1/*
2 * QEMU NVM Express Controller
3 *
4 * Copyright (c) 2012, Intel Corporation
5 *
6 * Written by Keith Busch <keith.busch@intel.com>
7 *
8 * This code is licensed under the GNU GPL v2 or later.
9 */
10
11/**
12 * Reference Specs: http://www.nvmexpress.org, 1.2, 1.1, 1.0e
13 *
14 * http://www.nvmexpress.org/resources/
15 */
16
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 * num_queues=<N[optional]>
23 *
24 * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
25 * offset 0 in BAR2 and supports only WDS, RDS and SQS for now.
26 */
27
28#include "qemu/osdep.h"
29#include "qemu/units.h"
30#include "hw/block/block.h"
31#include "hw/pci/msix.h"
32#include "hw/pci/pci.h"
33#include "hw/qdev-properties.h"
34#include "migration/vmstate.h"
35#include "sysemu/sysemu.h"
36#include "qapi/error.h"
37#include "qapi/visitor.h"
38#include "sysemu/block-backend.h"
39
40#include "qemu/log.h"
41#include "qemu/module.h"
42#include "qemu/cutils.h"
43#include "trace.h"
44#include "nvme.h"
45
46#define NVME_GUEST_ERR(trace, fmt, ...) \
47 do { \
48 (trace_##trace)(__VA_ARGS__); \
49 qemu_log_mask(LOG_GUEST_ERROR, #trace \
50 " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
51 } while (0)
52
53static void nvme_process_sq(void *opaque);
54
55static void nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
56{
57 if (n->cmbsz && addr >= n->ctrl_mem.addr &&
58 addr < (n->ctrl_mem.addr + int128_get64(n->ctrl_mem.size))) {
59 memcpy(buf, (void *)&n->cmbuf[addr - n->ctrl_mem.addr], size);
60 } else {
61 pci_dma_read(&n->parent_obj, addr, buf, size);
62 }
63}
64
65static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
66{
67 return sqid < n->num_queues && n->sq[sqid] != NULL ? 0 : -1;
68}
69
70static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
71{
72 return cqid < n->num_queues && n->cq[cqid] != NULL ? 0 : -1;
73}
74
75static void nvme_inc_cq_tail(NvmeCQueue *cq)
76{
77 cq->tail++;
78 if (cq->tail >= cq->size) {
79 cq->tail = 0;
80 cq->phase = !cq->phase;
81 }
82}
83
84static void nvme_inc_sq_head(NvmeSQueue *sq)
85{
86 sq->head = (sq->head + 1) % sq->size;
87}
88
89static uint8_t nvme_cq_full(NvmeCQueue *cq)
90{
91 return (cq->tail + 1) % cq->size == cq->head;
92}
93
94static uint8_t nvme_sq_empty(NvmeSQueue *sq)
95{
96 return sq->head == sq->tail;
97}
98
99static void nvme_irq_check(NvmeCtrl *n)
100{
101 if (msix_enabled(&(n->parent_obj))) {
102 return;
103 }
104 if (~n->bar.intms & n->irq_status) {
105 pci_irq_assert(&n->parent_obj);
106 } else {
107 pci_irq_deassert(&n->parent_obj);
108 }
109}
110
111static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
112{
113 if (cq->irq_enabled) {
114 if (msix_enabled(&(n->parent_obj))) {
115 trace_nvme_irq_msix(cq->vector);
116 msix_notify(&(n->parent_obj), cq->vector);
117 } else {
118 trace_nvme_irq_pin();
119 assert(cq->cqid < 64);
120 n->irq_status |= 1 << cq->cqid;
121 nvme_irq_check(n);
122 }
123 } else {
124 trace_nvme_irq_masked();
125 }
126}
127
128static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
129{
130 if (cq->irq_enabled) {
131 if (msix_enabled(&(n->parent_obj))) {
132 return;
133 } else {
134 assert(cq->cqid < 64);
135 n->irq_status &= ~(1 << cq->cqid);
136 nvme_irq_check(n);
137 }
138 }
139}
140
141static uint16_t nvme_map_prp(QEMUSGList *qsg, QEMUIOVector *iov, uint64_t prp1,
142 uint64_t prp2, uint32_t len, NvmeCtrl *n)
143{
144 hwaddr trans_len = n->page_size - (prp1 % n->page_size);
145 trans_len = MIN(len, trans_len);
146 int num_prps = (len >> n->page_bits) + 1;
147
148 if (unlikely(!prp1)) {
149 trace_nvme_err_invalid_prp();
150 return NVME_INVALID_FIELD | NVME_DNR;
151 } else if (n->cmbsz && prp1 >= n->ctrl_mem.addr &&
152 prp1 < n->ctrl_mem.addr + int128_get64(n->ctrl_mem.size)) {
153 qsg->nsg = 0;
154 qemu_iovec_init(iov, num_prps);
155 qemu_iovec_add(iov, (void *)&n->cmbuf[prp1 - n->ctrl_mem.addr], trans_len);
156 } else {
157 pci_dma_sglist_init(qsg, &n->parent_obj, num_prps);
158 qemu_sglist_add(qsg, prp1, trans_len);
159 }
160 len -= trans_len;
161 if (len) {
162 if (unlikely(!prp2)) {
163 trace_nvme_err_invalid_prp2_missing();
164 goto unmap;
165 }
166 if (len > n->page_size) {
167 uint64_t prp_list[n->max_prp_ents];
168 uint32_t nents, prp_trans;
169 int i = 0;
170
171 nents = (len + n->page_size - 1) >> n->page_bits;
172 prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
173 nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
174 while (len != 0) {
175 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
176
177 if (i == n->max_prp_ents - 1 && len > n->page_size) {
178 if (unlikely(!prp_ent || prp_ent & (n->page_size - 1))) {
179 trace_nvme_err_invalid_prplist_ent(prp_ent);
180 goto unmap;
181 }
182
183 i = 0;
184 nents = (len + n->page_size - 1) >> n->page_bits;
185 prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
186 nvme_addr_read(n, prp_ent, (void *)prp_list,
187 prp_trans);
188 prp_ent = le64_to_cpu(prp_list[i]);
189 }
190
191 if (unlikely(!prp_ent || prp_ent & (n->page_size - 1))) {
192 trace_nvme_err_invalid_prplist_ent(prp_ent);
193 goto unmap;
194 }
195
196 trans_len = MIN(len, n->page_size);
197 if (qsg->nsg){
198 qemu_sglist_add(qsg, prp_ent, trans_len);
199 } else {
200 qemu_iovec_add(iov, (void *)&n->cmbuf[prp_ent - n->ctrl_mem.addr], trans_len);
201 }
202 len -= trans_len;
203 i++;
204 }
205 } else {
206 if (unlikely(prp2 & (n->page_size - 1))) {
207 trace_nvme_err_invalid_prp2_align(prp2);
208 goto unmap;
209 }
210 if (qsg->nsg) {
211 qemu_sglist_add(qsg, prp2, len);
212 } else {
213 qemu_iovec_add(iov, (void *)&n->cmbuf[prp2 - n->ctrl_mem.addr], trans_len);
214 }
215 }
216 }
217 return NVME_SUCCESS;
218
219 unmap:
220 qemu_sglist_destroy(qsg);
221 return NVME_INVALID_FIELD | NVME_DNR;
222}
223
224static uint16_t nvme_dma_write_prp(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
225 uint64_t prp1, uint64_t prp2)
226{
227 QEMUSGList qsg;
228 QEMUIOVector iov;
229 uint16_t status = NVME_SUCCESS;
230
231 if (nvme_map_prp(&qsg, &iov, prp1, prp2, len, n)) {
232 return NVME_INVALID_FIELD | NVME_DNR;
233 }
234 if (qsg.nsg > 0) {
235 if (dma_buf_write(ptr, len, &qsg)) {
236 status = NVME_INVALID_FIELD | NVME_DNR;
237 }
238 qemu_sglist_destroy(&qsg);
239 } else {
240 if (qemu_iovec_to_buf(&iov, 0, ptr, len) != len) {
241 status = NVME_INVALID_FIELD | NVME_DNR;
242 }
243 qemu_iovec_destroy(&iov);
244 }
245 return status;
246}
247
248static uint16_t nvme_dma_read_prp(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
249 uint64_t prp1, uint64_t prp2)
250{
251 QEMUSGList qsg;
252 QEMUIOVector iov;
253 uint16_t status = NVME_SUCCESS;
254
255 trace_nvme_dma_read(prp1, prp2);
256
257 if (nvme_map_prp(&qsg, &iov, prp1, prp2, len, n)) {
258 return NVME_INVALID_FIELD | NVME_DNR;
259 }
260 if (qsg.nsg > 0) {
261 if (unlikely(dma_buf_read(ptr, len, &qsg))) {
262 trace_nvme_err_invalid_dma();
263 status = NVME_INVALID_FIELD | NVME_DNR;
264 }
265 qemu_sglist_destroy(&qsg);
266 } else {
267 if (unlikely(qemu_iovec_from_buf(&iov, 0, ptr, len) != len)) {
268 trace_nvme_err_invalid_dma();
269 status = NVME_INVALID_FIELD | NVME_DNR;
270 }
271 qemu_iovec_destroy(&iov);
272 }
273 return status;
274}
275
276static void nvme_post_cqes(void *opaque)
277{
278 NvmeCQueue *cq = opaque;
279 NvmeCtrl *n = cq->ctrl;
280 NvmeRequest *req, *next;
281
282 QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
283 NvmeSQueue *sq;
284 hwaddr addr;
285
286 if (nvme_cq_full(cq)) {
287 break;
288 }
289
290 QTAILQ_REMOVE(&cq->req_list, req, entry);
291 sq = req->sq;
292 req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
293 req->cqe.sq_id = cpu_to_le16(sq->sqid);
294 req->cqe.sq_head = cpu_to_le16(sq->head);
295 addr = cq->dma_addr + cq->tail * n->cqe_size;
296 nvme_inc_cq_tail(cq);
297 pci_dma_write(&n->parent_obj, addr, (void *)&req->cqe,
298 sizeof(req->cqe));
299 QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
300 }
301 if (cq->tail != cq->head) {
302 nvme_irq_assert(n, cq);
303 }
304}
305
306static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
307{
308 assert(cq->cqid == req->sq->cqid);
309 QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
310 QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
311 timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
312}
313
314static void nvme_rw_cb(void *opaque, int ret)
315{
316 NvmeRequest *req = opaque;
317 NvmeSQueue *sq = req->sq;
318 NvmeCtrl *n = sq->ctrl;
319 NvmeCQueue *cq = n->cq[sq->cqid];
320
321 if (!ret) {
322 block_acct_done(blk_get_stats(n->conf.blk), &req->acct);
323 req->status = NVME_SUCCESS;
324 } else {
325 block_acct_failed(blk_get_stats(n->conf.blk), &req->acct);
326 req->status = NVME_INTERNAL_DEV_ERROR;
327 }
328 if (req->has_sg) {
329 qemu_sglist_destroy(&req->qsg);
330 }
331 nvme_enqueue_req_completion(cq, req);
332}
333
334static uint16_t nvme_flush(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
335 NvmeRequest *req)
336{
337 req->has_sg = false;
338 block_acct_start(blk_get_stats(n->conf.blk), &req->acct, 0,
339 BLOCK_ACCT_FLUSH);
340 req->aiocb = blk_aio_flush(n->conf.blk, nvme_rw_cb, req);
341
342 return NVME_NO_COMPLETE;
343}
344
345static uint16_t nvme_write_zeros(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
346 NvmeRequest *req)
347{
348 NvmeRwCmd *rw = (NvmeRwCmd *)cmd;
349 const uint8_t lba_index = NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas);
350 const uint8_t data_shift = ns->id_ns.lbaf[lba_index].ds;
351 uint64_t slba = le64_to_cpu(rw->slba);
352 uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
353 uint64_t offset = slba << data_shift;
354 uint32_t count = nlb << data_shift;
355
356 if (unlikely(slba + nlb > ns->id_ns.nsze)) {
357 trace_nvme_err_invalid_lba_range(slba, nlb, ns->id_ns.nsze);
358 return NVME_LBA_RANGE | NVME_DNR;
359 }
360
361 req->has_sg = false;
362 block_acct_start(blk_get_stats(n->conf.blk), &req->acct, 0,
363 BLOCK_ACCT_WRITE);
364 req->aiocb = blk_aio_pwrite_zeroes(n->conf.blk, offset, count,
365 BDRV_REQ_MAY_UNMAP, nvme_rw_cb, req);
366 return NVME_NO_COMPLETE;
367}
368
369static uint16_t nvme_rw(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
370 NvmeRequest *req)
371{
372 NvmeRwCmd *rw = (NvmeRwCmd *)cmd;
373 uint32_t nlb = le32_to_cpu(rw->nlb) + 1;
374 uint64_t slba = le64_to_cpu(rw->slba);
375 uint64_t prp1 = le64_to_cpu(rw->prp1);
376 uint64_t prp2 = le64_to_cpu(rw->prp2);
377
378 uint8_t lba_index = NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas);
379 uint8_t data_shift = ns->id_ns.lbaf[lba_index].ds;
380 uint64_t data_size = (uint64_t)nlb << data_shift;
381 uint64_t data_offset = slba << data_shift;
382 int is_write = rw->opcode == NVME_CMD_WRITE ? 1 : 0;
383 enum BlockAcctType acct = is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ;
384
385 trace_nvme_rw(is_write ? "write" : "read", nlb, data_size, slba);
386
387 if (unlikely((slba + nlb) > ns->id_ns.nsze)) {
388 block_acct_invalid(blk_get_stats(n->conf.blk), acct);
389 trace_nvme_err_invalid_lba_range(slba, nlb, ns->id_ns.nsze);
390 return NVME_LBA_RANGE | NVME_DNR;
391 }
392
393 if (nvme_map_prp(&req->qsg, &req->iov, prp1, prp2, data_size, n)) {
394 block_acct_invalid(blk_get_stats(n->conf.blk), acct);
395 return NVME_INVALID_FIELD | NVME_DNR;
396 }
397
398 dma_acct_start(n->conf.blk, &req->acct, &req->qsg, acct);
399 if (req->qsg.nsg > 0) {
400 req->has_sg = true;
401 req->aiocb = is_write ?
402 dma_blk_write(n->conf.blk, &req->qsg, data_offset, BDRV_SECTOR_SIZE,
403 nvme_rw_cb, req) :
404 dma_blk_read(n->conf.blk, &req->qsg, data_offset, BDRV_SECTOR_SIZE,
405 nvme_rw_cb, req);
406 } else {
407 req->has_sg = false;
408 req->aiocb = is_write ?
409 blk_aio_pwritev(n->conf.blk, data_offset, &req->iov, 0, nvme_rw_cb,
410 req) :
411 blk_aio_preadv(n->conf.blk, data_offset, &req->iov, 0, nvme_rw_cb,
412 req);
413 }
414
415 return NVME_NO_COMPLETE;
416}
417
418static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
419{
420 NvmeNamespace *ns;
421 uint32_t nsid = le32_to_cpu(cmd->nsid);
422
423 if (unlikely(nsid == 0 || nsid > n->num_namespaces)) {
424 trace_nvme_err_invalid_ns(nsid, n->num_namespaces);
425 return NVME_INVALID_NSID | NVME_DNR;
426 }
427
428 ns = &n->namespaces[nsid - 1];
429 switch (cmd->opcode) {
430 case NVME_CMD_FLUSH:
431 return nvme_flush(n, ns, cmd, req);
432 case NVME_CMD_WRITE_ZEROS:
433 return nvme_write_zeros(n, ns, cmd, req);
434 case NVME_CMD_WRITE:
435 case NVME_CMD_READ:
436 return nvme_rw(n, ns, cmd, req);
437 default:
438 trace_nvme_err_invalid_opc(cmd->opcode);
439 return NVME_INVALID_OPCODE | NVME_DNR;
440 }
441}
442
443static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
444{
445 n->sq[sq->sqid] = NULL;
446 timer_del(sq->timer);
447 timer_free(sq->timer);
448 g_free(sq->io_req);
449 if (sq->sqid) {
450 g_free(sq);
451 }
452}
453
454static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeCmd *cmd)
455{
456 NvmeDeleteQ *c = (NvmeDeleteQ *)cmd;
457 NvmeRequest *req, *next;
458 NvmeSQueue *sq;
459 NvmeCQueue *cq;
460 uint16_t qid = le16_to_cpu(c->qid);
461
462 if (unlikely(!qid || nvme_check_sqid(n, qid))) {
463 trace_nvme_err_invalid_del_sq(qid);
464 return NVME_INVALID_QID | NVME_DNR;
465 }
466
467 trace_nvme_del_sq(qid);
468
469 sq = n->sq[qid];
470 while (!QTAILQ_EMPTY(&sq->out_req_list)) {
471 req = QTAILQ_FIRST(&sq->out_req_list);
472 assert(req->aiocb);
473 blk_aio_cancel(req->aiocb);
474 }
475 if (!nvme_check_cqid(n, sq->cqid)) {
476 cq = n->cq[sq->cqid];
477 QTAILQ_REMOVE(&cq->sq_list, sq, entry);
478
479 nvme_post_cqes(cq);
480 QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
481 if (req->sq == sq) {
482 QTAILQ_REMOVE(&cq->req_list, req, entry);
483 QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
484 }
485 }
486 }
487
488 nvme_free_sq(sq, n);
489 return NVME_SUCCESS;
490}
491
492static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
493 uint16_t sqid, uint16_t cqid, uint16_t size)
494{
495 int i;
496 NvmeCQueue *cq;
497
498 sq->ctrl = n;
499 sq->dma_addr = dma_addr;
500 sq->sqid = sqid;
501 sq->size = size;
502 sq->cqid = cqid;
503 sq->head = sq->tail = 0;
504 sq->io_req = g_new(NvmeRequest, sq->size);
505
506 QTAILQ_INIT(&sq->req_list);
507 QTAILQ_INIT(&sq->out_req_list);
508 for (i = 0; i < sq->size; i++) {
509 sq->io_req[i].sq = sq;
510 QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
511 }
512 sq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_process_sq, sq);
513
514 assert(n->cq[cqid]);
515 cq = n->cq[cqid];
516 QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
517 n->sq[sqid] = sq;
518}
519
520static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeCmd *cmd)
521{
522 NvmeSQueue *sq;
523 NvmeCreateSq *c = (NvmeCreateSq *)cmd;
524
525 uint16_t cqid = le16_to_cpu(c->cqid);
526 uint16_t sqid = le16_to_cpu(c->sqid);
527 uint16_t qsize = le16_to_cpu(c->qsize);
528 uint16_t qflags = le16_to_cpu(c->sq_flags);
529 uint64_t prp1 = le64_to_cpu(c->prp1);
530
531 trace_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
532
533 if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
534 trace_nvme_err_invalid_create_sq_cqid(cqid);
535 return NVME_INVALID_CQID | NVME_DNR;
536 }
537 if (unlikely(!sqid || !nvme_check_sqid(n, sqid))) {
538 trace_nvme_err_invalid_create_sq_sqid(sqid);
539 return NVME_INVALID_QID | NVME_DNR;
540 }
541 if (unlikely(!qsize || qsize > NVME_CAP_MQES(n->bar.cap))) {
542 trace_nvme_err_invalid_create_sq_size(qsize);
543 return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
544 }
545 if (unlikely(!prp1 || prp1 & (n->page_size - 1))) {
546 trace_nvme_err_invalid_create_sq_addr(prp1);
547 return NVME_INVALID_FIELD | NVME_DNR;
548 }
549 if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
550 trace_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
551 return NVME_INVALID_FIELD | NVME_DNR;
552 }
553 sq = g_malloc0(sizeof(*sq));
554 nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
555 return NVME_SUCCESS;
556}
557
558static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
559{
560 n->cq[cq->cqid] = NULL;
561 timer_del(cq->timer);
562 timer_free(cq->timer);
563 msix_vector_unuse(&n->parent_obj, cq->vector);
564 if (cq->cqid) {
565 g_free(cq);
566 }
567}
568
569static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeCmd *cmd)
570{
571 NvmeDeleteQ *c = (NvmeDeleteQ *)cmd;
572 NvmeCQueue *cq;
573 uint16_t qid = le16_to_cpu(c->qid);
574
575 if (unlikely(!qid || nvme_check_cqid(n, qid))) {
576 trace_nvme_err_invalid_del_cq_cqid(qid);
577 return NVME_INVALID_CQID | NVME_DNR;
578 }
579
580 cq = n->cq[qid];
581 if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
582 trace_nvme_err_invalid_del_cq_notempty(qid);
583 return NVME_INVALID_QUEUE_DEL;
584 }
585 nvme_irq_deassert(n, cq);
586 trace_nvme_del_cq(qid);
587 nvme_free_cq(cq, n);
588 return NVME_SUCCESS;
589}
590
591static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
592 uint16_t cqid, uint16_t vector, uint16_t size, uint16_t irq_enabled)
593{
594 cq->ctrl = n;
595 cq->cqid = cqid;
596 cq->size = size;
597 cq->dma_addr = dma_addr;
598 cq->phase = 1;
599 cq->irq_enabled = irq_enabled;
600 cq->vector = vector;
601 cq->head = cq->tail = 0;
602 QTAILQ_INIT(&cq->req_list);
603 QTAILQ_INIT(&cq->sq_list);
604 msix_vector_use(&n->parent_obj, cq->vector);
605 n->cq[cqid] = cq;
606 cq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_post_cqes, cq);
607}
608
609static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeCmd *cmd)
610{
611 NvmeCQueue *cq;
612 NvmeCreateCq *c = (NvmeCreateCq *)cmd;
613 uint16_t cqid = le16_to_cpu(c->cqid);
614 uint16_t vector = le16_to_cpu(c->irq_vector);
615 uint16_t qsize = le16_to_cpu(c->qsize);
616 uint16_t qflags = le16_to_cpu(c->cq_flags);
617 uint64_t prp1 = le64_to_cpu(c->prp1);
618
619 trace_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
620 NVME_CQ_FLAGS_IEN(qflags) != 0);
621
622 if (unlikely(!cqid || !nvme_check_cqid(n, cqid))) {
623 trace_nvme_err_invalid_create_cq_cqid(cqid);
624 return NVME_INVALID_CQID | NVME_DNR;
625 }
626 if (unlikely(!qsize || qsize > NVME_CAP_MQES(n->bar.cap))) {
627 trace_nvme_err_invalid_create_cq_size(qsize);
628 return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
629 }
630 if (unlikely(!prp1)) {
631 trace_nvme_err_invalid_create_cq_addr(prp1);
632 return NVME_INVALID_FIELD | NVME_DNR;
633 }
634 if (unlikely(vector > n->num_queues)) {
635 trace_nvme_err_invalid_create_cq_vector(vector);
636 return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
637 }
638 if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
639 trace_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
640 return NVME_INVALID_FIELD | NVME_DNR;
641 }
642
643 cq = g_malloc0(sizeof(*cq));
644 nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
645 NVME_CQ_FLAGS_IEN(qflags));
646 return NVME_SUCCESS;
647}
648
649static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeIdentify *c)
650{
651 uint64_t prp1 = le64_to_cpu(c->prp1);
652 uint64_t prp2 = le64_to_cpu(c->prp2);
653
654 trace_nvme_identify_ctrl();
655
656 return nvme_dma_read_prp(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl),
657 prp1, prp2);
658}
659
660static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeIdentify *c)
661{
662 NvmeNamespace *ns;
663 uint32_t nsid = le32_to_cpu(c->nsid);
664 uint64_t prp1 = le64_to_cpu(c->prp1);
665 uint64_t prp2 = le64_to_cpu(c->prp2);
666
667 trace_nvme_identify_ns(nsid);
668
669 if (unlikely(nsid == 0 || nsid > n->num_namespaces)) {
670 trace_nvme_err_invalid_ns(nsid, n->num_namespaces);
671 return NVME_INVALID_NSID | NVME_DNR;
672 }
673
674 ns = &n->namespaces[nsid - 1];
675
676 return nvme_dma_read_prp(n, (uint8_t *)&ns->id_ns, sizeof(ns->id_ns),
677 prp1, prp2);
678}
679
680static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeIdentify *c)
681{
682 static const int data_len = 4 * KiB;
683 uint32_t min_nsid = le32_to_cpu(c->nsid);
684 uint64_t prp1 = le64_to_cpu(c->prp1);
685 uint64_t prp2 = le64_to_cpu(c->prp2);
686 uint32_t *list;
687 uint16_t ret;
688 int i, j = 0;
689
690 trace_nvme_identify_nslist(min_nsid);
691
692 list = g_malloc0(data_len);
693 for (i = 0; i < n->num_namespaces; i++) {
694 if (i < min_nsid) {
695 continue;
696 }
697 list[j++] = cpu_to_le32(i + 1);
698 if (j == data_len / sizeof(uint32_t)) {
699 break;
700 }
701 }
702 ret = nvme_dma_read_prp(n, (uint8_t *)list, data_len, prp1, prp2);
703 g_free(list);
704 return ret;
705}
706
707static uint16_t nvme_identify(NvmeCtrl *n, NvmeCmd *cmd)
708{
709 NvmeIdentify *c = (NvmeIdentify *)cmd;
710
711 switch (le32_to_cpu(c->cns)) {
712 case 0x00:
713 return nvme_identify_ns(n, c);
714 case 0x01:
715 return nvme_identify_ctrl(n, c);
716 case 0x02:
717 return nvme_identify_nslist(n, c);
718 default:
719 trace_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
720 return NVME_INVALID_FIELD | NVME_DNR;
721 }
722}
723
724static inline void nvme_set_timestamp(NvmeCtrl *n, uint64_t ts)
725{
726 trace_nvme_setfeat_timestamp(ts);
727
728 n->host_timestamp = le64_to_cpu(ts);
729 n->timestamp_set_qemu_clock_ms = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
730}
731
732static inline uint64_t nvme_get_timestamp(const NvmeCtrl *n)
733{
734 uint64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL);
735 uint64_t elapsed_time = current_time - n->timestamp_set_qemu_clock_ms;
736
737 union nvme_timestamp {
738 struct {
739 uint64_t timestamp:48;
740 uint64_t sync:1;
741 uint64_t origin:3;
742 uint64_t rsvd1:12;
743 };
744 uint64_t all;
745 };
746
747 union nvme_timestamp ts;
748 ts.all = 0;
749
750 /*
751 * If the sum of the Timestamp value set by the host and the elapsed
752 * time exceeds 2^48, the value returned should be reduced modulo 2^48.
753 */
754 ts.timestamp = (n->host_timestamp + elapsed_time) & 0xffffffffffff;
755
756 /* If the host timestamp is non-zero, set the timestamp origin */
757 ts.origin = n->host_timestamp ? 0x01 : 0x00;
758
759 trace_nvme_getfeat_timestamp(ts.all);
760
761 return cpu_to_le64(ts.all);
762}
763
764static uint16_t nvme_get_feature_timestamp(NvmeCtrl *n, NvmeCmd *cmd)
765{
766 uint64_t prp1 = le64_to_cpu(cmd->prp1);
767 uint64_t prp2 = le64_to_cpu(cmd->prp2);
768
769 uint64_t timestamp = nvme_get_timestamp(n);
770
771 return nvme_dma_read_prp(n, (uint8_t *)&timestamp,
772 sizeof(timestamp), prp1, prp2);
773}
774
775static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
776{
777 uint32_t dw10 = le32_to_cpu(cmd->cdw10);
778 uint32_t result;
779
780 switch (dw10) {
781 case NVME_VOLATILE_WRITE_CACHE:
782 result = blk_enable_write_cache(n->conf.blk);
783 trace_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
784 break;
785 case NVME_NUMBER_OF_QUEUES:
786 result = cpu_to_le32((n->num_queues - 2) | ((n->num_queues - 2) << 16));
787 trace_nvme_getfeat_numq(result);
788 break;
789 case NVME_TIMESTAMP:
790 return nvme_get_feature_timestamp(n, cmd);
791 break;
792 default:
793 trace_nvme_err_invalid_getfeat(dw10);
794 return NVME_INVALID_FIELD | NVME_DNR;
795 }
796
797 req->cqe.result = result;
798 return NVME_SUCCESS;
799}
800
801static uint16_t nvme_set_feature_timestamp(NvmeCtrl *n, NvmeCmd *cmd)
802{
803 uint16_t ret;
804 uint64_t timestamp;
805 uint64_t prp1 = le64_to_cpu(cmd->prp1);
806 uint64_t prp2 = le64_to_cpu(cmd->prp2);
807
808 ret = nvme_dma_write_prp(n, (uint8_t *)&timestamp,
809 sizeof(timestamp), prp1, prp2);
810 if (ret != NVME_SUCCESS) {
811 return ret;
812 }
813
814 nvme_set_timestamp(n, timestamp);
815
816 return NVME_SUCCESS;
817}
818
819static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
820{
821 uint32_t dw10 = le32_to_cpu(cmd->cdw10);
822 uint32_t dw11 = le32_to_cpu(cmd->cdw11);
823
824 switch (dw10) {
825 case NVME_VOLATILE_WRITE_CACHE:
826 blk_set_enable_write_cache(n->conf.blk, dw11 & 1);
827 break;
828 case NVME_NUMBER_OF_QUEUES:
829 trace_nvme_setfeat_numq((dw11 & 0xFFFF) + 1,
830 ((dw11 >> 16) & 0xFFFF) + 1,
831 n->num_queues - 1, n->num_queues - 1);
832 req->cqe.result =
833 cpu_to_le32((n->num_queues - 2) | ((n->num_queues - 2) << 16));
834 break;
835
836 case NVME_TIMESTAMP:
837 return nvme_set_feature_timestamp(n, cmd);
838 break;
839
840 default:
841 trace_nvme_err_invalid_setfeat(dw10);
842 return NVME_INVALID_FIELD | NVME_DNR;
843 }
844 return NVME_SUCCESS;
845}
846
847static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
848{
849 switch (cmd->opcode) {
850 case NVME_ADM_CMD_DELETE_SQ:
851 return nvme_del_sq(n, cmd);
852 case NVME_ADM_CMD_CREATE_SQ:
853 return nvme_create_sq(n, cmd);
854 case NVME_ADM_CMD_DELETE_CQ:
855 return nvme_del_cq(n, cmd);
856 case NVME_ADM_CMD_CREATE_CQ:
857 return nvme_create_cq(n, cmd);
858 case NVME_ADM_CMD_IDENTIFY:
859 return nvme_identify(n, cmd);
860 case NVME_ADM_CMD_SET_FEATURES:
861 return nvme_set_feature(n, cmd, req);
862 case NVME_ADM_CMD_GET_FEATURES:
863 return nvme_get_feature(n, cmd, req);
864 default:
865 trace_nvme_err_invalid_admin_opc(cmd->opcode);
866 return NVME_INVALID_OPCODE | NVME_DNR;
867 }
868}
869
870static void nvme_process_sq(void *opaque)
871{
872 NvmeSQueue *sq = opaque;
873 NvmeCtrl *n = sq->ctrl;
874 NvmeCQueue *cq = n->cq[sq->cqid];
875
876 uint16_t status;
877 hwaddr addr;
878 NvmeCmd cmd;
879 NvmeRequest *req;
880
881 while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
882 addr = sq->dma_addr + sq->head * n->sqe_size;
883 nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd));
884 nvme_inc_sq_head(sq);
885
886 req = QTAILQ_FIRST(&sq->req_list);
887 QTAILQ_REMOVE(&sq->req_list, req, entry);
888 QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
889 memset(&req->cqe, 0, sizeof(req->cqe));
890 req->cqe.cid = cmd.cid;
891
892 status = sq->sqid ? nvme_io_cmd(n, &cmd, req) :
893 nvme_admin_cmd(n, &cmd, req);
894 if (status != NVME_NO_COMPLETE) {
895 req->status = status;
896 nvme_enqueue_req_completion(cq, req);
897 }
898 }
899}
900
901static void nvme_clear_ctrl(NvmeCtrl *n)
902{
903 int i;
904
905 blk_drain(n->conf.blk);
906
907 for (i = 0; i < n->num_queues; i++) {
908 if (n->sq[i] != NULL) {
909 nvme_free_sq(n->sq[i], n);
910 }
911 }
912 for (i = 0; i < n->num_queues; i++) {
913 if (n->cq[i] != NULL) {
914 nvme_free_cq(n->cq[i], n);
915 }
916 }
917
918 blk_flush(n->conf.blk);
919 n->bar.cc = 0;
920}
921
922static int nvme_start_ctrl(NvmeCtrl *n)
923{
924 uint32_t page_bits = NVME_CC_MPS(n->bar.cc) + 12;
925 uint32_t page_size = 1 << page_bits;
926
927 if (unlikely(n->cq[0])) {
928 trace_nvme_err_startfail_cq();
929 return -1;
930 }
931 if (unlikely(n->sq[0])) {
932 trace_nvme_err_startfail_sq();
933 return -1;
934 }
935 if (unlikely(!n->bar.asq)) {
936 trace_nvme_err_startfail_nbarasq();
937 return -1;
938 }
939 if (unlikely(!n->bar.acq)) {
940 trace_nvme_err_startfail_nbaracq();
941 return -1;
942 }
943 if (unlikely(n->bar.asq & (page_size - 1))) {
944 trace_nvme_err_startfail_asq_misaligned(n->bar.asq);
945 return -1;
946 }
947 if (unlikely(n->bar.acq & (page_size - 1))) {
948 trace_nvme_err_startfail_acq_misaligned(n->bar.acq);
949 return -1;
950 }
951 if (unlikely(NVME_CC_MPS(n->bar.cc) <
952 NVME_CAP_MPSMIN(n->bar.cap))) {
953 trace_nvme_err_startfail_page_too_small(
954 NVME_CC_MPS(n->bar.cc),
955 NVME_CAP_MPSMIN(n->bar.cap));
956 return -1;
957 }
958 if (unlikely(NVME_CC_MPS(n->bar.cc) >
959 NVME_CAP_MPSMAX(n->bar.cap))) {
960 trace_nvme_err_startfail_page_too_large(
961 NVME_CC_MPS(n->bar.cc),
962 NVME_CAP_MPSMAX(n->bar.cap));
963 return -1;
964 }
965 if (unlikely(NVME_CC_IOCQES(n->bar.cc) <
966 NVME_CTRL_CQES_MIN(n->id_ctrl.cqes))) {
967 trace_nvme_err_startfail_cqent_too_small(
968 NVME_CC_IOCQES(n->bar.cc),
969 NVME_CTRL_CQES_MIN(n->bar.cap));
970 return -1;
971 }
972 if (unlikely(NVME_CC_IOCQES(n->bar.cc) >
973 NVME_CTRL_CQES_MAX(n->id_ctrl.cqes))) {
974 trace_nvme_err_startfail_cqent_too_large(
975 NVME_CC_IOCQES(n->bar.cc),
976 NVME_CTRL_CQES_MAX(n->bar.cap));
977 return -1;
978 }
979 if (unlikely(NVME_CC_IOSQES(n->bar.cc) <
980 NVME_CTRL_SQES_MIN(n->id_ctrl.sqes))) {
981 trace_nvme_err_startfail_sqent_too_small(
982 NVME_CC_IOSQES(n->bar.cc),
983 NVME_CTRL_SQES_MIN(n->bar.cap));
984 return -1;
985 }
986 if (unlikely(NVME_CC_IOSQES(n->bar.cc) >
987 NVME_CTRL_SQES_MAX(n->id_ctrl.sqes))) {
988 trace_nvme_err_startfail_sqent_too_large(
989 NVME_CC_IOSQES(n->bar.cc),
990 NVME_CTRL_SQES_MAX(n->bar.cap));
991 return -1;
992 }
993 if (unlikely(!NVME_AQA_ASQS(n->bar.aqa))) {
994 trace_nvme_err_startfail_asqent_sz_zero();
995 return -1;
996 }
997 if (unlikely(!NVME_AQA_ACQS(n->bar.aqa))) {
998 trace_nvme_err_startfail_acqent_sz_zero();
999 return -1;
1000 }
1001
1002 n->page_bits = page_bits;
1003 n->page_size = page_size;
1004 n->max_prp_ents = n->page_size / sizeof(uint64_t);
1005 n->cqe_size = 1 << NVME_CC_IOCQES(n->bar.cc);
1006 n->sqe_size = 1 << NVME_CC_IOSQES(n->bar.cc);
1007 nvme_init_cq(&n->admin_cq, n, n->bar.acq, 0, 0,
1008 NVME_AQA_ACQS(n->bar.aqa) + 1, 1);
1009 nvme_init_sq(&n->admin_sq, n, n->bar.asq, 0, 0,
1010 NVME_AQA_ASQS(n->bar.aqa) + 1);
1011
1012 nvme_set_timestamp(n, 0ULL);
1013
1014 return 0;
1015}
1016
1017static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
1018 unsigned size)
1019{
1020 if (unlikely(offset & (sizeof(uint32_t) - 1))) {
1021 NVME_GUEST_ERR(nvme_ub_mmiowr_misaligned32,
1022 "MMIO write not 32-bit aligned,"
1023 " offset=0x%"PRIx64"", offset);
1024 /* should be ignored, fall through for now */
1025 }
1026
1027 if (unlikely(size < sizeof(uint32_t))) {
1028 NVME_GUEST_ERR(nvme_ub_mmiowr_toosmall,
1029 "MMIO write smaller than 32-bits,"
1030 " offset=0x%"PRIx64", size=%u",
1031 offset, size);
1032 /* should be ignored, fall through for now */
1033 }
1034
1035 switch (offset) {
1036 case 0xc: /* INTMS */
1037 if (unlikely(msix_enabled(&(n->parent_obj)))) {
1038 NVME_GUEST_ERR(nvme_ub_mmiowr_intmask_with_msix,
1039 "undefined access to interrupt mask set"
1040 " when MSI-X is enabled");
1041 /* should be ignored, fall through for now */
1042 }
1043 n->bar.intms |= data & 0xffffffff;
1044 n->bar.intmc = n->bar.intms;
1045 trace_nvme_mmio_intm_set(data & 0xffffffff,
1046 n->bar.intmc);
1047 nvme_irq_check(n);
1048 break;
1049 case 0x10: /* INTMC */
1050 if (unlikely(msix_enabled(&(n->parent_obj)))) {
1051 NVME_GUEST_ERR(nvme_ub_mmiowr_intmask_with_msix,
1052 "undefined access to interrupt mask clr"
1053 " when MSI-X is enabled");
1054 /* should be ignored, fall through for now */
1055 }
1056 n->bar.intms &= ~(data & 0xffffffff);
1057 n->bar.intmc = n->bar.intms;
1058 trace_nvme_mmio_intm_clr(data & 0xffffffff,
1059 n->bar.intmc);
1060 nvme_irq_check(n);
1061 break;
1062 case 0x14: /* CC */
1063 trace_nvme_mmio_cfg(data & 0xffffffff);
1064 /* Windows first sends data, then sends enable bit */
1065 if (!NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc) &&
1066 !NVME_CC_SHN(data) && !NVME_CC_SHN(n->bar.cc))
1067 {
1068 n->bar.cc = data;
1069 }
1070
1071 if (NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc)) {
1072 n->bar.cc = data;
1073 if (unlikely(nvme_start_ctrl(n))) {
1074 trace_nvme_err_startfail();
1075 n->bar.csts = NVME_CSTS_FAILED;
1076 } else {
1077 trace_nvme_mmio_start_success();
1078 n->bar.csts = NVME_CSTS_READY;
1079 }
1080 } else if (!NVME_CC_EN(data) && NVME_CC_EN(n->bar.cc)) {
1081 trace_nvme_mmio_stopped();
1082 nvme_clear_ctrl(n);
1083 n->bar.csts &= ~NVME_CSTS_READY;
1084 }
1085 if (NVME_CC_SHN(data) && !(NVME_CC_SHN(n->bar.cc))) {
1086 trace_nvme_mmio_shutdown_set();
1087 nvme_clear_ctrl(n);
1088 n->bar.cc = data;
1089 n->bar.csts |= NVME_CSTS_SHST_COMPLETE;
1090 } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(n->bar.cc)) {
1091 trace_nvme_mmio_shutdown_cleared();
1092 n->bar.csts &= ~NVME_CSTS_SHST_COMPLETE;
1093 n->bar.cc = data;
1094 }
1095 break;
1096 case 0x1C: /* CSTS */
1097 if (data & (1 << 4)) {
1098 NVME_GUEST_ERR(nvme_ub_mmiowr_ssreset_w1c_unsupported,
1099 "attempted to W1C CSTS.NSSRO"
1100 " but CAP.NSSRS is zero (not supported)");
1101 } else if (data != 0) {
1102 NVME_GUEST_ERR(nvme_ub_mmiowr_ro_csts,
1103 "attempted to set a read only bit"
1104 " of controller status");
1105 }
1106 break;
1107 case 0x20: /* NSSR */
1108 if (data == 0x4E564D65) {
1109 trace_nvme_ub_mmiowr_ssreset_unsupported();
1110 } else {
1111 /* The spec says that writes of other values have no effect */
1112 return;
1113 }
1114 break;
1115 case 0x24: /* AQA */
1116 n->bar.aqa = data & 0xffffffff;
1117 trace_nvme_mmio_aqattr(data & 0xffffffff);
1118 break;
1119 case 0x28: /* ASQ */
1120 n->bar.asq = data;
1121 trace_nvme_mmio_asqaddr(data);
1122 break;
1123 case 0x2c: /* ASQ hi */
1124 n->bar.asq |= data << 32;
1125 trace_nvme_mmio_asqaddr_hi(data, n->bar.asq);
1126 break;
1127 case 0x30: /* ACQ */
1128 trace_nvme_mmio_acqaddr(data);
1129 n->bar.acq = data;
1130 break;
1131 case 0x34: /* ACQ hi */
1132 n->bar.acq |= data << 32;
1133 trace_nvme_mmio_acqaddr_hi(data, n->bar.acq);
1134 break;
1135 case 0x38: /* CMBLOC */
1136 NVME_GUEST_ERR(nvme_ub_mmiowr_cmbloc_reserved,
1137 "invalid write to reserved CMBLOC"
1138 " when CMBSZ is zero, ignored");
1139 return;
1140 case 0x3C: /* CMBSZ */
1141 NVME_GUEST_ERR(nvme_ub_mmiowr_cmbsz_readonly,
1142 "invalid write to read only CMBSZ, ignored");
1143 return;
1144 default:
1145 NVME_GUEST_ERR(nvme_ub_mmiowr_invalid,
1146 "invalid MMIO write,"
1147 " offset=0x%"PRIx64", data=%"PRIx64"",
1148 offset, data);
1149 break;
1150 }
1151}
1152
1153static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
1154{
1155 NvmeCtrl *n = (NvmeCtrl *)opaque;
1156 uint8_t *ptr = (uint8_t *)&n->bar;
1157 uint64_t val = 0;
1158
1159 if (unlikely(addr & (sizeof(uint32_t) - 1))) {
1160 NVME_GUEST_ERR(nvme_ub_mmiord_misaligned32,
1161 "MMIO read not 32-bit aligned,"
1162 " offset=0x%"PRIx64"", addr);
1163 /* should RAZ, fall through for now */
1164 } else if (unlikely(size < sizeof(uint32_t))) {
1165 NVME_GUEST_ERR(nvme_ub_mmiord_toosmall,
1166 "MMIO read smaller than 32-bits,"
1167 " offset=0x%"PRIx64"", addr);
1168 /* should RAZ, fall through for now */
1169 }
1170
1171 if (addr < sizeof(n->bar)) {
1172 memcpy(&val, ptr + addr, size);
1173 } else {
1174 NVME_GUEST_ERR(nvme_ub_mmiord_invalid_ofs,
1175 "MMIO read beyond last register,"
1176 " offset=0x%"PRIx64", returning 0", addr);
1177 }
1178
1179 return val;
1180}
1181
1182static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
1183{
1184 uint32_t qid;
1185
1186 if (unlikely(addr & ((1 << 2) - 1))) {
1187 NVME_GUEST_ERR(nvme_ub_db_wr_misaligned,
1188 "doorbell write not 32-bit aligned,"
1189 " offset=0x%"PRIx64", ignoring", addr);
1190 return;
1191 }
1192
1193 if (((addr - 0x1000) >> 2) & 1) {
1194 /* Completion queue doorbell write */
1195
1196 uint16_t new_head = val & 0xffff;
1197 int start_sqs;
1198 NvmeCQueue *cq;
1199
1200 qid = (addr - (0x1000 + (1 << 2))) >> 3;
1201 if (unlikely(nvme_check_cqid(n, qid))) {
1202 NVME_GUEST_ERR(nvme_ub_db_wr_invalid_cq,
1203 "completion queue doorbell write"
1204 " for nonexistent queue,"
1205 " sqid=%"PRIu32", ignoring", qid);
1206 return;
1207 }
1208
1209 cq = n->cq[qid];
1210 if (unlikely(new_head >= cq->size)) {
1211 NVME_GUEST_ERR(nvme_ub_db_wr_invalid_cqhead,
1212 "completion queue doorbell write value"
1213 " beyond queue size, sqid=%"PRIu32","
1214 " new_head=%"PRIu16", ignoring",
1215 qid, new_head);
1216 return;
1217 }
1218
1219 start_sqs = nvme_cq_full(cq) ? 1 : 0;
1220 cq->head = new_head;
1221 if (start_sqs) {
1222 NvmeSQueue *sq;
1223 QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
1224 timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
1225 }
1226 timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
1227 }
1228
1229 if (cq->tail == cq->head) {
1230 nvme_irq_deassert(n, cq);
1231 }
1232 } else {
1233 /* Submission queue doorbell write */
1234
1235 uint16_t new_tail = val & 0xffff;
1236 NvmeSQueue *sq;
1237
1238 qid = (addr - 0x1000) >> 3;
1239 if (unlikely(nvme_check_sqid(n, qid))) {
1240 NVME_GUEST_ERR(nvme_ub_db_wr_invalid_sq,
1241 "submission queue doorbell write"
1242 " for nonexistent queue,"
1243 " sqid=%"PRIu32", ignoring", qid);
1244 return;
1245 }
1246
1247 sq = n->sq[qid];
1248 if (unlikely(new_tail >= sq->size)) {
1249 NVME_GUEST_ERR(nvme_ub_db_wr_invalid_sqtail,
1250 "submission queue doorbell write value"
1251 " beyond queue size, sqid=%"PRIu32","
1252 " new_tail=%"PRIu16", ignoring",
1253 qid, new_tail);
1254 return;
1255 }
1256
1257 sq->tail = new_tail;
1258 timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
1259 }
1260}
1261
1262static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
1263 unsigned size)
1264{
1265 NvmeCtrl *n = (NvmeCtrl *)opaque;
1266 if (addr < sizeof(n->bar)) {
1267 nvme_write_bar(n, addr, data, size);
1268 } else if (addr >= 0x1000) {
1269 nvme_process_db(n, addr, data);
1270 }
1271}
1272
1273static const MemoryRegionOps nvme_mmio_ops = {
1274 .read = nvme_mmio_read,
1275 .write = nvme_mmio_write,
1276 .endianness = DEVICE_LITTLE_ENDIAN,
1277 .impl = {
1278 .min_access_size = 2,
1279 .max_access_size = 8,
1280 },
1281};
1282
1283static void nvme_cmb_write(void *opaque, hwaddr addr, uint64_t data,
1284 unsigned size)
1285{
1286 NvmeCtrl *n = (NvmeCtrl *)opaque;
1287 stn_le_p(&n->cmbuf[addr], size, data);
1288}
1289
1290static uint64_t nvme_cmb_read(void *opaque, hwaddr addr, unsigned size)
1291{
1292 NvmeCtrl *n = (NvmeCtrl *)opaque;
1293 return ldn_le_p(&n->cmbuf[addr], size);
1294}
1295
1296static const MemoryRegionOps nvme_cmb_ops = {
1297 .read = nvme_cmb_read,
1298 .write = nvme_cmb_write,
1299 .endianness = DEVICE_LITTLE_ENDIAN,
1300 .impl = {
1301 .min_access_size = 1,
1302 .max_access_size = 8,
1303 },
1304};
1305
1306static void nvme_realize(PCIDevice *pci_dev, Error **errp)
1307{
1308 NvmeCtrl *n = NVME(pci_dev);
1309 NvmeIdCtrl *id = &n->id_ctrl;
1310
1311 int i;
1312 int64_t bs_size;
1313 uint8_t *pci_conf;
1314
1315 if (!n->num_queues) {
1316 error_setg(errp, "num_queues can't be zero");
1317 return;
1318 }
1319
1320 if (!n->conf.blk) {
1321 error_setg(errp, "drive property not set");
1322 return;
1323 }
1324
1325 bs_size = blk_getlength(n->conf.blk);
1326 if (bs_size < 0) {
1327 error_setg(errp, "could not get backing file size");
1328 return;
1329 }
1330
1331 if (!n->serial) {
1332 error_setg(errp, "serial property not set");
1333 return;
1334 }
1335 blkconf_blocksizes(&n->conf);
1336 if (!blkconf_apply_backend_options(&n->conf, blk_is_read_only(n->conf.blk),
1337 false, errp)) {
1338 return;
1339 }
1340
1341 pci_conf = pci_dev->config;
1342 pci_conf[PCI_INTERRUPT_PIN] = 1;
1343 pci_config_set_prog_interface(pci_dev->config, 0x2);
1344 pci_config_set_class(pci_dev->config, PCI_CLASS_STORAGE_EXPRESS);
1345 pcie_endpoint_cap_init(pci_dev, 0x80);
1346
1347 n->num_namespaces = 1;
1348 n->reg_size = pow2ceil(0x1004 + 2 * (n->num_queues + 1) * 4);
1349 n->ns_size = bs_size / (uint64_t)n->num_namespaces;
1350
1351 n->namespaces = g_new0(NvmeNamespace, n->num_namespaces);
1352 n->sq = g_new0(NvmeSQueue *, n->num_queues);
1353 n->cq = g_new0(NvmeCQueue *, n->num_queues);
1354
1355 memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n,
1356 "nvme", n->reg_size);
1357 pci_register_bar(pci_dev, 0,
1358 PCI_BASE_ADDRESS_SPACE_MEMORY | PCI_BASE_ADDRESS_MEM_TYPE_64,
1359 &n->iomem);
1360 msix_init_exclusive_bar(pci_dev, n->num_queues, 4, NULL);
1361
1362 id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
1363 id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
1364 strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
1365 strpadcpy((char *)id->fr, sizeof(id->fr), "1.0", ' ');
1366 strpadcpy((char *)id->sn, sizeof(id->sn), n->serial, ' ');
1367 id->rab = 6;
1368 id->ieee[0] = 0x00;
1369 id->ieee[1] = 0x02;
1370 id->ieee[2] = 0xb3;
1371 id->oacs = cpu_to_le16(0);
1372 id->frmw = 7 << 1;
1373 id->lpa = 1 << 0;
1374 id->sqes = (0x6 << 4) | 0x6;
1375 id->cqes = (0x4 << 4) | 0x4;
1376 id->nn = cpu_to_le32(n->num_namespaces);
1377 id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROS | NVME_ONCS_TIMESTAMP);
1378 id->psd[0].mp = cpu_to_le16(0x9c4);
1379 id->psd[0].enlat = cpu_to_le32(0x10);
1380 id->psd[0].exlat = cpu_to_le32(0x4);
1381 if (blk_enable_write_cache(n->conf.blk)) {
1382 id->vwc = 1;
1383 }
1384
1385 n->bar.cap = 0;
1386 NVME_CAP_SET_MQES(n->bar.cap, 0x7ff);
1387 NVME_CAP_SET_CQR(n->bar.cap, 1);
1388 NVME_CAP_SET_TO(n->bar.cap, 0xf);
1389 NVME_CAP_SET_CSS(n->bar.cap, 1);
1390 NVME_CAP_SET_MPSMAX(n->bar.cap, 4);
1391
1392 n->bar.vs = 0x00010200;
1393 n->bar.intmc = n->bar.intms = 0;
1394
1395 if (n->cmb_size_mb) {
1396
1397 NVME_CMBLOC_SET_BIR(n->bar.cmbloc, 2);
1398 NVME_CMBLOC_SET_OFST(n->bar.cmbloc, 0);
1399
1400 NVME_CMBSZ_SET_SQS(n->bar.cmbsz, 1);
1401 NVME_CMBSZ_SET_CQS(n->bar.cmbsz, 0);
1402 NVME_CMBSZ_SET_LISTS(n->bar.cmbsz, 0);
1403 NVME_CMBSZ_SET_RDS(n->bar.cmbsz, 1);
1404 NVME_CMBSZ_SET_WDS(n->bar.cmbsz, 1);
1405 NVME_CMBSZ_SET_SZU(n->bar.cmbsz, 2); /* MBs */
1406 NVME_CMBSZ_SET_SZ(n->bar.cmbsz, n->cmb_size_mb);
1407
1408 n->cmbloc = n->bar.cmbloc;
1409 n->cmbsz = n->bar.cmbsz;
1410
1411 n->cmbuf = g_malloc0(NVME_CMBSZ_GETSIZE(n->bar.cmbsz));
1412 memory_region_init_io(&n->ctrl_mem, OBJECT(n), &nvme_cmb_ops, n,
1413 "nvme-cmb", NVME_CMBSZ_GETSIZE(n->bar.cmbsz));
1414 pci_register_bar(pci_dev, NVME_CMBLOC_BIR(n->bar.cmbloc),
1415 PCI_BASE_ADDRESS_SPACE_MEMORY | PCI_BASE_ADDRESS_MEM_TYPE_64 |
1416 PCI_BASE_ADDRESS_MEM_PREFETCH, &n->ctrl_mem);
1417
1418 }
1419
1420 for (i = 0; i < n->num_namespaces; i++) {
1421 NvmeNamespace *ns = &n->namespaces[i];
1422 NvmeIdNs *id_ns = &ns->id_ns;
1423 id_ns->nsfeat = 0;
1424 id_ns->nlbaf = 0;
1425 id_ns->flbas = 0;
1426 id_ns->mc = 0;
1427 id_ns->dpc = 0;
1428 id_ns->dps = 0;
1429 id_ns->lbaf[0].ds = BDRV_SECTOR_BITS;
1430 id_ns->ncap = id_ns->nuse = id_ns->nsze =
1431 cpu_to_le64(n->ns_size >>
1432 id_ns->lbaf[NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas)].ds);
1433 }
1434}
1435
1436static void nvme_exit(PCIDevice *pci_dev)
1437{
1438 NvmeCtrl *n = NVME(pci_dev);
1439
1440 nvme_clear_ctrl(n);
1441 g_free(n->namespaces);
1442 g_free(n->cq);
1443 g_free(n->sq);
1444
1445 if (n->cmb_size_mb) {
1446 g_free(n->cmbuf);
1447 }
1448 msix_uninit_exclusive_bar(pci_dev);
1449}
1450
1451static Property nvme_props[] = {
1452 DEFINE_BLOCK_PROPERTIES(NvmeCtrl, conf),
1453 DEFINE_PROP_STRING("serial", NvmeCtrl, serial),
1454 DEFINE_PROP_UINT32("cmb_size_mb", NvmeCtrl, cmb_size_mb, 0),
1455 DEFINE_PROP_UINT32("num_queues", NvmeCtrl, num_queues, 64),
1456 DEFINE_PROP_END_OF_LIST(),
1457};
1458
1459static const VMStateDescription nvme_vmstate = {
1460 .name = "nvme",
1461 .unmigratable = 1,
1462};
1463
1464static void nvme_class_init(ObjectClass *oc, void *data)
1465{
1466 DeviceClass *dc = DEVICE_CLASS(oc);
1467 PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
1468
1469 pc->realize = nvme_realize;
1470 pc->exit = nvme_exit;
1471 pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
1472 pc->vendor_id = PCI_VENDOR_ID_INTEL;
1473 pc->device_id = 0x5845;
1474 pc->revision = 2;
1475
1476 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1477 dc->desc = "Non-Volatile Memory Express";
1478 dc->props = nvme_props;
1479 dc->vmsd = &nvme_vmstate;
1480}
1481
1482static void nvme_instance_init(Object *obj)
1483{
1484 NvmeCtrl *s = NVME(obj);
1485
1486 device_add_bootindex_property(obj, &s->conf.bootindex,
1487 "bootindex", "/namespace@1,0",
1488 DEVICE(obj), &error_abort);
1489}
1490
1491static const TypeInfo nvme_info = {
1492 .name = TYPE_NVME,
1493 .parent = TYPE_PCI_DEVICE,
1494 .instance_size = sizeof(NvmeCtrl),
1495 .class_init = nvme_class_init,
1496 .instance_init = nvme_instance_init,
1497 .interfaces = (InterfaceInfo[]) {
1498 { INTERFACE_PCIE_DEVICE },
1499 { }
1500 },
1501};
1502
1503static void nvme_register_types(void)
1504{
1505 type_register_static(&nvme_info);
1506}
1507
1508type_init(nvme_register_types)
1509