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
:
1758 status
= NVME_WRITE_FAULT
;
1761 status
= NVME_INTERNAL_DEV_ERROR
;
1765 trace_pci_nvme_err_aio(nvme_cid(req
), strerror(-ret
), status
);
1767 error_setg_errno(&local_err
, -ret
, "aio failed");
1768 error_report_err(local_err
);
1771 * Set the command status code to the first encountered error but allow a
1772 * subsequent Internal Device Error to trump it.
1774 if (req
->status
&& status
!= NVME_INTERNAL_DEV_ERROR
) {
1778 req
->status
= status
;
1781 static inline uint32_t nvme_zone_idx(NvmeNamespace
*ns
, uint64_t slba
)
1783 return ns
->zone_size_log2
> 0 ? slba
>> ns
->zone_size_log2
:
1784 slba
/ ns
->zone_size
;
1787 static inline NvmeZone
*nvme_get_zone_by_slba(NvmeNamespace
*ns
, uint64_t slba
)
1789 uint32_t zone_idx
= nvme_zone_idx(ns
, slba
);
1791 if (zone_idx
>= ns
->num_zones
) {
1795 return &ns
->zone_array
[zone_idx
];
1798 static uint16_t nvme_check_zone_state_for_write(NvmeZone
*zone
)
1800 uint64_t zslba
= zone
->d
.zslba
;
1802 switch (nvme_get_zone_state(zone
)) {
1803 case NVME_ZONE_STATE_EMPTY
:
1804 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1805 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1806 case NVME_ZONE_STATE_CLOSED
:
1807 return NVME_SUCCESS
;
1808 case NVME_ZONE_STATE_FULL
:
1809 trace_pci_nvme_err_zone_is_full(zslba
);
1810 return NVME_ZONE_FULL
;
1811 case NVME_ZONE_STATE_OFFLINE
:
1812 trace_pci_nvme_err_zone_is_offline(zslba
);
1813 return NVME_ZONE_OFFLINE
;
1814 case NVME_ZONE_STATE_READ_ONLY
:
1815 trace_pci_nvme_err_zone_is_read_only(zslba
);
1816 return NVME_ZONE_READ_ONLY
;
1821 return NVME_INTERNAL_DEV_ERROR
;
1824 static uint16_t nvme_check_zone_write(NvmeNamespace
*ns
, NvmeZone
*zone
,
1825 uint64_t slba
, uint32_t nlb
)
1827 uint64_t zcap
= nvme_zone_wr_boundary(zone
);
1830 status
= nvme_check_zone_state_for_write(zone
);
1835 if (zone
->d
.za
& NVME_ZA_ZRWA_VALID
) {
1836 uint64_t ezrwa
= zone
->w_ptr
+ 2 * ns
->zns
.zrwas
;
1838 if (slba
< zone
->w_ptr
|| slba
+ nlb
> ezrwa
) {
1839 trace_pci_nvme_err_zone_invalid_write(slba
, zone
->w_ptr
);
1840 return NVME_ZONE_INVALID_WRITE
;
1843 if (unlikely(slba
!= zone
->w_ptr
)) {
1844 trace_pci_nvme_err_write_not_at_wp(slba
, zone
->d
.zslba
,
1846 return NVME_ZONE_INVALID_WRITE
;
1850 if (unlikely((slba
+ nlb
) > zcap
)) {
1851 trace_pci_nvme_err_zone_boundary(slba
, nlb
, zcap
);
1852 return NVME_ZONE_BOUNDARY_ERROR
;
1855 return NVME_SUCCESS
;
1858 static uint16_t nvme_check_zone_state_for_read(NvmeZone
*zone
)
1860 switch (nvme_get_zone_state(zone
)) {
1861 case NVME_ZONE_STATE_EMPTY
:
1862 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1863 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1864 case NVME_ZONE_STATE_FULL
:
1865 case NVME_ZONE_STATE_CLOSED
:
1866 case NVME_ZONE_STATE_READ_ONLY
:
1867 return NVME_SUCCESS
;
1868 case NVME_ZONE_STATE_OFFLINE
:
1869 trace_pci_nvme_err_zone_is_offline(zone
->d
.zslba
);
1870 return NVME_ZONE_OFFLINE
;
1875 return NVME_INTERNAL_DEV_ERROR
;
1878 static uint16_t nvme_check_zone_read(NvmeNamespace
*ns
, uint64_t slba
,
1882 uint64_t bndry
, end
;
1885 zone
= nvme_get_zone_by_slba(ns
, slba
);
1888 bndry
= nvme_zone_rd_boundary(ns
, zone
);
1891 status
= nvme_check_zone_state_for_read(zone
);
1894 } else if (unlikely(end
> bndry
)) {
1895 if (!ns
->params
.cross_zone_read
) {
1896 status
= NVME_ZONE_BOUNDARY_ERROR
;
1899 * Read across zone boundary - check that all subsequent
1900 * zones that are being read have an appropriate state.
1904 status
= nvme_check_zone_state_for_read(zone
);
1908 } while (end
> nvme_zone_rd_boundary(ns
, zone
));
1915 static uint16_t nvme_zrm_finish(NvmeNamespace
*ns
, NvmeZone
*zone
)
1917 switch (nvme_get_zone_state(zone
)) {
1918 case NVME_ZONE_STATE_FULL
:
1919 return NVME_SUCCESS
;
1921 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1922 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1923 nvme_aor_dec_open(ns
);
1925 case NVME_ZONE_STATE_CLOSED
:
1926 nvme_aor_dec_active(ns
);
1928 if (zone
->d
.za
& NVME_ZA_ZRWA_VALID
) {
1929 zone
->d
.za
&= ~NVME_ZA_ZRWA_VALID
;
1930 if (ns
->params
.numzrwa
) {
1936 case NVME_ZONE_STATE_EMPTY
:
1937 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_FULL
);
1938 return NVME_SUCCESS
;
1941 return NVME_ZONE_INVAL_TRANSITION
;
1945 static uint16_t nvme_zrm_close(NvmeNamespace
*ns
, NvmeZone
*zone
)
1947 switch (nvme_get_zone_state(zone
)) {
1948 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1949 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1950 nvme_aor_dec_open(ns
);
1951 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_CLOSED
);
1953 case NVME_ZONE_STATE_CLOSED
:
1954 return NVME_SUCCESS
;
1957 return NVME_ZONE_INVAL_TRANSITION
;
1961 static uint16_t nvme_zrm_reset(NvmeNamespace
*ns
, NvmeZone
*zone
)
1963 switch (nvme_get_zone_state(zone
)) {
1964 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
1965 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
1966 nvme_aor_dec_open(ns
);
1968 case NVME_ZONE_STATE_CLOSED
:
1969 nvme_aor_dec_active(ns
);
1971 if (zone
->d
.za
& NVME_ZA_ZRWA_VALID
) {
1972 if (ns
->params
.numzrwa
) {
1978 case NVME_ZONE_STATE_FULL
:
1979 zone
->w_ptr
= zone
->d
.zslba
;
1980 zone
->d
.wp
= zone
->w_ptr
;
1981 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_EMPTY
);
1983 case NVME_ZONE_STATE_EMPTY
:
1984 return NVME_SUCCESS
;
1987 return NVME_ZONE_INVAL_TRANSITION
;
1991 static void nvme_zrm_auto_transition_zone(NvmeNamespace
*ns
)
1995 if (ns
->params
.max_open_zones
&&
1996 ns
->nr_open_zones
== ns
->params
.max_open_zones
) {
1997 zone
= QTAILQ_FIRST(&ns
->imp_open_zones
);
2000 * Automatically close this implicitly open zone.
2002 QTAILQ_REMOVE(&ns
->imp_open_zones
, zone
, entry
);
2003 nvme_zrm_close(ns
, zone
);
2009 NVME_ZRM_AUTO
= 1 << 0,
2010 NVME_ZRM_ZRWA
= 1 << 1,
2013 static uint16_t nvme_zrm_open_flags(NvmeCtrl
*n
, NvmeNamespace
*ns
,
2014 NvmeZone
*zone
, int flags
)
2019 switch (nvme_get_zone_state(zone
)) {
2020 case NVME_ZONE_STATE_EMPTY
:
2025 case NVME_ZONE_STATE_CLOSED
:
2026 if (n
->params
.auto_transition_zones
) {
2027 nvme_zrm_auto_transition_zone(ns
);
2029 status
= nvme_zns_check_resources(ns
, act
, 1,
2030 (flags
& NVME_ZRM_ZRWA
) ? 1 : 0);
2036 nvme_aor_inc_active(ns
);
2039 nvme_aor_inc_open(ns
);
2041 if (flags
& NVME_ZRM_AUTO
) {
2042 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_IMPLICITLY_OPEN
);
2043 return NVME_SUCCESS
;
2048 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
2049 if (flags
& NVME_ZRM_AUTO
) {
2050 return NVME_SUCCESS
;
2053 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_EXPLICITLY_OPEN
);
2057 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
2058 if (flags
& NVME_ZRM_ZRWA
) {
2061 zone
->d
.za
|= NVME_ZA_ZRWA_VALID
;
2064 return NVME_SUCCESS
;
2067 return NVME_ZONE_INVAL_TRANSITION
;
2071 static inline uint16_t nvme_zrm_auto(NvmeCtrl
*n
, NvmeNamespace
*ns
,
2074 return nvme_zrm_open_flags(n
, ns
, zone
, NVME_ZRM_AUTO
);
2077 static void nvme_advance_zone_wp(NvmeNamespace
*ns
, NvmeZone
*zone
,
2082 if (zone
->d
.wp
== nvme_zone_wr_boundary(zone
)) {
2083 nvme_zrm_finish(ns
, zone
);
2087 static void nvme_zoned_zrwa_implicit_flush(NvmeNamespace
*ns
, NvmeZone
*zone
,
2090 uint16_t nzrwafgs
= DIV_ROUND_UP(nlbc
, ns
->zns
.zrwafg
);
2092 nlbc
= nzrwafgs
* ns
->zns
.zrwafg
;
2094 trace_pci_nvme_zoned_zrwa_implicit_flush(zone
->d
.zslba
, nlbc
);
2096 zone
->w_ptr
+= nlbc
;
2098 nvme_advance_zone_wp(ns
, zone
, nlbc
);
2101 static void nvme_finalize_zoned_write(NvmeNamespace
*ns
, NvmeRequest
*req
)
2103 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2108 slba
= le64_to_cpu(rw
->slba
);
2109 nlb
= le16_to_cpu(rw
->nlb
) + 1;
2110 zone
= nvme_get_zone_by_slba(ns
, slba
);
2113 if (zone
->d
.za
& NVME_ZA_ZRWA_VALID
) {
2114 uint64_t ezrwa
= zone
->w_ptr
+ ns
->zns
.zrwas
- 1;
2115 uint64_t elba
= slba
+ nlb
- 1;
2118 nvme_zoned_zrwa_implicit_flush(ns
, zone
, elba
- ezrwa
);
2124 nvme_advance_zone_wp(ns
, zone
, nlb
);
2127 static inline bool nvme_is_write(NvmeRequest
*req
)
2129 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2131 return rw
->opcode
== NVME_CMD_WRITE
||
2132 rw
->opcode
== NVME_CMD_ZONE_APPEND
||
2133 rw
->opcode
== NVME_CMD_WRITE_ZEROES
;
2136 static AioContext
*nvme_get_aio_context(BlockAIOCB
*acb
)
2138 return qemu_get_aio_context();
2141 static void nvme_misc_cb(void *opaque
, int ret
)
2143 NvmeRequest
*req
= opaque
;
2145 trace_pci_nvme_misc_cb(nvme_cid(req
));
2148 nvme_aio_err(req
, ret
);
2151 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2154 void nvme_rw_complete_cb(void *opaque
, int ret
)
2156 NvmeRequest
*req
= opaque
;
2157 NvmeNamespace
*ns
= req
->ns
;
2158 BlockBackend
*blk
= ns
->blkconf
.blk
;
2159 BlockAcctCookie
*acct
= &req
->acct
;
2160 BlockAcctStats
*stats
= blk_get_stats(blk
);
2162 trace_pci_nvme_rw_complete_cb(nvme_cid(req
), blk_name(blk
));
2165 block_acct_failed(stats
, acct
);
2166 nvme_aio_err(req
, ret
);
2168 block_acct_done(stats
, acct
);
2171 if (ns
->params
.zoned
&& nvme_is_write(req
)) {
2172 nvme_finalize_zoned_write(ns
, req
);
2175 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2178 static void nvme_rw_cb(void *opaque
, int ret
)
2180 NvmeRequest
*req
= opaque
;
2181 NvmeNamespace
*ns
= req
->ns
;
2183 BlockBackend
*blk
= ns
->blkconf
.blk
;
2185 trace_pci_nvme_rw_cb(nvme_cid(req
), blk_name(blk
));
2192 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2193 uint64_t slba
= le64_to_cpu(rw
->slba
);
2194 uint32_t nlb
= (uint32_t)le16_to_cpu(rw
->nlb
) + 1;
2195 uint64_t offset
= nvme_moff(ns
, slba
);
2197 if (req
->cmd
.opcode
== NVME_CMD_WRITE_ZEROES
) {
2198 size_t mlen
= nvme_m2b(ns
, nlb
);
2200 req
->aiocb
= blk_aio_pwrite_zeroes(blk
, offset
, mlen
,
2202 nvme_rw_complete_cb
, req
);
2206 if (nvme_ns_ext(ns
) || req
->cmd
.mptr
) {
2209 nvme_sg_unmap(&req
->sg
);
2210 status
= nvme_map_mdata(nvme_ctrl(req
), nlb
, req
);
2216 if (req
->cmd
.opcode
== NVME_CMD_READ
) {
2217 return nvme_blk_read(blk
, offset
, 1, nvme_rw_complete_cb
, req
);
2220 return nvme_blk_write(blk
, offset
, 1, nvme_rw_complete_cb
, req
);
2225 nvme_rw_complete_cb(req
, ret
);
2228 static void nvme_verify_cb(void *opaque
, int ret
)
2230 NvmeBounceContext
*ctx
= opaque
;
2231 NvmeRequest
*req
= ctx
->req
;
2232 NvmeNamespace
*ns
= req
->ns
;
2233 BlockBackend
*blk
= ns
->blkconf
.blk
;
2234 BlockAcctCookie
*acct
= &req
->acct
;
2235 BlockAcctStats
*stats
= blk_get_stats(blk
);
2236 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2237 uint64_t slba
= le64_to_cpu(rw
->slba
);
2238 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
2239 uint16_t apptag
= le16_to_cpu(rw
->apptag
);
2240 uint16_t appmask
= le16_to_cpu(rw
->appmask
);
2241 uint64_t reftag
= le32_to_cpu(rw
->reftag
);
2242 uint64_t cdw3
= le32_to_cpu(rw
->cdw3
);
2245 reftag
|= cdw3
<< 32;
2247 trace_pci_nvme_verify_cb(nvme_cid(req
), prinfo
, apptag
, appmask
, reftag
);
2250 block_acct_failed(stats
, acct
);
2251 nvme_aio_err(req
, ret
);
2255 block_acct_done(stats
, acct
);
2257 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
2258 status
= nvme_dif_mangle_mdata(ns
, ctx
->mdata
.bounce
,
2259 ctx
->mdata
.iov
.size
, slba
);
2261 req
->status
= status
;
2265 req
->status
= nvme_dif_check(ns
, ctx
->data
.bounce
, ctx
->data
.iov
.size
,
2266 ctx
->mdata
.bounce
, ctx
->mdata
.iov
.size
,
2267 prinfo
, slba
, apptag
, appmask
, &reftag
);
2271 qemu_iovec_destroy(&ctx
->data
.iov
);
2272 g_free(ctx
->data
.bounce
);
2274 qemu_iovec_destroy(&ctx
->mdata
.iov
);
2275 g_free(ctx
->mdata
.bounce
);
2279 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2283 static void nvme_verify_mdata_in_cb(void *opaque
, int ret
)
2285 NvmeBounceContext
*ctx
= opaque
;
2286 NvmeRequest
*req
= ctx
->req
;
2287 NvmeNamespace
*ns
= req
->ns
;
2288 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2289 uint64_t slba
= le64_to_cpu(rw
->slba
);
2290 uint32_t nlb
= le16_to_cpu(rw
->nlb
) + 1;
2291 size_t mlen
= nvme_m2b(ns
, nlb
);
2292 uint64_t offset
= nvme_moff(ns
, slba
);
2293 BlockBackend
*blk
= ns
->blkconf
.blk
;
2295 trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req
), blk_name(blk
));
2301 ctx
->mdata
.bounce
= g_malloc(mlen
);
2303 qemu_iovec_reset(&ctx
->mdata
.iov
);
2304 qemu_iovec_add(&ctx
->mdata
.iov
, ctx
->mdata
.bounce
, mlen
);
2306 req
->aiocb
= blk_aio_preadv(blk
, offset
, &ctx
->mdata
.iov
, 0,
2307 nvme_verify_cb
, ctx
);
2311 nvme_verify_cb(ctx
, ret
);
2314 struct nvme_compare_ctx
{
2326 static void nvme_compare_mdata_cb(void *opaque
, int ret
)
2328 NvmeRequest
*req
= opaque
;
2329 NvmeNamespace
*ns
= req
->ns
;
2330 NvmeCtrl
*n
= nvme_ctrl(req
);
2331 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2332 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
2333 uint16_t apptag
= le16_to_cpu(rw
->apptag
);
2334 uint16_t appmask
= le16_to_cpu(rw
->appmask
);
2335 uint64_t reftag
= le32_to_cpu(rw
->reftag
);
2336 uint64_t cdw3
= le32_to_cpu(rw
->cdw3
);
2337 struct nvme_compare_ctx
*ctx
= req
->opaque
;
2338 g_autofree
uint8_t *buf
= NULL
;
2339 BlockBackend
*blk
= ns
->blkconf
.blk
;
2340 BlockAcctCookie
*acct
= &req
->acct
;
2341 BlockAcctStats
*stats
= blk_get_stats(blk
);
2342 uint16_t status
= NVME_SUCCESS
;
2344 reftag
|= cdw3
<< 32;
2346 trace_pci_nvme_compare_mdata_cb(nvme_cid(req
));
2349 block_acct_failed(stats
, acct
);
2350 nvme_aio_err(req
, ret
);
2354 buf
= g_malloc(ctx
->mdata
.iov
.size
);
2356 status
= nvme_bounce_mdata(n
, buf
, ctx
->mdata
.iov
.size
,
2357 NVME_TX_DIRECTION_TO_DEVICE
, req
);
2359 req
->status
= status
;
2363 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
2364 uint64_t slba
= le64_to_cpu(rw
->slba
);
2366 uint8_t *mbufp
= ctx
->mdata
.bounce
;
2367 uint8_t *end
= mbufp
+ ctx
->mdata
.iov
.size
;
2370 status
= nvme_dif_check(ns
, ctx
->data
.bounce
, ctx
->data
.iov
.size
,
2371 ctx
->mdata
.bounce
, ctx
->mdata
.iov
.size
, prinfo
,
2372 slba
, apptag
, appmask
, &reftag
);
2374 req
->status
= status
;
2379 * When formatted with protection information, do not compare the DIF
2382 if (!(ns
->id_ns
.dps
& NVME_ID_NS_DPS_FIRST_EIGHT
)) {
2383 pil
= ns
->lbaf
.ms
- nvme_pi_tuple_size(ns
);
2386 for (bufp
= buf
; mbufp
< end
; bufp
+= ns
->lbaf
.ms
, mbufp
+= ns
->lbaf
.ms
) {
2387 if (memcmp(bufp
+ pil
, mbufp
+ pil
, ns
->lbaf
.ms
- pil
)) {
2388 req
->status
= NVME_CMP_FAILURE
| NVME_DNR
;
2396 if (memcmp(buf
, ctx
->mdata
.bounce
, ctx
->mdata
.iov
.size
)) {
2397 req
->status
= NVME_CMP_FAILURE
| NVME_DNR
;
2401 block_acct_done(stats
, acct
);
2404 qemu_iovec_destroy(&ctx
->data
.iov
);
2405 g_free(ctx
->data
.bounce
);
2407 qemu_iovec_destroy(&ctx
->mdata
.iov
);
2408 g_free(ctx
->mdata
.bounce
);
2412 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2415 static void nvme_compare_data_cb(void *opaque
, int ret
)
2417 NvmeRequest
*req
= opaque
;
2418 NvmeCtrl
*n
= nvme_ctrl(req
);
2419 NvmeNamespace
*ns
= req
->ns
;
2420 BlockBackend
*blk
= ns
->blkconf
.blk
;
2421 BlockAcctCookie
*acct
= &req
->acct
;
2422 BlockAcctStats
*stats
= blk_get_stats(blk
);
2424 struct nvme_compare_ctx
*ctx
= req
->opaque
;
2425 g_autofree
uint8_t *buf
= NULL
;
2428 trace_pci_nvme_compare_data_cb(nvme_cid(req
));
2431 block_acct_failed(stats
, acct
);
2432 nvme_aio_err(req
, ret
);
2436 buf
= g_malloc(ctx
->data
.iov
.size
);
2438 status
= nvme_bounce_data(n
, buf
, ctx
->data
.iov
.size
,
2439 NVME_TX_DIRECTION_TO_DEVICE
, req
);
2441 req
->status
= status
;
2445 if (memcmp(buf
, ctx
->data
.bounce
, ctx
->data
.iov
.size
)) {
2446 req
->status
= NVME_CMP_FAILURE
| NVME_DNR
;
2451 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2452 uint64_t slba
= le64_to_cpu(rw
->slba
);
2453 uint32_t nlb
= le16_to_cpu(rw
->nlb
) + 1;
2454 size_t mlen
= nvme_m2b(ns
, nlb
);
2455 uint64_t offset
= nvme_moff(ns
, slba
);
2457 ctx
->mdata
.bounce
= g_malloc(mlen
);
2459 qemu_iovec_init(&ctx
->mdata
.iov
, 1);
2460 qemu_iovec_add(&ctx
->mdata
.iov
, ctx
->mdata
.bounce
, mlen
);
2462 req
->aiocb
= blk_aio_preadv(blk
, offset
, &ctx
->mdata
.iov
, 0,
2463 nvme_compare_mdata_cb
, req
);
2467 block_acct_done(stats
, acct
);
2470 qemu_iovec_destroy(&ctx
->data
.iov
);
2471 g_free(ctx
->data
.bounce
);
2474 nvme_enqueue_req_completion(nvme_cq(req
), req
);
2477 typedef struct NvmeDSMAIOCB
{
2483 NvmeDsmRange
*range
;
2488 static void nvme_dsm_cancel(BlockAIOCB
*aiocb
)
2490 NvmeDSMAIOCB
*iocb
= container_of(aiocb
, NvmeDSMAIOCB
, common
);
2492 /* break nvme_dsm_cb loop */
2493 iocb
->idx
= iocb
->nr
;
2494 iocb
->ret
= -ECANCELED
;
2497 blk_aio_cancel_async(iocb
->aiocb
);
2501 * We only reach this if nvme_dsm_cancel() has already been called or
2502 * the command ran to completion.
2504 assert(iocb
->idx
== iocb
->nr
);
2508 static const AIOCBInfo nvme_dsm_aiocb_info
= {
2509 .aiocb_size
= sizeof(NvmeDSMAIOCB
),
2510 .cancel_async
= nvme_dsm_cancel
,
2513 static void nvme_dsm_cb(void *opaque
, int ret
);
2515 static void nvme_dsm_md_cb(void *opaque
, int ret
)
2517 NvmeDSMAIOCB
*iocb
= opaque
;
2518 NvmeRequest
*req
= iocb
->req
;
2519 NvmeNamespace
*ns
= req
->ns
;
2520 NvmeDsmRange
*range
;
2524 if (ret
< 0 || iocb
->ret
< 0 || !ns
->lbaf
.ms
) {
2528 range
= &iocb
->range
[iocb
->idx
- 1];
2529 slba
= le64_to_cpu(range
->slba
);
2530 nlb
= le32_to_cpu(range
->nlb
);
2533 * Check that all block were discarded (zeroed); otherwise we do not zero
2537 ret
= nvme_block_status_all(ns
, slba
, nlb
, BDRV_BLOCK_ZERO
);
2543 nvme_dsm_cb(iocb
, 0);
2547 iocb
->aiocb
= blk_aio_pwrite_zeroes(ns
->blkconf
.blk
, nvme_moff(ns
, slba
),
2548 nvme_m2b(ns
, nlb
), BDRV_REQ_MAY_UNMAP
,
2553 nvme_dsm_cb(iocb
, ret
);
2556 static void nvme_dsm_cb(void *opaque
, int ret
)
2558 NvmeDSMAIOCB
*iocb
= opaque
;
2559 NvmeRequest
*req
= iocb
->req
;
2560 NvmeCtrl
*n
= nvme_ctrl(req
);
2561 NvmeNamespace
*ns
= req
->ns
;
2562 NvmeDsmRange
*range
;
2566 if (iocb
->ret
< 0) {
2568 } else if (ret
< 0) {
2574 if (iocb
->idx
== iocb
->nr
) {
2578 range
= &iocb
->range
[iocb
->idx
++];
2579 slba
= le64_to_cpu(range
->slba
);
2580 nlb
= le32_to_cpu(range
->nlb
);
2582 trace_pci_nvme_dsm_deallocate(slba
, nlb
);
2584 if (nlb
> n
->dmrsl
) {
2585 trace_pci_nvme_dsm_single_range_limit_exceeded(nlb
, n
->dmrsl
);
2589 if (nvme_check_bounds(ns
, slba
, nlb
)) {
2590 trace_pci_nvme_err_invalid_lba_range(slba
, nlb
,
2595 iocb
->aiocb
= blk_aio_pdiscard(ns
->blkconf
.blk
, nvme_l2b(ns
, slba
),
2597 nvme_dsm_md_cb
, iocb
);
2602 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
2603 qemu_aio_unref(iocb
);
2606 static uint16_t nvme_dsm(NvmeCtrl
*n
, NvmeRequest
*req
)
2608 NvmeNamespace
*ns
= req
->ns
;
2609 NvmeDsmCmd
*dsm
= (NvmeDsmCmd
*) &req
->cmd
;
2610 uint32_t attr
= le32_to_cpu(dsm
->attributes
);
2611 uint32_t nr
= (le32_to_cpu(dsm
->nr
) & 0xff) + 1;
2612 uint16_t status
= NVME_SUCCESS
;
2614 trace_pci_nvme_dsm(nr
, attr
);
2616 if (attr
& NVME_DSMGMT_AD
) {
2617 NvmeDSMAIOCB
*iocb
= blk_aio_get(&nvme_dsm_aiocb_info
, ns
->blkconf
.blk
,
2622 iocb
->range
= g_new(NvmeDsmRange
, nr
);
2626 status
= nvme_h2c(n
, (uint8_t *)iocb
->range
, sizeof(NvmeDsmRange
) * nr
,
2629 g_free(iocb
->range
);
2630 qemu_aio_unref(iocb
);
2635 req
->aiocb
= &iocb
->common
;
2636 nvme_dsm_cb(iocb
, 0);
2638 return NVME_NO_COMPLETE
;
2644 static uint16_t nvme_verify(NvmeCtrl
*n
, NvmeRequest
*req
)
2646 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
2647 NvmeNamespace
*ns
= req
->ns
;
2648 BlockBackend
*blk
= ns
->blkconf
.blk
;
2649 uint64_t slba
= le64_to_cpu(rw
->slba
);
2650 uint32_t nlb
= le16_to_cpu(rw
->nlb
) + 1;
2651 size_t len
= nvme_l2b(ns
, nlb
);
2652 int64_t offset
= nvme_l2b(ns
, slba
);
2653 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
2654 uint32_t reftag
= le32_to_cpu(rw
->reftag
);
2655 NvmeBounceContext
*ctx
= NULL
;
2658 trace_pci_nvme_verify(nvme_cid(req
), nvme_nsid(ns
), slba
, nlb
);
2660 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
2661 status
= nvme_check_prinfo(ns
, prinfo
, slba
, reftag
);
2666 if (prinfo
& NVME_PRINFO_PRACT
) {
2667 return NVME_INVALID_PROT_INFO
| NVME_DNR
;
2671 if (len
> n
->page_size
<< n
->params
.vsl
) {
2672 return NVME_INVALID_FIELD
| NVME_DNR
;
2675 status
= nvme_check_bounds(ns
, slba
, nlb
);
2680 if (NVME_ERR_REC_DULBE(ns
->features
.err_rec
)) {
2681 status
= nvme_check_dulbe(ns
, slba
, nlb
);
2687 ctx
= g_new0(NvmeBounceContext
, 1);
2690 ctx
->data
.bounce
= g_malloc(len
);
2692 qemu_iovec_init(&ctx
->data
.iov
, 1);
2693 qemu_iovec_add(&ctx
->data
.iov
, ctx
->data
.bounce
, len
);
2695 block_acct_start(blk_get_stats(blk
), &req
->acct
, ctx
->data
.iov
.size
,
2698 req
->aiocb
= blk_aio_preadv(ns
->blkconf
.blk
, offset
, &ctx
->data
.iov
, 0,
2699 nvme_verify_mdata_in_cb
, ctx
);
2700 return NVME_NO_COMPLETE
;
2703 typedef struct NvmeCopyAIOCB
{
2710 unsigned int format
;
2717 BlockAcctCookie read
;
2718 BlockAcctCookie write
;
2727 static void nvme_copy_cancel(BlockAIOCB
*aiocb
)
2729 NvmeCopyAIOCB
*iocb
= container_of(aiocb
, NvmeCopyAIOCB
, common
);
2731 iocb
->ret
= -ECANCELED
;
2734 blk_aio_cancel_async(iocb
->aiocb
);
2739 static const AIOCBInfo nvme_copy_aiocb_info
= {
2740 .aiocb_size
= sizeof(NvmeCopyAIOCB
),
2741 .cancel_async
= nvme_copy_cancel
,
2744 static void nvme_copy_done(NvmeCopyAIOCB
*iocb
)
2746 NvmeRequest
*req
= iocb
->req
;
2747 NvmeNamespace
*ns
= req
->ns
;
2748 BlockAcctStats
*stats
= blk_get_stats(ns
->blkconf
.blk
);
2750 if (iocb
->idx
!= iocb
->nr
) {
2751 req
->cqe
.result
= cpu_to_le32(iocb
->idx
);
2754 qemu_iovec_destroy(&iocb
->iov
);
2755 g_free(iocb
->bounce
);
2757 if (iocb
->ret
< 0) {
2758 block_acct_failed(stats
, &iocb
->acct
.read
);
2759 block_acct_failed(stats
, &iocb
->acct
.write
);
2761 block_acct_done(stats
, &iocb
->acct
.read
);
2762 block_acct_done(stats
, &iocb
->acct
.write
);
2765 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
2766 qemu_aio_unref(iocb
);
2769 static void nvme_do_copy(NvmeCopyAIOCB
*iocb
);
2771 static void nvme_copy_source_range_parse_format0(void *ranges
, int idx
,
2772 uint64_t *slba
, uint32_t *nlb
,
2777 NvmeCopySourceRangeFormat0
*_ranges
= ranges
;
2780 *slba
= le64_to_cpu(_ranges
[idx
].slba
);
2784 *nlb
= le16_to_cpu(_ranges
[idx
].nlb
) + 1;
2788 *apptag
= le16_to_cpu(_ranges
[idx
].apptag
);
2792 *appmask
= le16_to_cpu(_ranges
[idx
].appmask
);
2796 *reftag
= le32_to_cpu(_ranges
[idx
].reftag
);
2800 static void nvme_copy_source_range_parse_format1(void *ranges
, int idx
,
2801 uint64_t *slba
, uint32_t *nlb
,
2806 NvmeCopySourceRangeFormat1
*_ranges
= ranges
;
2809 *slba
= le64_to_cpu(_ranges
[idx
].slba
);
2813 *nlb
= le16_to_cpu(_ranges
[idx
].nlb
) + 1;
2817 *apptag
= le16_to_cpu(_ranges
[idx
].apptag
);
2821 *appmask
= le16_to_cpu(_ranges
[idx
].appmask
);
2827 *reftag
|= (uint64_t)_ranges
[idx
].sr
[4] << 40;
2828 *reftag
|= (uint64_t)_ranges
[idx
].sr
[5] << 32;
2829 *reftag
|= (uint64_t)_ranges
[idx
].sr
[6] << 24;
2830 *reftag
|= (uint64_t)_ranges
[idx
].sr
[7] << 16;
2831 *reftag
|= (uint64_t)_ranges
[idx
].sr
[8] << 8;
2832 *reftag
|= (uint64_t)_ranges
[idx
].sr
[9];
2836 static void nvme_copy_source_range_parse(void *ranges
, int idx
, uint8_t format
,
2837 uint64_t *slba
, uint32_t *nlb
,
2838 uint16_t *apptag
, uint16_t *appmask
,
2842 case NVME_COPY_FORMAT_0
:
2843 nvme_copy_source_range_parse_format0(ranges
, idx
, slba
, nlb
, apptag
,
2847 case NVME_COPY_FORMAT_1
:
2848 nvme_copy_source_range_parse_format1(ranges
, idx
, slba
, nlb
, apptag
,
2857 static void nvme_copy_out_completed_cb(void *opaque
, int ret
)
2859 NvmeCopyAIOCB
*iocb
= opaque
;
2860 NvmeRequest
*req
= iocb
->req
;
2861 NvmeNamespace
*ns
= req
->ns
;
2864 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, NULL
,
2865 &nlb
, NULL
, NULL
, NULL
);
2870 } else if (iocb
->ret
< 0) {
2874 if (ns
->params
.zoned
) {
2875 nvme_advance_zone_wp(ns
, iocb
->zone
, nlb
);
2884 static void nvme_copy_out_cb(void *opaque
, int ret
)
2886 NvmeCopyAIOCB
*iocb
= opaque
;
2887 NvmeRequest
*req
= iocb
->req
;
2888 NvmeNamespace
*ns
= req
->ns
;
2893 if (ret
< 0 || iocb
->ret
< 0 || !ns
->lbaf
.ms
) {
2897 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, NULL
,
2898 &nlb
, NULL
, NULL
, NULL
);
2900 mlen
= nvme_m2b(ns
, nlb
);
2901 mbounce
= iocb
->bounce
+ nvme_l2b(ns
, nlb
);
2903 qemu_iovec_reset(&iocb
->iov
);
2904 qemu_iovec_add(&iocb
->iov
, mbounce
, mlen
);
2906 iocb
->aiocb
= blk_aio_pwritev(ns
->blkconf
.blk
, nvme_moff(ns
, iocb
->slba
),
2907 &iocb
->iov
, 0, nvme_copy_out_completed_cb
,
2913 nvme_copy_out_completed_cb(iocb
, ret
);
2916 static void nvme_copy_in_completed_cb(void *opaque
, int ret
)
2918 NvmeCopyAIOCB
*iocb
= opaque
;
2919 NvmeRequest
*req
= iocb
->req
;
2920 NvmeNamespace
*ns
= req
->ns
;
2923 uint16_t apptag
, appmask
;
2931 } else if (iocb
->ret
< 0) {
2935 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, &slba
,
2936 &nlb
, &apptag
, &appmask
, &reftag
);
2937 len
= nvme_l2b(ns
, nlb
);
2939 trace_pci_nvme_copy_out(iocb
->slba
, nlb
);
2941 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
2942 NvmeCopyCmd
*copy
= (NvmeCopyCmd
*)&req
->cmd
;
2944 uint16_t prinfor
= ((copy
->control
[0] >> 4) & 0xf);
2945 uint16_t prinfow
= ((copy
->control
[2] >> 2) & 0xf);
2947 size_t mlen
= nvme_m2b(ns
, nlb
);
2948 uint8_t *mbounce
= iocb
->bounce
+ nvme_l2b(ns
, nlb
);
2950 status
= nvme_dif_mangle_mdata(ns
, mbounce
, mlen
, slba
);
2954 status
= nvme_dif_check(ns
, iocb
->bounce
, len
, mbounce
, mlen
, prinfor
,
2955 slba
, apptag
, appmask
, &reftag
);
2960 apptag
= le16_to_cpu(copy
->apptag
);
2961 appmask
= le16_to_cpu(copy
->appmask
);
2963 if (prinfow
& NVME_PRINFO_PRACT
) {
2964 status
= nvme_check_prinfo(ns
, prinfow
, iocb
->slba
, iocb
->reftag
);
2969 nvme_dif_pract_generate_dif(ns
, iocb
->bounce
, len
, mbounce
, mlen
,
2970 apptag
, &iocb
->reftag
);
2972 status
= nvme_dif_check(ns
, iocb
->bounce
, len
, mbounce
, mlen
,
2973 prinfow
, iocb
->slba
, apptag
, appmask
,
2981 status
= nvme_check_bounds(ns
, iocb
->slba
, nlb
);
2986 if (ns
->params
.zoned
) {
2987 status
= nvme_check_zone_write(ns
, iocb
->zone
, iocb
->slba
, nlb
);
2992 if (!(iocb
->zone
->d
.za
& NVME_ZA_ZRWA_VALID
)) {
2993 iocb
->zone
->w_ptr
+= nlb
;
2997 qemu_iovec_reset(&iocb
->iov
);
2998 qemu_iovec_add(&iocb
->iov
, iocb
->bounce
, len
);
3000 iocb
->aiocb
= blk_aio_pwritev(ns
->blkconf
.blk
, nvme_l2b(ns
, iocb
->slba
),
3001 &iocb
->iov
, 0, nvme_copy_out_cb
, iocb
);
3006 req
->status
= status
;
3012 static void nvme_copy_in_cb(void *opaque
, int ret
)
3014 NvmeCopyAIOCB
*iocb
= opaque
;
3015 NvmeRequest
*req
= iocb
->req
;
3016 NvmeNamespace
*ns
= req
->ns
;
3020 if (ret
< 0 || iocb
->ret
< 0 || !ns
->lbaf
.ms
) {
3024 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, &slba
,
3025 &nlb
, NULL
, NULL
, NULL
);
3027 qemu_iovec_reset(&iocb
->iov
);
3028 qemu_iovec_add(&iocb
->iov
, iocb
->bounce
+ nvme_l2b(ns
, nlb
),
3031 iocb
->aiocb
= blk_aio_preadv(ns
->blkconf
.blk
, nvme_moff(ns
, slba
),
3032 &iocb
->iov
, 0, nvme_copy_in_completed_cb
,
3037 nvme_copy_in_completed_cb(iocb
, ret
);
3040 static void nvme_do_copy(NvmeCopyAIOCB
*iocb
)
3042 NvmeRequest
*req
= iocb
->req
;
3043 NvmeNamespace
*ns
= req
->ns
;
3049 if (iocb
->ret
< 0) {
3053 if (iocb
->idx
== iocb
->nr
) {
3057 nvme_copy_source_range_parse(iocb
->ranges
, iocb
->idx
, iocb
->format
, &slba
,
3058 &nlb
, NULL
, NULL
, NULL
);
3059 len
= nvme_l2b(ns
, nlb
);
3061 trace_pci_nvme_copy_source_range(slba
, nlb
);
3063 if (nlb
> le16_to_cpu(ns
->id_ns
.mssrl
)) {
3064 status
= NVME_CMD_SIZE_LIMIT
| NVME_DNR
;
3068 status
= nvme_check_bounds(ns
, slba
, nlb
);
3073 if (NVME_ERR_REC_DULBE(ns
->features
.err_rec
)) {
3074 status
= nvme_check_dulbe(ns
, slba
, nlb
);
3080 if (ns
->params
.zoned
) {
3081 status
= nvme_check_zone_read(ns
, slba
, nlb
);
3087 qemu_iovec_reset(&iocb
->iov
);
3088 qemu_iovec_add(&iocb
->iov
, iocb
->bounce
, len
);
3090 iocb
->aiocb
= blk_aio_preadv(ns
->blkconf
.blk
, nvme_l2b(ns
, slba
),
3091 &iocb
->iov
, 0, nvme_copy_in_cb
, iocb
);
3095 req
->status
= status
;
3098 nvme_copy_done(iocb
);
3101 static uint16_t nvme_copy(NvmeCtrl
*n
, NvmeRequest
*req
)
3103 NvmeNamespace
*ns
= req
->ns
;
3104 NvmeCopyCmd
*copy
= (NvmeCopyCmd
*)&req
->cmd
;
3105 NvmeCopyAIOCB
*iocb
= blk_aio_get(&nvme_copy_aiocb_info
, ns
->blkconf
.blk
,
3107 uint16_t nr
= copy
->nr
+ 1;
3108 uint8_t format
= copy
->control
[0] & 0xf;
3109 uint16_t prinfor
= ((copy
->control
[0] >> 4) & 0xf);
3110 uint16_t prinfow
= ((copy
->control
[2] >> 2) & 0xf);
3111 size_t len
= sizeof(NvmeCopySourceRangeFormat0
);
3115 trace_pci_nvme_copy(nvme_cid(req
), nvme_nsid(ns
), nr
, format
);
3117 iocb
->ranges
= NULL
;
3120 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
) &&
3121 ((prinfor
& NVME_PRINFO_PRACT
) != (prinfow
& NVME_PRINFO_PRACT
))) {
3122 status
= NVME_INVALID_FIELD
| NVME_DNR
;
3126 if (!(n
->id_ctrl
.ocfs
& (1 << format
))) {
3127 trace_pci_nvme_err_copy_invalid_format(format
);
3128 status
= NVME_INVALID_FIELD
| NVME_DNR
;
3132 if (nr
> ns
->id_ns
.msrc
+ 1) {
3133 status
= NVME_CMD_SIZE_LIMIT
| NVME_DNR
;
3137 if ((ns
->pif
== 0x0 && format
!= 0x0) ||
3138 (ns
->pif
!= 0x0 && format
!= 0x1)) {
3139 status
= NVME_INVALID_FORMAT
| NVME_DNR
;
3144 len
= sizeof(NvmeCopySourceRangeFormat1
);
3147 iocb
->format
= format
;
3148 iocb
->ranges
= g_malloc_n(nr
, len
);
3149 status
= nvme_h2c(n
, (uint8_t *)iocb
->ranges
, len
* nr
, req
);
3154 iocb
->slba
= le64_to_cpu(copy
->sdlba
);
3156 if (ns
->params
.zoned
) {
3157 iocb
->zone
= nvme_get_zone_by_slba(ns
, iocb
->slba
);
3159 status
= NVME_LBA_RANGE
| NVME_DNR
;
3163 status
= nvme_zrm_auto(n
, ns
, iocb
->zone
);
3173 iocb
->reftag
= le32_to_cpu(copy
->reftag
);
3174 iocb
->reftag
|= (uint64_t)le32_to_cpu(copy
->cdw3
) << 32;
3175 iocb
->bounce
= g_malloc_n(le16_to_cpu(ns
->id_ns
.mssrl
),
3176 ns
->lbasz
+ ns
->lbaf
.ms
);
3178 qemu_iovec_init(&iocb
->iov
, 1);
3180 block_acct_start(blk_get_stats(ns
->blkconf
.blk
), &iocb
->acct
.read
, 0,
3182 block_acct_start(blk_get_stats(ns
->blkconf
.blk
), &iocb
->acct
.write
, 0,
3185 req
->aiocb
= &iocb
->common
;
3188 return NVME_NO_COMPLETE
;
3191 g_free(iocb
->ranges
);
3192 qemu_aio_unref(iocb
);
3196 static uint16_t nvme_compare(NvmeCtrl
*n
, NvmeRequest
*req
)
3198 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
3199 NvmeNamespace
*ns
= req
->ns
;
3200 BlockBackend
*blk
= ns
->blkconf
.blk
;
3201 uint64_t slba
= le64_to_cpu(rw
->slba
);
3202 uint32_t nlb
= le16_to_cpu(rw
->nlb
) + 1;
3203 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
3204 size_t data_len
= nvme_l2b(ns
, nlb
);
3205 size_t len
= data_len
;
3206 int64_t offset
= nvme_l2b(ns
, slba
);
3207 struct nvme_compare_ctx
*ctx
= NULL
;
3210 trace_pci_nvme_compare(nvme_cid(req
), nvme_nsid(ns
), slba
, nlb
);
3212 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
) && (prinfo
& NVME_PRINFO_PRACT
)) {
3213 return NVME_INVALID_PROT_INFO
| NVME_DNR
;
3216 if (nvme_ns_ext(ns
)) {
3217 len
+= nvme_m2b(ns
, nlb
);
3220 status
= nvme_check_mdts(n
, len
);
3225 status
= nvme_check_bounds(ns
, slba
, nlb
);
3230 if (NVME_ERR_REC_DULBE(ns
->features
.err_rec
)) {
3231 status
= nvme_check_dulbe(ns
, slba
, nlb
);
3237 status
= nvme_map_dptr(n
, &req
->sg
, len
, &req
->cmd
);
3242 ctx
= g_new(struct nvme_compare_ctx
, 1);
3243 ctx
->data
.bounce
= g_malloc(data_len
);
3247 qemu_iovec_init(&ctx
->data
.iov
, 1);
3248 qemu_iovec_add(&ctx
->data
.iov
, ctx
->data
.bounce
, data_len
);
3250 block_acct_start(blk_get_stats(blk
), &req
->acct
, data_len
,
3252 req
->aiocb
= blk_aio_preadv(blk
, offset
, &ctx
->data
.iov
, 0,
3253 nvme_compare_data_cb
, req
);
3255 return NVME_NO_COMPLETE
;
3258 typedef struct NvmeFlushAIOCB
{
3269 static void nvme_flush_cancel(BlockAIOCB
*acb
)
3271 NvmeFlushAIOCB
*iocb
= container_of(acb
, NvmeFlushAIOCB
, common
);
3273 iocb
->ret
= -ECANCELED
;
3276 blk_aio_cancel_async(iocb
->aiocb
);
3281 static const AIOCBInfo nvme_flush_aiocb_info
= {
3282 .aiocb_size
= sizeof(NvmeFlushAIOCB
),
3283 .cancel_async
= nvme_flush_cancel
,
3284 .get_aio_context
= nvme_get_aio_context
,
3287 static void nvme_do_flush(NvmeFlushAIOCB
*iocb
);
3289 static void nvme_flush_ns_cb(void *opaque
, int ret
)
3291 NvmeFlushAIOCB
*iocb
= opaque
;
3292 NvmeNamespace
*ns
= iocb
->ns
;
3297 } else if (iocb
->ret
< 0) {
3302 trace_pci_nvme_flush_ns(iocb
->nsid
);
3305 iocb
->aiocb
= blk_aio_flush(ns
->blkconf
.blk
, nvme_flush_ns_cb
, iocb
);
3310 nvme_do_flush(iocb
);
3313 static void nvme_do_flush(NvmeFlushAIOCB
*iocb
)
3315 NvmeRequest
*req
= iocb
->req
;
3316 NvmeCtrl
*n
= nvme_ctrl(req
);
3319 if (iocb
->ret
< 0) {
3323 if (iocb
->broadcast
) {
3324 for (i
= iocb
->nsid
+ 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
3325 iocb
->ns
= nvme_ns(n
, i
);
3337 nvme_flush_ns_cb(iocb
, 0);
3341 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
3342 qemu_aio_unref(iocb
);
3345 static uint16_t nvme_flush(NvmeCtrl
*n
, NvmeRequest
*req
)
3347 NvmeFlushAIOCB
*iocb
;
3348 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
3351 iocb
= qemu_aio_get(&nvme_flush_aiocb_info
, NULL
, nvme_misc_cb
, req
);
3357 iocb
->broadcast
= (nsid
== NVME_NSID_BROADCAST
);
3359 if (!iocb
->broadcast
) {
3360 if (!nvme_nsid_valid(n
, nsid
)) {
3361 status
= NVME_INVALID_NSID
| NVME_DNR
;
3365 iocb
->ns
= nvme_ns(n
, nsid
);
3367 status
= NVME_INVALID_FIELD
| NVME_DNR
;
3374 req
->aiocb
= &iocb
->common
;
3375 nvme_do_flush(iocb
);
3377 return NVME_NO_COMPLETE
;
3380 qemu_aio_unref(iocb
);
3385 static uint16_t nvme_read(NvmeCtrl
*n
, NvmeRequest
*req
)
3387 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
3388 NvmeNamespace
*ns
= req
->ns
;
3389 uint64_t slba
= le64_to_cpu(rw
->slba
);
3390 uint32_t nlb
= (uint32_t)le16_to_cpu(rw
->nlb
) + 1;
3391 uint8_t prinfo
= NVME_RW_PRINFO(le16_to_cpu(rw
->control
));
3392 uint64_t data_size
= nvme_l2b(ns
, nlb
);
3393 uint64_t mapped_size
= data_size
;
3394 uint64_t data_offset
;
3395 BlockBackend
*blk
= ns
->blkconf
.blk
;
3398 if (nvme_ns_ext(ns
)) {
3399 mapped_size
+= nvme_m2b(ns
, nlb
);
3401 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3402 bool pract
= prinfo
& NVME_PRINFO_PRACT
;
3404 if (pract
&& ns
->lbaf
.ms
== nvme_pi_tuple_size(ns
)) {
3405 mapped_size
= data_size
;
3410 trace_pci_nvme_read(nvme_cid(req
), nvme_nsid(ns
), nlb
, mapped_size
, slba
);
3412 status
= nvme_check_mdts(n
, mapped_size
);
3417 status
= nvme_check_bounds(ns
, slba
, nlb
);
3422 if (ns
->params
.zoned
) {
3423 status
= nvme_check_zone_read(ns
, slba
, nlb
);
3425 trace_pci_nvme_err_zone_read_not_ok(slba
, nlb
, status
);
3430 if (NVME_ERR_REC_DULBE(ns
->features
.err_rec
)) {
3431 status
= nvme_check_dulbe(ns
, slba
, nlb
);
3437 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3438 return nvme_dif_rw(n
, req
);
3441 status
= nvme_map_data(n
, nlb
, req
);
3446 data_offset
= nvme_l2b(ns
, slba
);
3448 block_acct_start(blk_get_stats(blk
), &req
->acct
, data_size
,
3450 nvme_blk_read(blk
, data_offset
, BDRV_SECTOR_SIZE
, nvme_rw_cb
, req
);
3451 return NVME_NO_COMPLETE
;
3454 block_acct_invalid(blk_get_stats(blk
), BLOCK_ACCT_READ
);
3455 return status
| NVME_DNR
;
3458 static void nvme_do_write_fdp(NvmeCtrl
*n
, NvmeRequest
*req
, uint64_t slba
,
3461 NvmeNamespace
*ns
= req
->ns
;
3462 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
3463 uint64_t data_size
= nvme_l2b(ns
, nlb
);
3464 uint32_t dw12
= le32_to_cpu(req
->cmd
.cdw12
);
3465 uint8_t dtype
= (dw12
>> 20) & 0xf;
3466 uint16_t pid
= le16_to_cpu(rw
->dspec
);
3467 uint16_t ph
, rg
, ruhid
;
3468 NvmeReclaimUnit
*ru
;
3470 if (dtype
!= NVME_DIRECTIVE_DATA_PLACEMENT
||
3471 !nvme_parse_pid(ns
, pid
, &ph
, &rg
)) {
3476 ruhid
= ns
->fdp
.phs
[ph
];
3477 ru
= &ns
->endgrp
->fdp
.ruhs
[ruhid
].rus
[rg
];
3479 nvme_fdp_stat_inc(&ns
->endgrp
->fdp
.hbmw
, data_size
);
3480 nvme_fdp_stat_inc(&ns
->endgrp
->fdp
.mbmw
, data_size
);
3483 if (nlb
< ru
->ruamw
) {
3489 nvme_update_ruh(n
, ns
, pid
);
3493 static uint16_t nvme_do_write(NvmeCtrl
*n
, NvmeRequest
*req
, bool append
,
3496 NvmeRwCmd
*rw
= (NvmeRwCmd
*)&req
->cmd
;
3497 NvmeNamespace
*ns
= req
->ns
;
3498 uint64_t slba
= le64_to_cpu(rw
->slba
);
3499 uint32_t nlb
= (uint32_t)le16_to_cpu(rw
->nlb
) + 1;
3500 uint16_t ctrl
= le16_to_cpu(rw
->control
);
3501 uint8_t prinfo
= NVME_RW_PRINFO(ctrl
);
3502 uint64_t data_size
= nvme_l2b(ns
, nlb
);
3503 uint64_t mapped_size
= data_size
;
3504 uint64_t data_offset
;
3506 NvmeZonedResult
*res
= (NvmeZonedResult
*)&req
->cqe
;
3507 BlockBackend
*blk
= ns
->blkconf
.blk
;
3510 if (nvme_ns_ext(ns
)) {
3511 mapped_size
+= nvme_m2b(ns
, nlb
);
3513 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3514 bool pract
= prinfo
& NVME_PRINFO_PRACT
;
3516 if (pract
&& ns
->lbaf
.ms
== nvme_pi_tuple_size(ns
)) {
3517 mapped_size
-= nvme_m2b(ns
, nlb
);
3522 trace_pci_nvme_write(nvme_cid(req
), nvme_io_opc_str(rw
->opcode
),
3523 nvme_nsid(ns
), nlb
, mapped_size
, slba
);
3526 status
= nvme_check_mdts(n
, mapped_size
);
3532 status
= nvme_check_bounds(ns
, slba
, nlb
);
3537 if (ns
->params
.zoned
) {
3538 zone
= nvme_get_zone_by_slba(ns
, slba
);
3542 bool piremap
= !!(ctrl
& NVME_RW_PIREMAP
);
3544 if (unlikely(zone
->d
.za
& NVME_ZA_ZRWA_VALID
)) {
3545 return NVME_INVALID_ZONE_OP
| NVME_DNR
;
3548 if (unlikely(slba
!= zone
->d
.zslba
)) {
3549 trace_pci_nvme_err_append_not_at_start(slba
, zone
->d
.zslba
);
3550 status
= NVME_INVALID_FIELD
;
3554 if (n
->params
.zasl
&&
3555 data_size
> (uint64_t)n
->page_size
<< n
->params
.zasl
) {
3556 trace_pci_nvme_err_zasl(data_size
);
3557 return NVME_INVALID_FIELD
| NVME_DNR
;
3561 rw
->slba
= cpu_to_le64(slba
);
3562 res
->slba
= cpu_to_le64(slba
);
3564 switch (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3565 case NVME_ID_NS_DPS_TYPE_1
:
3567 return NVME_INVALID_PROT_INFO
| NVME_DNR
;
3572 case NVME_ID_NS_DPS_TYPE_2
:
3574 uint32_t reftag
= le32_to_cpu(rw
->reftag
);
3575 rw
->reftag
= cpu_to_le32(reftag
+ (slba
- zone
->d
.zslba
));
3580 case NVME_ID_NS_DPS_TYPE_3
:
3582 return NVME_INVALID_PROT_INFO
| NVME_DNR
;
3589 status
= nvme_check_zone_write(ns
, zone
, slba
, nlb
);
3594 status
= nvme_zrm_auto(n
, ns
, zone
);
3599 if (!(zone
->d
.za
& NVME_ZA_ZRWA_VALID
)) {
3602 } else if (ns
->endgrp
&& ns
->endgrp
->fdp
.enabled
) {
3603 nvme_do_write_fdp(n
, req
, slba
, nlb
);
3606 data_offset
= nvme_l2b(ns
, slba
);
3608 if (NVME_ID_NS_DPS_TYPE(ns
->id_ns
.dps
)) {
3609 return nvme_dif_rw(n
, req
);
3613 status
= nvme_map_data(n
, nlb
, req
);
3618 block_acct_start(blk_get_stats(blk
), &req
->acct
, data_size
,
3620 nvme_blk_write(blk
, data_offset
, BDRV_SECTOR_SIZE
, nvme_rw_cb
, req
);
3622 req
->aiocb
= blk_aio_pwrite_zeroes(blk
, data_offset
, data_size
,
3623 BDRV_REQ_MAY_UNMAP
, nvme_rw_cb
,
3627 return NVME_NO_COMPLETE
;
3630 block_acct_invalid(blk_get_stats(blk
), BLOCK_ACCT_WRITE
);
3631 return status
| NVME_DNR
;
3634 static inline uint16_t nvme_write(NvmeCtrl
*n
, NvmeRequest
*req
)
3636 return nvme_do_write(n
, req
, false, false);
3639 static inline uint16_t nvme_write_zeroes(NvmeCtrl
*n
, NvmeRequest
*req
)
3641 return nvme_do_write(n
, req
, false, true);
3644 static inline uint16_t nvme_zone_append(NvmeCtrl
*n
, NvmeRequest
*req
)
3646 return nvme_do_write(n
, req
, true, false);
3649 static uint16_t nvme_get_mgmt_zone_slba_idx(NvmeNamespace
*ns
, NvmeCmd
*c
,
3650 uint64_t *slba
, uint32_t *zone_idx
)
3652 uint32_t dw10
= le32_to_cpu(c
->cdw10
);
3653 uint32_t dw11
= le32_to_cpu(c
->cdw11
);
3655 if (!ns
->params
.zoned
) {
3656 trace_pci_nvme_err_invalid_opc(c
->opcode
);
3657 return NVME_INVALID_OPCODE
| NVME_DNR
;
3660 *slba
= ((uint64_t)dw11
) << 32 | dw10
;
3661 if (unlikely(*slba
>= ns
->id_ns
.nsze
)) {
3662 trace_pci_nvme_err_invalid_lba_range(*slba
, 0, ns
->id_ns
.nsze
);
3664 return NVME_LBA_RANGE
| NVME_DNR
;
3667 *zone_idx
= nvme_zone_idx(ns
, *slba
);
3668 assert(*zone_idx
< ns
->num_zones
);
3670 return NVME_SUCCESS
;
3673 typedef uint16_t (*op_handler_t
)(NvmeNamespace
*, NvmeZone
*, NvmeZoneState
,
3676 enum NvmeZoneProcessingMask
{
3677 NVME_PROC_CURRENT_ZONE
= 0,
3678 NVME_PROC_OPENED_ZONES
= 1 << 0,
3679 NVME_PROC_CLOSED_ZONES
= 1 << 1,
3680 NVME_PROC_READ_ONLY_ZONES
= 1 << 2,
3681 NVME_PROC_FULL_ZONES
= 1 << 3,
3684 static uint16_t nvme_open_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3685 NvmeZoneState state
, NvmeRequest
*req
)
3687 NvmeZoneSendCmd
*cmd
= (NvmeZoneSendCmd
*)&req
->cmd
;
3690 if (cmd
->zsflags
& NVME_ZSFLAG_ZRWA_ALLOC
) {
3691 uint16_t ozcs
= le16_to_cpu(ns
->id_ns_zoned
->ozcs
);
3693 if (!(ozcs
& NVME_ID_NS_ZONED_OZCS_ZRWASUP
)) {
3694 return NVME_INVALID_ZONE_OP
| NVME_DNR
;
3697 if (zone
->w_ptr
% ns
->zns
.zrwafg
) {
3698 return NVME_NOZRWA
| NVME_DNR
;
3701 flags
= NVME_ZRM_ZRWA
;
3704 return nvme_zrm_open_flags(nvme_ctrl(req
), ns
, zone
, flags
);
3707 static uint16_t nvme_close_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3708 NvmeZoneState state
, NvmeRequest
*req
)
3710 return nvme_zrm_close(ns
, zone
);
3713 static uint16_t nvme_finish_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3714 NvmeZoneState state
, NvmeRequest
*req
)
3716 return nvme_zrm_finish(ns
, zone
);
3719 static uint16_t nvme_offline_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3720 NvmeZoneState state
, NvmeRequest
*req
)
3723 case NVME_ZONE_STATE_READ_ONLY
:
3724 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_OFFLINE
);
3726 case NVME_ZONE_STATE_OFFLINE
:
3727 return NVME_SUCCESS
;
3729 return NVME_ZONE_INVAL_TRANSITION
;
3733 static uint16_t nvme_set_zd_ext(NvmeNamespace
*ns
, NvmeZone
*zone
)
3736 uint8_t state
= nvme_get_zone_state(zone
);
3738 if (state
== NVME_ZONE_STATE_EMPTY
) {
3739 status
= nvme_aor_check(ns
, 1, 0);
3743 nvme_aor_inc_active(ns
);
3744 zone
->d
.za
|= NVME_ZA_ZD_EXT_VALID
;
3745 nvme_assign_zone_state(ns
, zone
, NVME_ZONE_STATE_CLOSED
);
3746 return NVME_SUCCESS
;
3749 return NVME_ZONE_INVAL_TRANSITION
;
3752 static uint16_t nvme_bulk_proc_zone(NvmeNamespace
*ns
, NvmeZone
*zone
,
3753 enum NvmeZoneProcessingMask proc_mask
,
3754 op_handler_t op_hndlr
, NvmeRequest
*req
)
3756 uint16_t status
= NVME_SUCCESS
;
3757 NvmeZoneState zs
= nvme_get_zone_state(zone
);
3761 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
3762 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
3763 proc_zone
= proc_mask
& NVME_PROC_OPENED_ZONES
;
3765 case NVME_ZONE_STATE_CLOSED
:
3766 proc_zone
= proc_mask
& NVME_PROC_CLOSED_ZONES
;
3768 case NVME_ZONE_STATE_READ_ONLY
:
3769 proc_zone
= proc_mask
& NVME_PROC_READ_ONLY_ZONES
;
3771 case NVME_ZONE_STATE_FULL
:
3772 proc_zone
= proc_mask
& NVME_PROC_FULL_ZONES
;
3779 status
= op_hndlr(ns
, zone
, zs
, req
);
3785 static uint16_t nvme_do_zone_op(NvmeNamespace
*ns
, NvmeZone
*zone
,
3786 enum NvmeZoneProcessingMask proc_mask
,
3787 op_handler_t op_hndlr
, NvmeRequest
*req
)
3790 uint16_t status
= NVME_SUCCESS
;
3794 status
= op_hndlr(ns
, zone
, nvme_get_zone_state(zone
), req
);
3796 if (proc_mask
& NVME_PROC_CLOSED_ZONES
) {
3797 QTAILQ_FOREACH_SAFE(zone
, &ns
->closed_zones
, entry
, next
) {
3798 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3800 if (status
&& status
!= NVME_NO_COMPLETE
) {
3805 if (proc_mask
& NVME_PROC_OPENED_ZONES
) {
3806 QTAILQ_FOREACH_SAFE(zone
, &ns
->imp_open_zones
, entry
, next
) {
3807 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3809 if (status
&& status
!= NVME_NO_COMPLETE
) {
3814 QTAILQ_FOREACH_SAFE(zone
, &ns
->exp_open_zones
, entry
, next
) {
3815 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3817 if (status
&& status
!= NVME_NO_COMPLETE
) {
3822 if (proc_mask
& NVME_PROC_FULL_ZONES
) {
3823 QTAILQ_FOREACH_SAFE(zone
, &ns
->full_zones
, entry
, next
) {
3824 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3826 if (status
&& status
!= NVME_NO_COMPLETE
) {
3832 if (proc_mask
& NVME_PROC_READ_ONLY_ZONES
) {
3833 for (i
= 0; i
< ns
->num_zones
; i
++, zone
++) {
3834 status
= nvme_bulk_proc_zone(ns
, zone
, proc_mask
, op_hndlr
,
3836 if (status
&& status
!= NVME_NO_COMPLETE
) {
3847 typedef struct NvmeZoneResetAIOCB
{
3856 } NvmeZoneResetAIOCB
;
3858 static void nvme_zone_reset_cancel(BlockAIOCB
*aiocb
)
3860 NvmeZoneResetAIOCB
*iocb
= container_of(aiocb
, NvmeZoneResetAIOCB
, common
);
3861 NvmeRequest
*req
= iocb
->req
;
3862 NvmeNamespace
*ns
= req
->ns
;
3864 iocb
->idx
= ns
->num_zones
;
3866 iocb
->ret
= -ECANCELED
;
3869 blk_aio_cancel_async(iocb
->aiocb
);
3874 static const AIOCBInfo nvme_zone_reset_aiocb_info
= {
3875 .aiocb_size
= sizeof(NvmeZoneResetAIOCB
),
3876 .cancel_async
= nvme_zone_reset_cancel
,
3879 static void nvme_zone_reset_cb(void *opaque
, int ret
);
3881 static void nvme_zone_reset_epilogue_cb(void *opaque
, int ret
)
3883 NvmeZoneResetAIOCB
*iocb
= opaque
;
3884 NvmeRequest
*req
= iocb
->req
;
3885 NvmeNamespace
*ns
= req
->ns
;
3889 if (ret
< 0 || iocb
->ret
< 0 || !ns
->lbaf
.ms
) {
3893 moff
= nvme_moff(ns
, iocb
->zone
->d
.zslba
);
3894 count
= nvme_m2b(ns
, ns
->zone_size
);
3896 iocb
->aiocb
= blk_aio_pwrite_zeroes(ns
->blkconf
.blk
, moff
, count
,
3898 nvme_zone_reset_cb
, iocb
);
3902 nvme_zone_reset_cb(iocb
, ret
);
3905 static void nvme_zone_reset_cb(void *opaque
, int ret
)
3907 NvmeZoneResetAIOCB
*iocb
= opaque
;
3908 NvmeRequest
*req
= iocb
->req
;
3909 NvmeNamespace
*ns
= req
->ns
;
3911 if (iocb
->ret
< 0) {
3913 } else if (ret
< 0) {
3919 nvme_zrm_reset(ns
, iocb
->zone
);
3926 while (iocb
->idx
< ns
->num_zones
) {
3927 NvmeZone
*zone
= &ns
->zone_array
[iocb
->idx
++];
3929 switch (nvme_get_zone_state(zone
)) {
3930 case NVME_ZONE_STATE_EMPTY
:
3937 case NVME_ZONE_STATE_EXPLICITLY_OPEN
:
3938 case NVME_ZONE_STATE_IMPLICITLY_OPEN
:
3939 case NVME_ZONE_STATE_CLOSED
:
3940 case NVME_ZONE_STATE_FULL
:
3948 trace_pci_nvme_zns_zone_reset(zone
->d
.zslba
);
3950 iocb
->aiocb
= blk_aio_pwrite_zeroes(ns
->blkconf
.blk
,
3951 nvme_l2b(ns
, zone
->d
.zslba
),
3952 nvme_l2b(ns
, ns
->zone_size
),
3954 nvme_zone_reset_epilogue_cb
,
3962 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
3963 qemu_aio_unref(iocb
);
3966 static uint16_t nvme_zone_mgmt_send_zrwa_flush(NvmeCtrl
*n
, NvmeZone
*zone
,
3967 uint64_t elba
, NvmeRequest
*req
)
3969 NvmeNamespace
*ns
= req
->ns
;
3970 uint16_t ozcs
= le16_to_cpu(ns
->id_ns_zoned
->ozcs
);
3971 uint64_t wp
= zone
->d
.wp
;
3972 uint32_t nlb
= elba
- wp
+ 1;
3976 if (!(ozcs
& NVME_ID_NS_ZONED_OZCS_ZRWASUP
)) {
3977 return NVME_INVALID_ZONE_OP
| NVME_DNR
;
3980 if (!(zone
->d
.za
& NVME_ZA_ZRWA_VALID
)) {
3981 return NVME_INVALID_FIELD
| NVME_DNR
;
3984 if (elba
< wp
|| elba
> wp
+ ns
->zns
.zrwas
) {
3985 return NVME_ZONE_BOUNDARY_ERROR
| NVME_DNR
;
3988 if (nlb
% ns
->zns
.zrwafg
) {
3989 return NVME_INVALID_FIELD
| NVME_DNR
;
3992 status
= nvme_zrm_auto(n
, ns
, zone
);
3999 nvme_advance_zone_wp(ns
, zone
, nlb
);
4001 return NVME_SUCCESS
;
4004 static uint16_t nvme_zone_mgmt_send(NvmeCtrl
*n
, NvmeRequest
*req
)
4006 NvmeZoneSendCmd
*cmd
= (NvmeZoneSendCmd
*)&req
->cmd
;
4007 NvmeNamespace
*ns
= req
->ns
;
4009 NvmeZoneResetAIOCB
*iocb
;
4012 uint32_t zone_idx
= 0;
4014 uint8_t action
= cmd
->zsa
;
4016 enum NvmeZoneProcessingMask proc_mask
= NVME_PROC_CURRENT_ZONE
;
4018 all
= cmd
->zsflags
& NVME_ZSFLAG_SELECT_ALL
;
4020 req
->status
= NVME_SUCCESS
;
4023 status
= nvme_get_mgmt_zone_slba_idx(ns
, &req
->cmd
, &slba
, &zone_idx
);
4029 zone
= &ns
->zone_array
[zone_idx
];
4030 if (slba
!= zone
->d
.zslba
&& action
!= NVME_ZONE_ACTION_ZRWA_FLUSH
) {
4031 trace_pci_nvme_err_unaligned_zone_cmd(action
, slba
, zone
->d
.zslba
);
4032 return NVME_INVALID_FIELD
| NVME_DNR
;
4037 case NVME_ZONE_ACTION_OPEN
:
4039 proc_mask
= NVME_PROC_CLOSED_ZONES
;
4041 trace_pci_nvme_open_zone(slba
, zone_idx
, all
);
4042 status
= nvme_do_zone_op(ns
, zone
, proc_mask
, nvme_open_zone
, req
);
4045 case NVME_ZONE_ACTION_CLOSE
:
4047 proc_mask
= NVME_PROC_OPENED_ZONES
;
4049 trace_pci_nvme_close_zone(slba
, zone_idx
, all
);
4050 status
= nvme_do_zone_op(ns
, zone
, proc_mask
, nvme_close_zone
, req
);
4053 case NVME_ZONE_ACTION_FINISH
:
4055 proc_mask
= NVME_PROC_OPENED_ZONES
| NVME_PROC_CLOSED_ZONES
;
4057 trace_pci_nvme_finish_zone(slba
, zone_idx
, all
);
4058 status
= nvme_do_zone_op(ns
, zone
, proc_mask
, nvme_finish_zone
, req
);
4061 case NVME_ZONE_ACTION_RESET
:
4062 trace_pci_nvme_reset_zone(slba
, zone_idx
, all
);
4064 iocb
= blk_aio_get(&nvme_zone_reset_aiocb_info
, ns
->blkconf
.blk
,
4070 iocb
->idx
= zone_idx
;
4073 req
->aiocb
= &iocb
->common
;
4074 nvme_zone_reset_cb(iocb
, 0);
4076 return NVME_NO_COMPLETE
;
4078 case NVME_ZONE_ACTION_OFFLINE
:
4080 proc_mask
= NVME_PROC_READ_ONLY_ZONES
;
4082 trace_pci_nvme_offline_zone(slba
, zone_idx
, all
);
4083 status
= nvme_do_zone_op(ns
, zone
, proc_mask
, nvme_offline_zone
, req
);
4086 case NVME_ZONE_ACTION_SET_ZD_EXT
:
4087 trace_pci_nvme_set_descriptor_extension(slba
, zone_idx
);
4088 if (all
|| !ns
->params
.zd_extension_size
) {
4089 return NVME_INVALID_FIELD
| NVME_DNR
;
4091 zd_ext
= nvme_get_zd_extension(ns
, zone_idx
);
4092 status
= nvme_h2c(n
, zd_ext
, ns
->params
.zd_extension_size
, req
);
4094 trace_pci_nvme_err_zd_extension_map_error(zone_idx
);
4098 status
= nvme_set_zd_ext(ns
, zone
);
4099 if (status
== NVME_SUCCESS
) {
4100 trace_pci_nvme_zd_extension_set(zone_idx
);
4105 case NVME_ZONE_ACTION_ZRWA_FLUSH
:
4107 return NVME_INVALID_FIELD
| NVME_DNR
;
4110 return nvme_zone_mgmt_send_zrwa_flush(n
, zone
, slba
, req
);
4113 trace_pci_nvme_err_invalid_mgmt_action(action
);
4114 status
= NVME_INVALID_FIELD
;
4117 if (status
== NVME_ZONE_INVAL_TRANSITION
) {
4118 trace_pci_nvme_err_invalid_zone_state_transition(action
, slba
,
4128 static bool nvme_zone_matches_filter(uint32_t zafs
, NvmeZone
*zl
)
4130 NvmeZoneState zs
= nvme_get_zone_state(zl
);
4133 case NVME_ZONE_REPORT_ALL
:
4135 case NVME_ZONE_REPORT_EMPTY
:
4136 return zs
== NVME_ZONE_STATE_EMPTY
;
4137 case NVME_ZONE_REPORT_IMPLICITLY_OPEN
:
4138 return zs
== NVME_ZONE_STATE_IMPLICITLY_OPEN
;
4139 case NVME_ZONE_REPORT_EXPLICITLY_OPEN
:
4140 return zs
== NVME_ZONE_STATE_EXPLICITLY_OPEN
;
4141 case NVME_ZONE_REPORT_CLOSED
:
4142 return zs
== NVME_ZONE_STATE_CLOSED
;
4143 case NVME_ZONE_REPORT_FULL
:
4144 return zs
== NVME_ZONE_STATE_FULL
;
4145 case NVME_ZONE_REPORT_READ_ONLY
:
4146 return zs
== NVME_ZONE_STATE_READ_ONLY
;
4147 case NVME_ZONE_REPORT_OFFLINE
:
4148 return zs
== NVME_ZONE_STATE_OFFLINE
;
4154 static uint16_t nvme_zone_mgmt_recv(NvmeCtrl
*n
, NvmeRequest
*req
)
4156 NvmeCmd
*cmd
= (NvmeCmd
*)&req
->cmd
;
4157 NvmeNamespace
*ns
= req
->ns
;
4158 /* cdw12 is zero-based number of dwords to return. Convert to bytes */
4159 uint32_t data_size
= (le32_to_cpu(cmd
->cdw12
) + 1) << 2;
4160 uint32_t dw13
= le32_to_cpu(cmd
->cdw13
);
4161 uint32_t zone_idx
, zra
, zrasf
, partial
;
4162 uint64_t max_zones
, nr_zones
= 0;
4167 NvmeZoneReportHeader
*header
;
4169 size_t zone_entry_sz
;
4172 req
->status
= NVME_SUCCESS
;
4174 status
= nvme_get_mgmt_zone_slba_idx(ns
, cmd
, &slba
, &zone_idx
);
4180 if (zra
!= NVME_ZONE_REPORT
&& zra
!= NVME_ZONE_REPORT_EXTENDED
) {
4181 return NVME_INVALID_FIELD
| NVME_DNR
;
4183 if (zra
== NVME_ZONE_REPORT_EXTENDED
&& !ns
->params
.zd_extension_size
) {
4184 return NVME_INVALID_FIELD
| NVME_DNR
;
4187 zrasf
= (dw13
>> 8) & 0xff;
4188 if (zrasf
> NVME_ZONE_REPORT_OFFLINE
) {
4189 return NVME_INVALID_FIELD
| NVME_DNR
;
4192 if (data_size
< sizeof(NvmeZoneReportHeader
)) {
4193 return NVME_INVALID_FIELD
| NVME_DNR
;
4196 status
= nvme_check_mdts(n
, data_size
);
4201 partial
= (dw13
>> 16) & 0x01;
4203 zone_entry_sz
= sizeof(NvmeZoneDescr
);
4204 if (zra
== NVME_ZONE_REPORT_EXTENDED
) {
4205 zone_entry_sz
+= ns
->params
.zd_extension_size
;
4208 max_zones
= (data_size
- sizeof(NvmeZoneReportHeader
)) / zone_entry_sz
;
4209 buf
= g_malloc0(data_size
);
4211 zone
= &ns
->zone_array
[zone_idx
];
4212 for (i
= zone_idx
; i
< ns
->num_zones
; i
++) {
4213 if (partial
&& nr_zones
>= max_zones
) {
4216 if (nvme_zone_matches_filter(zrasf
, zone
++)) {
4221 header
->nr_zones
= cpu_to_le64(nr_zones
);
4223 buf_p
= buf
+ sizeof(NvmeZoneReportHeader
);
4224 for (; zone_idx
< ns
->num_zones
&& max_zones
> 0; zone_idx
++) {
4225 zone
= &ns
->zone_array
[zone_idx
];
4226 if (nvme_zone_matches_filter(zrasf
, zone
)) {
4228 buf_p
+= sizeof(NvmeZoneDescr
);
4232 z
->zcap
= cpu_to_le64(zone
->d
.zcap
);
4233 z
->zslba
= cpu_to_le64(zone
->d
.zslba
);
4236 if (nvme_wp_is_valid(zone
)) {
4237 z
->wp
= cpu_to_le64(zone
->d
.wp
);
4239 z
->wp
= cpu_to_le64(~0ULL);
4242 if (zra
== NVME_ZONE_REPORT_EXTENDED
) {
4243 if (zone
->d
.za
& NVME_ZA_ZD_EXT_VALID
) {
4244 memcpy(buf_p
, nvme_get_zd_extension(ns
, zone_idx
),
4245 ns
->params
.zd_extension_size
);
4247 buf_p
+= ns
->params
.zd_extension_size
;
4254 status
= nvme_c2h(n
, (uint8_t *)buf
, data_size
, req
);
4261 static uint16_t nvme_io_mgmt_recv_ruhs(NvmeCtrl
*n
, NvmeRequest
*req
,
4264 NvmeNamespace
*ns
= req
->ns
;
4265 NvmeEnduranceGroup
*endgrp
;
4267 NvmeRuhStatusDescr
*ruhsd
;
4268 unsigned int nruhsd
;
4269 uint16_t rg
, ph
, *ruhid
;
4271 g_autofree
uint8_t *buf
= NULL
;
4274 return NVME_INVALID_FIELD
| NVME_DNR
;
4277 if (ns
->params
.nsid
== 0 || ns
->params
.nsid
== 0xffffffff) {
4278 return NVME_INVALID_NSID
| NVME_DNR
;
4281 if (!n
->subsys
->endgrp
.fdp
.enabled
) {
4282 return NVME_FDP_DISABLED
| NVME_DNR
;
4285 endgrp
= ns
->endgrp
;
4287 nruhsd
= ns
->fdp
.nphs
* endgrp
->fdp
.nrg
;
4288 trans_len
= sizeof(NvmeRuhStatus
) + nruhsd
* sizeof(NvmeRuhStatusDescr
);
4289 buf
= g_malloc(trans_len
);
4291 trans_len
= MIN(trans_len
, len
);
4293 hdr
= (NvmeRuhStatus
*)buf
;
4294 ruhsd
= (NvmeRuhStatusDescr
*)(buf
+ sizeof(NvmeRuhStatus
));
4296 hdr
->nruhsd
= cpu_to_le16(nruhsd
);
4298 ruhid
= ns
->fdp
.phs
;
4300 for (ph
= 0; ph
< ns
->fdp
.nphs
; ph
++, ruhid
++) {
4301 NvmeRuHandle
*ruh
= &endgrp
->fdp
.ruhs
[*ruhid
];
4303 for (rg
= 0; rg
< endgrp
->fdp
.nrg
; rg
++, ruhsd
++) {
4304 uint16_t pid
= nvme_make_pid(ns
, rg
, ph
);
4306 ruhsd
->pid
= cpu_to_le16(pid
);
4307 ruhsd
->ruhid
= *ruhid
;
4309 ruhsd
->ruamw
= cpu_to_le64(ruh
->rus
[rg
].ruamw
);
4313 return nvme_c2h(n
, buf
, trans_len
, req
);
4316 static uint16_t nvme_io_mgmt_recv(NvmeCtrl
*n
, NvmeRequest
*req
)
4318 NvmeCmd
*cmd
= &req
->cmd
;
4319 uint32_t cdw10
= le32_to_cpu(cmd
->cdw10
);
4320 uint32_t numd
= le32_to_cpu(cmd
->cdw11
);
4321 uint8_t mo
= (cdw10
& 0xff);
4322 size_t len
= (numd
+ 1) << 2;
4325 case NVME_IOMR_MO_NOP
:
4327 case NVME_IOMR_MO_RUH_STATUS
:
4328 return nvme_io_mgmt_recv_ruhs(n
, req
, len
);
4330 return NVME_INVALID_FIELD
| NVME_DNR
;
4334 static uint16_t nvme_io_mgmt_send_ruh_update(NvmeCtrl
*n
, NvmeRequest
*req
)
4336 NvmeCmd
*cmd
= &req
->cmd
;
4337 NvmeNamespace
*ns
= req
->ns
;
4338 uint32_t cdw10
= le32_to_cpu(cmd
->cdw10
);
4339 uint16_t ret
= NVME_SUCCESS
;
4340 uint32_t npid
= (cdw10
>> 1) + 1;
4342 g_autofree
uint16_t *pids
= NULL
;
4343 uint32_t maxnpid
= n
->subsys
->endgrp
.fdp
.nrg
* n
->subsys
->endgrp
.fdp
.nruh
;
4345 if (unlikely(npid
>= MIN(NVME_FDP_MAXPIDS
, maxnpid
))) {
4346 return NVME_INVALID_FIELD
| NVME_DNR
;
4349 pids
= g_new(uint16_t, npid
);
4351 ret
= nvme_h2c(n
, pids
, npid
* sizeof(uint16_t), req
);
4356 for (; i
< npid
; i
++) {
4357 if (!nvme_update_ruh(n
, ns
, pids
[i
])) {
4358 return NVME_INVALID_FIELD
| NVME_DNR
;
4365 static uint16_t nvme_io_mgmt_send(NvmeCtrl
*n
, NvmeRequest
*req
)
4367 NvmeCmd
*cmd
= &req
->cmd
;
4368 uint32_t cdw10
= le32_to_cpu(cmd
->cdw10
);
4369 uint8_t mo
= (cdw10
& 0xff);
4372 case NVME_IOMS_MO_NOP
:
4374 case NVME_IOMS_MO_RUH_UPDATE
:
4375 return nvme_io_mgmt_send_ruh_update(n
, req
);
4377 return NVME_INVALID_FIELD
| NVME_DNR
;
4381 static uint16_t nvme_io_cmd(NvmeCtrl
*n
, NvmeRequest
*req
)
4384 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
4386 trace_pci_nvme_io_cmd(nvme_cid(req
), nsid
, nvme_sqid(req
),
4387 req
->cmd
.opcode
, nvme_io_opc_str(req
->cmd
.opcode
));
4389 if (!nvme_nsid_valid(n
, nsid
)) {
4390 return NVME_INVALID_NSID
| NVME_DNR
;
4394 * In the base NVM command set, Flush may apply to all namespaces
4395 * (indicated by NSID being set to FFFFFFFFh). But if that feature is used
4396 * along with TP 4056 (Namespace Types), it may be pretty screwed up.
4398 * If NSID is indeed set to FFFFFFFFh, we simply cannot associate the
4399 * opcode with a specific command since we cannot determine a unique I/O
4400 * command set. Opcode 0h could have any other meaning than something
4401 * equivalent to flushing and say it DOES have completely different
4402 * semantics in some other command set - does an NSID of FFFFFFFFh then
4403 * mean "for all namespaces, apply whatever command set specific command
4404 * that uses the 0h opcode?" Or does it mean "for all namespaces, apply
4405 * whatever command that uses the 0h opcode if, and only if, it allows NSID
4408 * Anyway (and luckily), for now, we do not care about this since the
4409 * device only supports namespace types that includes the NVM Flush command
4410 * (NVM and Zoned), so always do an NVM Flush.
4412 if (req
->cmd
.opcode
== NVME_CMD_FLUSH
) {
4413 return nvme_flush(n
, req
);
4416 ns
= nvme_ns(n
, nsid
);
4417 if (unlikely(!ns
)) {
4418 return NVME_INVALID_FIELD
| NVME_DNR
;
4421 if (!(ns
->iocs
[req
->cmd
.opcode
] & NVME_CMD_EFF_CSUPP
)) {
4422 trace_pci_nvme_err_invalid_opc(req
->cmd
.opcode
);
4423 return NVME_INVALID_OPCODE
| NVME_DNR
;
4430 if (NVME_CMD_FLAGS_FUSE(req
->cmd
.flags
)) {
4431 return NVME_INVALID_FIELD
;
4436 switch (req
->cmd
.opcode
) {
4437 case NVME_CMD_WRITE_ZEROES
:
4438 return nvme_write_zeroes(n
, req
);
4439 case NVME_CMD_ZONE_APPEND
:
4440 return nvme_zone_append(n
, req
);
4441 case NVME_CMD_WRITE
:
4442 return nvme_write(n
, req
);
4444 return nvme_read(n
, req
);
4445 case NVME_CMD_COMPARE
:
4446 return nvme_compare(n
, req
);
4448 return nvme_dsm(n
, req
);
4449 case NVME_CMD_VERIFY
:
4450 return nvme_verify(n
, req
);
4452 return nvme_copy(n
, req
);
4453 case NVME_CMD_ZONE_MGMT_SEND
:
4454 return nvme_zone_mgmt_send(n
, req
);
4455 case NVME_CMD_ZONE_MGMT_RECV
:
4456 return nvme_zone_mgmt_recv(n
, req
);
4457 case NVME_CMD_IO_MGMT_RECV
:
4458 return nvme_io_mgmt_recv(n
, req
);
4459 case NVME_CMD_IO_MGMT_SEND
:
4460 return nvme_io_mgmt_send(n
, req
);
4465 return NVME_INVALID_OPCODE
| NVME_DNR
;
4468 static void nvme_cq_notifier(EventNotifier
*e
)
4470 NvmeCQueue
*cq
= container_of(e
, NvmeCQueue
, notifier
);
4471 NvmeCtrl
*n
= cq
->ctrl
;
4473 if (!event_notifier_test_and_clear(e
)) {
4477 nvme_update_cq_head(cq
);
4479 if (cq
->tail
== cq
->head
) {
4480 if (cq
->irq_enabled
) {
4484 nvme_irq_deassert(n
, cq
);
4487 qemu_bh_schedule(cq
->bh
);
4490 static int nvme_init_cq_ioeventfd(NvmeCQueue
*cq
)
4492 NvmeCtrl
*n
= cq
->ctrl
;
4493 uint16_t offset
= (cq
->cqid
<< 3) + (1 << 2);
4496 ret
= event_notifier_init(&cq
->notifier
, 0);
4501 event_notifier_set_handler(&cq
->notifier
, nvme_cq_notifier
);
4502 memory_region_add_eventfd(&n
->iomem
,
4503 0x1000 + offset
, 4, false, 0, &cq
->notifier
);
4508 static void nvme_sq_notifier(EventNotifier
*e
)
4510 NvmeSQueue
*sq
= container_of(e
, NvmeSQueue
, notifier
);
4512 if (!event_notifier_test_and_clear(e
)) {
4516 nvme_process_sq(sq
);
4519 static int nvme_init_sq_ioeventfd(NvmeSQueue
*sq
)
4521 NvmeCtrl
*n
= sq
->ctrl
;
4522 uint16_t offset
= sq
->sqid
<< 3;
4525 ret
= event_notifier_init(&sq
->notifier
, 0);
4530 event_notifier_set_handler(&sq
->notifier
, nvme_sq_notifier
);
4531 memory_region_add_eventfd(&n
->iomem
,
4532 0x1000 + offset
, 4, false, 0, &sq
->notifier
);
4537 static void nvme_free_sq(NvmeSQueue
*sq
, NvmeCtrl
*n
)
4539 uint16_t offset
= sq
->sqid
<< 3;
4541 n
->sq
[sq
->sqid
] = NULL
;
4542 qemu_bh_delete(sq
->bh
);
4543 if (sq
->ioeventfd_enabled
) {
4544 memory_region_del_eventfd(&n
->iomem
,
4545 0x1000 + offset
, 4, false, 0, &sq
->notifier
);
4546 event_notifier_set_handler(&sq
->notifier
, NULL
);
4547 event_notifier_cleanup(&sq
->notifier
);
4555 static uint16_t nvme_del_sq(NvmeCtrl
*n
, NvmeRequest
*req
)
4557 NvmeDeleteQ
*c
= (NvmeDeleteQ
*)&req
->cmd
;
4558 NvmeRequest
*r
, *next
;
4561 uint16_t qid
= le16_to_cpu(c
->qid
);
4563 if (unlikely(!qid
|| nvme_check_sqid(n
, qid
))) {
4564 trace_pci_nvme_err_invalid_del_sq(qid
);
4565 return NVME_INVALID_QID
| NVME_DNR
;
4568 trace_pci_nvme_del_sq(qid
);
4571 while (!QTAILQ_EMPTY(&sq
->out_req_list
)) {
4572 r
= QTAILQ_FIRST(&sq
->out_req_list
);
4574 blk_aio_cancel(r
->aiocb
);
4577 assert(QTAILQ_EMPTY(&sq
->out_req_list
));
4579 if (!nvme_check_cqid(n
, sq
->cqid
)) {
4580 cq
= n
->cq
[sq
->cqid
];
4581 QTAILQ_REMOVE(&cq
->sq_list
, sq
, entry
);
4584 QTAILQ_FOREACH_SAFE(r
, &cq
->req_list
, entry
, next
) {
4586 QTAILQ_REMOVE(&cq
->req_list
, r
, entry
);
4587 QTAILQ_INSERT_TAIL(&sq
->req_list
, r
, entry
);
4592 nvme_free_sq(sq
, n
);
4593 return NVME_SUCCESS
;
4596 static void nvme_init_sq(NvmeSQueue
*sq
, NvmeCtrl
*n
, uint64_t dma_addr
,
4597 uint16_t sqid
, uint16_t cqid
, uint16_t size
)
4603 sq
->dma_addr
= dma_addr
;
4607 sq
->head
= sq
->tail
= 0;
4608 sq
->io_req
= g_new0(NvmeRequest
, sq
->size
);
4610 QTAILQ_INIT(&sq
->req_list
);
4611 QTAILQ_INIT(&sq
->out_req_list
);
4612 for (i
= 0; i
< sq
->size
; i
++) {
4613 sq
->io_req
[i
].sq
= sq
;
4614 QTAILQ_INSERT_TAIL(&(sq
->req_list
), &sq
->io_req
[i
], entry
);
4617 sq
->bh
= qemu_bh_new_guarded(nvme_process_sq
, sq
,
4618 &DEVICE(sq
->ctrl
)->mem_reentrancy_guard
);
4620 if (n
->dbbuf_enabled
) {
4621 sq
->db_addr
= n
->dbbuf_dbs
+ (sqid
<< 3);
4622 sq
->ei_addr
= n
->dbbuf_eis
+ (sqid
<< 3);
4624 if (n
->params
.ioeventfd
&& sq
->sqid
!= 0) {
4625 if (!nvme_init_sq_ioeventfd(sq
)) {
4626 sq
->ioeventfd_enabled
= true;
4631 assert(n
->cq
[cqid
]);
4633 QTAILQ_INSERT_TAIL(&(cq
->sq_list
), sq
, entry
);
4637 static uint16_t nvme_create_sq(NvmeCtrl
*n
, NvmeRequest
*req
)
4640 NvmeCreateSq
*c
= (NvmeCreateSq
*)&req
->cmd
;
4642 uint16_t cqid
= le16_to_cpu(c
->cqid
);
4643 uint16_t sqid
= le16_to_cpu(c
->sqid
);
4644 uint16_t qsize
= le16_to_cpu(c
->qsize
);
4645 uint16_t qflags
= le16_to_cpu(c
->sq_flags
);
4646 uint64_t prp1
= le64_to_cpu(c
->prp1
);
4648 trace_pci_nvme_create_sq(prp1
, sqid
, cqid
, qsize
, qflags
);
4650 if (unlikely(!cqid
|| nvme_check_cqid(n
, cqid
))) {
4651 trace_pci_nvme_err_invalid_create_sq_cqid(cqid
);
4652 return NVME_INVALID_CQID
| NVME_DNR
;
4654 if (unlikely(!sqid
|| sqid
> n
->conf_ioqpairs
|| n
->sq
[sqid
] != NULL
)) {
4655 trace_pci_nvme_err_invalid_create_sq_sqid(sqid
);
4656 return NVME_INVALID_QID
| NVME_DNR
;
4658 if (unlikely(!qsize
|| qsize
> NVME_CAP_MQES(ldq_le_p(&n
->bar
.cap
)))) {
4659 trace_pci_nvme_err_invalid_create_sq_size(qsize
);
4660 return NVME_MAX_QSIZE_EXCEEDED
| NVME_DNR
;
4662 if (unlikely(prp1
& (n
->page_size
- 1))) {
4663 trace_pci_nvme_err_invalid_create_sq_addr(prp1
);
4664 return NVME_INVALID_PRP_OFFSET
| NVME_DNR
;
4666 if (unlikely(!(NVME_SQ_FLAGS_PC(qflags
)))) {
4667 trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags
));
4668 return NVME_INVALID_FIELD
| NVME_DNR
;
4670 sq
= g_malloc0(sizeof(*sq
));
4671 nvme_init_sq(sq
, n
, prp1
, sqid
, cqid
, qsize
+ 1);
4672 return NVME_SUCCESS
;
4676 uint64_t units_read
;
4677 uint64_t units_written
;
4678 uint64_t read_commands
;
4679 uint64_t write_commands
;
4682 static void nvme_set_blk_stats(NvmeNamespace
*ns
, struct nvme_stats
*stats
)
4684 BlockAcctStats
*s
= blk_get_stats(ns
->blkconf
.blk
);
4686 stats
->units_read
+= s
->nr_bytes
[BLOCK_ACCT_READ
];
4687 stats
->units_written
+= s
->nr_bytes
[BLOCK_ACCT_WRITE
];
4688 stats
->read_commands
+= s
->nr_ops
[BLOCK_ACCT_READ
];
4689 stats
->write_commands
+= s
->nr_ops
[BLOCK_ACCT_WRITE
];
4692 static uint16_t nvme_smart_info(NvmeCtrl
*n
, uint8_t rae
, uint32_t buf_len
,
4693 uint64_t off
, NvmeRequest
*req
)
4695 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
4696 struct nvme_stats stats
= { 0 };
4697 NvmeSmartLog smart
= { 0 };
4701 uint64_t u_read
, u_written
;
4703 if (off
>= sizeof(smart
)) {
4704 return NVME_INVALID_FIELD
| NVME_DNR
;
4707 if (nsid
!= 0xffffffff) {
4708 ns
= nvme_ns(n
, nsid
);
4710 return NVME_INVALID_NSID
| NVME_DNR
;
4712 nvme_set_blk_stats(ns
, &stats
);
4716 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
4721 nvme_set_blk_stats(ns
, &stats
);
4725 trans_len
= MIN(sizeof(smart
) - off
, buf_len
);
4726 smart
.critical_warning
= n
->smart_critical_warning
;
4728 u_read
= DIV_ROUND_UP(stats
.units_read
>> BDRV_SECTOR_BITS
, 1000);
4729 u_written
= DIV_ROUND_UP(stats
.units_written
>> BDRV_SECTOR_BITS
, 1000);
4731 smart
.data_units_read
[0] = cpu_to_le64(u_read
);
4732 smart
.data_units_written
[0] = cpu_to_le64(u_written
);
4733 smart
.host_read_commands
[0] = cpu_to_le64(stats
.read_commands
);
4734 smart
.host_write_commands
[0] = cpu_to_le64(stats
.write_commands
);
4736 smart
.temperature
= cpu_to_le16(n
->temperature
);
4738 if ((n
->temperature
>= n
->features
.temp_thresh_hi
) ||
4739 (n
->temperature
<= n
->features
.temp_thresh_low
)) {
4740 smart
.critical_warning
|= NVME_SMART_TEMPERATURE
;
4743 current_ms
= qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
);
4744 smart
.power_on_hours
[0] =
4745 cpu_to_le64((((current_ms
- n
->starttime_ms
) / 1000) / 60) / 60);
4748 nvme_clear_events(n
, NVME_AER_TYPE_SMART
);
4751 return nvme_c2h(n
, (uint8_t *) &smart
+ off
, trans_len
, req
);
4754 static uint16_t nvme_endgrp_info(NvmeCtrl
*n
, uint8_t rae
, uint32_t buf_len
,
4755 uint64_t off
, NvmeRequest
*req
)
4757 uint32_t dw11
= le32_to_cpu(req
->cmd
.cdw11
);
4758 uint16_t endgrpid
= (dw11
>> 16) & 0xffff;
4759 struct nvme_stats stats
= {};
4760 NvmeEndGrpLog info
= {};
4763 if (!n
->subsys
|| endgrpid
!= 0x1) {
4764 return NVME_INVALID_FIELD
| NVME_DNR
;
4767 if (off
>= sizeof(info
)) {
4768 return NVME_INVALID_FIELD
| NVME_DNR
;
4771 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
4772 NvmeNamespace
*ns
= nvme_subsys_ns(n
->subsys
, i
);
4777 nvme_set_blk_stats(ns
, &stats
);
4780 info
.data_units_read
[0] =
4781 cpu_to_le64(DIV_ROUND_UP(stats
.units_read
/ 1000000000, 1000000000));
4782 info
.data_units_written
[0] =
4783 cpu_to_le64(DIV_ROUND_UP(stats
.units_written
/ 1000000000, 1000000000));
4784 info
.media_units_written
[0] =
4785 cpu_to_le64(DIV_ROUND_UP(stats
.units_written
/ 1000000000, 1000000000));
4787 info
.host_read_commands
[0] = cpu_to_le64(stats
.read_commands
);
4788 info
.host_write_commands
[0] = cpu_to_le64(stats
.write_commands
);
4790 buf_len
= MIN(sizeof(info
) - off
, buf_len
);
4792 return nvme_c2h(n
, (uint8_t *)&info
+ off
, buf_len
, req
);
4796 static uint16_t nvme_fw_log_info(NvmeCtrl
*n
, uint32_t buf_len
, uint64_t off
,
4800 NvmeFwSlotInfoLog fw_log
= {
4804 if (off
>= sizeof(fw_log
)) {
4805 return NVME_INVALID_FIELD
| NVME_DNR
;
4808 strpadcpy((char *)&fw_log
.frs1
, sizeof(fw_log
.frs1
), "1.0", ' ');
4809 trans_len
= MIN(sizeof(fw_log
) - off
, buf_len
);
4811 return nvme_c2h(n
, (uint8_t *) &fw_log
+ off
, trans_len
, req
);
4814 static uint16_t nvme_error_info(NvmeCtrl
*n
, uint8_t rae
, uint32_t buf_len
,
4815 uint64_t off
, NvmeRequest
*req
)
4818 NvmeErrorLog errlog
;
4820 if (off
>= sizeof(errlog
)) {
4821 return NVME_INVALID_FIELD
| NVME_DNR
;
4825 nvme_clear_events(n
, NVME_AER_TYPE_ERROR
);
4828 memset(&errlog
, 0x0, sizeof(errlog
));
4829 trans_len
= MIN(sizeof(errlog
) - off
, buf_len
);
4831 return nvme_c2h(n
, (uint8_t *)&errlog
, trans_len
, req
);
4834 static uint16_t nvme_changed_nslist(NvmeCtrl
*n
, uint8_t rae
, uint32_t buf_len
,
4835 uint64_t off
, NvmeRequest
*req
)
4837 uint32_t nslist
[1024];
4842 if (off
>= sizeof(nslist
)) {
4843 trace_pci_nvme_err_invalid_log_page_offset(off
, sizeof(nslist
));
4844 return NVME_INVALID_FIELD
| NVME_DNR
;
4847 memset(nslist
, 0x0, sizeof(nslist
));
4848 trans_len
= MIN(sizeof(nslist
) - off
, buf_len
);
4850 while ((nsid
= find_first_bit(n
->changed_nsids
, NVME_CHANGED_NSID_SIZE
)) !=
4851 NVME_CHANGED_NSID_SIZE
) {
4853 * If more than 1024 namespaces, the first entry in the log page should
4854 * be set to FFFFFFFFh and the others to 0 as spec.
4856 if (i
== ARRAY_SIZE(nslist
)) {
4857 memset(nslist
, 0x0, sizeof(nslist
));
4858 nslist
[0] = 0xffffffff;
4863 clear_bit(nsid
, n
->changed_nsids
);
4867 * Remove all the remaining list entries in case returns directly due to
4868 * more than 1024 namespaces.
4870 if (nslist
[0] == 0xffffffff) {
4871 bitmap_zero(n
->changed_nsids
, NVME_CHANGED_NSID_SIZE
);
4875 nvme_clear_events(n
, NVME_AER_TYPE_NOTICE
);
4878 return nvme_c2h(n
, ((uint8_t *)nslist
) + off
, trans_len
, req
);
4881 static uint16_t nvme_cmd_effects(NvmeCtrl
*n
, uint8_t csi
, uint32_t buf_len
,
4882 uint64_t off
, NvmeRequest
*req
)
4884 NvmeEffectsLog log
= {};
4885 const uint32_t *src_iocs
= NULL
;
4888 if (off
>= sizeof(log
)) {
4889 trace_pci_nvme_err_invalid_log_page_offset(off
, sizeof(log
));
4890 return NVME_INVALID_FIELD
| NVME_DNR
;
4893 switch (NVME_CC_CSS(ldl_le_p(&n
->bar
.cc
))) {
4894 case NVME_CC_CSS_NVM
:
4895 src_iocs
= nvme_cse_iocs_nvm
;
4897 case NVME_CC_CSS_ADMIN_ONLY
:
4899 case NVME_CC_CSS_CSI
:
4902 src_iocs
= nvme_cse_iocs_nvm
;
4904 case NVME_CSI_ZONED
:
4905 src_iocs
= nvme_cse_iocs_zoned
;
4910 memcpy(log
.acs
, nvme_cse_acs
, sizeof(nvme_cse_acs
));
4913 memcpy(log
.iocs
, src_iocs
, sizeof(log
.iocs
));
4916 trans_len
= MIN(sizeof(log
) - off
, buf_len
);
4918 return nvme_c2h(n
, ((uint8_t *)&log
) + off
, trans_len
, req
);
4921 static size_t sizeof_fdp_conf_descr(size_t nruh
, size_t vss
)
4923 size_t entry_siz
= sizeof(NvmeFdpDescrHdr
) + nruh
* sizeof(NvmeRuhDescr
)
4925 return ROUND_UP(entry_siz
, 8);
4928 static uint16_t nvme_fdp_confs(NvmeCtrl
*n
, uint32_t endgrpid
, uint32_t buf_len
,
4929 uint64_t off
, NvmeRequest
*req
)
4931 uint32_t log_size
, trans_len
;
4932 g_autofree
uint8_t *buf
= NULL
;
4933 NvmeFdpDescrHdr
*hdr
;
4935 NvmeEnduranceGroup
*endgrp
;
4936 NvmeFdpConfsHdr
*log
;
4937 size_t nruh
, fdp_descr_size
;
4940 if (endgrpid
!= 1 || !n
->subsys
) {
4941 return NVME_INVALID_FIELD
| NVME_DNR
;
4944 endgrp
= &n
->subsys
->endgrp
;
4946 if (endgrp
->fdp
.enabled
) {
4947 nruh
= endgrp
->fdp
.nruh
;
4952 fdp_descr_size
= sizeof_fdp_conf_descr(nruh
, FDPVSS
);
4953 log_size
= sizeof(NvmeFdpConfsHdr
) + fdp_descr_size
;
4955 if (off
>= log_size
) {
4956 return NVME_INVALID_FIELD
| NVME_DNR
;
4959 trans_len
= MIN(log_size
- off
, buf_len
);
4961 buf
= g_malloc0(log_size
);
4962 log
= (NvmeFdpConfsHdr
*)buf
;
4963 hdr
= (NvmeFdpDescrHdr
*)(log
+ 1);
4964 ruhd
= (NvmeRuhDescr
*)(buf
+ sizeof(*log
) + sizeof(*hdr
));
4966 log
->num_confs
= cpu_to_le16(0);
4967 log
->size
= cpu_to_le32(log_size
);
4969 hdr
->descr_size
= cpu_to_le16(fdp_descr_size
);
4970 if (endgrp
->fdp
.enabled
) {
4971 hdr
->fdpa
= FIELD_DP8(hdr
->fdpa
, FDPA
, VALID
, 1);
4972 hdr
->fdpa
= FIELD_DP8(hdr
->fdpa
, FDPA
, RGIF
, endgrp
->fdp
.rgif
);
4973 hdr
->nrg
= cpu_to_le16(endgrp
->fdp
.nrg
);
4974 hdr
->nruh
= cpu_to_le16(endgrp
->fdp
.nruh
);
4975 hdr
->maxpids
= cpu_to_le16(NVME_FDP_MAXPIDS
- 1);
4976 hdr
->nnss
= cpu_to_le32(NVME_MAX_NAMESPACES
);
4977 hdr
->runs
= cpu_to_le64(endgrp
->fdp
.runs
);
4979 for (i
= 0; i
< nruh
; i
++) {
4980 ruhd
->ruht
= NVME_RUHT_INITIALLY_ISOLATED
;
4984 /* 1 bit for RUH in PIF -> 2 RUHs max. */
4985 hdr
->nrg
= cpu_to_le16(1);
4986 hdr
->nruh
= cpu_to_le16(1);
4987 hdr
->maxpids
= cpu_to_le16(NVME_FDP_MAXPIDS
- 1);
4988 hdr
->nnss
= cpu_to_le32(1);
4989 hdr
->runs
= cpu_to_le64(96 * MiB
);
4991 ruhd
->ruht
= NVME_RUHT_INITIALLY_ISOLATED
;
4994 return nvme_c2h(n
, (uint8_t *)buf
+ off
, trans_len
, req
);
4997 static uint16_t nvme_fdp_ruh_usage(NvmeCtrl
*n
, uint32_t endgrpid
,
4998 uint32_t dw10
, uint32_t dw12
,
4999 uint32_t buf_len
, uint64_t off
,
5004 NvmeRuhuDescr
*ruhud
;
5005 NvmeEnduranceGroup
*endgrp
;
5006 g_autofree
uint8_t *buf
= NULL
;
5007 uint32_t log_size
, trans_len
;
5010 if (endgrpid
!= 1 || !n
->subsys
) {
5011 return NVME_INVALID_FIELD
| NVME_DNR
;
5014 endgrp
= &n
->subsys
->endgrp
;
5016 if (!endgrp
->fdp
.enabled
) {
5017 return NVME_FDP_DISABLED
| NVME_DNR
;
5020 log_size
= sizeof(NvmeRuhuLog
) + endgrp
->fdp
.nruh
* sizeof(NvmeRuhuDescr
);
5022 if (off
>= log_size
) {
5023 return NVME_INVALID_FIELD
| NVME_DNR
;
5026 trans_len
= MIN(log_size
- off
, buf_len
);
5028 buf
= g_malloc0(log_size
);
5029 hdr
= (NvmeRuhuLog
*)buf
;
5030 ruhud
= (NvmeRuhuDescr
*)(hdr
+ 1);
5032 ruh
= endgrp
->fdp
.ruhs
;
5033 hdr
->nruh
= cpu_to_le16(endgrp
->fdp
.nruh
);
5035 for (i
= 0; i
< endgrp
->fdp
.nruh
; i
++, ruhud
++, ruh
++) {
5036 ruhud
->ruha
= ruh
->ruha
;
5039 return nvme_c2h(n
, (uint8_t *)buf
+ off
, trans_len
, req
);
5042 static uint16_t nvme_fdp_stats(NvmeCtrl
*n
, uint32_t endgrpid
, uint32_t buf_len
,
5043 uint64_t off
, NvmeRequest
*req
)
5045 NvmeEnduranceGroup
*endgrp
;
5046 NvmeFdpStatsLog log
= {};
5049 if (off
>= sizeof(NvmeFdpStatsLog
)) {
5050 return NVME_INVALID_FIELD
| NVME_DNR
;
5053 if (endgrpid
!= 1 || !n
->subsys
) {
5054 return NVME_INVALID_FIELD
| NVME_DNR
;
5057 if (!n
->subsys
->endgrp
.fdp
.enabled
) {
5058 return NVME_FDP_DISABLED
| NVME_DNR
;
5061 endgrp
= &n
->subsys
->endgrp
;
5063 trans_len
= MIN(sizeof(log
) - off
, buf_len
);
5065 /* spec value is 128 bit, we only use 64 bit */
5066 log
.hbmw
[0] = cpu_to_le64(endgrp
->fdp
.hbmw
);
5067 log
.mbmw
[0] = cpu_to_le64(endgrp
->fdp
.mbmw
);
5068 log
.mbe
[0] = cpu_to_le64(endgrp
->fdp
.mbe
);
5070 return nvme_c2h(n
, (uint8_t *)&log
+ off
, trans_len
, req
);
5073 static uint16_t nvme_fdp_events(NvmeCtrl
*n
, uint32_t endgrpid
,
5074 uint32_t buf_len
, uint64_t off
,
5077 NvmeEnduranceGroup
*endgrp
;
5078 NvmeCmd
*cmd
= &req
->cmd
;
5079 bool host_events
= (cmd
->cdw10
>> 8) & 0x1;
5080 uint32_t log_size
, trans_len
;
5081 NvmeFdpEventBuffer
*ebuf
;
5082 g_autofree NvmeFdpEventsLog
*elog
= NULL
;
5083 NvmeFdpEvent
*event
;
5085 if (endgrpid
!= 1 || !n
->subsys
) {
5086 return NVME_INVALID_FIELD
| NVME_DNR
;
5089 endgrp
= &n
->subsys
->endgrp
;
5091 if (!endgrp
->fdp
.enabled
) {
5092 return NVME_FDP_DISABLED
| NVME_DNR
;
5096 ebuf
= &endgrp
->fdp
.host_events
;
5098 ebuf
= &endgrp
->fdp
.ctrl_events
;
5101 log_size
= sizeof(NvmeFdpEventsLog
) + ebuf
->nelems
* sizeof(NvmeFdpEvent
);
5102 trans_len
= MIN(log_size
- off
, buf_len
);
5103 elog
= g_malloc0(log_size
);
5104 elog
->num_events
= cpu_to_le32(ebuf
->nelems
);
5105 event
= (NvmeFdpEvent
*)(elog
+ 1);
5107 if (ebuf
->nelems
&& ebuf
->start
== ebuf
->next
) {
5108 unsigned int nelems
= (NVME_FDP_MAX_EVENTS
- ebuf
->start
);
5109 /* wrap over, copy [start;NVME_FDP_MAX_EVENTS[ and [0; next[ */
5110 memcpy(event
, &ebuf
->events
[ebuf
->start
],
5111 sizeof(NvmeFdpEvent
) * nelems
);
5112 memcpy(event
+ nelems
, ebuf
->events
,
5113 sizeof(NvmeFdpEvent
) * ebuf
->next
);
5114 } else if (ebuf
->start
< ebuf
->next
) {
5115 memcpy(event
, &ebuf
->events
[ebuf
->start
],
5116 sizeof(NvmeFdpEvent
) * (ebuf
->next
- ebuf
->start
));
5119 return nvme_c2h(n
, (uint8_t *)elog
+ off
, trans_len
, req
);
5122 static uint16_t nvme_get_log(NvmeCtrl
*n
, NvmeRequest
*req
)
5124 NvmeCmd
*cmd
= &req
->cmd
;
5126 uint32_t dw10
= le32_to_cpu(cmd
->cdw10
);
5127 uint32_t dw11
= le32_to_cpu(cmd
->cdw11
);
5128 uint32_t dw12
= le32_to_cpu(cmd
->cdw12
);
5129 uint32_t dw13
= le32_to_cpu(cmd
->cdw13
);
5130 uint8_t lid
= dw10
& 0xff;
5131 uint8_t lsp
= (dw10
>> 8) & 0xf;
5132 uint8_t rae
= (dw10
>> 15) & 0x1;
5133 uint8_t csi
= le32_to_cpu(cmd
->cdw14
) >> 24;
5134 uint32_t numdl
, numdu
, lspi
;
5135 uint64_t off
, lpol
, lpou
;
5139 numdl
= (dw10
>> 16);
5140 numdu
= (dw11
& 0xffff);
5141 lspi
= (dw11
>> 16);
5145 len
= (((numdu
<< 16) | numdl
) + 1) << 2;
5146 off
= (lpou
<< 32ULL) | lpol
;
5149 return NVME_INVALID_FIELD
| NVME_DNR
;
5152 trace_pci_nvme_get_log(nvme_cid(req
), lid
, lsp
, rae
, len
, off
);
5154 status
= nvme_check_mdts(n
, len
);
5160 case NVME_LOG_ERROR_INFO
:
5161 return nvme_error_info(n
, rae
, len
, off
, req
);
5162 case NVME_LOG_SMART_INFO
:
5163 return nvme_smart_info(n
, rae
, len
, off
, req
);
5164 case NVME_LOG_FW_SLOT_INFO
:
5165 return nvme_fw_log_info(n
, len
, off
, req
);
5166 case NVME_LOG_CHANGED_NSLIST
:
5167 return nvme_changed_nslist(n
, rae
, len
, off
, req
);
5168 case NVME_LOG_CMD_EFFECTS
:
5169 return nvme_cmd_effects(n
, csi
, len
, off
, req
);
5170 case NVME_LOG_ENDGRP
:
5171 return nvme_endgrp_info(n
, rae
, len
, off
, req
);
5172 case NVME_LOG_FDP_CONFS
:
5173 return nvme_fdp_confs(n
, lspi
, len
, off
, req
);
5174 case NVME_LOG_FDP_RUH_USAGE
:
5175 return nvme_fdp_ruh_usage(n
, lspi
, dw10
, dw12
, len
, off
, req
);
5176 case NVME_LOG_FDP_STATS
:
5177 return nvme_fdp_stats(n
, lspi
, len
, off
, req
);
5178 case NVME_LOG_FDP_EVENTS
:
5179 return nvme_fdp_events(n
, lspi
, len
, off
, req
);
5181 trace_pci_nvme_err_invalid_log_page(nvme_cid(req
), lid
);
5182 return NVME_INVALID_FIELD
| NVME_DNR
;
5186 static void nvme_free_cq(NvmeCQueue
*cq
, NvmeCtrl
*n
)
5188 PCIDevice
*pci
= PCI_DEVICE(n
);
5189 uint16_t offset
= (cq
->cqid
<< 3) + (1 << 2);
5191 n
->cq
[cq
->cqid
] = NULL
;
5192 qemu_bh_delete(cq
->bh
);
5193 if (cq
->ioeventfd_enabled
) {
5194 memory_region_del_eventfd(&n
->iomem
,
5195 0x1000 + offset
, 4, false, 0, &cq
->notifier
);
5196 event_notifier_set_handler(&cq
->notifier
, NULL
);
5197 event_notifier_cleanup(&cq
->notifier
);
5199 if (msix_enabled(pci
)) {
5200 msix_vector_unuse(pci
, cq
->vector
);
5207 static uint16_t nvme_del_cq(NvmeCtrl
*n
, NvmeRequest
*req
)
5209 NvmeDeleteQ
*c
= (NvmeDeleteQ
*)&req
->cmd
;
5211 uint16_t qid
= le16_to_cpu(c
->qid
);
5213 if (unlikely(!qid
|| nvme_check_cqid(n
, qid
))) {
5214 trace_pci_nvme_err_invalid_del_cq_cqid(qid
);
5215 return NVME_INVALID_CQID
| NVME_DNR
;
5219 if (unlikely(!QTAILQ_EMPTY(&cq
->sq_list
))) {
5220 trace_pci_nvme_err_invalid_del_cq_notempty(qid
);
5221 return NVME_INVALID_QUEUE_DEL
;
5224 if (cq
->irq_enabled
&& cq
->tail
!= cq
->head
) {
5228 nvme_irq_deassert(n
, cq
);
5229 trace_pci_nvme_del_cq(qid
);
5230 nvme_free_cq(cq
, n
);
5231 return NVME_SUCCESS
;
5234 static void nvme_init_cq(NvmeCQueue
*cq
, NvmeCtrl
*n
, uint64_t dma_addr
,
5235 uint16_t cqid
, uint16_t vector
, uint16_t size
,
5236 uint16_t irq_enabled
)
5238 PCIDevice
*pci
= PCI_DEVICE(n
);
5240 if (msix_enabled(pci
)) {
5241 msix_vector_use(pci
, vector
);
5246 cq
->dma_addr
= dma_addr
;
5248 cq
->irq_enabled
= irq_enabled
;
5249 cq
->vector
= vector
;
5250 cq
->head
= cq
->tail
= 0;
5251 QTAILQ_INIT(&cq
->req_list
);
5252 QTAILQ_INIT(&cq
->sq_list
);
5253 if (n
->dbbuf_enabled
) {
5254 cq
->db_addr
= n
->dbbuf_dbs
+ (cqid
<< 3) + (1 << 2);
5255 cq
->ei_addr
= n
->dbbuf_eis
+ (cqid
<< 3) + (1 << 2);
5257 if (n
->params
.ioeventfd
&& cqid
!= 0) {
5258 if (!nvme_init_cq_ioeventfd(cq
)) {
5259 cq
->ioeventfd_enabled
= true;
5264 cq
->bh
= qemu_bh_new_guarded(nvme_post_cqes
, cq
,
5265 &DEVICE(cq
->ctrl
)->mem_reentrancy_guard
);
5268 static uint16_t nvme_create_cq(NvmeCtrl
*n
, NvmeRequest
*req
)
5271 NvmeCreateCq
*c
= (NvmeCreateCq
*)&req
->cmd
;
5272 uint16_t cqid
= le16_to_cpu(c
->cqid
);
5273 uint16_t vector
= le16_to_cpu(c
->irq_vector
);
5274 uint16_t qsize
= le16_to_cpu(c
->qsize
);
5275 uint16_t qflags
= le16_to_cpu(c
->cq_flags
);
5276 uint64_t prp1
= le64_to_cpu(c
->prp1
);
5278 trace_pci_nvme_create_cq(prp1
, cqid
, vector
, qsize
, qflags
,
5279 NVME_CQ_FLAGS_IEN(qflags
) != 0);
5281 if (unlikely(!cqid
|| cqid
> n
->conf_ioqpairs
|| n
->cq
[cqid
] != NULL
)) {
5282 trace_pci_nvme_err_invalid_create_cq_cqid(cqid
);
5283 return NVME_INVALID_QID
| NVME_DNR
;
5285 if (unlikely(!qsize
|| qsize
> NVME_CAP_MQES(ldq_le_p(&n
->bar
.cap
)))) {
5286 trace_pci_nvme_err_invalid_create_cq_size(qsize
);
5287 return NVME_MAX_QSIZE_EXCEEDED
| NVME_DNR
;
5289 if (unlikely(prp1
& (n
->page_size
- 1))) {
5290 trace_pci_nvme_err_invalid_create_cq_addr(prp1
);
5291 return NVME_INVALID_PRP_OFFSET
| NVME_DNR
;
5293 if (unlikely(!msix_enabled(PCI_DEVICE(n
)) && vector
)) {
5294 trace_pci_nvme_err_invalid_create_cq_vector(vector
);
5295 return NVME_INVALID_IRQ_VECTOR
| NVME_DNR
;
5297 if (unlikely(vector
>= n
->conf_msix_qsize
)) {
5298 trace_pci_nvme_err_invalid_create_cq_vector(vector
);
5299 return NVME_INVALID_IRQ_VECTOR
| NVME_DNR
;
5301 if (unlikely(!(NVME_CQ_FLAGS_PC(qflags
)))) {
5302 trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags
));
5303 return NVME_INVALID_FIELD
| NVME_DNR
;
5306 cq
= g_malloc0(sizeof(*cq
));
5307 nvme_init_cq(cq
, n
, prp1
, cqid
, vector
, qsize
+ 1,
5308 NVME_CQ_FLAGS_IEN(qflags
));
5311 * It is only required to set qs_created when creating a completion queue;
5312 * creating a submission queue without a matching completion queue will
5315 n
->qs_created
= true;
5316 return NVME_SUCCESS
;
5319 static uint16_t nvme_rpt_empty_id_struct(NvmeCtrl
*n
, NvmeRequest
*req
)
5321 uint8_t id
[NVME_IDENTIFY_DATA_SIZE
] = {};
5323 return nvme_c2h(n
, id
, sizeof(id
), req
);
5326 static uint16_t nvme_identify_ctrl(NvmeCtrl
*n
, NvmeRequest
*req
)
5328 trace_pci_nvme_identify_ctrl();
5330 return nvme_c2h(n
, (uint8_t *)&n
->id_ctrl
, sizeof(n
->id_ctrl
), req
);
5333 static uint16_t nvme_identify_ctrl_csi(NvmeCtrl
*n
, NvmeRequest
*req
)
5335 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5336 uint8_t id
[NVME_IDENTIFY_DATA_SIZE
] = {};
5337 NvmeIdCtrlNvm
*id_nvm
= (NvmeIdCtrlNvm
*)&id
;
5339 trace_pci_nvme_identify_ctrl_csi(c
->csi
);
5343 id_nvm
->vsl
= n
->params
.vsl
;
5344 id_nvm
->dmrsl
= cpu_to_le32(n
->dmrsl
);
5347 case NVME_CSI_ZONED
:
5348 ((NvmeIdCtrlZoned
*)&id
)->zasl
= n
->params
.zasl
;
5352 return NVME_INVALID_FIELD
| NVME_DNR
;
5355 return nvme_c2h(n
, id
, sizeof(id
), req
);
5358 static uint16_t nvme_identify_ns(NvmeCtrl
*n
, NvmeRequest
*req
, bool active
)
5361 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5362 uint32_t nsid
= le32_to_cpu(c
->nsid
);
5364 trace_pci_nvme_identify_ns(nsid
);
5366 if (!nvme_nsid_valid(n
, nsid
) || nsid
== NVME_NSID_BROADCAST
) {
5367 return NVME_INVALID_NSID
| NVME_DNR
;
5370 ns
= nvme_ns(n
, nsid
);
5371 if (unlikely(!ns
)) {
5373 ns
= nvme_subsys_ns(n
->subsys
, nsid
);
5375 return nvme_rpt_empty_id_struct(n
, req
);
5378 return nvme_rpt_empty_id_struct(n
, req
);
5382 if (active
|| ns
->csi
== NVME_CSI_NVM
) {
5383 return nvme_c2h(n
, (uint8_t *)&ns
->id_ns
, sizeof(NvmeIdNs
), req
);
5386 return NVME_INVALID_CMD_SET
| NVME_DNR
;
5389 static uint16_t nvme_identify_ctrl_list(NvmeCtrl
*n
, NvmeRequest
*req
,
5392 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5393 uint32_t nsid
= le32_to_cpu(c
->nsid
);
5394 uint16_t min_id
= le16_to_cpu(c
->ctrlid
);
5395 uint16_t list
[NVME_CONTROLLER_LIST_SIZE
] = {};
5396 uint16_t *ids
= &list
[1];
5399 int cntlid
, nr_ids
= 0;
5401 trace_pci_nvme_identify_ctrl_list(c
->cns
, min_id
);
5404 return NVME_INVALID_FIELD
| NVME_DNR
;
5408 if (nsid
== NVME_NSID_BROADCAST
) {
5409 return NVME_INVALID_FIELD
| NVME_DNR
;
5412 ns
= nvme_subsys_ns(n
->subsys
, nsid
);
5414 return NVME_INVALID_FIELD
| NVME_DNR
;
5418 for (cntlid
= min_id
; cntlid
< ARRAY_SIZE(n
->subsys
->ctrls
); cntlid
++) {
5419 ctrl
= nvme_subsys_ctrl(n
->subsys
, cntlid
);
5424 if (attached
&& !nvme_ns(ctrl
, nsid
)) {
5428 ids
[nr_ids
++] = cntlid
;
5433 return nvme_c2h(n
, (uint8_t *)list
, sizeof(list
), req
);
5436 static uint16_t nvme_identify_pri_ctrl_cap(NvmeCtrl
*n
, NvmeRequest
*req
)
5438 trace_pci_nvme_identify_pri_ctrl_cap(le16_to_cpu(n
->pri_ctrl_cap
.cntlid
));
5440 return nvme_c2h(n
, (uint8_t *)&n
->pri_ctrl_cap
,
5441 sizeof(NvmePriCtrlCap
), req
);
5444 static uint16_t nvme_identify_sec_ctrl_list(NvmeCtrl
*n
, NvmeRequest
*req
)
5446 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5447 uint16_t pri_ctrl_id
= le16_to_cpu(n
->pri_ctrl_cap
.cntlid
);
5448 uint16_t min_id
= le16_to_cpu(c
->ctrlid
);
5449 uint8_t num_sec_ctrl
= n
->sec_ctrl_list
.numcntl
;
5450 NvmeSecCtrlList list
= {0};
5453 for (i
= 0; i
< num_sec_ctrl
; i
++) {
5454 if (n
->sec_ctrl_list
.sec
[i
].scid
>= min_id
) {
5455 list
.numcntl
= num_sec_ctrl
- i
;
5456 memcpy(&list
.sec
, n
->sec_ctrl_list
.sec
+ i
,
5457 list
.numcntl
* sizeof(NvmeSecCtrlEntry
));
5462 trace_pci_nvme_identify_sec_ctrl_list(pri_ctrl_id
, list
.numcntl
);
5464 return nvme_c2h(n
, (uint8_t *)&list
, sizeof(list
), req
);
5467 static uint16_t nvme_identify_ns_csi(NvmeCtrl
*n
, NvmeRequest
*req
,
5471 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5472 uint32_t nsid
= le32_to_cpu(c
->nsid
);
5474 trace_pci_nvme_identify_ns_csi(nsid
, c
->csi
);
5476 if (!nvme_nsid_valid(n
, nsid
) || nsid
== NVME_NSID_BROADCAST
) {
5477 return NVME_INVALID_NSID
| NVME_DNR
;
5480 ns
= nvme_ns(n
, nsid
);
5481 if (unlikely(!ns
)) {
5483 ns
= nvme_subsys_ns(n
->subsys
, nsid
);
5485 return nvme_rpt_empty_id_struct(n
, req
);
5488 return nvme_rpt_empty_id_struct(n
, req
);
5492 if (c
->csi
== NVME_CSI_NVM
) {
5493 return nvme_c2h(n
, (uint8_t *)&ns
->id_ns_nvm
, sizeof(NvmeIdNsNvm
),
5495 } else if (c
->csi
== NVME_CSI_ZONED
&& ns
->csi
== NVME_CSI_ZONED
) {
5496 return nvme_c2h(n
, (uint8_t *)ns
->id_ns_zoned
, sizeof(NvmeIdNsZoned
),
5500 return NVME_INVALID_FIELD
| NVME_DNR
;
5503 static uint16_t nvme_identify_nslist(NvmeCtrl
*n
, NvmeRequest
*req
,
5507 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5508 uint32_t min_nsid
= le32_to_cpu(c
->nsid
);
5509 uint8_t list
[NVME_IDENTIFY_DATA_SIZE
] = {};
5510 static const int data_len
= sizeof(list
);
5511 uint32_t *list_ptr
= (uint32_t *)list
;
5514 trace_pci_nvme_identify_nslist(min_nsid
);
5517 * Both FFFFFFFFh (NVME_NSID_BROADCAST) and FFFFFFFFEh are invalid values
5518 * since the Active Namespace ID List should return namespaces with ids
5519 * *higher* than the NSID specified in the command. This is also specified
5520 * in the spec (NVM Express v1.3d, Section 5.15.4).
5522 if (min_nsid
>= NVME_NSID_BROADCAST
- 1) {
5523 return NVME_INVALID_NSID
| NVME_DNR
;
5526 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
5530 ns
= nvme_subsys_ns(n
->subsys
, i
);
5538 if (ns
->params
.nsid
<= min_nsid
) {
5541 list_ptr
[j
++] = cpu_to_le32(ns
->params
.nsid
);
5542 if (j
== data_len
/ sizeof(uint32_t)) {
5547 return nvme_c2h(n
, list
, data_len
, req
);
5550 static uint16_t nvme_identify_nslist_csi(NvmeCtrl
*n
, NvmeRequest
*req
,
5554 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5555 uint32_t min_nsid
= le32_to_cpu(c
->nsid
);
5556 uint8_t list
[NVME_IDENTIFY_DATA_SIZE
] = {};
5557 static const int data_len
= sizeof(list
);
5558 uint32_t *list_ptr
= (uint32_t *)list
;
5561 trace_pci_nvme_identify_nslist_csi(min_nsid
, c
->csi
);
5564 * Same as in nvme_identify_nslist(), FFFFFFFFh/FFFFFFFFEh are invalid.
5566 if (min_nsid
>= NVME_NSID_BROADCAST
- 1) {
5567 return NVME_INVALID_NSID
| NVME_DNR
;
5570 if (c
->csi
!= NVME_CSI_NVM
&& c
->csi
!= NVME_CSI_ZONED
) {
5571 return NVME_INVALID_FIELD
| NVME_DNR
;
5574 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
5578 ns
= nvme_subsys_ns(n
->subsys
, i
);
5586 if (ns
->params
.nsid
<= min_nsid
|| c
->csi
!= ns
->csi
) {
5589 list_ptr
[j
++] = cpu_to_le32(ns
->params
.nsid
);
5590 if (j
== data_len
/ sizeof(uint32_t)) {
5595 return nvme_c2h(n
, list
, data_len
, req
);
5598 static uint16_t nvme_identify_ns_descr_list(NvmeCtrl
*n
, NvmeRequest
*req
)
5601 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5602 uint32_t nsid
= le32_to_cpu(c
->nsid
);
5603 uint8_t list
[NVME_IDENTIFY_DATA_SIZE
] = {};
5604 uint8_t *pos
= list
;
5607 uint8_t v
[NVME_NIDL_UUID
];
5608 } QEMU_PACKED uuid
= {};
5612 } QEMU_PACKED eui64
= {};
5616 } QEMU_PACKED csi
= {};
5618 trace_pci_nvme_identify_ns_descr_list(nsid
);
5620 if (!nvme_nsid_valid(n
, nsid
) || nsid
== NVME_NSID_BROADCAST
) {
5621 return NVME_INVALID_NSID
| NVME_DNR
;
5624 ns
= nvme_ns(n
, nsid
);
5625 if (unlikely(!ns
)) {
5626 return NVME_INVALID_FIELD
| NVME_DNR
;
5629 if (!qemu_uuid_is_null(&ns
->params
.uuid
)) {
5630 uuid
.hdr
.nidt
= NVME_NIDT_UUID
;
5631 uuid
.hdr
.nidl
= NVME_NIDL_UUID
;
5632 memcpy(uuid
.v
, ns
->params
.uuid
.data
, NVME_NIDL_UUID
);
5633 memcpy(pos
, &uuid
, sizeof(uuid
));
5634 pos
+= sizeof(uuid
);
5637 if (ns
->params
.eui64
) {
5638 eui64
.hdr
.nidt
= NVME_NIDT_EUI64
;
5639 eui64
.hdr
.nidl
= NVME_NIDL_EUI64
;
5640 eui64
.v
= cpu_to_be64(ns
->params
.eui64
);
5641 memcpy(pos
, &eui64
, sizeof(eui64
));
5642 pos
+= sizeof(eui64
);
5645 csi
.hdr
.nidt
= NVME_NIDT_CSI
;
5646 csi
.hdr
.nidl
= NVME_NIDL_CSI
;
5648 memcpy(pos
, &csi
, sizeof(csi
));
5651 return nvme_c2h(n
, list
, sizeof(list
), req
);
5654 static uint16_t nvme_identify_cmd_set(NvmeCtrl
*n
, NvmeRequest
*req
)
5656 uint8_t list
[NVME_IDENTIFY_DATA_SIZE
] = {};
5657 static const int data_len
= sizeof(list
);
5659 trace_pci_nvme_identify_cmd_set();
5661 NVME_SET_CSI(*list
, NVME_CSI_NVM
);
5662 NVME_SET_CSI(*list
, NVME_CSI_ZONED
);
5664 return nvme_c2h(n
, list
, data_len
, req
);
5667 static uint16_t nvme_identify(NvmeCtrl
*n
, NvmeRequest
*req
)
5669 NvmeIdentify
*c
= (NvmeIdentify
*)&req
->cmd
;
5671 trace_pci_nvme_identify(nvme_cid(req
), c
->cns
, le16_to_cpu(c
->ctrlid
),
5675 case NVME_ID_CNS_NS
:
5676 return nvme_identify_ns(n
, req
, true);
5677 case NVME_ID_CNS_NS_PRESENT
:
5678 return nvme_identify_ns(n
, req
, false);
5679 case NVME_ID_CNS_NS_ATTACHED_CTRL_LIST
:
5680 return nvme_identify_ctrl_list(n
, req
, true);
5681 case NVME_ID_CNS_CTRL_LIST
:
5682 return nvme_identify_ctrl_list(n
, req
, false);
5683 case NVME_ID_CNS_PRIMARY_CTRL_CAP
:
5684 return nvme_identify_pri_ctrl_cap(n
, req
);
5685 case NVME_ID_CNS_SECONDARY_CTRL_LIST
:
5686 return nvme_identify_sec_ctrl_list(n
, req
);
5687 case NVME_ID_CNS_CS_NS
:
5688 return nvme_identify_ns_csi(n
, req
, true);
5689 case NVME_ID_CNS_CS_NS_PRESENT
:
5690 return nvme_identify_ns_csi(n
, req
, false);
5691 case NVME_ID_CNS_CTRL
:
5692 return nvme_identify_ctrl(n
, req
);
5693 case NVME_ID_CNS_CS_CTRL
:
5694 return nvme_identify_ctrl_csi(n
, req
);
5695 case NVME_ID_CNS_NS_ACTIVE_LIST
:
5696 return nvme_identify_nslist(n
, req
, true);
5697 case NVME_ID_CNS_NS_PRESENT_LIST
:
5698 return nvme_identify_nslist(n
, req
, false);
5699 case NVME_ID_CNS_CS_NS_ACTIVE_LIST
:
5700 return nvme_identify_nslist_csi(n
, req
, true);
5701 case NVME_ID_CNS_CS_NS_PRESENT_LIST
:
5702 return nvme_identify_nslist_csi(n
, req
, false);
5703 case NVME_ID_CNS_NS_DESCR_LIST
:
5704 return nvme_identify_ns_descr_list(n
, req
);
5705 case NVME_ID_CNS_IO_COMMAND_SET
:
5706 return nvme_identify_cmd_set(n
, req
);
5708 trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c
->cns
));
5709 return NVME_INVALID_FIELD
| NVME_DNR
;
5713 static uint16_t nvme_abort(NvmeCtrl
*n
, NvmeRequest
*req
)
5715 uint16_t sqid
= le32_to_cpu(req
->cmd
.cdw10
) & 0xffff;
5717 req
->cqe
.result
= 1;
5718 if (nvme_check_sqid(n
, sqid
)) {
5719 return NVME_INVALID_FIELD
| NVME_DNR
;
5722 return NVME_SUCCESS
;
5725 static inline void nvme_set_timestamp(NvmeCtrl
*n
, uint64_t ts
)
5727 trace_pci_nvme_setfeat_timestamp(ts
);
5729 n
->host_timestamp
= le64_to_cpu(ts
);
5730 n
->timestamp_set_qemu_clock_ms
= qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
);
5733 static inline uint64_t nvme_get_timestamp(const NvmeCtrl
*n
)
5735 uint64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
);
5736 uint64_t elapsed_time
= current_time
- n
->timestamp_set_qemu_clock_ms
;
5738 union nvme_timestamp
{
5740 uint64_t timestamp
:48;
5748 union nvme_timestamp ts
;
5750 ts
.timestamp
= n
->host_timestamp
+ elapsed_time
;
5752 /* If the host timestamp is non-zero, set the timestamp origin */
5753 ts
.origin
= n
->host_timestamp
? 0x01 : 0x00;
5755 trace_pci_nvme_getfeat_timestamp(ts
.all
);
5757 return cpu_to_le64(ts
.all
);
5760 static uint16_t nvme_get_feature_timestamp(NvmeCtrl
*n
, NvmeRequest
*req
)
5762 uint64_t timestamp
= nvme_get_timestamp(n
);
5764 return nvme_c2h(n
, (uint8_t *)×tamp
, sizeof(timestamp
), req
);
5767 static int nvme_get_feature_fdp(NvmeCtrl
*n
, uint32_t endgrpid
,
5772 if (!n
->subsys
|| !n
->subsys
->endgrp
.fdp
.enabled
) {
5773 return NVME_INVALID_FIELD
| NVME_DNR
;
5776 *result
= FIELD_DP16(0, FEAT_FDP
, FDPE
, 1);
5777 *result
= FIELD_DP16(*result
, FEAT_FDP
, CONF_NDX
, 0);
5779 return NVME_SUCCESS
;
5782 static uint16_t nvme_get_feature_fdp_events(NvmeCtrl
*n
, NvmeNamespace
*ns
,
5783 NvmeRequest
*req
, uint32_t *result
)
5785 NvmeCmd
*cmd
= &req
->cmd
;
5786 uint32_t cdw11
= le32_to_cpu(cmd
->cdw11
);
5787 uint16_t ph
= cdw11
& 0xffff;
5788 uint8_t noet
= (cdw11
>> 16) & 0xff;
5789 uint16_t ruhid
, ret
;
5790 uint32_t nentries
= 0;
5791 uint8_t s_events_ndx
= 0;
5792 size_t s_events_siz
= sizeof(NvmeFdpEventDescr
) * noet
;
5793 g_autofree NvmeFdpEventDescr
*s_events
= g_malloc0(s_events_siz
);
5795 NvmeFdpEventDescr
*s_event
;
5797 if (!n
->subsys
|| !n
->subsys
->endgrp
.fdp
.enabled
) {
5798 return NVME_FDP_DISABLED
| NVME_DNR
;
5801 if (!nvme_ph_valid(ns
, ph
)) {
5802 return NVME_INVALID_FIELD
| NVME_DNR
;
5805 ruhid
= ns
->fdp
.phs
[ph
];
5806 ruh
= &n
->subsys
->endgrp
.fdp
.ruhs
[ruhid
];
5810 if (unlikely(noet
== 0)) {
5811 return NVME_INVALID_FIELD
| NVME_DNR
;
5814 for (uint8_t event_type
= 0; event_type
< FDP_EVT_MAX
; event_type
++) {
5815 uint8_t shift
= nvme_fdp_evf_shifts
[event_type
];
5816 if (!shift
&& event_type
) {
5818 * only first entry (event_type == 0) has a shift value of 0
5819 * other entries are simply unpopulated.
5826 s_event
= &s_events
[s_events_ndx
];
5827 s_event
->evt
= event_type
;
5828 s_event
->evta
= (ruh
->event_filter
>> shift
) & 0x1;
5830 /* break if all `noet` entries are filled */
5831 if ((++s_events_ndx
) == noet
) {
5836 ret
= nvme_c2h(n
, s_events
, s_events_siz
, req
);
5842 return NVME_SUCCESS
;
5845 static uint16_t nvme_get_feature(NvmeCtrl
*n
, NvmeRequest
*req
)
5847 NvmeCmd
*cmd
= &req
->cmd
;
5848 uint32_t dw10
= le32_to_cpu(cmd
->cdw10
);
5849 uint32_t dw11
= le32_to_cpu(cmd
->cdw11
);
5850 uint32_t nsid
= le32_to_cpu(cmd
->nsid
);
5852 uint8_t fid
= NVME_GETSETFEAT_FID(dw10
);
5853 NvmeGetFeatureSelect sel
= NVME_GETFEAT_SELECT(dw10
);
5857 uint16_t endgrpid
= 0, ret
= NVME_SUCCESS
;
5859 static const uint32_t nvme_feature_default
[NVME_FID_MAX
] = {
5860 [NVME_ARBITRATION
] = NVME_ARB_AB_NOLIMIT
,
5863 trace_pci_nvme_getfeat(nvme_cid(req
), nsid
, fid
, sel
, dw11
);
5865 if (!nvme_feature_support
[fid
]) {
5866 return NVME_INVALID_FIELD
| NVME_DNR
;
5869 if (nvme_feature_cap
[fid
] & NVME_FEAT_CAP_NS
) {
5870 if (!nvme_nsid_valid(n
, nsid
) || nsid
== NVME_NSID_BROADCAST
) {
5872 * The Reservation Notification Mask and Reservation Persistence
5873 * features require a status code of Invalid Field in Command when
5874 * NSID is FFFFFFFFh. Since the device does not support those
5875 * features we can always return Invalid Namespace or Format as we
5876 * should do for all other features.
5878 return NVME_INVALID_NSID
| NVME_DNR
;
5881 if (!nvme_ns(n
, nsid
)) {
5882 return NVME_INVALID_FIELD
| NVME_DNR
;
5887 case NVME_GETFEAT_SELECT_CURRENT
:
5889 case NVME_GETFEAT_SELECT_SAVED
:
5890 /* no features are saveable by the controller; fallthrough */
5891 case NVME_GETFEAT_SELECT_DEFAULT
:
5893 case NVME_GETFEAT_SELECT_CAP
:
5894 result
= nvme_feature_cap
[fid
];
5899 case NVME_TEMPERATURE_THRESHOLD
:
5903 * The controller only implements the Composite Temperature sensor, so
5904 * return 0 for all other sensors.
5906 if (NVME_TEMP_TMPSEL(dw11
) != NVME_TEMP_TMPSEL_COMPOSITE
) {
5910 switch (NVME_TEMP_THSEL(dw11
)) {
5911 case NVME_TEMP_THSEL_OVER
:
5912 result
= n
->features
.temp_thresh_hi
;
5914 case NVME_TEMP_THSEL_UNDER
:
5915 result
= n
->features
.temp_thresh_low
;
5919 return NVME_INVALID_FIELD
| NVME_DNR
;
5920 case NVME_ERROR_RECOVERY
:
5921 if (!nvme_nsid_valid(n
, nsid
)) {
5922 return NVME_INVALID_NSID
| NVME_DNR
;
5925 ns
= nvme_ns(n
, nsid
);
5926 if (unlikely(!ns
)) {
5927 return NVME_INVALID_FIELD
| NVME_DNR
;
5930 result
= ns
->features
.err_rec
;
5932 case NVME_VOLATILE_WRITE_CACHE
:
5934 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
5940 result
= blk_enable_write_cache(ns
->blkconf
.blk
);
5945 trace_pci_nvme_getfeat_vwcache(result
? "enabled" : "disabled");
5947 case NVME_ASYNCHRONOUS_EVENT_CONF
:
5948 result
= n
->features
.async_config
;
5950 case NVME_TIMESTAMP
:
5951 return nvme_get_feature_timestamp(n
, req
);
5952 case NVME_HOST_BEHAVIOR_SUPPORT
:
5953 return nvme_c2h(n
, (uint8_t *)&n
->features
.hbs
,
5954 sizeof(n
->features
.hbs
), req
);
5956 endgrpid
= dw11
& 0xff;
5958 if (endgrpid
!= 0x1) {
5959 return NVME_INVALID_FIELD
| NVME_DNR
;
5962 ret
= nvme_get_feature_fdp(n
, endgrpid
, &result
);
5967 case NVME_FDP_EVENTS
:
5968 if (!nvme_nsid_valid(n
, nsid
)) {
5969 return NVME_INVALID_NSID
| NVME_DNR
;
5972 ns
= nvme_ns(n
, nsid
);
5973 if (unlikely(!ns
)) {
5974 return NVME_INVALID_FIELD
| NVME_DNR
;
5977 ret
= nvme_get_feature_fdp_events(n
, ns
, req
, &result
);
5988 case NVME_TEMPERATURE_THRESHOLD
:
5991 if (NVME_TEMP_TMPSEL(dw11
) != NVME_TEMP_TMPSEL_COMPOSITE
) {
5995 if (NVME_TEMP_THSEL(dw11
) == NVME_TEMP_THSEL_OVER
) {
5996 result
= NVME_TEMPERATURE_WARNING
;
6000 case NVME_NUMBER_OF_QUEUES
:
6001 result
= (n
->conf_ioqpairs
- 1) | ((n
->conf_ioqpairs
- 1) << 16);
6002 trace_pci_nvme_getfeat_numq(result
);
6004 case NVME_INTERRUPT_VECTOR_CONF
:
6006 if (iv
>= n
->conf_ioqpairs
+ 1) {
6007 return NVME_INVALID_FIELD
| NVME_DNR
;
6011 if (iv
== n
->admin_cq
.vector
) {
6012 result
|= NVME_INTVC_NOCOALESCING
;
6016 endgrpid
= dw11
& 0xff;
6018 if (endgrpid
!= 0x1) {
6019 return NVME_INVALID_FIELD
| NVME_DNR
;
6022 ret
= nvme_get_feature_fdp(n
, endgrpid
, &result
);
6030 result
= nvme_feature_default
[fid
];
6035 req
->cqe
.result
= cpu_to_le32(result
);
6039 static uint16_t nvme_set_feature_timestamp(NvmeCtrl
*n
, NvmeRequest
*req
)
6044 ret
= nvme_h2c(n
, (uint8_t *)×tamp
, sizeof(timestamp
), req
);
6049 nvme_set_timestamp(n
, timestamp
);
6051 return NVME_SUCCESS
;
6054 static uint16_t nvme_set_feature_fdp_events(NvmeCtrl
*n
, NvmeNamespace
*ns
,
6057 NvmeCmd
*cmd
= &req
->cmd
;
6058 uint32_t cdw11
= le32_to_cpu(cmd
->cdw11
);
6059 uint16_t ph
= cdw11
& 0xffff;
6060 uint8_t noet
= (cdw11
>> 16) & 0xff;
6061 uint16_t ret
, ruhid
;
6062 uint8_t enable
= le32_to_cpu(cmd
->cdw12
) & 0x1;
6063 uint8_t event_mask
= 0;
6065 g_autofree
uint8_t *events
= g_malloc0(noet
);
6066 NvmeRuHandle
*ruh
= NULL
;
6070 if (!n
->subsys
|| !n
->subsys
->endgrp
.fdp
.enabled
) {
6071 return NVME_FDP_DISABLED
| NVME_DNR
;
6074 if (!nvme_ph_valid(ns
, ph
)) {
6075 return NVME_INVALID_FIELD
| NVME_DNR
;
6078 ruhid
= ns
->fdp
.phs
[ph
];
6079 ruh
= &n
->subsys
->endgrp
.fdp
.ruhs
[ruhid
];
6081 ret
= nvme_h2c(n
, events
, noet
, req
);
6086 for (i
= 0; i
< noet
; i
++) {
6087 event_mask
|= (1 << nvme_fdp_evf_shifts
[events
[i
]]);
6091 ruh
->event_filter
|= event_mask
;
6093 ruh
->event_filter
= ruh
->event_filter
& ~event_mask
;
6096 return NVME_SUCCESS
;
6099 static uint16_t nvme_set_feature(NvmeCtrl
*n
, NvmeRequest
*req
)
6101 NvmeNamespace
*ns
= NULL
;
6103 NvmeCmd
*cmd
= &req
->cmd
;
6104 uint32_t dw10
= le32_to_cpu(cmd
->cdw10
);
6105 uint32_t dw11
= le32_to_cpu(cmd
->cdw11
);
6106 uint32_t nsid
= le32_to_cpu(cmd
->nsid
);
6107 uint8_t fid
= NVME_GETSETFEAT_FID(dw10
);
6108 uint8_t save
= NVME_SETFEAT_SAVE(dw10
);
6112 trace_pci_nvme_setfeat(nvme_cid(req
), nsid
, fid
, save
, dw11
);
6114 if (save
&& !(nvme_feature_cap
[fid
] & NVME_FEAT_CAP_SAVE
)) {
6115 return NVME_FID_NOT_SAVEABLE
| NVME_DNR
;
6118 if (!nvme_feature_support
[fid
]) {
6119 return NVME_INVALID_FIELD
| NVME_DNR
;
6122 if (nvme_feature_cap
[fid
] & NVME_FEAT_CAP_NS
) {
6123 if (nsid
!= NVME_NSID_BROADCAST
) {
6124 if (!nvme_nsid_valid(n
, nsid
)) {
6125 return NVME_INVALID_NSID
| NVME_DNR
;
6128 ns
= nvme_ns(n
, nsid
);
6129 if (unlikely(!ns
)) {
6130 return NVME_INVALID_FIELD
| NVME_DNR
;
6133 } else if (nsid
&& nsid
!= NVME_NSID_BROADCAST
) {
6134 if (!nvme_nsid_valid(n
, nsid
)) {
6135 return NVME_INVALID_NSID
| NVME_DNR
;
6138 return NVME_FEAT_NOT_NS_SPEC
| NVME_DNR
;
6141 if (!(nvme_feature_cap
[fid
] & NVME_FEAT_CAP_CHANGE
)) {
6142 return NVME_FEAT_NOT_CHANGEABLE
| NVME_DNR
;
6146 case NVME_TEMPERATURE_THRESHOLD
:
6147 if (NVME_TEMP_TMPSEL(dw11
) != NVME_TEMP_TMPSEL_COMPOSITE
) {
6151 switch (NVME_TEMP_THSEL(dw11
)) {
6152 case NVME_TEMP_THSEL_OVER
:
6153 n
->features
.temp_thresh_hi
= NVME_TEMP_TMPTH(dw11
);
6155 case NVME_TEMP_THSEL_UNDER
:
6156 n
->features
.temp_thresh_low
= NVME_TEMP_TMPTH(dw11
);
6159 return NVME_INVALID_FIELD
| NVME_DNR
;
6162 if ((n
->temperature
>= n
->features
.temp_thresh_hi
) ||
6163 (n
->temperature
<= n
->features
.temp_thresh_low
)) {
6164 nvme_smart_event(n
, NVME_SMART_TEMPERATURE
);
6168 case NVME_ERROR_RECOVERY
:
6169 if (nsid
== NVME_NSID_BROADCAST
) {
6170 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
6177 if (NVME_ID_NS_NSFEAT_DULBE(ns
->id_ns
.nsfeat
)) {
6178 ns
->features
.err_rec
= dw11
;
6186 if (NVME_ID_NS_NSFEAT_DULBE(ns
->id_ns
.nsfeat
)) {
6187 ns
->features
.err_rec
= dw11
;
6190 case NVME_VOLATILE_WRITE_CACHE
:
6191 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
6197 if (!(dw11
& 0x1) && blk_enable_write_cache(ns
->blkconf
.blk
)) {
6198 blk_flush(ns
->blkconf
.blk
);
6201 blk_set_enable_write_cache(ns
->blkconf
.blk
, dw11
& 1);
6206 case NVME_NUMBER_OF_QUEUES
:
6207 if (n
->qs_created
) {
6208 return NVME_CMD_SEQ_ERROR
| NVME_DNR
;
6212 * NVMe v1.3, Section 5.21.1.7: FFFFh is not an allowed value for NCQR
6215 if ((dw11
& 0xffff) == 0xffff || ((dw11
>> 16) & 0xffff) == 0xffff) {
6216 return NVME_INVALID_FIELD
| NVME_DNR
;
6219 trace_pci_nvme_setfeat_numq((dw11
& 0xffff) + 1,
6220 ((dw11
>> 16) & 0xffff) + 1,
6223 req
->cqe
.result
= cpu_to_le32((n
->conf_ioqpairs
- 1) |
6224 ((n
->conf_ioqpairs
- 1) << 16));
6226 case NVME_ASYNCHRONOUS_EVENT_CONF
:
6227 n
->features
.async_config
= dw11
;
6229 case NVME_TIMESTAMP
:
6230 return nvme_set_feature_timestamp(n
, req
);
6231 case NVME_HOST_BEHAVIOR_SUPPORT
:
6232 status
= nvme_h2c(n
, (uint8_t *)&n
->features
.hbs
,
6233 sizeof(n
->features
.hbs
), req
);
6238 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
6245 ns
->id_ns
.nlbaf
= ns
->nlbaf
- 1;
6246 if (!n
->features
.hbs
.lbafee
) {
6247 ns
->id_ns
.nlbaf
= MIN(ns
->id_ns
.nlbaf
, 15);
6252 case NVME_COMMAND_SET_PROFILE
:
6254 trace_pci_nvme_err_invalid_iocsci(dw11
& 0x1ff);
6255 return NVME_CMD_SET_CMB_REJECTED
| NVME_DNR
;
6259 /* spec: abort with cmd seq err if there's one or more NS' in endgrp */
6260 return NVME_CMD_SEQ_ERROR
| NVME_DNR
;
6261 case NVME_FDP_EVENTS
:
6262 return nvme_set_feature_fdp_events(n
, ns
, req
);
6264 return NVME_FEAT_NOT_CHANGEABLE
| NVME_DNR
;
6266 return NVME_SUCCESS
;
6269 static uint16_t nvme_aer(NvmeCtrl
*n
, NvmeRequest
*req
)
6271 trace_pci_nvme_aer(nvme_cid(req
));
6273 if (n
->outstanding_aers
> n
->params
.aerl
) {
6274 trace_pci_nvme_aer_aerl_exceeded();
6275 return NVME_AER_LIMIT_EXCEEDED
;
6278 n
->aer_reqs
[n
->outstanding_aers
] = req
;
6279 n
->outstanding_aers
++;
6281 if (!QTAILQ_EMPTY(&n
->aer_queue
)) {
6282 nvme_process_aers(n
);
6285 return NVME_NO_COMPLETE
;
6288 static void nvme_update_dmrsl(NvmeCtrl
*n
)
6292 for (nsid
= 1; nsid
<= NVME_MAX_NAMESPACES
; nsid
++) {
6293 NvmeNamespace
*ns
= nvme_ns(n
, nsid
);
6298 n
->dmrsl
= MIN_NON_ZERO(n
->dmrsl
,
6299 BDRV_REQUEST_MAX_BYTES
/ nvme_l2b(ns
, 1));
6303 static void nvme_select_iocs_ns(NvmeCtrl
*n
, NvmeNamespace
*ns
)
6305 uint32_t cc
= ldl_le_p(&n
->bar
.cc
);
6307 ns
->iocs
= nvme_cse_iocs_none
;
6310 if (NVME_CC_CSS(cc
) != NVME_CC_CSS_ADMIN_ONLY
) {
6311 ns
->iocs
= nvme_cse_iocs_nvm
;
6314 case NVME_CSI_ZONED
:
6315 if (NVME_CC_CSS(cc
) == NVME_CC_CSS_CSI
) {
6316 ns
->iocs
= nvme_cse_iocs_zoned
;
6317 } else if (NVME_CC_CSS(cc
) == NVME_CC_CSS_NVM
) {
6318 ns
->iocs
= nvme_cse_iocs_nvm
;
6324 static uint16_t nvme_ns_attachment(NvmeCtrl
*n
, NvmeRequest
*req
)
6328 uint16_t list
[NVME_CONTROLLER_LIST_SIZE
] = {};
6329 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
6330 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6331 uint8_t sel
= dw10
& 0xf;
6332 uint16_t *nr_ids
= &list
[0];
6333 uint16_t *ids
= &list
[1];
6337 trace_pci_nvme_ns_attachment(nvme_cid(req
), dw10
& 0xf);
6339 if (!nvme_nsid_valid(n
, nsid
)) {
6340 return NVME_INVALID_NSID
| NVME_DNR
;
6343 ns
= nvme_subsys_ns(n
->subsys
, nsid
);
6345 return NVME_INVALID_FIELD
| NVME_DNR
;
6348 ret
= nvme_h2c(n
, (uint8_t *)list
, 4096, req
);
6354 return NVME_NS_CTRL_LIST_INVALID
| NVME_DNR
;
6357 *nr_ids
= MIN(*nr_ids
, NVME_CONTROLLER_LIST_SIZE
- 1);
6358 for (i
= 0; i
< *nr_ids
; i
++) {
6359 ctrl
= nvme_subsys_ctrl(n
->subsys
, ids
[i
]);
6361 return NVME_NS_CTRL_LIST_INVALID
| NVME_DNR
;
6365 case NVME_NS_ATTACHMENT_ATTACH
:
6366 if (nvme_ns(ctrl
, nsid
)) {
6367 return NVME_NS_ALREADY_ATTACHED
| NVME_DNR
;
6370 if (ns
->attached
&& !ns
->params
.shared
) {
6371 return NVME_NS_PRIVATE
| NVME_DNR
;
6374 nvme_attach_ns(ctrl
, ns
);
6375 nvme_select_iocs_ns(ctrl
, ns
);
6379 case NVME_NS_ATTACHMENT_DETACH
:
6380 if (!nvme_ns(ctrl
, nsid
)) {
6381 return NVME_NS_NOT_ATTACHED
| NVME_DNR
;
6384 ctrl
->namespaces
[nsid
] = NULL
;
6387 nvme_update_dmrsl(ctrl
);
6392 return NVME_INVALID_FIELD
| NVME_DNR
;
6396 * Add namespace id to the changed namespace id list for event clearing
6397 * via Get Log Page command.
6399 if (!test_and_set_bit(nsid
, ctrl
->changed_nsids
)) {
6400 nvme_enqueue_event(ctrl
, NVME_AER_TYPE_NOTICE
,
6401 NVME_AER_INFO_NOTICE_NS_ATTR_CHANGED
,
6402 NVME_LOG_CHANGED_NSLIST
);
6406 return NVME_SUCCESS
;
6409 typedef struct NvmeFormatAIOCB
{
6426 static void nvme_format_cancel(BlockAIOCB
*aiocb
)
6428 NvmeFormatAIOCB
*iocb
= container_of(aiocb
, NvmeFormatAIOCB
, common
);
6430 iocb
->ret
= -ECANCELED
;
6433 blk_aio_cancel_async(iocb
->aiocb
);
6438 static const AIOCBInfo nvme_format_aiocb_info
= {
6439 .aiocb_size
= sizeof(NvmeFormatAIOCB
),
6440 .cancel_async
= nvme_format_cancel
,
6441 .get_aio_context
= nvme_get_aio_context
,
6444 static void nvme_format_set(NvmeNamespace
*ns
, uint8_t lbaf
, uint8_t mset
,
6445 uint8_t pi
, uint8_t pil
)
6447 uint8_t lbafl
= lbaf
& 0xf;
6448 uint8_t lbafu
= lbaf
>> 4;
6450 trace_pci_nvme_format_set(ns
->params
.nsid
, lbaf
, mset
, pi
, pil
);
6452 ns
->id_ns
.dps
= (pil
<< 3) | pi
;
6453 ns
->id_ns
.flbas
= (lbafu
<< 5) | (mset
<< 4) | lbafl
;
6455 nvme_ns_init_format(ns
);
6458 static void nvme_do_format(NvmeFormatAIOCB
*iocb
);
6460 static void nvme_format_ns_cb(void *opaque
, int ret
)
6462 NvmeFormatAIOCB
*iocb
= opaque
;
6463 NvmeNamespace
*ns
= iocb
->ns
;
6466 if (iocb
->ret
< 0) {
6468 } else if (ret
< 0) {
6475 if (iocb
->offset
< ns
->size
) {
6476 bytes
= MIN(BDRV_REQUEST_MAX_BYTES
, ns
->size
- iocb
->offset
);
6478 iocb
->aiocb
= blk_aio_pwrite_zeroes(ns
->blkconf
.blk
, iocb
->offset
,
6479 bytes
, BDRV_REQ_MAY_UNMAP
,
6480 nvme_format_ns_cb
, iocb
);
6482 iocb
->offset
+= bytes
;
6486 nvme_format_set(ns
, iocb
->lbaf
, iocb
->mset
, iocb
->pi
, iocb
->pil
);
6492 nvme_do_format(iocb
);
6495 static uint16_t nvme_format_check(NvmeNamespace
*ns
, uint8_t lbaf
, uint8_t pi
)
6497 if (ns
->params
.zoned
) {
6498 return NVME_INVALID_FORMAT
| NVME_DNR
;
6501 if (lbaf
> ns
->id_ns
.nlbaf
) {
6502 return NVME_INVALID_FORMAT
| NVME_DNR
;
6505 if (pi
&& (ns
->id_ns
.lbaf
[lbaf
].ms
< nvme_pi_tuple_size(ns
))) {
6506 return NVME_INVALID_FORMAT
| NVME_DNR
;
6509 if (pi
&& pi
> NVME_ID_NS_DPS_TYPE_3
) {
6510 return NVME_INVALID_FIELD
| NVME_DNR
;
6513 return NVME_SUCCESS
;
6516 static void nvme_do_format(NvmeFormatAIOCB
*iocb
)
6518 NvmeRequest
*req
= iocb
->req
;
6519 NvmeCtrl
*n
= nvme_ctrl(req
);
6520 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6521 uint8_t lbaf
= dw10
& 0xf;
6522 uint8_t pi
= (dw10
>> 5) & 0x7;
6526 if (iocb
->ret
< 0) {
6530 if (iocb
->broadcast
) {
6531 for (i
= iocb
->nsid
+ 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
6532 iocb
->ns
= nvme_ns(n
, i
);
6544 status
= nvme_format_check(iocb
->ns
, lbaf
, pi
);
6546 req
->status
= status
;
6550 iocb
->ns
->status
= NVME_FORMAT_IN_PROGRESS
;
6551 nvme_format_ns_cb(iocb
, 0);
6555 iocb
->common
.cb(iocb
->common
.opaque
, iocb
->ret
);
6556 qemu_aio_unref(iocb
);
6559 static uint16_t nvme_format(NvmeCtrl
*n
, NvmeRequest
*req
)
6561 NvmeFormatAIOCB
*iocb
;
6562 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
6563 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6564 uint8_t lbaf
= dw10
& 0xf;
6565 uint8_t mset
= (dw10
>> 4) & 0x1;
6566 uint8_t pi
= (dw10
>> 5) & 0x7;
6567 uint8_t pil
= (dw10
>> 8) & 0x1;
6568 uint8_t lbafu
= (dw10
>> 12) & 0x3;
6571 iocb
= qemu_aio_get(&nvme_format_aiocb_info
, NULL
, nvme_misc_cb
, req
);
6581 iocb
->broadcast
= (nsid
== NVME_NSID_BROADCAST
);
6584 if (n
->features
.hbs
.lbafee
) {
6585 iocb
->lbaf
|= lbafu
<< 4;
6588 if (!iocb
->broadcast
) {
6589 if (!nvme_nsid_valid(n
, nsid
)) {
6590 status
= NVME_INVALID_NSID
| NVME_DNR
;
6594 iocb
->ns
= nvme_ns(n
, nsid
);
6596 status
= NVME_INVALID_FIELD
| NVME_DNR
;
6601 req
->aiocb
= &iocb
->common
;
6602 nvme_do_format(iocb
);
6604 return NVME_NO_COMPLETE
;
6607 qemu_aio_unref(iocb
);
6612 static void nvme_get_virt_res_num(NvmeCtrl
*n
, uint8_t rt
, int *num_total
,
6613 int *num_prim
, int *num_sec
)
6615 *num_total
= le32_to_cpu(rt
?
6616 n
->pri_ctrl_cap
.vifrt
: n
->pri_ctrl_cap
.vqfrt
);
6617 *num_prim
= le16_to_cpu(rt
?
6618 n
->pri_ctrl_cap
.virfap
: n
->pri_ctrl_cap
.vqrfap
);
6619 *num_sec
= le16_to_cpu(rt
? n
->pri_ctrl_cap
.virfa
: n
->pri_ctrl_cap
.vqrfa
);
6622 static uint16_t nvme_assign_virt_res_to_prim(NvmeCtrl
*n
, NvmeRequest
*req
,
6623 uint16_t cntlid
, uint8_t rt
,
6626 int num_total
, num_prim
, num_sec
;
6628 if (cntlid
!= n
->cntlid
) {
6629 return NVME_INVALID_CTRL_ID
| NVME_DNR
;
6632 nvme_get_virt_res_num(n
, rt
, &num_total
, &num_prim
, &num_sec
);
6634 if (nr
> num_total
) {
6635 return NVME_INVALID_NUM_RESOURCES
| NVME_DNR
;
6638 if (nr
> num_total
- num_sec
) {
6639 return NVME_INVALID_RESOURCE_ID
| NVME_DNR
;
6643 n
->next_pri_ctrl_cap
.virfap
= cpu_to_le16(nr
);
6645 n
->next_pri_ctrl_cap
.vqrfap
= cpu_to_le16(nr
);
6648 req
->cqe
.result
= cpu_to_le32(nr
);
6652 static void nvme_update_virt_res(NvmeCtrl
*n
, NvmeSecCtrlEntry
*sctrl
,
6655 int prev_nr
, prev_total
;
6658 prev_nr
= le16_to_cpu(sctrl
->nvi
);
6659 prev_total
= le32_to_cpu(n
->pri_ctrl_cap
.virfa
);
6660 sctrl
->nvi
= cpu_to_le16(nr
);
6661 n
->pri_ctrl_cap
.virfa
= cpu_to_le32(prev_total
+ nr
- prev_nr
);
6663 prev_nr
= le16_to_cpu(sctrl
->nvq
);
6664 prev_total
= le32_to_cpu(n
->pri_ctrl_cap
.vqrfa
);
6665 sctrl
->nvq
= cpu_to_le16(nr
);
6666 n
->pri_ctrl_cap
.vqrfa
= cpu_to_le32(prev_total
+ nr
- prev_nr
);
6670 static uint16_t nvme_assign_virt_res_to_sec(NvmeCtrl
*n
, NvmeRequest
*req
,
6671 uint16_t cntlid
, uint8_t rt
, int nr
)
6673 int num_total
, num_prim
, num_sec
, num_free
, diff
, limit
;
6674 NvmeSecCtrlEntry
*sctrl
;
6676 sctrl
= nvme_sctrl_for_cntlid(n
, cntlid
);
6678 return NVME_INVALID_CTRL_ID
| NVME_DNR
;
6682 return NVME_INVALID_SEC_CTRL_STATE
| NVME_DNR
;
6685 limit
= le16_to_cpu(rt
? n
->pri_ctrl_cap
.vifrsm
: n
->pri_ctrl_cap
.vqfrsm
);
6687 return NVME_INVALID_NUM_RESOURCES
| NVME_DNR
;
6690 nvme_get_virt_res_num(n
, rt
, &num_total
, &num_prim
, &num_sec
);
6691 num_free
= num_total
- num_prim
- num_sec
;
6692 diff
= nr
- le16_to_cpu(rt
? sctrl
->nvi
: sctrl
->nvq
);
6694 if (diff
> num_free
) {
6695 return NVME_INVALID_RESOURCE_ID
| NVME_DNR
;
6698 nvme_update_virt_res(n
, sctrl
, rt
, nr
);
6699 req
->cqe
.result
= cpu_to_le32(nr
);
6704 static uint16_t nvme_virt_set_state(NvmeCtrl
*n
, uint16_t cntlid
, bool online
)
6706 PCIDevice
*pci
= PCI_DEVICE(n
);
6707 NvmeCtrl
*sn
= NULL
;
6708 NvmeSecCtrlEntry
*sctrl
;
6711 sctrl
= nvme_sctrl_for_cntlid(n
, cntlid
);
6713 return NVME_INVALID_CTRL_ID
| NVME_DNR
;
6716 if (!pci_is_vf(pci
)) {
6717 vf_index
= le16_to_cpu(sctrl
->vfn
) - 1;
6718 sn
= NVME(pcie_sriov_get_vf_at_index(pci
, vf_index
));
6722 if (!sctrl
->nvi
|| (le16_to_cpu(sctrl
->nvq
) < 2) || !sn
) {
6723 return NVME_INVALID_SEC_CTRL_STATE
| NVME_DNR
;
6728 nvme_ctrl_reset(sn
, NVME_RESET_FUNCTION
);
6731 nvme_update_virt_res(n
, sctrl
, NVME_VIRT_RES_INTERRUPT
, 0);
6732 nvme_update_virt_res(n
, sctrl
, NVME_VIRT_RES_QUEUE
, 0);
6737 nvme_ctrl_reset(sn
, NVME_RESET_FUNCTION
);
6742 return NVME_SUCCESS
;
6745 static uint16_t nvme_virt_mngmt(NvmeCtrl
*n
, NvmeRequest
*req
)
6747 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6748 uint32_t dw11
= le32_to_cpu(req
->cmd
.cdw11
);
6749 uint8_t act
= dw10
& 0xf;
6750 uint8_t rt
= (dw10
>> 8) & 0x7;
6751 uint16_t cntlid
= (dw10
>> 16) & 0xffff;
6752 int nr
= dw11
& 0xffff;
6754 trace_pci_nvme_virt_mngmt(nvme_cid(req
), act
, cntlid
, rt
? "VI" : "VQ", nr
);
6756 if (rt
!= NVME_VIRT_RES_QUEUE
&& rt
!= NVME_VIRT_RES_INTERRUPT
) {
6757 return NVME_INVALID_RESOURCE_ID
| NVME_DNR
;
6761 case NVME_VIRT_MNGMT_ACTION_SEC_ASSIGN
:
6762 return nvme_assign_virt_res_to_sec(n
, req
, cntlid
, rt
, nr
);
6763 case NVME_VIRT_MNGMT_ACTION_PRM_ALLOC
:
6764 return nvme_assign_virt_res_to_prim(n
, req
, cntlid
, rt
, nr
);
6765 case NVME_VIRT_MNGMT_ACTION_SEC_ONLINE
:
6766 return nvme_virt_set_state(n
, cntlid
, true);
6767 case NVME_VIRT_MNGMT_ACTION_SEC_OFFLINE
:
6768 return nvme_virt_set_state(n
, cntlid
, false);
6770 return NVME_INVALID_FIELD
| NVME_DNR
;
6774 static uint16_t nvme_dbbuf_config(NvmeCtrl
*n
, const NvmeRequest
*req
)
6776 PCIDevice
*pci
= PCI_DEVICE(n
);
6777 uint64_t dbs_addr
= le64_to_cpu(req
->cmd
.dptr
.prp1
);
6778 uint64_t eis_addr
= le64_to_cpu(req
->cmd
.dptr
.prp2
);
6781 /* Address should be page aligned */
6782 if (dbs_addr
& (n
->page_size
- 1) || eis_addr
& (n
->page_size
- 1)) {
6783 return NVME_INVALID_FIELD
| NVME_DNR
;
6786 /* Save shadow buffer base addr for use during queue creation */
6787 n
->dbbuf_dbs
= dbs_addr
;
6788 n
->dbbuf_eis
= eis_addr
;
6789 n
->dbbuf_enabled
= true;
6791 for (i
= 0; i
< n
->params
.max_ioqpairs
+ 1; i
++) {
6792 NvmeSQueue
*sq
= n
->sq
[i
];
6793 NvmeCQueue
*cq
= n
->cq
[i
];
6797 * CAP.DSTRD is 0, so offset of ith sq db_addr is (i<<3)
6798 * nvme_process_db() uses this hard-coded way to calculate
6799 * doorbell offsets. Be consistent with that here.
6801 sq
->db_addr
= dbs_addr
+ (i
<< 3);
6802 sq
->ei_addr
= eis_addr
+ (i
<< 3);
6803 pci_dma_write(pci
, sq
->db_addr
, &sq
->tail
, sizeof(sq
->tail
));
6805 if (n
->params
.ioeventfd
&& sq
->sqid
!= 0) {
6806 if (!nvme_init_sq_ioeventfd(sq
)) {
6807 sq
->ioeventfd_enabled
= true;
6813 /* CAP.DSTRD is 0, so offset of ith cq db_addr is (i<<3)+(1<<2) */
6814 cq
->db_addr
= dbs_addr
+ (i
<< 3) + (1 << 2);
6815 cq
->ei_addr
= eis_addr
+ (i
<< 3) + (1 << 2);
6816 pci_dma_write(pci
, cq
->db_addr
, &cq
->head
, sizeof(cq
->head
));
6818 if (n
->params
.ioeventfd
&& cq
->cqid
!= 0) {
6819 if (!nvme_init_cq_ioeventfd(cq
)) {
6820 cq
->ioeventfd_enabled
= true;
6826 trace_pci_nvme_dbbuf_config(dbs_addr
, eis_addr
);
6828 return NVME_SUCCESS
;
6831 static uint16_t nvme_directive_send(NvmeCtrl
*n
, NvmeRequest
*req
)
6833 return NVME_INVALID_FIELD
| NVME_DNR
;
6836 static uint16_t nvme_directive_receive(NvmeCtrl
*n
, NvmeRequest
*req
)
6839 uint32_t dw10
= le32_to_cpu(req
->cmd
.cdw10
);
6840 uint32_t dw11
= le32_to_cpu(req
->cmd
.cdw11
);
6841 uint32_t nsid
= le32_to_cpu(req
->cmd
.nsid
);
6842 uint8_t doper
, dtype
;
6843 uint32_t numd
, trans_len
;
6844 NvmeDirectiveIdentify id
= {
6845 .supported
= 1 << NVME_DIRECTIVE_IDENTIFY
,
6846 .enabled
= 1 << NVME_DIRECTIVE_IDENTIFY
,
6850 doper
= dw11
& 0xff;
6851 dtype
= (dw11
>> 8) & 0xff;
6853 trans_len
= MIN(sizeof(NvmeDirectiveIdentify
), numd
<< 2);
6855 if (nsid
== NVME_NSID_BROADCAST
|| dtype
!= NVME_DIRECTIVE_IDENTIFY
||
6856 doper
!= NVME_DIRECTIVE_RETURN_PARAMS
) {
6857 return NVME_INVALID_FIELD
| NVME_DNR
;
6860 ns
= nvme_ns(n
, nsid
);
6862 return NVME_INVALID_FIELD
| NVME_DNR
;
6866 case NVME_DIRECTIVE_IDENTIFY
:
6868 case NVME_DIRECTIVE_RETURN_PARAMS
:
6869 if (ns
->endgrp
->fdp
.enabled
) {
6870 id
.supported
|= 1 << NVME_DIRECTIVE_DATA_PLACEMENT
;
6871 id
.enabled
|= 1 << NVME_DIRECTIVE_DATA_PLACEMENT
;
6872 id
.persistent
|= 1 << NVME_DIRECTIVE_DATA_PLACEMENT
;
6875 return nvme_c2h(n
, (uint8_t *)&id
, trans_len
, req
);
6878 return NVME_INVALID_FIELD
| NVME_DNR
;
6882 return NVME_INVALID_FIELD
;
6886 static uint16_t nvme_admin_cmd(NvmeCtrl
*n
, NvmeRequest
*req
)
6888 trace_pci_nvme_admin_cmd(nvme_cid(req
), nvme_sqid(req
), req
->cmd
.opcode
,
6889 nvme_adm_opc_str(req
->cmd
.opcode
));
6891 if (!(nvme_cse_acs
[req
->cmd
.opcode
] & NVME_CMD_EFF_CSUPP
)) {
6892 trace_pci_nvme_err_invalid_admin_opc(req
->cmd
.opcode
);
6893 return NVME_INVALID_OPCODE
| NVME_DNR
;
6896 /* SGLs shall not be used for Admin commands in NVMe over PCIe */
6897 if (NVME_CMD_FLAGS_PSDT(req
->cmd
.flags
) != NVME_PSDT_PRP
) {
6898 return NVME_INVALID_FIELD
| NVME_DNR
;
6901 if (NVME_CMD_FLAGS_FUSE(req
->cmd
.flags
)) {
6902 return NVME_INVALID_FIELD
;
6905 switch (req
->cmd
.opcode
) {
6906 case NVME_ADM_CMD_DELETE_SQ
:
6907 return nvme_del_sq(n
, req
);
6908 case NVME_ADM_CMD_CREATE_SQ
:
6909 return nvme_create_sq(n
, req
);
6910 case NVME_ADM_CMD_GET_LOG_PAGE
:
6911 return nvme_get_log(n
, req
);
6912 case NVME_ADM_CMD_DELETE_CQ
:
6913 return nvme_del_cq(n
, req
);
6914 case NVME_ADM_CMD_CREATE_CQ
:
6915 return nvme_create_cq(n
, req
);
6916 case NVME_ADM_CMD_IDENTIFY
:
6917 return nvme_identify(n
, req
);
6918 case NVME_ADM_CMD_ABORT
:
6919 return nvme_abort(n
, req
);
6920 case NVME_ADM_CMD_SET_FEATURES
:
6921 return nvme_set_feature(n
, req
);
6922 case NVME_ADM_CMD_GET_FEATURES
:
6923 return nvme_get_feature(n
, req
);
6924 case NVME_ADM_CMD_ASYNC_EV_REQ
:
6925 return nvme_aer(n
, req
);
6926 case NVME_ADM_CMD_NS_ATTACHMENT
:
6927 return nvme_ns_attachment(n
, req
);
6928 case NVME_ADM_CMD_VIRT_MNGMT
:
6929 return nvme_virt_mngmt(n
, req
);
6930 case NVME_ADM_CMD_DBBUF_CONFIG
:
6931 return nvme_dbbuf_config(n
, req
);
6932 case NVME_ADM_CMD_FORMAT_NVM
:
6933 return nvme_format(n
, req
);
6934 case NVME_ADM_CMD_DIRECTIVE_SEND
:
6935 return nvme_directive_send(n
, req
);
6936 case NVME_ADM_CMD_DIRECTIVE_RECV
:
6937 return nvme_directive_receive(n
, req
);
6942 return NVME_INVALID_OPCODE
| NVME_DNR
;
6945 static void nvme_update_sq_eventidx(const NvmeSQueue
*sq
)
6947 uint32_t v
= cpu_to_le32(sq
->tail
);
6949 trace_pci_nvme_update_sq_eventidx(sq
->sqid
, sq
->tail
);
6951 pci_dma_write(PCI_DEVICE(sq
->ctrl
), sq
->ei_addr
, &v
, sizeof(v
));
6954 static void nvme_update_sq_tail(NvmeSQueue
*sq
)
6958 pci_dma_read(PCI_DEVICE(sq
->ctrl
), sq
->db_addr
, &v
, sizeof(v
));
6960 sq
->tail
= le32_to_cpu(v
);
6962 trace_pci_nvme_update_sq_tail(sq
->sqid
, sq
->tail
);
6965 static void nvme_process_sq(void *opaque
)
6967 NvmeSQueue
*sq
= opaque
;
6968 NvmeCtrl
*n
= sq
->ctrl
;
6969 NvmeCQueue
*cq
= n
->cq
[sq
->cqid
];
6976 if (n
->dbbuf_enabled
) {
6977 nvme_update_sq_tail(sq
);
6980 while (!(nvme_sq_empty(sq
) || QTAILQ_EMPTY(&sq
->req_list
))) {
6981 addr
= sq
->dma_addr
+ sq
->head
* n
->sqe_size
;
6982 if (nvme_addr_read(n
, addr
, (void *)&cmd
, sizeof(cmd
))) {
6983 trace_pci_nvme_err_addr_read(addr
);
6984 trace_pci_nvme_err_cfs();
6985 stl_le_p(&n
->bar
.csts
, NVME_CSTS_FAILED
);
6988 nvme_inc_sq_head(sq
);
6990 req
= QTAILQ_FIRST(&sq
->req_list
);
6991 QTAILQ_REMOVE(&sq
->req_list
, req
, entry
);
6992 QTAILQ_INSERT_TAIL(&sq
->out_req_list
, req
, entry
);
6993 nvme_req_clear(req
);
6994 req
->cqe
.cid
= cmd
.cid
;
6995 memcpy(&req
->cmd
, &cmd
, sizeof(NvmeCmd
));
6997 status
= sq
->sqid
? nvme_io_cmd(n
, req
) :
6998 nvme_admin_cmd(n
, req
);
6999 if (status
!= NVME_NO_COMPLETE
) {
7000 req
->status
= status
;
7001 nvme_enqueue_req_completion(cq
, req
);
7004 if (n
->dbbuf_enabled
) {
7005 nvme_update_sq_eventidx(sq
);
7006 nvme_update_sq_tail(sq
);
7011 static void nvme_update_msixcap_ts(PCIDevice
*pci_dev
, uint32_t table_size
)
7015 if (!msix_present(pci_dev
)) {
7019 assert(table_size
> 0 && table_size
<= pci_dev
->msix_entries_nr
);
7021 config
= pci_dev
->config
+ pci_dev
->msix_cap
;
7022 pci_set_word_by_mask(config
+ PCI_MSIX_FLAGS
, PCI_MSIX_FLAGS_QSIZE
,
7026 static void nvme_activate_virt_res(NvmeCtrl
*n
)
7028 PCIDevice
*pci_dev
= PCI_DEVICE(n
);
7029 NvmePriCtrlCap
*cap
= &n
->pri_ctrl_cap
;
7030 NvmeSecCtrlEntry
*sctrl
;
7032 /* -1 to account for the admin queue */
7033 if (pci_is_vf(pci_dev
)) {
7034 sctrl
= nvme_sctrl(n
);
7035 cap
->vqprt
= sctrl
->nvq
;
7036 cap
->viprt
= sctrl
->nvi
;
7037 n
->conf_ioqpairs
= sctrl
->nvq
? le16_to_cpu(sctrl
->nvq
) - 1 : 0;
7038 n
->conf_msix_qsize
= sctrl
->nvi
? le16_to_cpu(sctrl
->nvi
) : 1;
7040 cap
->vqrfap
= n
->next_pri_ctrl_cap
.vqrfap
;
7041 cap
->virfap
= n
->next_pri_ctrl_cap
.virfap
;
7042 n
->conf_ioqpairs
= le16_to_cpu(cap
->vqprt
) +
7043 le16_to_cpu(cap
->vqrfap
) - 1;
7044 n
->conf_msix_qsize
= le16_to_cpu(cap
->viprt
) +
7045 le16_to_cpu(cap
->virfap
);
7049 static void nvme_ctrl_reset(NvmeCtrl
*n
, NvmeResetType rst
)
7051 PCIDevice
*pci_dev
= PCI_DEVICE(n
);
7052 NvmeSecCtrlEntry
*sctrl
;
7056 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
7065 for (i
= 0; i
< n
->params
.max_ioqpairs
+ 1; i
++) {
7066 if (n
->sq
[i
] != NULL
) {
7067 nvme_free_sq(n
->sq
[i
], n
);
7070 for (i
= 0; i
< n
->params
.max_ioqpairs
+ 1; i
++) {
7071 if (n
->cq
[i
] != NULL
) {
7072 nvme_free_cq(n
->cq
[i
], n
);
7076 while (!QTAILQ_EMPTY(&n
->aer_queue
)) {
7077 NvmeAsyncEvent
*event
= QTAILQ_FIRST(&n
->aer_queue
);
7078 QTAILQ_REMOVE(&n
->aer_queue
, event
, entry
);
7082 if (n
->params
.sriov_max_vfs
) {
7083 if (!pci_is_vf(pci_dev
)) {
7084 for (i
= 0; i
< n
->sec_ctrl_list
.numcntl
; i
++) {
7085 sctrl
= &n
->sec_ctrl_list
.sec
[i
];
7086 nvme_virt_set_state(n
, le16_to_cpu(sctrl
->scid
), false);
7089 if (rst
!= NVME_RESET_CONTROLLER
) {
7090 pcie_sriov_pf_disable_vfs(pci_dev
);
7094 if (rst
!= NVME_RESET_CONTROLLER
) {
7095 nvme_activate_virt_res(n
);
7101 n
->outstanding_aers
= 0;
7102 n
->qs_created
= false;
7104 nvme_update_msixcap_ts(pci_dev
, n
->conf_msix_qsize
);
7106 if (pci_is_vf(pci_dev
)) {
7107 sctrl
= nvme_sctrl(n
);
7109 stl_le_p(&n
->bar
.csts
, sctrl
->scs
? 0 : NVME_CSTS_FAILED
);
7111 stl_le_p(&n
->bar
.csts
, 0);
7114 stl_le_p(&n
->bar
.intms
, 0);
7115 stl_le_p(&n
->bar
.intmc
, 0);
7116 stl_le_p(&n
->bar
.cc
, 0);
7120 n
->dbbuf_enabled
= false;
7123 static void nvme_ctrl_shutdown(NvmeCtrl
*n
)
7129 memory_region_msync(&n
->pmr
.dev
->mr
, 0, n
->pmr
.dev
->size
);
7132 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
7138 nvme_ns_shutdown(ns
);
7142 static void nvme_select_iocs(NvmeCtrl
*n
)
7147 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
7153 nvme_select_iocs_ns(n
, ns
);
7157 static int nvme_start_ctrl(NvmeCtrl
*n
)
7159 uint64_t cap
= ldq_le_p(&n
->bar
.cap
);
7160 uint32_t cc
= ldl_le_p(&n
->bar
.cc
);
7161 uint32_t aqa
= ldl_le_p(&n
->bar
.aqa
);
7162 uint64_t asq
= ldq_le_p(&n
->bar
.asq
);
7163 uint64_t acq
= ldq_le_p(&n
->bar
.acq
);
7164 uint32_t page_bits
= NVME_CC_MPS(cc
) + 12;
7165 uint32_t page_size
= 1 << page_bits
;
7166 NvmeSecCtrlEntry
*sctrl
= nvme_sctrl(n
);
7168 if (pci_is_vf(PCI_DEVICE(n
)) && !sctrl
->scs
) {
7169 trace_pci_nvme_err_startfail_virt_state(le16_to_cpu(sctrl
->nvi
),
7170 le16_to_cpu(sctrl
->nvq
));
7173 if (unlikely(n
->cq
[0])) {
7174 trace_pci_nvme_err_startfail_cq();
7177 if (unlikely(n
->sq
[0])) {
7178 trace_pci_nvme_err_startfail_sq();
7181 if (unlikely(asq
& (page_size
- 1))) {
7182 trace_pci_nvme_err_startfail_asq_misaligned(asq
);
7185 if (unlikely(acq
& (page_size
- 1))) {
7186 trace_pci_nvme_err_startfail_acq_misaligned(acq
);
7189 if (unlikely(!(NVME_CAP_CSS(cap
) & (1 << NVME_CC_CSS(cc
))))) {
7190 trace_pci_nvme_err_startfail_css(NVME_CC_CSS(cc
));
7193 if (unlikely(NVME_CC_MPS(cc
) < NVME_CAP_MPSMIN(cap
))) {
7194 trace_pci_nvme_err_startfail_page_too_small(
7196 NVME_CAP_MPSMIN(cap
));
7199 if (unlikely(NVME_CC_MPS(cc
) >
7200 NVME_CAP_MPSMAX(cap
))) {
7201 trace_pci_nvme_err_startfail_page_too_large(
7203 NVME_CAP_MPSMAX(cap
));
7206 if (unlikely(NVME_CC_IOCQES(cc
) <
7207 NVME_CTRL_CQES_MIN(n
->id_ctrl
.cqes
))) {
7208 trace_pci_nvme_err_startfail_cqent_too_small(
7210 NVME_CTRL_CQES_MIN(cap
));
7213 if (unlikely(NVME_CC_IOCQES(cc
) >
7214 NVME_CTRL_CQES_MAX(n
->id_ctrl
.cqes
))) {
7215 trace_pci_nvme_err_startfail_cqent_too_large(
7217 NVME_CTRL_CQES_MAX(cap
));
7220 if (unlikely(NVME_CC_IOSQES(cc
) <
7221 NVME_CTRL_SQES_MIN(n
->id_ctrl
.sqes
))) {
7222 trace_pci_nvme_err_startfail_sqent_too_small(
7224 NVME_CTRL_SQES_MIN(cap
));
7227 if (unlikely(NVME_CC_IOSQES(cc
) >
7228 NVME_CTRL_SQES_MAX(n
->id_ctrl
.sqes
))) {
7229 trace_pci_nvme_err_startfail_sqent_too_large(
7231 NVME_CTRL_SQES_MAX(cap
));
7234 if (unlikely(!NVME_AQA_ASQS(aqa
))) {
7235 trace_pci_nvme_err_startfail_asqent_sz_zero();
7238 if (unlikely(!NVME_AQA_ACQS(aqa
))) {
7239 trace_pci_nvme_err_startfail_acqent_sz_zero();
7243 n
->page_bits
= page_bits
;
7244 n
->page_size
= page_size
;
7245 n
->max_prp_ents
= n
->page_size
/ sizeof(uint64_t);
7246 n
->cqe_size
= 1 << NVME_CC_IOCQES(cc
);
7247 n
->sqe_size
= 1 << NVME_CC_IOSQES(cc
);
7248 nvme_init_cq(&n
->admin_cq
, n
, acq
, 0, 0, NVME_AQA_ACQS(aqa
) + 1, 1);
7249 nvme_init_sq(&n
->admin_sq
, n
, asq
, 0, 0, NVME_AQA_ASQS(aqa
) + 1);
7251 nvme_set_timestamp(n
, 0ULL);
7253 nvme_select_iocs(n
);
7258 static void nvme_cmb_enable_regs(NvmeCtrl
*n
)
7260 uint32_t cmbloc
= ldl_le_p(&n
->bar
.cmbloc
);
7261 uint32_t cmbsz
= ldl_le_p(&n
->bar
.cmbsz
);
7263 NVME_CMBLOC_SET_CDPCILS(cmbloc
, 1);
7264 NVME_CMBLOC_SET_CDPMLS(cmbloc
, 1);
7265 NVME_CMBLOC_SET_BIR(cmbloc
, NVME_CMB_BIR
);
7266 stl_le_p(&n
->bar
.cmbloc
, cmbloc
);
7268 NVME_CMBSZ_SET_SQS(cmbsz
, 1);
7269 NVME_CMBSZ_SET_CQS(cmbsz
, 0);
7270 NVME_CMBSZ_SET_LISTS(cmbsz
, 1);
7271 NVME_CMBSZ_SET_RDS(cmbsz
, 1);
7272 NVME_CMBSZ_SET_WDS(cmbsz
, 1);
7273 NVME_CMBSZ_SET_SZU(cmbsz
, 2); /* MBs */
7274 NVME_CMBSZ_SET_SZ(cmbsz
, n
->params
.cmb_size_mb
);
7275 stl_le_p(&n
->bar
.cmbsz
, cmbsz
);
7278 static void nvme_write_bar(NvmeCtrl
*n
, hwaddr offset
, uint64_t data
,
7281 PCIDevice
*pci
= PCI_DEVICE(n
);
7282 uint64_t cap
= ldq_le_p(&n
->bar
.cap
);
7283 uint32_t cc
= ldl_le_p(&n
->bar
.cc
);
7284 uint32_t intms
= ldl_le_p(&n
->bar
.intms
);
7285 uint32_t csts
= ldl_le_p(&n
->bar
.csts
);
7286 uint32_t pmrsts
= ldl_le_p(&n
->bar
.pmrsts
);
7288 if (unlikely(offset
& (sizeof(uint32_t) - 1))) {
7289 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32
,
7290 "MMIO write not 32-bit aligned,"
7291 " offset=0x%"PRIx64
"", offset
);
7292 /* should be ignored, fall through for now */
7295 if (unlikely(size
< sizeof(uint32_t))) {
7296 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall
,
7297 "MMIO write smaller than 32-bits,"
7298 " offset=0x%"PRIx64
", size=%u",
7300 /* should be ignored, fall through for now */
7304 case NVME_REG_INTMS
:
7305 if (unlikely(msix_enabled(pci
))) {
7306 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix
,
7307 "undefined access to interrupt mask set"
7308 " when MSI-X is enabled");
7309 /* should be ignored, fall through for now */
7312 stl_le_p(&n
->bar
.intms
, intms
);
7313 n
->bar
.intmc
= n
->bar
.intms
;
7314 trace_pci_nvme_mmio_intm_set(data
& 0xffffffff, intms
);
7317 case NVME_REG_INTMC
:
7318 if (unlikely(msix_enabled(pci
))) {
7319 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix
,
7320 "undefined access to interrupt mask clr"
7321 " when MSI-X is enabled");
7322 /* should be ignored, fall through for now */
7325 stl_le_p(&n
->bar
.intms
, intms
);
7326 n
->bar
.intmc
= n
->bar
.intms
;
7327 trace_pci_nvme_mmio_intm_clr(data
& 0xffffffff, intms
);
7331 stl_le_p(&n
->bar
.cc
, data
);
7333 trace_pci_nvme_mmio_cfg(data
& 0xffffffff);
7335 if (NVME_CC_SHN(data
) && !(NVME_CC_SHN(cc
))) {
7336 trace_pci_nvme_mmio_shutdown_set();
7337 nvme_ctrl_shutdown(n
);
7338 csts
&= ~(CSTS_SHST_MASK
<< CSTS_SHST_SHIFT
);
7339 csts
|= NVME_CSTS_SHST_COMPLETE
;
7340 } else if (!NVME_CC_SHN(data
) && NVME_CC_SHN(cc
)) {
7341 trace_pci_nvme_mmio_shutdown_cleared();
7342 csts
&= ~(CSTS_SHST_MASK
<< CSTS_SHST_SHIFT
);
7345 if (NVME_CC_EN(data
) && !NVME_CC_EN(cc
)) {
7346 if (unlikely(nvme_start_ctrl(n
))) {
7347 trace_pci_nvme_err_startfail();
7348 csts
= NVME_CSTS_FAILED
;
7350 trace_pci_nvme_mmio_start_success();
7351 csts
= NVME_CSTS_READY
;
7353 } else if (!NVME_CC_EN(data
) && NVME_CC_EN(cc
)) {
7354 trace_pci_nvme_mmio_stopped();
7355 nvme_ctrl_reset(n
, NVME_RESET_CONTROLLER
);
7360 stl_le_p(&n
->bar
.csts
, csts
);
7364 if (data
& (1 << 4)) {
7365 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported
,
7366 "attempted to W1C CSTS.NSSRO"
7367 " but CAP.NSSRS is zero (not supported)");
7368 } else if (data
!= 0) {
7369 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts
,
7370 "attempted to set a read only bit"
7371 " of controller status");
7375 if (data
== 0x4e564d65) {
7376 trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
7378 /* The spec says that writes of other values have no effect */
7383 stl_le_p(&n
->bar
.aqa
, data
);
7384 trace_pci_nvme_mmio_aqattr(data
& 0xffffffff);
7387 stn_le_p(&n
->bar
.asq
, size
, data
);
7388 trace_pci_nvme_mmio_asqaddr(data
);
7390 case NVME_REG_ASQ
+ 4:
7391 stl_le_p((uint8_t *)&n
->bar
.asq
+ 4, data
);
7392 trace_pci_nvme_mmio_asqaddr_hi(data
, ldq_le_p(&n
->bar
.asq
));
7395 trace_pci_nvme_mmio_acqaddr(data
);
7396 stn_le_p(&n
->bar
.acq
, size
, data
);
7398 case NVME_REG_ACQ
+ 4:
7399 stl_le_p((uint8_t *)&n
->bar
.acq
+ 4, data
);
7400 trace_pci_nvme_mmio_acqaddr_hi(data
, ldq_le_p(&n
->bar
.acq
));
7402 case NVME_REG_CMBLOC
:
7403 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved
,
7404 "invalid write to reserved CMBLOC"
7405 " when CMBSZ is zero, ignored");
7407 case NVME_REG_CMBSZ
:
7408 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly
,
7409 "invalid write to read only CMBSZ, ignored");
7411 case NVME_REG_CMBMSC
:
7412 if (!NVME_CAP_CMBS(cap
)) {
7416 stn_le_p(&n
->bar
.cmbmsc
, size
, data
);
7417 n
->cmb
.cmse
= false;
7419 if (NVME_CMBMSC_CRE(data
)) {
7420 nvme_cmb_enable_regs(n
);
7422 if (NVME_CMBMSC_CMSE(data
)) {
7423 uint64_t cmbmsc
= ldq_le_p(&n
->bar
.cmbmsc
);
7424 hwaddr cba
= NVME_CMBMSC_CBA(cmbmsc
) << CMBMSC_CBA_SHIFT
;
7425 if (cba
+ int128_get64(n
->cmb
.mem
.size
) < cba
) {
7426 uint32_t cmbsts
= ldl_le_p(&n
->bar
.cmbsts
);
7427 NVME_CMBSTS_SET_CBAI(cmbsts
, 1);
7428 stl_le_p(&n
->bar
.cmbsts
, cmbsts
);
7441 case NVME_REG_CMBMSC
+ 4:
7442 stl_le_p((uint8_t *)&n
->bar
.cmbmsc
+ 4, data
);
7445 case NVME_REG_PMRCAP
:
7446 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly
,
7447 "invalid write to PMRCAP register, ignored");
7449 case NVME_REG_PMRCTL
:
7450 if (!NVME_CAP_PMRS(cap
)) {
7454 stl_le_p(&n
->bar
.pmrctl
, data
);
7455 if (NVME_PMRCTL_EN(data
)) {
7456 memory_region_set_enabled(&n
->pmr
.dev
->mr
, true);
7459 memory_region_set_enabled(&n
->pmr
.dev
->mr
, false);
7460 NVME_PMRSTS_SET_NRDY(pmrsts
, 1);
7461 n
->pmr
.cmse
= false;
7463 stl_le_p(&n
->bar
.pmrsts
, pmrsts
);
7465 case NVME_REG_PMRSTS
:
7466 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly
,
7467 "invalid write to PMRSTS register, ignored");
7469 case NVME_REG_PMREBS
:
7470 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly
,
7471 "invalid write to PMREBS register, ignored");
7473 case NVME_REG_PMRSWTP
:
7474 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly
,
7475 "invalid write to PMRSWTP register, ignored");
7477 case NVME_REG_PMRMSCL
:
7478 if (!NVME_CAP_PMRS(cap
)) {
7482 stl_le_p(&n
->bar
.pmrmscl
, data
);
7483 n
->pmr
.cmse
= false;
7485 if (NVME_PMRMSCL_CMSE(data
)) {
7486 uint64_t pmrmscu
= ldl_le_p(&n
->bar
.pmrmscu
);
7487 hwaddr cba
= pmrmscu
<< 32 |
7488 (NVME_PMRMSCL_CBA(data
) << PMRMSCL_CBA_SHIFT
);
7489 if (cba
+ int128_get64(n
->pmr
.dev
->mr
.size
) < cba
) {
7490 NVME_PMRSTS_SET_CBAI(pmrsts
, 1);
7491 stl_le_p(&n
->bar
.pmrsts
, pmrsts
);
7500 case NVME_REG_PMRMSCU
:
7501 if (!NVME_CAP_PMRS(cap
)) {
7505 stl_le_p(&n
->bar
.pmrmscu
, data
);
7508 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid
,
7509 "invalid MMIO write,"
7510 " offset=0x%"PRIx64
", data=%"PRIx64
"",
7516 static uint64_t nvme_mmio_read(void *opaque
, hwaddr addr
, unsigned size
)
7518 NvmeCtrl
*n
= (NvmeCtrl
*)opaque
;
7519 uint8_t *ptr
= (uint8_t *)&n
->bar
;
7521 trace_pci_nvme_mmio_read(addr
, size
);
7523 if (unlikely(addr
& (sizeof(uint32_t) - 1))) {
7524 NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32
,
7525 "MMIO read not 32-bit aligned,"
7526 " offset=0x%"PRIx64
"", addr
);
7527 /* should RAZ, fall through for now */
7528 } else if (unlikely(size
< sizeof(uint32_t))) {
7529 NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall
,
7530 "MMIO read smaller than 32-bits,"
7531 " offset=0x%"PRIx64
"", addr
);
7532 /* should RAZ, fall through for now */
7535 if (addr
> sizeof(n
->bar
) - size
) {
7536 NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs
,
7537 "MMIO read beyond last register,"
7538 " offset=0x%"PRIx64
", returning 0", addr
);
7543 if (pci_is_vf(PCI_DEVICE(n
)) && !nvme_sctrl(n
)->scs
&&
7544 addr
!= NVME_REG_CSTS
) {
7545 trace_pci_nvme_err_ignored_mmio_vf_offline(addr
, size
);
7550 * When PMRWBM bit 1 is set then read from
7551 * from PMRSTS should ensure prior writes
7552 * made it to persistent media
7554 if (addr
== NVME_REG_PMRSTS
&&
7555 (NVME_PMRCAP_PMRWBM(ldl_le_p(&n
->bar
.pmrcap
)) & 0x02)) {
7556 memory_region_msync(&n
->pmr
.dev
->mr
, 0, n
->pmr
.dev
->size
);
7559 return ldn_le_p(ptr
+ addr
, size
);
7562 static void nvme_process_db(NvmeCtrl
*n
, hwaddr addr
, int val
)
7564 PCIDevice
*pci
= PCI_DEVICE(n
);
7567 if (unlikely(addr
& ((1 << 2) - 1))) {
7568 NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned
,
7569 "doorbell write not 32-bit aligned,"
7570 " offset=0x%"PRIx64
", ignoring", addr
);
7574 if (((addr
- 0x1000) >> 2) & 1) {
7575 /* Completion queue doorbell write */
7577 uint16_t new_head
= val
& 0xffff;
7581 qid
= (addr
- (0x1000 + (1 << 2))) >> 3;
7582 if (unlikely(nvme_check_cqid(n
, qid
))) {
7583 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq
,
7584 "completion queue doorbell write"
7585 " for nonexistent queue,"
7586 " sqid=%"PRIu32
", ignoring", qid
);
7589 * NVM Express v1.3d, Section 4.1 state: "If host software writes
7590 * an invalid value to the Submission Queue Tail Doorbell or
7591 * Completion Queue Head Doorbell regiter and an Asynchronous Event
7592 * Request command is outstanding, then an asynchronous event is
7593 * posted to the Admin Completion Queue with a status code of
7594 * Invalid Doorbell Write Value."
7596 * Also note that the spec includes the "Invalid Doorbell Register"
7597 * status code, but nowhere does it specify when to use it.
7598 * However, it seems reasonable to use it here in a similar
7601 if (n
->outstanding_aers
) {
7602 nvme_enqueue_event(n
, NVME_AER_TYPE_ERROR
,
7603 NVME_AER_INFO_ERR_INVALID_DB_REGISTER
,
7604 NVME_LOG_ERROR_INFO
);
7611 if (unlikely(new_head
>= cq
->size
)) {
7612 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead
,
7613 "completion queue doorbell write value"
7614 " beyond queue size, sqid=%"PRIu32
","
7615 " new_head=%"PRIu16
", ignoring",
7618 if (n
->outstanding_aers
) {
7619 nvme_enqueue_event(n
, NVME_AER_TYPE_ERROR
,
7620 NVME_AER_INFO_ERR_INVALID_DB_VALUE
,
7621 NVME_LOG_ERROR_INFO
);
7627 trace_pci_nvme_mmio_doorbell_cq(cq
->cqid
, new_head
);
7629 start_sqs
= nvme_cq_full(cq
) ? 1 : 0;
7630 cq
->head
= new_head
;
7631 if (!qid
&& n
->dbbuf_enabled
) {
7632 pci_dma_write(pci
, cq
->db_addr
, &cq
->head
, sizeof(cq
->head
));
7636 QTAILQ_FOREACH(sq
, &cq
->sq_list
, entry
) {
7637 qemu_bh_schedule(sq
->bh
);
7639 qemu_bh_schedule(cq
->bh
);
7642 if (cq
->tail
== cq
->head
) {
7643 if (cq
->irq_enabled
) {
7647 nvme_irq_deassert(n
, cq
);
7650 /* Submission queue doorbell write */
7652 uint16_t new_tail
= val
& 0xffff;
7655 qid
= (addr
- 0x1000) >> 3;
7656 if (unlikely(nvme_check_sqid(n
, qid
))) {
7657 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq
,
7658 "submission queue doorbell write"
7659 " for nonexistent queue,"
7660 " sqid=%"PRIu32
", ignoring", qid
);
7662 if (n
->outstanding_aers
) {
7663 nvme_enqueue_event(n
, NVME_AER_TYPE_ERROR
,
7664 NVME_AER_INFO_ERR_INVALID_DB_REGISTER
,
7665 NVME_LOG_ERROR_INFO
);
7672 if (unlikely(new_tail
>= sq
->size
)) {
7673 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail
,
7674 "submission queue doorbell write value"
7675 " beyond queue size, sqid=%"PRIu32
","
7676 " new_tail=%"PRIu16
", ignoring",
7679 if (n
->outstanding_aers
) {
7680 nvme_enqueue_event(n
, NVME_AER_TYPE_ERROR
,
7681 NVME_AER_INFO_ERR_INVALID_DB_VALUE
,
7682 NVME_LOG_ERROR_INFO
);
7688 trace_pci_nvme_mmio_doorbell_sq(sq
->sqid
, new_tail
);
7690 sq
->tail
= new_tail
;
7691 if (!qid
&& n
->dbbuf_enabled
) {
7693 * The spec states "the host shall also update the controller's
7694 * corresponding doorbell property to match the value of that entry
7695 * in the Shadow Doorbell buffer."
7697 * Since this context is currently a VM trap, we can safely enforce
7698 * the requirement from the device side in case the host is
7701 * Note, we shouldn't have to do this, but various drivers
7702 * including ones that run on Linux, are not updating Admin Queues,
7703 * so we can't trust reading it for an appropriate sq tail.
7705 pci_dma_write(pci
, sq
->db_addr
, &sq
->tail
, sizeof(sq
->tail
));
7708 qemu_bh_schedule(sq
->bh
);
7712 static void nvme_mmio_write(void *opaque
, hwaddr addr
, uint64_t data
,
7715 NvmeCtrl
*n
= (NvmeCtrl
*)opaque
;
7717 trace_pci_nvme_mmio_write(addr
, data
, size
);
7719 if (pci_is_vf(PCI_DEVICE(n
)) && !nvme_sctrl(n
)->scs
&&
7720 addr
!= NVME_REG_CSTS
) {
7721 trace_pci_nvme_err_ignored_mmio_vf_offline(addr
, size
);
7725 if (addr
< sizeof(n
->bar
)) {
7726 nvme_write_bar(n
, addr
, data
, size
);
7728 nvme_process_db(n
, addr
, data
);
7732 static const MemoryRegionOps nvme_mmio_ops
= {
7733 .read
= nvme_mmio_read
,
7734 .write
= nvme_mmio_write
,
7735 .endianness
= DEVICE_LITTLE_ENDIAN
,
7737 .min_access_size
= 2,
7738 .max_access_size
= 8,
7742 static void nvme_cmb_write(void *opaque
, hwaddr addr
, uint64_t data
,
7745 NvmeCtrl
*n
= (NvmeCtrl
*)opaque
;
7746 stn_le_p(&n
->cmb
.buf
[addr
], size
, data
);
7749 static uint64_t nvme_cmb_read(void *opaque
, hwaddr addr
, unsigned size
)
7751 NvmeCtrl
*n
= (NvmeCtrl
*)opaque
;
7752 return ldn_le_p(&n
->cmb
.buf
[addr
], size
);
7755 static const MemoryRegionOps nvme_cmb_ops
= {
7756 .read
= nvme_cmb_read
,
7757 .write
= nvme_cmb_write
,
7758 .endianness
= DEVICE_LITTLE_ENDIAN
,
7760 .min_access_size
= 1,
7761 .max_access_size
= 8,
7765 static bool nvme_check_params(NvmeCtrl
*n
, Error
**errp
)
7767 NvmeParams
*params
= &n
->params
;
7769 if (params
->num_queues
) {
7770 warn_report("num_queues is deprecated; please use max_ioqpairs "
7773 params
->max_ioqpairs
= params
->num_queues
- 1;
7776 if (n
->namespace.blkconf
.blk
&& n
->subsys
) {
7777 error_setg(errp
, "subsystem support is unavailable with legacy "
7778 "namespace ('drive' property)");
7782 if (params
->max_ioqpairs
< 1 ||
7783 params
->max_ioqpairs
> NVME_MAX_IOQPAIRS
) {
7784 error_setg(errp
, "max_ioqpairs must be between 1 and %d",
7789 if (params
->msix_qsize
< 1 ||
7790 params
->msix_qsize
> PCI_MSIX_FLAGS_QSIZE
+ 1) {
7791 error_setg(errp
, "msix_qsize must be between 1 and %d",
7792 PCI_MSIX_FLAGS_QSIZE
+ 1);
7796 if (!params
->serial
) {
7797 error_setg(errp
, "serial property not set");
7802 if (host_memory_backend_is_mapped(n
->pmr
.dev
)) {
7803 error_setg(errp
, "can't use already busy memdev: %s",
7804 object_get_canonical_path_component(OBJECT(n
->pmr
.dev
)));
7808 if (!is_power_of_2(n
->pmr
.dev
->size
)) {
7809 error_setg(errp
, "pmr backend size needs to be power of 2 in size");
7813 host_memory_backend_set_mapped(n
->pmr
.dev
, true);
7816 if (n
->params
.zasl
> n
->params
.mdts
) {
7817 error_setg(errp
, "zoned.zasl (Zone Append Size Limit) must be less "
7818 "than or equal to mdts (Maximum Data Transfer Size)");
7822 if (!n
->params
.vsl
) {
7823 error_setg(errp
, "vsl must be non-zero");
7827 if (params
->sriov_max_vfs
) {
7829 error_setg(errp
, "subsystem is required for the use of SR-IOV");
7833 if (params
->sriov_max_vfs
> NVME_MAX_VFS
) {
7834 error_setg(errp
, "sriov_max_vfs must be between 0 and %d",
7839 if (params
->cmb_size_mb
) {
7840 error_setg(errp
, "CMB is not supported with SR-IOV");
7845 error_setg(errp
, "PMR is not supported with SR-IOV");
7849 if (!params
->sriov_vq_flexible
|| !params
->sriov_vi_flexible
) {
7850 error_setg(errp
, "both sriov_vq_flexible and sriov_vi_flexible"
7851 " must be set for the use of SR-IOV");
7855 if (params
->sriov_vq_flexible
< params
->sriov_max_vfs
* 2) {
7856 error_setg(errp
, "sriov_vq_flexible must be greater than or equal"
7857 " to %d (sriov_max_vfs * 2)", params
->sriov_max_vfs
* 2);
7861 if (params
->max_ioqpairs
< params
->sriov_vq_flexible
+ 2) {
7862 error_setg(errp
, "(max_ioqpairs - sriov_vq_flexible) must be"
7863 " greater than or equal to 2");
7867 if (params
->sriov_vi_flexible
< params
->sriov_max_vfs
) {
7868 error_setg(errp
, "sriov_vi_flexible must be greater than or equal"
7869 " to %d (sriov_max_vfs)", params
->sriov_max_vfs
);
7873 if (params
->msix_qsize
< params
->sriov_vi_flexible
+ 1) {
7874 error_setg(errp
, "(msix_qsize - sriov_vi_flexible) must be"
7875 " greater than or equal to 1");
7879 if (params
->sriov_max_vi_per_vf
&&
7880 (params
->sriov_max_vi_per_vf
- 1) % NVME_VF_RES_GRANULARITY
) {
7881 error_setg(errp
, "sriov_max_vi_per_vf must meet:"
7882 " (sriov_max_vi_per_vf - 1) %% %d == 0 and"
7883 " sriov_max_vi_per_vf >= 1", NVME_VF_RES_GRANULARITY
);
7887 if (params
->sriov_max_vq_per_vf
&&
7888 (params
->sriov_max_vq_per_vf
< 2 ||
7889 (params
->sriov_max_vq_per_vf
- 1) % NVME_VF_RES_GRANULARITY
)) {
7890 error_setg(errp
, "sriov_max_vq_per_vf must meet:"
7891 " (sriov_max_vq_per_vf - 1) %% %d == 0 and"
7892 " sriov_max_vq_per_vf >= 2", NVME_VF_RES_GRANULARITY
);
7900 static void nvme_init_state(NvmeCtrl
*n
)
7902 NvmePriCtrlCap
*cap
= &n
->pri_ctrl_cap
;
7903 NvmeSecCtrlList
*list
= &n
->sec_ctrl_list
;
7904 NvmeSecCtrlEntry
*sctrl
;
7905 PCIDevice
*pci
= PCI_DEVICE(n
);
7909 if (pci_is_vf(pci
)) {
7910 sctrl
= nvme_sctrl(n
);
7912 n
->conf_ioqpairs
= sctrl
->nvq
? le16_to_cpu(sctrl
->nvq
) - 1 : 0;
7913 n
->conf_msix_qsize
= sctrl
->nvi
? le16_to_cpu(sctrl
->nvi
) : 1;
7915 max_vfs
= n
->params
.sriov_max_vfs
;
7916 n
->conf_ioqpairs
= n
->params
.max_ioqpairs
;
7917 n
->conf_msix_qsize
= n
->params
.msix_qsize
;
7920 n
->sq
= g_new0(NvmeSQueue
*, n
->params
.max_ioqpairs
+ 1);
7921 n
->cq
= g_new0(NvmeCQueue
*, n
->params
.max_ioqpairs
+ 1);
7922 n
->temperature
= NVME_TEMPERATURE
;
7923 n
->features
.temp_thresh_hi
= NVME_TEMPERATURE_WARNING
;
7924 n
->starttime_ms
= qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
);
7925 n
->aer_reqs
= g_new0(NvmeRequest
*, n
->params
.aerl
+ 1);
7926 QTAILQ_INIT(&n
->aer_queue
);
7928 list
->numcntl
= cpu_to_le16(max_vfs
);
7929 for (i
= 0; i
< max_vfs
; i
++) {
7930 sctrl
= &list
->sec
[i
];
7931 sctrl
->pcid
= cpu_to_le16(n
->cntlid
);
7932 sctrl
->vfn
= cpu_to_le16(i
+ 1);
7935 cap
->cntlid
= cpu_to_le16(n
->cntlid
);
7936 cap
->crt
= NVME_CRT_VQ
| NVME_CRT_VI
;
7938 if (pci_is_vf(pci
)) {
7939 cap
->vqprt
= cpu_to_le16(1 + n
->conf_ioqpairs
);
7941 cap
->vqprt
= cpu_to_le16(1 + n
->params
.max_ioqpairs
-
7942 n
->params
.sriov_vq_flexible
);
7943 cap
->vqfrt
= cpu_to_le32(n
->params
.sriov_vq_flexible
);
7944 cap
->vqrfap
= cap
->vqfrt
;
7945 cap
->vqgran
= cpu_to_le16(NVME_VF_RES_GRANULARITY
);
7946 cap
->vqfrsm
= n
->params
.sriov_max_vq_per_vf
?
7947 cpu_to_le16(n
->params
.sriov_max_vq_per_vf
) :
7948 cap
->vqfrt
/ MAX(max_vfs
, 1);
7951 if (pci_is_vf(pci
)) {
7952 cap
->viprt
= cpu_to_le16(n
->conf_msix_qsize
);
7954 cap
->viprt
= cpu_to_le16(n
->params
.msix_qsize
-
7955 n
->params
.sriov_vi_flexible
);
7956 cap
->vifrt
= cpu_to_le32(n
->params
.sriov_vi_flexible
);
7957 cap
->virfap
= cap
->vifrt
;
7958 cap
->vigran
= cpu_to_le16(NVME_VF_RES_GRANULARITY
);
7959 cap
->vifrsm
= n
->params
.sriov_max_vi_per_vf
?
7960 cpu_to_le16(n
->params
.sriov_max_vi_per_vf
) :
7961 cap
->vifrt
/ MAX(max_vfs
, 1);
7965 static void nvme_init_cmb(NvmeCtrl
*n
, PCIDevice
*pci_dev
)
7967 uint64_t cmb_size
= n
->params
.cmb_size_mb
* MiB
;
7968 uint64_t cap
= ldq_le_p(&n
->bar
.cap
);
7970 n
->cmb
.buf
= g_malloc0(cmb_size
);
7971 memory_region_init_io(&n
->cmb
.mem
, OBJECT(n
), &nvme_cmb_ops
, n
,
7972 "nvme-cmb", cmb_size
);
7973 pci_register_bar(pci_dev
, NVME_CMB_BIR
,
7974 PCI_BASE_ADDRESS_SPACE_MEMORY
|
7975 PCI_BASE_ADDRESS_MEM_TYPE_64
|
7976 PCI_BASE_ADDRESS_MEM_PREFETCH
, &n
->cmb
.mem
);
7978 NVME_CAP_SET_CMBS(cap
, 1);
7979 stq_le_p(&n
->bar
.cap
, cap
);
7981 if (n
->params
.legacy_cmb
) {
7982 nvme_cmb_enable_regs(n
);
7987 static void nvme_init_pmr(NvmeCtrl
*n
, PCIDevice
*pci_dev
)
7989 uint32_t pmrcap
= ldl_le_p(&n
->bar
.pmrcap
);
7991 NVME_PMRCAP_SET_RDS(pmrcap
, 1);
7992 NVME_PMRCAP_SET_WDS(pmrcap
, 1);
7993 NVME_PMRCAP_SET_BIR(pmrcap
, NVME_PMR_BIR
);
7994 /* Turn on bit 1 support */
7995 NVME_PMRCAP_SET_PMRWBM(pmrcap
, 0x02);
7996 NVME_PMRCAP_SET_CMSS(pmrcap
, 1);
7997 stl_le_p(&n
->bar
.pmrcap
, pmrcap
);
7999 pci_register_bar(pci_dev
, NVME_PMR_BIR
,
8000 PCI_BASE_ADDRESS_SPACE_MEMORY
|
8001 PCI_BASE_ADDRESS_MEM_TYPE_64
|
8002 PCI_BASE_ADDRESS_MEM_PREFETCH
, &n
->pmr
.dev
->mr
);
8004 memory_region_set_enabled(&n
->pmr
.dev
->mr
, false);
8007 static uint64_t nvme_bar_size(unsigned total_queues
, unsigned total_irqs
,
8008 unsigned *msix_table_offset
,
8009 unsigned *msix_pba_offset
)
8011 uint64_t bar_size
, msix_table_size
, msix_pba_size
;
8013 bar_size
= sizeof(NvmeBar
) + 2 * total_queues
* NVME_DB_SIZE
;
8014 bar_size
= QEMU_ALIGN_UP(bar_size
, 4 * KiB
);
8016 if (msix_table_offset
) {
8017 *msix_table_offset
= bar_size
;
8020 msix_table_size
= PCI_MSIX_ENTRY_SIZE
* total_irqs
;
8021 bar_size
+= msix_table_size
;
8022 bar_size
= QEMU_ALIGN_UP(bar_size
, 4 * KiB
);
8024 if (msix_pba_offset
) {
8025 *msix_pba_offset
= bar_size
;
8028 msix_pba_size
= QEMU_ALIGN_UP(total_irqs
, 64) / 8;
8029 bar_size
+= msix_pba_size
;
8031 bar_size
= pow2ceil(bar_size
);
8035 static void nvme_init_sriov(NvmeCtrl
*n
, PCIDevice
*pci_dev
, uint16_t offset
)
8037 uint16_t vf_dev_id
= n
->params
.use_intel_id
?
8038 PCI_DEVICE_ID_INTEL_NVME
: PCI_DEVICE_ID_REDHAT_NVME
;
8039 NvmePriCtrlCap
*cap
= &n
->pri_ctrl_cap
;
8040 uint64_t bar_size
= nvme_bar_size(le16_to_cpu(cap
->vqfrsm
),
8041 le16_to_cpu(cap
->vifrsm
),
8044 pcie_sriov_pf_init(pci_dev
, offset
, "nvme", vf_dev_id
,
8045 n
->params
.sriov_max_vfs
, n
->params
.sriov_max_vfs
,
8046 NVME_VF_OFFSET
, NVME_VF_STRIDE
);
8048 pcie_sriov_pf_init_vf_bar(pci_dev
, 0, PCI_BASE_ADDRESS_SPACE_MEMORY
|
8049 PCI_BASE_ADDRESS_MEM_TYPE_64
, bar_size
);
8052 static int nvme_add_pm_capability(PCIDevice
*pci_dev
, uint8_t offset
)
8057 ret
= pci_add_capability(pci_dev
, PCI_CAP_ID_PM
, offset
,
8058 PCI_PM_SIZEOF
, &err
);
8060 error_report_err(err
);
8064 pci_set_word(pci_dev
->config
+ offset
+ PCI_PM_PMC
,
8065 PCI_PM_CAP_VER_1_2
);
8066 pci_set_word(pci_dev
->config
+ offset
+ PCI_PM_CTRL
,
8067 PCI_PM_CTRL_NO_SOFT_RESET
);
8068 pci_set_word(pci_dev
->wmask
+ offset
+ PCI_PM_CTRL
,
8069 PCI_PM_CTRL_STATE_MASK
);
8074 static bool nvme_init_pci(NvmeCtrl
*n
, PCIDevice
*pci_dev
, Error
**errp
)
8077 uint8_t *pci_conf
= pci_dev
->config
;
8079 unsigned msix_table_offset
, msix_pba_offset
;
8082 pci_conf
[PCI_INTERRUPT_PIN
] = 1;
8083 pci_config_set_prog_interface(pci_conf
, 0x2);
8085 if (n
->params
.use_intel_id
) {
8086 pci_config_set_vendor_id(pci_conf
, PCI_VENDOR_ID_INTEL
);
8087 pci_config_set_device_id(pci_conf
, PCI_DEVICE_ID_INTEL_NVME
);
8089 pci_config_set_vendor_id(pci_conf
, PCI_VENDOR_ID_REDHAT
);
8090 pci_config_set_device_id(pci_conf
, PCI_DEVICE_ID_REDHAT_NVME
);
8093 pci_config_set_class(pci_conf
, PCI_CLASS_STORAGE_EXPRESS
);
8094 nvme_add_pm_capability(pci_dev
, 0x60);
8095 pcie_endpoint_cap_init(pci_dev
, 0x80);
8096 pcie_cap_flr_init(pci_dev
);
8097 if (n
->params
.sriov_max_vfs
) {
8098 pcie_ari_init(pci_dev
, 0x100, 1);
8101 /* add one to max_ioqpairs to account for the admin queue pair */
8102 bar_size
= nvme_bar_size(n
->params
.max_ioqpairs
+ 1, n
->params
.msix_qsize
,
8103 &msix_table_offset
, &msix_pba_offset
);
8105 memory_region_init(&n
->bar0
, OBJECT(n
), "nvme-bar0", bar_size
);
8106 memory_region_init_io(&n
->iomem
, OBJECT(n
), &nvme_mmio_ops
, n
, "nvme",
8108 memory_region_add_subregion(&n
->bar0
, 0, &n
->iomem
);
8110 if (pci_is_vf(pci_dev
)) {
8111 pcie_sriov_vf_register_bar(pci_dev
, 0, &n
->bar0
);
8113 pci_register_bar(pci_dev
, 0, PCI_BASE_ADDRESS_SPACE_MEMORY
|
8114 PCI_BASE_ADDRESS_MEM_TYPE_64
, &n
->bar0
);
8116 ret
= msix_init(pci_dev
, n
->params
.msix_qsize
,
8117 &n
->bar0
, 0, msix_table_offset
,
8118 &n
->bar0
, 0, msix_pba_offset
, 0, errp
);
8119 if (ret
== -ENOTSUP
) {
8120 /* report that msix is not supported, but do not error out */
8121 warn_report_err(*errp
);
8123 } else if (ret
< 0) {
8124 /* propagate error to caller */
8128 nvme_update_msixcap_ts(pci_dev
, n
->conf_msix_qsize
);
8130 if (n
->params
.cmb_size_mb
) {
8131 nvme_init_cmb(n
, pci_dev
);
8135 nvme_init_pmr(n
, pci_dev
);
8138 if (!pci_is_vf(pci_dev
) && n
->params
.sriov_max_vfs
) {
8139 nvme_init_sriov(n
, pci_dev
, 0x120);
8145 static void nvme_init_subnqn(NvmeCtrl
*n
)
8147 NvmeSubsystem
*subsys
= n
->subsys
;
8148 NvmeIdCtrl
*id
= &n
->id_ctrl
;
8151 snprintf((char *)id
->subnqn
, sizeof(id
->subnqn
),
8152 "nqn.2019-08.org.qemu:%s", n
->params
.serial
);
8154 pstrcpy((char *)id
->subnqn
, sizeof(id
->subnqn
), (char*)subsys
->subnqn
);
8158 static void nvme_init_ctrl(NvmeCtrl
*n
, PCIDevice
*pci_dev
)
8160 NvmeIdCtrl
*id
= &n
->id_ctrl
;
8161 uint8_t *pci_conf
= pci_dev
->config
;
8162 uint64_t cap
= ldq_le_p(&n
->bar
.cap
);
8163 NvmeSecCtrlEntry
*sctrl
= nvme_sctrl(n
);
8166 id
->vid
= cpu_to_le16(pci_get_word(pci_conf
+ PCI_VENDOR_ID
));
8167 id
->ssvid
= cpu_to_le16(pci_get_word(pci_conf
+ PCI_SUBSYSTEM_VENDOR_ID
));
8168 strpadcpy((char *)id
->mn
, sizeof(id
->mn
), "QEMU NVMe Ctrl", ' ');
8169 strpadcpy((char *)id
->fr
, sizeof(id
->fr
), QEMU_VERSION
, ' ');
8170 strpadcpy((char *)id
->sn
, sizeof(id
->sn
), n
->params
.serial
, ' ');
8172 id
->cntlid
= cpu_to_le16(n
->cntlid
);
8174 id
->oaes
= cpu_to_le32(NVME_OAES_NS_ATTR
);
8175 ctratt
= NVME_CTRATT_ELBAS
;
8179 if (n
->params
.use_intel_id
) {
8189 id
->mdts
= n
->params
.mdts
;
8190 id
->ver
= cpu_to_le32(NVME_SPEC_VER
);
8192 cpu_to_le16(NVME_OACS_NS_MGMT
| NVME_OACS_FORMAT
| NVME_OACS_DBBUF
|
8193 NVME_OACS_DIRECTIVES
);
8194 id
->cntrltype
= 0x1;
8197 * Because the controller always completes the Abort command immediately,
8198 * there can never be more than one concurrently executing Abort command,
8199 * so this value is never used for anything. Note that there can easily be
8200 * many Abort commands in the queues, but they are not considered
8201 * "executing" until processed by nvme_abort.
8203 * The specification recommends a value of 3 for Abort Command Limit (four
8204 * concurrently outstanding Abort commands), so lets use that though it is
8208 id
->aerl
= n
->params
.aerl
;
8209 id
->frmw
= (NVME_NUM_FW_SLOTS
<< 1) | NVME_FRMW_SLOT1_RO
;
8210 id
->lpa
= NVME_LPA_NS_SMART
| NVME_LPA_CSE
| NVME_LPA_EXTENDED
;
8212 /* recommended default value (~70 C) */
8213 id
->wctemp
= cpu_to_le16(NVME_TEMPERATURE_WARNING
);
8214 id
->cctemp
= cpu_to_le16(NVME_TEMPERATURE_CRITICAL
);
8216 id
->sqes
= (0x6 << 4) | 0x6;
8217 id
->cqes
= (0x4 << 4) | 0x4;
8218 id
->nn
= cpu_to_le32(NVME_MAX_NAMESPACES
);
8219 id
->oncs
= cpu_to_le16(NVME_ONCS_WRITE_ZEROES
| NVME_ONCS_TIMESTAMP
|
8220 NVME_ONCS_FEATURES
| NVME_ONCS_DSM
|
8221 NVME_ONCS_COMPARE
| NVME_ONCS_COPY
);
8224 * NOTE: If this device ever supports a command set that does NOT use 0x0
8225 * as a Flush-equivalent operation, support for the broadcast NSID in Flush
8226 * should probably be removed.
8228 * See comment in nvme_io_cmd.
8230 id
->vwc
= NVME_VWC_NSID_BROADCAST_SUPPORT
| NVME_VWC_PRESENT
;
8232 id
->ocfs
= cpu_to_le16(NVME_OCFS_COPY_FORMAT_0
| NVME_OCFS_COPY_FORMAT_1
);
8233 id
->sgls
= cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN
);
8235 nvme_init_subnqn(n
);
8237 id
->psd
[0].mp
= cpu_to_le16(0x9c4);
8238 id
->psd
[0].enlat
= cpu_to_le32(0x10);
8239 id
->psd
[0].exlat
= cpu_to_le32(0x4);
8242 id
->cmic
|= NVME_CMIC_MULTI_CTRL
;
8243 ctratt
|= NVME_CTRATT_ENDGRPS
;
8245 id
->endgidmax
= cpu_to_le16(0x1);
8247 if (n
->subsys
->endgrp
.fdp
.enabled
) {
8248 ctratt
|= NVME_CTRATT_FDPS
;
8252 id
->ctratt
= cpu_to_le32(ctratt
);
8254 NVME_CAP_SET_MQES(cap
, 0x7ff);
8255 NVME_CAP_SET_CQR(cap
, 1);
8256 NVME_CAP_SET_TO(cap
, 0xf);
8257 NVME_CAP_SET_CSS(cap
, NVME_CAP_CSS_NVM
);
8258 NVME_CAP_SET_CSS(cap
, NVME_CAP_CSS_CSI_SUPP
);
8259 NVME_CAP_SET_CSS(cap
, NVME_CAP_CSS_ADMIN_ONLY
);
8260 NVME_CAP_SET_MPSMAX(cap
, 4);
8261 NVME_CAP_SET_CMBS(cap
, n
->params
.cmb_size_mb
? 1 : 0);
8262 NVME_CAP_SET_PMRS(cap
, n
->pmr
.dev
? 1 : 0);
8263 stq_le_p(&n
->bar
.cap
, cap
);
8265 stl_le_p(&n
->bar
.vs
, NVME_SPEC_VER
);
8266 n
->bar
.intmc
= n
->bar
.intms
= 0;
8268 if (pci_is_vf(pci_dev
) && !sctrl
->scs
) {
8269 stl_le_p(&n
->bar
.csts
, NVME_CSTS_FAILED
);
8273 static int nvme_init_subsys(NvmeCtrl
*n
, Error
**errp
)
8281 cntlid
= nvme_subsys_register_ctrl(n
, errp
);
8291 void nvme_attach_ns(NvmeCtrl
*n
, NvmeNamespace
*ns
)
8293 uint32_t nsid
= ns
->params
.nsid
;
8294 assert(nsid
&& nsid
<= NVME_MAX_NAMESPACES
);
8296 n
->namespaces
[nsid
] = ns
;
8299 n
->dmrsl
= MIN_NON_ZERO(n
->dmrsl
,
8300 BDRV_REQUEST_MAX_BYTES
/ nvme_l2b(ns
, 1));
8303 static void nvme_realize(PCIDevice
*pci_dev
, Error
**errp
)
8305 NvmeCtrl
*n
= NVME(pci_dev
);
8306 DeviceState
*dev
= DEVICE(pci_dev
);
8308 NvmeCtrl
*pn
= NVME(pcie_sriov_get_pf(pci_dev
));
8310 if (pci_is_vf(pci_dev
)) {
8312 * VFs derive settings from the parent. PF's lifespan exceeds
8313 * that of VF's, so it's safe to share params.serial.
8315 memcpy(&n
->params
, &pn
->params
, sizeof(NvmeParams
));
8316 n
->subsys
= pn
->subsys
;
8319 if (!nvme_check_params(n
, errp
)) {
8323 qbus_init(&n
->bus
, sizeof(NvmeBus
), TYPE_NVME_BUS
, dev
, dev
->id
);
8325 if (nvme_init_subsys(n
, errp
)) {
8329 if (!nvme_init_pci(n
, pci_dev
, errp
)) {
8332 nvme_init_ctrl(n
, pci_dev
);
8334 /* setup a namespace if the controller drive property was given */
8335 if (n
->namespace.blkconf
.blk
) {
8337 ns
->params
.nsid
= 1;
8339 if (nvme_ns_setup(ns
, errp
)) {
8343 nvme_attach_ns(n
, ns
);
8347 static void nvme_exit(PCIDevice
*pci_dev
)
8349 NvmeCtrl
*n
= NVME(pci_dev
);
8353 nvme_ctrl_reset(n
, NVME_RESET_FUNCTION
);
8356 for (i
= 1; i
<= NVME_MAX_NAMESPACES
; i
++) {
8363 nvme_subsys_unregister_ctrl(n
->subsys
, n
);
8368 g_free(n
->aer_reqs
);
8370 if (n
->params
.cmb_size_mb
) {
8375 host_memory_backend_set_mapped(n
->pmr
.dev
, false);
8378 if (!pci_is_vf(pci_dev
) && n
->params
.sriov_max_vfs
) {
8379 pcie_sriov_pf_exit(pci_dev
);
8382 msix_uninit(pci_dev
, &n
->bar0
, &n
->bar0
);
8383 memory_region_del_subregion(&n
->bar0
, &n
->iomem
);
8386 static Property nvme_props
[] = {
8387 DEFINE_BLOCK_PROPERTIES(NvmeCtrl
, namespace.blkconf
),
8388 DEFINE_PROP_LINK("pmrdev", NvmeCtrl
, pmr
.dev
, TYPE_MEMORY_BACKEND
,
8389 HostMemoryBackend
*),
8390 DEFINE_PROP_LINK("subsys", NvmeCtrl
, subsys
, TYPE_NVME_SUBSYS
,
8392 DEFINE_PROP_STRING("serial", NvmeCtrl
, params
.serial
),
8393 DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl
, params
.cmb_size_mb
, 0),
8394 DEFINE_PROP_UINT32("num_queues", NvmeCtrl
, params
.num_queues
, 0),
8395 DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl
, params
.max_ioqpairs
, 64),
8396 DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl
, params
.msix_qsize
, 65),
8397 DEFINE_PROP_UINT8("aerl", NvmeCtrl
, params
.aerl
, 3),
8398 DEFINE_PROP_UINT32("aer_max_queued", NvmeCtrl
, params
.aer_max_queued
, 64),
8399 DEFINE_PROP_UINT8("mdts", NvmeCtrl
, params
.mdts
, 7),
8400 DEFINE_PROP_UINT8("vsl", NvmeCtrl
, params
.vsl
, 7),
8401 DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl
, params
.use_intel_id
, false),
8402 DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl
, params
.legacy_cmb
, false),
8403 DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl
, params
.ioeventfd
, false),
8404 DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl
, params
.zasl
, 0),
8405 DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl
,
8406 params
.auto_transition_zones
, true),
8407 DEFINE_PROP_UINT8("sriov_max_vfs", NvmeCtrl
, params
.sriov_max_vfs
, 0),
8408 DEFINE_PROP_UINT16("sriov_vq_flexible", NvmeCtrl
,
8409 params
.sriov_vq_flexible
, 0),
8410 DEFINE_PROP_UINT16("sriov_vi_flexible", NvmeCtrl
,
8411 params
.sriov_vi_flexible
, 0),
8412 DEFINE_PROP_UINT8("sriov_max_vi_per_vf", NvmeCtrl
,
8413 params
.sriov_max_vi_per_vf
, 0),
8414 DEFINE_PROP_UINT8("sriov_max_vq_per_vf", NvmeCtrl
,
8415 params
.sriov_max_vq_per_vf
, 0),
8416 DEFINE_PROP_END_OF_LIST(),
8419 static void nvme_get_smart_warning(Object
*obj
, Visitor
*v
, const char *name
,
8420 void *opaque
, Error
**errp
)
8422 NvmeCtrl
*n
= NVME(obj
);
8423 uint8_t value
= n
->smart_critical_warning
;
8425 visit_type_uint8(v
, name
, &value
, errp
);
8428 static void nvme_set_smart_warning(Object
*obj
, Visitor
*v
, const char *name
,
8429 void *opaque
, Error
**errp
)
8431 NvmeCtrl
*n
= NVME(obj
);
8432 uint8_t value
, old_value
, cap
= 0, index
, event
;
8434 if (!visit_type_uint8(v
, name
, &value
, errp
)) {
8438 cap
= NVME_SMART_SPARE
| NVME_SMART_TEMPERATURE
| NVME_SMART_RELIABILITY
8439 | NVME_SMART_MEDIA_READ_ONLY
| NVME_SMART_FAILED_VOLATILE_MEDIA
;
8440 if (NVME_CAP_PMRS(ldq_le_p(&n
->bar
.cap
))) {
8441 cap
|= NVME_SMART_PMR_UNRELIABLE
;
8444 if ((value
& cap
) != value
) {
8445 error_setg(errp
, "unsupported smart critical warning bits: 0x%x",
8450 old_value
= n
->smart_critical_warning
;
8451 n
->smart_critical_warning
= value
;
8453 /* only inject new bits of smart critical warning */
8454 for (index
= 0; index
< NVME_SMART_WARN_MAX
; index
++) {
8456 if (value
& ~old_value
& event
)
8457 nvme_smart_event(n
, event
);
8461 static void nvme_pci_reset(DeviceState
*qdev
)
8463 PCIDevice
*pci_dev
= PCI_DEVICE(qdev
);
8464 NvmeCtrl
*n
= NVME(pci_dev
);
8466 trace_pci_nvme_pci_reset();
8467 nvme_ctrl_reset(n
, NVME_RESET_FUNCTION
);
8470 static void nvme_sriov_pre_write_ctrl(PCIDevice
*dev
, uint32_t address
,
8471 uint32_t val
, int len
)
8473 NvmeCtrl
*n
= NVME(dev
);
8474 NvmeSecCtrlEntry
*sctrl
;
8475 uint16_t sriov_cap
= dev
->exp
.sriov_cap
;
8476 uint32_t off
= address
- sriov_cap
;
8483 if (range_covers_byte(off
, len
, PCI_SRIOV_CTRL
)) {
8484 if (!(val
& PCI_SRIOV_CTRL_VFE
)) {
8485 num_vfs
= pci_get_word(dev
->config
+ sriov_cap
+ PCI_SRIOV_NUM_VF
);
8486 for (i
= 0; i
< num_vfs
; i
++) {
8487 sctrl
= &n
->sec_ctrl_list
.sec
[i
];
8488 nvme_virt_set_state(n
, le16_to_cpu(sctrl
->scid
), false);
8494 static void nvme_pci_write_config(PCIDevice
*dev
, uint32_t address
,
8495 uint32_t val
, int len
)
8497 nvme_sriov_pre_write_ctrl(dev
, address
, val
, len
);
8498 pci_default_write_config(dev
, address
, val
, len
);
8499 pcie_cap_flr_write_config(dev
, address
, val
, len
);
8502 static const VMStateDescription nvme_vmstate
= {
8507 static void nvme_class_init(ObjectClass
*oc
, void *data
)
8509 DeviceClass
*dc
= DEVICE_CLASS(oc
);
8510 PCIDeviceClass
*pc
= PCI_DEVICE_CLASS(oc
);
8512 pc
->realize
= nvme_realize
;
8513 pc
->config_write
= nvme_pci_write_config
;
8514 pc
->exit
= nvme_exit
;
8515 pc
->class_id
= PCI_CLASS_STORAGE_EXPRESS
;
8518 set_bit(DEVICE_CATEGORY_STORAGE
, dc
->categories
);
8519 dc
->desc
= "Non-Volatile Memory Express";
8520 device_class_set_props(dc
, nvme_props
);
8521 dc
->vmsd
= &nvme_vmstate
;
8522 dc
->reset
= nvme_pci_reset
;
8525 static void nvme_instance_init(Object
*obj
)
8527 NvmeCtrl
*n
= NVME(obj
);
8529 device_add_bootindex_property(obj
, &n
->namespace.blkconf
.bootindex
,
8530 "bootindex", "/namespace@1,0",
8533 object_property_add(obj
, "smart_critical_warning", "uint8",
8534 nvme_get_smart_warning
,
8535 nvme_set_smart_warning
, NULL
, NULL
);
8538 static const TypeInfo nvme_info
= {
8540 .parent
= TYPE_PCI_DEVICE
,
8541 .instance_size
= sizeof(NvmeCtrl
),
8542 .instance_init
= nvme_instance_init
,
8543 .class_init
= nvme_class_init
,
8544 .interfaces
= (InterfaceInfo
[]) {
8545 { INTERFACE_PCIE_DEVICE
},
8550 static const TypeInfo nvme_bus_info
= {
8551 .name
= TYPE_NVME_BUS
,
8553 .instance_size
= sizeof(NvmeBus
),
8556 static void nvme_register_types(void)
8558 type_register_static(&nvme_info
);
8559 type_register_static(&nvme_bus_info
);
8562 type_init(nvme_register_types
)