audio: deprecate -soundhw gus
[qemu/ar7.git] / hw / block / nvme.c
blob1aee042d4cb2540a50565f6f75feface2c3128bf
1 /*
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.
9 */
11 /**
12 * Reference Specs: http://www.nvmexpress.org, 1.2, 1.1, 1.0e
14 * http://www.nvmexpress.org/resources/
17 /**
18 * Usage: add options:
19 * -drive file=<file>,if=none,id=<drive_id>
20 * -device nvme,drive=<drive_id>,serial=<serial>,id=<id[optional]>, \
21 * cmb_size_mb=<cmb_size_mb[optional]>, \
22 * [pmrdev=<mem_backend_file_id>,] \
23 * max_ioqpairs=<N[optional]>
25 * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
26 * offset 0 in BAR2 and supports only WDS, RDS and SQS for now.
28 * cmb_size_mb= and pmrdev= options are mutually exclusive due to limitation
29 * in available BAR's. cmb_size_mb= will take precedence over pmrdev= when
30 * both provided.
31 * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
32 * For example:
33 * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
34 * size=<size> .... -device nvme,...,pmrdev=<mem_id>
37 #include "qemu/osdep.h"
38 #include "qemu/units.h"
39 #include "qemu/error-report.h"
40 #include "hw/block/block.h"
41 #include "hw/pci/msix.h"
42 #include "hw/pci/pci.h"
43 #include "hw/qdev-properties.h"
44 #include "migration/vmstate.h"
45 #include "sysemu/sysemu.h"
46 #include "qapi/error.h"
47 #include "qapi/visitor.h"
48 #include "sysemu/hostmem.h"
49 #include "sysemu/block-backend.h"
50 #include "exec/memory.h"
51 #include "qemu/log.h"
52 #include "qemu/module.h"
53 #include "qemu/cutils.h"
54 #include "trace.h"
55 #include "nvme.h"
57 #define NVME_MAX_IOQPAIRS 0xffff
58 #define NVME_REG_SIZE 0x1000
59 #define NVME_DB_SIZE 4
60 #define NVME_CMB_BIR 2
61 #define NVME_PMR_BIR 2
63 #define NVME_GUEST_ERR(trace, fmt, ...) \
64 do { \
65 (trace_##trace)(__VA_ARGS__); \
66 qemu_log_mask(LOG_GUEST_ERROR, #trace \
67 " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
68 } while (0)
70 static void nvme_process_sq(void *opaque);
72 static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
74 hwaddr low = n->ctrl_mem.addr;
75 hwaddr hi = n->ctrl_mem.addr + int128_get64(n->ctrl_mem.size);
77 return addr >= low && addr < hi;
80 static void nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
82 if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr)) {
83 memcpy(buf, (void *)&n->cmbuf[addr - n->ctrl_mem.addr], size);
84 return;
87 pci_dma_read(&n->parent_obj, addr, buf, size);
90 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
92 return sqid < n->params.max_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
95 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
97 return cqid < n->params.max_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
100 static void nvme_inc_cq_tail(NvmeCQueue *cq)
102 cq->tail++;
103 if (cq->tail >= cq->size) {
104 cq->tail = 0;
105 cq->phase = !cq->phase;
109 static void nvme_inc_sq_head(NvmeSQueue *sq)
111 sq->head = (sq->head + 1) % sq->size;
114 static uint8_t nvme_cq_full(NvmeCQueue *cq)
116 return (cq->tail + 1) % cq->size == cq->head;
119 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
121 return sq->head == sq->tail;
124 static void nvme_irq_check(NvmeCtrl *n)
126 if (msix_enabled(&(n->parent_obj))) {
127 return;
129 if (~n->bar.intms & n->irq_status) {
130 pci_irq_assert(&n->parent_obj);
131 } else {
132 pci_irq_deassert(&n->parent_obj);
136 static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
138 if (cq->irq_enabled) {
139 if (msix_enabled(&(n->parent_obj))) {
140 trace_pci_nvme_irq_msix(cq->vector);
141 msix_notify(&(n->parent_obj), cq->vector);
142 } else {
143 trace_pci_nvme_irq_pin();
144 assert(cq->vector < 32);
145 n->irq_status |= 1 << cq->vector;
146 nvme_irq_check(n);
148 } else {
149 trace_pci_nvme_irq_masked();
153 static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
155 if (cq->irq_enabled) {
156 if (msix_enabled(&(n->parent_obj))) {
157 return;
158 } else {
159 assert(cq->vector < 32);
160 n->irq_status &= ~(1 << cq->vector);
161 nvme_irq_check(n);
166 static uint16_t nvme_map_prp(QEMUSGList *qsg, QEMUIOVector *iov, uint64_t prp1,
167 uint64_t prp2, uint32_t len, NvmeCtrl *n)
169 hwaddr trans_len = n->page_size - (prp1 % n->page_size);
170 trans_len = MIN(len, trans_len);
171 int num_prps = (len >> n->page_bits) + 1;
173 if (unlikely(!prp1)) {
174 trace_pci_nvme_err_invalid_prp();
175 return NVME_INVALID_FIELD | NVME_DNR;
176 } else if (n->bar.cmbsz && prp1 >= n->ctrl_mem.addr &&
177 prp1 < n->ctrl_mem.addr + int128_get64(n->ctrl_mem.size)) {
178 qsg->nsg = 0;
179 qemu_iovec_init(iov, num_prps);
180 qemu_iovec_add(iov, (void *)&n->cmbuf[prp1 - n->ctrl_mem.addr], trans_len);
181 } else {
182 pci_dma_sglist_init(qsg, &n->parent_obj, num_prps);
183 qemu_sglist_add(qsg, prp1, trans_len);
185 len -= trans_len;
186 if (len) {
187 if (unlikely(!prp2)) {
188 trace_pci_nvme_err_invalid_prp2_missing();
189 goto unmap;
191 if (len > n->page_size) {
192 uint64_t prp_list[n->max_prp_ents];
193 uint32_t nents, prp_trans;
194 int i = 0;
196 nents = (len + n->page_size - 1) >> n->page_bits;
197 prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
198 nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
199 while (len != 0) {
200 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
202 if (i == n->max_prp_ents - 1 && len > n->page_size) {
203 if (unlikely(!prp_ent || prp_ent & (n->page_size - 1))) {
204 trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
205 goto unmap;
208 i = 0;
209 nents = (len + n->page_size - 1) >> n->page_bits;
210 prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
211 nvme_addr_read(n, prp_ent, (void *)prp_list,
212 prp_trans);
213 prp_ent = le64_to_cpu(prp_list[i]);
216 if (unlikely(!prp_ent || prp_ent & (n->page_size - 1))) {
217 trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
218 goto unmap;
221 trans_len = MIN(len, n->page_size);
222 if (qsg->nsg){
223 qemu_sglist_add(qsg, prp_ent, trans_len);
224 } else {
225 qemu_iovec_add(iov, (void *)&n->cmbuf[prp_ent - n->ctrl_mem.addr], trans_len);
227 len -= trans_len;
228 i++;
230 } else {
231 if (unlikely(prp2 & (n->page_size - 1))) {
232 trace_pci_nvme_err_invalid_prp2_align(prp2);
233 goto unmap;
235 if (qsg->nsg) {
236 qemu_sglist_add(qsg, prp2, len);
237 } else {
238 qemu_iovec_add(iov, (void *)&n->cmbuf[prp2 - n->ctrl_mem.addr], trans_len);
242 return NVME_SUCCESS;
244 unmap:
245 qemu_sglist_destroy(qsg);
246 return NVME_INVALID_FIELD | NVME_DNR;
249 static uint16_t nvme_dma_write_prp(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
250 uint64_t prp1, uint64_t prp2)
252 QEMUSGList qsg;
253 QEMUIOVector iov;
254 uint16_t status = NVME_SUCCESS;
256 if (nvme_map_prp(&qsg, &iov, prp1, prp2, len, n)) {
257 return NVME_INVALID_FIELD | NVME_DNR;
259 if (qsg.nsg > 0) {
260 if (dma_buf_write(ptr, len, &qsg)) {
261 status = NVME_INVALID_FIELD | NVME_DNR;
263 qemu_sglist_destroy(&qsg);
264 } else {
265 if (qemu_iovec_to_buf(&iov, 0, ptr, len) != len) {
266 status = NVME_INVALID_FIELD | NVME_DNR;
268 qemu_iovec_destroy(&iov);
270 return status;
273 static uint16_t nvme_dma_read_prp(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
274 uint64_t prp1, uint64_t prp2)
276 QEMUSGList qsg;
277 QEMUIOVector iov;
278 uint16_t status = NVME_SUCCESS;
280 trace_pci_nvme_dma_read(prp1, prp2);
282 if (nvme_map_prp(&qsg, &iov, prp1, prp2, len, n)) {
283 return NVME_INVALID_FIELD | NVME_DNR;
285 if (qsg.nsg > 0) {
286 if (unlikely(dma_buf_read(ptr, len, &qsg))) {
287 trace_pci_nvme_err_invalid_dma();
288 status = NVME_INVALID_FIELD | NVME_DNR;
290 qemu_sglist_destroy(&qsg);
291 } else {
292 if (unlikely(qemu_iovec_from_buf(&iov, 0, ptr, len) != len)) {
293 trace_pci_nvme_err_invalid_dma();
294 status = NVME_INVALID_FIELD | NVME_DNR;
296 qemu_iovec_destroy(&iov);
298 return status;
301 static void nvme_post_cqes(void *opaque)
303 NvmeCQueue *cq = opaque;
304 NvmeCtrl *n = cq->ctrl;
305 NvmeRequest *req, *next;
307 QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
308 NvmeSQueue *sq;
309 hwaddr addr;
311 if (nvme_cq_full(cq)) {
312 break;
315 QTAILQ_REMOVE(&cq->req_list, req, entry);
316 sq = req->sq;
317 req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
318 req->cqe.sq_id = cpu_to_le16(sq->sqid);
319 req->cqe.sq_head = cpu_to_le16(sq->head);
320 addr = cq->dma_addr + cq->tail * n->cqe_size;
321 nvme_inc_cq_tail(cq);
322 pci_dma_write(&n->parent_obj, addr, (void *)&req->cqe,
323 sizeof(req->cqe));
324 QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
326 if (cq->tail != cq->head) {
327 nvme_irq_assert(n, cq);
331 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
333 assert(cq->cqid == req->sq->cqid);
334 QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
335 QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
336 timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
339 static void nvme_rw_cb(void *opaque, int ret)
341 NvmeRequest *req = opaque;
342 NvmeSQueue *sq = req->sq;
343 NvmeCtrl *n = sq->ctrl;
344 NvmeCQueue *cq = n->cq[sq->cqid];
346 if (!ret) {
347 block_acct_done(blk_get_stats(n->conf.blk), &req->acct);
348 req->status = NVME_SUCCESS;
349 } else {
350 block_acct_failed(blk_get_stats(n->conf.blk), &req->acct);
351 req->status = NVME_INTERNAL_DEV_ERROR;
353 if (req->has_sg) {
354 qemu_sglist_destroy(&req->qsg);
356 nvme_enqueue_req_completion(cq, req);
359 static uint16_t nvme_flush(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
360 NvmeRequest *req)
362 req->has_sg = false;
363 block_acct_start(blk_get_stats(n->conf.blk), &req->acct, 0,
364 BLOCK_ACCT_FLUSH);
365 req->aiocb = blk_aio_flush(n->conf.blk, nvme_rw_cb, req);
367 return NVME_NO_COMPLETE;
370 static uint16_t nvme_write_zeros(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
371 NvmeRequest *req)
373 NvmeRwCmd *rw = (NvmeRwCmd *)cmd;
374 const uint8_t lba_index = NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas);
375 const uint8_t data_shift = ns->id_ns.lbaf[lba_index].ds;
376 uint64_t slba = le64_to_cpu(rw->slba);
377 uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
378 uint64_t offset = slba << data_shift;
379 uint32_t count = nlb << data_shift;
381 if (unlikely(slba + nlb > ns->id_ns.nsze)) {
382 trace_pci_nvme_err_invalid_lba_range(slba, nlb, ns->id_ns.nsze);
383 return NVME_LBA_RANGE | NVME_DNR;
386 req->has_sg = false;
387 block_acct_start(blk_get_stats(n->conf.blk), &req->acct, 0,
388 BLOCK_ACCT_WRITE);
389 req->aiocb = blk_aio_pwrite_zeroes(n->conf.blk, offset, count,
390 BDRV_REQ_MAY_UNMAP, nvme_rw_cb, req);
391 return NVME_NO_COMPLETE;
394 static uint16_t nvme_rw(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
395 NvmeRequest *req)
397 NvmeRwCmd *rw = (NvmeRwCmd *)cmd;
398 uint32_t nlb = le32_to_cpu(rw->nlb) + 1;
399 uint64_t slba = le64_to_cpu(rw->slba);
400 uint64_t prp1 = le64_to_cpu(rw->prp1);
401 uint64_t prp2 = le64_to_cpu(rw->prp2);
403 uint8_t lba_index = NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas);
404 uint8_t data_shift = ns->id_ns.lbaf[lba_index].ds;
405 uint64_t data_size = (uint64_t)nlb << data_shift;
406 uint64_t data_offset = slba << data_shift;
407 int is_write = rw->opcode == NVME_CMD_WRITE ? 1 : 0;
408 enum BlockAcctType acct = is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ;
410 trace_pci_nvme_rw(is_write ? "write" : "read", nlb, data_size, slba);
412 if (unlikely((slba + nlb) > ns->id_ns.nsze)) {
413 block_acct_invalid(blk_get_stats(n->conf.blk), acct);
414 trace_pci_nvme_err_invalid_lba_range(slba, nlb, ns->id_ns.nsze);
415 return NVME_LBA_RANGE | NVME_DNR;
418 if (nvme_map_prp(&req->qsg, &req->iov, prp1, prp2, data_size, n)) {
419 block_acct_invalid(blk_get_stats(n->conf.blk), acct);
420 return NVME_INVALID_FIELD | NVME_DNR;
423 dma_acct_start(n->conf.blk, &req->acct, &req->qsg, acct);
424 if (req->qsg.nsg > 0) {
425 req->has_sg = true;
426 req->aiocb = is_write ?
427 dma_blk_write(n->conf.blk, &req->qsg, data_offset, BDRV_SECTOR_SIZE,
428 nvme_rw_cb, req) :
429 dma_blk_read(n->conf.blk, &req->qsg, data_offset, BDRV_SECTOR_SIZE,
430 nvme_rw_cb, req);
431 } else {
432 req->has_sg = false;
433 req->aiocb = is_write ?
434 blk_aio_pwritev(n->conf.blk, data_offset, &req->iov, 0, nvme_rw_cb,
435 req) :
436 blk_aio_preadv(n->conf.blk, data_offset, &req->iov, 0, nvme_rw_cb,
437 req);
440 return NVME_NO_COMPLETE;
443 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
445 NvmeNamespace *ns;
446 uint32_t nsid = le32_to_cpu(cmd->nsid);
448 if (unlikely(nsid == 0 || nsid > n->num_namespaces)) {
449 trace_pci_nvme_err_invalid_ns(nsid, n->num_namespaces);
450 return NVME_INVALID_NSID | NVME_DNR;
453 ns = &n->namespaces[nsid - 1];
454 switch (cmd->opcode) {
455 case NVME_CMD_FLUSH:
456 return nvme_flush(n, ns, cmd, req);
457 case NVME_CMD_WRITE_ZEROS:
458 return nvme_write_zeros(n, ns, cmd, req);
459 case NVME_CMD_WRITE:
460 case NVME_CMD_READ:
461 return nvme_rw(n, ns, cmd, req);
462 default:
463 trace_pci_nvme_err_invalid_opc(cmd->opcode);
464 return NVME_INVALID_OPCODE | NVME_DNR;
468 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
470 n->sq[sq->sqid] = NULL;
471 timer_del(sq->timer);
472 timer_free(sq->timer);
473 g_free(sq->io_req);
474 if (sq->sqid) {
475 g_free(sq);
479 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeCmd *cmd)
481 NvmeDeleteQ *c = (NvmeDeleteQ *)cmd;
482 NvmeRequest *req, *next;
483 NvmeSQueue *sq;
484 NvmeCQueue *cq;
485 uint16_t qid = le16_to_cpu(c->qid);
487 if (unlikely(!qid || nvme_check_sqid(n, qid))) {
488 trace_pci_nvme_err_invalid_del_sq(qid);
489 return NVME_INVALID_QID | NVME_DNR;
492 trace_pci_nvme_del_sq(qid);
494 sq = n->sq[qid];
495 while (!QTAILQ_EMPTY(&sq->out_req_list)) {
496 req = QTAILQ_FIRST(&sq->out_req_list);
497 assert(req->aiocb);
498 blk_aio_cancel(req->aiocb);
500 if (!nvme_check_cqid(n, sq->cqid)) {
501 cq = n->cq[sq->cqid];
502 QTAILQ_REMOVE(&cq->sq_list, sq, entry);
504 nvme_post_cqes(cq);
505 QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
506 if (req->sq == sq) {
507 QTAILQ_REMOVE(&cq->req_list, req, entry);
508 QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
513 nvme_free_sq(sq, n);
514 return NVME_SUCCESS;
517 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
518 uint16_t sqid, uint16_t cqid, uint16_t size)
520 int i;
521 NvmeCQueue *cq;
523 sq->ctrl = n;
524 sq->dma_addr = dma_addr;
525 sq->sqid = sqid;
526 sq->size = size;
527 sq->cqid = cqid;
528 sq->head = sq->tail = 0;
529 sq->io_req = g_new(NvmeRequest, sq->size);
531 QTAILQ_INIT(&sq->req_list);
532 QTAILQ_INIT(&sq->out_req_list);
533 for (i = 0; i < sq->size; i++) {
534 sq->io_req[i].sq = sq;
535 QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
537 sq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_process_sq, sq);
539 assert(n->cq[cqid]);
540 cq = n->cq[cqid];
541 QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
542 n->sq[sqid] = sq;
545 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeCmd *cmd)
547 NvmeSQueue *sq;
548 NvmeCreateSq *c = (NvmeCreateSq *)cmd;
550 uint16_t cqid = le16_to_cpu(c->cqid);
551 uint16_t sqid = le16_to_cpu(c->sqid);
552 uint16_t qsize = le16_to_cpu(c->qsize);
553 uint16_t qflags = le16_to_cpu(c->sq_flags);
554 uint64_t prp1 = le64_to_cpu(c->prp1);
556 trace_pci_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
558 if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
559 trace_pci_nvme_err_invalid_create_sq_cqid(cqid);
560 return NVME_INVALID_CQID | NVME_DNR;
562 if (unlikely(!sqid || !nvme_check_sqid(n, sqid))) {
563 trace_pci_nvme_err_invalid_create_sq_sqid(sqid);
564 return NVME_INVALID_QID | NVME_DNR;
566 if (unlikely(!qsize || qsize > NVME_CAP_MQES(n->bar.cap))) {
567 trace_pci_nvme_err_invalid_create_sq_size(qsize);
568 return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
570 if (unlikely(!prp1 || prp1 & (n->page_size - 1))) {
571 trace_pci_nvme_err_invalid_create_sq_addr(prp1);
572 return NVME_INVALID_FIELD | NVME_DNR;
574 if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
575 trace_pci_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
576 return NVME_INVALID_FIELD | NVME_DNR;
578 sq = g_malloc0(sizeof(*sq));
579 nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
580 return NVME_SUCCESS;
583 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
585 n->cq[cq->cqid] = NULL;
586 timer_del(cq->timer);
587 timer_free(cq->timer);
588 msix_vector_unuse(&n->parent_obj, cq->vector);
589 if (cq->cqid) {
590 g_free(cq);
594 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeCmd *cmd)
596 NvmeDeleteQ *c = (NvmeDeleteQ *)cmd;
597 NvmeCQueue *cq;
598 uint16_t qid = le16_to_cpu(c->qid);
600 if (unlikely(!qid || nvme_check_cqid(n, qid))) {
601 trace_pci_nvme_err_invalid_del_cq_cqid(qid);
602 return NVME_INVALID_CQID | NVME_DNR;
605 cq = n->cq[qid];
606 if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
607 trace_pci_nvme_err_invalid_del_cq_notempty(qid);
608 return NVME_INVALID_QUEUE_DEL;
610 nvme_irq_deassert(n, cq);
611 trace_pci_nvme_del_cq(qid);
612 nvme_free_cq(cq, n);
613 return NVME_SUCCESS;
616 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
617 uint16_t cqid, uint16_t vector, uint16_t size, uint16_t irq_enabled)
619 int ret;
621 ret = msix_vector_use(&n->parent_obj, vector);
622 assert(ret == 0);
623 cq->ctrl = n;
624 cq->cqid = cqid;
625 cq->size = size;
626 cq->dma_addr = dma_addr;
627 cq->phase = 1;
628 cq->irq_enabled = irq_enabled;
629 cq->vector = vector;
630 cq->head = cq->tail = 0;
631 QTAILQ_INIT(&cq->req_list);
632 QTAILQ_INIT(&cq->sq_list);
633 n->cq[cqid] = cq;
634 cq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_post_cqes, cq);
637 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeCmd *cmd)
639 NvmeCQueue *cq;
640 NvmeCreateCq *c = (NvmeCreateCq *)cmd;
641 uint16_t cqid = le16_to_cpu(c->cqid);
642 uint16_t vector = le16_to_cpu(c->irq_vector);
643 uint16_t qsize = le16_to_cpu(c->qsize);
644 uint16_t qflags = le16_to_cpu(c->cq_flags);
645 uint64_t prp1 = le64_to_cpu(c->prp1);
647 trace_pci_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
648 NVME_CQ_FLAGS_IEN(qflags) != 0);
650 if (unlikely(!cqid || !nvme_check_cqid(n, cqid))) {
651 trace_pci_nvme_err_invalid_create_cq_cqid(cqid);
652 return NVME_INVALID_CQID | NVME_DNR;
654 if (unlikely(!qsize || qsize > NVME_CAP_MQES(n->bar.cap))) {
655 trace_pci_nvme_err_invalid_create_cq_size(qsize);
656 return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
658 if (unlikely(!prp1)) {
659 trace_pci_nvme_err_invalid_create_cq_addr(prp1);
660 return NVME_INVALID_FIELD | NVME_DNR;
662 if (unlikely(!msix_enabled(&n->parent_obj) && vector)) {
663 trace_pci_nvme_err_invalid_create_cq_vector(vector);
664 return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
666 if (unlikely(vector >= n->params.msix_qsize)) {
667 trace_pci_nvme_err_invalid_create_cq_vector(vector);
668 return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
670 if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
671 trace_pci_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
672 return NVME_INVALID_FIELD | NVME_DNR;
675 cq = g_malloc0(sizeof(*cq));
676 nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
677 NVME_CQ_FLAGS_IEN(qflags));
678 return NVME_SUCCESS;
681 static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeIdentify *c)
683 uint64_t prp1 = le64_to_cpu(c->prp1);
684 uint64_t prp2 = le64_to_cpu(c->prp2);
686 trace_pci_nvme_identify_ctrl();
688 return nvme_dma_read_prp(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl),
689 prp1, prp2);
692 static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeIdentify *c)
694 NvmeNamespace *ns;
695 uint32_t nsid = le32_to_cpu(c->nsid);
696 uint64_t prp1 = le64_to_cpu(c->prp1);
697 uint64_t prp2 = le64_to_cpu(c->prp2);
699 trace_pci_nvme_identify_ns(nsid);
701 if (unlikely(nsid == 0 || nsid > n->num_namespaces)) {
702 trace_pci_nvme_err_invalid_ns(nsid, n->num_namespaces);
703 return NVME_INVALID_NSID | NVME_DNR;
706 ns = &n->namespaces[nsid - 1];
708 return nvme_dma_read_prp(n, (uint8_t *)&ns->id_ns, sizeof(ns->id_ns),
709 prp1, prp2);
712 static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeIdentify *c)
714 static const int data_len = NVME_IDENTIFY_DATA_SIZE;
715 uint32_t min_nsid = le32_to_cpu(c->nsid);
716 uint64_t prp1 = le64_to_cpu(c->prp1);
717 uint64_t prp2 = le64_to_cpu(c->prp2);
718 uint32_t *list;
719 uint16_t ret;
720 int i, j = 0;
722 trace_pci_nvme_identify_nslist(min_nsid);
724 list = g_malloc0(data_len);
725 for (i = 0; i < n->num_namespaces; i++) {
726 if (i < min_nsid) {
727 continue;
729 list[j++] = cpu_to_le32(i + 1);
730 if (j == data_len / sizeof(uint32_t)) {
731 break;
734 ret = nvme_dma_read_prp(n, (uint8_t *)list, data_len, prp1, prp2);
735 g_free(list);
736 return ret;
739 static uint16_t nvme_identify(NvmeCtrl *n, NvmeCmd *cmd)
741 NvmeIdentify *c = (NvmeIdentify *)cmd;
743 switch (le32_to_cpu(c->cns)) {
744 case NVME_ID_CNS_NS:
745 return nvme_identify_ns(n, c);
746 case NVME_ID_CNS_CTRL:
747 return nvme_identify_ctrl(n, c);
748 case NVME_ID_CNS_NS_ACTIVE_LIST:
749 return nvme_identify_nslist(n, c);
750 default:
751 trace_pci_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
752 return NVME_INVALID_FIELD | NVME_DNR;
756 static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
758 trace_pci_nvme_setfeat_timestamp(ts);
760 n->host_timestamp = le64_to_cpu(ts);
761 n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
764 static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
766 uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
767 uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
769 union nvme_timestamp {
770 struct {
771 uint64_t timestamp:48;
772 uint64_t sync:1;
773 uint64_t origin:3;
774 uint64_t rsvd1:12;
776 uint64_t all;
779 union nvme_timestamp ts;
780 ts.all = 0;
783 * If the sum of the Timestamp value set by the host and the elapsed
784 * time exceeds 2^48, the value returned should be reduced modulo 2^48.
786 ts.timestamp = (n->host_timestamp + elapsed_time) & 0xffffffffffff;
788 /* If the host timestamp is non-zero, set the timestamp origin */
789 ts.origin = n->host_timestamp ? 0x01 : 0x00;
791 trace_pci_nvme_getfeat_timestamp(ts.all);
793 return cpu_to_le64(ts.all);
796 static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeCmd *cmd)
798 uint64_t prp1 = le64_to_cpu(cmd->prp1);
799 uint64_t prp2 = le64_to_cpu(cmd->prp2);
801 uint64_t timestamp = nvme_get_timestamp(n);
803 return nvme_dma_read_prp(n, (uint8_t *)&timestamp,
804 sizeof(timestamp), prp1, prp2);
807 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
809 uint32_t dw10 = le32_to_cpu(cmd->cdw10);
810 uint32_t result;
812 switch (dw10) {
813 case NVME_VOLATILE_WRITE_CACHE:
814 result = blk_enable_write_cache(n->conf.blk);
815 trace_pci_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
816 break;
817 case NVME_NUMBER_OF_QUEUES:
818 result = cpu_to_le32((n->params.max_ioqpairs - 1) |
819 ((n->params.max_ioqpairs - 1) << 16));
820 trace_pci_nvme_getfeat_numq(result);
821 break;
822 case NVME_TIMESTAMP:
823 return nvme_get_feature_timestamp(n, cmd);
824 default:
825 trace_pci_nvme_err_invalid_getfeat(dw10);
826 return NVME_INVALID_FIELD | NVME_DNR;
829 req->cqe.result = result;
830 return NVME_SUCCESS;
833 static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeCmd *cmd)
835 uint16_t ret;
836 uint64_t timestamp;
837 uint64_t prp1 = le64_to_cpu(cmd->prp1);
838 uint64_t prp2 = le64_to_cpu(cmd->prp2);
840 ret = nvme_dma_write_prp(n, (uint8_t *)&timestamp,
841 sizeof(timestamp), prp1, prp2);
842 if (ret != NVME_SUCCESS) {
843 return ret;
846 nvme_set_timestamp(n, timestamp);
848 return NVME_SUCCESS;
851 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
853 uint32_t dw10 = le32_to_cpu(cmd->cdw10);
854 uint32_t dw11 = le32_to_cpu(cmd->cdw11);
856 switch (dw10) {
857 case NVME_VOLATILE_WRITE_CACHE:
858 blk_set_enable_write_cache(n->conf.blk, dw11 & 1);
859 break;
860 case NVME_NUMBER_OF_QUEUES:
861 trace_pci_nvme_setfeat_numq((dw11 & 0xFFFF) + 1,
862 ((dw11 >> 16) & 0xFFFF) + 1,
863 n->params.max_ioqpairs,
864 n->params.max_ioqpairs);
865 req->cqe.result = cpu_to_le32((n->params.max_ioqpairs - 1) |
866 ((n->params.max_ioqpairs - 1) << 16));
867 break;
868 case NVME_TIMESTAMP:
869 return nvme_set_feature_timestamp(n, cmd);
870 default:
871 trace_pci_nvme_err_invalid_setfeat(dw10);
872 return NVME_INVALID_FIELD | NVME_DNR;
874 return NVME_SUCCESS;
877 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
879 switch (cmd->opcode) {
880 case NVME_ADM_CMD_DELETE_SQ:
881 return nvme_del_sq(n, cmd);
882 case NVME_ADM_CMD_CREATE_SQ:
883 return nvme_create_sq(n, cmd);
884 case NVME_ADM_CMD_DELETE_CQ:
885 return nvme_del_cq(n, cmd);
886 case NVME_ADM_CMD_CREATE_CQ:
887 return nvme_create_cq(n, cmd);
888 case NVME_ADM_CMD_IDENTIFY:
889 return nvme_identify(n, cmd);
890 case NVME_ADM_CMD_SET_FEATURES:
891 return nvme_set_feature(n, cmd, req);
892 case NVME_ADM_CMD_GET_FEATURES:
893 return nvme_get_feature(n, cmd, req);
894 default:
895 trace_pci_nvme_err_invalid_admin_opc(cmd->opcode);
896 return NVME_INVALID_OPCODE | NVME_DNR;
900 static void nvme_process_sq(void *opaque)
902 NvmeSQueue *sq = opaque;
903 NvmeCtrl *n = sq->ctrl;
904 NvmeCQueue *cq = n->cq[sq->cqid];
906 uint16_t status;
907 hwaddr addr;
908 NvmeCmd cmd;
909 NvmeRequest *req;
911 while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
912 addr = sq->dma_addr + sq->head * n->sqe_size;
913 nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd));
914 nvme_inc_sq_head(sq);
916 req = QTAILQ_FIRST(&sq->req_list);
917 QTAILQ_REMOVE(&sq->req_list, req, entry);
918 QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
919 memset(&req->cqe, 0, sizeof(req->cqe));
920 req->cqe.cid = cmd.cid;
922 status = sq->sqid ? nvme_io_cmd(n, &cmd, req) :
923 nvme_admin_cmd(n, &cmd, req);
924 if (status != NVME_NO_COMPLETE) {
925 req->status = status;
926 nvme_enqueue_req_completion(cq, req);
931 static void nvme_clear_ctrl(NvmeCtrl *n)
933 int i;
935 blk_drain(n->conf.blk);
937 for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
938 if (n->sq[i] != NULL) {
939 nvme_free_sq(n->sq[i], n);
942 for (i = 0; i < n->params.max_ioqpairs + 1; i++) {
943 if (n->cq[i] != NULL) {
944 nvme_free_cq(n->cq[i], n);
948 blk_flush(n->conf.blk);
949 n->bar.cc = 0;
952 static int nvme_start_ctrl(NvmeCtrl *n)
954 uint32_t page_bits = NVME_CC_MPS(n->bar.cc) + 12;
955 uint32_t page_size = 1 << page_bits;
957 if (unlikely(n->cq[0])) {
958 trace_pci_nvme_err_startfail_cq();
959 return -1;
961 if (unlikely(n->sq[0])) {
962 trace_pci_nvme_err_startfail_sq();
963 return -1;
965 if (unlikely(!n->bar.asq)) {
966 trace_pci_nvme_err_startfail_nbarasq();
967 return -1;
969 if (unlikely(!n->bar.acq)) {
970 trace_pci_nvme_err_startfail_nbaracq();
971 return -1;
973 if (unlikely(n->bar.asq & (page_size - 1))) {
974 trace_pci_nvme_err_startfail_asq_misaligned(n->bar.asq);
975 return -1;
977 if (unlikely(n->bar.acq & (page_size - 1))) {
978 trace_pci_nvme_err_startfail_acq_misaligned(n->bar.acq);
979 return -1;
981 if (unlikely(NVME_CC_MPS(n->bar.cc) <
982 NVME_CAP_MPSMIN(n->bar.cap))) {
983 trace_pci_nvme_err_startfail_page_too_small(
984 NVME_CC_MPS(n->bar.cc),
985 NVME_CAP_MPSMIN(n->bar.cap));
986 return -1;
988 if (unlikely(NVME_CC_MPS(n->bar.cc) >
989 NVME_CAP_MPSMAX(n->bar.cap))) {
990 trace_pci_nvme_err_startfail_page_too_large(
991 NVME_CC_MPS(n->bar.cc),
992 NVME_CAP_MPSMAX(n->bar.cap));
993 return -1;
995 if (unlikely(NVME_CC_IOCQES(n->bar.cc) <
996 NVME_CTRL_CQES_MIN(n->id_ctrl.cqes))) {
997 trace_pci_nvme_err_startfail_cqent_too_small(
998 NVME_CC_IOCQES(n->bar.cc),
999 NVME_CTRL_CQES_MIN(n->bar.cap));
1000 return -1;
1002 if (unlikely(NVME_CC_IOCQES(n->bar.cc) >
1003 NVME_CTRL_CQES_MAX(n->id_ctrl.cqes))) {
1004 trace_pci_nvme_err_startfail_cqent_too_large(
1005 NVME_CC_IOCQES(n->bar.cc),
1006 NVME_CTRL_CQES_MAX(n->bar.cap));
1007 return -1;
1009 if (unlikely(NVME_CC_IOSQES(n->bar.cc) <
1010 NVME_CTRL_SQES_MIN(n->id_ctrl.sqes))) {
1011 trace_pci_nvme_err_startfail_sqent_too_small(
1012 NVME_CC_IOSQES(n->bar.cc),
1013 NVME_CTRL_SQES_MIN(n->bar.cap));
1014 return -1;
1016 if (unlikely(NVME_CC_IOSQES(n->bar.cc) >
1017 NVME_CTRL_SQES_MAX(n->id_ctrl.sqes))) {
1018 trace_pci_nvme_err_startfail_sqent_too_large(
1019 NVME_CC_IOSQES(n->bar.cc),
1020 NVME_CTRL_SQES_MAX(n->bar.cap));
1021 return -1;
1023 if (unlikely(!NVME_AQA_ASQS(n->bar.aqa))) {
1024 trace_pci_nvme_err_startfail_asqent_sz_zero();
1025 return -1;
1027 if (unlikely(!NVME_AQA_ACQS(n->bar.aqa))) {
1028 trace_pci_nvme_err_startfail_acqent_sz_zero();
1029 return -1;
1032 n->page_bits = page_bits;
1033 n->page_size = page_size;
1034 n->max_prp_ents = n->page_size / sizeof(uint64_t);
1035 n->cqe_size = 1 << NVME_CC_IOCQES(n->bar.cc);
1036 n->sqe_size = 1 << NVME_CC_IOSQES(n->bar.cc);
1037 nvme_init_cq(&n->admin_cq, n, n->bar.acq, 0, 0,
1038 NVME_AQA_ACQS(n->bar.aqa) + 1, 1);
1039 nvme_init_sq(&n->admin_sq, n, n->bar.asq, 0, 0,
1040 NVME_AQA_ASQS(n->bar.aqa) + 1);
1042 nvme_set_timestamp(n, 0ULL);
1044 return 0;
1047 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
1048 unsigned size)
1050 if (unlikely(offset & (sizeof(uint32_t) - 1))) {
1051 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_misaligned32,
1052 "MMIO write not 32-bit aligned,"
1053 " offset=0x%"PRIx64"", offset);
1054 /* should be ignored, fall through for now */
1057 if (unlikely(size < sizeof(uint32_t))) {
1058 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_toosmall,
1059 "MMIO write smaller than 32-bits,"
1060 " offset=0x%"PRIx64", size=%u",
1061 offset, size);
1062 /* should be ignored, fall through for now */
1065 switch (offset) {
1066 case 0xc: /* INTMS */
1067 if (unlikely(msix_enabled(&(n->parent_obj)))) {
1068 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
1069 "undefined access to interrupt mask set"
1070 " when MSI-X is enabled");
1071 /* should be ignored, fall through for now */
1073 n->bar.intms |= data & 0xffffffff;
1074 n->bar.intmc = n->bar.intms;
1075 trace_pci_nvme_mmio_intm_set(data & 0xffffffff, n->bar.intmc);
1076 nvme_irq_check(n);
1077 break;
1078 case 0x10: /* INTMC */
1079 if (unlikely(msix_enabled(&(n->parent_obj)))) {
1080 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_intmask_with_msix,
1081 "undefined access to interrupt mask clr"
1082 " when MSI-X is enabled");
1083 /* should be ignored, fall through for now */
1085 n->bar.intms &= ~(data & 0xffffffff);
1086 n->bar.intmc = n->bar.intms;
1087 trace_pci_nvme_mmio_intm_clr(data & 0xffffffff, n->bar.intmc);
1088 nvme_irq_check(n);
1089 break;
1090 case 0x14: /* CC */
1091 trace_pci_nvme_mmio_cfg(data & 0xffffffff);
1092 /* Windows first sends data, then sends enable bit */
1093 if (!NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc) &&
1094 !NVME_CC_SHN(data) && !NVME_CC_SHN(n->bar.cc))
1096 n->bar.cc = data;
1099 if (NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc)) {
1100 n->bar.cc = data;
1101 if (unlikely(nvme_start_ctrl(n))) {
1102 trace_pci_nvme_err_startfail();
1103 n->bar.csts = NVME_CSTS_FAILED;
1104 } else {
1105 trace_pci_nvme_mmio_start_success();
1106 n->bar.csts = NVME_CSTS_READY;
1108 } else if (!NVME_CC_EN(data) && NVME_CC_EN(n->bar.cc)) {
1109 trace_pci_nvme_mmio_stopped();
1110 nvme_clear_ctrl(n);
1111 n->bar.csts &= ~NVME_CSTS_READY;
1113 if (NVME_CC_SHN(data) && !(NVME_CC_SHN(n->bar.cc))) {
1114 trace_pci_nvme_mmio_shutdown_set();
1115 nvme_clear_ctrl(n);
1116 n->bar.cc = data;
1117 n->bar.csts |= NVME_CSTS_SHST_COMPLETE;
1118 } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(n->bar.cc)) {
1119 trace_pci_nvme_mmio_shutdown_cleared();
1120 n->bar.csts &= ~NVME_CSTS_SHST_COMPLETE;
1121 n->bar.cc = data;
1123 break;
1124 case 0x1C: /* CSTS */
1125 if (data & (1 << 4)) {
1126 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ssreset_w1c_unsupported,
1127 "attempted to W1C CSTS.NSSRO"
1128 " but CAP.NSSRS is zero (not supported)");
1129 } else if (data != 0) {
1130 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_ro_csts,
1131 "attempted to set a read only bit"
1132 " of controller status");
1134 break;
1135 case 0x20: /* NSSR */
1136 if (data == 0x4E564D65) {
1137 trace_pci_nvme_ub_mmiowr_ssreset_unsupported();
1138 } else {
1139 /* The spec says that writes of other values have no effect */
1140 return;
1142 break;
1143 case 0x24: /* AQA */
1144 n->bar.aqa = data & 0xffffffff;
1145 trace_pci_nvme_mmio_aqattr(data & 0xffffffff);
1146 break;
1147 case 0x28: /* ASQ */
1148 n->bar.asq = data;
1149 trace_pci_nvme_mmio_asqaddr(data);
1150 break;
1151 case 0x2c: /* ASQ hi */
1152 n->bar.asq |= data << 32;
1153 trace_pci_nvme_mmio_asqaddr_hi(data, n->bar.asq);
1154 break;
1155 case 0x30: /* ACQ */
1156 trace_pci_nvme_mmio_acqaddr(data);
1157 n->bar.acq = data;
1158 break;
1159 case 0x34: /* ACQ hi */
1160 n->bar.acq |= data << 32;
1161 trace_pci_nvme_mmio_acqaddr_hi(data, n->bar.acq);
1162 break;
1163 case 0x38: /* CMBLOC */
1164 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbloc_reserved,
1165 "invalid write to reserved CMBLOC"
1166 " when CMBSZ is zero, ignored");
1167 return;
1168 case 0x3C: /* CMBSZ */
1169 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_cmbsz_readonly,
1170 "invalid write to read only CMBSZ, ignored");
1171 return;
1172 case 0xE00: /* PMRCAP */
1173 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrcap_readonly,
1174 "invalid write to PMRCAP register, ignored");
1175 return;
1176 case 0xE04: /* TODO PMRCTL */
1177 break;
1178 case 0xE08: /* PMRSTS */
1179 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrsts_readonly,
1180 "invalid write to PMRSTS register, ignored");
1181 return;
1182 case 0xE0C: /* PMREBS */
1183 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrebs_readonly,
1184 "invalid write to PMREBS register, ignored");
1185 return;
1186 case 0xE10: /* PMRSWTP */
1187 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_pmrswtp_readonly,
1188 "invalid write to PMRSWTP register, ignored");
1189 return;
1190 case 0xE14: /* TODO PMRMSC */
1191 break;
1192 default:
1193 NVME_GUEST_ERR(pci_nvme_ub_mmiowr_invalid,
1194 "invalid MMIO write,"
1195 " offset=0x%"PRIx64", data=%"PRIx64"",
1196 offset, data);
1197 break;
1201 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
1203 NvmeCtrl *n = (NvmeCtrl *)opaque;
1204 uint8_t *ptr = (uint8_t *)&n->bar;
1205 uint64_t val = 0;
1207 if (unlikely(addr & (sizeof(uint32_t) - 1))) {
1208 NVME_GUEST_ERR(pci_nvme_ub_mmiord_misaligned32,
1209 "MMIO read not 32-bit aligned,"
1210 " offset=0x%"PRIx64"", addr);
1211 /* should RAZ, fall through for now */
1212 } else if (unlikely(size < sizeof(uint32_t))) {
1213 NVME_GUEST_ERR(pci_nvme_ub_mmiord_toosmall,
1214 "MMIO read smaller than 32-bits,"
1215 " offset=0x%"PRIx64"", addr);
1216 /* should RAZ, fall through for now */
1219 if (addr < sizeof(n->bar)) {
1221 * When PMRWBM bit 1 is set then read from
1222 * from PMRSTS should ensure prior writes
1223 * made it to persistent media
1225 if (addr == 0xE08 &&
1226 (NVME_PMRCAP_PMRWBM(n->bar.pmrcap) & 0x02)) {
1227 memory_region_msync(&n->pmrdev->mr, 0, n->pmrdev->size);
1229 memcpy(&val, ptr + addr, size);
1230 } else {
1231 NVME_GUEST_ERR(pci_nvme_ub_mmiord_invalid_ofs,
1232 "MMIO read beyond last register,"
1233 " offset=0x%"PRIx64", returning 0", addr);
1236 return val;
1239 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
1241 uint32_t qid;
1243 if (unlikely(addr & ((1 << 2) - 1))) {
1244 NVME_GUEST_ERR(pci_nvme_ub_db_wr_misaligned,
1245 "doorbell write not 32-bit aligned,"
1246 " offset=0x%"PRIx64", ignoring", addr);
1247 return;
1250 if (((addr - 0x1000) >> 2) & 1) {
1251 /* Completion queue doorbell write */
1253 uint16_t new_head = val & 0xffff;
1254 int start_sqs;
1255 NvmeCQueue *cq;
1257 qid = (addr - (0x1000 + (1 << 2))) >> 3;
1258 if (unlikely(nvme_check_cqid(n, qid))) {
1259 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cq,
1260 "completion queue doorbell write"
1261 " for nonexistent queue,"
1262 " sqid=%"PRIu32", ignoring", qid);
1263 return;
1266 cq = n->cq[qid];
1267 if (unlikely(new_head >= cq->size)) {
1268 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_cqhead,
1269 "completion queue doorbell write value"
1270 " beyond queue size, sqid=%"PRIu32","
1271 " new_head=%"PRIu16", ignoring",
1272 qid, new_head);
1273 return;
1276 start_sqs = nvme_cq_full(cq) ? 1 : 0;
1277 cq->head = new_head;
1278 if (start_sqs) {
1279 NvmeSQueue *sq;
1280 QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
1281 timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
1283 timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
1286 if (cq->tail == cq->head) {
1287 nvme_irq_deassert(n, cq);
1289 } else {
1290 /* Submission queue doorbell write */
1292 uint16_t new_tail = val & 0xffff;
1293 NvmeSQueue *sq;
1295 qid = (addr - 0x1000) >> 3;
1296 if (unlikely(nvme_check_sqid(n, qid))) {
1297 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sq,
1298 "submission queue doorbell write"
1299 " for nonexistent queue,"
1300 " sqid=%"PRIu32", ignoring", qid);
1301 return;
1304 sq = n->sq[qid];
1305 if (unlikely(new_tail >= sq->size)) {
1306 NVME_GUEST_ERR(pci_nvme_ub_db_wr_invalid_sqtail,
1307 "submission queue doorbell write value"
1308 " beyond queue size, sqid=%"PRIu32","
1309 " new_tail=%"PRIu16", ignoring",
1310 qid, new_tail);
1311 return;
1314 sq->tail = new_tail;
1315 timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
1319 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
1320 unsigned size)
1322 NvmeCtrl *n = (NvmeCtrl *)opaque;
1323 if (addr < sizeof(n->bar)) {
1324 nvme_write_bar(n, addr, data, size);
1325 } else if (addr >= 0x1000) {
1326 nvme_process_db(n, addr, data);
1330 static const MemoryRegionOps nvme_mmio_ops = {
1331 .read = nvme_mmio_read,
1332 .write = nvme_mmio_write,
1333 .endianness = DEVICE_LITTLE_ENDIAN,
1334 .impl = {
1335 .min_access_size = 2,
1336 .max_access_size = 8,
1340 static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
1341 unsigned size)
1343 NvmeCtrl *n = (NvmeCtrl *)opaque;
1344 stn_le_p(&n->cmbuf[addr], size, data);
1347 static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
1349 NvmeCtrl *n = (NvmeCtrl *)opaque;
1350 return ldn_le_p(&n->cmbuf[addr], size);
1353 static const MemoryRegionOps nvme_cmb_ops = {
1354 .read = nvme_cmb_read,
1355 .write = nvme_cmb_write,
1356 .endianness = DEVICE_LITTLE_ENDIAN,
1357 .impl = {
1358 .min_access_size = 1,
1359 .max_access_size = 8,
1363 static void nvme_check_constraints(NvmeCtrl *n, Error **errp)
1365 NvmeParams *params = &n->params;
1367 if (params->num_queues) {
1368 warn_report("num_queues is deprecated; please use max_ioqpairs "
1369 "instead");
1371 params->max_ioqpairs = params->num_queues - 1;
1374 if (params->max_ioqpairs < 1 ||
1375 params->max_ioqpairs > NVME_MAX_IOQPAIRS) {
1376 error_setg(errp, "max_ioqpairs must be between 1 and %d",
1377 NVME_MAX_IOQPAIRS);
1378 return;
1381 if (params->msix_qsize < 1 ||
1382 params->msix_qsize > PCI_MSIX_FLAGS_QSIZE + 1) {
1383 error_setg(errp, "msix_qsize must be between 1 and %d",
1384 PCI_MSIX_FLAGS_QSIZE + 1);
1385 return;
1388 if (!n->conf.blk) {
1389 error_setg(errp, "drive property not set");
1390 return;
1393 if (!params->serial) {
1394 error_setg(errp, "serial property not set");
1395 return;
1398 if (!n->params.cmb_size_mb && n->pmrdev) {
1399 if (host_memory_backend_is_mapped(n->pmrdev)) {
1400 char *path = object_get_canonical_path_component(OBJECT(n->pmrdev));
1401 error_setg(errp, "can't use already busy memdev: %s", path);
1402 g_free(path);
1403 return;
1406 if (!is_power_of_2(n->pmrdev->size)) {
1407 error_setg(errp, "pmr backend size needs to be power of 2 in size");
1408 return;
1411 host_memory_backend_set_mapped(n->pmrdev, true);
1415 static void nvme_init_state(NvmeCtrl *n)
1417 n->num_namespaces = 1;
1418 /* add one to max_ioqpairs to account for the admin queue pair */
1419 n->reg_size = pow2ceil(NVME_REG_SIZE +
1420 2 * (n->params.max_ioqpairs + 1) * NVME_DB_SIZE);
1421 n->namespaces = g_new0(NvmeNamespace, n->num_namespaces);
1422 n->sq = g_new0(NvmeSQueue *, n->params.max_ioqpairs + 1);
1423 n->cq = g_new0(NvmeCQueue *, n->params.max_ioqpairs + 1);
1426 static void nvme_init_blk(NvmeCtrl *n, Error **errp)
1428 if (!blkconf_blocksizes(&n->conf, errp)) {
1429 return;
1431 blkconf_apply_backend_options(&n->conf, blk_is_read_only(n->conf.blk),
1432 false, errp);
1435 static void nvme_init_namespace(NvmeCtrl *n, NvmeNamespace *ns, Error **errp)
1437 int64_t bs_size;
1438 NvmeIdNs *id_ns = &ns->id_ns;
1440 bs_size = blk_getlength(n->conf.blk);
1441 if (bs_size < 0) {
1442 error_setg_errno(errp, -bs_size, "could not get backing file size");
1443 return;
1446 n->ns_size = bs_size;
1448 id_ns->lbaf[0].ds = BDRV_SECTOR_BITS;
1449 id_ns->nsze = cpu_to_le64(nvme_ns_nlbas(n, ns));
1451 /* no thin provisioning */
1452 id_ns->ncap = id_ns->nsze;
1453 id_ns->nuse = id_ns->ncap;
1456 static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
1458 NVME_CMBLOC_SET_BIR(n->bar.cmbloc, NVME_CMB_BIR);
1459 NVME_CMBLOC_SET_OFST(n->bar.cmbloc, 0);
1461 NVME_CMBSZ_SET_SQS(n->bar.cmbsz, 1);
1462 NVME_CMBSZ_SET_CQS(n->bar.cmbsz, 0);
1463 NVME_CMBSZ_SET_LISTS(n->bar.cmbsz, 0);
1464 NVME_CMBSZ_SET_RDS(n->bar.cmbsz, 1);
1465 NVME_CMBSZ_SET_WDS(n->bar.cmbsz, 1);
1466 NVME_CMBSZ_SET_SZU(n->bar.cmbsz, 2); /* MBs */
1467 NVME_CMBSZ_SET_SZ(n->bar.cmbsz, n->params.cmb_size_mb);
1469 n->cmbuf = g_malloc0(NVME_CMBSZ_GETSIZE(n->bar.cmbsz));
1470 memory_region_init_io(&n->ctrl_mem, OBJECT(n), &nvme_cmb_ops, n,
1471 "nvme-cmb", NVME_CMBSZ_GETSIZE(n->bar.cmbsz));
1472 pci_register_bar(pci_dev, NVME_CMBLOC_BIR(n->bar.cmbloc),
1473 PCI_BASE_ADDRESS_SPACE_MEMORY |
1474 PCI_BASE_ADDRESS_MEM_TYPE_64 |
1475 PCI_BASE_ADDRESS_MEM_PREFETCH, &n->ctrl_mem);
1478 static void nvme_init_pmr(NvmeCtrl *n, PCIDevice *pci_dev)
1480 /* Controller Capabilities register */
1481 NVME_CAP_SET_PMRS(n->bar.cap, 1);
1483 /* PMR Capabities register */
1484 n->bar.pmrcap = 0;
1485 NVME_PMRCAP_SET_RDS(n->bar.pmrcap, 0);
1486 NVME_PMRCAP_SET_WDS(n->bar.pmrcap, 0);
1487 NVME_PMRCAP_SET_BIR(n->bar.pmrcap, NVME_PMR_BIR);
1488 NVME_PMRCAP_SET_PMRTU(n->bar.pmrcap, 0);
1489 /* Turn on bit 1 support */
1490 NVME_PMRCAP_SET_PMRWBM(n->bar.pmrcap, 0x02);
1491 NVME_PMRCAP_SET_PMRTO(n->bar.pmrcap, 0);
1492 NVME_PMRCAP_SET_CMSS(n->bar.pmrcap, 0);
1494 /* PMR Control register */
1495 n->bar.pmrctl = 0;
1496 NVME_PMRCTL_SET_EN(n->bar.pmrctl, 0);
1498 /* PMR Status register */
1499 n->bar.pmrsts = 0;
1500 NVME_PMRSTS_SET_ERR(n->bar.pmrsts, 0);
1501 NVME_PMRSTS_SET_NRDY(n->bar.pmrsts, 0);
1502 NVME_PMRSTS_SET_HSTS(n->bar.pmrsts, 0);
1503 NVME_PMRSTS_SET_CBAI(n->bar.pmrsts, 0);
1505 /* PMR Elasticity Buffer Size register */
1506 n->bar.pmrebs = 0;
1507 NVME_PMREBS_SET_PMRSZU(n->bar.pmrebs, 0);
1508 NVME_PMREBS_SET_RBB(n->bar.pmrebs, 0);
1509 NVME_PMREBS_SET_PMRWBZ(n->bar.pmrebs, 0);
1511 /* PMR Sustained Write Throughput register */
1512 n->bar.pmrswtp = 0;
1513 NVME_PMRSWTP_SET_PMRSWTU(n->bar.pmrswtp, 0);
1514 NVME_PMRSWTP_SET_PMRSWTV(n->bar.pmrswtp, 0);
1516 /* PMR Memory Space Control register */
1517 n->bar.pmrmsc = 0;
1518 NVME_PMRMSC_SET_CMSE(n->bar.pmrmsc, 0);
1519 NVME_PMRMSC_SET_CBA(n->bar.pmrmsc, 0);
1521 pci_register_bar(pci_dev, NVME_PMRCAP_BIR(n->bar.pmrcap),
1522 PCI_BASE_ADDRESS_SPACE_MEMORY |
1523 PCI_BASE_ADDRESS_MEM_TYPE_64 |
1524 PCI_BASE_ADDRESS_MEM_PREFETCH, &n->pmrdev->mr);
1527 static void nvme_init_pci(NvmeCtrl *n, PCIDevice *pci_dev, Error **errp)
1529 uint8_t *pci_conf = pci_dev->config;
1531 pci_conf[PCI_INTERRUPT_PIN] = 1;
1532 pci_config_set_prog_interface(pci_conf, 0x2);
1533 pci_config_set_class(pci_conf, PCI_CLASS_STORAGE_EXPRESS);
1534 pcie_endpoint_cap_init(pci_dev, 0x80);
1536 memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n, "nvme",
1537 n->reg_size);
1538 pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY |
1539 PCI_BASE_ADDRESS_MEM_TYPE_64, &n->iomem);
1540 if (msix_init_exclusive_bar(pci_dev, n->params.msix_qsize, 4, errp)) {
1541 return;
1544 if (n->params.cmb_size_mb) {
1545 nvme_init_cmb(n, pci_dev);
1546 } else if (n->pmrdev) {
1547 nvme_init_pmr(n, pci_dev);
1551 static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev)
1553 NvmeIdCtrl *id = &n->id_ctrl;
1554 uint8_t *pci_conf = pci_dev->config;
1556 id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
1557 id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
1558 strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
1559 strpadcpy((char *)id->fr, sizeof(id->fr), "1.0", ' ');
1560 strpadcpy((char *)id->sn, sizeof(id->sn), n->params.serial, ' ');
1561 id->rab = 6;
1562 id->ieee[0] = 0x00;
1563 id->ieee[1] = 0x02;
1564 id->ieee[2] = 0xb3;
1565 id->oacs = cpu_to_le16(0);
1566 id->frmw = 7 << 1;
1567 id->lpa = 1 << 0;
1568 id->sqes = (0x6 << 4) | 0x6;
1569 id->cqes = (0x4 << 4) | 0x4;
1570 id->nn = cpu_to_le32(n->num_namespaces);
1571 id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROS | NVME_ONCS_TIMESTAMP);
1572 id->psd[0].mp = cpu_to_le16(0x9c4);
1573 id->psd[0].enlat = cpu_to_le32(0x10);
1574 id->psd[0].exlat = cpu_to_le32(0x4);
1575 if (blk_enable_write_cache(n->conf.blk)) {
1576 id->vwc = 1;
1579 n->bar.cap = 0;
1580 NVME_CAP_SET_MQES(n->bar.cap, 0x7ff);
1581 NVME_CAP_SET_CQR(n->bar.cap, 1);
1582 NVME_CAP_SET_TO(n->bar.cap, 0xf);
1583 NVME_CAP_SET_CSS(n->bar.cap, 1);
1584 NVME_CAP_SET_MPSMAX(n->bar.cap, 4);
1586 n->bar.vs = 0x00010200;
1587 n->bar.intmc = n->bar.intms = 0;
1590 static void nvme_realize(PCIDevice *pci_dev, Error **errp)
1592 NvmeCtrl *n = NVME(pci_dev);
1593 Error *local_err = NULL;
1595 int i;
1597 nvme_check_constraints(n, &local_err);
1598 if (local_err) {
1599 error_propagate(errp, local_err);
1600 return;
1603 nvme_init_state(n);
1604 nvme_init_blk(n, &local_err);
1605 if (local_err) {
1606 error_propagate(errp, local_err);
1607 return;
1610 nvme_init_pci(n, pci_dev, &local_err);
1611 if (local_err) {
1612 error_propagate(errp, local_err);
1613 return;
1616 nvme_init_ctrl(n, pci_dev);
1618 for (i = 0; i < n->num_namespaces; i++) {
1619 nvme_init_namespace(n, &n->namespaces[i], &local_err);
1620 if (local_err) {
1621 error_propagate(errp, local_err);
1622 return;
1627 static void nvme_exit(PCIDevice *pci_dev)
1629 NvmeCtrl *n = NVME(pci_dev);
1631 nvme_clear_ctrl(n);
1632 g_free(n->namespaces);
1633 g_free(n->cq);
1634 g_free(n->sq);
1636 if (n->params.cmb_size_mb) {
1637 g_free(n->cmbuf);
1640 if (n->pmrdev) {
1641 host_memory_backend_set_mapped(n->pmrdev, false);
1643 msix_uninit_exclusive_bar(pci_dev);
1646 static Property nvme_props[] = {
1647 DEFINE_BLOCK_PROPERTIES(NvmeCtrl, conf),
1648 DEFINE_PROP_LINK("pmrdev", NvmeCtrl, pmrdev, TYPE_MEMORY_BACKEND,
1649 HostMemoryBackend *),
1650 DEFINE_PROP_STRING("serial", NvmeCtrl, params.serial),
1651 DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, params.cmb_size_mb, 0),
1652 DEFINE_PROP_UINT32("num_queues", NvmeCtrl, params.num_queues, 0),
1653 DEFINE_PROP_UINT32("max_ioqpairs", NvmeCtrl, params.max_ioqpairs, 64),
1654 DEFINE_PROP_UINT16("msix_qsize", NvmeCtrl, params.msix_qsize, 65),
1655 DEFINE_PROP_END_OF_LIST(),
1658 static const VMStateDescription nvme_vmstate = {
1659 .name = "nvme",
1660 .unmigratable = 1,
1663 static void nvme_class_init(ObjectClass *oc, void *data)
1665 DeviceClass *dc = DEVICE_CLASS(oc);
1666 PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
1668 pc->realize = nvme_realize;
1669 pc->exit = nvme_exit;
1670 pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
1671 pc->vendor_id = PCI_VENDOR_ID_INTEL;
1672 pc->device_id = 0x5845;
1673 pc->revision = 2;
1675 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1676 dc->desc = "Non-Volatile Memory Express";
1677 device_class_set_props(dc, nvme_props);
1678 dc->vmsd = &nvme_vmstate;
1681 static void nvme_instance_init(Object *obj)
1683 NvmeCtrl *s = NVME(obj);
1685 device_add_bootindex_property(obj, &s->conf.bootindex,
1686 "bootindex", "/namespace@1,0",
1687 DEVICE(obj));
1690 static const TypeInfo nvme_info = {
1691 .name = TYPE_NVME,
1692 .parent = TYPE_PCI_DEVICE,
1693 .instance_size = sizeof(NvmeCtrl),
1694 .class_init = nvme_class_init,
1695 .instance_init = nvme_instance_init,
1696 .interfaces = (InterfaceInfo[]) {
1697 { INTERFACE_PCIE_DEVICE },
1702 static void nvme_register_types(void)
1704 type_register_static(&nvme_info);
1707 type_init(nvme_register_types)