2 * QEMU NVM Express Controller
4 * Copyright (c) 2012, Intel Corporation
6 * Written by Keith Busch <keith.busch@intel.com>
8 * This code is licensed under the GNU GPL v2 or later.
12 * Reference Specs: http://www.nvmexpress.org, 1.4, 1.3, 1.2, 1.1, 1.0e
14 * https://nvmexpress.org/developers/nvme-specification/
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).
25 * See docs/system/nvme.rst for extensive documentation.
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]> \
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]>
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).
60 * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
62 * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
63 * size=<size> .... -device nvme,...,pmrdev=<mem_id>
65 * The PMR will use BAR 4/5 exclusively.
67 * To place controller(s) and namespace(s) to a subsystem, then provide
68 * nvme-subsys device as above.
70 * nvme subsystem device parameters
71 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
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>`).
79 * nvme device parameters
80 * ~~~~~~~~~~~~~~~~~~~~~~
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).
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.
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.
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).
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
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`).
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.
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.
127 * NOTE: Single Root I/O Virtualization support is experimental.
128 * All the related parameters may be subject to change.
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)`.
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)`.
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)`.
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)`.
150 * nvme namespace device parameters
151 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
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
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
167 * Setting `zoned` to true selects Zoned Command Set at the namespace.
168 * In this case, the following namespace properties are available to configure
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-.
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.
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.
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.
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.
189 * zoned.cross_read=<enable RAZB, default: false>
190 * Setting this property to true enables Read Across Zone Boundaries.
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"
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
227 #define NVME_GUEST_ERR(trace, fmt, ...) \
229 (trace_##trace)(__VA_ARGS__); \
230 qemu_log_mask(LOG_GUEST_ERROR, #trace \
231 " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
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,
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
,
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
,
284 static const uint32_t nvme_cse_iocs_none
[256];
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
,
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
,
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
);
317 static uint16_t nvme_sqid(NvmeRequest
*req
)
319 return le16_to_cpu(req
->sq
->sqid
);
322 static inline uint16_t nvme_make_pid(NvmeNamespace
*ns
, uint16_t rg
,
325 uint16_t rgif
= ns
->endgrp
->fdp
.rgif
;
331 return (rg
<< (16 - rgif
)) | ph
;
334 static inline bool nvme_ph_valid(NvmeNamespace
*ns
, uint16_t ph
)
336 return ph
< ns
->fdp
.nphs
;
339 static inline bool nvme_rg_valid(NvmeEnduranceGroup
*endgrp
, uint16_t rg
)
341 return rg
< endgrp
->fdp
.nrg
;
344 static inline uint16_t nvme_pid2ph(NvmeNamespace
*ns
, uint16_t pid
)
346 uint16_t rgif
= ns
->endgrp
->fdp
.rgif
;
352 return pid
& ((1 << (15 - rgif
)) - 1);
355 static inline uint16_t nvme_pid2rg(NvmeNamespace
*ns
, uint16_t pid
)
357 uint16_t rgif
= ns
->endgrp
->fdp
.rgif
;
363 return pid
>> (16 - rgif
);
366 static inline bool nvme_parse_pid(NvmeNamespace
*ns
, uint16_t pid
,
367 uint16_t *ph
, uint16_t *rg
)
369 *rg
= nvme_pid2rg(ns
, pid
);
370 *ph
= nvme_pid2ph(ns
, pid
);
372 return nvme_ph_valid(ns
, *ph
) && nvme_rg_valid(ns
->endgrp
, *rg
);
375 static void nvme_assign_zone_state(NvmeNamespace
*ns
, NvmeZone
*zone
,
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
);
383 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
384 QTAILQ_REMOVE(&ns
->imp_open_zones
, zone
, entry
);
386 case NVME_ZONE_STATE_CLOSED
:
387 QTAILQ_REMOVE(&ns
->closed_zones
, zone
, entry
);
389 case NVME_ZONE_STATE_FULL
:
390 QTAILQ_REMOVE(&ns
->full_zones
, zone
, entry
);
396 nvme_set_zone_state(zone
, state
);
399 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
400 QTAILQ_INSERT_TAIL(&ns
->exp_open_zones
, zone
, entry
);
402 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
403 QTAILQ_INSERT_TAIL(&ns
->imp_open_zones
, zone
, entry
);
405 case NVME_ZONE_STATE_CLOSED
:
406 QTAILQ_INSERT_TAIL(&ns
->closed_zones
, zone
, entry
);
408 case NVME_ZONE_STATE_FULL
:
409 QTAILQ_INSERT_TAIL(&ns
->full_zones
, zone
, entry
);
410 case NVME_ZONE_STATE_READ_ONLY
:
417 static uint16_t nvme_zns_check_resources(NvmeNamespace
*ns
, uint32_t act
,
418 uint32_t opn
, uint32_t zrwa
)
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
;
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
;
432 if (zrwa
> ns
->zns
.numzrwa
) {
433 return NVME_NOZRWA
| NVME_DNR
;
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).
443 static uint16_t nvme_aor_check(NvmeNamespace
*ns
, uint32_t act
, uint32_t opn
)
445 return nvme_zns_check_resources(ns
, act
, opn
, 0);
448 static NvmeFdpEvent
*nvme_fdp_alloc_event(NvmeCtrl
*n
, NvmeFdpEventBuffer
*ebuf
)
450 NvmeFdpEvent
*ret
= NULL
;
451 bool is_full
= ebuf
->next
== ebuf
->start
&& ebuf
->nelems
;
453 ret
= &ebuf
->events
[ebuf
->next
++];
454 if (unlikely(ebuf
->next
== NVME_FDP_MAX_EVENTS
)) {
458 ebuf
->start
= ebuf
->next
;
463 memset(ret
, 0, sizeof(NvmeFdpEvent
));
464 ret
->timestamp
= nvme_get_timestamp(n
);
469 static inline int log_event(NvmeRuHandle
*ruh
, uint8_t event_type
)
471 return (ruh
->event_filter
>> nvme_fdp_evf_shifts
[event_type
]) & 0x1;
474 static bool nvme_update_ruh(NvmeCtrl
*n
, NvmeNamespace
*ns
, uint16_t pid
)
476 NvmeEnduranceGroup
*endgrp
= ns
->endgrp
;
479 NvmeFdpEvent
*e
= NULL
;
480 uint16_t ph
, rg
, ruhid
;
482 if (!nvme_parse_pid(ns
, pid
, &ph
, &rg
)) {
486 ruhid
= ns
->fdp
.phs
[ph
];
488 ruh
= &endgrp
->fdp
.ruhs
[ruhid
];
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
);
502 /* log (eventual) GC overhead of prematurely swapping the RU */
503 nvme_fdp_stat_inc(&endgrp
->fdp
.mbmw
, nvme_l2b(ns
, ru
->ruamw
));
506 ru
->ruamw
= ruh
->ruamw
;
511 static bool nvme_addr_is_cmb(NvmeCtrl
*n
, hwaddr addr
)
519 lo
= n
->params
.legacy_cmb
? n
->cmb
.mem
.addr
: n
->cmb
.cba
;
520 hi
= lo
+ int128_get64(n
->cmb
.mem
.size
);
522 return addr
>= lo
&& addr
< hi
;
525 static inline void *nvme_addr_to_cmb(NvmeCtrl
*n
, hwaddr addr
)
527 hwaddr base
= n
->params
.legacy_cmb
? n
->cmb
.mem
.addr
: n
->cmb
.cba
;
528 return &n
->cmb
.buf
[addr
- base
];
531 static bool nvme_addr_is_pmr(NvmeCtrl
*n
, hwaddr addr
)
539 hi
= n
->pmr
.cba
+ int128_get64(n
->pmr
.dev
->mr
.size
);
541 return addr
>= n
->pmr
.cba
&& addr
< hi
;
544 static inline void *nvme_addr_to_pmr(NvmeCtrl
*n
, hwaddr addr
)
546 return memory_region_get_ram_ptr(&n
->pmr
.dev
->mr
) + (addr
- n
->pmr
.cba
);
549 static inline bool nvme_addr_is_iomem(NvmeCtrl
*n
, hwaddr addr
)
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.
562 hi
= lo
+ int128_get64(n
->bar0
.size
);
564 return addr
>= lo
&& addr
< hi
;
567 static int nvme_addr_read(NvmeCtrl
*n
, hwaddr addr
, void *buf
, int size
)
569 hwaddr hi
= addr
+ size
- 1;
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
);
579 if (nvme_addr_is_pmr(n
, addr
) && nvme_addr_is_pmr(n
, hi
)) {
580 memcpy(buf
, nvme_addr_to_pmr(n
, addr
), size
);
584 return pci_dma_read(PCI_DEVICE(n
), addr
, buf
, size
);
587 static int nvme_addr_write(NvmeCtrl
*n
, hwaddr addr
, const void *buf
, int size
)
589 hwaddr hi
= addr
+ size
- 1;
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
);
599 if (nvme_addr_is_pmr(n
, addr
) && nvme_addr_is_pmr(n
, hi
)) {
600 memcpy(nvme_addr_to_pmr(n
, addr
), buf
, size
);
604 return pci_dma_write(PCI_DEVICE(n
), addr
, buf
, size
);
607 static bool nvme_nsid_valid(NvmeCtrl
*n
, uint32_t nsid
)
610 (nsid
== NVME_NSID_BROADCAST
|| nsid
<= NVME_MAX_NAMESPACES
);
613 static int nvme_check_sqid(NvmeCtrl
*n
, uint16_t sqid
)
615 return sqid
< n
->conf_ioqpairs
+ 1 && n
->sq
[sqid
] != NULL
? 0 : -1;
618 static int nvme_check_cqid(NvmeCtrl
*n
, uint16_t cqid
)
620 return cqid
< n
->conf_ioqpairs
+ 1 && n
->cq
[cqid
] != NULL
? 0 : -1;
623 static void nvme_inc_cq_tail(NvmeCQueue
*cq
)
626 if (cq
->tail
>= cq
->size
) {
628 cq
->phase
= !cq
->phase
;
632 static void nvme_inc_sq_head(NvmeSQueue
*sq
)
634 sq
->head
= (sq
->head
+ 1) % sq
->size
;
637 static uint8_t nvme_cq_full(NvmeCQueue
*cq
)
639 return (cq
->tail
+ 1) % cq
->size
== cq
->head
;
642 static uint8_t nvme_sq_empty(NvmeSQueue
*sq
)
644 return sq
->head
== sq
->tail
;
647 static void nvme_irq_check(NvmeCtrl
*n
)
649 PCIDevice
*pci
= PCI_DEVICE(n
);
650 uint32_t intms
= ldl_le_p(&n
->bar
.intms
);
652 if (msix_enabled(pci
)) {
655 if (~intms
& n
->irq_status
) {
658 pci_irq_deassert(pci
);
662 static void nvme_irq_assert(NvmeCtrl
*n
, NvmeCQueue
*cq
)
664 PCIDevice
*pci
= PCI_DEVICE(n
);
666 if (cq
->irq_enabled
) {
667 if (msix_enabled(pci
)) {
668 trace_pci_nvme_irq_msix(cq
->vector
);
669 msix_notify(pci
, cq
->vector
);
671 trace_pci_nvme_irq_pin();
672 assert(cq
->vector
< 32);
673 n
->irq_status
|= 1 << cq
->vector
;
677 trace_pci_nvme_irq_masked();
681 static void nvme_irq_deassert(NvmeCtrl
*n
, NvmeCQueue
*cq
)
683 if (cq
->irq_enabled
) {
684 if (msix_enabled(PCI_DEVICE(n
))) {
687 assert(cq
->vector
< 32);
688 if (!n
->cq_pending
) {
689 n
->irq_status
&= ~(1 << cq
->vector
);
696 static void nvme_req_clear(NvmeRequest
*req
)
701 memset(&req
->cqe
, 0x0, sizeof(req
->cqe
));
702 req
->status
= NVME_SUCCESS
;
705 static inline void nvme_sg_init(NvmeCtrl
*n
, NvmeSg
*sg
, bool dma
)
708 pci_dma_sglist_init(&sg
->qsg
, PCI_DEVICE(n
), 0);
709 sg
->flags
= NVME_SG_DMA
;
711 qemu_iovec_init(&sg
->iov
, 0);
714 sg
->flags
|= NVME_SG_ALLOC
;
717 static inline void nvme_sg_unmap(NvmeSg
*sg
)
719 if (!(sg
->flags
& NVME_SG_ALLOC
)) {
723 if (sg
->flags
& NVME_SG_DMA
) {
724 qemu_sglist_destroy(&sg
->qsg
);
726 qemu_iovec_destroy(&sg
->iov
);
729 memset(sg
, 0x0, sizeof(*sg
));
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.
737 static void nvme_sg_split(NvmeSg
*sg
, NvmeNamespace
*ns
, NvmeSg
*data
,
741 uint32_t trans_len
, count
= ns
->lbasz
;
743 bool dma
= sg
->flags
& NVME_SG_DMA
;
745 size_t sg_len
= dma
? sg
->qsg
.size
: sg
->iov
.size
;
748 assert(sg
->flags
& NVME_SG_ALLOC
);
751 sge_len
= dma
? sg
->qsg
.sg
[sg_idx
].len
: sg
->iov
.iov
[sg_idx
].iov_len
;
753 trans_len
= MIN(sg_len
, count
);
754 trans_len
= MIN(trans_len
, sge_len
- offset
);
758 qemu_sglist_add(&dst
->qsg
, sg
->qsg
.sg
[sg_idx
].base
+ offset
,
761 qemu_iovec_add(&dst
->iov
,
762 sg
->iov
.iov
[sg_idx
].iov_base
+ offset
,
772 dst
= (dst
== data
) ? mdata
: data
;
773 count
= (dst
== data
) ? ns
->lbasz
: ns
->lbaf
.ms
;
776 if (sge_len
== offset
) {
783 static uint16_t nvme_map_addr_cmb(NvmeCtrl
*n
, QEMUIOVector
*iov
, hwaddr addr
,
790 trace_pci_nvme_map_addr_cmb(addr
, len
);
792 if (!nvme_addr_is_cmb(n
, addr
) || !nvme_addr_is_cmb(n
, addr
+ len
- 1)) {
793 return NVME_DATA_TRAS_ERROR
;
796 qemu_iovec_add(iov
, nvme_addr_to_cmb(n
, addr
), len
);
801 static uint16_t nvme_map_addr_pmr(NvmeCtrl
*n
, QEMUIOVector
*iov
, hwaddr addr
,
808 if (!nvme_addr_is_pmr(n
, addr
) || !nvme_addr_is_pmr(n
, addr
+ len
- 1)) {
809 return NVME_DATA_TRAS_ERROR
;
812 qemu_iovec_add(iov
, nvme_addr_to_pmr(n
, addr
), len
);
817 static uint16_t nvme_map_addr(NvmeCtrl
*n
, NvmeSg
*sg
, hwaddr addr
, size_t len
)
819 bool cmb
= false, pmr
= false;
825 trace_pci_nvme_map_addr(addr
, len
);
827 if (nvme_addr_is_iomem(n
, addr
)) {
828 return NVME_DATA_TRAS_ERROR
;
831 if (nvme_addr_is_cmb(n
, addr
)) {
833 } else if (nvme_addr_is_pmr(n
, addr
)) {
838 if (sg
->flags
& NVME_SG_DMA
) {
839 return NVME_INVALID_USE_OF_CMB
| NVME_DNR
;
842 if (sg
->iov
.niov
+ 1 > IOV_MAX
) {
843 goto max_mappings_exceeded
;
847 return nvme_map_addr_cmb(n
, &sg
->iov
, addr
, len
);
849 return nvme_map_addr_pmr(n
, &sg
->iov
, addr
, len
);
853 if (!(sg
->flags
& NVME_SG_DMA
)) {
854 return NVME_INVALID_USE_OF_CMB
| NVME_DNR
;
857 if (sg
->qsg
.nsg
+ 1 > IOV_MAX
) {
858 goto max_mappings_exceeded
;
861 qemu_sglist_add(&sg
->qsg
, addr
, len
);
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
;
871 static inline bool nvme_addr_is_dma(NvmeCtrl
*n
, hwaddr addr
)
873 return !(nvme_addr_is_cmb(n
, addr
) || nvme_addr_is_pmr(n
, addr
));
876 static uint16_t nvme_map_prp(NvmeCtrl
*n
, NvmeSg
*sg
, uint64_t prp1
,
877 uint64_t prp2
, uint32_t len
)
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;
885 trace_pci_nvme_map_prp(trans_len
, len
, prp1
, prp2
, num_prps
);
887 nvme_sg_init(n
, sg
, nvme_addr_is_dma(n
, prp1
));
889 status
= nvme_map_addr(n
, sg
, prp1
, trans_len
);
896 if (len
> n
->page_size
) {
897 uint64_t prp_list
[n
->max_prp_ents
];
898 uint32_t nents
, prp_trans
;
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
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
);
910 trace_pci_nvme_err_addr_read(prp2
);
911 status
= NVME_DATA_TRAS_ERROR
;
915 uint64_t prp_ent
= le64_to_cpu(prp_list
[i
]);
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
;
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
,
931 trace_pci_nvme_err_addr_read(prp_ent
);
932 status
= NVME_DATA_TRAS_ERROR
;
935 prp_ent
= le64_to_cpu(prp_list
[i
]);
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
;
944 trans_len
= MIN(len
, n
->page_size
);
945 status
= nvme_map_addr(n
, sg
, prp_ent
, trans_len
);
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
;
959 status
= nvme_map_addr(n
, sg
, prp2
, len
);
974 * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
975 * number of bytes mapped in len.
977 static uint16_t nvme_map_sgl_data(NvmeCtrl
*n
, NvmeSg
*sg
,
978 NvmeSglDescriptor
*segment
, uint64_t nsgld
,
979 size_t *len
, NvmeCmd
*cmd
)
981 dma_addr_t addr
, trans_len
;
985 for (int i
= 0; i
< nsgld
; i
++) {
986 uint8_t type
= NVME_SGL_TYPE(segment
[i
].type
);
989 case NVME_SGL_DESCR_TYPE_DATA_BLOCK
:
991 case NVME_SGL_DESCR_TYPE_SEGMENT
:
992 case NVME_SGL_DESCR_TYPE_LAST_SEGMENT
:
993 return NVME_INVALID_NUM_SGL_DESCRS
| NVME_DNR
;
995 return NVME_SGL_DESCR_TYPE_INVALID
| NVME_DNR
;
998 dlen
= le32_to_cpu(segment
[i
].len
);
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.
1010 uint32_t sgls
= le32_to_cpu(n
->id_ctrl
.sgls
);
1011 if (sgls
& NVME_CTRL_SGLS_EXCESS_LENGTH
) {
1015 trace_pci_nvme_err_invalid_sgl_excess_length(dlen
);
1016 return NVME_DATA_SGL_LEN_INVALID
| NVME_DNR
;
1019 trans_len
= MIN(*len
, dlen
);
1021 addr
= le64_to_cpu(segment
[i
].addr
);
1023 if (UINT64_MAX
- addr
< dlen
) {
1024 return NVME_DATA_SGL_LEN_INVALID
| NVME_DNR
;
1027 status
= nvme_map_addr(n
, sg
, addr
, trans_len
);
1035 return NVME_SUCCESS
;
1038 static uint16_t nvme_map_sgl(NvmeCtrl
*n
, NvmeSg
*sg
, NvmeSglDescriptor sgl
,
1039 size_t len
, NvmeCmd
*cmd
)
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.
1048 const int SEG_CHUNK_SIZE
= 256;
1050 NvmeSglDescriptor segment
[SEG_CHUNK_SIZE
], *sgld
, *last_sgld
;
1058 addr
= le64_to_cpu(sgl
.addr
);
1060 trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl
.type
), len
);
1062 nvme_sg_init(n
, sg
, nvme_addr_is_dma(n
, addr
));
1065 * If the entire transfer can be described with a single data block it can
1066 * be mapped directly.
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
);
1078 switch (NVME_SGL_TYPE(sgld
->type
)) {
1079 case NVME_SGL_DESCR_TYPE_SEGMENT
:
1080 case NVME_SGL_DESCR_TYPE_LAST_SEGMENT
:
1083 return NVME_INVALID_SGL_SEG_DESCR
| NVME_DNR
;
1086 seg_len
= le32_to_cpu(sgld
->len
);
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
;
1093 if (UINT64_MAX
- addr
< seg_len
) {
1094 return NVME_DATA_SGL_LEN_INVALID
| NVME_DNR
;
1097 nsgld
= seg_len
/ sizeof(NvmeSglDescriptor
);
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
;
1106 status
= nvme_map_sgl_data(n
, sg
, segment
, SEG_CHUNK_SIZE
,
1112 nsgld
-= SEG_CHUNK_SIZE
;
1113 addr
+= SEG_CHUNK_SIZE
* sizeof(NvmeSglDescriptor
);
1116 ret
= nvme_addr_read(n
, addr
, segment
, nsgld
*
1117 sizeof(NvmeSglDescriptor
));
1119 trace_pci_nvme_err_addr_read(addr
);
1120 status
= NVME_DATA_TRAS_ERROR
;
1124 last_sgld
= &segment
[nsgld
- 1];
1127 * If the segment ends with a Data Block, then we are done.
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
);
1139 * If the last descriptor was not a Data Block, then the current
1140 * segment must not be a Last Segment.
1142 if (NVME_SGL_TYPE(sgld
->type
) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT
) {
1143 status
= NVME_INVALID_SGL_SEG_DESCR
| NVME_DNR
;
1148 addr
= le64_to_cpu(sgld
->addr
);
1151 * Do not map the last descriptor; it will be a Segment or Last Segment
1152 * descriptor and is handled by the next iteration.
1154 status
= nvme_map_sgl_data(n
, sg
, segment
, nsgld
- 1, &len
, cmd
);
1161 /* if there is any residual left in len, the SGL was too short */
1163 status
= NVME_DATA_SGL_LEN_INVALID
| NVME_DNR
;
1167 return NVME_SUCCESS
;
1174 uint16_t nvme_map_dptr(NvmeCtrl
*n
, NvmeSg
*sg
, size_t len
,
1177 uint64_t prp1
, prp2
;
1179 switch (NVME_CMD_FLAGS_PSDT(cmd
->flags
)) {
1181 prp1
= le64_to_cpu(cmd
->dptr
.prp1
);
1182 prp2
= le64_to_cpu(cmd
->dptr
.prp2
);
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
);
1189 return NVME_INVALID_FIELD
;
1193 static uint16_t nvme_map_mptr(NvmeCtrl
*n
, NvmeSg
*sg
, size_t len
,
1196 int psdt
= NVME_CMD_FLAGS_PSDT(cmd
->flags
);
1197 hwaddr mptr
= le64_to_cpu(cmd
->mptr
);
1200 if (psdt
== NVME_PSDT_SGL_MPTR_SGL
) {
1201 NvmeSglDescriptor sgl
;
1203 if (nvme_addr_read(n
, mptr
, &sgl
, sizeof(sgl
))) {
1204 return NVME_DATA_TRAS_ERROR
;
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
;
1215 nvme_sg_init(n
, sg
, nvme_addr_is_dma(n
, mptr
));
1216 status
= nvme_map_addr(n
, sg
, mptr
, len
);
1224 static uint16_t nvme_map_data(NvmeCtrl
*n
, uint32_t nlb
, NvmeRequest
*req
)
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
);
1233 if (nvme_ns_ext(ns
) &&
1234 !(pi
&& pract
&& ns
->lbaf
.ms
== nvme_pi_tuple_size(ns
))) {
1237 len
+= nvme_m2b(ns
, nlb
);
1239 status
= nvme_map_dptr(n
, &sg
, len
, &req
->cmd
);
1244 nvme_sg_init(n
, &req
->sg
, sg
.flags
& NVME_SG_DMA
);
1245 nvme_sg_split(&sg
, ns
, &req
->sg
, NULL
);
1248 return NVME_SUCCESS
;
1251 return nvme_map_dptr(n
, &req
->sg
, len
, &req
->cmd
);
1254 static uint16_t nvme_map_mdata(NvmeCtrl
*n
, uint32_t nlb
, NvmeRequest
*req
)
1256 NvmeNamespace
*ns
= req
->ns
;
1257 size_t len
= nvme_m2b(ns
, nlb
);
1260 if (nvme_ns_ext(ns
)) {
1263 len
+= nvme_l2b(ns
, nlb
);
1265 status
= nvme_map_dptr(n
, &sg
, len
, &req
->cmd
);
1270 nvme_sg_init(n
, &req
->sg
, sg
.flags
& NVME_SG_DMA
);
1271 nvme_sg_split(&sg
, ns
, NULL
, &req
->sg
);
1274 return NVME_SUCCESS
;
1277 return nvme_map_mptr(n
, &req
->sg
, len
, &req
->cmd
);
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
)
1286 uint32_t trans_len
, count
= bytes
;
1287 bool dma
= sg
->flags
& NVME_SG_DMA
;
1292 assert(sg
->flags
& NVME_SG_ALLOC
);
1295 sge_len
= dma
? sg
->qsg
.sg
[sg_idx
].len
: sg
->iov
.iov
[sg_idx
].iov_len
;
1297 if (sge_len
- offset
< 0) {
1303 if (sge_len
== offset
) {
1309 trans_len
= MIN(len
, count
);
1310 trans_len
= MIN(trans_len
, sge_len
- offset
);
1313 addr
= sg
->qsg
.sg
[sg_idx
].base
+ offset
;
1315 addr
= (hwaddr
)(uintptr_t)sg
->iov
.iov
[sg_idx
].iov_base
+ offset
;
1318 if (dir
== NVME_TX_DIRECTION_TO_DEVICE
) {
1319 ret
= nvme_addr_read(n
, addr
, ptr
, trans_len
);
1321 ret
= nvme_addr_write(n
, addr
, ptr
, trans_len
);
1325 return NVME_DATA_TRAS_ERROR
;
1331 offset
+= trans_len
;
1335 offset
+= skip_bytes
;
1339 return NVME_SUCCESS
;
1342 static uint16_t nvme_tx(NvmeCtrl
*n
, NvmeSg
*sg
, void *ptr
, uint32_t len
,
1343 NvmeTxDirection dir
)
1345 assert(sg
->flags
& NVME_SG_ALLOC
);
1347 if (sg
->flags
& NVME_SG_DMA
) {
1348 const MemTxAttrs attrs
= MEMTXATTRS_UNSPECIFIED
;
1349 dma_addr_t residual
;
1351 if (dir
== NVME_TX_DIRECTION_TO_DEVICE
) {
1352 dma_buf_write(ptr
, len
, &residual
, &sg
->qsg
, attrs
);
1354 dma_buf_read(ptr
, len
, &residual
, &sg
->qsg
, attrs
);
1357 if (unlikely(residual
)) {
1358 trace_pci_nvme_err_invalid_dma();
1359 return NVME_INVALID_FIELD
| NVME_DNR
;
1364 if (dir
== NVME_TX_DIRECTION_TO_DEVICE
) {
1365 bytes
= qemu_iovec_to_buf(&sg
->iov
, 0, ptr
, len
);
1367 bytes
= qemu_iovec_from_buf(&sg
->iov
, 0, ptr
, len
);
1370 if (unlikely(bytes
!= len
)) {
1371 trace_pci_nvme_err_invalid_dma();
1372 return NVME_INVALID_FIELD
| NVME_DNR
;
1376 return NVME_SUCCESS
;
1379 static inline uint16_t nvme_c2h(NvmeCtrl
*n
, void *ptr
, uint32_t len
,
1384 status
= nvme_map_dptr(n
, &req
->sg
, len
, &req
->cmd
);
1389 return nvme_tx(n
, &req
->sg
, ptr
, len
, NVME_TX_DIRECTION_FROM_DEVICE
);
1392 static inline uint16_t nvme_h2c(NvmeCtrl
*n
, void *ptr
, uint32_t len
,
1397 status
= nvme_map_dptr(n
, &req
->sg
, len
, &req
->cmd
);
1402 return nvme_tx(n
, &req
->sg
, ptr
, len
, NVME_TX_DIRECTION_TO_DEVICE
);
1405 uint16_t nvme_bounce_data(NvmeCtrl
*n
, void *ptr
, uint32_t len
,
1406 NvmeTxDirection dir
, NvmeRequest
*req
)
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
);
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
);
1419 return nvme_tx(n
, &req
->sg
, ptr
, len
, dir
);
1422 uint16_t nvme_bounce_mdata(NvmeCtrl
*n
, void *ptr
, uint32_t len
,
1423 NvmeTxDirection dir
, NvmeRequest
*req
)
1425 NvmeNamespace
*ns
= req
->ns
;
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
);
1433 nvme_sg_unmap(&req
->sg
);
1435 status
= nvme_map_mptr(n
, &req
->sg
, len
, &req
->cmd
);
1440 return nvme_tx(n
, &req
->sg
, ptr
, len
, dir
);
1443 static inline void nvme_blk_read(BlockBackend
*blk
, int64_t offset
,
1444 uint32_t align
, BlockCompletionFunc
*cb
,
1447 assert(req
->sg
.flags
& NVME_SG_ALLOC
);
1449 if (req
->sg
.flags
& NVME_SG_DMA
) {
1450 req
->aiocb
= dma_blk_read(blk
, &req
->sg
.qsg
, offset
, align
, cb
, req
);
1452 req
->aiocb
= blk_aio_preadv(blk
, offset
, &req
->sg
.iov
, 0, cb
, req
);
1456 static inline void nvme_blk_write(BlockBackend
*blk
, int64_t offset
,
1457 uint32_t align
, BlockCompletionFunc
*cb
,
1460 assert(req
->sg
.flags
& NVME_SG_ALLOC
);
1462 if (req
->sg
.flags
& NVME_SG_DMA
) {
1463 req
->aiocb
= dma_blk_write(blk
, &req
->sg
.qsg
, offset
, align
, cb
, req
);
1465 req
->aiocb
= blk_aio_pwritev(blk
, offset
, &req
->sg
.iov
, 0, cb
, req
);
1469 static void nvme_update_cq_eventidx(const NvmeCQueue
*cq
)
1471 uint32_t v
= cpu_to_le32(cq
->head
);
1473 trace_pci_nvme_update_cq_eventidx(cq
->cqid
, cq
->head
);
1475 pci_dma_write(PCI_DEVICE(cq
->ctrl
), cq
->ei_addr
, &v
, sizeof(v
));
1478 static void nvme_update_cq_head(NvmeCQueue
*cq
)
1482 pci_dma_read(PCI_DEVICE(cq
->ctrl
), cq
->db_addr
, &v
, sizeof(v
));
1484 cq
->head
= le32_to_cpu(v
);
1486 trace_pci_nvme_update_cq_head(cq
->cqid
, cq
->head
);
1489 static void nvme_post_cqes(void *opaque
)
1491 NvmeCQueue
*cq
= opaque
;
1492 NvmeCtrl
*n
= cq
->ctrl
;
1493 NvmeRequest
*req
, *next
;
1494 bool pending
= cq
->head
!= cq
->tail
;
1497 QTAILQ_FOREACH_SAFE(req
, &cq
->req_list
, entry
, next
) {
1501 if (n
->dbbuf_enabled
) {
1502 nvme_update_cq_eventidx(cq
);
1503 nvme_update_cq_head(cq
);
1506 if (nvme_cq_full(cq
)) {
1511 req
->cqe
.status
= cpu_to_le16((req
->status
<< 1) | cq
->phase
);
1512 req
->cqe
.sq_id
= cpu_to_le16(sq
->sqid
);
1513 req
->cqe
.sq_head
= cpu_to_le16(sq
->head
);
1514 addr
= cq
->dma_addr
+ cq
->tail
* n
->cqe_size
;
1515 ret
= pci_dma_write(PCI_DEVICE(n
), addr
, (void *)&req
->cqe
,
1518 trace_pci_nvme_err_addr_write(addr
);
1519 trace_pci_nvme_err_cfs();
1520 stl_le_p(&n
->bar
.csts
, NVME_CSTS_FAILED
);
1523 QTAILQ_REMOVE(&cq
->req_list
, req
, entry
);
1524 nvme_inc_cq_tail(cq
);
1525 nvme_sg_unmap(&req
->sg
);
1526 QTAILQ_INSERT_TAIL(&sq
->req_list
, req
, entry
);
1528 if (cq
->tail
!= cq
->head
) {
1529 if (cq
->irq_enabled
&& !pending
) {
1533 nvme_irq_assert(n
, cq
);
1537 static void nvme_enqueue_req_completion(NvmeCQueue
*cq
, NvmeRequest
*req
)
1539 assert(cq
->cqid
== req
->sq
->cqid
);
1540 trace_pci_nvme_enqueue_req_completion(nvme_cid(req
), cq
->cqid
,
1541 le32_to_cpu(req
->cqe
.result
),
1542 le32_to_cpu(req
->cqe
.dw1
),
1546 trace_pci_nvme_err_req_status(nvme_cid(req
), nvme_nsid(req
->ns
),
1547 req
->status
, req
->cmd
.opcode
);
1550 QTAILQ_REMOVE(&req
->sq
->out_req_list
, req
, entry
);
1551 QTAILQ_INSERT_TAIL(&cq
->req_list
, req
, entry
);
1553 qemu_bh_schedule(cq
->bh
);
1556 static void nvme_process_aers(void *opaque
)
1558 NvmeCtrl
*n
= opaque
;
1559 NvmeAsyncEvent
*event
, *next
;
1561 trace_pci_nvme_process_aers(n
->aer_queued
);
1563 QTAILQ_FOREACH_SAFE(event
, &n
->aer_queue
, entry
, next
) {
1565 NvmeAerResult
*result
;
1567 /* can't post cqe if there is nothing to complete */
1568 if (!n
->outstanding_aers
) {
1569 trace_pci_nvme_no_outstanding_aers();
1573 /* ignore if masked (cqe posted, but event not cleared) */
1574 if (n
->aer_mask
& (1 << event
->result
.event_type
)) {
1575 trace_pci_nvme_aer_masked(event
->result
.event_type
, n
->aer_mask
);
1579 QTAILQ_REMOVE(&n
->aer_queue
, event
, entry
);
1582 n
->aer_mask
|= 1 << event
->result
.event_type
;
1583 n
->outstanding_aers
--;
1585 req
= n
->aer_reqs
[n
->outstanding_aers
];
1587 result
= (NvmeAerResult
*) &req
->cqe
.result
;
1588 result
->event_type
= event
->result
.event_type
;
1589 result
->event_info
= event
->result
.event_info
;
1590 result
->log_page
= event
->result
.log_page
;
1593 trace_pci_nvme_aer_post_cqe(result
->event_type
, result
->event_info
,
1596 nvme_enqueue_req_completion(&n
->admin_cq
, req
);
1600 static void nvme_enqueue_event(NvmeCtrl
*n
, uint8_t event_type
,
1601 uint8_t event_info
, uint8_t log_page
)
1603 NvmeAsyncEvent
*event
;
1605 trace_pci_nvme_enqueue_event(event_type
, event_info
, log_page
);
1607 if (n
->aer_queued
== n
->params
.aer_max_queued
) {
1608 trace_pci_nvme_enqueue_event_noqueue(n
->aer_queued
);
1612 event
= g_new(NvmeAsyncEvent
, 1);
1613 event
->result
= (NvmeAerResult
) {
1614 .event_type
= event_type
,
1615 .event_info
= event_info
,
1616 .log_page
= log_page
,
1619 QTAILQ_INSERT_TAIL(&n
->aer_queue
, event
, entry
);
1622 nvme_process_aers(n
);
1625 static void nvme_smart_event(NvmeCtrl
*n
, uint8_t event
)
1629 /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
1630 if (!(NVME_AEC_SMART(n
->features
.async_config
) & event
)) {
1635 case NVME_SMART_SPARE
:
1636 aer_info
= NVME_AER_INFO_SMART_SPARE_THRESH
;
1638 case NVME_SMART_TEMPERATURE
:
1639 aer_info
= NVME_AER_INFO_SMART_TEMP_THRESH
;
1641 case NVME_SMART_RELIABILITY
:
1642 case NVME_SMART_MEDIA_READ_ONLY
:
1643 case NVME_SMART_FAILED_VOLATILE_MEDIA
:
1644 case NVME_SMART_PMR_UNRELIABLE
:
1645 aer_info
= NVME_AER_INFO_SMART_RELIABILITY
;
1651 nvme_enqueue_event(n
, NVME_AER_TYPE_SMART
, aer_info
, NVME_LOG_SMART_INFO
);
1654 static void nvme_clear_events(NvmeCtrl
*n
, uint8_t event_type
)
1656 n
->aer_mask
&= ~(1 << event_type
);
1657 if (!QTAILQ_EMPTY(&n
->aer_queue
)) {
1658 nvme_process_aers(n
);
1662 static inline uint16_t nvme_check_mdts(NvmeCtrl
*n
, size_t len
)
1664 uint8_t mdts
= n
->params
.mdts
;
1666 if (mdts
&& len
> n
->page_size
<< mdts
) {
1667 trace_pci_nvme_err_mdts(len
);
1668 return NVME_INVALID_FIELD
| NVME_DNR
;
1671 return NVME_SUCCESS
;
1674 static inline uint16_t nvme_check_bounds(NvmeNamespace
*ns
, uint64_t slba
,
1677 uint64_t nsze
= le64_to_cpu(ns
->id_ns
.nsze
);
1679 if (unlikely(UINT64_MAX
- slba
< nlb
|| slba
+ nlb
> nsze
)) {
1680 trace_pci_nvme_err_invalid_lba_range(slba
, nlb
, nsze
);
1681 return NVME_LBA_RANGE
| NVME_DNR
;
1684 return NVME_SUCCESS
;
1687 static int nvme_block_status_all(NvmeNamespace
*ns
, uint64_t slba
,
1688 uint32_t nlb
, int flags
)
1690 BlockDriverState
*bs
= blk_bs(ns
->blkconf
.blk
);
1692 int64_t pnum
= 0, bytes
= nvme_l2b(ns
, nlb
);
1693 int64_t offset
= nvme_l2b(ns
, slba
);
1697 * `pnum` holds the number of bytes after offset that shares the same
1698 * allocation status as the byte at offset. If `pnum` is different from
1699 * `bytes`, we should check the allocation status of the next range and
1700 * continue this until all bytes have been checked.
1705 ret
= bdrv_block_status(bs
, offset
, bytes
, &pnum
, NULL
, NULL
);
1711 trace_pci_nvme_block_status(offset
, bytes
, pnum
, ret
,
1712 !!(ret
& BDRV_BLOCK_ZERO
));
1714 if (!(ret
& flags
)) {
1719 } while (pnum
!= bytes
);
1724 static uint16_t nvme_check_dulbe(NvmeNamespace
*ns
, uint64_t slba
,
1730 ret
= nvme_block_status_all(ns
, slba
, nlb
, BDRV_BLOCK_DATA
);
1733 error_setg_errno(&err
, -ret
, "unable to get block status");
1734 error_report_err(err
);
1736 return NVME_INTERNAL_DEV_ERROR
;
1742 return NVME_SUCCESS
;
1745 static void nvme_aio_err(NvmeRequest
*req
, int ret
)
1747 uint16_t status
= NVME_SUCCESS
;
1748 Error
*local_err
= NULL
;
1750 switch (req
->cmd
.opcode
) {
1752 status
= NVME_UNRECOVERED_READ
;
1754 case NVME_CMD_FLUSH
:
1755 case NVME_CMD_WRITE
:
1756 case NVME_CMD_WRITE_ZEROES
:
1757 case NVME_CMD_ZONE_APPEND
:
1759 status
= NVME_WRITE_FAULT
;
1762 status
= NVME_INTERNAL_DEV_ERROR
;
1766 trace_pci_nvme_err_aio(nvme_cid(req
), strerror(-ret
), status
);
1768 error_setg_errno(&local_err
, -ret
, "aio failed");
1769 error_report_err(local_err
);
1772 * Set the command status code to the first encountered error but allow a
1773 * subsequent Internal Device Error to trump it.
1775 if (req
->status
&& status
!= NVME_INTERNAL_DEV_ERROR
) {
1779 req
->status
= status
;
1782 static inline uint32_t nvme_zone_idx(NvmeNamespace
*ns
, uint64_t slba
)
1784 return ns
->zone_size_log2
> 0 ? slba
>> ns
->zone_size_log2
:
1785 slba
/ ns
->zone_size
;
1788 static inline NvmeZone
*nvme_get_zone_by_slba(NvmeNamespace
*ns
, uint64_t slba
)
1790 uint32_t zone_idx
= nvme_zone_idx(ns
, slba
);
1792 if (zone_idx
>= ns
->num_zones
) {
1796 return &ns
->zone_array
[zone_idx
];
1799 static uint16_t nvme_check_zone_state_for_write(NvmeZone
*zone
)
1801 uint64_t zslba
= zone
->d
.zslba
;
1803 switch (nvme_get_zone_state(zone
)) {
1804 case NVME_ZONE_STATE_EMPTY
:
1805 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1806 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1807 case NVME_ZONE_STATE_CLOSED
:
1808 return NVME_SUCCESS
;
1809 case NVME_ZONE_STATE_FULL
:
1810 trace_pci_nvme_err_zone_is_full(zslba
);
1811 return NVME_ZONE_FULL
;
1812 case NVME_ZONE_STATE_OFFLINE
:
1813 trace_pci_nvme_err_zone_is_offline(zslba
);
1814 return NVME_ZONE_OFFLINE
;
1815 case NVME_ZONE_STATE_READ_ONLY
:
1816 trace_pci_nvme_err_zone_is_read_only(zslba
);
1817 return NVME_ZONE_READ_ONLY
;
1822 return NVME_INTERNAL_DEV_ERROR
;
1825 static uint16_t nvme_check_zone_write(NvmeNamespace
*ns
, NvmeZone
*zone
,
1826 uint64_t slba
, uint32_t nlb
)
1828 uint64_t zcap
= nvme_zone_wr_boundary(zone
);
1831 status
= nvme_check_zone_state_for_write(zone
);
1836 if (zone
->d
.za
& NVME_ZA_ZRWA_VALID
) {
1837 uint64_t ezrwa
= zone
->w_ptr
+ 2 * ns
->zns
.zrwas
;
1839 if (slba
< zone
->w_ptr
|| slba
+ nlb
> ezrwa
) {
1840 trace_pci_nvme_err_zone_invalid_write(slba
, zone
->w_ptr
);
1841 return NVME_ZONE_INVALID_WRITE
;
1844 if (unlikely(slba
!= zone
->w_ptr
)) {
1845 trace_pci_nvme_err_write_not_at_wp(slba
, zone
->d
.zslba
,
1847 return NVME_ZONE_INVALID_WRITE
;
1851 if (unlikely((slba
+ nlb
) > zcap
)) {
1852 trace_pci_nvme_err_zone_boundary(slba
, nlb
, zcap
);
1853 return NVME_ZONE_BOUNDARY_ERROR
;
1856 return NVME_SUCCESS
;
1859 static uint16_t nvme_check_zone_state_for_read(NvmeZone
*zone
)
1861 switch (nvme_get_zone_state(zone
)) {
1862 case NVME_ZONE_STATE_EMPTY
:
1863 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1864 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1865 case NVME_ZONE_STATE_FULL
:
1866 case NVME_ZONE_STATE_CLOSED
:
1867 case NVME_ZONE_STATE_READ_ONLY
:
1868 return NVME_SUCCESS
;
1869 case NVME_ZONE_STATE_OFFLINE
:
1870 trace_pci_nvme_err_zone_is_offline(zone
->d
.zslba
);
1871 return NVME_ZONE_OFFLINE
;
1876 return NVME_INTERNAL_DEV_ERROR
;
1879 static uint16_t nvme_check_zone_read(NvmeNamespace
*ns
, uint64_t slba
,
1883 uint64_t bndry
, end
;
1886 zone
= nvme_get_zone_by_slba(ns
, slba
);
1889 bndry
= nvme_zone_rd_boundary(ns
, zone
);
1892 status
= nvme_check_zone_state_for_read(zone
);
1895 } else if (unlikely(end
> bndry
)) {
1896 if (!ns
->params
.cross_zone_read
) {
1897 status
= NVME_ZONE_BOUNDARY_ERROR
;
1900 * Read across zone boundary - check that all subsequent
1901 * zones that are being read have an appropriate state.
1905 status
= nvme_check_zone_state_for_read(zone
);
1909 } while (end
> nvme_zone_rd_boundary(ns
, zone
));
1916 static uint16_t nvme_zrm_finish(NvmeNamespace
*ns
, NvmeZone
*zone
)
1918 switch (nvme_get_zone_state(zone
)) {
1919 case NVME_ZONE_STATE_FULL
:
1920 return NVME_SUCCESS
;
1922 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1923 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1924 nvme_aor_dec_open(ns
);
1926 case NVME_ZONE_STATE_CLOSED
:
1927 nvme_aor_dec_active(ns
);
1929 if (zone
->d
.za
& NVME_ZA_ZRWA_VALID
) {
1930 zone
->d
.za
&= ~NVME_ZA_ZRWA_VALID
;
1931 if (ns
->params
.numzrwa
) {
1937 case NVME_ZONE_STATE_EMPTY
:
1938 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_FULL
);
1939 return NVME_SUCCESS
;
1942 return NVME_ZONE_INVAL_TRANSITION
;
1946 static uint16_t nvme_zrm_close(NvmeNamespace
*ns
, NvmeZone
*zone
)
1948 switch (nvme_get_zone_state(zone
)) {
1949 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1950 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1951 nvme_aor_dec_open(ns
);
1952 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_CLOSED
);
1954 case NVME_ZONE_STATE_CLOSED
:
1955 return NVME_SUCCESS
;
1958 return NVME_ZONE_INVAL_TRANSITION
;
1962 static uint16_t nvme_zrm_reset(NvmeNamespace
*ns
, NvmeZone
*zone
)
1964 switch (nvme_get_zone_state(zone
)) {
1965 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1966 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1967 nvme_aor_dec_open(ns
);
1969 case NVME_ZONE_STATE_CLOSED
:
1970 nvme_aor_dec_active(ns
);
1972 if (zone
->d
.za
& NVME_ZA_ZRWA_VALID
) {
1973 if (ns
->params
.numzrwa
) {
1979 case NVME_ZONE_STATE_FULL
:
1980 zone
->w_ptr
= zone
->d
.zslba
;
1981 zone
->d
.wp
= zone
->w_ptr
;
1982 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_EMPTY
);
1984 case NVME_ZONE_STATE_EMPTY
:
1985 return NVME_SUCCESS
;
1988 return NVME_ZONE_INVAL_TRANSITION
;
1992 static void nvme_zrm_auto_transition_zone(NvmeNamespace
*ns
)
1996 if (ns
->params
.max_open_zones
&&
1997 ns
->nr_open_zones
== ns
->params
.max_open_zones
) {
1998 zone
= QTAILQ_FIRST(&ns
->imp_open_zones
);
2001 * Automatically close this implicitly open zone.
2003 QTAILQ_REMOVE(&ns
->imp_open_zones
, zone
, entry
);
2004 nvme_zrm_close(ns
, zone
);
2010 NVME_ZRM_AUTO
= 1 << 0,
2011 NVME_ZRM_ZRWA
= 1 << 1,
2014 static uint16_t nvme_zrm_open_flags(NvmeCtrl
*n
, NvmeNamespace
*ns
,
2015 NvmeZone
*zone
, int flags
)
2020 switch (nvme_get_zone_state(zone
)) {
2021 case NVME_ZONE_STATE_EMPTY
:
2026 case NVME_ZONE_STATE_CLOSED
:
2027 if (n
->params
.auto_transition_zones
) {
2028 nvme_zrm_auto_transition_zone(ns
);
2030 status
= nvme_zns_check_resources(ns
, act
, 1,
2031 (flags
& NVME_ZRM_ZRWA
) ? 1 : 0);
2037 nvme_aor_inc_active(ns
);
2040 nvme_aor_inc_open(ns
);
2042 if (flags
& NVME_ZRM_AUTO
) {
2043 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_IMPLICITLY_OPEN
);
2044 return NVME_SUCCESS
;
2049 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
2050 if (flags
& NVME_ZRM_AUTO
) {
2051 return NVME_SUCCESS
;
2054 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_EXPLICITLY_OPEN
);
2058 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
2059 if (flags
& NVME_ZRM_ZRWA
) {
2062 zone
->d
.za
|= NVME_ZA_ZRWA_VALID
;
2065 return NVME_SUCCESS
;
2068 return NVME_ZONE_INVAL_TRANSITION
;
2072 static inline uint16_t nvme_zrm_auto(NvmeCtrl
*n
, NvmeNamespace
*ns
,
2075 return nvme_zrm_open_flags(n
, ns
, zone
, NVME_ZRM_AUTO
);
2078 static void nvme_advance_zone_wp(NvmeNamespace
*ns
, NvmeZone
*zone
,
2083 if (zone
->d
.wp
== nvme_zone_wr_boundary(zone
)) {
2084 nvme_zrm_finish(ns
, zone
);
2088 static void nvme_zoned_zrwa_implicit_flush(NvmeNamespace
*ns
, NvmeZone
*zone
,
2091 uint16_t nzrwafgs
= DIV_ROUND_UP(nlbc
, ns
->zns
.zrwafg
);
2093 nlbc
= nzrwafgs
* ns
->zns
.zrwafg
;
2095 trace_pci_nvme_zoned_zrwa_implicit_flush(zone
->d
.zslba
, nlbc
);
2097 zone
->w_ptr
+= nlbc
;
2099 nvme_advance_zone_wp(ns
, zone
, nlbc
);
2102 static void nvme_finalize_zoned_write(NvmeNamespace
*ns
, NvmeRequest
*req
)
2104 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2109 slba
= le64_to_cpu(rw
->slba
);
2110 nlb
= le16_to_cpu(rw
->nlb
) + 1;
2111 zone
= nvme_get_zone_by_slba(ns
, slba
);
2114 if (zone
->d
.za
& NVME_ZA_ZRWA_VALID
) {
2115 uint64_t ezrwa
= zone
->w_ptr
+ ns
->zns
.zrwas
- 1;
2116 uint64_t elba
= slba
+ nlb
- 1;
2119 nvme_zoned_zrwa_implicit_flush(ns
, zone
, elba
- ezrwa
);
2125 nvme_advance_zone_wp(ns
, zone
, nlb
);
2128 static inline bool nvme_is_write(NvmeRequest
*req
)
2130 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2132 return rw
->opcode
== NVME_CMD_WRITE
||
2133 rw
->opcode
== NVME_CMD_ZONE_APPEND
||
2134 rw
->opcode
== NVME_CMD_WRITE_ZEROES
;
2137 static AioContext
*nvme_get_aio_context(BlockAIOCB
*acb
)
2139 return qemu_get_aio_context();
2142 static void nvme_misc_cb(void *opaque
, int ret
)
2144 NvmeRequest
*req
= opaque
;
2146 trace_pci_nvme_misc_cb(nvme_cid(req
));
2149 nvme_aio_err(req
, ret
);
2152 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2155 void nvme_rw_complete_cb(void *opaque
, int ret
)
2157 NvmeRequest
*req
= opaque
;
2158 NvmeNamespace
*ns
= req
->ns
;
2159 BlockBackend
*blk
= ns
->blkconf
.blk
;
2160 BlockAcctCookie
*acct
= &req
->acct
;
2161 BlockAcctStats
*stats
= blk_get_stats(blk
);
2163 trace_pci_nvme_rw_complete_cb(nvme_cid(req
), blk_name(blk
));
2166 block_acct_failed(stats
, acct
);
2167 nvme_aio_err(req
, ret
);
2169 block_acct_done(stats
, acct
);
2172 if (ns
->params
.zoned
&& nvme_is_write(req
)) {
2173 nvme_finalize_zoned_write(ns
, req
);
2176 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2179 static void nvme_rw_cb(void *opaque
, int ret
)
2181 NvmeRequest
*req
= opaque
;
2182 NvmeNamespace
*ns
= req
->ns
;
2184 BlockBackend
*blk
= ns
->blkconf
.blk
;
2186 trace_pci_nvme_rw_cb(nvme_cid(req
), blk_name(blk
));
2193 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2194 uint64_t slba
= le64_to_cpu(rw
->slba
);
2195 uint32_t nlb
= (uint32_t)le16_to_cpu(rw
->nlb
) + 1;
2196 uint64_t offset
= nvme_moff(ns
, slba
);
2198 if (req
->cmd
.opcode
== NVME_CMD_WRITE_ZEROES
) {
2199 size_t mlen
= nvme_m2b(ns
, nlb
);
2201 req
->aiocb
= blk_aio_pwrite_zeroes(blk
, offset
, mlen
,
2203 nvme_rw_complete_cb
, req
);
2207 if (nvme_ns_ext(ns
) || req
->cmd
.mptr
) {
2210 nvme_sg_unmap(&req
->sg
);
2211 status
= nvme_map_mdata(nvme_ctrl(req
), nlb
, req
);
2217 if (req
->cmd
.opcode
== NVME_CMD_READ
) {
2218 return nvme_blk_read(blk
, offset
, 1, nvme_rw_complete_cb
, req
);
2221 return nvme_blk_write(blk
, offset
, 1, nvme_rw_complete_cb
, req
);
2226 nvme_rw_complete_cb(req
, ret
);
2229 static void nvme_verify_cb(void *opaque
, int ret
)
2231 NvmeBounceContext
*ctx
= opaque
;
2232 NvmeRequest
*req
= ctx
->req
;
2233 NvmeNamespace
*ns
= req
->ns
;
2234 BlockBackend
*blk
= ns
->blkconf
.blk
;
2235 BlockAcctCookie
*acct
= &req
->acct
;
2236 BlockAcctStats
*stats
= blk_get_stats(blk
);
2237 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2238 uint64_t slba
= le64_to_cpu(rw
->slba
);
2239 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
2240 uint16_t apptag
= le16_to_cpu(rw
->apptag
);
2241 uint16_t appmask
= le16_to_cpu(rw
->appmask
);
2242 uint64_t reftag
= le32_to_cpu(rw
->reftag
);
2243 uint64_t cdw3
= le32_to_cpu(rw
->cdw3
);
2246 reftag
|= cdw3
<< 32;
2248 trace_pci_nvme_verify_cb(nvme_cid(req
), prinfo
, apptag
, appmask
, reftag
);
2251 block_acct_failed(stats
, acct
);
2252 nvme_aio_err(req
, ret
);
2256 block_acct_done(stats
, acct
);
2258 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
2259 status
= nvme_dif_mangle_mdata(ns
, ctx
->mdata
.bounce
,
2260 ctx
->mdata
.iov
.size
, slba
);
2262 req
->status
= status
;
2266 req
->status
= nvme_dif_check(ns
, ctx
->data
.bounce
, ctx
->data
.iov
.size
,
2267 ctx
->mdata
.bounce
, ctx
->mdata
.iov
.size
,
2268 prinfo
, slba
, apptag
, appmask
, &reftag
);
2272 qemu_iovec_destroy(&ctx
->data
.iov
);
2273 g_free(ctx
->data
.bounce
);
2275 qemu_iovec_destroy(&ctx
->mdata
.iov
);
2276 g_free(ctx
->mdata
.bounce
);
2280 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2284 static void nvme_verify_mdata_in_cb(void *opaque
, int ret
)
2286 NvmeBounceContext
*ctx
= opaque
;
2287 NvmeRequest
*req
= ctx
->req
;
2288 NvmeNamespace
*ns
= req
->ns
;
2289 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2290 uint64_t slba
= le64_to_cpu(rw
->slba
);
2291 uint32_t nlb
= le16_to_cpu(rw
->nlb
) + 1;
2292 size_t mlen
= nvme_m2b(ns
, nlb
);
2293 uint64_t offset
= nvme_moff(ns
, slba
);
2294 BlockBackend
*blk
= ns
->blkconf
.blk
;
2296 trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req
), blk_name(blk
));
2302 ctx
->mdata
.bounce
= g_malloc(mlen
);
2304 qemu_iovec_reset(&ctx
->mdata
.iov
);
2305 qemu_iovec_add(&ctx
->mdata
.iov
, ctx
->mdata
.bounce
, mlen
);
2307 req
->aiocb
= blk_aio_preadv(blk
, offset
, &ctx
->mdata
.iov
, 0,
2308 nvme_verify_cb
, ctx
);
2312 nvme_verify_cb(ctx
, ret
);
2315 struct nvme_compare_ctx
{
2327 static void nvme_compare_mdata_cb(void *opaque
, int ret
)
2329 NvmeRequest
*req
= opaque
;
2330 NvmeNamespace
*ns
= req
->ns
;
2331 NvmeCtrl
*n
= nvme_ctrl(req
);
2332 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2333 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
2334 uint16_t apptag
= le16_to_cpu(rw
->apptag
);
2335 uint16_t appmask
= le16_to_cpu(rw
->appmask
);
2336 uint64_t reftag
= le32_to_cpu(rw
->reftag
);
2337 uint64_t cdw3
= le32_to_cpu(rw
->cdw3
);
2338 struct nvme_compare_ctx
*ctx
= req
->opaque
;
2339 g_autofree
uint8_t *buf
= NULL
;
2340 BlockBackend
*blk
= ns
->blkconf
.blk
;
2341 BlockAcctCookie
*acct
= &req
->acct
;
2342 BlockAcctStats
*stats
= blk_get_stats(blk
);
2343 uint16_t status
= NVME_SUCCESS
;
2345 reftag
|= cdw3
<< 32;
2347 trace_pci_nvme_compare_mdata_cb(nvme_cid(req
));
2350 block_acct_failed(stats
, acct
);
2351 nvme_aio_err(req
, ret
);
2355 buf
= g_malloc(ctx
->mdata
.iov
.size
);
2357 status
= nvme_bounce_mdata(n
, buf
, ctx
->mdata
.iov
.size
,
2358 NVME_TX_DIRECTION_TO_DEVICE
, req
);
2360 req
->status
= status
;
2364 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
2365 uint64_t slba
= le64_to_cpu(rw
->slba
);
2367 uint8_t *mbufp
= ctx
->mdata
.bounce
;
2368 uint8_t *end
= mbufp
+ ctx
->mdata
.iov
.size
;
2371 status
= nvme_dif_check(ns
, ctx
->data
.bounce
, ctx
->data
.iov
.size
,
2372 ctx
->mdata
.bounce
, ctx
->mdata
.iov
.size
, prinfo
,
2373 slba
, apptag
, appmask
, &reftag
);
2375 req
->status
= status
;
2380 * When formatted with protection information, do not compare the DIF
2383 if (!(ns
->id_ns
.dps
& NVME_ID_NS_DPS_FIRST_EIGHT
)) {
2384 pil
= ns
->lbaf
.ms
- nvme_pi_tuple_size(ns
);
2387 for (bufp
= buf
; mbufp
< end
; bufp
+= ns
->lbaf
.ms
, mbufp
+= ns
->lbaf
.ms
) {
2388 if (memcmp(bufp
+ pil
, mbufp
+ pil
, ns
->lbaf
.ms
- pil
)) {
2389 req
->status
= NVME_CMP_FAILURE
| NVME_DNR
;
2397 if (memcmp(buf
, ctx
->mdata
.bounce
, ctx
->mdata
.iov
.size
)) {
2398 req
->status
= NVME_CMP_FAILURE
| NVME_DNR
;
2402 block_acct_done(stats
, acct
);
2405 qemu_iovec_destroy(&ctx
->data
.iov
);
2406 g_free(ctx
->data
.bounce
);
2408 qemu_iovec_destroy(&ctx
->mdata
.iov
);
2409 g_free(ctx
->mdata
.bounce
);
2413 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2416 static void nvme_compare_data_cb(void *opaque
, int ret
)
2418 NvmeRequest
*req
= opaque
;
2419 NvmeCtrl
*n
= nvme_ctrl(req
);
2420 NvmeNamespace
*ns
= req
->ns
;
2421 BlockBackend
*blk
= ns
->blkconf
.blk
;
2422 BlockAcctCookie
*acct
= &req
->acct
;
2423 BlockAcctStats
*stats
= blk_get_stats(blk
);
2425 struct nvme_compare_ctx
*ctx
= req
->opaque
;
2426 g_autofree
uint8_t *buf
= NULL
;
2429 trace_pci_nvme_compare_data_cb(nvme_cid(req
));
2432 block_acct_failed(stats
, acct
);
2433 nvme_aio_err(req
, ret
);
2437 buf
= g_malloc(ctx
->data
.iov
.size
);
2439 status
= nvme_bounce_data(n
, buf
, ctx
->data
.iov
.size
,
2440 NVME_TX_DIRECTION_TO_DEVICE
, req
);
2442 req
->status
= status
;
2446 if (memcmp(buf
, ctx
->data
.bounce
, ctx
->data
.iov
.size
)) {
2447 req
->status
= NVME_CMP_FAILURE
| NVME_DNR
;
2452 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2453 uint64_t slba
= le64_to_cpu(rw
->slba
);
2454 uint32_t nlb
= le16_to_cpu(rw
->nlb
) + 1;
2455 size_t mlen
= nvme_m2b(ns
, nlb
);
2456 uint64_t offset
= nvme_moff(ns
, slba
);
2458 ctx
->mdata
.bounce
= g_malloc(mlen
);
2460 qemu_iovec_init(&ctx
->mdata
.iov
, 1);
2461 qemu_iovec_add(&ctx
->mdata
.iov
, ctx
->mdata
.bounce
, mlen
);
2463 req
->aiocb
= blk_aio_preadv(blk
, offset
, &ctx
->mdata
.iov
, 0,
2464 nvme_compare_mdata_cb
, req
);
2468 block_acct_done(stats
, acct
);
2471 qemu_iovec_destroy(&ctx
->data
.iov
);
2472 g_free(ctx
->data
.bounce
);
2475 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2478 typedef struct NvmeDSMAIOCB
{
2484 NvmeDsmRange
*range
;
2489 static void nvme_dsm_cancel(BlockAIOCB
*aiocb
)
2491 NvmeDSMAIOCB
*iocb
= container_of(aiocb
, NvmeDSMAIOCB
, common
);
2493 /* break nvme_dsm_cb loop */
2494 iocb
->idx
= iocb
->nr
;
2495 iocb
->ret
= -ECANCELED
;
2498 blk_aio_cancel_async(iocb
->aiocb
);
2502 * We only reach this if nvme_dsm_cancel() has already been called or
2503 * the command ran to completion.
2505 assert(iocb
->idx
== iocb
->nr
);
2509 static const AIOCBInfo nvme_dsm_aiocb_info
= {
2510 .aiocb_size
= sizeof(NvmeDSMAIOCB
),
2511 .cancel_async
= nvme_dsm_cancel
,
2514 static void nvme_dsm_cb(void *opaque
, int ret
);
2516 static void nvme_dsm_md_cb(void *opaque
, int ret
)
2518 NvmeDSMAIOCB
*iocb
= opaque
;
2519 NvmeRequest
*req
= iocb
->req
;
2520 NvmeNamespace
*ns
= req
->ns
;
2521 NvmeDsmRange
*range
;
2525 if (ret
< 0 || iocb
->ret
< 0 || !ns
->lbaf
.ms
) {
2529 range
= &iocb
->range
[iocb
->idx
- 1];
2530 slba
= le64_to_cpu(range
->slba
);
2531 nlb
= le32_to_cpu(range
->nlb
);
2534 * Check that all block were discarded (zeroed); otherwise we do not zero
2538 ret
= nvme_block_status_all(ns
, slba
, nlb
, BDRV_BLOCK_ZERO
);
2544 nvme_dsm_cb(iocb
, 0);
2548 iocb
->aiocb
= blk_aio_pwrite_zeroes(ns
->blkconf
.blk
, nvme_moff(ns
, slba
),
2549 nvme_m2b(ns
, nlb
), BDRV_REQ_MAY_UNMAP
,
2554 nvme_dsm_cb(iocb
, ret
);
2557 static void nvme_dsm_cb(void *opaque
, int ret
)
2559 NvmeDSMAIOCB
*iocb
= opaque
;
2560 NvmeRequest
*req
= iocb
->req
;
2561 NvmeCtrl
*n
= nvme_ctrl(req
);
2562 NvmeNamespace
*ns
= req
->ns
;
2563 NvmeDsmRange
*range
;
2567 if (iocb
->ret
< 0) {
2569 } else if (ret
< 0) {
2575 if (iocb
->idx
== iocb
->nr
) {
2579 range
= &iocb
->range
[iocb
->idx
++];
2580 slba
= le64_to_cpu(range
->slba
);
2581 nlb
= le32_to_cpu(range
->nlb
);
2583 trace_pci_nvme_dsm_deallocate(slba
, nlb
);
2585 if (nlb
> n
->dmrsl
) {
2586 trace_pci_nvme_dsm_single_range_limit_exceeded(nlb
, n
->dmrsl
);
2590 if (nvme_check_bounds(ns
, slba
, nlb
)) {
2591 trace_pci_nvme_err_invalid_lba_range(slba
, nlb
,
2596 iocb
->aiocb
= blk_aio_pdiscard(ns
->blkconf
.blk
, nvme_l2b(ns
, slba
),
2598 nvme_dsm_md_cb
, iocb
);
2603 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
2604 qemu_aio_unref(iocb
);
2607 static uint16_t nvme_dsm(NvmeCtrl
*n
, NvmeRequest
*req
)
2609 NvmeNamespace
*ns
= req
->ns
;
2610 NvmeDsmCmd
*dsm
= (NvmeDsmCmd
*) &req
->cmd
;
2611 uint32_t attr
= le32_to_cpu(dsm
->attributes
);
2612 uint32_t nr
= (le32_to_cpu(dsm
->nr
) & 0xff) + 1;
2613 uint16_t status
= NVME_SUCCESS
;
2615 trace_pci_nvme_dsm(nr
, attr
);
2617 if (attr
& NVME_DSMGMT_AD
) {
2618 NvmeDSMAIOCB
*iocb
= blk_aio_get(&nvme_dsm_aiocb_info
, ns
->blkconf
.blk
,
2623 iocb
->range
= g_new(NvmeDsmRange
, nr
);
2627 status
= nvme_h2c(n
, (uint8_t *)iocb
->range
, sizeof(NvmeDsmRange
) * nr
,
2630 g_free(iocb
->range
);
2631 qemu_aio_unref(iocb
);
2636 req
->aiocb
= &iocb
->common
;
2637 nvme_dsm_cb(iocb
, 0);
2639 return NVME_NO_COMPLETE
;
2645 static uint16_t nvme_verify(NvmeCtrl
*n
, NvmeRequest
*req
)
2647 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2648 NvmeNamespace
*ns
= req
->ns
;
2649 BlockBackend
*blk
= ns
->blkconf
.blk
;
2650 uint64_t slba
= le64_to_cpu(rw
->slba
);
2651 uint32_t nlb
= le16_to_cpu(rw
->nlb
) + 1;
2652 size_t len
= nvme_l2b(ns
, nlb
);
2653 int64_t offset
= nvme_l2b(ns
, slba
);
2654 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
2655 uint32_t reftag
= le32_to_cpu(rw
->reftag
);
2656 NvmeBounceContext
*ctx
= NULL
;
2659 trace_pci_nvme_verify(nvme_cid(req
), nvme_nsid(ns
), slba
, nlb
);
2661 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
2662 status
= nvme_check_prinfo(ns
, prinfo
, slba
, reftag
);
2667 if (prinfo
& NVME_PRINFO_PRACT
) {
2668 return NVME_INVALID_PROT_INFO
| NVME_DNR
;
2672 if (len
> n
->page_size
<< n
->params
.vsl
) {
2673 return NVME_INVALID_FIELD
| NVME_DNR
;
2676 status
= nvme_check_bounds(ns
, slba
, nlb
);
2681 if (NVME_ERR_REC_DULBE(ns
->features
.err_rec
)) {
2682 status
= nvme_check_dulbe(ns
, slba
, nlb
);
2688 ctx
= g_new0(NvmeBounceContext
, 1);
2691 ctx
->data
.bounce
= g_malloc(len
);
2693 qemu_iovec_init(&ctx
->data
.iov
, 1);
2694 qemu_iovec_add(&ctx
->data
.iov
, ctx
->data
.bounce
, len
);
2696 block_acct_start(blk_get_stats(blk
), &req
->acct
, ctx
->data
.iov
.size
,
2699 req
->aiocb
= blk_aio_preadv(ns
->blkconf
.blk
, offset
, &ctx
->data
.iov
, 0,
2700 nvme_verify_mdata_in_cb
, ctx
);
2701 return NVME_NO_COMPLETE
;
2704 typedef struct NvmeCopyAIOCB
{
2711 unsigned int format
;
2718 BlockAcctCookie read
;
2719 BlockAcctCookie write
;
2728 static void nvme_copy_cancel(BlockAIOCB
*aiocb
)
2730 NvmeCopyAIOCB
*iocb
= container_of(aiocb
, NvmeCopyAIOCB
, common
);
2732 iocb
->ret
= -ECANCELED
;
2735 blk_aio_cancel_async(iocb
->aiocb
);
2740 static const AIOCBInfo nvme_copy_aiocb_info
= {
2741 .aiocb_size
= sizeof(NvmeCopyAIOCB
),
2742 .cancel_async
= nvme_copy_cancel
,
2745 static void nvme_copy_done(NvmeCopyAIOCB
*iocb
)
2747 NvmeRequest
*req
= iocb
->req
;
2748 NvmeNamespace
*ns
= req
->ns
;
2749 BlockAcctStats
*stats
= blk_get_stats(ns
->blkconf
.blk
);
2751 if (iocb
->idx
!= iocb
->nr
) {
2752 req
->cqe
.result
= cpu_to_le32(iocb
->idx
);
2755 qemu_iovec_destroy(&iocb
->iov
);
2756 g_free(iocb
->bounce
);
2758 if (iocb
->ret
< 0) {
2759 block_acct_failed(stats
, &iocb
->acct
.read
);
2760 block_acct_failed(stats
, &iocb
->acct
.write
);
2762 block_acct_done(stats
, &iocb
->acct
.read
);
2763 block_acct_done(stats
, &iocb
->acct
.write
);
2766 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
2767 qemu_aio_unref(iocb
);
2770 static void nvme_do_copy(NvmeCopyAIOCB
*iocb
);
2772 static void nvme_copy_source_range_parse_format0(void *ranges
, int idx
,
2773 uint64_t *slba
, uint32_t *nlb
,
2778 NvmeCopySourceRangeFormat0
*_ranges
= ranges
;
2781 *slba
= le64_to_cpu(_ranges
[idx
].slba
);
2785 *nlb
= le16_to_cpu(_ranges
[idx
].nlb
) + 1;
2789 *apptag
= le16_to_cpu(_ranges
[idx
].apptag
);
2793 *appmask
= le16_to_cpu(_ranges
[idx
].appmask
);
2797 *reftag
= le32_to_cpu(_ranges
[idx
].reftag
);
2801 static void nvme_copy_source_range_parse_format1(void *ranges
, int idx
,
2802 uint64_t *slba
, uint32_t *nlb
,
2807 NvmeCopySourceRangeFormat1
*_ranges
= ranges
;
2810 *slba
= le64_to_cpu(_ranges
[idx
].slba
);
2814 *nlb
= le16_to_cpu(_ranges
[idx
].nlb
) + 1;
2818 *apptag
= le16_to_cpu(_ranges
[idx
].apptag
);
2822 *appmask
= le16_to_cpu(_ranges
[idx
].appmask
);
2828 *reftag
|= (uint64_t)_ranges
[idx
].sr
[4] << 40;
2829 *reftag
|= (uint64_t)_ranges
[idx
].sr
[5] << 32;
2830 *reftag
|= (uint64_t)_ranges
[idx
].sr
[6] << 24;
2831 *reftag
|= (uint64_t)_ranges
[idx
].sr
[7] << 16;
2832 *reftag
|= (uint64_t)_ranges
[idx
].sr
[8] << 8;
2833 *reftag
|= (uint64_t)_ranges
[idx
].sr
[9];
2837 static void nvme_copy_source_range_parse(void *ranges
, int idx
, uint8_t format
,
2838 uint64_t *slba
, uint32_t *nlb
,
2839 uint16_t *apptag
, uint16_t *appmask
,
2843 case NVME_COPY_FORMAT_0
:
2844 nvme_copy_source_range_parse_format0(ranges
, idx
, slba
, nlb
, apptag
,
2848 case NVME_COPY_FORMAT_1
:
2849 nvme_copy_source_range_parse_format1(ranges
, idx
, slba
, nlb
, apptag
,
2858 static void nvme_copy_out_completed_cb(void *opaque
, int ret
)
2860 NvmeCopyAIOCB
*iocb
= opaque
;
2861 NvmeRequest
*req
= iocb
->req
;
2862 NvmeNamespace
*ns
= req
->ns
;
2865 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, NULL
,
2866 &nlb
, NULL
, NULL
, NULL
);
2871 } else if (iocb
->ret
< 0) {
2875 if (ns
->params
.zoned
) {
2876 nvme_advance_zone_wp(ns
, iocb
->zone
, nlb
);
2885 static void nvme_copy_out_cb(void *opaque
, int ret
)
2887 NvmeCopyAIOCB
*iocb
= opaque
;
2888 NvmeRequest
*req
= iocb
->req
;
2889 NvmeNamespace
*ns
= req
->ns
;
2894 if (ret
< 0 || iocb
->ret
< 0 || !ns
->lbaf
.ms
) {
2898 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, NULL
,
2899 &nlb
, NULL
, NULL
, NULL
);
2901 mlen
= nvme_m2b(ns
, nlb
);
2902 mbounce
= iocb
->bounce
+ nvme_l2b(ns
, nlb
);
2904 qemu_iovec_reset(&iocb
->iov
);
2905 qemu_iovec_add(&iocb
->iov
, mbounce
, mlen
);
2907 iocb
->aiocb
= blk_aio_pwritev(ns
->blkconf
.blk
, nvme_moff(ns
, iocb
->slba
),
2908 &iocb
->iov
, 0, nvme_copy_out_completed_cb
,
2914 nvme_copy_out_completed_cb(iocb
, ret
);
2917 static void nvme_copy_in_completed_cb(void *opaque
, int ret
)
2919 NvmeCopyAIOCB
*iocb
= opaque
;
2920 NvmeRequest
*req
= iocb
->req
;
2921 NvmeNamespace
*ns
= req
->ns
;
2924 uint16_t apptag
, appmask
;
2932 } else if (iocb
->ret
< 0) {
2936 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, &slba
,
2937 &nlb
, &apptag
, &appmask
, &reftag
);
2938 len
= nvme_l2b(ns
, nlb
);
2940 trace_pci_nvme_copy_out(iocb
->slba
, nlb
);
2942 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
2943 NvmeCopyCmd
*copy
= (NvmeCopyCmd
*)&req
->cmd
;
2945 uint16_t prinfor
= ((copy
->control
[0] >> 4) & 0xf);
2946 uint16_t prinfow
= ((copy
->control
[2] >> 2) & 0xf);
2948 size_t mlen
= nvme_m2b(ns
, nlb
);
2949 uint8_t *mbounce
= iocb
->bounce
+ nvme_l2b(ns
, nlb
);
2951 status
= nvme_dif_mangle_mdata(ns
, mbounce
, mlen
, slba
);
2955 status
= nvme_dif_check(ns
, iocb
->bounce
, len
, mbounce
, mlen
, prinfor
,
2956 slba
, apptag
, appmask
, &reftag
);
2961 apptag
= le16_to_cpu(copy
->apptag
);
2962 appmask
= le16_to_cpu(copy
->appmask
);
2964 if (prinfow
& NVME_PRINFO_PRACT
) {
2965 status
= nvme_check_prinfo(ns
, prinfow
, iocb
->slba
, iocb
->reftag
);
2970 nvme_dif_pract_generate_dif(ns
, iocb
->bounce
, len
, mbounce
, mlen
,
2971 apptag
, &iocb
->reftag
);
2973 status
= nvme_dif_check(ns
, iocb
->bounce
, len
, mbounce
, mlen
,
2974 prinfow
, iocb
->slba
, apptag
, appmask
,
2982 status
= nvme_check_bounds(ns
, iocb
->slba
, nlb
);
2987 if (ns
->params
.zoned
) {
2988 status
= nvme_check_zone_write(ns
, iocb
->zone
, iocb
->slba
, nlb
);
2993 if (!(iocb
->zone
->d
.za
& NVME_ZA_ZRWA_VALID
)) {
2994 iocb
->zone
->w_ptr
+= nlb
;
2998 qemu_iovec_reset(&iocb
->iov
);
2999 qemu_iovec_add(&iocb
->iov
, iocb
->bounce
, len
);
3001 iocb
->aiocb
= blk_aio_pwritev(ns
->blkconf
.blk
, nvme_l2b(ns
, iocb
->slba
),
3002 &iocb
->iov
, 0, nvme_copy_out_cb
, iocb
);
3007 req
->status
= status
;
3013 static void nvme_copy_in_cb(void *opaque
, int ret
)
3015 NvmeCopyAIOCB
*iocb
= opaque
;
3016 NvmeRequest
*req
= iocb
->req
;
3017 NvmeNamespace
*ns
= req
->ns
;
3021 if (ret
< 0 || iocb
->ret
< 0 || !ns
->lbaf
.ms
) {
3025 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, &slba
,
3026 &nlb
, NULL
, NULL
, NULL
);
3028 qemu_iovec_reset(&iocb
->iov
);
3029 qemu_iovec_add(&iocb
->iov
, iocb
->bounce
+ nvme_l2b(ns
, nlb
),
3032 iocb
->aiocb
= blk_aio_preadv(ns
->blkconf
.blk
, nvme_moff(ns
, slba
),
3033 &iocb
->iov
, 0, nvme_copy_in_completed_cb
,
3038 nvme_copy_in_completed_cb(iocb
, ret
);
3041 static void nvme_do_copy(NvmeCopyAIOCB
*iocb
)
3043 NvmeRequest
*req
= iocb
->req
;
3044 NvmeNamespace
*ns
= req
->ns
;
3050 if (iocb
->ret
< 0) {
3054 if (iocb
->idx
== iocb
->nr
) {
3058 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, &slba
,
3059 &nlb
, NULL
, NULL
, NULL
);
3060 len
= nvme_l2b(ns
, nlb
);
3062 trace_pci_nvme_copy_source_range(slba
, nlb
);
3064 if (nlb
> le16_to_cpu(ns
->id_ns
.mssrl
)) {
3065 status
= NVME_CMD_SIZE_LIMIT
| NVME_DNR
;
3069 status
= nvme_check_bounds(ns
, slba
, nlb
);
3074 if (NVME_ERR_REC_DULBE(ns
->features
.err_rec
)) {
3075 status
= nvme_check_dulbe(ns
, slba
, nlb
);
3081 if (ns
->params
.zoned
) {
3082 status
= nvme_check_zone_read(ns
, slba
, nlb
);
3088 qemu_iovec_reset(&iocb
->iov
);
3089 qemu_iovec_add(&iocb
->iov
, iocb
->bounce
, len
);
3091 iocb
->aiocb
= blk_aio_preadv(ns
->blkconf
.blk
, nvme_l2b(ns
, slba
),
3092 &iocb
->iov
, 0, nvme_copy_in_cb
, iocb
);
3096 req
->status
= status
;
3099 nvme_copy_done(iocb
);
3102 static uint16_t nvme_copy(NvmeCtrl
*n
, NvmeRequest
*req
)
3104 NvmeNamespace
*ns
= req
->ns
;
3105 NvmeCopyCmd
*copy
= (NvmeCopyCmd
*)&req
->cmd
;
3106 NvmeCopyAIOCB
*iocb
= blk_aio_get(&nvme_copy_aiocb_info
, ns
->blkconf
.blk
,
3108 uint16_t nr
= copy
->nr
+ 1;
3109 uint8_t format
= copy
->control
[0] & 0xf;
3110 uint16_t prinfor
= ((copy
->control
[0] >> 4) & 0xf);
3111 uint16_t prinfow
= ((copy
->control
[2] >> 2) & 0xf);
3112 size_t len
= sizeof(NvmeCopySourceRangeFormat0
);
3116 trace_pci_nvme_copy(nvme_cid(req
), nvme_nsid(ns
), nr
, format
);
3118 iocb
->ranges
= NULL
;
3121 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
) &&
3122 ((prinfor
& NVME_PRINFO_PRACT
) != (prinfow
& NVME_PRINFO_PRACT
))) {
3123 status
= NVME_INVALID_FIELD
| NVME_DNR
;
3127 if (!(n
->id_ctrl
.ocfs
& (1 << format
))) {
3128 trace_pci_nvme_err_copy_invalid_format(format
);
3129 status
= NVME_INVALID_FIELD
| NVME_DNR
;
3133 if (nr
> ns
->id_ns
.msrc
+ 1) {
3134 status
= NVME_CMD_SIZE_LIMIT
| NVME_DNR
;
3138 if ((ns
->pif
== 0x0 && format
!= 0x0) ||
3139 (ns
->pif
!= 0x0 && format
!= 0x1)) {
3140 status
= NVME_INVALID_FORMAT
| NVME_DNR
;
3145 len
= sizeof(NvmeCopySourceRangeFormat1
);
3148 iocb
->format
= format
;
3149 iocb
->ranges
= g_malloc_n(nr
, len
);
3150 status
= nvme_h2c(n
, (uint8_t *)iocb
->ranges
, len
* nr
, req
);
3155 iocb
->slba
= le64_to_cpu(copy
->sdlba
);
3157 if (ns
->params
.zoned
) {
3158 iocb
->zone
= nvme_get_zone_by_slba(ns
, iocb
->slba
);
3160 status
= NVME_LBA_RANGE
| NVME_DNR
;
3164 status
= nvme_zrm_auto(n
, ns
, iocb
->zone
);
3174 iocb
->reftag
= le32_to_cpu(copy
->reftag
);
3175 iocb
->reftag
|= (uint64_t)le32_to_cpu(copy
->cdw3
) << 32;
3176 iocb
->bounce
= g_malloc_n(le16_to_cpu(ns
->id_ns
.mssrl
),
3177 ns
->lbasz
+ ns
->lbaf
.ms
);
3179 qemu_iovec_init(&iocb
->iov
, 1);
3181 block_acct_start(blk_get_stats(ns
->blkconf
.blk
), &iocb
->acct
.read
, 0,
3183 block_acct_start(blk_get_stats(ns
->blkconf
.blk
), &iocb
->acct
.write
, 0,
3186 req
->aiocb
= &iocb
->common
;
3189 return NVME_NO_COMPLETE
;
3192 g_free(iocb
->ranges
);
3193 qemu_aio_unref(iocb
);
3197 static uint16_t nvme_compare(NvmeCtrl
*n
, NvmeRequest
*req
)
3199 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
3200 NvmeNamespace
*ns
= req
->ns
;
3201 BlockBackend
*blk
= ns
->blkconf
.blk
;
3202 uint64_t slba
= le64_to_cpu(rw
->slba
);
3203 uint32_t nlb
= le16_to_cpu(rw
->nlb
) + 1;
3204 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
3205 size_t data_len
= nvme_l2b(ns
, nlb
);
3206 size_t len
= data_len
;
3207 int64_t offset
= nvme_l2b(ns
, slba
);
3208 struct nvme_compare_ctx
*ctx
= NULL
;
3211 trace_pci_nvme_compare(nvme_cid(req
), nvme_nsid(ns
), slba
, nlb
);
3213 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
) && (prinfo
& NVME_PRINFO_PRACT
)) {
3214 return NVME_INVALID_PROT_INFO
| NVME_DNR
;
3217 if (nvme_ns_ext(ns
)) {
3218 len
+= nvme_m2b(ns
, nlb
);
3221 status
= nvme_check_mdts(n
, len
);
3226 status
= nvme_check_bounds(ns
, slba
, nlb
);
3231 if (NVME_ERR_REC_DULBE(ns
->features
.err_rec
)) {
3232 status
= nvme_check_dulbe(ns
, slba
, nlb
);
3238 status
= nvme_map_dptr(n
, &req
->sg
, len
, &req
->cmd
);
3243 ctx
= g_new(struct nvme_compare_ctx
, 1);
3244 ctx
->data
.bounce
= g_malloc(data_len
);
3248 qemu_iovec_init(&ctx
->data
.iov
, 1);
3249 qemu_iovec_add(&ctx
->data
.iov
, ctx
->data
.bounce
, data_len
);
3251 block_acct_start(blk_get_stats(blk
), &req
->acct
, data_len
,
3253 req
->aiocb
= blk_aio_preadv(blk
, offset
, &ctx
->data
.iov
, 0,
3254 nvme_compare_data_cb
, req
);
3256 return NVME_NO_COMPLETE
;
3259 typedef struct NvmeFlushAIOCB
{
3270 static void nvme_flush_cancel(BlockAIOCB
*acb
)
3272 NvmeFlushAIOCB
*iocb
= container_of(acb
, NvmeFlushAIOCB
, common
);
3274 iocb
->ret
= -ECANCELED
;
3277 blk_aio_cancel_async(iocb
->aiocb
);
3282 static const AIOCBInfo nvme_flush_aiocb_info
= {
3283 .aiocb_size
= sizeof(NvmeFlushAIOCB
),
3284 .cancel_async
= nvme_flush_cancel
,
3285 .get_aio_context
= nvme_get_aio_context
,
3288 static void nvme_do_flush(NvmeFlushAIOCB
*iocb
);
3290 static void nvme_flush_ns_cb(void *opaque
, int ret
)
3292 NvmeFlushAIOCB
*iocb
= opaque
;
3293 NvmeNamespace
*ns
= iocb
->ns
;
3298 } else if (iocb
->ret
< 0) {
3303 trace_pci_nvme_flush_ns(iocb
->nsid
);
3306 iocb
->aiocb
= blk_aio_flush(ns
->blkconf
.blk
, nvme_flush_ns_cb
, iocb
);
3311 nvme_do_flush(iocb
);
3314 static void nvme_do_flush(NvmeFlushAIOCB
*iocb
)
3316 NvmeRequest
*req
= iocb
->req
;
3317 NvmeCtrl
*n
= nvme_ctrl(req
);
3320 if (iocb
->ret
< 0) {
3324 if (iocb
->broadcast
) {
3325 for (i
= iocb
->nsid
+ 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
3326 iocb
->ns
= nvme_ns(n
, i
);
3338 nvme_flush_ns_cb(iocb
, 0);
3342 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
3343 qemu_aio_unref(iocb
);
3346 static uint16_t nvme_flush(NvmeCtrl
*n
, NvmeRequest
*req
)
3348 NvmeFlushAIOCB
*iocb
;
3349 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
3352 iocb
= qemu_aio_get(&nvme_flush_aiocb_info
, NULL
, nvme_misc_cb
, req
);
3358 iocb
->broadcast
= (nsid
== NVME_NSID_BROADCAST
);
3360 if (!iocb
->broadcast
) {
3361 if (!nvme_nsid_valid(n
, nsid
)) {
3362 status
= NVME_INVALID_NSID
| NVME_DNR
;
3366 iocb
->ns
= nvme_ns(n
, nsid
);
3368 status
= NVME_INVALID_FIELD
| NVME_DNR
;
3375 req
->aiocb
= &iocb
->common
;
3376 nvme_do_flush(iocb
);
3378 return NVME_NO_COMPLETE
;
3381 qemu_aio_unref(iocb
);
3386 static uint16_t nvme_read(NvmeCtrl
*n
, NvmeRequest
*req
)
3388 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
3389 NvmeNamespace
*ns
= req
->ns
;
3390 uint64_t slba
= le64_to_cpu(rw
->slba
);
3391 uint32_t nlb
= (uint32_t)le16_to_cpu(rw
->nlb
) + 1;
3392 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
3393 uint64_t data_size
= nvme_l2b(ns
, nlb
);
3394 uint64_t mapped_size
= data_size
;
3395 uint64_t data_offset
;
3396 BlockBackend
*blk
= ns
->blkconf
.blk
;
3399 if (nvme_ns_ext(ns
)) {
3400 mapped_size
+= nvme_m2b(ns
, nlb
);
3402 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3403 bool pract
= prinfo
& NVME_PRINFO_PRACT
;
3405 if (pract
&& ns
->lbaf
.ms
== nvme_pi_tuple_size(ns
)) {
3406 mapped_size
= data_size
;
3411 trace_pci_nvme_read(nvme_cid(req
), nvme_nsid(ns
), nlb
, mapped_size
, slba
);
3413 status
= nvme_check_mdts(n
, mapped_size
);
3418 status
= nvme_check_bounds(ns
, slba
, nlb
);
3423 if (ns
->params
.zoned
) {
3424 status
= nvme_check_zone_read(ns
, slba
, nlb
);
3426 trace_pci_nvme_err_zone_read_not_ok(slba
, nlb
, status
);
3431 if (NVME_ERR_REC_DULBE(ns
->features
.err_rec
)) {
3432 status
= nvme_check_dulbe(ns
, slba
, nlb
);
3438 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3439 return nvme_dif_rw(n
, req
);
3442 status
= nvme_map_data(n
, nlb
, req
);
3447 data_offset
= nvme_l2b(ns
, slba
);
3449 block_acct_start(blk_get_stats(blk
), &req
->acct
, data_size
,
3451 nvme_blk_read(blk
, data_offset
, BDRV_SECTOR_SIZE
, nvme_rw_cb
, req
);
3452 return NVME_NO_COMPLETE
;
3455 block_acct_invalid(blk_get_stats(blk
), BLOCK_ACCT_READ
);
3456 return status
| NVME_DNR
;
3459 static void nvme_do_write_fdp(NvmeCtrl
*n
, NvmeRequest
*req
, uint64_t slba
,
3462 NvmeNamespace
*ns
= req
->ns
;
3463 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
3464 uint64_t data_size
= nvme_l2b(ns
, nlb
);
3465 uint32_t dw12
= le32_to_cpu(req
->cmd
.cdw12
);
3466 uint8_t dtype
= (dw12
>> 20) & 0xf;
3467 uint16_t pid
= le16_to_cpu(rw
->dspec
);
3468 uint16_t ph
, rg
, ruhid
;
3469 NvmeReclaimUnit
*ru
;
3471 if (dtype
!= NVME_DIRECTIVE_DATA_PLACEMENT
||
3472 !nvme_parse_pid(ns
, pid
, &ph
, &rg
)) {
3477 ruhid
= ns
->fdp
.phs
[ph
];
3478 ru
= &ns
->endgrp
->fdp
.ruhs
[ruhid
].rus
[rg
];
3480 nvme_fdp_stat_inc(&ns
->endgrp
->fdp
.hbmw
, data_size
);
3481 nvme_fdp_stat_inc(&ns
->endgrp
->fdp
.mbmw
, data_size
);
3484 if (nlb
< ru
->ruamw
) {
3490 nvme_update_ruh(n
, ns
, pid
);
3494 static uint16_t nvme_do_write(NvmeCtrl
*n
, NvmeRequest
*req
, bool append
,
3497 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
3498 NvmeNamespace
*ns
= req
->ns
;
3499 uint64_t slba
= le64_to_cpu(rw
->slba
);
3500 uint32_t nlb
= (uint32_t)le16_to_cpu(rw
->nlb
) + 1;
3501 uint16_t ctrl
= le16_to_cpu(rw
->control
);
3502 uint8_t prinfo
= NVME_RW_PRINFO(ctrl
);
3503 uint64_t data_size
= nvme_l2b(ns
, nlb
);
3504 uint64_t mapped_size
= data_size
;
3505 uint64_t data_offset
;
3507 NvmeZonedResult
*res
= (NvmeZonedResult
*)&req
->cqe
;
3508 BlockBackend
*blk
= ns
->blkconf
.blk
;
3511 if (nvme_ns_ext(ns
)) {
3512 mapped_size
+= nvme_m2b(ns
, nlb
);
3514 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3515 bool pract
= prinfo
& NVME_PRINFO_PRACT
;
3517 if (pract
&& ns
->lbaf
.ms
== nvme_pi_tuple_size(ns
)) {
3518 mapped_size
-= nvme_m2b(ns
, nlb
);
3523 trace_pci_nvme_write(nvme_cid(req
), nvme_io_opc_str(rw
->opcode
),
3524 nvme_nsid(ns
), nlb
, mapped_size
, slba
);
3527 status
= nvme_check_mdts(n
, mapped_size
);
3533 status
= nvme_check_bounds(ns
, slba
, nlb
);
3538 if (ns
->params
.zoned
) {
3539 zone
= nvme_get_zone_by_slba(ns
, slba
);
3543 bool piremap
= !!(ctrl
& NVME_RW_PIREMAP
);
3545 if (unlikely(zone
->d
.za
& NVME_ZA_ZRWA_VALID
)) {
3546 return NVME_INVALID_ZONE_OP
| NVME_DNR
;
3549 if (unlikely(slba
!= zone
->d
.zslba
)) {
3550 trace_pci_nvme_err_append_not_at_start(slba
, zone
->d
.zslba
);
3551 status
= NVME_INVALID_FIELD
;
3555 if (n
->params
.zasl
&&
3556 data_size
> (uint64_t)n
->page_size
<< n
->params
.zasl
) {
3557 trace_pci_nvme_err_zasl(data_size
);
3558 return NVME_INVALID_FIELD
| NVME_DNR
;
3562 rw
->slba
= cpu_to_le64(slba
);
3563 res
->slba
= cpu_to_le64(slba
);
3565 switch (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3566 case NVME_ID_NS_DPS_TYPE_1
:
3568 return NVME_INVALID_PROT_INFO
| NVME_DNR
;
3573 case NVME_ID_NS_DPS_TYPE_2
:
3575 uint32_t reftag
= le32_to_cpu(rw
->reftag
);
3576 rw
->reftag
= cpu_to_le32(reftag
+ (slba
- zone
->d
.zslba
));
3581 case NVME_ID_NS_DPS_TYPE_3
:
3583 return NVME_INVALID_PROT_INFO
| NVME_DNR
;
3590 status
= nvme_check_zone_write(ns
, zone
, slba
, nlb
);
3595 status
= nvme_zrm_auto(n
, ns
, zone
);
3600 if (!(zone
->d
.za
& NVME_ZA_ZRWA_VALID
)) {
3603 } else if (ns
->endgrp
&& ns
->endgrp
->fdp
.enabled
) {
3604 nvme_do_write_fdp(n
, req
, slba
, nlb
);
3607 data_offset
= nvme_l2b(ns
, slba
);
3609 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3610 return nvme_dif_rw(n
, req
);
3614 status
= nvme_map_data(n
, nlb
, req
);
3619 block_acct_start(blk_get_stats(blk
), &req
->acct
, data_size
,
3621 nvme_blk_write(blk
, data_offset
, BDRV_SECTOR_SIZE
, nvme_rw_cb
, req
);
3623 req
->aiocb
= blk_aio_pwrite_zeroes(blk
, data_offset
, data_size
,
3624 BDRV_REQ_MAY_UNMAP
, nvme_rw_cb
,
3628 return NVME_NO_COMPLETE
;
3631 block_acct_invalid(blk_get_stats(blk
), BLOCK_ACCT_WRITE
);
3632 return status
| NVME_DNR
;
3635 static inline uint16_t nvme_write(NvmeCtrl
*n
, NvmeRequest
*req
)
3637 return nvme_do_write(n
, req
, false, false);
3640 static inline uint16_t nvme_write_zeroes(NvmeCtrl
*n
, NvmeRequest
*req
)
3642 return nvme_do_write(n
, req
, false, true);
3645 static inline uint16_t nvme_zone_append(NvmeCtrl
*n
, NvmeRequest
*req
)
3647 return nvme_do_write(n
, req
, true, false);
3650 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace
*ns
, NvmeCmd
*c
,
3651 uint64_t *slba
, uint32_t *zone_idx
)
3653 uint32_t dw10
= le32_to_cpu(c
->cdw10
);
3654 uint32_t dw11
= le32_to_cpu(c
->cdw11
);
3656 if (!ns
->params
.zoned
) {
3657 trace_pci_nvme_err_invalid_opc(c
->opcode
);
3658 return NVME_INVALID_OPCODE
| NVME_DNR
;
3661 *slba
= ((uint64_t)dw11
) << 32 | dw10
;
3662 if (unlikely(*slba
>= ns
->id_ns
.nsze
)) {
3663 trace_pci_nvme_err_invalid_lba_range(*slba
, 0, ns
->id_ns
.nsze
);
3665 return NVME_LBA_RANGE
| NVME_DNR
;
3668 *zone_idx
= nvme_zone_idx(ns
, *slba
);
3669 assert(*zone_idx
< ns
->num_zones
);
3671 return NVME_SUCCESS
;
3674 typedef uint16_t (*op_handler_t
)(NvmeNamespace
*, NvmeZone
*, NvmeZoneState
,
3677 enum NvmeZoneProcessingMask
{
3678 NVME_PROC_CURRENT_ZONE
= 0,
3679 NVME_PROC_OPENED_ZONES
= 1 << 0,
3680 NVME_PROC_CLOSED_ZONES
= 1 << 1,
3681 NVME_PROC_READ_ONLY_ZONES
= 1 << 2,
3682 NVME_PROC_FULL_ZONES
= 1 << 3,
3685 static uint16_t nvme_open_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3686 NvmeZoneState state
, NvmeRequest
*req
)
3688 NvmeZoneSendCmd
*cmd
= (NvmeZoneSendCmd
*)&req
->cmd
;
3691 if (cmd
->zsflags
& NVME_ZSFLAG_ZRWA_ALLOC
) {
3692 uint16_t ozcs
= le16_to_cpu(ns
->id_ns_zoned
->ozcs
);
3694 if (!(ozcs
& NVME_ID_NS_ZONED_OZCS_ZRWASUP
)) {
3695 return NVME_INVALID_ZONE_OP
| NVME_DNR
;
3698 if (zone
->w_ptr
% ns
->zns
.zrwafg
) {
3699 return NVME_NOZRWA
| NVME_DNR
;
3702 flags
= NVME_ZRM_ZRWA
;
3705 return nvme_zrm_open_flags(nvme_ctrl(req
), ns
, zone
, flags
);
3708 static uint16_t nvme_close_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3709 NvmeZoneState state
, NvmeRequest
*req
)
3711 return nvme_zrm_close(ns
, zone
);
3714 static uint16_t nvme_finish_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3715 NvmeZoneState state
, NvmeRequest
*req
)
3717 return nvme_zrm_finish(ns
, zone
);
3720 static uint16_t nvme_offline_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3721 NvmeZoneState state
, NvmeRequest
*req
)
3724 case NVME_ZONE_STATE_READ_ONLY
:
3725 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_OFFLINE
);
3727 case NVME_ZONE_STATE_OFFLINE
:
3728 return NVME_SUCCESS
;
3730 return NVME_ZONE_INVAL_TRANSITION
;
3734 static uint16_t nvme_set_zd_ext(NvmeNamespace
*ns
, NvmeZone
*zone
)
3737 uint8_t state
= nvme_get_zone_state(zone
);
3739 if (state
== NVME_ZONE_STATE_EMPTY
) {
3740 status
= nvme_aor_check(ns
, 1, 0);
3744 nvme_aor_inc_active(ns
);
3745 zone
->d
.za
|= NVME_ZA_ZD_EXT_VALID
;
3746 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_CLOSED
);
3747 return NVME_SUCCESS
;
3750 return NVME_ZONE_INVAL_TRANSITION
;
3753 static uint16_t nvme_bulk_proc_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3754 enum NvmeZoneProcessingMask proc_mask
,
3755 op_handler_t op_hndlr
, NvmeRequest
*req
)
3757 uint16_t status
= NVME_SUCCESS
;
3758 NvmeZoneState zs
= nvme_get_zone_state(zone
);
3762 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
3763 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
3764 proc_zone
= proc_mask
& NVME_PROC_OPENED_ZONES
;
3766 case NVME_ZONE_STATE_CLOSED
:
3767 proc_zone
= proc_mask
& NVME_PROC_CLOSED_ZONES
;
3769 case NVME_ZONE_STATE_READ_ONLY
:
3770 proc_zone
= proc_mask
& NVME_PROC_READ_ONLY_ZONES
;
3772 case NVME_ZONE_STATE_FULL
:
3773 proc_zone
= proc_mask
& NVME_PROC_FULL_ZONES
;
3780 status
= op_hndlr(ns
, zone
, zs
, req
);
3786 static uint16_t nvme_do_zone_op(NvmeNamespace
*ns
, NvmeZone
*zone
,
3787 enum NvmeZoneProcessingMask proc_mask
,
3788 op_handler_t op_hndlr
, NvmeRequest
*req
)
3791 uint16_t status
= NVME_SUCCESS
;
3795 status
= op_hndlr(ns
, zone
, nvme_get_zone_state(zone
), req
);
3797 if (proc_mask
& NVME_PROC_CLOSED_ZONES
) {
3798 QTAILQ_FOREACH_SAFE(zone
, &ns
->closed_zones
, entry
, next
) {
3799 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3801 if (status
&& status
!= NVME_NO_COMPLETE
) {
3806 if (proc_mask
& NVME_PROC_OPENED_ZONES
) {
3807 QTAILQ_FOREACH_SAFE(zone
, &ns
->imp_open_zones
, entry
, next
) {
3808 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3810 if (status
&& status
!= NVME_NO_COMPLETE
) {
3815 QTAILQ_FOREACH_SAFE(zone
, &ns
->exp_open_zones
, entry
, next
) {
3816 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3818 if (status
&& status
!= NVME_NO_COMPLETE
) {
3823 if (proc_mask
& NVME_PROC_FULL_ZONES
) {
3824 QTAILQ_FOREACH_SAFE(zone
, &ns
->full_zones
, entry
, next
) {
3825 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3827 if (status
&& status
!= NVME_NO_COMPLETE
) {
3833 if (proc_mask
& NVME_PROC_READ_ONLY_ZONES
) {
3834 for (i
= 0; i
< ns
->num_zones
; i
++, zone
++) {
3835 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3837 if (status
&& status
!= NVME_NO_COMPLETE
) {
3848 typedef struct NvmeZoneResetAIOCB
{
3857 } NvmeZoneResetAIOCB
;
3859 static void nvme_zone_reset_cancel(BlockAIOCB
*aiocb
)
3861 NvmeZoneResetAIOCB
*iocb
= container_of(aiocb
, NvmeZoneResetAIOCB
, common
);
3862 NvmeRequest
*req
= iocb
->req
;
3863 NvmeNamespace
*ns
= req
->ns
;
3865 iocb
->idx
= ns
->num_zones
;
3867 iocb
->ret
= -ECANCELED
;
3870 blk_aio_cancel_async(iocb
->aiocb
);
3875 static const AIOCBInfo nvme_zone_reset_aiocb_info
= {
3876 .aiocb_size
= sizeof(NvmeZoneResetAIOCB
),
3877 .cancel_async
= nvme_zone_reset_cancel
,
3880 static void nvme_zone_reset_cb(void *opaque
, int ret
);
3882 static void nvme_zone_reset_epilogue_cb(void *opaque
, int ret
)
3884 NvmeZoneResetAIOCB
*iocb
= opaque
;
3885 NvmeRequest
*req
= iocb
->req
;
3886 NvmeNamespace
*ns
= req
->ns
;
3890 if (ret
< 0 || iocb
->ret
< 0 || !ns
->lbaf
.ms
) {
3894 moff
= nvme_moff(ns
, iocb
->zone
->d
.zslba
);
3895 count
= nvme_m2b(ns
, ns
->zone_size
);
3897 iocb
->aiocb
= blk_aio_pwrite_zeroes(ns
->blkconf
.blk
, moff
, count
,
3899 nvme_zone_reset_cb
, iocb
);
3903 nvme_zone_reset_cb(iocb
, ret
);
3906 static void nvme_zone_reset_cb(void *opaque
, int ret
)
3908 NvmeZoneResetAIOCB
*iocb
= opaque
;
3909 NvmeRequest
*req
= iocb
->req
;
3910 NvmeNamespace
*ns
= req
->ns
;
3912 if (iocb
->ret
< 0) {
3914 } else if (ret
< 0) {
3920 nvme_zrm_reset(ns
, iocb
->zone
);
3927 while (iocb
->idx
< ns
->num_zones
) {
3928 NvmeZone
*zone
= &ns
->zone_array
[iocb
->idx
++];
3930 switch (nvme_get_zone_state(zone
)) {
3931 case NVME_ZONE_STATE_EMPTY
:
3938 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
3939 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
3940 case NVME_ZONE_STATE_CLOSED
:
3941 case NVME_ZONE_STATE_FULL
:
3949 trace_pci_nvme_zns_zone_reset(zone
->d
.zslba
);
3951 iocb
->aiocb
= blk_aio_pwrite_zeroes(ns
->blkconf
.blk
,
3952 nvme_l2b(ns
, zone
->d
.zslba
),
3953 nvme_l2b(ns
, ns
->zone_size
),
3955 nvme_zone_reset_epilogue_cb
,
3963 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
3964 qemu_aio_unref(iocb
);
3967 static uint16_t nvme_zone_mgmt_send_zrwa_flush(NvmeCtrl
*n
, NvmeZone
*zone
,
3968 uint64_t elba
, NvmeRequest
*req
)
3970 NvmeNamespace
*ns
= req
->ns
;
3971 uint16_t ozcs
= le16_to_cpu(ns
->id_ns_zoned
->ozcs
);
3972 uint64_t wp
= zone
->d
.wp
;
3973 uint32_t nlb
= elba
- wp
+ 1;
3977 if (!(ozcs
& NVME_ID_NS_ZONED_OZCS_ZRWASUP
)) {
3978 return NVME_INVALID_ZONE_OP
| NVME_DNR
;
3981 if (!(zone
->d
.za
& NVME_ZA_ZRWA_VALID
)) {
3982 return NVME_INVALID_FIELD
| NVME_DNR
;
3985 if (elba
< wp
|| elba
> wp
+ ns
->zns
.zrwas
) {
3986 return NVME_ZONE_BOUNDARY_ERROR
| NVME_DNR
;
3989 if (nlb
% ns
->zns
.zrwafg
) {
3990 return NVME_INVALID_FIELD
| NVME_DNR
;
3993 status
= nvme_zrm_auto(n
, ns
, zone
);
4000 nvme_advance_zone_wp(ns
, zone
, nlb
);
4002 return NVME_SUCCESS
;
4005 static uint16_t nvme_zone_mgmt_send(NvmeCtrl
*n
, NvmeRequest
*req
)
4007 NvmeZoneSendCmd
*cmd
= (NvmeZoneSendCmd
*)&req
->cmd
;
4008 NvmeNamespace
*ns
= req
->ns
;
4010 NvmeZoneResetAIOCB
*iocb
;
4013 uint32_t zone_idx
= 0;
4015 uint8_t action
= cmd
->zsa
;
4017 enum NvmeZoneProcessingMask proc_mask
= NVME_PROC_CURRENT_ZONE
;
4019 all
= cmd
->zsflags
& NVME_ZSFLAG_SELECT_ALL
;
4021 req
->status
= NVME_SUCCESS
;
4024 status
= nvme_get_mgmt_zone_slba_idx(ns
, &req
->cmd
, &slba
, &zone_idx
);
4030 zone
= &ns
->zone_array
[zone_idx
];
4031 if (slba
!= zone
->d
.zslba
&& action
!= NVME_ZONE_ACTION_ZRWA_FLUSH
) {
4032 trace_pci_nvme_err_unaligned_zone_cmd(action
, slba
, zone
->d
.zslba
);
4033 return NVME_INVALID_FIELD
| NVME_DNR
;
4038 case NVME_ZONE_ACTION_OPEN
:
4040 proc_mask
= NVME_PROC_CLOSED_ZONES
;
4042 trace_pci_nvme_open_zone(slba
, zone_idx
, all
);
4043 status
= nvme_do_zone_op(ns
, zone
, proc_mask
, nvme_open_zone
, req
);
4046 case NVME_ZONE_ACTION_CLOSE
:
4048 proc_mask
= NVME_PROC_OPENED_ZONES
;
4050 trace_pci_nvme_close_zone(slba
, zone_idx
, all
);
4051 status
= nvme_do_zone_op(ns
, zone
, proc_mask
, nvme_close_zone
, req
);
4054 case NVME_ZONE_ACTION_FINISH
:
4056 proc_mask
= NVME_PROC_OPENED_ZONES
| NVME_PROC_CLOSED_ZONES
;
4058 trace_pci_nvme_finish_zone(slba
, zone_idx
, all
);
4059 status
= nvme_do_zone_op(ns
, zone
, proc_mask
, nvme_finish_zone
, req
);
4062 case NVME_ZONE_ACTION_RESET
:
4063 trace_pci_nvme_reset_zone(slba
, zone_idx
, all
);
4065 iocb
= blk_aio_get(&nvme_zone_reset_aiocb_info
, ns
->blkconf
.blk
,
4071 iocb
->idx
= zone_idx
;
4074 req
->aiocb
= &iocb
->common
;
4075 nvme_zone_reset_cb(iocb
, 0);
4077 return NVME_NO_COMPLETE
;
4079 case NVME_ZONE_ACTION_OFFLINE
:
4081 proc_mask
= NVME_PROC_READ_ONLY_ZONES
;
4083 trace_pci_nvme_offline_zone(slba
, zone_idx
, all
);
4084 status
= nvme_do_zone_op(ns
, zone
, proc_mask
, nvme_offline_zone
, req
);
4087 case NVME_ZONE_ACTION_SET_ZD_EXT
:
4088 trace_pci_nvme_set_descriptor_extension(slba
, zone_idx
);
4089 if (all
|| !ns
->params
.zd_extension_size
) {
4090 return NVME_INVALID_FIELD
| NVME_DNR
;
4092 zd_ext
= nvme_get_zd_extension(ns
, zone_idx
);
4093 status
= nvme_h2c(n
, zd_ext
, ns
->params
.zd_extension_size
, req
);
4095 trace_pci_nvme_err_zd_extension_map_error(zone_idx
);
4099 status
= nvme_set_zd_ext(ns
, zone
);
4100 if (status
== NVME_SUCCESS
) {
4101 trace_pci_nvme_zd_extension_set(zone_idx
);
4106 case NVME_ZONE_ACTION_ZRWA_FLUSH
:
4108 return NVME_INVALID_FIELD
| NVME_DNR
;
4111 return nvme_zone_mgmt_send_zrwa_flush(n
, zone
, slba
, req
);
4114 trace_pci_nvme_err_invalid_mgmt_action(action
);
4115 status
= NVME_INVALID_FIELD
;
4118 if (status
== NVME_ZONE_INVAL_TRANSITION
) {
4119 trace_pci_nvme_err_invalid_zone_state_transition(action
, slba
,
4129 static bool nvme_zone_matches_filter(uint32_t zafs
, NvmeZone
*zl
)
4131 NvmeZoneState zs
= nvme_get_zone_state(zl
);
4134 case NVME_ZONE_REPORT_ALL
:
4136 case NVME_ZONE_REPORT_EMPTY
:
4137 return zs
== NVME_ZONE_STATE_EMPTY
;
4138 case NVME_ZONE_REPORT_IMPLICITLY_OPEN
:
4139 return zs
== NVME_ZONE_STATE_IMPLICITLY_OPEN
;
4140 case NVME_ZONE_REPORT_EXPLICITLY_OPEN
:
4141 return zs
== NVME_ZONE_STATE_EXPLICITLY_OPEN
;
4142 case NVME_ZONE_REPORT_CLOSED
:
4143 return zs
== NVME_ZONE_STATE_CLOSED
;
4144 case NVME_ZONE_REPORT_FULL
:
4145 return zs
== NVME_ZONE_STATE_FULL
;
4146 case NVME_ZONE_REPORT_READ_ONLY
:
4147 return zs
== NVME_ZONE_STATE_READ_ONLY
;
4148 case NVME_ZONE_REPORT_OFFLINE
:
4149 return zs
== NVME_ZONE_STATE_OFFLINE
;
4155 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl
*n
, NvmeRequest
*req
)
4157 NvmeCmd
*cmd
= (NvmeCmd
*)&req
->cmd
;
4158 NvmeNamespace
*ns
= req
->ns
;
4159 /* cdw12 is zero-based number of dwords to return. Convert to bytes */
4160 uint32_t data_size
= (le32_to_cpu(cmd
->cdw12
) + 1) << 2;
4161 uint32_t dw13
= le32_to_cpu(cmd
->cdw13
);
4162 uint32_t zone_idx
, zra
, zrasf
, partial
;
4163 uint64_t max_zones
, nr_zones
= 0;
4168 NvmeZoneReportHeader
*header
;
4170 size_t zone_entry_sz
;
4173 req
->status
= NVME_SUCCESS
;
4175 status
= nvme_get_mgmt_zone_slba_idx(ns
, cmd
, &slba
, &zone_idx
);
4181 if (zra
!= NVME_ZONE_REPORT
&& zra
!= NVME_ZONE_REPORT_EXTENDED
) {
4182 return NVME_INVALID_FIELD
| NVME_DNR
;
4184 if (zra
== NVME_ZONE_REPORT_EXTENDED
&& !ns
->params
.zd_extension_size
) {
4185 return NVME_INVALID_FIELD
| NVME_DNR
;
4188 zrasf
= (dw13
>> 8) & 0xff;
4189 if (zrasf
> NVME_ZONE_REPORT_OFFLINE
) {
4190 return NVME_INVALID_FIELD
| NVME_DNR
;
4193 if (data_size
< sizeof(NvmeZoneReportHeader
)) {
4194 return NVME_INVALID_FIELD
| NVME_DNR
;
4197 status
= nvme_check_mdts(n
, data_size
);
4202 partial
= (dw13
>> 16) & 0x01;
4204 zone_entry_sz
= sizeof(NvmeZoneDescr
);
4205 if (zra
== NVME_ZONE_REPORT_EXTENDED
) {
4206 zone_entry_sz
+= ns
->params
.zd_extension_size
;
4209 max_zones
= (data_size
- sizeof(NvmeZoneReportHeader
)) / zone_entry_sz
;
4210 buf
= g_malloc0(data_size
);
4212 zone
= &ns
->zone_array
[zone_idx
];
4213 for (i
= zone_idx
; i
< ns
->num_zones
; i
++) {
4214 if (partial
&& nr_zones
>= max_zones
) {
4217 if (nvme_zone_matches_filter(zrasf
, zone
++)) {
4222 header
->nr_zones
= cpu_to_le64(nr_zones
);
4224 buf_p
= buf
+ sizeof(NvmeZoneReportHeader
);
4225 for (; zone_idx
< ns
->num_zones
&& max_zones
> 0; zone_idx
++) {
4226 zone
= &ns
->zone_array
[zone_idx
];
4227 if (nvme_zone_matches_filter(zrasf
, zone
)) {
4229 buf_p
+= sizeof(NvmeZoneDescr
);
4233 z
->zcap
= cpu_to_le64(zone
->d
.zcap
);
4234 z
->zslba
= cpu_to_le64(zone
->d
.zslba
);
4237 if (nvme_wp_is_valid(zone
)) {
4238 z
->wp
= cpu_to_le64(zone
->d
.wp
);
4240 z
->wp
= cpu_to_le64(~0ULL);
4243 if (zra
== NVME_ZONE_REPORT_EXTENDED
) {
4244 if (zone
->d
.za
& NVME_ZA_ZD_EXT_VALID
) {
4245 memcpy(buf_p
, nvme_get_zd_extension(ns
, zone_idx
),
4246 ns
->params
.zd_extension_size
);
4248 buf_p
+= ns
->params
.zd_extension_size
;
4255 status
= nvme_c2h(n
, (uint8_t *)buf
, data_size
, req
);
4262 static uint16_t nvme_io_mgmt_recv_ruhs(NvmeCtrl
*n
, NvmeRequest
*req
,
4265 NvmeNamespace
*ns
= req
->ns
;
4266 NvmeEnduranceGroup
*endgrp
;
4268 NvmeRuhStatusDescr
*ruhsd
;
4269 unsigned int nruhsd
;
4270 uint16_t rg
, ph
, *ruhid
;
4272 g_autofree
uint8_t *buf
= NULL
;
4275 return NVME_INVALID_FIELD
| NVME_DNR
;
4278 if (ns
->params
.nsid
== 0 || ns
->params
.nsid
== 0xffffffff) {
4279 return NVME_INVALID_NSID
| NVME_DNR
;
4282 if (!n
->subsys
->endgrp
.fdp
.enabled
) {
4283 return NVME_FDP_DISABLED
| NVME_DNR
;
4286 endgrp
= ns
->endgrp
;
4288 nruhsd
= ns
->fdp
.nphs
* endgrp
->fdp
.nrg
;
4289 trans_len
= sizeof(NvmeRuhStatus
) + nruhsd
* sizeof(NvmeRuhStatusDescr
);
4290 buf
= g_malloc(trans_len
);
4292 trans_len
= MIN(trans_len
, len
);
4294 hdr
= (NvmeRuhStatus
*)buf
;
4295 ruhsd
= (NvmeRuhStatusDescr
*)(buf
+ sizeof(NvmeRuhStatus
));
4297 hdr
->nruhsd
= cpu_to_le16(nruhsd
);
4299 ruhid
= ns
->fdp
.phs
;
4301 for (ph
= 0; ph
< ns
->fdp
.nphs
; ph
++, ruhid
++) {
4302 NvmeRuHandle
*ruh
= &endgrp
->fdp
.ruhs
[*ruhid
];
4304 for (rg
= 0; rg
< endgrp
->fdp
.nrg
; rg
++, ruhsd
++) {
4305 uint16_t pid
= nvme_make_pid(ns
, rg
, ph
);
4307 ruhsd
->pid
= cpu_to_le16(pid
);
4308 ruhsd
->ruhid
= *ruhid
;
4310 ruhsd
->ruamw
= cpu_to_le64(ruh
->rus
[rg
].ruamw
);
4314 return nvme_c2h(n
, buf
, trans_len
, req
);
4317 static uint16_t nvme_io_mgmt_recv(NvmeCtrl
*n
, NvmeRequest
*req
)
4319 NvmeCmd
*cmd
= &req
->cmd
;
4320 uint32_t cdw10
= le32_to_cpu(cmd
->cdw10
);
4321 uint32_t numd
= le32_to_cpu(cmd
->cdw11
);
4322 uint8_t mo
= (cdw10
& 0xff);
4323 size_t len
= (numd
+ 1) << 2;
4326 case NVME_IOMR_MO_NOP
:
4328 case NVME_IOMR_MO_RUH_STATUS
:
4329 return nvme_io_mgmt_recv_ruhs(n
, req
, len
);
4331 return NVME_INVALID_FIELD
| NVME_DNR
;
4335 static uint16_t nvme_io_mgmt_send_ruh_update(NvmeCtrl
*n
, NvmeRequest
*req
)
4337 NvmeCmd
*cmd
= &req
->cmd
;
4338 NvmeNamespace
*ns
= req
->ns
;
4339 uint32_t cdw10
= le32_to_cpu(cmd
->cdw10
);
4340 uint16_t ret
= NVME_SUCCESS
;
4341 uint32_t npid
= (cdw10
>> 1) + 1;
4343 g_autofree
uint16_t *pids
= NULL
;
4344 uint32_t maxnpid
= n
->subsys
->endgrp
.fdp
.nrg
* n
->subsys
->endgrp
.fdp
.nruh
;
4346 if (unlikely(npid
>= MIN(NVME_FDP_MAXPIDS
, maxnpid
))) {
4347 return NVME_INVALID_FIELD
| NVME_DNR
;
4350 pids
= g_new(uint16_t, npid
);
4352 ret
= nvme_h2c(n
, pids
, npid
* sizeof(uint16_t), req
);
4357 for (; i
< npid
; i
++) {
4358 if (!nvme_update_ruh(n
, ns
, pids
[i
])) {
4359 return NVME_INVALID_FIELD
| NVME_DNR
;
4366 static uint16_t nvme_io_mgmt_send(NvmeCtrl
*n
, NvmeRequest
*req
)
4368 NvmeCmd
*cmd
= &req
->cmd
;
4369 uint32_t cdw10
= le32_to_cpu(cmd
->cdw10
);
4370 uint8_t mo
= (cdw10
& 0xff);
4373 case NVME_IOMS_MO_NOP
:
4375 case NVME_IOMS_MO_RUH_UPDATE
:
4376 return nvme_io_mgmt_send_ruh_update(n
, req
);
4378 return NVME_INVALID_FIELD
| NVME_DNR
;
4382 static uint16_t nvme_io_cmd(NvmeCtrl
*n
, NvmeRequest
*req
)
4385 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
4387 trace_pci_nvme_io_cmd(nvme_cid(req
), nsid
, nvme_sqid(req
),
4388 req
->cmd
.opcode
, nvme_io_opc_str(req
->cmd
.opcode
));
4390 if (!nvme_nsid_valid(n
, nsid
)) {
4391 return NVME_INVALID_NSID
| NVME_DNR
;
4395 * In the base NVM command set, Flush may apply to all namespaces
4396 * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4397 * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4399 * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4400 * opcode with a specific command since we cannot determine a unique I/O
4401 * command set. Opcode 0h could have any other meaning than something
4402 * equivalent to flushing and say it DOES have completely different
4403 * semantics in some other command set - does an NSID of FFFFFFFFh then
4404 * mean "for all namespaces, apply whatever command set specific command
4405 * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4406 * whatever command that uses the 0h opcode if, and only if, it allows NSID
4409 * Anyway (and luckily), for now, we do not care about this since the
4410 * device only supports namespace types that includes the NVM Flush command
4411 * (NVM and Zoned), so always do an NVM Flush.
4413 if (req
->cmd
.opcode
== NVME_CMD_FLUSH
) {
4414 return nvme_flush(n
, req
);
4417 ns
= nvme_ns(n
, nsid
);
4418 if (unlikely(!ns
)) {
4419 return NVME_INVALID_FIELD
| NVME_DNR
;
4422 if (!(ns
->iocs
[req
->cmd
.opcode
] & NVME_CMD_EFF_CSUPP
)) {
4423 trace_pci_nvme_err_invalid_opc(req
->cmd
.opcode
);
4424 return NVME_INVALID_OPCODE
| NVME_DNR
;
4431 if (NVME_CMD_FLAGS_FUSE(req
->cmd
.flags
)) {
4432 return NVME_INVALID_FIELD
;
4437 switch (req
->cmd
.opcode
) {
4438 case NVME_CMD_WRITE_ZEROES
:
4439 return nvme_write_zeroes(n
, req
);
4440 case NVME_CMD_ZONE_APPEND
:
4441 return nvme_zone_append(n
, req
);
4442 case NVME_CMD_WRITE
:
4443 return nvme_write(n
, req
);
4445 return nvme_read(n
, req
);
4446 case NVME_CMD_COMPARE
:
4447 return nvme_compare(n
, req
);
4449 return nvme_dsm(n
, req
);
4450 case NVME_CMD_VERIFY
:
4451 return nvme_verify(n
, req
);
4453 return nvme_copy(n
, req
);
4454 case NVME_CMD_ZONE_MGMT_SEND
:
4455 return nvme_zone_mgmt_send(n
, req
);
4456 case NVME_CMD_ZONE_MGMT_RECV
:
4457 return nvme_zone_mgmt_recv(n
, req
);
4458 case NVME_CMD_IO_MGMT_RECV
:
4459 return nvme_io_mgmt_recv(n
, req
);
4460 case NVME_CMD_IO_MGMT_SEND
:
4461 return nvme_io_mgmt_send(n
, req
);
4466 return NVME_INVALID_OPCODE
| NVME_DNR
;
4469 static void nvme_cq_notifier(EventNotifier
*e
)
4471 NvmeCQueue
*cq
= container_of(e
, NvmeCQueue
, notifier
);
4472 NvmeCtrl
*n
= cq
->ctrl
;
4474 if (!event_notifier_test_and_clear(e
)) {
4478 nvme_update_cq_head(cq
);
4480 if (cq
->tail
== cq
->head
) {
4481 if (cq
->irq_enabled
) {
4485 nvme_irq_deassert(n
, cq
);
4488 qemu_bh_schedule(cq
->bh
);
4491 static int nvme_init_cq_ioeventfd(NvmeCQueue
*cq
)
4493 NvmeCtrl
*n
= cq
->ctrl
;
4494 uint16_t offset
= (cq
->cqid
<< 3) + (1 << 2);
4497 ret
= event_notifier_init(&cq
->notifier
, 0);
4502 event_notifier_set_handler(&cq
->notifier
, nvme_cq_notifier
);
4503 memory_region_add_eventfd(&n
->iomem
,
4504 0x1000 + offset
, 4, false, 0, &cq
->notifier
);
4509 static void nvme_sq_notifier(EventNotifier
*e
)
4511 NvmeSQueue
*sq
= container_of(e
, NvmeSQueue
, notifier
);
4513 if (!event_notifier_test_and_clear(e
)) {
4517 nvme_process_sq(sq
);
4520 static int nvme_init_sq_ioeventfd(NvmeSQueue
*sq
)
4522 NvmeCtrl
*n
= sq
->ctrl
;
4523 uint16_t offset
= sq
->sqid
<< 3;
4526 ret
= event_notifier_init(&sq
->notifier
, 0);
4531 event_notifier_set_handler(&sq
->notifier
, nvme_sq_notifier
);
4532 memory_region_add_eventfd(&n
->iomem
,
4533 0x1000 + offset
, 4, false, 0, &sq
->notifier
);
4538 static void nvme_free_sq(NvmeSQueue
*sq
, NvmeCtrl
*n
)
4540 uint16_t offset
= sq
->sqid
<< 3;
4542 n
->sq
[sq
->sqid
] = NULL
;
4543 qemu_bh_delete(sq
->bh
);
4544 if (sq
->ioeventfd_enabled
) {
4545 memory_region_del_eventfd(&n
->iomem
,
4546 0x1000 + offset
, 4, false, 0, &sq
->notifier
);
4547 event_notifier_set_handler(&sq
->notifier
, NULL
);
4548 event_notifier_cleanup(&sq
->notifier
);
4556 static uint16_t nvme_del_sq(NvmeCtrl
*n
, NvmeRequest
*req
)
4558 NvmeDeleteQ
*c
= (NvmeDeleteQ
*)&req
->cmd
;
4559 NvmeRequest
*r
, *next
;
4562 uint16_t qid
= le16_to_cpu(c
->qid
);
4564 if (unlikely(!qid
|| nvme_check_sqid(n
, qid
))) {
4565 trace_pci_nvme_err_invalid_del_sq(qid
);
4566 return NVME_INVALID_QID
| NVME_DNR
;
4569 trace_pci_nvme_del_sq(qid
);
4572 while (!QTAILQ_EMPTY(&sq
->out_req_list
)) {
4573 r
= QTAILQ_FIRST(&sq
->out_req_list
);
4575 blk_aio_cancel(r
->aiocb
);
4578 assert(QTAILQ_EMPTY(&sq
->out_req_list
));
4580 if (!nvme_check_cqid(n
, sq
->cqid
)) {
4581 cq
= n
->cq
[sq
->cqid
];
4582 QTAILQ_REMOVE(&cq
->sq_list
, sq
, entry
);
4585 QTAILQ_FOREACH_SAFE(r
, &cq
->req_list
, entry
, next
) {
4587 QTAILQ_REMOVE(&cq
->req_list
, r
, entry
);
4588 QTAILQ_INSERT_TAIL(&sq
->req_list
, r
, entry
);
4593 nvme_free_sq(sq
, n
);
4594 return NVME_SUCCESS
;
4597 static void nvme_init_sq(NvmeSQueue
*sq
, NvmeCtrl
*n
, uint64_t dma_addr
,
4598 uint16_t sqid
, uint16_t cqid
, uint16_t size
)
4604 sq
->dma_addr
= dma_addr
;
4608 sq
->head
= sq
->tail
= 0;
4609 sq
->io_req
= g_new0(NvmeRequest
, sq
->size
);
4611 QTAILQ_INIT(&sq
->req_list
);
4612 QTAILQ_INIT(&sq
->out_req_list
);
4613 for (i
= 0; i
< sq
->size
; i
++) {
4614 sq
->io_req
[i
].sq
= sq
;
4615 QTAILQ_INSERT_TAIL(&(sq
->req_list
), &sq
->io_req
[i
], entry
);
4618 sq
->bh
= qemu_bh_new_guarded(nvme_process_sq
, sq
,
4619 &DEVICE(sq
->ctrl
)->mem_reentrancy_guard
);
4621 if (n
->dbbuf_enabled
) {
4622 sq
->db_addr
= n
->dbbuf_dbs
+ (sqid
<< 3);
4623 sq
->ei_addr
= n
->dbbuf_eis
+ (sqid
<< 3);
4625 if (n
->params
.ioeventfd
&& sq
->sqid
!= 0) {
4626 if (!nvme_init_sq_ioeventfd(sq
)) {
4627 sq
->ioeventfd_enabled
= true;
4632 assert(n
->cq
[cqid
]);
4634 QTAILQ_INSERT_TAIL(&(cq
->sq_list
), sq
, entry
);
4638 static uint16_t nvme_create_sq(NvmeCtrl
*n
, NvmeRequest
*req
)
4641 NvmeCreateSq
*c
= (NvmeCreateSq
*)&req
->cmd
;
4643 uint16_t cqid
= le16_to_cpu(c
->cqid
);
4644 uint16_t sqid
= le16_to_cpu(c
->sqid
);
4645 uint16_t qsize
= le16_to_cpu(c
->qsize
);
4646 uint16_t qflags
= le16_to_cpu(c
->sq_flags
);
4647 uint64_t prp1
= le64_to_cpu(c
->prp1
);
4649 trace_pci_nvme_create_sq(prp1
, sqid
, cqid
, qsize
, qflags
);
4651 if (unlikely(!cqid
|| nvme_check_cqid(n
, cqid
))) {
4652 trace_pci_nvme_err_invalid_create_sq_cqid(cqid
);
4653 return NVME_INVALID_CQID
| NVME_DNR
;
4655 if (unlikely(!sqid
|| sqid
> n
->conf_ioqpairs
|| n
->sq
[sqid
] != NULL
)) {
4656 trace_pci_nvme_err_invalid_create_sq_sqid(sqid
);
4657 return NVME_INVALID_QID
| NVME_DNR
;
4659 if (unlikely(!qsize
|| qsize
> NVME_CAP_MQES(ldq_le_p(&n
->bar
.cap
)))) {
4660 trace_pci_nvme_err_invalid_create_sq_size(qsize
);
4661 return NVME_MAX_QSIZE_EXCEEDED
| NVME_DNR
;
4663 if (unlikely(prp1
& (n
->page_size
- 1))) {
4664 trace_pci_nvme_err_invalid_create_sq_addr(prp1
);
4665 return NVME_INVALID_PRP_OFFSET
| NVME_DNR
;
4667 if (unlikely(!(NVME_SQ_FLAGS_PC(qflags
)))) {
4668 trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags
));
4669 return NVME_INVALID_FIELD
| NVME_DNR
;
4671 sq
= g_malloc0(sizeof(*sq
));
4672 nvme_init_sq(sq
, n
, prp1
, sqid
, cqid
, qsize
+ 1);
4673 return NVME_SUCCESS
;
4677 uint64_t units_read
;
4678 uint64_t units_written
;
4679 uint64_t read_commands
;
4680 uint64_t write_commands
;
4683 static void nvme_set_blk_stats(NvmeNamespace
*ns
, struct nvme_stats
*stats
)
4685 BlockAcctStats
*s
= blk_get_stats(ns
->blkconf
.blk
);
4687 stats
->units_read
+= s
->nr_bytes
[BLOCK_ACCT_READ
];
4688 stats
->units_written
+= s
->nr_bytes
[BLOCK_ACCT_WRITE
];
4689 stats
->read_commands
+= s
->nr_ops
[BLOCK_ACCT_READ
];
4690 stats
->write_commands
+= s
->nr_ops
[BLOCK_ACCT_WRITE
];
4693 static uint16_t nvme_smart_info(NvmeCtrl
*n
, uint8_t rae
, uint32_t buf_len
,
4694 uint64_t off
, NvmeRequest
*req
)
4696 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
4697 struct nvme_stats stats
= { 0 };
4698 NvmeSmartLog smart
= { 0 };
4702 uint64_t u_read
, u_written
;
4704 if (off
>= sizeof(smart
)) {
4705 return NVME_INVALID_FIELD
| NVME_DNR
;
4708 if (nsid
!= 0xffffffff) {
4709 ns
= nvme_ns(n
, nsid
);
4711 return NVME_INVALID_NSID
| NVME_DNR
;
4713 nvme_set_blk_stats(ns
, &stats
);
4717 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
4722 nvme_set_blk_stats(ns
, &stats
);
4726 trans_len
= MIN(sizeof(smart
) - off
, buf_len
);
4727 smart
.critical_warning
= n
->smart_critical_warning
;
4729 u_read
= DIV_ROUND_UP(stats
.units_read
>> BDRV_SECTOR_BITS
, 1000);
4730 u_written
= DIV_ROUND_UP(stats
.units_written
>> BDRV_SECTOR_BITS
, 1000);
4732 smart
.data_units_read
[0] = cpu_to_le64(u_read
);
4733 smart
.data_units_written
[0] = cpu_to_le64(u_written
);
4734 smart
.host_read_commands
[0] = cpu_to_le64(stats
.read_commands
);
4735 smart
.host_write_commands
[0] = cpu_to_le64(stats
.write_commands
);
4737 smart
.temperature
= cpu_to_le16(n
->temperature
);
4739 if ((n
->temperature
>= n
->features
.temp_thresh_hi
) ||
4740 (n
->temperature
<= n
->features
.temp_thresh_low
)) {
4741 smart
.critical_warning
|= NVME_SMART_TEMPERATURE
;
4744 current_ms
= qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
);
4745 smart
.power_on_hours
[0] =
4746 cpu_to_le64((((current_ms
- n
->starttime_ms
) / 1000) / 60) / 60);
4749 nvme_clear_events(n
, NVME_AER_TYPE_SMART
);
4752 return nvme_c2h(n
, (uint8_t *) &smart
+ off
, trans_len
, req
);
4755 static uint16_t nvme_endgrp_info(NvmeCtrl
*n
, uint8_t rae
, uint32_t buf_len
,
4756 uint64_t off
, NvmeRequest
*req
)
4758 uint32_t dw11
= le32_to_cpu(req
->cmd
.cdw11
);
4759 uint16_t endgrpid
= (dw11
>> 16) & 0xffff;
4760 struct nvme_stats stats
= {};
4761 NvmeEndGrpLog info
= {};
4764 if (!n
->subsys
|| endgrpid
!= 0x1) {
4765 return NVME_INVALID_FIELD
| NVME_DNR
;
4768 if (off
>= sizeof(info
)) {
4769 return NVME_INVALID_FIELD
| NVME_DNR
;
4772 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
4773 NvmeNamespace
*ns
= nvme_subsys_ns(n
->subsys
, i
);
4778 nvme_set_blk_stats(ns
, &stats
);
4781 info
.data_units_read
[0] =
4782 cpu_to_le64(DIV_ROUND_UP(stats
.units_read
/ 1000000000, 1000000000));
4783 info
.data_units_written
[0] =
4784 cpu_to_le64(DIV_ROUND_UP(stats
.units_written
/ 1000000000, 1000000000));
4785 info
.media_units_written
[0] =
4786 cpu_to_le64(DIV_ROUND_UP(stats
.units_written
/ 1000000000, 1000000000));
4788 info
.host_read_commands
[0] = cpu_to_le64(stats
.read_commands
);
4789 info
.host_write_commands
[0] = cpu_to_le64(stats
.write_commands
);
4791 buf_len
= MIN(sizeof(info
) - off
, buf_len
);
4793 return nvme_c2h(n
, (uint8_t *)&info
+ off
, buf_len
, req
);
4797 static uint16_t nvme_fw_log_info(NvmeCtrl
*n
, uint32_t buf_len
, uint64_t off
,
4801 NvmeFwSlotInfoLog fw_log
= {
4805 if (off
>= sizeof(fw_log
)) {
4806 return NVME_INVALID_FIELD
| NVME_DNR
;
4809 strpadcpy((char *)&fw_log
.frs1
, sizeof(fw_log
.frs1
), "1.0", ' ');
4810 trans_len
= MIN(sizeof(fw_log
) - off
, buf_len
);
4812 return nvme_c2h(n
, (uint8_t *) &fw_log
+ off
, trans_len
, req
);
4815 static uint16_t nvme_error_info(NvmeCtrl
*n
, uint8_t rae
, uint32_t buf_len
,
4816 uint64_t off
, NvmeRequest
*req
)
4819 NvmeErrorLog errlog
;
4821 if (off
>= sizeof(errlog
)) {
4822 return NVME_INVALID_FIELD
| NVME_DNR
;
4826 nvme_clear_events(n
, NVME_AER_TYPE_ERROR
);
4829 memset(&errlog
, 0x0, sizeof(errlog
));
4830 trans_len
= MIN(sizeof(errlog
) - off
, buf_len
);
4832 return nvme_c2h(n
, (uint8_t *)&errlog
, trans_len
, req
);
4835 static uint16_t nvme_changed_nslist(NvmeCtrl
*n
, uint8_t rae
, uint32_t buf_len
,
4836 uint64_t off
, NvmeRequest
*req
)
4838 uint32_t nslist
[1024];
4843 if (off
>= sizeof(nslist
)) {
4844 trace_pci_nvme_err_invalid_log_page_offset(off
, sizeof(nslist
));
4845 return NVME_INVALID_FIELD
| NVME_DNR
;
4848 memset(nslist
, 0x0, sizeof(nslist
));
4849 trans_len
= MIN(sizeof(nslist
) - off
, buf_len
);
4851 while ((nsid
= find_first_bit(n
->changed_nsids
, NVME_CHANGED_NSID_SIZE
)) !=
4852 NVME_CHANGED_NSID_SIZE
) {
4854 * If more than 1024 namespaces, the first entry in the log page should
4855 * be set to FFFFFFFFh and the others to 0 as spec.
4857 if (i
== ARRAY_SIZE(nslist
)) {
4858 memset(nslist
, 0x0, sizeof(nslist
));
4859 nslist
[0] = 0xffffffff;
4864 clear_bit(nsid
, n
->changed_nsids
);
4868 * Remove all the remaining list entries in case returns directly due to
4869 * more than 1024 namespaces.
4871 if (nslist
[0] == 0xffffffff) {
4872 bitmap_zero(n
->changed_nsids
, NVME_CHANGED_NSID_SIZE
);
4876 nvme_clear_events(n
, NVME_AER_TYPE_NOTICE
);
4879 return nvme_c2h(n
, ((uint8_t *)nslist
) + off
, trans_len
, req
);
4882 static uint16_t nvme_cmd_effects(NvmeCtrl
*n
, uint8_t csi
, uint32_t buf_len
,
4883 uint64_t off
, NvmeRequest
*req
)
4885 NvmeEffectsLog log
= {};
4886 const uint32_t *src_iocs
= NULL
;
4889 if (off
>= sizeof(log
)) {
4890 trace_pci_nvme_err_invalid_log_page_offset(off
, sizeof(log
));
4891 return NVME_INVALID_FIELD
| NVME_DNR
;
4894 switch (NVME_CC_CSS(ldl_le_p(&n
->bar
.cc
))) {
4895 case NVME_CC_CSS_NVM
:
4896 src_iocs
= nvme_cse_iocs_nvm
;
4898 case NVME_CC_CSS_ADMIN_ONLY
:
4900 case NVME_CC_CSS_CSI
:
4903 src_iocs
= nvme_cse_iocs_nvm
;
4905 case NVME_CSI_ZONED
:
4906 src_iocs
= nvme_cse_iocs_zoned
;
4911 memcpy(log
.acs
, nvme_cse_acs
, sizeof(nvme_cse_acs
));
4914 memcpy(log
.iocs
, src_iocs
, sizeof(log
.iocs
));
4917 trans_len
= MIN(sizeof(log
) - off
, buf_len
);
4919 return nvme_c2h(n
, ((uint8_t *)&log
) + off
, trans_len
, req
);
4922 static size_t sizeof_fdp_conf_descr(size_t nruh
, size_t vss
)
4924 size_t entry_siz
= sizeof(NvmeFdpDescrHdr
) + nruh
* sizeof(NvmeRuhDescr
)
4926 return ROUND_UP(entry_siz
, 8);
4929 static uint16_t nvme_fdp_confs(NvmeCtrl
*n
, uint32_t endgrpid
, uint32_t buf_len
,
4930 uint64_t off
, NvmeRequest
*req
)
4932 uint32_t log_size
, trans_len
;
4933 g_autofree
uint8_t *buf
= NULL
;
4934 NvmeFdpDescrHdr
*hdr
;
4936 NvmeEnduranceGroup
*endgrp
;
4937 NvmeFdpConfsHdr
*log
;
4938 size_t nruh
, fdp_descr_size
;
4941 if (endgrpid
!= 1 || !n
->subsys
) {
4942 return NVME_INVALID_FIELD
| NVME_DNR
;
4945 endgrp
= &n
->subsys
->endgrp
;
4947 if (endgrp
->fdp
.enabled
) {
4948 nruh
= endgrp
->fdp
.nruh
;
4953 fdp_descr_size
= sizeof_fdp_conf_descr(nruh
, FDPVSS
);
4954 log_size
= sizeof(NvmeFdpConfsHdr
) + fdp_descr_size
;
4956 if (off
>= log_size
) {
4957 return NVME_INVALID_FIELD
| NVME_DNR
;
4960 trans_len
= MIN(log_size
- off
, buf_len
);
4962 buf
= g_malloc0(log_size
);
4963 log
= (NvmeFdpConfsHdr
*)buf
;
4964 hdr
= (NvmeFdpDescrHdr
*)(log
+ 1);
4965 ruhd
= (NvmeRuhDescr
*)(buf
+ sizeof(*log
) + sizeof(*hdr
));
4967 log
->num_confs
= cpu_to_le16(0);
4968 log
->size
= cpu_to_le32(log_size
);
4970 hdr
->descr_size
= cpu_to_le16(fdp_descr_size
);
4971 if (endgrp
->fdp
.enabled
) {
4972 hdr
->fdpa
= FIELD_DP8(hdr
->fdpa
, FDPA
, VALID
, 1);
4973 hdr
->fdpa
= FIELD_DP8(hdr
->fdpa
, FDPA
, RGIF
, endgrp
->fdp
.rgif
);
4974 hdr
->nrg
= cpu_to_le16(endgrp
->fdp
.nrg
);
4975 hdr
->nruh
= cpu_to_le16(endgrp
->fdp
.nruh
);
4976 hdr
->maxpids
= cpu_to_le16(NVME_FDP_MAXPIDS
- 1);
4977 hdr
->nnss
= cpu_to_le32(NVME_MAX_NAMESPACES
);
4978 hdr
->runs
= cpu_to_le64(endgrp
->fdp
.runs
);
4980 for (i
= 0; i
< nruh
; i
++) {
4981 ruhd
->ruht
= NVME_RUHT_INITIALLY_ISOLATED
;
4985 /* 1 bit for RUH in PIF -> 2 RUHs max. */
4986 hdr
->nrg
= cpu_to_le16(1);
4987 hdr
->nruh
= cpu_to_le16(1);
4988 hdr
->maxpids
= cpu_to_le16(NVME_FDP_MAXPIDS
- 1);
4989 hdr
->nnss
= cpu_to_le32(1);
4990 hdr
->runs
= cpu_to_le64(96 * MiB
);
4992 ruhd
->ruht
= NVME_RUHT_INITIALLY_ISOLATED
;
4995 return nvme_c2h(n
, (uint8_t *)buf
+ off
, trans_len
, req
);
4998 static uint16_t nvme_fdp_ruh_usage(NvmeCtrl
*n
, uint32_t endgrpid
,
4999 uint32_t dw10
, uint32_t dw12
,
5000 uint32_t buf_len
, uint64_t off
,
5005 NvmeRuhuDescr
*ruhud
;
5006 NvmeEnduranceGroup
*endgrp
;
5007 g_autofree
uint8_t *buf
= NULL
;
5008 uint32_t log_size
, trans_len
;
5011 if (endgrpid
!= 1 || !n
->subsys
) {
5012 return NVME_INVALID_FIELD
| NVME_DNR
;
5015 endgrp
= &n
->subsys
->endgrp
;
5017 if (!endgrp
->fdp
.enabled
) {
5018 return NVME_FDP_DISABLED
| NVME_DNR
;
5021 log_size
= sizeof(NvmeRuhuLog
) + endgrp
->fdp
.nruh
* sizeof(NvmeRuhuDescr
);
5023 if (off
>= log_size
) {
5024 return NVME_INVALID_FIELD
| NVME_DNR
;
5027 trans_len
= MIN(log_size
- off
, buf_len
);
5029 buf
= g_malloc0(log_size
);
5030 hdr
= (NvmeRuhuLog
*)buf
;
5031 ruhud
= (NvmeRuhuDescr
*)(hdr
+ 1);
5033 ruh
= endgrp
->fdp
.ruhs
;
5034 hdr
->nruh
= cpu_to_le16(endgrp
->fdp
.nruh
);
5036 for (i
= 0; i
< endgrp
->fdp
.nruh
; i
++, ruhud
++, ruh
++) {
5037 ruhud
->ruha
= ruh
->ruha
;
5040 return nvme_c2h(n
, (uint8_t *)buf
+ off
, trans_len
, req
);
5043 static uint16_t nvme_fdp_stats(NvmeCtrl
*n
, uint32_t endgrpid
, uint32_t buf_len
,
5044 uint64_t off
, NvmeRequest
*req
)
5046 NvmeEnduranceGroup
*endgrp
;
5047 NvmeFdpStatsLog log
= {};
5050 if (off
>= sizeof(NvmeFdpStatsLog
)) {
5051 return NVME_INVALID_FIELD
| NVME_DNR
;
5054 if (endgrpid
!= 1 || !n
->subsys
) {
5055 return NVME_INVALID_FIELD
| NVME_DNR
;
5058 if (!n
->subsys
->endgrp
.fdp
.enabled
) {
5059 return NVME_FDP_DISABLED
| NVME_DNR
;
5062 endgrp
= &n
->subsys
->endgrp
;
5064 trans_len
= MIN(sizeof(log
) - off
, buf_len
);
5066 /* spec value is 128 bit, we only use 64 bit */
5067 log
.hbmw
[0] = cpu_to_le64(endgrp
->fdp
.hbmw
);
5068 log
.mbmw
[0] = cpu_to_le64(endgrp
->fdp
.mbmw
);
5069 log
.mbe
[0] = cpu_to_le64(endgrp
->fdp
.mbe
);
5071 return nvme_c2h(n
, (uint8_t *)&log
+ off
, trans_len
, req
);
5074 static uint16_t nvme_fdp_events(NvmeCtrl
*n
, uint32_t endgrpid
,
5075 uint32_t buf_len
, uint64_t off
,
5078 NvmeEnduranceGroup
*endgrp
;
5079 NvmeCmd
*cmd
= &req
->cmd
;
5080 bool host_events
= (cmd
->cdw10
>> 8) & 0x1;
5081 uint32_t log_size
, trans_len
;
5082 NvmeFdpEventBuffer
*ebuf
;
5083 g_autofree NvmeFdpEventsLog
*elog
= NULL
;
5084 NvmeFdpEvent
*event
;
5086 if (endgrpid
!= 1 || !n
->subsys
) {
5087 return NVME_INVALID_FIELD
| NVME_DNR
;
5090 endgrp
= &n
->subsys
->endgrp
;
5092 if (!endgrp
->fdp
.enabled
) {
5093 return NVME_FDP_DISABLED
| NVME_DNR
;
5097 ebuf
= &endgrp
->fdp
.host_events
;
5099 ebuf
= &endgrp
->fdp
.ctrl_events
;
5102 log_size
= sizeof(NvmeFdpEventsLog
) + ebuf
->nelems
* sizeof(NvmeFdpEvent
);
5103 trans_len
= MIN(log_size
- off
, buf_len
);
5104 elog
= g_malloc0(log_size
);
5105 elog
->num_events
= cpu_to_le32(ebuf
->nelems
);
5106 event
= (NvmeFdpEvent
*)(elog
+ 1);
5108 if (ebuf
->nelems
&& ebuf
->start
== ebuf
->next
) {
5109 unsigned int nelems
= (NVME_FDP_MAX_EVENTS
- ebuf
->start
);
5110 /* wrap over, copy [start;NVME_FDP_MAX_EVENTS[ and [0; next[ */
5111 memcpy(event
, &ebuf
->events
[ebuf
->start
],
5112 sizeof(NvmeFdpEvent
) * nelems
);
5113 memcpy(event
+ nelems
, ebuf
->events
,
5114 sizeof(NvmeFdpEvent
) * ebuf
->next
);
5115 } else if (ebuf
->start
< ebuf
->next
) {
5116 memcpy(event
, &ebuf
->events
[ebuf
->start
],
5117 sizeof(NvmeFdpEvent
) * (ebuf
->next
- ebuf
->start
));
5120 return nvme_c2h(n
, (uint8_t *)elog
+ off
, trans_len
, req
);
5123 static uint16_t nvme_get_log(NvmeCtrl
*n
, NvmeRequest
*req
)
5125 NvmeCmd
*cmd
= &req
->cmd
;
5127 uint32_t dw10
= le32_to_cpu(cmd
->cdw10
);
5128 uint32_t dw11
= le32_to_cpu(cmd
->cdw11
);
5129 uint32_t dw12
= le32_to_cpu(cmd
->cdw12
);
5130 uint32_t dw13
= le32_to_cpu(cmd
->cdw13
);
5131 uint8_t lid
= dw10
& 0xff;
5132 uint8_t lsp
= (dw10
>> 8) & 0xf;
5133 uint8_t rae
= (dw10
>> 15) & 0x1;
5134 uint8_t csi
= le32_to_cpu(cmd
->cdw14
) >> 24;
5135 uint32_t numdl
, numdu
, lspi
;
5136 uint64_t off
, lpol
, lpou
;
5140 numdl
= (dw10
>> 16);
5141 numdu
= (dw11
& 0xffff);
5142 lspi
= (dw11
>> 16);
5146 len
= (((numdu
<< 16) | numdl
) + 1) << 2;
5147 off
= (lpou
<< 32ULL) | lpol
;
5150 return NVME_INVALID_FIELD
| NVME_DNR
;
5153 trace_pci_nvme_get_log(nvme_cid(req
), lid
, lsp
, rae
, len
, off
);
5155 status
= nvme_check_mdts(n
, len
);
5161 case NVME_LOG_ERROR_INFO
:
5162 return nvme_error_info(n
, rae
, len
, off
, req
);
5163 case NVME_LOG_SMART_INFO
:
5164 return nvme_smart_info(n
, rae
, len
, off
, req
);
5165 case NVME_LOG_FW_SLOT_INFO
:
5166 return nvme_fw_log_info(n
, len
, off
, req
);
5167 case NVME_LOG_CHANGED_NSLIST
:
5168 return nvme_changed_nslist(n
, rae
, len
, off
, req
);
5169 case NVME_LOG_CMD_EFFECTS
:
5170 return nvme_cmd_effects(n
, csi
, len
, off
, req
);
5171 case NVME_LOG_ENDGRP
:
5172 return nvme_endgrp_info(n
, rae
, len
, off
, req
);
5173 case NVME_LOG_FDP_CONFS
:
5174 return nvme_fdp_confs(n
, lspi
, len
, off
, req
);
5175 case NVME_LOG_FDP_RUH_USAGE
:
5176 return nvme_fdp_ruh_usage(n
, lspi
, dw10
, dw12
, len
, off
, req
);
5177 case NVME_LOG_FDP_STATS
:
5178 return nvme_fdp_stats(n
, lspi
, len
, off
, req
);
5179 case NVME_LOG_FDP_EVENTS
:
5180 return nvme_fdp_events(n
, lspi
, len
, off
, req
);
5182 trace_pci_nvme_err_invalid_log_page(nvme_cid(req
), lid
);
5183 return NVME_INVALID_FIELD
| NVME_DNR
;
5187 static void nvme_free_cq(NvmeCQueue
*cq
, NvmeCtrl
*n
)
5189 PCIDevice
*pci
= PCI_DEVICE(n
);
5190 uint16_t offset
= (cq
->cqid
<< 3) + (1 << 2);
5192 n
->cq
[cq
->cqid
] = NULL
;
5193 qemu_bh_delete(cq
->bh
);
5194 if (cq
->ioeventfd_enabled
) {
5195 memory_region_del_eventfd(&n
->iomem
,
5196 0x1000 + offset
, 4, false, 0, &cq
->notifier
);
5197 event_notifier_set_handler(&cq
->notifier
, NULL
);
5198 event_notifier_cleanup(&cq
->notifier
);
5200 if (msix_enabled(pci
)) {
5201 msix_vector_unuse(pci
, cq
->vector
);
5208 static uint16_t nvme_del_cq(NvmeCtrl
*n
, NvmeRequest
*req
)
5210 NvmeDeleteQ
*c
= (NvmeDeleteQ
*)&req
->cmd
;
5212 uint16_t qid
= le16_to_cpu(c
->qid
);
5214 if (unlikely(!qid
|| nvme_check_cqid(n
, qid
))) {
5215 trace_pci_nvme_err_invalid_del_cq_cqid(qid
);
5216 return NVME_INVALID_CQID
| NVME_DNR
;
5220 if (unlikely(!QTAILQ_EMPTY(&cq
->sq_list
))) {
5221 trace_pci_nvme_err_invalid_del_cq_notempty(qid
);
5222 return NVME_INVALID_QUEUE_DEL
;
5225 if (cq
->irq_enabled
&& cq
->tail
!= cq
->head
) {
5229 nvme_irq_deassert(n
, cq
);
5230 trace_pci_nvme_del_cq(qid
);
5231 nvme_free_cq(cq
, n
);
5232 return NVME_SUCCESS
;
5235 static void nvme_init_cq(NvmeCQueue
*cq
, NvmeCtrl
*n
, uint64_t dma_addr
,
5236 uint16_t cqid
, uint16_t vector
, uint16_t size
,
5237 uint16_t irq_enabled
)
5239 PCIDevice
*pci
= PCI_DEVICE(n
);
5241 if (msix_enabled(pci
)) {
5242 msix_vector_use(pci
, vector
);
5247 cq
->dma_addr
= dma_addr
;
5249 cq
->irq_enabled
= irq_enabled
;
5250 cq
->vector
= vector
;
5251 cq
->head
= cq
->tail
= 0;
5252 QTAILQ_INIT(&cq
->req_list
);
5253 QTAILQ_INIT(&cq
->sq_list
);
5254 if (n
->dbbuf_enabled
) {
5255 cq
->db_addr
= n
->dbbuf_dbs
+ (cqid
<< 3) + (1 << 2);
5256 cq
->ei_addr
= n
->dbbuf_eis
+ (cqid
<< 3) + (1 << 2);
5258 if (n
->params
.ioeventfd
&& cqid
!= 0) {
5259 if (!nvme_init_cq_ioeventfd(cq
)) {
5260 cq
->ioeventfd_enabled
= true;
5265 cq
->bh
= qemu_bh_new_guarded(nvme_post_cqes
, cq
,
5266 &DEVICE(cq
->ctrl
)->mem_reentrancy_guard
);
5269 static uint16_t nvme_create_cq(NvmeCtrl
*n
, NvmeRequest
*req
)
5272 NvmeCreateCq
*c
= (NvmeCreateCq
*)&req
->cmd
;
5273 uint16_t cqid
= le16_to_cpu(c
->cqid
);
5274 uint16_t vector
= le16_to_cpu(c
->irq_vector
);
5275 uint16_t qsize
= le16_to_cpu(c
->qsize
);
5276 uint16_t qflags
= le16_to_cpu(c
->cq_flags
);
5277 uint64_t prp1
= le64_to_cpu(c
->prp1
);
5279 trace_pci_nvme_create_cq(prp1
, cqid
, vector
, qsize
, qflags
,
5280 NVME_CQ_FLAGS_IEN(qflags
) != 0);
5282 if (unlikely(!cqid
|| cqid
> n
->conf_ioqpairs
|| n
->cq
[cqid
] != NULL
)) {
5283 trace_pci_nvme_err_invalid_create_cq_cqid(cqid
);
5284 return NVME_INVALID_QID
| NVME_DNR
;
5286 if (unlikely(!qsize
|| qsize
> NVME_CAP_MQES(ldq_le_p(&n
->bar
.cap
)))) {
5287 trace_pci_nvme_err_invalid_create_cq_size(qsize
);
5288 return NVME_MAX_QSIZE_EXCEEDED
| NVME_DNR
;
5290 if (unlikely(prp1
& (n
->page_size
- 1))) {
5291 trace_pci_nvme_err_invalid_create_cq_addr(prp1
);
5292 return NVME_INVALID_PRP_OFFSET
| NVME_DNR
;
5294 if (unlikely(!msix_enabled(PCI_DEVICE(n
)) && vector
)) {
5295 trace_pci_nvme_err_invalid_create_cq_vector(vector
);
5296 return NVME_INVALID_IRQ_VECTOR
| NVME_DNR
;
5298 if (unlikely(vector
>= n
->conf_msix_qsize
)) {
5299 trace_pci_nvme_err_invalid_create_cq_vector(vector
);
5300 return NVME_INVALID_IRQ_VECTOR
| NVME_DNR
;
5302 if (unlikely(!(NVME_CQ_FLAGS_PC(qflags
)))) {
5303 trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags
));
5304 return NVME_INVALID_FIELD
| NVME_DNR
;
5307 cq
= g_malloc0(sizeof(*cq
));
5308 nvme_init_cq(cq
, n
, prp1
, cqid
, vector
, qsize
+ 1,
5309 NVME_CQ_FLAGS_IEN(qflags
));
5312 * It is only required to set qs_created when creating a completion queue;
5313 * creating a submission queue without a matching completion queue will
5316 n
->qs_created
= true;
5317 return NVME_SUCCESS
;
5320 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl
*n
, NvmeRequest
*req
)
5322 uint8_t id
[NVME_IDENTIFY_DATA_SIZE
] = {};
5324 return nvme_c2h(n
, id
, sizeof(id
), req
);
5327 static uint16_t nvme_identify_ctrl(NvmeCtrl
*n
, NvmeRequest
*req
)
5329 trace_pci_nvme_identify_ctrl();
5331 return nvme_c2h(n
, (uint8_t *)&n
->id_ctrl
, sizeof(n
->id_ctrl
), req
);
5334 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl
*n
, NvmeRequest
*req
)
5336 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5337 uint8_t id
[NVME_IDENTIFY_DATA_SIZE
] = {};
5338 NvmeIdCtrlNvm
*id_nvm
= (NvmeIdCtrlNvm
*)&id
;
5340 trace_pci_nvme_identify_ctrl_csi(c
->csi
);
5344 id_nvm
->vsl
= n
->params
.vsl
;
5345 id_nvm
->dmrsl
= cpu_to_le32(n
->dmrsl
);
5348 case NVME_CSI_ZONED
:
5349 ((NvmeIdCtrlZoned
*)&id
)->zasl
= n
->params
.zasl
;
5353 return NVME_INVALID_FIELD
| NVME_DNR
;
5356 return nvme_c2h(n
, id
, sizeof(id
), req
);
5359 static uint16_t nvme_identify_ns(NvmeCtrl
*n
, NvmeRequest
*req
, bool active
)
5362 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5363 uint32_t nsid
= le32_to_cpu(c
->nsid
);
5365 trace_pci_nvme_identify_ns(nsid
);
5367 if (!nvme_nsid_valid(n
, nsid
) || nsid
== NVME_NSID_BROADCAST
) {
5368 return NVME_INVALID_NSID
| NVME_DNR
;
5371 ns
= nvme_ns(n
, nsid
);
5372 if (unlikely(!ns
)) {
5374 ns
= nvme_subsys_ns(n
->subsys
, nsid
);
5376 return nvme_rpt_empty_id_struct(n
, req
);
5379 return nvme_rpt_empty_id_struct(n
, req
);
5383 if (active
|| ns
->csi
== NVME_CSI_NVM
) {
5384 return nvme_c2h(n
, (uint8_t *)&ns
->id_ns
, sizeof(NvmeIdNs
), req
);
5387 return NVME_INVALID_CMD_SET
| NVME_DNR
;
5390 static uint16_t nvme_identify_ctrl_list(NvmeCtrl
*n
, NvmeRequest
*req
,
5393 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5394 uint32_t nsid
= le32_to_cpu(c
->nsid
);
5395 uint16_t min_id
= le16_to_cpu(c
->ctrlid
);
5396 uint16_t list
[NVME_CONTROLLER_LIST_SIZE
] = {};
5397 uint16_t *ids
= &list
[1];
5400 int cntlid
, nr_ids
= 0;
5402 trace_pci_nvme_identify_ctrl_list(c
->cns
, min_id
);
5405 return NVME_INVALID_FIELD
| NVME_DNR
;
5409 if (nsid
== NVME_NSID_BROADCAST
) {
5410 return NVME_INVALID_FIELD
| NVME_DNR
;
5413 ns
= nvme_subsys_ns(n
->subsys
, nsid
);
5415 return NVME_INVALID_FIELD
| NVME_DNR
;
5419 for (cntlid
= min_id
; cntlid
< ARRAY_SIZE(n
->subsys
->ctrls
); cntlid
++) {
5420 ctrl
= nvme_subsys_ctrl(n
->subsys
, cntlid
);
5425 if (attached
&& !nvme_ns(ctrl
, nsid
)) {
5429 ids
[nr_ids
++] = cntlid
;
5434 return nvme_c2h(n
, (uint8_t *)list
, sizeof(list
), req
);
5437 static uint16_t nvme_identify_pri_ctrl_cap(NvmeCtrl
*n
, NvmeRequest
*req
)
5439 trace_pci_nvme_identify_pri_ctrl_cap(le16_to_cpu(n
->pri_ctrl_cap
.cntlid
));
5441 return nvme_c2h(n
, (uint8_t *)&n
->pri_ctrl_cap
,
5442 sizeof(NvmePriCtrlCap
), req
);
5445 static uint16_t nvme_identify_sec_ctrl_list(NvmeCtrl
*n
, NvmeRequest
*req
)
5447 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5448 uint16_t pri_ctrl_id
= le16_to_cpu(n
->pri_ctrl_cap
.cntlid
);
5449 uint16_t min_id
= le16_to_cpu(c
->ctrlid
);
5450 uint8_t num_sec_ctrl
= n
->sec_ctrl_list
.numcntl
;
5451 NvmeSecCtrlList list
= {0};
5454 for (i
= 0; i
< num_sec_ctrl
; i
++) {
5455 if (n
->sec_ctrl_list
.sec
[i
].scid
>= min_id
) {
5456 list
.numcntl
= num_sec_ctrl
- i
;
5457 memcpy(&list
.sec
, n
->sec_ctrl_list
.sec
+ i
,
5458 list
.numcntl
* sizeof(NvmeSecCtrlEntry
));
5463 trace_pci_nvme_identify_sec_ctrl_list(pri_ctrl_id
, list
.numcntl
);
5465 return nvme_c2h(n
, (uint8_t *)&list
, sizeof(list
), req
);
5468 static uint16_t nvme_identify_ns_csi(NvmeCtrl
*n
, NvmeRequest
*req
,
5472 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5473 uint32_t nsid
= le32_to_cpu(c
->nsid
);
5475 trace_pci_nvme_identify_ns_csi(nsid
, c
->csi
);
5477 if (!nvme_nsid_valid(n
, nsid
) || nsid
== NVME_NSID_BROADCAST
) {
5478 return NVME_INVALID_NSID
| NVME_DNR
;
5481 ns
= nvme_ns(n
, nsid
);
5482 if (unlikely(!ns
)) {
5484 ns
= nvme_subsys_ns(n
->subsys
, nsid
);
5486 return nvme_rpt_empty_id_struct(n
, req
);
5489 return nvme_rpt_empty_id_struct(n
, req
);
5493 if (c
->csi
== NVME_CSI_NVM
) {
5494 return nvme_c2h(n
, (uint8_t *)&ns
->id_ns_nvm
, sizeof(NvmeIdNsNvm
),
5496 } else if (c
->csi
== NVME_CSI_ZONED
&& ns
->csi
== NVME_CSI_ZONED
) {
5497 return nvme_c2h(n
, (uint8_t *)ns
->id_ns_zoned
, sizeof(NvmeIdNsZoned
),
5501 return NVME_INVALID_FIELD
| NVME_DNR
;
5504 static uint16_t nvme_identify_nslist(NvmeCtrl
*n
, NvmeRequest
*req
,
5508 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5509 uint32_t min_nsid
= le32_to_cpu(c
->nsid
);
5510 uint8_t list
[NVME_IDENTIFY_DATA_SIZE
] = {};
5511 static const int data_len
= sizeof(list
);
5512 uint32_t *list_ptr
= (uint32_t *)list
;
5515 trace_pci_nvme_identify_nslist(min_nsid
);
5518 * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
5519 * since the Active Namespace ID List should return namespaces with ids
5520 * *higher* than the NSID specified in the command. This is also specified
5521 * in the spec (NVM Express v1.3d, Section 5.15.4).
5523 if (min_nsid
>= NVME_NSID_BROADCAST
- 1) {
5524 return NVME_INVALID_NSID
| NVME_DNR
;
5527 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
5531 ns
= nvme_subsys_ns(n
->subsys
, i
);
5539 if (ns
->params
.nsid
<= min_nsid
) {
5542 list_ptr
[j
++] = cpu_to_le32(ns
->params
.nsid
);
5543 if (j
== data_len
/ sizeof(uint32_t)) {
5548 return nvme_c2h(n
, list
, data_len
, req
);
5551 static uint16_t nvme_identify_nslist_csi(NvmeCtrl
*n
, NvmeRequest
*req
,
5555 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5556 uint32_t min_nsid
= le32_to_cpu(c
->nsid
);
5557 uint8_t list
[NVME_IDENTIFY_DATA_SIZE
] = {};
5558 static const int data_len
= sizeof(list
);
5559 uint32_t *list_ptr
= (uint32_t *)list
;
5562 trace_pci_nvme_identify_nslist_csi(min_nsid
, c
->csi
);
5565 * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
5567 if (min_nsid
>= NVME_NSID_BROADCAST
- 1) {
5568 return NVME_INVALID_NSID
| NVME_DNR
;
5571 if (c
->csi
!= NVME_CSI_NVM
&& c
->csi
!= NVME_CSI_ZONED
) {
5572 return NVME_INVALID_FIELD
| NVME_DNR
;
5575 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
5579 ns
= nvme_subsys_ns(n
->subsys
, i
);
5587 if (ns
->params
.nsid
<= min_nsid
|| c
->csi
!= ns
->csi
) {
5590 list_ptr
[j
++] = cpu_to_le32(ns
->params
.nsid
);
5591 if (j
== data_len
/ sizeof(uint32_t)) {
5596 return nvme_c2h(n
, list
, data_len
, req
);
5599 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl
*n
, NvmeRequest
*req
)
5602 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5603 uint32_t nsid
= le32_to_cpu(c
->nsid
);
5604 uint8_t list
[NVME_IDENTIFY_DATA_SIZE
] = {};
5605 uint8_t *pos
= list
;
5608 uint8_t v
[NVME_NIDL_UUID
];
5609 } QEMU_PACKED uuid
= {};
5613 } QEMU_PACKED eui64
= {};
5617 } QEMU_PACKED csi
= {};
5619 trace_pci_nvme_identify_ns_descr_list(nsid
);
5621 if (!nvme_nsid_valid(n
, nsid
) || nsid
== NVME_NSID_BROADCAST
) {
5622 return NVME_INVALID_NSID
| NVME_DNR
;
5625 ns
= nvme_ns(n
, nsid
);
5626 if (unlikely(!ns
)) {
5627 return NVME_INVALID_FIELD
| NVME_DNR
;
5630 if (!qemu_uuid_is_null(&ns
->params
.uuid
)) {
5631 uuid
.hdr
.nidt
= NVME_NIDT_UUID
;
5632 uuid
.hdr
.nidl
= NVME_NIDL_UUID
;
5633 memcpy(uuid
.v
, ns
->params
.uuid
.data
, NVME_NIDL_UUID
);
5634 memcpy(pos
, &uuid
, sizeof(uuid
));
5635 pos
+= sizeof(uuid
);
5638 if (ns
->params
.eui64
) {
5639 eui64
.hdr
.nidt
= NVME_NIDT_EUI64
;
5640 eui64
.hdr
.nidl
= NVME_NIDL_EUI64
;
5641 eui64
.v
= cpu_to_be64(ns
->params
.eui64
);
5642 memcpy(pos
, &eui64
, sizeof(eui64
));
5643 pos
+= sizeof(eui64
);
5646 csi
.hdr
.nidt
= NVME_NIDT_CSI
;
5647 csi
.hdr
.nidl
= NVME_NIDL_CSI
;
5649 memcpy(pos
, &csi
, sizeof(csi
));
5652 return nvme_c2h(n
, list
, sizeof(list
), req
);
5655 static uint16_t nvme_identify_cmd_set(NvmeCtrl
*n
, NvmeRequest
*req
)
5657 uint8_t list
[NVME_IDENTIFY_DATA_SIZE
] = {};
5658 static const int data_len
= sizeof(list
);
5660 trace_pci_nvme_identify_cmd_set();
5662 NVME_SET_CSI(*list
, NVME_CSI_NVM
);
5663 NVME_SET_CSI(*list
, NVME_CSI_ZONED
);
5665 return nvme_c2h(n
, list
, data_len
, req
);
5668 static uint16_t nvme_identify(NvmeCtrl
*n
, NvmeRequest
*req
)
5670 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5672 trace_pci_nvme_identify(nvme_cid(req
), c
->cns
, le16_to_cpu(c
->ctrlid
),
5676 case NVME_ID_CNS_NS
:
5677 return nvme_identify_ns(n
, req
, true);
5678 case NVME_ID_CNS_NS_PRESENT
:
5679 return nvme_identify_ns(n
, req
, false);
5680 case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST
:
5681 return nvme_identify_ctrl_list(n
, req
, true);
5682 case NVME_ID_CNS_CTRL_LIST
:
5683 return nvme_identify_ctrl_list(n
, req
, false);
5684 case NVME_ID_CNS_PRIMARY_CTRL_CAP
:
5685 return nvme_identify_pri_ctrl_cap(n
, req
);
5686 case NVME_ID_CNS_SECONDARY_CTRL_LIST
:
5687 return nvme_identify_sec_ctrl_list(n
, req
);
5688 case NVME_ID_CNS_CS_NS
:
5689 return nvme_identify_ns_csi(n
, req
, true);
5690 case NVME_ID_CNS_CS_NS_PRESENT
:
5691 return nvme_identify_ns_csi(n
, req
, false);
5692 case NVME_ID_CNS_CTRL
:
5693 return nvme_identify_ctrl(n
, req
);
5694 case NVME_ID_CNS_CS_CTRL
:
5695 return nvme_identify_ctrl_csi(n
, req
);
5696 case NVME_ID_CNS_NS_ACTIVE_LIST
:
5697 return nvme_identify_nslist(n
, req
, true);
5698 case NVME_ID_CNS_NS_PRESENT_LIST
:
5699 return nvme_identify_nslist(n
, req
, false);
5700 case NVME_ID_CNS_CS_NS_ACTIVE_LIST
:
5701 return nvme_identify_nslist_csi(n
, req
, true);
5702 case NVME_ID_CNS_CS_NS_PRESENT_LIST
:
5703 return nvme_identify_nslist_csi(n
, req
, false);
5704 case NVME_ID_CNS_NS_DESCR_LIST
:
5705 return nvme_identify_ns_descr_list(n
, req
);
5706 case NVME_ID_CNS_IO_COMMAND_SET
:
5707 return nvme_identify_cmd_set(n
, req
);
5709 trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c
->cns
));
5710 return NVME_INVALID_FIELD
| NVME_DNR
;
5714 static uint16_t nvme_abort(NvmeCtrl
*n
, NvmeRequest
*req
)
5716 uint16_t sqid
= le32_to_cpu(req
->cmd
.cdw10
) & 0xffff;
5718 req
->cqe
.result
= 1;
5719 if (nvme_check_sqid(n
, sqid
)) {
5720 return NVME_INVALID_FIELD
| NVME_DNR
;
5723 return NVME_SUCCESS
;
5726 static inline void nvme_set_timestamp(NvmeCtrl
*n
, uint64_t ts
)
5728 trace_pci_nvme_setfeat_timestamp(ts
);
5730 n
->host_timestamp
= le64_to_cpu(ts
);
5731 n
->timestamp_set_qemu_clock_ms
= qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
);
5734 static inline uint64_t nvme_get_timestamp(const NvmeCtrl
*n
)
5736 uint64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
);
5737 uint64_t elapsed_time
= current_time
- n
->timestamp_set_qemu_clock_ms
;
5739 union nvme_timestamp
{
5741 uint64_t timestamp
:48;
5749 union nvme_timestamp ts
;
5751 ts
.timestamp
= n
->host_timestamp
+ elapsed_time
;
5753 /* If the host timestamp is non-zero, set the timestamp origin */
5754 ts
.origin
= n
->host_timestamp
? 0x01 : 0x00;
5756 trace_pci_nvme_getfeat_timestamp(ts
.all
);
5758 return cpu_to_le64(ts
.all
);
5761 static uint16_t nvme_get_feature_timestamp(NvmeCtrl
*n
, NvmeRequest
*req
)
5763 uint64_t timestamp
= nvme_get_timestamp(n
);
5765 return nvme_c2h(n
, (uint8_t *)×tamp
, sizeof(timestamp
), req
);
5768 static int nvme_get_feature_fdp(NvmeCtrl
*n
, uint32_t endgrpid
,
5773 if (!n
->subsys
|| !n
->subsys
->endgrp
.fdp
.enabled
) {
5774 return NVME_INVALID_FIELD
| NVME_DNR
;
5777 *result
= FIELD_DP16(0, FEAT_FDP
, FDPE
, 1);
5778 *result
= FIELD_DP16(*result
, FEAT_FDP
, CONF_NDX
, 0);
5780 return NVME_SUCCESS
;
5783 static uint16_t nvme_get_feature_fdp_events(NvmeCtrl
*n
, NvmeNamespace
*ns
,
5784 NvmeRequest
*req
, uint32_t *result
)
5786 NvmeCmd
*cmd
= &req
->cmd
;
5787 uint32_t cdw11
= le32_to_cpu(cmd
->cdw11
);
5788 uint16_t ph
= cdw11
& 0xffff;
5789 uint8_t noet
= (cdw11
>> 16) & 0xff;
5790 uint16_t ruhid
, ret
;
5791 uint32_t nentries
= 0;
5792 uint8_t s_events_ndx
= 0;
5793 size_t s_events_siz
= sizeof(NvmeFdpEventDescr
) * noet
;
5794 g_autofree NvmeFdpEventDescr
*s_events
= g_malloc0(s_events_siz
);
5796 NvmeFdpEventDescr
*s_event
;
5798 if (!n
->subsys
|| !n
->subsys
->endgrp
.fdp
.enabled
) {
5799 return NVME_FDP_DISABLED
| NVME_DNR
;
5802 if (!nvme_ph_valid(ns
, ph
)) {
5803 return NVME_INVALID_FIELD
| NVME_DNR
;
5806 ruhid
= ns
->fdp
.phs
[ph
];
5807 ruh
= &n
->subsys
->endgrp
.fdp
.ruhs
[ruhid
];
5811 if (unlikely(noet
== 0)) {
5812 return NVME_INVALID_FIELD
| NVME_DNR
;
5815 for (uint8_t event_type
= 0; event_type
< FDP_EVT_MAX
; event_type
++) {
5816 uint8_t shift
= nvme_fdp_evf_shifts
[event_type
];
5817 if (!shift
&& event_type
) {
5819 * only first entry (event_type == 0) has a shift value of 0
5820 * other entries are simply unpopulated.
5827 s_event
= &s_events
[s_events_ndx
];
5828 s_event
->evt
= event_type
;
5829 s_event
->evta
= (ruh
->event_filter
>> shift
) & 0x1;
5831 /* break if all `noet` entries are filled */
5832 if ((++s_events_ndx
) == noet
) {
5837 ret
= nvme_c2h(n
, s_events
, s_events_siz
, req
);
5843 return NVME_SUCCESS
;
5846 static uint16_t nvme_get_feature(NvmeCtrl
*n
, NvmeRequest
*req
)
5848 NvmeCmd
*cmd
= &req
->cmd
;
5849 uint32_t dw10
= le32_to_cpu(cmd
->cdw10
);
5850 uint32_t dw11
= le32_to_cpu(cmd
->cdw11
);
5851 uint32_t nsid
= le32_to_cpu(cmd
->nsid
);
5853 uint8_t fid
= NVME_GETSETFEAT_FID(dw10
);
5854 NvmeGetFeatureSelect sel
= NVME_GETFEAT_SELECT(dw10
);
5858 uint16_t endgrpid
= 0, ret
= NVME_SUCCESS
;
5860 static const uint32_t nvme_feature_default
[NVME_FID_MAX
] = {
5861 [NVME_ARBITRATION
] = NVME_ARB_AB_NOLIMIT
,
5864 trace_pci_nvme_getfeat(nvme_cid(req
), nsid
, fid
, sel
, dw11
);
5866 if (!nvme_feature_support
[fid
]) {
5867 return NVME_INVALID_FIELD
| NVME_DNR
;
5870 if (nvme_feature_cap
[fid
] & NVME_FEAT_CAP_NS
) {
5871 if (!nvme_nsid_valid(n
, nsid
) || nsid
== NVME_NSID_BROADCAST
) {
5873 * The Reservation Notification Mask and Reservation Persistence
5874 * features require a status code of Invalid Field in Command when
5875 * NSID is FFFFFFFFh. Since the device does not support those
5876 * features we can always return Invalid Namespace or Format as we
5877 * should do for all other features.
5879 return NVME_INVALID_NSID
| NVME_DNR
;
5882 if (!nvme_ns(n
, nsid
)) {
5883 return NVME_INVALID_FIELD
| NVME_DNR
;
5888 case NVME_GETFEAT_SELECT_CURRENT
:
5890 case NVME_GETFEAT_SELECT_SAVED
:
5891 /* no features are saveable by the controller; fallthrough */
5892 case NVME_GETFEAT_SELECT_DEFAULT
:
5894 case NVME_GETFEAT_SELECT_CAP
:
5895 result
= nvme_feature_cap
[fid
];
5900 case NVME_TEMPERATURE_THRESHOLD
:
5904 * The controller only implements the Composite Temperature sensor, so
5905 * return 0 for all other sensors.
5907 if (NVME_TEMP_TMPSEL(dw11
) != NVME_TEMP_TMPSEL_COMPOSITE
) {
5911 switch (NVME_TEMP_THSEL(dw11
)) {
5912 case NVME_TEMP_THSEL_OVER
:
5913 result
= n
->features
.temp_thresh_hi
;
5915 case NVME_TEMP_THSEL_UNDER
:
5916 result
= n
->features
.temp_thresh_low
;
5920 return NVME_INVALID_FIELD
| NVME_DNR
;
5921 case NVME_ERROR_RECOVERY
:
5922 if (!nvme_nsid_valid(n
, nsid
)) {
5923 return NVME_INVALID_NSID
| NVME_DNR
;
5926 ns
= nvme_ns(n
, nsid
);
5927 if (unlikely(!ns
)) {
5928 return NVME_INVALID_FIELD
| NVME_DNR
;
5931 result
= ns
->features
.err_rec
;
5933 case NVME_VOLATILE_WRITE_CACHE
:
5935 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
5941 result
= blk_enable_write_cache(ns
->blkconf
.blk
);
5946 trace_pci_nvme_getfeat_vwcache(result
? "enabled" : "disabled");
5948 case NVME_ASYNCHRONOUS_EVENT_CONF
:
5949 result
= n
->features
.async_config
;
5951 case NVME_TIMESTAMP
:
5952 return nvme_get_feature_timestamp(n
, req
);
5953 case NVME_HOST_BEHAVIOR_SUPPORT
:
5954 return nvme_c2h(n
, (uint8_t *)&n
->features
.hbs
,
5955 sizeof(n
->features
.hbs
), req
);
5957 endgrpid
= dw11
& 0xff;
5959 if (endgrpid
!= 0x1) {
5960 return NVME_INVALID_FIELD
| NVME_DNR
;
5963 ret
= nvme_get_feature_fdp(n
, endgrpid
, &result
);
5968 case NVME_FDP_EVENTS
:
5969 if (!nvme_nsid_valid(n
, nsid
)) {
5970 return NVME_INVALID_NSID
| NVME_DNR
;
5973 ns
= nvme_ns(n
, nsid
);
5974 if (unlikely(!ns
)) {
5975 return NVME_INVALID_FIELD
| NVME_DNR
;
5978 ret
= nvme_get_feature_fdp_events(n
, ns
, req
, &result
);
5989 case NVME_TEMPERATURE_THRESHOLD
:
5992 if (NVME_TEMP_TMPSEL(dw11
) != NVME_TEMP_TMPSEL_COMPOSITE
) {
5996 if (NVME_TEMP_THSEL(dw11
) == NVME_TEMP_THSEL_OVER
) {
5997 result
= NVME_TEMPERATURE_WARNING
;
6001 case NVME_NUMBER_OF_QUEUES
:
6002 result
= (n
->conf_ioqpairs
- 1) | ((n
->conf_ioqpairs
- 1) << 16);
6003 trace_pci_nvme_getfeat_numq(result
);
6005 case NVME_INTERRUPT_VECTOR_CONF
:
6007 if (iv
>= n
->conf_ioqpairs
+ 1) {
6008 return NVME_INVALID_FIELD
| NVME_DNR
;
6012 if (iv
== n
->admin_cq
.vector
) {
6013 result
|= NVME_INTVC_NOCOALESCING
;
6017 endgrpid
= dw11
& 0xff;
6019 if (endgrpid
!= 0x1) {
6020 return NVME_INVALID_FIELD
| NVME_DNR
;
6023 ret
= nvme_get_feature_fdp(n
, endgrpid
, &result
);
6031 result
= nvme_feature_default
[fid
];
6036 req
->cqe
.result
= cpu_to_le32(result
);
6040 static uint16_t nvme_set_feature_timestamp(NvmeCtrl
*n
, NvmeRequest
*req
)
6045 ret
= nvme_h2c(n
, (uint8_t *)×tamp
, sizeof(timestamp
), req
);
6050 nvme_set_timestamp(n
, timestamp
);
6052 return NVME_SUCCESS
;
6055 static uint16_t nvme_set_feature_fdp_events(NvmeCtrl
*n
, NvmeNamespace
*ns
,
6058 NvmeCmd
*cmd
= &req
->cmd
;
6059 uint32_t cdw11
= le32_to_cpu(cmd
->cdw11
);
6060 uint16_t ph
= cdw11
& 0xffff;
6061 uint8_t noet
= (cdw11
>> 16) & 0xff;
6062 uint16_t ret
, ruhid
;
6063 uint8_t enable
= le32_to_cpu(cmd
->cdw12
) & 0x1;
6064 uint8_t event_mask
= 0;
6066 g_autofree
uint8_t *events
= g_malloc0(noet
);
6067 NvmeRuHandle
*ruh
= NULL
;
6071 if (!n
->subsys
|| !n
->subsys
->endgrp
.fdp
.enabled
) {
6072 return NVME_FDP_DISABLED
| NVME_DNR
;
6075 if (!nvme_ph_valid(ns
, ph
)) {
6076 return NVME_INVALID_FIELD
| NVME_DNR
;
6079 ruhid
= ns
->fdp
.phs
[ph
];
6080 ruh
= &n
->subsys
->endgrp
.fdp
.ruhs
[ruhid
];
6082 ret
= nvme_h2c(n
, events
, noet
, req
);
6087 for (i
= 0; i
< noet
; i
++) {
6088 event_mask
|= (1 << nvme_fdp_evf_shifts
[events
[i
]]);
6092 ruh
->event_filter
|= event_mask
;
6094 ruh
->event_filter
= ruh
->event_filter
& ~event_mask
;
6097 return NVME_SUCCESS
;
6100 static uint16_t nvme_set_feature(NvmeCtrl
*n
, NvmeRequest
*req
)
6102 NvmeNamespace
*ns
= NULL
;
6104 NvmeCmd
*cmd
= &req
->cmd
;
6105 uint32_t dw10
= le32_to_cpu(cmd
->cdw10
);
6106 uint32_t dw11
= le32_to_cpu(cmd
->cdw11
);
6107 uint32_t nsid
= le32_to_cpu(cmd
->nsid
);
6108 uint8_t fid
= NVME_GETSETFEAT_FID(dw10
);
6109 uint8_t save
= NVME_SETFEAT_SAVE(dw10
);
6113 trace_pci_nvme_setfeat(nvme_cid(req
), nsid
, fid
, save
, dw11
);
6115 if (save
&& !(nvme_feature_cap
[fid
] & NVME_FEAT_CAP_SAVE
)) {
6116 return NVME_FID_NOT_SAVEABLE
| NVME_DNR
;
6119 if (!nvme_feature_support
[fid
]) {
6120 return NVME_INVALID_FIELD
| NVME_DNR
;
6123 if (nvme_feature_cap
[fid
] & NVME_FEAT_CAP_NS
) {
6124 if (nsid
!= NVME_NSID_BROADCAST
) {
6125 if (!nvme_nsid_valid(n
, nsid
)) {
6126 return NVME_INVALID_NSID
| NVME_DNR
;
6129 ns
= nvme_ns(n
, nsid
);
6130 if (unlikely(!ns
)) {
6131 return NVME_INVALID_FIELD
| NVME_DNR
;
6134 } else if (nsid
&& nsid
!= NVME_NSID_BROADCAST
) {
6135 if (!nvme_nsid_valid(n
, nsid
)) {
6136 return NVME_INVALID_NSID
| NVME_DNR
;
6139 return NVME_FEAT_NOT_NS_SPEC
| NVME_DNR
;
6142 if (!(nvme_feature_cap
[fid
] & NVME_FEAT_CAP_CHANGE
)) {
6143 return NVME_FEAT_NOT_CHANGEABLE
| NVME_DNR
;
6147 case NVME_TEMPERATURE_THRESHOLD
:
6148 if (NVME_TEMP_TMPSEL(dw11
) != NVME_TEMP_TMPSEL_COMPOSITE
) {
6152 switch (NVME_TEMP_THSEL(dw11
)) {
6153 case NVME_TEMP_THSEL_OVER
:
6154 n
->features
.temp_thresh_hi
= NVME_TEMP_TMPTH(dw11
);
6156 case NVME_TEMP_THSEL_UNDER
:
6157 n
->features
.temp_thresh_low
= NVME_TEMP_TMPTH(dw11
);
6160 return NVME_INVALID_FIELD
| NVME_DNR
;
6163 if ((n
->temperature
>= n
->features
.temp_thresh_hi
) ||
6164 (n
->temperature
<= n
->features
.temp_thresh_low
)) {
6165 nvme_smart_event(n
, NVME_SMART_TEMPERATURE
);
6169 case NVME_ERROR_RECOVERY
:
6170 if (nsid
== NVME_NSID_BROADCAST
) {
6171 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
6178 if (NVME_ID_NS_NSFEAT_DULBE(ns
->id_ns
.nsfeat
)) {
6179 ns
->features
.err_rec
= dw11
;
6187 if (NVME_ID_NS_NSFEAT_DULBE(ns
->id_ns
.nsfeat
)) {
6188 ns
->features
.err_rec
= dw11
;
6191 case NVME_VOLATILE_WRITE_CACHE
:
6192 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
6198 if (!(dw11
& 0x1) && blk_enable_write_cache(ns
->blkconf
.blk
)) {
6199 blk_flush(ns
->blkconf
.blk
);
6202 blk_set_enable_write_cache(ns
->blkconf
.blk
, dw11
& 1);
6207 case NVME_NUMBER_OF_QUEUES
:
6208 if (n
->qs_created
) {
6209 return NVME_CMD_SEQ_ERROR
| NVME_DNR
;
6213 * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
6216 if ((dw11
& 0xffff) == 0xffff || ((dw11
>> 16) & 0xffff) == 0xffff) {
6217 return NVME_INVALID_FIELD
| NVME_DNR
;
6220 trace_pci_nvme_setfeat_numq((dw11
& 0xffff) + 1,
6221 ((dw11
>> 16) & 0xffff) + 1,
6224 req
->cqe
.result
= cpu_to_le32((n
->conf_ioqpairs
- 1) |
6225 ((n
->conf_ioqpairs
- 1) << 16));
6227 case NVME_ASYNCHRONOUS_EVENT_CONF
:
6228 n
->features
.async_config
= dw11
;
6230 case NVME_TIMESTAMP
:
6231 return nvme_set_feature_timestamp(n
, req
);
6232 case NVME_HOST_BEHAVIOR_SUPPORT
:
6233 status
= nvme_h2c(n
, (uint8_t *)&n
->features
.hbs
,
6234 sizeof(n
->features
.hbs
), req
);
6239 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
6246 ns
->id_ns
.nlbaf
= ns
->nlbaf
- 1;
6247 if (!n
->features
.hbs
.lbafee
) {
6248 ns
->id_ns
.nlbaf
= MIN(ns
->id_ns
.nlbaf
, 15);
6253 case NVME_COMMAND_SET_PROFILE
:
6255 trace_pci_nvme_err_invalid_iocsci(dw11
& 0x1ff);
6256 return NVME_CMD_SET_CMB_REJECTED
| NVME_DNR
;
6260 /* spec: abort with cmd seq err if there's one or more NS' in endgrp */
6261 return NVME_CMD_SEQ_ERROR
| NVME_DNR
;
6262 case NVME_FDP_EVENTS
:
6263 return nvme_set_feature_fdp_events(n
, ns
, req
);
6265 return NVME_FEAT_NOT_CHANGEABLE
| NVME_DNR
;
6267 return NVME_SUCCESS
;
6270 static uint16_t nvme_aer(NvmeCtrl
*n
, NvmeRequest
*req
)
6272 trace_pci_nvme_aer(nvme_cid(req
));
6274 if (n
->outstanding_aers
> n
->params
.aerl
) {
6275 trace_pci_nvme_aer_aerl_exceeded();
6276 return NVME_AER_LIMIT_EXCEEDED
;
6279 n
->aer_reqs
[n
->outstanding_aers
] = req
;
6280 n
->outstanding_aers
++;
6282 if (!QTAILQ_EMPTY(&n
->aer_queue
)) {
6283 nvme_process_aers(n
);
6286 return NVME_NO_COMPLETE
;
6289 static void nvme_update_dmrsl(NvmeCtrl
*n
)
6293 for (nsid
= 1; nsid
<= NVME_MAX_NAMESPACES
; nsid
++) {
6294 NvmeNamespace
*ns
= nvme_ns(n
, nsid
);
6299 n
->dmrsl
= MIN_NON_ZERO(n
->dmrsl
,
6300 BDRV_REQUEST_MAX_BYTES
/ nvme_l2b(ns
, 1));
6304 static void nvme_select_iocs_ns(NvmeCtrl
*n
, NvmeNamespace
*ns
)
6306 uint32_t cc
= ldl_le_p(&n
->bar
.cc
);
6308 ns
->iocs
= nvme_cse_iocs_none
;
6311 if (NVME_CC_CSS(cc
) != NVME_CC_CSS_ADMIN_ONLY
) {
6312 ns
->iocs
= nvme_cse_iocs_nvm
;
6315 case NVME_CSI_ZONED
:
6316 if (NVME_CC_CSS(cc
) == NVME_CC_CSS_CSI
) {
6317 ns
->iocs
= nvme_cse_iocs_zoned
;
6318 } else if (NVME_CC_CSS(cc
) == NVME_CC_CSS_NVM
) {
6319 ns
->iocs
= nvme_cse_iocs_nvm
;
6325 static uint16_t nvme_ns_attachment(NvmeCtrl
*n
, NvmeRequest
*req
)
6329 uint16_t list
[NVME_CONTROLLER_LIST_SIZE
] = {};
6330 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
6331 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6332 uint8_t sel
= dw10
& 0xf;
6333 uint16_t *nr_ids
= &list
[0];
6334 uint16_t *ids
= &list
[1];
6338 trace_pci_nvme_ns_attachment(nvme_cid(req
), dw10
& 0xf);
6340 if (!nvme_nsid_valid(n
, nsid
)) {
6341 return NVME_INVALID_NSID
| NVME_DNR
;
6344 ns
= nvme_subsys_ns(n
->subsys
, nsid
);
6346 return NVME_INVALID_FIELD
| NVME_DNR
;
6349 ret
= nvme_h2c(n
, (uint8_t *)list
, 4096, req
);
6355 return NVME_NS_CTRL_LIST_INVALID
| NVME_DNR
;
6358 *nr_ids
= MIN(*nr_ids
, NVME_CONTROLLER_LIST_SIZE
- 1);
6359 for (i
= 0; i
< *nr_ids
; i
++) {
6360 ctrl
= nvme_subsys_ctrl(n
->subsys
, ids
[i
]);
6362 return NVME_NS_CTRL_LIST_INVALID
| NVME_DNR
;
6366 case NVME_NS_ATTACHMENT_ATTACH
:
6367 if (nvme_ns(ctrl
, nsid
)) {
6368 return NVME_NS_ALREADY_ATTACHED
| NVME_DNR
;
6371 if (ns
->attached
&& !ns
->params
.shared
) {
6372 return NVME_NS_PRIVATE
| NVME_DNR
;
6375 nvme_attach_ns(ctrl
, ns
);
6376 nvme_select_iocs_ns(ctrl
, ns
);
6380 case NVME_NS_ATTACHMENT_DETACH
:
6381 if (!nvme_ns(ctrl
, nsid
)) {
6382 return NVME_NS_NOT_ATTACHED
| NVME_DNR
;
6385 ctrl
->namespaces
[nsid
] = NULL
;
6388 nvme_update_dmrsl(ctrl
);
6393 return NVME_INVALID_FIELD
| NVME_DNR
;
6397 * Add namespace id to the changed namespace id list for event clearing
6398 * via Get Log Page command.
6400 if (!test_and_set_bit(nsid
, ctrl
->changed_nsids
)) {
6401 nvme_enqueue_event(ctrl
, NVME_AER_TYPE_NOTICE
,
6402 NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED
,
6403 NVME_LOG_CHANGED_NSLIST
);
6407 return NVME_SUCCESS
;
6410 typedef struct NvmeFormatAIOCB
{
6427 static void nvme_format_cancel(BlockAIOCB
*aiocb
)
6429 NvmeFormatAIOCB
*iocb
= container_of(aiocb
, NvmeFormatAIOCB
, common
);
6431 iocb
->ret
= -ECANCELED
;
6434 blk_aio_cancel_async(iocb
->aiocb
);
6439 static const AIOCBInfo nvme_format_aiocb_info
= {
6440 .aiocb_size
= sizeof(NvmeFormatAIOCB
),
6441 .cancel_async
= nvme_format_cancel
,
6442 .get_aio_context
= nvme_get_aio_context
,
6445 static void nvme_format_set(NvmeNamespace
*ns
, uint8_t lbaf
, uint8_t mset
,
6446 uint8_t pi
, uint8_t pil
)
6448 uint8_t lbafl
= lbaf
& 0xf;
6449 uint8_t lbafu
= lbaf
>> 4;
6451 trace_pci_nvme_format_set(ns
->params
.nsid
, lbaf
, mset
, pi
, pil
);
6453 ns
->id_ns
.dps
= (pil
<< 3) | pi
;
6454 ns
->id_ns
.flbas
= (lbafu
<< 5) | (mset
<< 4) | lbafl
;
6456 nvme_ns_init_format(ns
);
6459 static void nvme_do_format(NvmeFormatAIOCB
*iocb
);
6461 static void nvme_format_ns_cb(void *opaque
, int ret
)
6463 NvmeFormatAIOCB
*iocb
= opaque
;
6464 NvmeNamespace
*ns
= iocb
->ns
;
6467 if (iocb
->ret
< 0) {
6469 } else if (ret
< 0) {
6476 if (iocb
->offset
< ns
->size
) {
6477 bytes
= MIN(BDRV_REQUEST_MAX_BYTES
, ns
->size
- iocb
->offset
);
6479 iocb
->aiocb
= blk_aio_pwrite_zeroes(ns
->blkconf
.blk
, iocb
->offset
,
6480 bytes
, BDRV_REQ_MAY_UNMAP
,
6481 nvme_format_ns_cb
, iocb
);
6483 iocb
->offset
+= bytes
;
6487 nvme_format_set(ns
, iocb
->lbaf
, iocb
->mset
, iocb
->pi
, iocb
->pil
);
6493 nvme_do_format(iocb
);
6496 static uint16_t nvme_format_check(NvmeNamespace
*ns
, uint8_t lbaf
, uint8_t pi
)
6498 if (ns
->params
.zoned
) {
6499 return NVME_INVALID_FORMAT
| NVME_DNR
;
6502 if (lbaf
> ns
->id_ns
.nlbaf
) {
6503 return NVME_INVALID_FORMAT
| NVME_DNR
;
6506 if (pi
&& (ns
->id_ns
.lbaf
[lbaf
].ms
< nvme_pi_tuple_size(ns
))) {
6507 return NVME_INVALID_FORMAT
| NVME_DNR
;
6510 if (pi
&& pi
> NVME_ID_NS_DPS_TYPE_3
) {
6511 return NVME_INVALID_FIELD
| NVME_DNR
;
6514 return NVME_SUCCESS
;
6517 static void nvme_do_format(NvmeFormatAIOCB
*iocb
)
6519 NvmeRequest
*req
= iocb
->req
;
6520 NvmeCtrl
*n
= nvme_ctrl(req
);
6521 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6522 uint8_t lbaf
= dw10
& 0xf;
6523 uint8_t pi
= (dw10
>> 5) & 0x7;
6527 if (iocb
->ret
< 0) {
6531 if (iocb
->broadcast
) {
6532 for (i
= iocb
->nsid
+ 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
6533 iocb
->ns
= nvme_ns(n
, i
);
6545 status
= nvme_format_check(iocb
->ns
, lbaf
, pi
);
6547 req
->status
= status
;
6551 iocb
->ns
->status
= NVME_FORMAT_IN_PROGRESS
;
6552 nvme_format_ns_cb(iocb
, 0);
6556 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
6557 qemu_aio_unref(iocb
);
6560 static uint16_t nvme_format(NvmeCtrl
*n
, NvmeRequest
*req
)
6562 NvmeFormatAIOCB
*iocb
;
6563 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
6564 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6565 uint8_t lbaf
= dw10
& 0xf;
6566 uint8_t mset
= (dw10
>> 4) & 0x1;
6567 uint8_t pi
= (dw10
>> 5) & 0x7;
6568 uint8_t pil
= (dw10
>> 8) & 0x1;
6569 uint8_t lbafu
= (dw10
>> 12) & 0x3;
6572 iocb
= qemu_aio_get(&nvme_format_aiocb_info
, NULL
, nvme_misc_cb
, req
);
6582 iocb
->broadcast
= (nsid
== NVME_NSID_BROADCAST
);
6585 if (n
->features
.hbs
.lbafee
) {
6586 iocb
->lbaf
|= lbafu
<< 4;
6589 if (!iocb
->broadcast
) {
6590 if (!nvme_nsid_valid(n
, nsid
)) {
6591 status
= NVME_INVALID_NSID
| NVME_DNR
;
6595 iocb
->ns
= nvme_ns(n
, nsid
);
6597 status
= NVME_INVALID_FIELD
| NVME_DNR
;
6602 req
->aiocb
= &iocb
->common
;
6603 nvme_do_format(iocb
);
6605 return NVME_NO_COMPLETE
;
6608 qemu_aio_unref(iocb
);
6613 static void nvme_get_virt_res_num(NvmeCtrl
*n
, uint8_t rt
, int *num_total
,
6614 int *num_prim
, int *num_sec
)
6616 *num_total
= le32_to_cpu(rt
?
6617 n
->pri_ctrl_cap
.vifrt
: n
->pri_ctrl_cap
.vqfrt
);
6618 *num_prim
= le16_to_cpu(rt
?
6619 n
->pri_ctrl_cap
.virfap
: n
->pri_ctrl_cap
.vqrfap
);
6620 *num_sec
= le16_to_cpu(rt
? n
->pri_ctrl_cap
.virfa
: n
->pri_ctrl_cap
.vqrfa
);
6623 static uint16_t nvme_assign_virt_res_to_prim(NvmeCtrl
*n
, NvmeRequest
*req
,
6624 uint16_t cntlid
, uint8_t rt
,
6627 int num_total
, num_prim
, num_sec
;
6629 if (cntlid
!= n
->cntlid
) {
6630 return NVME_INVALID_CTRL_ID
| NVME_DNR
;
6633 nvme_get_virt_res_num(n
, rt
, &num_total
, &num_prim
, &num_sec
);
6635 if (nr
> num_total
) {
6636 return NVME_INVALID_NUM_RESOURCES
| NVME_DNR
;
6639 if (nr
> num_total
- num_sec
) {
6640 return NVME_INVALID_RESOURCE_ID
| NVME_DNR
;
6644 n
->next_pri_ctrl_cap
.virfap
= cpu_to_le16(nr
);
6646 n
->next_pri_ctrl_cap
.vqrfap
= cpu_to_le16(nr
);
6649 req
->cqe
.result
= cpu_to_le32(nr
);
6653 static void nvme_update_virt_res(NvmeCtrl
*n
, NvmeSecCtrlEntry
*sctrl
,
6656 int prev_nr
, prev_total
;
6659 prev_nr
= le16_to_cpu(sctrl
->nvi
);
6660 prev_total
= le32_to_cpu(n
->pri_ctrl_cap
.virfa
);
6661 sctrl
->nvi
= cpu_to_le16(nr
);
6662 n
->pri_ctrl_cap
.virfa
= cpu_to_le32(prev_total
+ nr
- prev_nr
);
6664 prev_nr
= le16_to_cpu(sctrl
->nvq
);
6665 prev_total
= le32_to_cpu(n
->pri_ctrl_cap
.vqrfa
);
6666 sctrl
->nvq
= cpu_to_le16(nr
);
6667 n
->pri_ctrl_cap
.vqrfa
= cpu_to_le32(prev_total
+ nr
- prev_nr
);
6671 static uint16_t nvme_assign_virt_res_to_sec(NvmeCtrl
*n
, NvmeRequest
*req
,
6672 uint16_t cntlid
, uint8_t rt
, int nr
)
6674 int num_total
, num_prim
, num_sec
, num_free
, diff
, limit
;
6675 NvmeSecCtrlEntry
*sctrl
;
6677 sctrl
= nvme_sctrl_for_cntlid(n
, cntlid
);
6679 return NVME_INVALID_CTRL_ID
| NVME_DNR
;
6683 return NVME_INVALID_SEC_CTRL_STATE
| NVME_DNR
;
6686 limit
= le16_to_cpu(rt
? n
->pri_ctrl_cap
.vifrsm
: n
->pri_ctrl_cap
.vqfrsm
);
6688 return NVME_INVALID_NUM_RESOURCES
| NVME_DNR
;
6691 nvme_get_virt_res_num(n
, rt
, &num_total
, &num_prim
, &num_sec
);
6692 num_free
= num_total
- num_prim
- num_sec
;
6693 diff
= nr
- le16_to_cpu(rt
? sctrl
->nvi
: sctrl
->nvq
);
6695 if (diff
> num_free
) {
6696 return NVME_INVALID_RESOURCE_ID
| NVME_DNR
;
6699 nvme_update_virt_res(n
, sctrl
, rt
, nr
);
6700 req
->cqe
.result
= cpu_to_le32(nr
);
6705 static uint16_t nvme_virt_set_state(NvmeCtrl
*n
, uint16_t cntlid
, bool online
)
6707 PCIDevice
*pci
= PCI_DEVICE(n
);
6708 NvmeCtrl
*sn
= NULL
;
6709 NvmeSecCtrlEntry
*sctrl
;
6712 sctrl
= nvme_sctrl_for_cntlid(n
, cntlid
);
6714 return NVME_INVALID_CTRL_ID
| NVME_DNR
;
6717 if (!pci_is_vf(pci
)) {
6718 vf_index
= le16_to_cpu(sctrl
->vfn
) - 1;
6719 sn
= NVME(pcie_sriov_get_vf_at_index(pci
, vf_index
));
6723 if (!sctrl
->nvi
|| (le16_to_cpu(sctrl
->nvq
) < 2) || !sn
) {
6724 return NVME_INVALID_SEC_CTRL_STATE
| NVME_DNR
;
6729 nvme_ctrl_reset(sn
, NVME_RESET_FUNCTION
);
6732 nvme_update_virt_res(n
, sctrl
, NVME_VIRT_RES_INTERRUPT
, 0);
6733 nvme_update_virt_res(n
, sctrl
, NVME_VIRT_RES_QUEUE
, 0);
6738 nvme_ctrl_reset(sn
, NVME_RESET_FUNCTION
);
6743 return NVME_SUCCESS
;
6746 static uint16_t nvme_virt_mngmt(NvmeCtrl
*n
, NvmeRequest
*req
)
6748 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6749 uint32_t dw11
= le32_to_cpu(req
->cmd
.cdw11
);
6750 uint8_t act
= dw10
& 0xf;
6751 uint8_t rt
= (dw10
>> 8) & 0x7;
6752 uint16_t cntlid
= (dw10
>> 16) & 0xffff;
6753 int nr
= dw11
& 0xffff;
6755 trace_pci_nvme_virt_mngmt(nvme_cid(req
), act
, cntlid
, rt
? "VI" : "VQ", nr
);
6757 if (rt
!= NVME_VIRT_RES_QUEUE
&& rt
!= NVME_VIRT_RES_INTERRUPT
) {
6758 return NVME_INVALID_RESOURCE_ID
| NVME_DNR
;
6762 case NVME_VIRT_MNGMT_ACTION_SEC_ASSIGN
:
6763 return nvme_assign_virt_res_to_sec(n
, req
, cntlid
, rt
, nr
);
6764 case NVME_VIRT_MNGMT_ACTION_PRM_ALLOC
:
6765 return nvme_assign_virt_res_to_prim(n
, req
, cntlid
, rt
, nr
);
6766 case NVME_VIRT_MNGMT_ACTION_SEC_ONLINE
:
6767 return nvme_virt_set_state(n
, cntlid
, true);
6768 case NVME_VIRT_MNGMT_ACTION_SEC_OFFLINE
:
6769 return nvme_virt_set_state(n
, cntlid
, false);
6771 return NVME_INVALID_FIELD
| NVME_DNR
;
6775 static uint16_t nvme_dbbuf_config(NvmeCtrl
*n
, const NvmeRequest
*req
)
6777 PCIDevice
*pci
= PCI_DEVICE(n
);
6778 uint64_t dbs_addr
= le64_to_cpu(req
->cmd
.dptr
.prp1
);
6779 uint64_t eis_addr
= le64_to_cpu(req
->cmd
.dptr
.prp2
);
6782 /* Address should be page aligned */
6783 if (dbs_addr
& (n
->page_size
- 1) || eis_addr
& (n
->page_size
- 1)) {
6784 return NVME_INVALID_FIELD
| NVME_DNR
;
6787 /* Save shadow buffer base addr for use during queue creation */
6788 n
->dbbuf_dbs
= dbs_addr
;
6789 n
->dbbuf_eis
= eis_addr
;
6790 n
->dbbuf_enabled
= true;
6792 for (i
= 0; i
< n
->params
.max_ioqpairs
+ 1; i
++) {
6793 NvmeSQueue
*sq
= n
->sq
[i
];
6794 NvmeCQueue
*cq
= n
->cq
[i
];
6798 * CAP.DSTRD is 0, so offset of ith sq db_addr is (i<<3)
6799 * nvme_process_db() uses this hard-coded way to calculate
6800 * doorbell offsets. Be consistent with that here.
6802 sq
->db_addr
= dbs_addr
+ (i
<< 3);
6803 sq
->ei_addr
= eis_addr
+ (i
<< 3);
6804 pci_dma_write(pci
, sq
->db_addr
, &sq
->tail
, sizeof(sq
->tail
));
6806 if (n
->params
.ioeventfd
&& sq
->sqid
!= 0) {
6807 if (!nvme_init_sq_ioeventfd(sq
)) {
6808 sq
->ioeventfd_enabled
= true;
6814 /* CAP.DSTRD is 0, so offset of ith cq db_addr is (i<<3)+(1<<2) */
6815 cq
->db_addr
= dbs_addr
+ (i
<< 3) + (1 << 2);
6816 cq
->ei_addr
= eis_addr
+ (i
<< 3) + (1 << 2);
6817 pci_dma_write(pci
, cq
->db_addr
, &cq
->head
, sizeof(cq
->head
));
6819 if (n
->params
.ioeventfd
&& cq
->cqid
!= 0) {
6820 if (!nvme_init_cq_ioeventfd(cq
)) {
6821 cq
->ioeventfd_enabled
= true;
6827 trace_pci_nvme_dbbuf_config(dbs_addr
, eis_addr
);
6829 return NVME_SUCCESS
;
6832 static uint16_t nvme_directive_send(NvmeCtrl
*n
, NvmeRequest
*req
)
6834 return NVME_INVALID_FIELD
| NVME_DNR
;
6837 static uint16_t nvme_directive_receive(NvmeCtrl
*n
, NvmeRequest
*req
)
6840 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6841 uint32_t dw11
= le32_to_cpu(req
->cmd
.cdw11
);
6842 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
6843 uint8_t doper
, dtype
;
6844 uint32_t numd
, trans_len
;
6845 NvmeDirectiveIdentify id
= {
6846 .supported
= 1 << NVME_DIRECTIVE_IDENTIFY
,
6847 .enabled
= 1 << NVME_DIRECTIVE_IDENTIFY
,
6851 doper
= dw11
& 0xff;
6852 dtype
= (dw11
>> 8) & 0xff;
6854 trans_len
= MIN(sizeof(NvmeDirectiveIdentify
), numd
<< 2);
6856 if (nsid
== NVME_NSID_BROADCAST
|| dtype
!= NVME_DIRECTIVE_IDENTIFY
||
6857 doper
!= NVME_DIRECTIVE_RETURN_PARAMS
) {
6858 return NVME_INVALID_FIELD
| NVME_DNR
;
6861 ns
= nvme_ns(n
, nsid
);
6863 return NVME_INVALID_FIELD
| NVME_DNR
;
6867 case NVME_DIRECTIVE_IDENTIFY
:
6869 case NVME_DIRECTIVE_RETURN_PARAMS
:
6870 if (ns
->endgrp
->fdp
.enabled
) {
6871 id
.supported
|= 1 << NVME_DIRECTIVE_DATA_PLACEMENT
;
6872 id
.enabled
|= 1 << NVME_DIRECTIVE_DATA_PLACEMENT
;
6873 id
.persistent
|= 1 << NVME_DIRECTIVE_DATA_PLACEMENT
;
6876 return nvme_c2h(n
, (uint8_t *)&id
, trans_len
, req
);
6879 return NVME_INVALID_FIELD
| NVME_DNR
;
6883 return NVME_INVALID_FIELD
;
6887 static uint16_t nvme_admin_cmd(NvmeCtrl
*n
, NvmeRequest
*req
)
6889 trace_pci_nvme_admin_cmd(nvme_cid(req
), nvme_sqid(req
), req
->cmd
.opcode
,
6890 nvme_adm_opc_str(req
->cmd
.opcode
));
6892 if (!(nvme_cse_acs
[req
->cmd
.opcode
] & NVME_CMD_EFF_CSUPP
)) {
6893 trace_pci_nvme_err_invalid_admin_opc(req
->cmd
.opcode
);
6894 return NVME_INVALID_OPCODE
| NVME_DNR
;
6897 /* SGLs shall not be used for Admin commands in NVMe over PCIe */
6898 if (NVME_CMD_FLAGS_PSDT(req
->cmd
.flags
) != NVME_PSDT_PRP
) {
6899 return NVME_INVALID_FIELD
| NVME_DNR
;
6902 if (NVME_CMD_FLAGS_FUSE(req
->cmd
.flags
)) {
6903 return NVME_INVALID_FIELD
;
6906 switch (req
->cmd
.opcode
) {
6907 case NVME_ADM_CMD_DELETE_SQ
:
6908 return nvme_del_sq(n
, req
);
6909 case NVME_ADM_CMD_CREATE_SQ
:
6910 return nvme_create_sq(n
, req
);
6911 case NVME_ADM_CMD_GET_LOG_PAGE
:
6912 return nvme_get_log(n
, req
);
6913 case NVME_ADM_CMD_DELETE_CQ
:
6914 return nvme_del_cq(n
, req
);
6915 case NVME_ADM_CMD_CREATE_CQ
:
6916 return nvme_create_cq(n
, req
);
6917 case NVME_ADM_CMD_IDENTIFY
:
6918 return nvme_identify(n
, req
);
6919 case NVME_ADM_CMD_ABORT
:
6920 return nvme_abort(n
, req
);
6921 case NVME_ADM_CMD_SET_FEATURES
:
6922 return nvme_set_feature(n
, req
);
6923 case NVME_ADM_CMD_GET_FEATURES
:
6924 return nvme_get_feature(n
, req
);
6925 case NVME_ADM_CMD_ASYNC_EV_REQ
:
6926 return nvme_aer(n
, req
);
6927 case NVME_ADM_CMD_NS_ATTACHMENT
:
6928 return nvme_ns_attachment(n
, req
);
6929 case NVME_ADM_CMD_VIRT_MNGMT
:
6930 return nvme_virt_mngmt(n
, req
);
6931 case NVME_ADM_CMD_DBBUF_CONFIG
:
6932 return nvme_dbbuf_config(n
, req
);
6933 case NVME_ADM_CMD_FORMAT_NVM
:
6934 return nvme_format(n
, req
);
6935 case NVME_ADM_CMD_DIRECTIVE_SEND
:
6936 return nvme_directive_send(n
, req
);
6937 case NVME_ADM_CMD_DIRECTIVE_RECV
:
6938 return nvme_directive_receive(n
, req
);
6943 return NVME_INVALID_OPCODE
| NVME_DNR
;
6946 static void nvme_update_sq_eventidx(const NvmeSQueue
*sq
)
6948 uint32_t v
= cpu_to_le32(sq
->tail
);
6950 trace_pci_nvme_update_sq_eventidx(sq
->sqid
, sq
->tail
);
6952 pci_dma_write(PCI_DEVICE(sq
->ctrl
), sq
->ei_addr
, &v
, sizeof(v
));
6955 static void nvme_update_sq_tail(NvmeSQueue
*sq
)
6959 pci_dma_read(PCI_DEVICE(sq
->ctrl
), sq
->db_addr
, &v
, sizeof(v
));
6961 sq
->tail
= le32_to_cpu(v
);
6963 trace_pci_nvme_update_sq_tail(sq
->sqid
, sq
->tail
);
6966 static void nvme_process_sq(void *opaque
)
6968 NvmeSQueue
*sq
= opaque
;
6969 NvmeCtrl
*n
= sq
->ctrl
;
6970 NvmeCQueue
*cq
= n
->cq
[sq
->cqid
];
6977 if (n
->dbbuf_enabled
) {
6978 nvme_update_sq_tail(sq
);
6981 while (!(nvme_sq_empty(sq
) || QTAILQ_EMPTY(&sq
->req_list
))) {
6982 addr
= sq
->dma_addr
+ sq
->head
* n
->sqe_size
;
6983 if (nvme_addr_read(n
, addr
, (void *)&cmd
, sizeof(cmd
))) {
6984 trace_pci_nvme_err_addr_read(addr
);
6985 trace_pci_nvme_err_cfs();
6986 stl_le_p(&n
->bar
.csts
, NVME_CSTS_FAILED
);
6989 nvme_inc_sq_head(sq
);
6991 req
= QTAILQ_FIRST(&sq
->req_list
);
6992 QTAILQ_REMOVE(&sq
->req_list
, req
, entry
);
6993 QTAILQ_INSERT_TAIL(&sq
->out_req_list
, req
, entry
);
6994 nvme_req_clear(req
);
6995 req
->cqe
.cid
= cmd
.cid
;
6996 memcpy(&req
->cmd
, &cmd
, sizeof(NvmeCmd
));
6998 status
= sq
->sqid
? nvme_io_cmd(n
, req
) :
6999 nvme_admin_cmd(n
, req
);
7000 if (status
!= NVME_NO_COMPLETE
) {
7001 req
->status
= status
;
7002 nvme_enqueue_req_completion(cq
, req
);
7005 if (n
->dbbuf_enabled
) {
7006 nvme_update_sq_eventidx(sq
);
7007 nvme_update_sq_tail(sq
);
7012 static void nvme_update_msixcap_ts(PCIDevice
*pci_dev
, uint32_t table_size
)
7016 if (!msix_present(pci_dev
)) {
7020 assert(table_size
> 0 && table_size
<= pci_dev
->msix_entries_nr
);
7022 config
= pci_dev
->config
+ pci_dev
->msix_cap
;
7023 pci_set_word_by_mask(config
+ PCI_MSIX_FLAGS
, PCI_MSIX_FLAGS_QSIZE
,
7027 static void nvme_activate_virt_res(NvmeCtrl
*n
)
7029 PCIDevice
*pci_dev
= PCI_DEVICE(n
);
7030 NvmePriCtrlCap
*cap
= &n
->pri_ctrl_cap
;
7031 NvmeSecCtrlEntry
*sctrl
;
7033 /* -1 to account for the admin queue */
7034 if (pci_is_vf(pci_dev
)) {
7035 sctrl
= nvme_sctrl(n
);
7036 cap
->vqprt
= sctrl
->nvq
;
7037 cap
->viprt
= sctrl
->nvi
;
7038 n
->conf_ioqpairs
= sctrl
->nvq
? le16_to_cpu(sctrl
->nvq
) - 1 : 0;
7039 n
->conf_msix_qsize
= sctrl
->nvi
? le16_to_cpu(sctrl
->nvi
) : 1;
7041 cap
->vqrfap
= n
->next_pri_ctrl_cap
.vqrfap
;
7042 cap
->virfap
= n
->next_pri_ctrl_cap
.virfap
;
7043 n
->conf_ioqpairs
= le16_to_cpu(cap
->vqprt
) +
7044 le16_to_cpu(cap
->vqrfap
) - 1;
7045 n
->conf_msix_qsize
= le16_to_cpu(cap
->viprt
) +
7046 le16_to_cpu(cap
->virfap
);
7050 static void nvme_ctrl_reset(NvmeCtrl
*n
, NvmeResetType rst
)
7052 PCIDevice
*pci_dev
= PCI_DEVICE(n
);
7053 NvmeSecCtrlEntry
*sctrl
;
7057 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
7066 for (i
= 0; i
< n
->params
.max_ioqpairs
+ 1; i
++) {
7067 if (n
->sq
[i
] != NULL
) {
7068 nvme_free_sq(n
->sq
[i
], n
);
7071 for (i
= 0; i
< n
->params
.max_ioqpairs
+ 1; i
++) {
7072 if (n
->cq
[i
] != NULL
) {
7073 nvme_free_cq(n
->cq
[i
], n
);
7077 while (!QTAILQ_EMPTY(&n
->aer_queue
)) {
7078 NvmeAsyncEvent
*event
= QTAILQ_FIRST(&n
->aer_queue
);
7079 QTAILQ_REMOVE(&n
->aer_queue
, event
, entry
);
7083 if (n
->params
.sriov_max_vfs
) {
7084 if (!pci_is_vf(pci_dev
)) {
7085 for (i
= 0; i
< n
->sec_ctrl_list
.numcntl
; i
++) {
7086 sctrl
= &n
->sec_ctrl_list
.sec
[i
];
7087 nvme_virt_set_state(n
, le16_to_cpu(sctrl
->scid
), false);
7090 if (rst
!= NVME_RESET_CONTROLLER
) {
7091 pcie_sriov_pf_disable_vfs(pci_dev
);
7095 if (rst
!= NVME_RESET_CONTROLLER
) {
7096 nvme_activate_virt_res(n
);
7102 n
->outstanding_aers
= 0;
7103 n
->qs_created
= false;
7105 nvme_update_msixcap_ts(pci_dev
, n
->conf_msix_qsize
);
7107 if (pci_is_vf(pci_dev
)) {
7108 sctrl
= nvme_sctrl(n
);
7110 stl_le_p(&n
->bar
.csts
, sctrl
->scs
? 0 : NVME_CSTS_FAILED
);
7112 stl_le_p(&n
->bar
.csts
, 0);
7115 stl_le_p(&n
->bar
.intms
, 0);
7116 stl_le_p(&n
->bar
.intmc
, 0);
7117 stl_le_p(&n
->bar
.cc
, 0);
7121 n
->dbbuf_enabled
= false;
7124 static void nvme_ctrl_shutdown(NvmeCtrl
*n
)
7130 memory_region_msync(&n
->pmr
.dev
->mr
, 0, n
->pmr
.dev
->size
);
7133 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
7139 nvme_ns_shutdown(ns
);
7143 static void nvme_select_iocs(NvmeCtrl
*n
)
7148 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
7154 nvme_select_iocs_ns(n
, ns
);
7158 static int nvme_start_ctrl(NvmeCtrl
*n
)
7160 uint64_t cap
= ldq_le_p(&n
->bar
.cap
);
7161 uint32_t cc
= ldl_le_p(&n
->bar
.cc
);
7162 uint32_t aqa
= ldl_le_p(&n
->bar
.aqa
);
7163 uint64_t asq
= ldq_le_p(&n
->bar
.asq
);
7164 uint64_t acq
= ldq_le_p(&n
->bar
.acq
);
7165 uint32_t page_bits
= NVME_CC_MPS(cc
) + 12;
7166 uint32_t page_size
= 1 << page_bits
;
7167 NvmeSecCtrlEntry
*sctrl
= nvme_sctrl(n
);
7169 if (pci_is_vf(PCI_DEVICE(n
)) && !sctrl
->scs
) {
7170 trace_pci_nvme_err_startfail_virt_state(le16_to_cpu(sctrl
->nvi
),
7171 le16_to_cpu(sctrl
->nvq
));
7174 if (unlikely(n
->cq
[0])) {
7175 trace_pci_nvme_err_startfail_cq();
7178 if (unlikely(n
->sq
[0])) {
7179 trace_pci_nvme_err_startfail_sq();
7182 if (unlikely(asq
& (page_size
- 1))) {
7183 trace_pci_nvme_err_startfail_asq_misaligned(asq
);
7186 if (unlikely(acq
& (page_size
- 1))) {
7187 trace_pci_nvme_err_startfail_acq_misaligned(acq
);
7190 if (unlikely(!(NVME_CAP_CSS(cap
) & (1 << NVME_CC_CSS(cc
))))) {
7191 trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc
));
7194 if (unlikely(NVME_CC_MPS(cc
) < NVME_CAP_MPSMIN(cap
))) {
7195 trace_pci_nvme_err_startfail_page_too_small(
7197 NVME_CAP_MPSMIN(cap
));
7200 if (unlikely(NVME_CC_MPS(cc
) >
7201 NVME_CAP_MPSMAX(cap
))) {
7202 trace_pci_nvme_err_startfail_page_too_large(
7204 NVME_CAP_MPSMAX(cap
));
7207 if (unlikely(NVME_CC_IOCQES(cc
) <
7208 NVME_CTRL_CQES_MIN(n
->id_ctrl
.cqes
))) {
7209 trace_pci_nvme_err_startfail_cqent_too_small(
7211 NVME_CTRL_CQES_MIN(cap
));
7214 if (unlikely(NVME_CC_IOCQES(cc
) >
7215 NVME_CTRL_CQES_MAX(n
->id_ctrl
.cqes
))) {
7216 trace_pci_nvme_err_startfail_cqent_too_large(
7218 NVME_CTRL_CQES_MAX(cap
));
7221 if (unlikely(NVME_CC_IOSQES(cc
) <
7222 NVME_CTRL_SQES_MIN(n
->id_ctrl
.sqes
))) {
7223 trace_pci_nvme_err_startfail_sqent_too_small(
7225 NVME_CTRL_SQES_MIN(cap
));
7228 if (unlikely(NVME_CC_IOSQES(cc
) >
7229 NVME_CTRL_SQES_MAX(n
->id_ctrl
.sqes
))) {
7230 trace_pci_nvme_err_startfail_sqent_too_large(
7232 NVME_CTRL_SQES_MAX(cap
));
7235 if (unlikely(!NVME_AQA_ASQS(aqa
))) {
7236 trace_pci_nvme_err_startfail_asqent_sz_zero();
7239 if (unlikely(!NVME_AQA_ACQS(aqa
))) {
7240 trace_pci_nvme_err_startfail_acqent_sz_zero();
7244 n
->page_bits
= page_bits
;
7245 n
->page_size
= page_size
;
7246 n
->max_prp_ents
= n
->page_size
/ sizeof(uint64_t);
7247 n
->cqe_size
= 1 << NVME_CC_IOCQES(cc
);
7248 n
->sqe_size
= 1 << NVME_CC_IOSQES(cc
);
7249 nvme_init_cq(&n
->admin_cq
, n
, acq
, 0, 0, NVME_AQA_ACQS(aqa
) + 1, 1);
7250 nvme_init_sq(&n
->admin_sq
, n
, asq
, 0, 0, NVME_AQA_ASQS(aqa
) + 1);
7252 nvme_set_timestamp(n
, 0ULL);
7254 nvme_select_iocs(n
);
7259 static void nvme_cmb_enable_regs(NvmeCtrl
*n
)
7261 uint32_t cmbloc
= ldl_le_p(&n
->bar
.cmbloc
);
7262 uint32_t cmbsz
= ldl_le_p(&n
->bar
.cmbsz
);
7264 NVME_CMBLOC_SET_CDPCILS(cmbloc
, 1);
7265 NVME_CMBLOC_SET_CDPMLS(cmbloc
, 1);
7266 NVME_CMBLOC_SET_BIR(cmbloc
, NVME_CMB_BIR
);
7267 stl_le_p(&n
->bar
.cmbloc
, cmbloc
);
7269 NVME_CMBSZ_SET_SQS(cmbsz
, 1);
7270 NVME_CMBSZ_SET_CQS(cmbsz
, 0);
7271 NVME_CMBSZ_SET_LISTS(cmbsz
, 1);
7272 NVME_CMBSZ_SET_RDS(cmbsz
, 1);
7273 NVME_CMBSZ_SET_WDS(cmbsz
, 1);
7274 NVME_CMBSZ_SET_SZU(cmbsz
, 2); /* MBs */
7275 NVME_CMBSZ_SET_SZ(cmbsz
, n
->params
.cmb_size_mb
);
7276 stl_le_p(&n
->bar
.cmbsz
, cmbsz
);
7279 static void nvme_write_bar(NvmeCtrl
*n
, hwaddr offset
, uint64_t data
,
7282 PCIDevice
*pci
= PCI_DEVICE(n
);
7283 uint64_t cap
= ldq_le_p(&n
->bar
.cap
);
7284 uint32_t cc
= ldl_le_p(&n
->bar
.cc
);
7285 uint32_t intms
= ldl_le_p(&n
->bar
.intms
);
7286 uint32_t csts
= ldl_le_p(&n
->bar
.csts
);
7287 uint32_t pmrsts
= ldl_le_p(&n
->bar
.pmrsts
);
7289 if (unlikely(offset
& (sizeof(uint32_t) - 1))) {
7290 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32
,
7291 "MMIO write not 32-bit aligned,"
7292 " offset=0x%"PRIx64
"", offset
);
7293 /* should be ignored, fall through for now */
7296 if (unlikely(size
< sizeof(uint32_t))) {
7297 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall
,
7298 "MMIO write smaller than 32-bits,"
7299 " offset=0x%"PRIx64
", size=%u",
7301 /* should be ignored, fall through for now */
7305 case NVME_REG_INTMS
:
7306 if (unlikely(msix_enabled(pci
))) {
7307 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix
,
7308 "undefined access to interrupt mask set"
7309 " when MSI-X is enabled");
7310 /* should be ignored, fall through for now */
7313 stl_le_p(&n
->bar
.intms
, intms
);
7314 n
->bar
.intmc
= n
->bar
.intms
;
7315 trace_pci_nvme_mmio_intm_set(data
& 0xffffffff, intms
);
7318 case NVME_REG_INTMC
:
7319 if (unlikely(msix_enabled(pci
))) {
7320 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix
,
7321 "undefined access to interrupt mask clr"
7322 " when MSI-X is enabled");
7323 /* should be ignored, fall through for now */
7326 stl_le_p(&n
->bar
.intms
, intms
);
7327 n
->bar
.intmc
= n
->bar
.intms
;
7328 trace_pci_nvme_mmio_intm_clr(data
& 0xffffffff, intms
);
7332 stl_le_p(&n
->bar
.cc
, data
);
7334 trace_pci_nvme_mmio_cfg(data
& 0xffffffff);
7336 if (NVME_CC_SHN(data
) && !(NVME_CC_SHN(cc
))) {
7337 trace_pci_nvme_mmio_shutdown_set();
7338 nvme_ctrl_shutdown(n
);
7339 csts
&= ~(CSTS_SHST_MASK
<< CSTS_SHST_SHIFT
);
7340 csts
|= NVME_CSTS_SHST_COMPLETE
;
7341 } else if (!NVME_CC_SHN(data
) && NVME_CC_SHN(cc
)) {
7342 trace_pci_nvme_mmio_shutdown_cleared();
7343 csts
&= ~(CSTS_SHST_MASK
<< CSTS_SHST_SHIFT
);
7346 if (NVME_CC_EN(data
) && !NVME_CC_EN(cc
)) {
7347 if (unlikely(nvme_start_ctrl(n
))) {
7348 trace_pci_nvme_err_startfail();
7349 csts
= NVME_CSTS_FAILED
;
7351 trace_pci_nvme_mmio_start_success();
7352 csts
= NVME_CSTS_READY
;
7354 } else if (!NVME_CC_EN(data
) && NVME_CC_EN(cc
)) {
7355 trace_pci_nvme_mmio_stopped();
7356 nvme_ctrl_reset(n
, NVME_RESET_CONTROLLER
);
7361 stl_le_p(&n
->bar
.csts
, csts
);
7365 if (data
& (1 << 4)) {
7366 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported
,
7367 "attempted to W1C CSTS.NSSRO"
7368 " but CAP.NSSRS is zero (not supported)");
7369 } else if (data
!= 0) {
7370 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts
,
7371 "attempted to set a read only bit"
7372 " of controller status");
7376 if (data
== 0x4e564d65) {
7377 trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
7379 /* The spec says that writes of other values have no effect */
7384 stl_le_p(&n
->bar
.aqa
, data
);
7385 trace_pci_nvme_mmio_aqattr(data
& 0xffffffff);
7388 stn_le_p(&n
->bar
.asq
, size
, data
);
7389 trace_pci_nvme_mmio_asqaddr(data
);
7391 case NVME_REG_ASQ
+ 4:
7392 stl_le_p((uint8_t *)&n
->bar
.asq
+ 4, data
);
7393 trace_pci_nvme_mmio_asqaddr_hi(data
, ldq_le_p(&n
->bar
.asq
));
7396 trace_pci_nvme_mmio_acqaddr(data
);
7397 stn_le_p(&n
->bar
.acq
, size
, data
);
7399 case NVME_REG_ACQ
+ 4:
7400 stl_le_p((uint8_t *)&n
->bar
.acq
+ 4, data
);
7401 trace_pci_nvme_mmio_acqaddr_hi(data
, ldq_le_p(&n
->bar
.acq
));
7403 case NVME_REG_CMBLOC
:
7404 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved
,
7405 "invalid write to reserved CMBLOC"
7406 " when CMBSZ is zero, ignored");
7408 case NVME_REG_CMBSZ
:
7409 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly
,
7410 "invalid write to read only CMBSZ, ignored");
7412 case NVME_REG_CMBMSC
:
7413 if (!NVME_CAP_CMBS(cap
)) {
7417 stn_le_p(&n
->bar
.cmbmsc
, size
, data
);
7418 n
->cmb
.cmse
= false;
7420 if (NVME_CMBMSC_CRE(data
)) {
7421 nvme_cmb_enable_regs(n
);
7423 if (NVME_CMBMSC_CMSE(data
)) {
7424 uint64_t cmbmsc
= ldq_le_p(&n
->bar
.cmbmsc
);
7425 hwaddr cba
= NVME_CMBMSC_CBA(cmbmsc
) << CMBMSC_CBA_SHIFT
;
7426 if (cba
+ int128_get64(n
->cmb
.mem
.size
) < cba
) {
7427 uint32_t cmbsts
= ldl_le_p(&n
->bar
.cmbsts
);
7428 NVME_CMBSTS_SET_CBAI(cmbsts
, 1);
7429 stl_le_p(&n
->bar
.cmbsts
, cmbsts
);
7442 case NVME_REG_CMBMSC
+ 4:
7443 stl_le_p((uint8_t *)&n
->bar
.cmbmsc
+ 4, data
);
7446 case NVME_REG_PMRCAP
:
7447 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly
,
7448 "invalid write to PMRCAP register, ignored");
7450 case NVME_REG_PMRCTL
:
7451 if (!NVME_CAP_PMRS(cap
)) {
7455 stl_le_p(&n
->bar
.pmrctl
, data
);
7456 if (NVME_PMRCTL_EN(data
)) {
7457 memory_region_set_enabled(&n
->pmr
.dev
->mr
, true);
7460 memory_region_set_enabled(&n
->pmr
.dev
->mr
, false);
7461 NVME_PMRSTS_SET_NRDY(pmrsts
, 1);
7462 n
->pmr
.cmse
= false;
7464 stl_le_p(&n
->bar
.pmrsts
, pmrsts
);
7466 case NVME_REG_PMRSTS
:
7467 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly
,
7468 "invalid write to PMRSTS register, ignored");
7470 case NVME_REG_PMREBS
:
7471 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly
,
7472 "invalid write to PMREBS register, ignored");
7474 case NVME_REG_PMRSWTP
:
7475 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly
,
7476 "invalid write to PMRSWTP register, ignored");
7478 case NVME_REG_PMRMSCL
:
7479 if (!NVME_CAP_PMRS(cap
)) {
7483 stl_le_p(&n
->bar
.pmrmscl
, data
);
7484 n
->pmr
.cmse
= false;
7486 if (NVME_PMRMSCL_CMSE(data
)) {
7487 uint64_t pmrmscu
= ldl_le_p(&n
->bar
.pmrmscu
);
7488 hwaddr cba
= pmrmscu
<< 32 |
7489 (NVME_PMRMSCL_CBA(data
) << PMRMSCL_CBA_SHIFT
);
7490 if (cba
+ int128_get64(n
->pmr
.dev
->mr
.size
) < cba
) {
7491 NVME_PMRSTS_SET_CBAI(pmrsts
, 1);
7492 stl_le_p(&n
->bar
.pmrsts
, pmrsts
);
7501 case NVME_REG_PMRMSCU
:
7502 if (!NVME_CAP_PMRS(cap
)) {
7506 stl_le_p(&n
->bar
.pmrmscu
, data
);
7509 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid
,
7510 "invalid MMIO write,"
7511 " offset=0x%"PRIx64
", data=%"PRIx64
"",
7517 static uint64_t nvme_mmio_read(void *opaque
, hwaddr addr
, unsigned size
)
7519 NvmeCtrl
*n
= (NvmeCtrl
*)opaque
;
7520 uint8_t *ptr
= (uint8_t *)&n
->bar
;
7522 trace_pci_nvme_mmio_read(addr
, size
);
7524 if (unlikely(addr
& (sizeof(uint32_t) - 1))) {
7525 NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32
,
7526 "MMIO read not 32-bit aligned,"
7527 " offset=0x%"PRIx64
"", addr
);
7528 /* should RAZ, fall through for now */
7529 } else if (unlikely(size
< sizeof(uint32_t))) {
7530 NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall
,
7531 "MMIO read smaller than 32-bits,"
7532 " offset=0x%"PRIx64
"", addr
);
7533 /* should RAZ, fall through for now */
7536 if (addr
> sizeof(n
->bar
) - size
) {
7537 NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs
,
7538 "MMIO read beyond last register,"
7539 " offset=0x%"PRIx64
", returning 0", addr
);
7544 if (pci_is_vf(PCI_DEVICE(n
)) && !nvme_sctrl(n
)->scs
&&
7545 addr
!= NVME_REG_CSTS
) {
7546 trace_pci_nvme_err_ignored_mmio_vf_offline(addr
, size
);
7551 * When PMRWBM bit 1 is set then read from
7552 * from PMRSTS should ensure prior writes
7553 * made it to persistent media
7555 if (addr
== NVME_REG_PMRSTS
&&
7556 (NVME_PMRCAP_PMRWBM(ldl_le_p(&n
->bar
.pmrcap
)) & 0x02)) {
7557 memory_region_msync(&n
->pmr
.dev
->mr
, 0, n
->pmr
.dev
->size
);
7560 return ldn_le_p(ptr
+ addr
, size
);
7563 static void nvme_process_db(NvmeCtrl
*n
, hwaddr addr
, int val
)
7565 PCIDevice
*pci
= PCI_DEVICE(n
);
7568 if (unlikely(addr
& ((1 << 2) - 1))) {
7569 NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned
,
7570 "doorbell write not 32-bit aligned,"
7571 " offset=0x%"PRIx64
", ignoring", addr
);
7575 if (((addr
- 0x1000) >> 2) & 1) {
7576 /* Completion queue doorbell write */
7578 uint16_t new_head
= val
& 0xffff;
7582 qid
= (addr
- (0x1000 + (1 << 2))) >> 3;
7583 if (unlikely(nvme_check_cqid(n
, qid
))) {
7584 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq
,
7585 "completion queue doorbell write"
7586 " for nonexistent queue,"
7587 " sqid=%"PRIu32
", ignoring", qid
);
7590 * NVM Express v1.3d, Section 4.1 state: "If host software writes
7591 * an invalid value to the Submission Queue Tail Doorbell or
7592 * Completion Queue Head Doorbell regiter and an Asynchronous Event
7593 * Request command is outstanding, then an asynchronous event is
7594 * posted to the Admin Completion Queue with a status code of
7595 * Invalid Doorbell Write Value."
7597 * Also note that the spec includes the "Invalid Doorbell Register"
7598 * status code, but nowhere does it specify when to use it.
7599 * However, it seems reasonable to use it here in a similar
7602 if (n
->outstanding_aers
) {
7603 nvme_enqueue_event(n
, NVME_AER_TYPE_ERROR
,
7604 NVME_AER_INFO_ERR_INVALID_DB_REGISTER
,
7605 NVME_LOG_ERROR_INFO
);
7612 if (unlikely(new_head
>= cq
->size
)) {
7613 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead
,
7614 "completion queue doorbell write value"
7615 " beyond queue size, sqid=%"PRIu32
","
7616 " new_head=%"PRIu16
", ignoring",
7619 if (n
->outstanding_aers
) {
7620 nvme_enqueue_event(n
, NVME_AER_TYPE_ERROR
,
7621 NVME_AER_INFO_ERR_INVALID_DB_VALUE
,
7622 NVME_LOG_ERROR_INFO
);
7628 trace_pci_nvme_mmio_doorbell_cq(cq
->cqid
, new_head
);
7630 start_sqs
= nvme_cq_full(cq
) ? 1 : 0;
7631 cq
->head
= new_head
;
7632 if (!qid
&& n
->dbbuf_enabled
) {
7633 pci_dma_write(pci
, cq
->db_addr
, &cq
->head
, sizeof(cq
->head
));
7637 QTAILQ_FOREACH(sq
, &cq
->sq_list
, entry
) {
7638 qemu_bh_schedule(sq
->bh
);
7640 qemu_bh_schedule(cq
->bh
);
7643 if (cq
->tail
== cq
->head
) {
7644 if (cq
->irq_enabled
) {
7648 nvme_irq_deassert(n
, cq
);
7651 /* Submission queue doorbell write */
7653 uint16_t new_tail
= val
& 0xffff;
7656 qid
= (addr
- 0x1000) >> 3;
7657 if (unlikely(nvme_check_sqid(n
, qid
))) {
7658 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq
,
7659 "submission queue doorbell write"
7660 " for nonexistent queue,"
7661 " sqid=%"PRIu32
", ignoring", qid
);
7663 if (n
->outstanding_aers
) {
7664 nvme_enqueue_event(n
, NVME_AER_TYPE_ERROR
,
7665 NVME_AER_INFO_ERR_INVALID_DB_REGISTER
,
7666 NVME_LOG_ERROR_INFO
);
7673 if (unlikely(new_tail
>= sq
->size
)) {
7674 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail
,
7675 "submission queue doorbell write value"
7676 " beyond queue size, sqid=%"PRIu32
","
7677 " new_tail=%"PRIu16
", ignoring",
7680 if (n
->outstanding_aers
) {
7681 nvme_enqueue_event(n
, NVME_AER_TYPE_ERROR
,
7682 NVME_AER_INFO_ERR_INVALID_DB_VALUE
,
7683 NVME_LOG_ERROR_INFO
);
7689 trace_pci_nvme_mmio_doorbell_sq(sq
->sqid
, new_tail
);
7691 sq
->tail
= new_tail
;
7692 if (!qid
&& n
->dbbuf_enabled
) {
7694 * The spec states "the host shall also update the controller's
7695 * corresponding doorbell property to match the value of that entry
7696 * in the Shadow Doorbell buffer."
7698 * Since this context is currently a VM trap, we can safely enforce
7699 * the requirement from the device side in case the host is
7702 * Note, we shouldn't have to do this, but various drivers
7703 * including ones that run on Linux, are not updating Admin Queues,
7704 * so we can't trust reading it for an appropriate sq tail.
7706 pci_dma_write(pci
, sq
->db_addr
, &sq
->tail
, sizeof(sq
->tail
));
7709 qemu_bh_schedule(sq
->bh
);
7713 static void nvme_mmio_write(void *opaque
, hwaddr addr
, uint64_t data
,
7716 NvmeCtrl
*n
= (NvmeCtrl
*)opaque
;
7718 trace_pci_nvme_mmio_write(addr
, data
, size
);
7720 if (pci_is_vf(PCI_DEVICE(n
)) && !nvme_sctrl(n
)->scs
&&
7721 addr
!= NVME_REG_CSTS
) {
7722 trace_pci_nvme_err_ignored_mmio_vf_offline(addr
, size
);
7726 if (addr
< sizeof(n
->bar
)) {
7727 nvme_write_bar(n
, addr
, data
, size
);
7729 nvme_process_db(n
, addr
, data
);
7733 static const MemoryRegionOps nvme_mmio_ops
= {
7734 .read
= nvme_mmio_read
,
7735 .write
= nvme_mmio_write
,
7736 .endianness
= DEVICE_LITTLE_ENDIAN
,
7738 .min_access_size
= 2,
7739 .max_access_size
= 8,
7743 static void nvme_cmb_write(void *opaque
, hwaddr addr
, uint64_t data
,
7746 NvmeCtrl
*n
= (NvmeCtrl
*)opaque
;
7747 stn_le_p(&n
->cmb
.buf
[addr
], size
, data
);
7750 static uint64_t nvme_cmb_read(void *opaque
, hwaddr addr
, unsigned size
)
7752 NvmeCtrl
*n
= (NvmeCtrl
*)opaque
;
7753 return ldn_le_p(&n
->cmb
.buf
[addr
], size
);
7756 static const MemoryRegionOps nvme_cmb_ops
= {
7757 .read
= nvme_cmb_read
,
7758 .write
= nvme_cmb_write
,
7759 .endianness
= DEVICE_LITTLE_ENDIAN
,
7761 .min_access_size
= 1,
7762 .max_access_size
= 8,
7766 static bool nvme_check_params(NvmeCtrl
*n
, Error
**errp
)
7768 NvmeParams
*params
= &n
->params
;
7770 if (params
->num_queues
) {
7771 warn_report("num_queues is deprecated; please use max_ioqpairs "
7774 params
->max_ioqpairs
= params
->num_queues
- 1;
7777 if (n
->namespace.blkconf
.blk
&& n
->subsys
) {
7778 error_setg(errp
, "subsystem support is unavailable with legacy "
7779 "namespace ('drive' property)");
7783 if (params
->max_ioqpairs
< 1 ||
7784 params
->max_ioqpairs
> NVME_MAX_IOQPAIRS
) {
7785 error_setg(errp
, "max_ioqpairs must be between 1 and %d",
7790 if (params
->msix_qsize
< 1 ||
7791 params
->msix_qsize
> PCI_MSIX_FLAGS_QSIZE
+ 1) {
7792 error_setg(errp
, "msix_qsize must be between 1 and %d",
7793 PCI_MSIX_FLAGS_QSIZE
+ 1);
7797 if (!params
->serial
) {
7798 error_setg(errp
, "serial property not set");
7803 if (host_memory_backend_is_mapped(n
->pmr
.dev
)) {
7804 error_setg(errp
, "can't use already busy memdev: %s",
7805 object_get_canonical_path_component(OBJECT(n
->pmr
.dev
)));
7809 if (!is_power_of_2(n
->pmr
.dev
->size
)) {
7810 error_setg(errp
, "pmr backend size needs to be power of 2 in size");
7814 host_memory_backend_set_mapped(n
->pmr
.dev
, true);
7817 if (n
->params
.zasl
> n
->params
.mdts
) {
7818 error_setg(errp
, "zoned.zasl (Zone Append Size Limit) must be less "
7819 "than or equal to mdts (Maximum Data Transfer Size)");
7823 if (!n
->params
.vsl
) {
7824 error_setg(errp
, "vsl must be non-zero");
7828 if (params
->sriov_max_vfs
) {
7830 error_setg(errp
, "subsystem is required for the use of SR-IOV");
7834 if (params
->sriov_max_vfs
> NVME_MAX_VFS
) {
7835 error_setg(errp
, "sriov_max_vfs must be between 0 and %d",
7840 if (params
->cmb_size_mb
) {
7841 error_setg(errp
, "CMB is not supported with SR-IOV");
7846 error_setg(errp
, "PMR is not supported with SR-IOV");
7850 if (!params
->sriov_vq_flexible
|| !params
->sriov_vi_flexible
) {
7851 error_setg(errp
, "both sriov_vq_flexible and sriov_vi_flexible"
7852 " must be set for the use of SR-IOV");
7856 if (params
->sriov_vq_flexible
< params
->sriov_max_vfs
* 2) {
7857 error_setg(errp
, "sriov_vq_flexible must be greater than or equal"
7858 " to %d (sriov_max_vfs * 2)", params
->sriov_max_vfs
* 2);
7862 if (params
->max_ioqpairs
< params
->sriov_vq_flexible
+ 2) {
7863 error_setg(errp
, "(max_ioqpairs - sriov_vq_flexible) must be"
7864 " greater than or equal to 2");
7868 if (params
->sriov_vi_flexible
< params
->sriov_max_vfs
) {
7869 error_setg(errp
, "sriov_vi_flexible must be greater than or equal"
7870 " to %d (sriov_max_vfs)", params
->sriov_max_vfs
);
7874 if (params
->msix_qsize
< params
->sriov_vi_flexible
+ 1) {
7875 error_setg(errp
, "(msix_qsize - sriov_vi_flexible) must be"
7876 " greater than or equal to 1");
7880 if (params
->sriov_max_vi_per_vf
&&
7881 (params
->sriov_max_vi_per_vf
- 1) % NVME_VF_RES_GRANULARITY
) {
7882 error_setg(errp
, "sriov_max_vi_per_vf must meet:"
7883 " (sriov_max_vi_per_vf - 1) %% %d == 0 and"
7884 " sriov_max_vi_per_vf >= 1", NVME_VF_RES_GRANULARITY
);
7888 if (params
->sriov_max_vq_per_vf
&&
7889 (params
->sriov_max_vq_per_vf
< 2 ||
7890 (params
->sriov_max_vq_per_vf
- 1) % NVME_VF_RES_GRANULARITY
)) {
7891 error_setg(errp
, "sriov_max_vq_per_vf must meet:"
7892 " (sriov_max_vq_per_vf - 1) %% %d == 0 and"
7893 " sriov_max_vq_per_vf >= 2", NVME_VF_RES_GRANULARITY
);
7901 static void nvme_init_state(NvmeCtrl
*n
)
7903 NvmePriCtrlCap
*cap
= &n
->pri_ctrl_cap
;
7904 NvmeSecCtrlList
*list
= &n
->sec_ctrl_list
;
7905 NvmeSecCtrlEntry
*sctrl
;
7906 PCIDevice
*pci
= PCI_DEVICE(n
);
7910 if (pci_is_vf(pci
)) {
7911 sctrl
= nvme_sctrl(n
);
7913 n
->conf_ioqpairs
= sctrl
->nvq
? le16_to_cpu(sctrl
->nvq
) - 1 : 0;
7914 n
->conf_msix_qsize
= sctrl
->nvi
? le16_to_cpu(sctrl
->nvi
) : 1;
7916 max_vfs
= n
->params
.sriov_max_vfs
;
7917 n
->conf_ioqpairs
= n
->params
.max_ioqpairs
;
7918 n
->conf_msix_qsize
= n
->params
.msix_qsize
;
7921 n
->sq
= g_new0(NvmeSQueue
*, n
->params
.max_ioqpairs
+ 1);
7922 n
->cq
= g_new0(NvmeCQueue
*, n
->params
.max_ioqpairs
+ 1);
7923 n
->temperature
= NVME_TEMPERATURE
;
7924 n
->features
.temp_thresh_hi
= NVME_TEMPERATURE_WARNING
;
7925 n
->starttime_ms
= qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
);
7926 n
->aer_reqs
= g_new0(NvmeRequest
*, n
->params
.aerl
+ 1);
7927 QTAILQ_INIT(&n
->aer_queue
);
7929 list
->numcntl
= cpu_to_le16(max_vfs
);
7930 for (i
= 0; i
< max_vfs
; i
++) {
7931 sctrl
= &list
->sec
[i
];
7932 sctrl
->pcid
= cpu_to_le16(n
->cntlid
);
7933 sctrl
->vfn
= cpu_to_le16(i
+ 1);
7936 cap
->cntlid
= cpu_to_le16(n
->cntlid
);
7937 cap
->crt
= NVME_CRT_VQ
| NVME_CRT_VI
;
7939 if (pci_is_vf(pci
)) {
7940 cap
->vqprt
= cpu_to_le16(1 + n
->conf_ioqpairs
);
7942 cap
->vqprt
= cpu_to_le16(1 + n
->params
.max_ioqpairs
-
7943 n
->params
.sriov_vq_flexible
);
7944 cap
->vqfrt
= cpu_to_le32(n
->params
.sriov_vq_flexible
);
7945 cap
->vqrfap
= cap
->vqfrt
;
7946 cap
->vqgran
= cpu_to_le16(NVME_VF_RES_GRANULARITY
);
7947 cap
->vqfrsm
= n
->params
.sriov_max_vq_per_vf
?
7948 cpu_to_le16(n
->params
.sriov_max_vq_per_vf
) :
7949 cap
->vqfrt
/ MAX(max_vfs
, 1);
7952 if (pci_is_vf(pci
)) {
7953 cap
->viprt
= cpu_to_le16(n
->conf_msix_qsize
);
7955 cap
->viprt
= cpu_to_le16(n
->params
.msix_qsize
-
7956 n
->params
.sriov_vi_flexible
);
7957 cap
->vifrt
= cpu_to_le32(n
->params
.sriov_vi_flexible
);
7958 cap
->virfap
= cap
->vifrt
;
7959 cap
->vigran
= cpu_to_le16(NVME_VF_RES_GRANULARITY
);
7960 cap
->vifrsm
= n
->params
.sriov_max_vi_per_vf
?
7961 cpu_to_le16(n
->params
.sriov_max_vi_per_vf
) :
7962 cap
->vifrt
/ MAX(max_vfs
, 1);
7966 static void nvme_init_cmb(NvmeCtrl
*n
, PCIDevice
*pci_dev
)
7968 uint64_t cmb_size
= n
->params
.cmb_size_mb
* MiB
;
7969 uint64_t cap
= ldq_le_p(&n
->bar
.cap
);
7971 n
->cmb
.buf
= g_malloc0(cmb_size
);
7972 memory_region_init_io(&n
->cmb
.mem
, OBJECT(n
), &nvme_cmb_ops
, n
,
7973 "nvme-cmb", cmb_size
);
7974 pci_register_bar(pci_dev
, NVME_CMB_BIR
,
7975 PCI_BASE_ADDRESS_SPACE_MEMORY
|
7976 PCI_BASE_ADDRESS_MEM_TYPE_64
|
7977 PCI_BASE_ADDRESS_MEM_PREFETCH
, &n
->cmb
.mem
);
7979 NVME_CAP_SET_CMBS(cap
, 1);
7980 stq_le_p(&n
->bar
.cap
, cap
);
7982 if (n
->params
.legacy_cmb
) {
7983 nvme_cmb_enable_regs(n
);
7988 static void nvme_init_pmr(NvmeCtrl
*n
, PCIDevice
*pci_dev
)
7990 uint32_t pmrcap
= ldl_le_p(&n
->bar
.pmrcap
);
7992 NVME_PMRCAP_SET_RDS(pmrcap
, 1);
7993 NVME_PMRCAP_SET_WDS(pmrcap
, 1);
7994 NVME_PMRCAP_SET_BIR(pmrcap
, NVME_PMR_BIR
);
7995 /* Turn on bit 1 support */
7996 NVME_PMRCAP_SET_PMRWBM(pmrcap
, 0x02);
7997 NVME_PMRCAP_SET_CMSS(pmrcap
, 1);
7998 stl_le_p(&n
->bar
.pmrcap
, pmrcap
);
8000 pci_register_bar(pci_dev
, NVME_PMR_BIR
,
8001 PCI_BASE_ADDRESS_SPACE_MEMORY
|
8002 PCI_BASE_ADDRESS_MEM_TYPE_64
|
8003 PCI_BASE_ADDRESS_MEM_PREFETCH
, &n
->pmr
.dev
->mr
);
8005 memory_region_set_enabled(&n
->pmr
.dev
->mr
, false);
8008 static uint64_t nvme_bar_size(unsigned total_queues
, unsigned total_irqs
,
8009 unsigned *msix_table_offset
,
8010 unsigned *msix_pba_offset
)
8012 uint64_t bar_size
, msix_table_size
, msix_pba_size
;
8014 bar_size
= sizeof(NvmeBar
) + 2 * total_queues
* NVME_DB_SIZE
;
8015 bar_size
= QEMU_ALIGN_UP(bar_size
, 4 * KiB
);
8017 if (msix_table_offset
) {
8018 *msix_table_offset
= bar_size
;
8021 msix_table_size
= PCI_MSIX_ENTRY_SIZE
* total_irqs
;
8022 bar_size
+= msix_table_size
;
8023 bar_size
= QEMU_ALIGN_UP(bar_size
, 4 * KiB
);
8025 if (msix_pba_offset
) {
8026 *msix_pba_offset
= bar_size
;
8029 msix_pba_size
= QEMU_ALIGN_UP(total_irqs
, 64) / 8;
8030 bar_size
+= msix_pba_size
;
8032 bar_size
= pow2ceil(bar_size
);
8036 static void nvme_init_sriov(NvmeCtrl
*n
, PCIDevice
*pci_dev
, uint16_t offset
)
8038 uint16_t vf_dev_id
= n
->params
.use_intel_id
?
8039 PCI_DEVICE_ID_INTEL_NVME
: PCI_DEVICE_ID_REDHAT_NVME
;
8040 NvmePriCtrlCap
*cap
= &n
->pri_ctrl_cap
;
8041 uint64_t bar_size
= nvme_bar_size(le16_to_cpu(cap
->vqfrsm
),
8042 le16_to_cpu(cap
->vifrsm
),
8045 pcie_sriov_pf_init(pci_dev
, offset
, "nvme", vf_dev_id
,
8046 n
->params
.sriov_max_vfs
, n
->params
.sriov_max_vfs
,
8047 NVME_VF_OFFSET
, NVME_VF_STRIDE
);
8049 pcie_sriov_pf_init_vf_bar(pci_dev
, 0, PCI_BASE_ADDRESS_SPACE_MEMORY
|
8050 PCI_BASE_ADDRESS_MEM_TYPE_64
, bar_size
);
8053 static int nvme_add_pm_capability(PCIDevice
*pci_dev
, uint8_t offset
)
8058 ret
= pci_add_capability(pci_dev
, PCI_CAP_ID_PM
, offset
,
8059 PCI_PM_SIZEOF
, &err
);
8061 error_report_err(err
);
8065 pci_set_word(pci_dev
->config
+ offset
+ PCI_PM_PMC
,
8066 PCI_PM_CAP_VER_1_2
);
8067 pci_set_word(pci_dev
->config
+ offset
+ PCI_PM_CTRL
,
8068 PCI_PM_CTRL_NO_SOFT_RESET
);
8069 pci_set_word(pci_dev
->wmask
+ offset
+ PCI_PM_CTRL
,
8070 PCI_PM_CTRL_STATE_MASK
);
8075 static bool nvme_init_pci(NvmeCtrl
*n
, PCIDevice
*pci_dev
, Error
**errp
)
8078 uint8_t *pci_conf
= pci_dev
->config
;
8080 unsigned msix_table_offset
, msix_pba_offset
;
8083 pci_conf
[PCI_INTERRUPT_PIN
] = 1;
8084 pci_config_set_prog_interface(pci_conf
, 0x2);
8086 if (n
->params
.use_intel_id
) {
8087 pci_config_set_vendor_id(pci_conf
, PCI_VENDOR_ID_INTEL
);
8088 pci_config_set_device_id(pci_conf
, PCI_DEVICE_ID_INTEL_NVME
);
8090 pci_config_set_vendor_id(pci_conf
, PCI_VENDOR_ID_REDHAT
);
8091 pci_config_set_device_id(pci_conf
, PCI_DEVICE_ID_REDHAT_NVME
);
8094 pci_config_set_class(pci_conf
, PCI_CLASS_STORAGE_EXPRESS
);
8095 nvme_add_pm_capability(pci_dev
, 0x60);
8096 pcie_endpoint_cap_init(pci_dev
, 0x80);
8097 pcie_cap_flr_init(pci_dev
);
8098 if (n
->params
.sriov_max_vfs
) {
8099 pcie_ari_init(pci_dev
, 0x100, 1);
8102 /* add one to max_ioqpairs to account for the admin queue pair */
8103 bar_size
= nvme_bar_size(n
->params
.max_ioqpairs
+ 1, n
->params
.msix_qsize
,
8104 &msix_table_offset
, &msix_pba_offset
);
8106 memory_region_init(&n
->bar0
, OBJECT(n
), "nvme-bar0", bar_size
);
8107 memory_region_init_io(&n
->iomem
, OBJECT(n
), &nvme_mmio_ops
, n
, "nvme",
8109 memory_region_add_subregion(&n
->bar0
, 0, &n
->iomem
);
8111 if (pci_is_vf(pci_dev
)) {
8112 pcie_sriov_vf_register_bar(pci_dev
, 0, &n
->bar0
);
8114 pci_register_bar(pci_dev
, 0, PCI_BASE_ADDRESS_SPACE_MEMORY
|
8115 PCI_BASE_ADDRESS_MEM_TYPE_64
, &n
->bar0
);
8117 ret
= msix_init(pci_dev
, n
->params
.msix_qsize
,
8118 &n
->bar0
, 0, msix_table_offset
,
8119 &n
->bar0
, 0, msix_pba_offset
, 0, errp
);
8120 if (ret
== -ENOTSUP
) {
8121 /* report that msix is not supported, but do not error out */
8122 warn_report_err(*errp
);
8124 } else if (ret
< 0) {
8125 /* propagate error to caller */
8129 nvme_update_msixcap_ts(pci_dev
, n
->conf_msix_qsize
);
8131 if (n
->params
.cmb_size_mb
) {
8132 nvme_init_cmb(n
, pci_dev
);
8136 nvme_init_pmr(n
, pci_dev
);
8139 if (!pci_is_vf(pci_dev
) && n
->params
.sriov_max_vfs
) {
8140 nvme_init_sriov(n
, pci_dev
, 0x120);
8146 static void nvme_init_subnqn(NvmeCtrl
*n
)
8148 NvmeSubsystem
*subsys
= n
->subsys
;
8149 NvmeIdCtrl
*id
= &n
->id_ctrl
;
8152 snprintf((char *)id
->subnqn
, sizeof(id
->subnqn
),
8153 "nqn.2019-08.org.qemu:%s", n
->params
.serial
);
8155 pstrcpy((char *)id
->subnqn
, sizeof(id
->subnqn
), (char*)subsys
->subnqn
);
8159 static void nvme_init_ctrl(NvmeCtrl
*n
, PCIDevice
*pci_dev
)
8161 NvmeIdCtrl
*id
= &n
->id_ctrl
;
8162 uint8_t *pci_conf
= pci_dev
->config
;
8163 uint64_t cap
= ldq_le_p(&n
->bar
.cap
);
8164 NvmeSecCtrlEntry
*sctrl
= nvme_sctrl(n
);
8167 id
->vid
= cpu_to_le16(pci_get_word(pci_conf
+ PCI_VENDOR_ID
));
8168 id
->ssvid
= cpu_to_le16(pci_get_word(pci_conf
+ PCI_SUBSYSTEM_VENDOR_ID
));
8169 strpadcpy((char *)id
->mn
, sizeof(id
->mn
), "QEMU NVMe Ctrl", ' ');
8170 strpadcpy((char *)id
->fr
, sizeof(id
->fr
), QEMU_VERSION
, ' ');
8171 strpadcpy((char *)id
->sn
, sizeof(id
->sn
), n
->params
.serial
, ' ');
8173 id
->cntlid
= cpu_to_le16(n
->cntlid
);
8175 id
->oaes
= cpu_to_le32(NVME_OAES_NS_ATTR
);
8176 ctratt
= NVME_CTRATT_ELBAS
;
8180 if (n
->params
.use_intel_id
) {
8190 id
->mdts
= n
->params
.mdts
;
8191 id
->ver
= cpu_to_le32(NVME_SPEC_VER
);
8193 cpu_to_le16(NVME_OACS_NS_MGMT
| NVME_OACS_FORMAT
| NVME_OACS_DBBUF
|
8194 NVME_OACS_DIRECTIVES
);
8195 id
->cntrltype
= 0x1;
8198 * Because the controller always completes the Abort command immediately,
8199 * there can never be more than one concurrently executing Abort command,
8200 * so this value is never used for anything. Note that there can easily be
8201 * many Abort commands in the queues, but they are not considered
8202 * "executing" until processed by nvme_abort.
8204 * The specification recommends a value of 3 for Abort Command Limit (four
8205 * concurrently outstanding Abort commands), so lets use that though it is
8209 id
->aerl
= n
->params
.aerl
;
8210 id
->frmw
= (NVME_NUM_FW_SLOTS
<< 1) | NVME_FRMW_SLOT1_RO
;
8211 id
->lpa
= NVME_LPA_NS_SMART
| NVME_LPA_CSE
| NVME_LPA_EXTENDED
;
8213 /* recommended default value (~70 C) */
8214 id
->wctemp
= cpu_to_le16(NVME_TEMPERATURE_WARNING
);
8215 id
->cctemp
= cpu_to_le16(NVME_TEMPERATURE_CRITICAL
);
8217 id
->sqes
= (0x6 << 4) | 0x6;
8218 id
->cqes
= (0x4 << 4) | 0x4;
8219 id
->nn
= cpu_to_le32(NVME_MAX_NAMESPACES
);
8220 id
->oncs
= cpu_to_le16(NVME_ONCS_WRITE_ZEROES
| NVME_ONCS_TIMESTAMP
|
8221 NVME_ONCS_FEATURES
| NVME_ONCS_DSM
|
8222 NVME_ONCS_COMPARE
| NVME_ONCS_COPY
);
8225 * NOTE: If this device ever supports a command set that does NOT use 0x0
8226 * as a Flush-equivalent operation, support for the broadcast NSID in Flush
8227 * should probably be removed.
8229 * See comment in nvme_io_cmd.
8231 id
->vwc
= NVME_VWC_NSID_BROADCAST_SUPPORT
| NVME_VWC_PRESENT
;
8233 id
->ocfs
= cpu_to_le16(NVME_OCFS_COPY_FORMAT_0
| NVME_OCFS_COPY_FORMAT_1
);
8234 id
->sgls
= cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN
);
8236 nvme_init_subnqn(n
);
8238 id
->psd
[0].mp
= cpu_to_le16(0x9c4);
8239 id
->psd
[0].enlat
= cpu_to_le32(0x10);
8240 id
->psd
[0].exlat
= cpu_to_le32(0x4);
8243 id
->cmic
|= NVME_CMIC_MULTI_CTRL
;
8244 ctratt
|= NVME_CTRATT_ENDGRPS
;
8246 id
->endgidmax
= cpu_to_le16(0x1);
8248 if (n
->subsys
->endgrp
.fdp
.enabled
) {
8249 ctratt
|= NVME_CTRATT_FDPS
;
8253 id
->ctratt
= cpu_to_le32(ctratt
);
8255 NVME_CAP_SET_MQES(cap
, 0x7ff);
8256 NVME_CAP_SET_CQR(cap
, 1);
8257 NVME_CAP_SET_TO(cap
, 0xf);
8258 NVME_CAP_SET_CSS(cap
, NVME_CAP_CSS_NVM
);
8259 NVME_CAP_SET_CSS(cap
, NVME_CAP_CSS_CSI_SUPP
);
8260 NVME_CAP_SET_CSS(cap
, NVME_CAP_CSS_ADMIN_ONLY
);
8261 NVME_CAP_SET_MPSMAX(cap
, 4);
8262 NVME_CAP_SET_CMBS(cap
, n
->params
.cmb_size_mb
? 1 : 0);
8263 NVME_CAP_SET_PMRS(cap
, n
->pmr
.dev
? 1 : 0);
8264 stq_le_p(&n
->bar
.cap
, cap
);
8266 stl_le_p(&n
->bar
.vs
, NVME_SPEC_VER
);
8267 n
->bar
.intmc
= n
->bar
.intms
= 0;
8269 if (pci_is_vf(pci_dev
) && !sctrl
->scs
) {
8270 stl_le_p(&n
->bar
.csts
, NVME_CSTS_FAILED
);
8274 static int nvme_init_subsys(NvmeCtrl
*n
, Error
**errp
)
8282 cntlid
= nvme_subsys_register_ctrl(n
, errp
);
8292 void nvme_attach_ns(NvmeCtrl
*n
, NvmeNamespace
*ns
)
8294 uint32_t nsid
= ns
->params
.nsid
;
8295 assert(nsid
&& nsid
<= NVME_MAX_NAMESPACES
);
8297 n
->namespaces
[nsid
] = ns
;
8300 n
->dmrsl
= MIN_NON_ZERO(n
->dmrsl
,
8301 BDRV_REQUEST_MAX_BYTES
/ nvme_l2b(ns
, 1));
8304 static void nvme_realize(PCIDevice
*pci_dev
, Error
**errp
)
8306 NvmeCtrl
*n
= NVME(pci_dev
);
8307 DeviceState
*dev
= DEVICE(pci_dev
);
8309 NvmeCtrl
*pn
= NVME(pcie_sriov_get_pf(pci_dev
));
8311 if (pci_is_vf(pci_dev
)) {
8313 * VFs derive settings from the parent. PF's lifespan exceeds
8314 * that of VF's, so it's safe to share params.serial.
8316 memcpy(&n
->params
, &pn
->params
, sizeof(NvmeParams
));
8317 n
->subsys
= pn
->subsys
;
8320 if (!nvme_check_params(n
, errp
)) {
8324 qbus_init(&n
->bus
, sizeof(NvmeBus
), TYPE_NVME_BUS
, dev
, dev
->id
);
8326 if (nvme_init_subsys(n
, errp
)) {
8330 if (!nvme_init_pci(n
, pci_dev
, errp
)) {
8333 nvme_init_ctrl(n
, pci_dev
);
8335 /* setup a namespace if the controller drive property was given */
8336 if (n
->namespace.blkconf
.blk
) {
8338 ns
->params
.nsid
= 1;
8340 if (nvme_ns_setup(ns
, errp
)) {
8344 nvme_attach_ns(n
, ns
);
8348 static void nvme_exit(PCIDevice
*pci_dev
)
8350 NvmeCtrl
*n
= NVME(pci_dev
);
8354 nvme_ctrl_reset(n
, NVME_RESET_FUNCTION
);
8357 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
8364 nvme_subsys_unregister_ctrl(n
->subsys
, n
);
8369 g_free(n
->aer_reqs
);
8371 if (n
->params
.cmb_size_mb
) {
8376 host_memory_backend_set_mapped(n
->pmr
.dev
, false);
8379 if (!pci_is_vf(pci_dev
) && n
->params
.sriov_max_vfs
) {
8380 pcie_sriov_pf_exit(pci_dev
);
8383 msix_uninit(pci_dev
, &n
->bar0
, &n
->bar0
);
8384 memory_region_del_subregion(&n
->bar0
, &n
->iomem
);
8387 static Property nvme_props
[] = {
8388 DEFINE_BLOCK_PROPERTIES(NvmeCtrl
, namespace.blkconf
),
8389 DEFINE_PROP_LINK("pmrdev", NvmeCtrl
, pmr
.dev
, TYPE_MEMORY_BACKEND
,
8390 HostMemoryBackend
*),
8391 DEFINE_PROP_LINK("subsys", NvmeCtrl
, subsys
, TYPE_NVME_SUBSYS
,
8393 DEFINE_PROP_STRING("serial", NvmeCtrl
, params
.serial
),
8394 DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl
, params
.cmb_size_mb
, 0),
8395 DEFINE_PROP_UINT32("num_queues", NvmeCtrl
, params
.num_queues
, 0),
8396 DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl
, params
.max_ioqpairs
, 64),
8397 DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl
, params
.msix_qsize
, 65),
8398 DEFINE_PROP_UINT8("aerl", NvmeCtrl
, params
.aerl
, 3),
8399 DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl
, params
.aer_max_queued
, 64),
8400 DEFINE_PROP_UINT8("mdts", NvmeCtrl
, params
.mdts
, 7),
8401 DEFINE_PROP_UINT8("vsl", NvmeCtrl
, params
.vsl
, 7),
8402 DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl
, params
.use_intel_id
, false),
8403 DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl
, params
.legacy_cmb
, false),
8404 DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl
, params
.ioeventfd
, false),
8405 DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl
, params
.zasl
, 0),
8406 DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl
,
8407 params
.auto_transition_zones
, true),
8408 DEFINE_PROP_UINT8("sriov_max_vfs", NvmeCtrl
, params
.sriov_max_vfs
, 0),
8409 DEFINE_PROP_UINT16("sriov_vq_flexible", NvmeCtrl
,
8410 params
.sriov_vq_flexible
, 0),
8411 DEFINE_PROP_UINT16("sriov_vi_flexible", NvmeCtrl
,
8412 params
.sriov_vi_flexible
, 0),
8413 DEFINE_PROP_UINT8("sriov_max_vi_per_vf", NvmeCtrl
,
8414 params
.sriov_max_vi_per_vf
, 0),
8415 DEFINE_PROP_UINT8("sriov_max_vq_per_vf", NvmeCtrl
,
8416 params
.sriov_max_vq_per_vf
, 0),
8417 DEFINE_PROP_END_OF_LIST(),
8420 static void nvme_get_smart_warning(Object
*obj
, Visitor
*v
, const char *name
,
8421 void *opaque
, Error
**errp
)
8423 NvmeCtrl
*n
= NVME(obj
);
8424 uint8_t value
= n
->smart_critical_warning
;
8426 visit_type_uint8(v
, name
, &value
, errp
);
8429 static void nvme_set_smart_warning(Object
*obj
, Visitor
*v
, const char *name
,
8430 void *opaque
, Error
**errp
)
8432 NvmeCtrl
*n
= NVME(obj
);
8433 uint8_t value
, old_value
, cap
= 0, index
, event
;
8435 if (!visit_type_uint8(v
, name
, &value
, errp
)) {
8439 cap
= NVME_SMART_SPARE
| NVME_SMART_TEMPERATURE
| NVME_SMART_RELIABILITY
8440 | NVME_SMART_MEDIA_READ_ONLY
| NVME_SMART_FAILED_VOLATILE_MEDIA
;
8441 if (NVME_CAP_PMRS(ldq_le_p(&n
->bar
.cap
))) {
8442 cap
|= NVME_SMART_PMR_UNRELIABLE
;
8445 if ((value
& cap
) != value
) {
8446 error_setg(errp
, "unsupported smart critical warning bits: 0x%x",
8451 old_value
= n
->smart_critical_warning
;
8452 n
->smart_critical_warning
= value
;
8454 /* only inject new bits of smart critical warning */
8455 for (index
= 0; index
< NVME_SMART_WARN_MAX
; index
++) {
8457 if (value
& ~old_value
& event
)
8458 nvme_smart_event(n
, event
);
8462 static void nvme_pci_reset(DeviceState
*qdev
)
8464 PCIDevice
*pci_dev
= PCI_DEVICE(qdev
);
8465 NvmeCtrl
*n
= NVME(pci_dev
);
8467 trace_pci_nvme_pci_reset();
8468 nvme_ctrl_reset(n
, NVME_RESET_FUNCTION
);
8471 static void nvme_sriov_pre_write_ctrl(PCIDevice
*dev
, uint32_t address
,
8472 uint32_t val
, int len
)
8474 NvmeCtrl
*n
= NVME(dev
);
8475 NvmeSecCtrlEntry
*sctrl
;
8476 uint16_t sriov_cap
= dev
->exp
.sriov_cap
;
8477 uint32_t off
= address
- sriov_cap
;
8484 if (range_covers_byte(off
, len
, PCI_SRIOV_CTRL
)) {
8485 if (!(val
& PCI_SRIOV_CTRL_VFE
)) {
8486 num_vfs
= pci_get_word(dev
->config
+ sriov_cap
+ PCI_SRIOV_NUM_VF
);
8487 for (i
= 0; i
< num_vfs
; i
++) {
8488 sctrl
= &n
->sec_ctrl_list
.sec
[i
];
8489 nvme_virt_set_state(n
, le16_to_cpu(sctrl
->scid
), false);
8495 static void nvme_pci_write_config(PCIDevice
*dev
, uint32_t address
,
8496 uint32_t val
, int len
)
8498 nvme_sriov_pre_write_ctrl(dev
, address
, val
, len
);
8499 pci_default_write_config(dev
, address
, val
, len
);
8500 pcie_cap_flr_write_config(dev
, address
, val
, len
);
8503 static const VMStateDescription nvme_vmstate
= {
8508 static void nvme_class_init(ObjectClass
*oc
, void *data
)
8510 DeviceClass
*dc
= DEVICE_CLASS(oc
);
8511 PCIDeviceClass
*pc
= PCI_DEVICE_CLASS(oc
);
8513 pc
->realize
= nvme_realize
;
8514 pc
->config_write
= nvme_pci_write_config
;
8515 pc
->exit
= nvme_exit
;
8516 pc
->class_id
= PCI_CLASS_STORAGE_EXPRESS
;
8519 set_bit(DEVICE_CATEGORY_STORAGE
, dc
->categories
);
8520 dc
->desc
= "Non-Volatile Memory Express";
8521 device_class_set_props(dc
, nvme_props
);
8522 dc
->vmsd
= &nvme_vmstate
;
8523 dc
->reset
= nvme_pci_reset
;
8526 static void nvme_instance_init(Object
*obj
)
8528 NvmeCtrl
*n
= NVME(obj
);
8530 device_add_bootindex_property(obj
, &n
->namespace.blkconf
.bootindex
,
8531 "bootindex", "/namespace@1,0",
8534 object_property_add(obj
, "smart_critical_warning", "uint8",
8535 nvme_get_smart_warning
,
8536 nvme_set_smart_warning
, NULL
, NULL
);
8539 static const TypeInfo nvme_info
= {
8541 .parent
= TYPE_PCI_DEVICE
,
8542 .instance_size
= sizeof(NvmeCtrl
),
8543 .instance_init
= nvme_instance_init
,
8544 .class_init
= nvme_class_init
,
8545 .interfaces
= (InterfaceInfo
[]) {
8546 { INTERFACE_PCIE_DEVICE
},
8551 static const TypeInfo nvme_bus_info
= {
8552 .name
= TYPE_NVME_BUS
,
8554 .instance_size
= sizeof(NvmeBus
),
8557 static void nvme_register_types(void)
8559 type_register_static(&nvme_info
);
8560 type_register_static(&nvme_bus_info
);
8563 type_init(nvme_register_types
)