap.vqrfap = cpu_to_le16(nr);
6667     }
6668 
6669     req->cqe.result = cpu_to_le32(nr);
6670     return req->status;
6671 }
6672 
6673 static void nvme_update_virt_res(NvmeCtrl *n, NvmeSecCtrlEntry *sctrl,
6674                                  uint8_t rt, int nr)
6675 {
6676     int prev_nr, prev_total;
6677 
6678     if (rt) {
6679         prev_nr = le16_to_cpu(sctrl->nvi);
6680         prev_total = le32_to_cpu(n->pri_ctrl_cap.virfa);
6681         sctrl->nvi = cpu_to_le16(nr);
6682         n->pri_ctrl_cap.virfa = cpu_to_le32(prev_total + nr - prev_nr);
6683     } else {
6684         prev_nr = le16_to_cpu(sctrl->nvq);
6685         prev_total = le32_to_cpu(n->pri_ctrl_cap.vqrfa);
6686         sctrl->nvq = cpu_to_le16(nr);
6687         n->pri_ctrl_cap.vqrfa = cpu_to_le32(prev_total + nr - prev_nr);
6688     }
6689 }
6690 
6691 static uint16_t nvme_assign_virt_res_to_sec(NvmeCtrl *n, NvmeRequest *req,
6692                                             uint16_t cntlid, uint8_t rt, int nr)
6693 {
6694     int num_total, num_prim, num_sec, num_free, diff, limit;
6695     NvmeSecCtrlEntry *sctrl;
6696 
6697     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6698     if (!sctrl) {
6699         return NVME_INVALID_CTRL_ID | NVME_DNR;
6700     }
6701 
6702     if (sctrl->scs) {
6703         return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6704     }
6705 
6706     limit = le16_to_cpu(rt ? n->pri_ctrl_cap.vifrsm : n->pri_ctrl_cap.vqfrsm);
6707     if (nr > limit) {
6708         return NVME_INVALID_NUM_RESOURCES | NVME_DNR;
6709     }
6710 
6711     nvme_get_virt_res_num(n, rt, &num_total, &num_prim, &num_sec);
6712     num_free = num_total - num_prim - num_sec;
6713     diff = nr - le16_to_cpu(rt ? sctrl->nvi : sctrl->nvq);
6714 
6715     if (diff > num_free) {
6716         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6717     }
6718 
6719     nvme_update_virt_res(n, sctrl, rt, nr);
6720     req->cqe.result = cpu_to_le32(nr);
6721 
6722     return req->status;
6723 }
6724 
6725 static uint16_t nvme_virt_set_state(NvmeCtrl *n, uint16_t cntlid, bool online)
6726 {
6727     PCIDevice *pci = PCI_DEVICE(n);
6728     NvmeCtrl *sn = NULL;
6729     NvmeSecCtrlEntry *sctrl;
6730     int vf_index;
6731 
6732     sctrl = nvme_sctrl_for_cntlid(n, cntlid);
6733     if (!sctrl) {
6734         return NVME_INVALID_CTRL_ID | NVME_DNR;
6735     }
6736 
6737     if (!pci_is_vf(pci)) {
6738         vf_index = le16_to_cpu(sctrl->vfn) - 1;
6739         sn = NVME(pcie_sriov_get_vf_at_index(pci, vf_index));
6740     }
6741 
6742     if (online) {
6743         if (!sctrl->nvi || (le16_to_cpu(sctrl->nvq) < 2) || !sn) {
6744             return NVME_INVALID_SEC_CTRL_STATE | NVME_DNR;
6745         }
6746 
6747         if (!sctrl->scs) {
6748             sctrl->scs = 0x1;
6749             nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6750         }
6751     } else {
6752         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_INTERRUPT, 0);
6753         nvme_update_virt_res(n, sctrl, NVME_VIRT_RES_QUEUE, 0);
6754 
6755         if (sctrl->scs) {
6756             sctrl->scs = 0x0;
6757             if (sn) {
6758                 nvme_ctrl_reset(sn, NVME_RESET_FUNCTION);
6759             }
6760         }
6761     }
6762 
6763     return NVME_SUCCESS;
6764 }
6765 
6766 static uint16_t nvme_virt_mngmt(NvmeCtrl *n, NvmeRequest *req)
6767 {
6768     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6769     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
6770     uint8_t act = dw10 & 0xf;
6771     uint8_t rt = (dw10 >> 8) & 0x7;
6772     uint16_t cntlid = (dw10 >> 16) & 0xffff;
6773     int nr = dw11 & 0xffff;
6774 
6775     trace_pci_nvme_virt_mngmt(nvme_cid(req), act, cntlid, rt ? "VI" : "VQ", nr);
6776 
6777     if (rt != NVME_VIRT_RES_QUEUE && rt != NVME_VIRT_RES_INTERRUPT) {
6778         return NVME_INVALID_RESOURCE_ID | NVME_DNR;
6779     }
6780 
6781     switch (act) {
6782     case NVME_VIRT_MNGMT_ACTION_SEC_ASSIGN:
6783         return nvme_assign_virt_res_to_sec(n, req, cntlid, rt, nr);
6784     case NVME_VIRT_MNGMT_ACTION_PRM_ALLOC:
6785         return nvme_assign_virt_res_to_prim(n, req, cntlid, rt, nr);
6786     case NVME_VIRT_MNGMT_ACTION_SEC_ONLINE:
6787         return nvme_virt_set_state(n, cntlid, true);
6788     case NVME_VIRT_MNGMT_ACTION_SEC_OFFLINE:
6789         return nvme_virt_set_state(n, cntlid, false);
6790     default:
6791         return NVME_INVALID_FIELD | NVME_DNR;
6792     }
6793 }
6794 
6795 static uint16_t nvme_dbbuf_config(NvmeCtrl *n, const NvmeRequest *req)
6796 {
6797     PCIDevice *pci = PCI_DEVICE(n);
6798     uint64_t dbs_addr = le64_to_cpu(req->cmd.dptr.prp1);
6799     uint64_t eis_addr = le64_to_cpu(req->cmd.dptr.prp2);
6800     int i;
6801 
6802     /* Address should be page aligned */
6803     if (dbs_addr & (n->page_size - 1) || eis_addr & (n->page_size - 1)) {
6804         return NVME_INVALID_FIELD | NVME_DNR;
6805     }
6806 
6807     /* Save shadow buffer base addr for use during queue creation */
6808     n->dbbuf_dbs = dbs_addr;
6809     n->dbbuf_eis = eis_addr;
6810     n->dbbuf_enabled = true;
6811 
6812     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
6813         NvmeSQueue *sq = n->sq[i];
6814         NvmeCQueue *cq = n->cq[i];
6815 
6816         if (sq) {
6817             /*
6818              * CAP.DSTRD is 0, so offset of ith sq db_addr is (i<<3)
6819              * nvme_process_db() uses this hard-coded way to calculate
6820              * doorbell offsets. Be consistent with that here.
6821              */
6822             sq->db_addr = dbs_addr + (i << 3);
6823             sq->ei_addr = eis_addr + (i << 3);
6824             stl_le_pci_dma(pci, sq->db_addr, sq->tail, MEMTXATTRS_UNSPECIFIED);
6825 
6826             if (n->params.ioeventfd && sq->sqid != 0) {
6827                 if (!nvme_init_sq_ioeventfd(sq)) {
6828                     sq->ioeventfd_enabled = true;
6829                 }
6830             }
6831         }
6832 
6833         if (cq) {
6834             /* CAP.DSTRD is 0, so offset of ith cq db_addr is (i<<3)+(1<<2) */
6835             cq->db_addr = dbs_addr + (i << 3) + (1 << 2);
6836             cq->ei_addr = eis_addr + (i << 3) + (1 << 2);
6837             stl_le_pci_dma(pci, cq->db_addr, cq->head, MEMTXATTRS_UNSPECIFIED);
6838 
6839             if (n->params.ioeventfd && cq->cqid != 0) {
6840                 if (!nvme_init_cq_ioeventfd(cq)) {
6841                     cq->ioeventfd_enabled = true;
6842                 }
6843             }
6844         }
6845     }
6846 
6847     trace_pci_nvme_dbbuf_config(dbs_addr, eis_addr);
6848 
6849     return NVME_SUCCESS;
6850 }
6851 
6852 static uint16_t nvme_directive_send(NvmeCtrl *n, NvmeRequest *req)
6853 {
6854     return NVME_INVALID_FIELD | NVME_DNR;
6855 }
6856 
6857 static uint16_t nvme_directive_receive(NvmeCtrl *n, NvmeRequest *req)
6858 {
6859     NvmeNamespace *ns;
6860     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
6861     uint32_t dw11 = le32_to_cpu(req->cmd.cdw11);
6862     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
6863     uint8_t doper, dtype;
6864     uint32_t numd, trans_len;
6865     NvmeDirectiveIdentify id = {
6866         .supported = 1 << NVME_DIRECTIVE_IDENTIFY,
6867         .enabled = 1 << NVME_DIRECTIVE_IDENTIFY,
6868     };
6869 
6870     numd = dw10 + 1;
6871     doper = dw11 & 0xff;
6872     dtype = (dw11 >> 8) & 0xff;
6873 
6874     trans_len = MIN(sizeof(NvmeDirectiveIdentify), numd << 2);
6875 
6876     if (nsid == NVME_NSID_BROADCAST || dtype != NVME_DIRECTIVE_IDENTIFY ||
6877         doper != NVME_DIRECTIVE_RETURN_PARAMS) {
6878         return NVME_INVALID_FIELD | NVME_DNR;
6879     }
6880 
6881     ns = nvme_ns(n, nsid);
6882     if (!ns) {
6883         return NVME_INVALID_FIELD | NVME_DNR;
6884     }
6885 
6886     switch (dtype) {
6887     case NVME_DIRECTIVE_IDENTIFY:
6888         switch (doper) {
6889         case NVME_DIRECTIVE_RETURN_PARAMS:
6890             if (ns->endgrp->fdp.enabled) {
6891                 id.supported |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6892                 id.enabled |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6893                 id.persistent |= 1 << NVME_DIRECTIVE_DATA_PLACEMENT;
6894             }
6895 
6896             return nvme_c2h(n, (uint8_t *)&id, trans_len, req);
6897 
6898         default:
6899             return NVME_INVALID_FIELD | NVME_DNR;
6900         }
6901 
6902     default:
6903         return NVME_INVALID_FIELD;
6904     }
6905 }
6906 
6907 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
6908 {
6909     trace_pci_nvme_admin_cmd(nvme_cid(req), nvme_sqid(req), req->cmd.opcode,
6910                              nvme_adm_opc_str(req->cmd.opcode));
6911 
6912     if (!(nvme_cse_acs[req->cmd.opcode] & NVME_CMD_EFF_CSUPP)) {
6913         trace_pci_nvme_err_invalid_admin_opc(req->cmd.opcode);
6914         return NVME_INVALID_OPCODE | NVME_DNR;
6915     }
6916 
6917     /* SGLs shall not be used for Admin commands in NVMe over PCIe */
6918     if (NVME_CMD_FLAGS_PSDT(req->cmd.flags) != NVME_PSDT_PRP) {
6919         return NVME_INVALID_FIELD | NVME_DNR;
6920     }
6921 
6922     if (NVME_CMD_FLAGS_FUSE(req->cmd.flags)) {
6923         return NVME_INVALID_FIELD;
6924     }
6925 
6926     switch (req->cmd.opcode) {
6927     case NVME_ADM_CMD_DELETE_SQ:
6928         return nvme_del_sq(n, req);
6929     case NVME_ADM_CMD_CREATE_SQ:
6930         return nvme_create_sq(n, req);
6931     case NVME_ADM_CMD_GET_LOG_PAGE:
6932         return nvme_get_log(n, req);
6933     case NVME_ADM_CMD_DELETE_CQ:
6934         return nvme_del_cq(n, req);
6935     case NVME_ADM_CMD_CREATE_CQ:
6936         return nvme_create_cq(n, req);
6937     case NVME_ADM_CMD_IDENTIFY:
6938         return nvme_identify(n, req);
6939     case NVME_ADM_CMD_ABORT:
6940         return nvme_abort(n, req);
6941     case NVME_ADM_CMD_SET_FEATURES:
6942         return nvme_set_feature(n, req);
6943     case NVME_ADM_CMD_GET_FEATURES:
6944         return nvme_get_feature(n, req);
6945     case NVME_ADM_CMD_ASYNC_EV_REQ:
6946         return nvme_aer(n, req);
6947     case NVME_ADM_CMD_NS_ATTACHMENT:
6948         return nvme_ns_attachment(n, req);
6949     case NVME_ADM_CMD_VIRT_MNGMT:
6950         return nvme_virt_mngmt(n, req);
6951     case NVME_ADM_CMD_DBBUF_CONFIG:
6952         return nvme_dbbuf_config(n, req);
6953     case NVME_ADM_CMD_FORMAT_NVM:
6954         return nvme_format(n, req);
6955     case NVME_ADM_CMD_DIRECTIVE_SEND:
6956         return nvme_directive_send(n, req);
6957     case NVME_ADM_CMD_DIRECTIVE_RECV:
6958         return nvme_directive_receive(n, req);
6959     default:
6960         assert(false);
6961     }
6962 
6963     return NVME_INVALID_OPCODE | NVME_DNR;
6964 }
6965 
6966 static void nvme_update_sq_eventidx(const NvmeSQueue *sq)
6967 {
6968     trace_pci_nvme_update_sq_eventidx(sq->sqid, sq->tail);
6969 
6970     stl_le_pci_dma(PCI_DEVICE(sq->ctrl), sq->ei_addr, sq->tail,
6971                    MEMTXATTRS_UNSPECIFIED);
6972 }
6973 
6974 static void nvme_update_sq_tail(NvmeSQueue *sq)
6975 {
6976     ldl_le_pci_dma(PCI_DEVICE(sq->ctrl), sq->db_addr, &sq->tail,
6977                    MEMTXATTRS_UNSPECIFIED);
6978 
6979     trace_pci_nvme_update_sq_tail(sq->sqid, sq->tail);
6980 }
6981 
6982 static void nvme_process_sq(void *opaque)
6983 {
6984     NvmeSQueue *sq = opaque;
6985     NvmeCtrl *n = sq->ctrl;
6986     NvmeCQueue *cq = n->cq[sq->cqid];
6987 
6988     uint16_t status;
6989     hwaddr addr;
6990     NvmeCmd cmd;
6991     NvmeRequest *req;
6992 
6993     if (n->dbbuf_enabled) {
6994         nvme_update_sq_tail(sq);
6995     }
6996 
6997     while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
6998         addr = sq->dma_addr + sq->head * n->sqe_size;
6999         if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
7000             trace_pci_nvme_err_addr_read(addr);
7001             trace_pci_nvme_err_cfs();
7002             stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
7003             break;
7004         }
7005         nvme_inc_sq_head(sq);
7006 
7007         req = QTAILQ_FIRST(&sq->req_list);
7008         QTAILQ_REMOVE(&sq->req_list, req, entry);
7009         QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
7010         nvme_req_clear(req);
7011         req->cqe.cid = cmd.cid;
7012         memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
7013 
7014         status = sq->sqid ? nvme_io_cmd(n, req) :
7015             nvme_admin_cmd(n, req);
7016         if (status != NVME_NO_COMPLETE) {
7017             req->status = status;
7018             nvme_enqueue_req_completion(cq, req);
7019         }
7020 
7021         if (n->dbbuf_enabled) {
7022             nvme_update_sq_eventidx(sq);
7023             nvme_update_sq_tail(sq);
7024         }
7025     }
7026 }
7027 
7028 static void nvme_update_msixcap_ts(PCIDevice *pci_dev, uint32_t table_size)
7029 {
7030     uint8_t *config;
7031 
7032     if (!msix_present(pci_dev)) {
7033         return;
7034     }
7035 
7036     assert(table_size > 0 && table_size <= pci_dev->msix_entries_nr);
7037 
7038     config = pci_dev->config + pci_dev->msix_cap;
7039     pci_set_word_by_mask(config + PCI_MSIX_FLAGS, PCI_MSIX_FLAGS_QSIZE,
7040                          table_size - 1);
7041 }
7042 
7043 static void nvme_activate_virt_res(NvmeCtrl *n)
7044 {
7045     PCIDevice *pci_dev = PCI_DEVICE(n);
7046     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7047     NvmeSecCtrlEntry *sctrl;
7048 
7049     /* -1 to account for the admin queue */
7050     if (pci_is_vf(pci_dev)) {
7051         sctrl = nvme_sctrl(n);
7052         cap->vqprt = sctrl->nvq;
7053         cap->viprt = sctrl->nvi;
7054         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7055         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7056     } else {
7057         cap->vqrfap = n->next_pri_ctrl_cap.vqrfap;
7058         cap->virfap = n->next_pri_ctrl_cap.virfap;
7059         n->conf_ioqpairs = le16_to_cpu(cap->vqprt) +
7060                            le16_to_cpu(cap->vqrfap) - 1;
7061         n->conf_msix_qsize = le16_to_cpu(cap->viprt) +
7062                              le16_to_cpu(cap->virfap);
7063     }
7064 }
7065 
7066 static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst)
7067 {
7068     PCIDevice *pci_dev = PCI_DEVICE(n);
7069     NvmeSecCtrlEntry *sctrl;
7070     NvmeNamespace *ns;
7071     int i;
7072 
7073     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7074         ns = nvme_ns(n, i);
7075         if (!ns) {
7076             continue;
7077         }
7078 
7079         nvme_ns_drain(ns);
7080     }
7081 
7082     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7083         if (n->sq[i] != NULL) {
7084             nvme_free_sq(n->sq[i], n);
7085         }
7086     }
7087     for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
7088         if (n->cq[i] != NULL) {
7089             nvme_free_cq(n->cq[i], n);
7090         }
7091     }
7092 
7093     while (!QTAILQ_EMPTY(&n->aer_queue)) {
7094         NvmeAsyncEvent *event = QTAILQ_FIRST(&n->aer_queue);
7095         QTAILQ_REMOVE(&n->aer_queue, event, entry);
7096         g_free(event);
7097     }
7098 
7099     if (n->params.sriov_max_vfs) {
7100         if (!pci_is_vf(pci_dev)) {
7101             for (i = 0; i < n->sec_ctrl_list.numcntl; i++) {
7102                 sctrl = &n->sec_ctrl_list.sec[i];
7103                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
7104             }
7105 
7106             if (rst != NVME_RESET_CONTROLLER) {
7107                 pcie_sriov_pf_disable_vfs(pci_dev);
7108             }
7109         }
7110 
7111         if (rst != NVME_RESET_CONTROLLER) {
7112             nvme_activate_virt_res(n);
7113         }
7114     }
7115 
7116     n->aer_queued = 0;
7117     n->aer_mask = 0;
7118     n->outstanding_aers = 0;
7119     n->qs_created = false;
7120 
7121     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
7122 
7123     if (pci_is_vf(pci_dev)) {
7124         sctrl = nvme_sctrl(n);
7125 
7126         stl_le_p(&n->bar.csts, sctrl->scs ? 0 : NVME_CSTS_FAILED);
7127     } else {
7128         stl_le_p(&n->bar.csts, 0);
7129     }
7130 
7131     stl_le_p(&n->bar.intms, 0);
7132     stl_le_p(&n->bar.intmc, 0);
7133     stl_le_p(&n->bar.cc, 0);
7134 
7135     n->dbbuf_dbs = 0;
7136     n->dbbuf_eis = 0;
7137     n->dbbuf_enabled = false;
7138 }
7139 
7140 static void nvme_ctrl_shutdown(NvmeCtrl *n)
7141 {
7142     NvmeNamespace *ns;
7143     int i;
7144 
7145     if (n->pmr.dev) {
7146         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7147     }
7148 
7149     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7150         ns = nvme_ns(n, i);
7151         if (!ns) {
7152             continue;
7153         }
7154 
7155         nvme_ns_shutdown(ns);
7156     }
7157 }
7158 
7159 static void nvme_select_iocs(NvmeCtrl *n)
7160 {
7161     NvmeNamespace *ns;
7162     int i;
7163 
7164     for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
7165         ns = nvme_ns(n, i);
7166         if (!ns) {
7167             continue;
7168         }
7169 
7170         nvme_select_iocs_ns(n, ns);
7171     }
7172 }
7173 
7174 static int nvme_start_ctrl(NvmeCtrl *n)
7175 {
7176     uint64_t cap = ldq_le_p(&n->bar.cap);
7177     uint32_t cc = ldl_le_p(&n->bar.cc);
7178     uint32_t aqa = ldl_le_p(&n->bar.aqa);
7179     uint64_t asq = ldq_le_p(&n->bar.asq);
7180     uint64_t acq = ldq_le_p(&n->bar.acq);
7181     uint32_t page_bits = NVME_CC_MPS(cc) + 12;
7182     uint32_t page_size = 1 << page_bits;
7183     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
7184 
7185     if (pci_is_vf(PCI_DEVICE(n)) && !sctrl->scs) {
7186         trace_pci_nvme_err_startfail_virt_state(le16_to_cpu(sctrl->nvi),
7187                                                 le16_to_cpu(sctrl->nvq));
7188         return -1;
7189     }
7190     if (unlikely(n->cq[0])) {
7191         trace_pci_nvme_err_startfail_cq();
7192         return -1;
7193     }
7194     if (unlikely(n->sq[0])) {
7195         trace_pci_nvme_err_startfail_sq();
7196         return -1;
7197     }
7198     if (unlikely(asq & (page_size - 1))) {
7199         trace_pci_nvme_err_startfail_asq_misaligned(asq);
7200         return -1;
7201     }
7202     if (unlikely(acq & (page_size - 1))) {
7203         trace_pci_nvme_err_startfail_acq_misaligned(acq);
7204         return -1;
7205     }
7206     if (unlikely(!(NVME_CAP_CSS(cap) & (1 << NVME_CC_CSS(cc))))) {
7207         trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc));
7208         return -1;
7209     }
7210     if (unlikely(NVME_CC_MPS(cc) < NVME_CAP_MPSMIN(cap))) {
7211         trace_pci_nvme_err_startfail_page_too_small(
7212                     NVME_CC_MPS(cc),
7213                     NVME_CAP_MPSMIN(cap));
7214         return -1;
7215     }
7216     if (unlikely(NVME_CC_MPS(cc) >
7217                  NVME_CAP_MPSMAX(cap))) {
7218         trace_pci_nvme_err_startfail_page_too_large(
7219                     NVME_CC_MPS(cc),
7220                     NVME_CAP_MPSMAX(cap));
7221         return -1;
7222     }
7223     if (unlikely(NVME_CC_IOCQES(cc) <
7224                  NVME_CTRL_CQES_MIN(n->id_ctrl.cqes))) {
7225         trace_pci_nvme_err_startfail_cqent_too_small(
7226                     NVME_CC_IOCQES(cc),
7227                     NVME_CTRL_CQES_MIN(cap));
7228         return -1;
7229     }
7230     if (unlikely(NVME_CC_IOCQES(cc) >
7231                  NVME_CTRL_CQES_MAX(n->id_ctrl.cqes))) {
7232         trace_pci_nvme_err_startfail_cqent_too_large(
7233                     NVME_CC_IOCQES(cc),
7234                     NVME_CTRL_CQES_MAX(cap));
7235         return -1;
7236     }
7237     if (unlikely(NVME_CC_IOSQES(cc) <
7238                  NVME_CTRL_SQES_MIN(n->id_ctrl.sqes))) {
7239         trace_pci_nvme_err_startfail_sqent_too_small(
7240                     NVME_CC_IOSQES(cc),
7241                     NVME_CTRL_SQES_MIN(cap));
7242         return -1;
7243     }
7244     if (unlikely(NVME_CC_IOSQES(cc) >
7245                  NVME_CTRL_SQES_MAX(n->id_ctrl.sqes))) {
7246         trace_pci_nvme_err_startfail_sqent_too_large(
7247                     NVME_CC_IOSQES(cc),
7248                     NVME_CTRL_SQES_MAX(cap));
7249         return -1;
7250     }
7251     if (unlikely(!NVME_AQA_ASQS(aqa))) {
7252         trace_pci_nvme_err_startfail_asqent_sz_zero();
7253         return -1;
7254     }
7255     if (unlikely(!NVME_AQA_ACQS(aqa))) {
7256         trace_pci_nvme_err_startfail_acqent_sz_zero();
7257         return -1;
7258     }
7259 
7260     n->page_bits = page_bits;
7261     n->page_size = page_size;
7262     n->max_prp_ents = n->page_size / sizeof(uint64_t);
7263     n->cqe_size = 1 << NVME_CC_IOCQES(cc);
7264     n->sqe_size = 1 << NVME_CC_IOSQES(cc);
7265     nvme_init_cq(&n->admin_cq, n, acq, 0, 0, NVME_AQA_ACQS(aqa) + 1, 1);
7266     nvme_init_sq(&n->admin_sq, n, asq, 0, 0, NVME_AQA_ASQS(aqa) + 1);
7267 
7268     nvme_set_timestamp(n, 0ULL);
7269 
7270     nvme_select_iocs(n);
7271 
7272     return 0;
7273 }
7274 
7275 static void nvme_cmb_enable_regs(NvmeCtrl *n)
7276 {
7277     uint32_t cmbloc = ldl_le_p(&n->bar.cmbloc);
7278     uint32_t cmbsz = ldl_le_p(&n->bar.cmbsz);
7279 
7280     NVME_CMBLOC_SET_CDPCILS(cmbloc, 1);
7281     NVME_CMBLOC_SET_CDPMLS(cmbloc, 1);
7282     NVME_CMBLOC_SET_BIR(cmbloc, NVME_CMB_BIR);
7283     stl_le_p(&n->bar.cmbloc, cmbloc);
7284 
7285     NVME_CMBSZ_SET_SQS(cmbsz, 1);
7286     NVME_CMBSZ_SET_CQS(cmbsz, 0);
7287     NVME_CMBSZ_SET_LISTS(cmbsz, 1);
7288     NVME_CMBSZ_SET_RDS(cmbsz, 1);
7289     NVME_CMBSZ_SET_WDS(cmbsz, 1);
7290     NVME_CMBSZ_SET_SZU(cmbsz, 2); /* MBs */
7291     NVME_CMBSZ_SET_SZ(cmbsz, n->params.cmb_size_mb);
7292     stl_le_p(&n->bar.cmbsz, cmbsz);
7293 }
7294 
7295 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
7296                            unsigned size)
7297 {
7298     PCIDevice *pci = PCI_DEVICE(n);
7299     uint64_t cap = ldq_le_p(&n->bar.cap);
7300     uint32_t cc = ldl_le_p(&n->bar.cc);
7301     uint32_t intms = ldl_le_p(&n->bar.intms);
7302     uint32_t csts = ldl_le_p(&n->bar.csts);
7303     uint32_t pmrsts = ldl_le_p(&n->bar.pmrsts);
7304 
7305     if (unlikely(offset & (sizeof(uint32_t) - 1))) {
7306         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
7307                        "MMIO write not 32-bit aligned,"
7308                        " offset=0x%"PRIx64"", offset);
7309         /* should be ignored, fall through for now */
7310     }
7311 
7312     if (unlikely(size < sizeof(uint32_t))) {
7313         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
7314                        "MMIO write smaller than 32-bits,"
7315                        " offset=0x%"PRIx64", size=%u",
7316                        offset, size);
7317         /* should be ignored, fall through for now */
7318     }
7319 
7320     switch (offset) {
7321     case NVME_REG_INTMS:
7322         if (unlikely(msix_enabled(pci))) {
7323             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7324                            "undefined access to interrupt mask set"
7325                            " when MSI-X is enabled");
7326             /* should be ignored, fall through for now */
7327         }
7328         intms |= data;
7329         stl_le_p(&n->bar.intms, intms);
7330         n->bar.intmc = n->bar.intms;
7331         trace_pci_nvme_mmio_intm_set(data & 0xffffffff, intms);
7332         nvme_irq_check(n);
7333         break;
7334     case NVME_REG_INTMC:
7335         if (unlikely(msix_enabled(pci))) {
7336             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
7337                            "undefined access to interrupt mask clr"
7338                            " when MSI-X is enabled");
7339             /* should be ignored, fall through for now */
7340         }
7341         intms &= ~data;
7342         stl_le_p(&n->bar.intms, intms);
7343         n->bar.intmc = n->bar.intms;
7344         trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, intms);
7345         nvme_irq_check(n);
7346         break;
7347     case NVME_REG_CC:
7348         stl_le_p(&n->bar.cc, data);
7349 
7350         trace_pci_nvme_mmio_cfg(data & 0xffffffff);
7351 
7352         if (NVME_CC_SHN(data) && !(NVME_CC_SHN(cc))) {
7353             trace_pci_nvme_mmio_shutdown_set();
7354             nvme_ctrl_shutdown(n);
7355             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7356             csts |= NVME_CSTS_SHST_COMPLETE;
7357         } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(cc)) {
7358             trace_pci_nvme_mmio_shutdown_cleared();
7359             csts &= ~(CSTS_SHST_MASK << CSTS_SHST_SHIFT);
7360         }
7361 
7362         if (NVME_CC_EN(data) && !NVME_CC_EN(cc)) {
7363             if (unlikely(nvme_start_ctrl(n))) {
7364                 trace_pci_nvme_err_startfail();
7365                 csts = NVME_CSTS_FAILED;
7366             } else {
7367                 trace_pci_nvme_mmio_start_success();
7368                 csts = NVME_CSTS_READY;
7369             }
7370         } else if (!NVME_CC_EN(data) && NVME_CC_EN(cc)) {
7371             trace_pci_nvme_mmio_stopped();
7372             nvme_ctrl_reset(n, NVME_RESET_CONTROLLER);
7373 
7374             break;
7375         }
7376 
7377         stl_le_p(&n->bar.csts, csts);
7378 
7379         break;
7380     case NVME_REG_CSTS:
7381         if (data & (1 << 4)) {
7382             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
7383                            "attempted to W1C CSTS.NSSRO"
7384                            " but CAP.NSSRS is zero (not supported)");
7385         } else if (data != 0) {
7386             NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
7387                            "attempted to set a read only bit"
7388                            " of controller status");
7389         }
7390         break;
7391     case NVME_REG_NSSR:
7392         if (data == 0x4e564d65) {
7393             trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
7394         } else {
7395             /* The spec says that writes of other values have no effect */
7396             return;
7397         }
7398         break;
7399     case NVME_REG_AQA:
7400         stl_le_p(&n->bar.aqa, data);
7401         trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
7402         break;
7403     case NVME_REG_ASQ:
7404         stn_le_p(&n->bar.asq, size, data);
7405         trace_pci_nvme_mmio_asqaddr(data);
7406         break;
7407     case NVME_REG_ASQ + 4:
7408         stl_le_p((uint8_t *)&n->bar.asq + 4, data);
7409         trace_pci_nvme_mmio_asqaddr_hi(data, ldq_le_p(&n->bar.asq));
7410         break;
7411     case NVME_REG_ACQ:
7412         trace_pci_nvme_mmio_acqaddr(data);
7413         stn_le_p(&n->bar.acq, size, data);
7414         break;
7415     case NVME_REG_ACQ + 4:
7416         stl_le_p((uint8_t *)&n->bar.acq + 4, data);
7417         trace_pci_nvme_mmio_acqaddr_hi(data, ldq_le_p(&n->bar.acq));
7418         break;
7419     case NVME_REG_CMBLOC:
7420         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
7421                        "invalid write to reserved CMBLOC"
7422                        " when CMBSZ is zero, ignored");
7423         return;
7424     case NVME_REG_CMBSZ:
7425         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
7426                        "invalid write to read only CMBSZ, ignored");
7427         return;
7428     case NVME_REG_CMBMSC:
7429         if (!NVME_CAP_CMBS(cap)) {
7430             return;
7431         }
7432 
7433         stn_le_p(&n->bar.cmbmsc, size, data);
7434         n->cmb.cmse = false;
7435 
7436         if (NVME_CMBMSC_CRE(data)) {
7437             nvme_cmb_enable_regs(n);
7438 
7439             if (NVME_CMBMSC_CMSE(data)) {
7440                 uint64_t cmbmsc = ldq_le_p(&n->bar.cmbmsc);
7441                 hwaddr cba = NVME_CMBMSC_CBA(cmbmsc) << CMBMSC_CBA_SHIFT;
7442                 if (cba + int128_get64(n->cmb.mem.size) < cba) {
7443                     uint32_t cmbsts = ldl_le_p(&n->bar.cmbsts);
7444                     NVME_CMBSTS_SET_CBAI(cmbsts, 1);
7445                     stl_le_p(&n->bar.cmbsts, cmbsts);
7446                     return;
7447                 }
7448 
7449                 n->cmb.cba = cba;
7450                 n->cmb.cmse = true;
7451             }
7452         } else {
7453             n->bar.cmbsz = 0;
7454             n->bar.cmbloc = 0;
7455         }
7456 
7457         return;
7458     case NVME_REG_CMBMSC + 4:
7459         stl_le_p((uint8_t *)&n->bar.cmbmsc + 4, data);
7460         return;
7461 
7462     case NVME_REG_PMRCAP:
7463         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
7464                        "invalid write to PMRCAP register, ignored");
7465         return;
7466     case NVME_REG_PMRCTL:
7467         if (!NVME_CAP_PMRS(cap)) {
7468             return;
7469         }
7470 
7471         stl_le_p(&n->bar.pmrctl, data);
7472         if (NVME_PMRCTL_EN(data)) {
7473             memory_region_set_enabled(&n->pmr.dev->mr, true);
7474             pmrsts = 0;
7475         } else {
7476             memory_region_set_enabled(&n->pmr.dev->mr, false);
7477             NVME_PMRSTS_SET_NRDY(pmrsts, 1);
7478             n->pmr.cmse = false;
7479         }
7480         stl_le_p(&n->bar.pmrsts, pmrsts);
7481         return;
7482     case NVME_REG_PMRSTS:
7483         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
7484                        "invalid write to PMRSTS register, ignored");
7485         return;
7486     case NVME_REG_PMREBS:
7487         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
7488                        "invalid write to PMREBS register, ignored");
7489         return;
7490     case NVME_REG_PMRSWTP:
7491         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
7492                        "invalid write to PMRSWTP register, ignored");
7493         return;
7494     case NVME_REG_PMRMSCL:
7495         if (!NVME_CAP_PMRS(cap)) {
7496             return;
7497         }
7498 
7499         stl_le_p(&n->bar.pmrmscl, data);
7500         n->pmr.cmse = false;
7501 
7502         if (NVME_PMRMSCL_CMSE(data)) {
7503             uint64_t pmrmscu = ldl_le_p(&n->bar.pmrmscu);
7504             hwaddr cba = pmrmscu << 32 |
7505                 (NVME_PMRMSCL_CBA(data) << PMRMSCL_CBA_SHIFT);
7506             if (cba + int128_get64(n->pmr.dev->mr.size) < cba) {
7507                 NVME_PMRSTS_SET_CBAI(pmrsts, 1);
7508                 stl_le_p(&n->bar.pmrsts, pmrsts);
7509                 return;
7510             }
7511 
7512             n->pmr.cmse = true;
7513             n->pmr.cba = cba;
7514         }
7515 
7516         return;
7517     case NVME_REG_PMRMSCU:
7518         if (!NVME_CAP_PMRS(cap)) {
7519             return;
7520         }
7521 
7522         stl_le_p(&n->bar.pmrmscu, data);
7523         return;
7524     default:
7525         NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
7526                        "invalid MMIO write,"
7527                        " offset=0x%"PRIx64", data=%"PRIx64"",
7528                        offset, data);
7529         break;
7530     }
7531 }
7532 
7533 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
7534 {
7535     NvmeCtrl *n = (NvmeCtrl *)opaque;
7536     uint8_t *ptr = (uint8_t *)&n->bar;
7537 
7538     trace_pci_nvme_mmio_read(addr, size);
7539 
7540     if (unlikely(addr & (sizeof(uint32_t) - 1))) {
7541         NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
7542                        "MMIO read not 32-bit aligned,"
7543                        " offset=0x%"PRIx64"", addr);
7544         /* should RAZ, fall through for now */
7545     } else if (unlikely(size < sizeof(uint32_t))) {
7546         NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
7547                        "MMIO read smaller than 32-bits,"
7548                        " offset=0x%"PRIx64"", addr);
7549         /* should RAZ, fall through for now */
7550     }
7551 
7552     if (addr > sizeof(n->bar) - size) {
7553         NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
7554                        "MMIO read beyond last register,"
7555                        " offset=0x%"PRIx64", returning 0", addr);
7556 
7557         return 0;
7558     }
7559 
7560     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7561         addr != NVME_REG_CSTS) {
7562         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7563         return 0;
7564     }
7565 
7566     /*
7567      * When PMRWBM bit 1 is set then read from
7568      * from PMRSTS should ensure prior writes
7569      * made it to persistent media
7570      */
7571     if (addr == NVME_REG_PMRSTS &&
7572         (NVME_PMRCAP_PMRWBM(ldl_le_p(&n->bar.pmrcap)) & 0x02)) {
7573         memory_region_msync(&n->pmr.dev->mr, 0, n->pmr.dev->size);
7574     }
7575 
7576     return ldn_le_p(ptr + addr, size);
7577 }
7578 
7579 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
7580 {
7581     PCIDevice *pci = PCI_DEVICE(n);
7582     uint32_t qid;
7583 
7584     if (unlikely(addr & ((1 << 2) - 1))) {
7585         NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
7586                        "doorbell write not 32-bit aligned,"
7587                        " offset=0x%"PRIx64", ignoring", addr);
7588         return;
7589     }
7590 
7591     if (((addr - 0x1000) >> 2) & 1) {
7592         /* Completion queue doorbell write */
7593 
7594         uint16_t new_head = val & 0xffff;
7595         int start_sqs;
7596         NvmeCQueue *cq;
7597 
7598         qid = (addr - (0x1000 + (1 << 2))) >> 3;
7599         if (unlikely(nvme_check_cqid(n, qid))) {
7600             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
7601                            "completion queue doorbell write"
7602                            " for nonexistent queue,"
7603                            " sqid=%"PRIu32", ignoring", qid);
7604 
7605             /*
7606              * NVM Express v1.3d, Section 4.1 state: "If host software writes
7607              * an invalid value to the Submission Queue Tail Doorbell or
7608              * Completion Queue Head Doorbell regiter and an Asynchronous Event
7609              * Request command is outstanding, then an asynchronous event is
7610              * posted to the Admin Completion Queue with a status code of
7611              * Invalid Doorbell Write Value."
7612              *
7613              * Also note that the spec includes the "Invalid Doorbell Register"
7614              * status code, but nowhere does it specify when to use it.
7615              * However, it seems reasonable to use it here in a similar
7616              * fashion.
7617              */
7618             if (n->outstanding_aers) {
7619                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7620                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7621                                    NVME_LOG_ERROR_INFO);
7622             }
7623 
7624             return;
7625         }
7626 
7627         cq = n->cq[qid];
7628         if (unlikely(new_head >= cq->size)) {
7629             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
7630                            "completion queue doorbell write value"
7631                            " beyond queue size, sqid=%"PRIu32","
7632                            " new_head=%"PRIu16", ignoring",
7633                            qid, new_head);
7634 
7635             if (n->outstanding_aers) {
7636                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7637                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7638                                    NVME_LOG_ERROR_INFO);
7639             }
7640 
7641             return;
7642         }
7643 
7644         trace_pci_nvme_mmio_doorbell_cq(cq->cqid, new_head);
7645 
7646         start_sqs = nvme_cq_full(cq) ? 1 : 0;
7647         cq->head = new_head;
7648         if (!qid && n->dbbuf_enabled) {
7649             stl_le_pci_dma(pci, cq->db_addr, cq->head, MEMTXATTRS_UNSPECIFIED);
7650         }
7651         if (start_sqs) {
7652             NvmeSQueue *sq;
7653             QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
7654                 qemu_bh_schedule(sq->bh);
7655             }
7656             qemu_bh_schedule(cq->bh);
7657         }
7658 
7659         if (cq->tail == cq->head) {
7660             if (cq->irq_enabled) {
7661                 n->cq_pending--;
7662             }
7663 
7664             nvme_irq_deassert(n, cq);
7665         }
7666     } else {
7667         /* Submission queue doorbell write */
7668 
7669         uint16_t new_tail = val & 0xffff;
7670         NvmeSQueue *sq;
7671 
7672         qid = (addr - 0x1000) >> 3;
7673         if (unlikely(nvme_check_sqid(n, qid))) {
7674             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
7675                            "submission queue doorbell write"
7676                            " for nonexistent queue,"
7677                            " sqid=%"PRIu32", ignoring", qid);
7678 
7679             if (n->outstanding_aers) {
7680                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7681                                    NVME_AER_INFO_ERR_INVALID_DB_REGISTER,
7682                                    NVME_LOG_ERROR_INFO);
7683             }
7684 
7685             return;
7686         }
7687 
7688         sq = n->sq[qid];
7689         if (unlikely(new_tail >= sq->size)) {
7690             NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
7691                            "submission queue doorbell write value"
7692                            " beyond queue size, sqid=%"PRIu32","
7693                            " new_tail=%"PRIu16", ignoring",
7694                            qid, new_tail);
7695 
7696             if (n->outstanding_aers) {
7697                 nvme_enqueue_event(n, NVME_AER_TYPE_ERROR,
7698                                    NVME_AER_INFO_ERR_INVALID_DB_VALUE,
7699                                    NVME_LOG_ERROR_INFO);
7700             }
7701 
7702             return;
7703         }
7704 
7705         trace_pci_nvme_mmio_doorbell_sq(sq->sqid, new_tail);
7706 
7707         sq->tail = new_tail;
7708         if (!qid && n->dbbuf_enabled) {
7709             /*
7710              * The spec states "the host shall also update the controller's
7711              * corresponding doorbell property to match the value of that entry
7712              * in the Shadow Doorbell buffer."
7713              *
7714              * Since this context is currently a VM trap, we can safely enforce
7715              * the requirement from the device side in case the host is
7716              * misbehaving.
7717              *
7718              * Note, we shouldn't have to do this, but various drivers
7719              * including ones that run on Linux, are not updating Admin Queues,
7720              * so we can't trust reading it for an appropriate sq tail.
7721              */
7722             stl_le_pci_dma(pci, sq->db_addr, sq->tail, MEMTXATTRS_UNSPECIFIED);
7723         }
7724 
7725         qemu_bh_schedule(sq->bh);
7726     }
7727 }
7728 
7729 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
7730                             unsigned size)
7731 {
7732     NvmeCtrl *n = (NvmeCtrl *)opaque;
7733 
7734     trace_pci_nvme_mmio_write(addr, data, size);
7735 
7736     if (pci_is_vf(PCI_DEVICE(n)) && !nvme_sctrl(n)->scs &&
7737         addr != NVME_REG_CSTS) {
7738         trace_pci_nvme_err_ignored_mmio_vf_offline(addr, size);
7739         return;
7740     }
7741 
7742     if (addr < sizeof(n->bar)) {
7743         nvme_write_bar(n, addr, data, size);
7744     } else {
7745         nvme_process_db(n, addr, data);
7746     }
7747 }
7748 
7749 static const MemoryRegionOps nvme_mmio_ops = {
7750     .read = nvme_mmio_read,
7751     .write = nvme_mmio_write,
7752     .endianness = DEVICE_LITTLE_ENDIAN,
7753     .impl = {
7754         .min_access_size = 2,
7755         .max_access_size = 8,
7756     },
7757 };
7758 
7759 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
7760                            unsigned size)
7761 {
7762     NvmeCtrl *n = (NvmeCtrl *)opaque;
7763     stn_le_p(&n->cmb.buf[addr], size, data);
7764 }
7765 
7766 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
7767 {
7768     NvmeCtrl *n = (NvmeCtrl *)opaque;
7769     return ldn_le_p(&n->cmb.buf[addr], size);
7770 }
7771 
7772 static const MemoryRegionOps nvme_cmb_ops = {
7773     .read = nvme_cmb_read,
7774     .write = nvme_cmb_write,
7775     .endianness = DEVICE_LITTLE_ENDIAN,
7776     .impl = {
7777         .min_access_size = 1,
7778         .max_access_size = 8,
7779     },
7780 };
7781 
7782 static bool nvme_check_params(NvmeCtrl *n, Error **errp)
7783 {
7784     NvmeParams *params = &n->params;
7785 
7786     if (params->num_queues) {
7787         warn_report("num_queues is deprecated; please use max_ioqpairs "
7788                     "instead");
7789 
7790         params->max_ioqpairs = params->num_queues - 1;
7791     }
7792 
7793     if (n->namespace.blkconf.blk && n->subsys) {
7794         error_setg(errp, "subsystem support is unavailable with legacy "
7795                    "namespace ('drive' property)");
7796         return false;
7797     }
7798 
7799     if (params->max_ioqpairs < 1 ||
7800         params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
7801         error_setg(errp, "max_ioqpairs must be between 1 and %d",
7802                    NVME_MAX_IOQPAIRS);
7803         return false;
7804     }
7805 
7806     if (params->msix_qsize < 1 ||
7807         params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
7808         error_setg(errp, "msix_qsize must be between 1 and %d",
7809                    PCI_MSIX_FLAGS_QSIZE + 1);
7810         return false;
7811     }
7812 
7813     if (!params->serial) {
7814         error_setg(errp, "serial property not set");
7815         return false;
7816     }
7817 
7818     if (n->pmr.dev) {
7819         if (host_memory_backend_is_mapped(n->pmr.dev)) {
7820             error_setg(errp, "can't use already busy memdev: %s",
7821                        object_get_canonical_path_component(OBJECT(n->pmr.dev)));
7822             return false;
7823         }
7824 
7825         if (!is_power_of_2(n->pmr.dev->size)) {
7826             error_setg(errp, "pmr backend size needs to be power of 2 in size");
7827             return false;
7828         }
7829 
7830         host_memory_backend_set_mapped(n->pmr.dev, true);
7831     }
7832 
7833     if (n->params.zasl > n->params.mdts) {
7834         error_setg(errp, "zoned.zasl (Zone Append Size Limit) must be less "
7835                    "than or equal to mdts (Maximum Data Transfer Size)");
7836         return false;
7837     }
7838 
7839     if (!n->params.vsl) {
7840         error_setg(errp, "vsl must be non-zero");
7841         return false;
7842     }
7843 
7844     if (params->sriov_max_vfs) {
7845         if (!n->subsys) {
7846             error_setg(errp, "subsystem is required for the use of SR-IOV");
7847             return false;
7848         }
7849 
7850         if (params->sriov_max_vfs > NVME_MAX_VFS) {
7851             error_setg(errp, "sriov_max_vfs must be between 0 and %d",
7852                        NVME_MAX_VFS);
7853             return false;
7854         }
7855 
7856         if (params->cmb_size_mb) {
7857             error_setg(errp, "CMB is not supported with SR-IOV");
7858             return false;
7859         }
7860 
7861         if (n->pmr.dev) {
7862             error_setg(errp, "PMR is not supported with SR-IOV");
7863             return false;
7864         }
7865 
7866         if (!params->sriov_vq_flexible || !params->sriov_vi_flexible) {
7867             error_setg(errp, "both sriov_vq_flexible and sriov_vi_flexible"
7868                        " must be set for the use of SR-IOV");
7869             return false;
7870         }
7871 
7872         if (params->sriov_vq_flexible < params->sriov_max_vfs * 2) {
7873             error_setg(errp, "sriov_vq_flexible must be greater than or equal"
7874                        " to %d (sriov_max_vfs * 2)", params->sriov_max_vfs * 2);
7875             return false;
7876         }
7877 
7878         if (params->max_ioqpairs < params->sriov_vq_flexible + 2) {
7879             error_setg(errp, "(max_ioqpairs - sriov_vq_flexible) must be"
7880                        " greater than or equal to 2");
7881             return false;
7882         }
7883 
7884         if (params->sriov_vi_flexible < params->sriov_max_vfs) {
7885             error_setg(errp, "sriov_vi_flexible must be greater than or equal"
7886                        " to %d (sriov_max_vfs)", params->sriov_max_vfs);
7887             return false;
7888         }
7889 
7890         if (params->msix_qsize < params->sriov_vi_flexible + 1) {
7891             error_setg(errp, "(msix_qsize - sriov_vi_flexible) must be"
7892                        " greater than or equal to 1");
7893             return false;
7894         }
7895 
7896         if (params->sriov_max_vi_per_vf &&
7897             (params->sriov_max_vi_per_vf - 1) % NVME_VF_RES_GRANULARITY) {
7898             error_setg(errp, "sriov_max_vi_per_vf must meet:"
7899                        " (sriov_max_vi_per_vf - 1) %% %d == 0 and"
7900                        " sriov_max_vi_per_vf >= 1", NVME_VF_RES_GRANULARITY);
7901             return false;
7902         }
7903 
7904         if (params->sriov_max_vq_per_vf &&
7905             (params->sriov_max_vq_per_vf < 2 ||
7906              (params->sriov_max_vq_per_vf - 1) % NVME_VF_RES_GRANULARITY)) {
7907             error_setg(errp, "sriov_max_vq_per_vf must meet:"
7908                        " (sriov_max_vq_per_vf - 1) %% %d == 0 and"
7909                        " sriov_max_vq_per_vf >= 2", NVME_VF_RES_GRANULARITY);
7910             return false;
7911         }
7912     }
7913 
7914     return true;
7915 }
7916 
7917 static void nvme_init_state(NvmeCtrl *n)
7918 {
7919     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
7920     NvmeSecCtrlList *list = &n->sec_ctrl_list;
7921     NvmeSecCtrlEntry *sctrl;
7922     PCIDevice *pci = PCI_DEVICE(n);
7923     uint8_t max_vfs;
7924     int i;
7925 
7926     if (pci_is_vf(pci)) {
7927         sctrl = nvme_sctrl(n);
7928         max_vfs = 0;
7929         n->conf_ioqpairs = sctrl->nvq ? le16_to_cpu(sctrl->nvq) - 1 : 0;
7930         n->conf_msix_qsize = sctrl->nvi ? le16_to_cpu(sctrl->nvi) : 1;
7931     } else {
7932         max_vfs = n->params.sriov_max_vfs;
7933         n->conf_ioqpairs = n->params.max_ioqpairs;
7934         n->conf_msix_qsize = n->params.msix_qsize;
7935     }
7936 
7937     n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
7938     n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
7939     n->temperature = NVME_TEMPERATURE;
7940     n->features.temp_thresh_hi = NVME_TEMPERATURE_WARNING;
7941     n->starttime_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
7942     n->aer_reqs = g_new0(NvmeRequest *, n->params.aerl + 1);
7943     QTAILQ_INIT(&n->aer_queue);
7944 
7945     list->numcntl = cpu_to_le16(max_vfs);
7946     for (i = 0; i < max_vfs; i++) {
7947         sctrl = &list->sec[i];
7948         sctrl->pcid = cpu_to_le16(n->cntlid);
7949         sctrl->vfn = cpu_to_le16(i + 1);
7950     }
7951 
7952     cap->cntlid = cpu_to_le16(n->cntlid);
7953     cap->crt = NVME_CRT_VQ | NVME_CRT_VI;
7954 
7955     if (pci_is_vf(pci)) {
7956         cap->vqprt = cpu_to_le16(1 + n->conf_ioqpairs);
7957     } else {
7958         cap->vqprt = cpu_to_le16(1 + n->params.max_ioqpairs -
7959                                  n->params.sriov_vq_flexible);
7960         cap->vqfrt = cpu_to_le32(n->params.sriov_vq_flexible);
7961         cap->vqrfap = cap->vqfrt;
7962         cap->vqgran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
7963         cap->vqfrsm = n->params.sriov_max_vq_per_vf ?
7964                         cpu_to_le16(n->params.sriov_max_vq_per_vf) :
7965                         cap->vqfrt / MAX(max_vfs, 1);
7966     }
7967 
7968     if (pci_is_vf(pci)) {
7969         cap->viprt = cpu_to_le16(n->conf_msix_qsize);
7970     } else {
7971         cap->viprt = cpu_to_le16(n->params.msix_qsize -
7972                                  n->params.sriov_vi_flexible);
7973         cap->vifrt = cpu_to_le32(n->params.sriov_vi_flexible);
7974         cap->virfap = cap->vifrt;
7975         cap->vigran = cpu_to_le16(NVME_VF_RES_GRANULARITY);
7976         cap->vifrsm = n->params.sriov_max_vi_per_vf ?
7977                         cpu_to_le16(n->params.sriov_max_vi_per_vf) :
7978                         cap->vifrt / MAX(max_vfs, 1);
7979     }
7980 }
7981 
7982 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
7983 {
7984     uint64_t cmb_size = n->params.cmb_size_mb * MiB;
7985     uint64_t cap = ldq_le_p(&n->bar.cap);
7986 
7987     n->cmb.buf = g_malloc0(cmb_size);
7988     memory_region_init_io(&n->cmb.mem, OBJECT(n), &nvme_cmb_ops, n,
7989                           "nvme-cmb", cmb_size);
7990     pci_register_bar(pci_dev, NVME_CMB_BIR,
7991                      PCI_BASE_ADDRESS_SPACE_MEMORY |
7992                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
7993                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->cmb.mem);
7994 
7995     NVME_CAP_SET_CMBS(cap, 1);
7996     stq_le_p(&n->bar.cap, cap);
7997 
7998     if (n->params.legacy_cmb) {
7999         nvme_cmb_enable_regs(n);
8000         n->cmb.cmse = true;
8001     }
8002 }
8003 
8004 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
8005 {
8006     uint32_t pmrcap = ldl_le_p(&n->bar.pmrcap);
8007 
8008     NVME_PMRCAP_SET_RDS(pmrcap, 1);
8009     NVME_PMRCAP_SET_WDS(pmrcap, 1);
8010     NVME_PMRCAP_SET_BIR(pmrcap, NVME_PMR_BIR);
8011     /* Turn on bit 1 support */
8012     NVME_PMRCAP_SET_PMRWBM(pmrcap, 0x02);
8013     NVME_PMRCAP_SET_CMSS(pmrcap, 1);
8014     stl_le_p(&n->bar.pmrcap, pmrcap);
8015 
8016     pci_register_bar(pci_dev, NVME_PMR_BIR,
8017                      PCI_BASE_ADDRESS_SPACE_MEMORY |
8018                      PCI_BASE_ADDRESS_MEM_TYPE_64 |
8019                      PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmr.dev->mr);
8020 
8021     memory_region_set_enabled(&n->pmr.dev->mr, false);
8022 }
8023 
8024 static uint64_t nvme_bar_size(unsigned total_queues, unsigned total_irqs,
8025                               unsigned *msix_table_offset,
8026                               unsigned *msix_pba_offset)
8027 {
8028     uint64_t bar_size, msix_table_size, msix_pba_size;
8029 
8030     bar_size = sizeof(NvmeBar) + 2 * total_queues * NVME_DB_SIZE;
8031     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8032 
8033     if (msix_table_offset) {
8034         *msix_table_offset = bar_size;
8035     }
8036 
8037     msix_table_size = PCI_MSIX_ENTRY_SIZE * total_irqs;
8038     bar_size += msix_table_size;
8039     bar_size = QEMU_ALIGN_UP(bar_size, 4 * KiB);
8040 
8041     if (msix_pba_offset) {
8042         *msix_pba_offset = bar_size;
8043     }
8044 
8045     msix_pba_size = QEMU_ALIGN_UP(total_irqs, 64) / 8;
8046     bar_size += msix_pba_size;
8047 
8048     bar_size = pow2ceil(bar_size);
8049     return bar_size;
8050 }
8051 
8052 static void nvme_init_sriov(NvmeCtrl *n, PCIDevice *pci_dev, uint16_t offset)
8053 {
8054     uint16_t vf_dev_id = n->params.use_intel_id ?
8055                          PCI_DEVICE_ID_INTEL_NVME : PCI_DEVICE_ID_REDHAT_NVME;
8056     NvmePriCtrlCap *cap = &n->pri_ctrl_cap;
8057     uint64_t bar_size = nvme_bar_size(le16_to_cpu(cap->vqfrsm),
8058                                       le16_to_cpu(cap->vifrsm),
8059                                       NULL, NULL);
8060 
8061     pcie_sriov_pf_init(pci_dev, offset, "nvme", vf_dev_id,
8062                        n->params.sriov_max_vfs, n->params.sriov_max_vfs,
8063                        NVME_VF_OFFSET, NVME_VF_STRIDE);
8064 
8065     pcie_sriov_pf_init_vf_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8066                               PCI_BASE_ADDRESS_MEM_TYPE_64, bar_size);
8067 }
8068 
8069 static int nvme_add_pm_capability(PCIDevice *pci_dev, uint8_t offset)
8070 {
8071     Error *err = NULL;
8072     int ret;
8073 
8074     ret = pci_add_capability(pci_dev, PCI_CAP_ID_PM, offset,
8075                              PCI_PM_SIZEOF, &err);
8076     if (err) {
8077         error_report_err(err);
8078         return ret;
8079     }
8080 
8081     pci_set_word(pci_dev->config + offset + PCI_PM_PMC,
8082                  PCI_PM_CAP_VER_1_2);
8083     pci_set_word(pci_dev->config + offset + PCI_PM_CTRL,
8084                  PCI_PM_CTRL_NO_SOFT_RESET);
8085     pci_set_word(pci_dev->wmask + offset + PCI_PM_CTRL,
8086                  PCI_PM_CTRL_STATE_MASK);
8087 
8088     return 0;
8089 }
8090 
8091 static bool nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
8092 {
8093     ERRP_GUARD();
8094     uint8_t *pci_conf = pci_dev->config;
8095     uint64_t bar_size;
8096     unsigned msix_table_offset, msix_pba_offset;
8097     int ret;
8098 
8099     pci_conf[PCI_INTERRUPT_PIN] = 1;
8100     pci_config_set_prog_interface(pci_conf, 0x2);
8101 
8102     if (n->params.use_intel_id) {
8103         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_INTEL);
8104         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_INTEL_NVME);
8105     } else {
8106         pci_config_set_vendor_id(pci_conf, PCI_VENDOR_ID_REDHAT);
8107         pci_config_set_device_id(pci_conf, PCI_DEVICE_ID_REDHAT_NVME);
8108     }
8109 
8110     pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
8111     nvme_add_pm_capability(pci_dev, 0x60);
8112     pcie_endpoint_cap_init(pci_dev, 0x80);
8113     pcie_cap_flr_init(pci_dev);
8114     if (n->params.sriov_max_vfs) {
8115         pcie_ari_init(pci_dev, 0x100);
8116     }
8117 
8118     /* add one to max_ioqpairs to account for the admin queue pair */
8119     bar_size = nvme_bar_size(n->params.max_ioqpairs + 1, n->params.msix_qsize,
8120                              &msix_table_offset, &msix_pba_offset);
8121 
8122     memory_region_init(&n->bar0, OBJECT(n), "nvme-bar0", bar_size);
8123     memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
8124                           msix_table_offset);
8125     memory_region_add_subregion(&n->bar0, 0, &n->iomem);
8126 
8127     if (pci_is_vf(pci_dev)) {
8128         pcie_sriov_vf_register_bar(pci_dev, 0, &n->bar0);
8129     } else {
8130         pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
8131                          PCI_BASE_ADDRESS_MEM_TYPE_64, &n->bar0);
8132     }
8133     ret = msix_init(pci_dev, n->params.msix_qsize,
8134                     &n->bar0, 0, msix_table_offset,
8135                     &n->bar0, 0, msix_pba_offset, 0, errp);
8136     if (ret == -ENOTSUP) {
8137         /* report that msix is not supported, but do not error out */
8138         warn_report_err(*errp);
8139         *errp = NULL;
8140     } else if (ret < 0) {
8141         /* propagate error to caller */
8142         return false;
8143     }
8144 
8145     nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
8146 
8147     if (n->params.cmb_size_mb) {
8148         nvme_init_cmb(n, pci_dev);
8149     }
8150 
8151     if (n->pmr.dev) {
8152         nvme_init_pmr(n, pci_dev);
8153     }
8154 
8155     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8156         nvme_init_sriov(n, pci_dev, 0x120);
8157     }
8158 
8159     return true;
8160 }
8161 
8162 static void nvme_init_subnqn(NvmeCtrl *n)
8163 {
8164     NvmeSubsystem *subsys = n->subsys;
8165     NvmeIdCtrl *id = &n->id_ctrl;
8166 
8167     if (!subsys) {
8168         snprintf((char *)id->subnqn, sizeof(id->subnqn),
8169                  "nqn.2019-08.org.qemu:%s", n->params.serial);
8170     } else {
8171         pstrcpy((char *)id->subnqn, sizeof(id->subnqn), (char*)subsys->subnqn);
8172     }
8173 }
8174 
8175 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
8176 {
8177     NvmeIdCtrl *id = &n->id_ctrl;
8178     uint8_t *pci_conf = pci_dev->config;
8179     uint64_t cap = ldq_le_p(&n->bar.cap);
8180     NvmeSecCtrlEntry *sctrl = nvme_sctrl(n);
8181     uint32_t ctratt;
8182 
8183     id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
8184     id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
8185     strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
8186     strpadcpy((char *)id->fr, sizeof(id->fr), QEMU_VERSION, ' ');
8187     strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
8188 
8189     id->cntlid = cpu_to_le16(n->cntlid);
8190 
8191     id->oaes = cpu_to_le32(NVME_OAES_NS_ATTR);
8192     ctratt = NVME_CTRATT_ELBAS;
8193 
8194     id->rab = 6;
8195 
8196     if (n->params.use_intel_id) {
8197         id->ieee[0] = 0xb3;
8198         id->ieee[1] = 0x02;
8199         id->ieee[2] = 0x00;
8200     } else {
8201         id->ieee[0] = 0x00;
8202         id->ieee[1] = 0x54;
8203         id->ieee[2] = 0x52;
8204     }
8205 
8206     id->mdts = n->params.mdts;
8207     id->ver = cpu_to_le32(NVME_SPEC_VER);
8208     id->oacs =
8209         cpu_to_le16(NVME_OACS_NS_MGMT | NVME_OACS_FORMAT | NVME_OACS_DBBUF |
8210                     NVME_OACS_DIRECTIVES);
8211     id->cntrltype = 0x1;
8212 
8213     /*
8214      * Because the controller always completes the Abort command immediately,
8215      * there can never be more than one concurrently executing Abort command,
8216      * so this value is never used for anything. Note that there can easily be
8217      * many Abort commands in the queues, but they are not considered
8218      * "executing" until processed by nvme_abort.
8219      *
8220      * The specification recommends a value of 3 for Abort Command Limit (four
8221      * concurrently outstanding Abort commands), so lets use that though it is
8222      * inconsequential.
8223      */
8224     id->acl = 3;
8225     id->aerl = n->params.aerl;
8226     id->frmw = (NVME_NUM_FW_SLOTS << 1) | NVME_FRMW_SLOT1_RO;
8227     id->lpa = NVME_LPA_NS_SMART | NVME_LPA_CSE | NVME_LPA_EXTENDED;
8228 
8229     /* recommended default value (~70 C) */
8230     id->wctemp = cpu_to_le16(NVME_TEMPERATURE_WARNING);
8231     id->cctemp = cpu_to_le16(NVME_TEMPERATURE_CRITICAL);
8232 
8233     id->sqes = (0x6 << 4) | 0x6;
8234     id->cqes = (0x4 << 4) | 0x4;
8235     id->nn = cpu_to_le32(NVME_MAX_NAMESPACES);
8236     id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP |
8237                            NVME_ONCS_FEATURES | NVME_ONCS_DSM |
8238                            NVME_ONCS_COMPARE | NVME_ONCS_COPY);
8239 
8240     /*
8241      * NOTE: If this device ever supports a command set that does NOT use 0x0
8242      * as a Flush-equivalent operation, support for the broadcast NSID in Flush
8243      * should probably be removed.
8244      *
8245      * See comment in nvme_io_cmd.
8246      */
8247     id->vwc = NVME_VWC_NSID_BROADCAST_SUPPORT | NVME_VWC_PRESENT;
8248 
8249     id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0 | NVME_OCFS_COPY_FORMAT_1);
8250     id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN);
8251 
8252     nvme_init_subnqn(n);
8253 
8254     id->psd[0].mp = cpu_to_le16(0x9c4);
8255     id->psd[0].enlat = cpu_to_le32(0x10);
8256     id->psd[0].exlat = cpu_to_le32(0x4);
8257 
8258     if (n->subsys) {
8259         id->cmic |= NVME_CMIC_MULTI_CTRL;
8260         ctratt |= NVME_CTRATT_ENDGRPS;
8261 
8262         id->endgidmax = cpu_to_le16(0x1);
8263 
8264         if (n->subsys->endgrp.fdp.enabled) {
8265             ctratt |= NVME_CTRATT_FDPS;
8266         }
8267     }
8268 
8269     id->ctratt = cpu_to_le32(ctratt);
8270 
8271     NVME_CAP_SET_MQES(cap, 0x7ff);
8272     NVME_CAP_SET_CQR(cap, 1);
8273     NVME_CAP_SET_TO(cap, 0xf);
8274     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_NVM);
8275     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_CSI_SUPP);
8276     NVME_CAP_SET_CSS(cap, NVME_CAP_CSS_ADMIN_ONLY);
8277     NVME_CAP_SET_MPSMAX(cap, 4);
8278     NVME_CAP_SET_CMBS(cap, n->params.cmb_size_mb ? 1 : 0);
8279     NVME_CAP_SET_PMRS(cap, n->pmr.dev ? 1 : 0);
8280     stq_le_p(&n->bar.cap, cap);
8281 
8282     stl_le_p(&n->bar.vs, NVME_SPEC_VER);
8283     n->bar.intmc = n->bar.intms = 0;
8284 
8285     if (pci_is_vf(pci_dev) && !sctrl->scs) {
8286         stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
8287     }
8288 }
8289 
8290 static int nvme_init_subsys(NvmeCtrl *n, Error **errp)
8291 {
8292     int cntlid;
8293 
8294     if (!n->subsys) {
8295         return 0;
8296     }
8297 
8298     cntlid = nvme_subsys_register_ctrl(n, errp);
8299     if (cntlid < 0) {
8300         return -1;
8301     }
8302 
8303     n->cntlid = cntlid;
8304 
8305     return 0;
8306 }
8307 
8308 void nvme_attach_ns(NvmeCtrl *n, NvmeNamespace *ns)
8309 {
8310     uint32_t nsid = ns->params.nsid;
8311     assert(nsid && nsid <= NVME_MAX_NAMESPACES);
8312 
8313     n->namespaces[nsid] = ns;
8314     ns->attached++;
8315 
8316     n->dmrsl = MIN_NON_ZERO(n->dmrsl,
8317                             BDRV_REQUEST_MAX_BYTES / nvme_l2b(ns, 1));
8318 }
8319 
8320 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
8321 {
8322     NvmeCtrl *n = NVME(pci_dev);
8323     DeviceState *dev = DEVICE(pci_dev);
8324     NvmeNamespace *ns;
8325     NvmeCtrl *pn = NVME(pcie_sriov_get_pf(pci_dev));
8326 
8327     if (pci_is_vf(pci_dev)) {
8328         /*
8329          * VFs derive settings from the parent. PF's lifespan exceeds
8330          * that of VF's, so it's safe to share params.serial.
8331          */
8332         memcpy(&n->params, &pn->params, sizeof(NvmeParams));
8333         n->subsys = pn->subsys;
8334     }
8335 
8336     if (!nvme_check_params(n, errp)) {
8337         return;
8338     }
8339 
8340     qbus_init(&n->bus, sizeof(NvmeBus), TYPE_NVME_BUS, dev, dev->id);
8341 
8342     if (nvme_init_subsys(n, errp)) {
8343         return;
8344     }
8345     nvme_init_state(n);
8346     if (!nvme_init_pci(n, pci_dev, errp)) {
8347         return;
8348     }
8349     nvme_init_ctrl(n, pci_dev);
8350 
8351     /* setup a namespace if the controller drive property was given */
8352     if (n->namespace.blkconf.blk) {
8353         ns = &n->namespace;
8354         ns->params.nsid = 1;
8355 
8356         if (nvme_ns_setup(ns, errp)) {
8357             return;
8358         }
8359 
8360         nvme_attach_ns(n, ns);
8361     }
8362 }
8363 
8364 static void nvme_exit(PCIDevice *pci_dev)
8365 {
8366     NvmeCtrl *n = NVME(pci_dev);
8367     NvmeNamespace *ns;
8368     int i;
8369 
8370     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8371 
8372     if (n->subsys) {
8373         for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
8374             ns = nvme_ns(n, i);
8375             if (ns) {
8376                 ns->attached--;
8377             }
8378         }
8379 
8380         nvme_subsys_unregister_ctrl(n->subsys, n);
8381     }
8382 
8383     g_free(n->cq);
8384     g_free(n->sq);
8385     g_free(n->aer_reqs);
8386 
8387     if (n->params.cmb_size_mb) {
8388         g_free(n->cmb.buf);
8389     }
8390 
8391     if (n->pmr.dev) {
8392         host_memory_backend_set_mapped(n->pmr.dev, false);
8393     }
8394 
8395     if (!pci_is_vf(pci_dev) && n->params.sriov_max_vfs) {
8396         pcie_sriov_pf_exit(pci_dev);
8397     }
8398 
8399     msix_uninit(pci_dev, &n->bar0, &n->bar0);
8400     memory_region_del_subregion(&n->bar0, &n->iomem);
8401 }
8402 
8403 static Property nvme_props[] = {
8404     DEFINE_BLOCK_PROPERTIES(NvmeCtrl, namespace.blkconf),
8405     DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmr.dev, TYPE_MEMORY_BACKEND,
8406                      HostMemoryBackend *),
8407     DEFINE_PROP_LINK("subsys", NvmeCtrl, subsys, TYPE_NVME_SUBSYS,
8408                      NvmeSubsystem *),
8409     DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
8410     DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
8411     DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
8412     DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
8413     DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
8414     DEFINE_PROP_UINT8("aerl", NvmeCtrl, params.aerl, 3),
8415     DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl, params.aer_max_queued, 64),
8416     DEFINE_PROP_UINT8("mdts", NvmeCtrl, params.mdts, 7),
8417     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
8418     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
8419     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
8420     DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl, params.ioeventfd, false),
8421     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
8422     DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl,
8423                      params.auto_transition_zones, true),
8424     DEFINE_PROP_UINT8("sriov_max_vfs", NvmeCtrl, params.sriov_max_vfs, 0),
8425     DEFINE_PROP_UINT16("sriov_vq_flexible", NvmeCtrl,
8426                        params.sriov_vq_flexible, 0),
8427     DEFINE_PROP_UINT16("sriov_vi_flexible", NvmeCtrl,
8428                        params.sriov_vi_flexible, 0),
8429     DEFINE_PROP_UINT8("sriov_max_vi_per_vf", NvmeCtrl,
8430                       params.sriov_max_vi_per_vf, 0),
8431     DEFINE_PROP_UINT8("sriov_max_vq_per_vf", NvmeCtrl,
8432                       params.sriov_max_vq_per_vf, 0),
8433     DEFINE_PROP_END_OF_LIST(),
8434 };
8435 
8436 static void nvme_get_smart_warning(Object *obj, Visitor *v, const char *name,
8437                                    void *opaque, Error **errp)
8438 {
8439     NvmeCtrl *n = NVME(obj);
8440     uint8_t value = n->smart_critical_warning;
8441 
8442     visit_type_uint8(v, name, &value, errp);
8443 }
8444 
8445 static void nvme_set_smart_warning(Object *obj, Visitor *v, const char *name,
8446                                    void *opaque, Error **errp)
8447 {
8448     NvmeCtrl *n = NVME(obj);
8449     uint8_t value, old_value, cap = 0, index, event;
8450 
8451     if (!visit_type_uint8(v, name, &value, errp)) {
8452         return;
8453     }
8454 
8455     cap = NVME_SMART_SPARE | NVME_SMART_TEMPERATURE | NVME_SMART_RELIABILITY
8456           | NVME_SMART_MEDIA_READ_ONLY | NVME_SMART_FAILED_VOLATILE_MEDIA;
8457     if (NVME_CAP_PMRS(ldq_le_p(&n->bar.cap))) {
8458         cap |= NVME_SMART_PMR_UNRELIABLE;
8459     }
8460 
8461     if ((value & cap) != value) {
8462         error_setg(errp, "unsupported smart critical warning bits: 0x%x",
8463                    value & ~cap);
8464         return;
8465     }
8466 
8467     old_value = n->smart_critical_warning;
8468     n->smart_critical_warning = value;
8469 
8470     /* only inject new bits of smart critical warning */
8471     for (index = 0; index < NVME_SMART_WARN_MAX; index++) {
8472         event = 1 << index;
8473         if (value & ~old_value & event)
8474             nvme_smart_event(n, event);
8475     }
8476 }
8477 
8478 static void nvme_pci_reset(DeviceState *qdev)
8479 {
8480     PCIDevice *pci_dev = PCI_DEVICE(qdev);
8481     NvmeCtrl *n = NVME(pci_dev);
8482 
8483     trace_pci_nvme_pci_reset();
8484     nvme_ctrl_reset(n, NVME_RESET_FUNCTION);
8485 }
8486 
8487 static void nvme_sriov_pre_write_ctrl(PCIDevice *dev, uint32_t address,
8488                                       uint32_t val, int len)
8489 {
8490     NvmeCtrl *n = NVME(dev);
8491     NvmeSecCtrlEntry *sctrl;
8492     uint16_t sriov_cap = dev->exp.sriov_cap;
8493     uint32_t off = address - sriov_cap;
8494     int i, num_vfs;
8495 
8496     if (!sriov_cap) {
8497         return;
8498     }
8499 
8500     if (range_covers_byte(off, len, PCI_SRIOV_CTRL)) {
8501         if (!(val & PCI_SRIOV_CTRL_VFE)) {
8502             num_vfs = pci_get_word(dev->config + sriov_cap + PCI_SRIOV_NUM_VF);
8503             for (i = 0; i < num_vfs; i++) {
8504                 sctrl = &n->sec_ctrl_list.sec[i];
8505                 nvme_virt_set_state(n, le16_to_cpu(sctrl->scid), false);
8506             }
8507         }
8508     }
8509 }
8510 
8511 static void nvme_pci_write_config(PCIDevice *dev, uint32_t address,
8512                                   uint32_t val, int len)
8513 {
8514     nvme_sriov_pre_write_ctrl(dev, address, val, len);
8515     pci_default_write_config(dev, address, val, len);
8516     pcie_cap_flr_write_config(dev, address, val, len);
8517 }
8518 
8519 static const VMStateDescription nvme_vmstate = {
8520     .name = "nvme",
8521     .unmigratable = 1,
8522 };
8523 
8524 static void nvme_class_init(ObjectClass *oc, void *data)
8525 {
8526     DeviceClass *dc = DEVICE_CLASS(oc);
8527     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
8528 
8529     pc->realize = nvme_realize;
8530     pc->config_write = nvme_pci_write_config;
8531     pc->exit = nvme_exit;
8532     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
8533     pc->revision = 2;
8534 
8535     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
8536     dc->desc = "Non-Volatile Memory Express";
8537     device_class_set_props(dc, nvme_props);
8538     dc->vmsd = &nvme_vmstate;
8539     dc->reset = nvme_pci_reset;
8540 }
8541 
8542 static void nvme_instance_init(Object *obj)
8543 {
8544     NvmeCtrl *n = NVME(obj);
8545 
8546     device_add_bootindex_property(obj, &n->namespace.blkconf.bootindex,
8547                                   "bootindex", "/namespace@1,0",
8548                                   DEVICE(obj));
8549 
8550     object_property_add(obj, "smart_critical_warning", "uint8",
8551                         nvme_get_smart_warning,
8552                         nvme_set_smart_warning, NULL, NULL);
8553 }
8554 
8555 static const TypeInfo nvme_info = {
8556     .name          = TYPE_NVME,
8557     .parent        = TYPE_PCI_DEVICE,
8558     .instance_size = sizeof(NvmeCtrl),
8559     .instance_init = nvme_instance_init,
8560     .class_init    = nvme_class_init,
8561     .interfaces = (InterfaceInfo[]) {
8562         { INTERFACE_PCIE_DEVICE },
8563         { }
8564     },
8565 };
8566 
8567 static const TypeInfo nvme_bus_info = {
8568     .name = TYPE_NVME_BUS,
8569     .parent = TYPE_BUS,
8570     .instance_size = sizeof(NvmeBus),
8571 };
8572 
8573 static void nvme_register_types(void)
8574 {
8575     type_register_static(&nvme_info);
8576     type_register_static(&nvme_bus_info);
8577 }
8578 
8579 type_init(nvme_register_types)
8580