Merge remote-tracking branch 'remotes/agraf/tags/signed-s390-for-upstream' into staging
[qemu/ar7.git] / hw / block / nvme.c
blobad988d7c24edc56f3d3692578ef03bf3a8417a51
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, 1.1, 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]>
23 #include <hw/block/block.h>
24 #include <hw/hw.h>
25 #include <hw/pci/msix.h>
26 #include <hw/pci/pci.h>
27 #include "sysemu/sysemu.h"
28 #include "qapi/visitor.h"
29 #include "sysemu/block-backend.h"
31 #include "nvme.h"
33 static void nvme_process_sq(void *opaque);
35 static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
37 return sqid < n->num_queues && n->sq[sqid] != NULL ? 0 : -1;
40 static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
42 return cqid < n->num_queues && n->cq[cqid] != NULL ? 0 : -1;
45 static void nvme_inc_cq_tail(NvmeCQueue *cq)
47 cq->tail++;
48 if (cq->tail >= cq->size) {
49 cq->tail = 0;
50 cq->phase = !cq->phase;
54 static void nvme_inc_sq_head(NvmeSQueue *sq)
56 sq->head = (sq->head + 1) % sq->size;
59 static uint8_t nvme_cq_full(NvmeCQueue *cq)
61 return (cq->tail + 1) % cq->size == cq->head;
64 static uint8_t nvme_sq_empty(NvmeSQueue *sq)
66 return sq->head == sq->tail;
69 static void nvme_isr_notify(NvmeCtrl *n, NvmeCQueue *cq)
71 if (cq->irq_enabled) {
72 if (msix_enabled(&(n->parent_obj))) {
73 msix_notify(&(n->parent_obj), cq->vector);
74 } else {
75 pci_irq_pulse(&n->parent_obj);
80 static uint16_t nvme_map_prp(QEMUSGList *qsg, uint64_t prp1, uint64_t prp2,
81 uint32_t len, NvmeCtrl *n)
83 hwaddr trans_len = n->page_size - (prp1 % n->page_size);
84 trans_len = MIN(len, trans_len);
85 int num_prps = (len >> n->page_bits) + 1;
87 if (!prp1) {
88 return NVME_INVALID_FIELD | NVME_DNR;
91 pci_dma_sglist_init(qsg, &n->parent_obj, num_prps);
92 qemu_sglist_add(qsg, prp1, trans_len);
93 len -= trans_len;
94 if (len) {
95 if (!prp2) {
96 goto unmap;
98 if (len > n->page_size) {
99 uint64_t prp_list[n->max_prp_ents];
100 uint32_t nents, prp_trans;
101 int i = 0;
103 nents = (len + n->page_size - 1) >> n->page_bits;
104 prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
105 pci_dma_read(&n->parent_obj, prp2, (void *)prp_list, prp_trans);
106 while (len != 0) {
107 uint64_t prp_ent = le64_to_cpu(prp_list[i]);
109 if (i == n->max_prp_ents - 1 && len > n->page_size) {
110 if (!prp_ent || prp_ent & (n->page_size - 1)) {
111 goto unmap;
114 i = 0;
115 nents = (len + n->page_size - 1) >> n->page_bits;
116 prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
117 pci_dma_read(&n->parent_obj, prp_ent, (void *)prp_list,
118 prp_trans);
119 prp_ent = le64_to_cpu(prp_list[i]);
122 if (!prp_ent || prp_ent & (n->page_size - 1)) {
123 goto unmap;
126 trans_len = MIN(len, n->page_size);
127 qemu_sglist_add(qsg, prp_ent, trans_len);
128 len -= trans_len;
129 i++;
131 } else {
132 if (prp2 & (n->page_size - 1)) {
133 goto unmap;
135 qemu_sglist_add(qsg, prp2, len);
138 return NVME_SUCCESS;
140 unmap:
141 qemu_sglist_destroy(qsg);
142 return NVME_INVALID_FIELD | NVME_DNR;
145 static uint16_t nvme_dma_read_prp(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
146 uint64_t prp1, uint64_t prp2)
148 QEMUSGList qsg;
150 if (nvme_map_prp(&qsg, prp1, prp2, len, n)) {
151 return NVME_INVALID_FIELD | NVME_DNR;
153 if (dma_buf_read(ptr, len, &qsg)) {
154 qemu_sglist_destroy(&qsg);
155 return NVME_INVALID_FIELD | NVME_DNR;
157 return NVME_SUCCESS;
160 static void nvme_post_cqes(void *opaque)
162 NvmeCQueue *cq = opaque;
163 NvmeCtrl *n = cq->ctrl;
164 NvmeRequest *req, *next;
166 QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
167 NvmeSQueue *sq;
168 hwaddr addr;
170 if (nvme_cq_full(cq)) {
171 break;
174 QTAILQ_REMOVE(&cq->req_list, req, entry);
175 sq = req->sq;
176 req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
177 req->cqe.sq_id = cpu_to_le16(sq->sqid);
178 req->cqe.sq_head = cpu_to_le16(sq->head);
179 addr = cq->dma_addr + cq->tail * n->cqe_size;
180 nvme_inc_cq_tail(cq);
181 pci_dma_write(&n->parent_obj, addr, (void *)&req->cqe,
182 sizeof(req->cqe));
183 QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
185 nvme_isr_notify(n, cq);
188 static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
190 assert(cq->cqid == req->sq->cqid);
191 QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
192 QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
193 timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
196 static void nvme_rw_cb(void *opaque, int ret)
198 NvmeRequest *req = opaque;
199 NvmeSQueue *sq = req->sq;
200 NvmeCtrl *n = sq->ctrl;
201 NvmeCQueue *cq = n->cq[sq->cqid];
203 block_acct_done(blk_get_stats(n->conf.blk), &req->acct);
204 if (!ret) {
205 req->status = NVME_SUCCESS;
206 } else {
207 req->status = NVME_INTERNAL_DEV_ERROR;
210 qemu_sglist_destroy(&req->qsg);
211 nvme_enqueue_req_completion(cq, req);
214 static uint16_t nvme_rw(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
215 NvmeRequest *req)
217 NvmeRwCmd *rw = (NvmeRwCmd *)cmd;
218 uint32_t nlb = le32_to_cpu(rw->nlb) + 1;
219 uint64_t slba = le64_to_cpu(rw->slba);
220 uint64_t prp1 = le64_to_cpu(rw->prp1);
221 uint64_t prp2 = le64_to_cpu(rw->prp2);
223 uint8_t lba_index = NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas);
224 uint8_t data_shift = ns->id_ns.lbaf[lba_index].ds;
225 uint64_t data_size = (uint64_t)nlb << data_shift;
226 uint64_t aio_slba = slba << (data_shift - BDRV_SECTOR_BITS);
227 int is_write = rw->opcode == NVME_CMD_WRITE ? 1 : 0;
229 if ((slba + nlb) > ns->id_ns.nsze) {
230 return NVME_LBA_RANGE | NVME_DNR;
232 if (nvme_map_prp(&req->qsg, prp1, prp2, data_size, n)) {
233 return NVME_INVALID_FIELD | NVME_DNR;
235 assert((nlb << data_shift) == req->qsg.size);
237 dma_acct_start(n->conf.blk, &req->acct, &req->qsg,
238 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
239 req->aiocb = is_write ?
240 dma_blk_write(n->conf.blk, &req->qsg, aio_slba, nvme_rw_cb, req) :
241 dma_blk_read(n->conf.blk, &req->qsg, aio_slba, nvme_rw_cb, req);
243 return NVME_NO_COMPLETE;
246 static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
248 NvmeNamespace *ns;
249 uint32_t nsid = le32_to_cpu(cmd->nsid);
251 if (nsid == 0 || nsid > n->num_namespaces) {
252 return NVME_INVALID_NSID | NVME_DNR;
255 ns = &n->namespaces[nsid - 1];
256 switch (cmd->opcode) {
257 case NVME_CMD_FLUSH:
258 return NVME_SUCCESS;
259 case NVME_CMD_WRITE:
260 case NVME_CMD_READ:
261 return nvme_rw(n, ns, cmd, req);
262 default:
263 return NVME_INVALID_OPCODE | NVME_DNR;
267 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
269 n->sq[sq->sqid] = NULL;
270 timer_del(sq->timer);
271 timer_free(sq->timer);
272 g_free(sq->io_req);
273 if (sq->sqid) {
274 g_free(sq);
278 static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeCmd *cmd)
280 NvmeDeleteQ *c = (NvmeDeleteQ *)cmd;
281 NvmeRequest *req, *next;
282 NvmeSQueue *sq;
283 NvmeCQueue *cq;
284 uint16_t qid = le16_to_cpu(c->qid);
286 if (!qid || nvme_check_sqid(n, qid)) {
287 return NVME_INVALID_QID | NVME_DNR;
290 sq = n->sq[qid];
291 while (!QTAILQ_EMPTY(&sq->out_req_list)) {
292 req = QTAILQ_FIRST(&sq->out_req_list);
293 assert(req->aiocb);
294 blk_aio_cancel(req->aiocb);
296 if (!nvme_check_cqid(n, sq->cqid)) {
297 cq = n->cq[sq->cqid];
298 QTAILQ_REMOVE(&cq->sq_list, sq, entry);
300 nvme_post_cqes(cq);
301 QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
302 if (req->sq == sq) {
303 QTAILQ_REMOVE(&cq->req_list, req, entry);
304 QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
309 nvme_free_sq(sq, n);
310 return NVME_SUCCESS;
313 static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
314 uint16_t sqid, uint16_t cqid, uint16_t size)
316 int i;
317 NvmeCQueue *cq;
319 sq->ctrl = n;
320 sq->dma_addr = dma_addr;
321 sq->sqid = sqid;
322 sq->size = size;
323 sq->cqid = cqid;
324 sq->head = sq->tail = 0;
325 sq->io_req = g_new(NvmeRequest, sq->size);
327 QTAILQ_INIT(&sq->req_list);
328 QTAILQ_INIT(&sq->out_req_list);
329 for (i = 0; i < sq->size; i++) {
330 sq->io_req[i].sq = sq;
331 QTAILQ_INSERT_TAIL(&(sq->req_list), &sq->io_req[i], entry);
333 sq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_process_sq, sq);
335 assert(n->cq[cqid]);
336 cq = n->cq[cqid];
337 QTAILQ_INSERT_TAIL(&(cq->sq_list), sq, entry);
338 n->sq[sqid] = sq;
341 static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeCmd *cmd)
343 NvmeSQueue *sq;
344 NvmeCreateSq *c = (NvmeCreateSq *)cmd;
346 uint16_t cqid = le16_to_cpu(c->cqid);
347 uint16_t sqid = le16_to_cpu(c->sqid);
348 uint16_t qsize = le16_to_cpu(c->qsize);
349 uint16_t qflags = le16_to_cpu(c->sq_flags);
350 uint64_t prp1 = le64_to_cpu(c->prp1);
352 if (!cqid || nvme_check_cqid(n, cqid)) {
353 return NVME_INVALID_CQID | NVME_DNR;
355 if (!sqid || (sqid && !nvme_check_sqid(n, sqid))) {
356 return NVME_INVALID_QID | NVME_DNR;
358 if (!qsize || qsize > NVME_CAP_MQES(n->bar.cap)) {
359 return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
361 if (!prp1 || prp1 & (n->page_size - 1)) {
362 return NVME_INVALID_FIELD | NVME_DNR;
364 if (!(NVME_SQ_FLAGS_PC(qflags))) {
365 return NVME_INVALID_FIELD | NVME_DNR;
367 sq = g_malloc0(sizeof(*sq));
368 nvme_init_sq(sq, n, prp1, sqid, cqid, qsize + 1);
369 return NVME_SUCCESS;
372 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
374 n->cq[cq->cqid] = NULL;
375 timer_del(cq->timer);
376 timer_free(cq->timer);
377 msix_vector_unuse(&n->parent_obj, cq->vector);
378 if (cq->cqid) {
379 g_free(cq);
383 static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeCmd *cmd)
385 NvmeDeleteQ *c = (NvmeDeleteQ *)cmd;
386 NvmeCQueue *cq;
387 uint16_t qid = le16_to_cpu(c->qid);
389 if (!qid || nvme_check_cqid(n, qid)) {
390 return NVME_INVALID_CQID | NVME_DNR;
393 cq = n->cq[qid];
394 if (!QTAILQ_EMPTY(&cq->sq_list)) {
395 return NVME_INVALID_QUEUE_DEL;
397 nvme_free_cq(cq, n);
398 return NVME_SUCCESS;
401 static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
402 uint16_t cqid, uint16_t vector, uint16_t size, uint16_t irq_enabled)
404 cq->ctrl = n;
405 cq->cqid = cqid;
406 cq->size = size;
407 cq->dma_addr = dma_addr;
408 cq->phase = 1;
409 cq->irq_enabled = irq_enabled;
410 cq->vector = vector;
411 cq->head = cq->tail = 0;
412 QTAILQ_INIT(&cq->req_list);
413 QTAILQ_INIT(&cq->sq_list);
414 msix_vector_use(&n->parent_obj, cq->vector);
415 n->cq[cqid] = cq;
416 cq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_post_cqes, cq);
419 static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeCmd *cmd)
421 NvmeCQueue *cq;
422 NvmeCreateCq *c = (NvmeCreateCq *)cmd;
423 uint16_t cqid = le16_to_cpu(c->cqid);
424 uint16_t vector = le16_to_cpu(c->irq_vector);
425 uint16_t qsize = le16_to_cpu(c->qsize);
426 uint16_t qflags = le16_to_cpu(c->cq_flags);
427 uint64_t prp1 = le64_to_cpu(c->prp1);
429 if (!cqid || (cqid && !nvme_check_cqid(n, cqid))) {
430 return NVME_INVALID_CQID | NVME_DNR;
432 if (!qsize || qsize > NVME_CAP_MQES(n->bar.cap)) {
433 return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
435 if (!prp1) {
436 return NVME_INVALID_FIELD | NVME_DNR;
438 if (vector > n->num_queues) {
439 return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
441 if (!(NVME_CQ_FLAGS_PC(qflags))) {
442 return NVME_INVALID_FIELD | NVME_DNR;
445 cq = g_malloc0(sizeof(*cq));
446 nvme_init_cq(cq, n, prp1, cqid, vector, qsize + 1,
447 NVME_CQ_FLAGS_IEN(qflags));
448 return NVME_SUCCESS;
451 static uint16_t nvme_identify(NvmeCtrl *n, NvmeCmd *cmd)
453 NvmeNamespace *ns;
454 NvmeIdentify *c = (NvmeIdentify *)cmd;
455 uint32_t cns = le32_to_cpu(c->cns);
456 uint32_t nsid = le32_to_cpu(c->nsid);
457 uint64_t prp1 = le64_to_cpu(c->prp1);
458 uint64_t prp2 = le64_to_cpu(c->prp2);
460 if (cns) {
461 return nvme_dma_read_prp(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl),
462 prp1, prp2);
464 if (nsid == 0 || nsid > n->num_namespaces) {
465 return NVME_INVALID_NSID | NVME_DNR;
468 ns = &n->namespaces[nsid - 1];
469 return nvme_dma_read_prp(n, (uint8_t *)&ns->id_ns, sizeof(ns->id_ns),
470 prp1, prp2);
473 static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
475 uint32_t dw10 = le32_to_cpu(cmd->cdw10);
477 switch (dw10) {
478 case NVME_NUMBER_OF_QUEUES:
479 req->cqe.result =
480 cpu_to_le32((n->num_queues - 1) | ((n->num_queues - 1) << 16));
481 break;
482 default:
483 return NVME_INVALID_FIELD | NVME_DNR;
485 return NVME_SUCCESS;
488 static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
490 uint32_t dw10 = le32_to_cpu(cmd->cdw10);
492 switch (dw10) {
493 case NVME_NUMBER_OF_QUEUES:
494 req->cqe.result =
495 cpu_to_le32((n->num_queues - 1) | ((n->num_queues - 1) << 16));
496 break;
497 default:
498 return NVME_INVALID_FIELD | NVME_DNR;
500 return NVME_SUCCESS;
503 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
505 switch (cmd->opcode) {
506 case NVME_ADM_CMD_DELETE_SQ:
507 return nvme_del_sq(n, cmd);
508 case NVME_ADM_CMD_CREATE_SQ:
509 return nvme_create_sq(n, cmd);
510 case NVME_ADM_CMD_DELETE_CQ:
511 return nvme_del_cq(n, cmd);
512 case NVME_ADM_CMD_CREATE_CQ:
513 return nvme_create_cq(n, cmd);
514 case NVME_ADM_CMD_IDENTIFY:
515 return nvme_identify(n, cmd);
516 case NVME_ADM_CMD_SET_FEATURES:
517 return nvme_set_feature(n, cmd, req);
518 case NVME_ADM_CMD_GET_FEATURES:
519 return nvme_get_feature(n, cmd, req);
520 default:
521 return NVME_INVALID_OPCODE | NVME_DNR;
525 static void nvme_process_sq(void *opaque)
527 NvmeSQueue *sq = opaque;
528 NvmeCtrl *n = sq->ctrl;
529 NvmeCQueue *cq = n->cq[sq->cqid];
531 uint16_t status;
532 hwaddr addr;
533 NvmeCmd cmd;
534 NvmeRequest *req;
536 while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
537 addr = sq->dma_addr + sq->head * n->sqe_size;
538 pci_dma_read(&n->parent_obj, addr, (void *)&cmd, sizeof(cmd));
539 nvme_inc_sq_head(sq);
541 req = QTAILQ_FIRST(&sq->req_list);
542 QTAILQ_REMOVE(&sq->req_list, req, entry);
543 QTAILQ_INSERT_TAIL(&sq->out_req_list, req, entry);
544 memset(&req->cqe, 0, sizeof(req->cqe));
545 req->cqe.cid = cmd.cid;
547 status = sq->sqid ? nvme_io_cmd(n, &cmd, req) :
548 nvme_admin_cmd(n, &cmd, req);
549 if (status != NVME_NO_COMPLETE) {
550 req->status = status;
551 nvme_enqueue_req_completion(cq, req);
556 static void nvme_clear_ctrl(NvmeCtrl *n)
558 int i;
560 for (i = 0; i < n->num_queues; i++) {
561 if (n->sq[i] != NULL) {
562 nvme_free_sq(n->sq[i], n);
565 for (i = 0; i < n->num_queues; i++) {
566 if (n->cq[i] != NULL) {
567 nvme_free_cq(n->cq[i], n);
571 blk_flush(n->conf.blk);
572 n->bar.cc = 0;
575 static int nvme_start_ctrl(NvmeCtrl *n)
577 uint32_t page_bits = NVME_CC_MPS(n->bar.cc) + 12;
578 uint32_t page_size = 1 << page_bits;
580 if (n->cq[0] || n->sq[0] || !n->bar.asq || !n->bar.acq ||
581 n->bar.asq & (page_size - 1) || n->bar.acq & (page_size - 1) ||
582 NVME_CC_MPS(n->bar.cc) < NVME_CAP_MPSMIN(n->bar.cap) ||
583 NVME_CC_MPS(n->bar.cc) > NVME_CAP_MPSMAX(n->bar.cap) ||
584 NVME_CC_IOCQES(n->bar.cc) < NVME_CTRL_CQES_MIN(n->id_ctrl.cqes) ||
585 NVME_CC_IOCQES(n->bar.cc) > NVME_CTRL_CQES_MAX(n->id_ctrl.cqes) ||
586 NVME_CC_IOSQES(n->bar.cc) < NVME_CTRL_SQES_MIN(n->id_ctrl.sqes) ||
587 NVME_CC_IOSQES(n->bar.cc) > NVME_CTRL_SQES_MAX(n->id_ctrl.sqes) ||
588 !NVME_AQA_ASQS(n->bar.aqa) || !NVME_AQA_ACQS(n->bar.aqa)) {
589 return -1;
592 n->page_bits = page_bits;
593 n->page_size = page_size;
594 n->max_prp_ents = n->page_size / sizeof(uint64_t);
595 n->cqe_size = 1 << NVME_CC_IOCQES(n->bar.cc);
596 n->sqe_size = 1 << NVME_CC_IOSQES(n->bar.cc);
597 nvme_init_cq(&n->admin_cq, n, n->bar.acq, 0, 0,
598 NVME_AQA_ACQS(n->bar.aqa) + 1, 1);
599 nvme_init_sq(&n->admin_sq, n, n->bar.asq, 0, 0,
600 NVME_AQA_ASQS(n->bar.aqa) + 1);
602 return 0;
605 static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
606 unsigned size)
608 switch (offset) {
609 case 0xc:
610 n->bar.intms |= data & 0xffffffff;
611 n->bar.intmc = n->bar.intms;
612 break;
613 case 0x10:
614 n->bar.intms &= ~(data & 0xffffffff);
615 n->bar.intmc = n->bar.intms;
616 break;
617 case 0x14:
618 /* Windows first sends data, then sends enable bit */
619 if (!NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc) &&
620 !NVME_CC_SHN(data) && !NVME_CC_SHN(n->bar.cc))
622 n->bar.cc = data;
625 if (NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc)) {
626 n->bar.cc = data;
627 if (nvme_start_ctrl(n)) {
628 n->bar.csts = NVME_CSTS_FAILED;
629 } else {
630 n->bar.csts = NVME_CSTS_READY;
632 } else if (!NVME_CC_EN(data) && NVME_CC_EN(n->bar.cc)) {
633 nvme_clear_ctrl(n);
634 n->bar.csts &= ~NVME_CSTS_READY;
636 if (NVME_CC_SHN(data) && !(NVME_CC_SHN(n->bar.cc))) {
637 nvme_clear_ctrl(n);
638 n->bar.cc = data;
639 n->bar.csts |= NVME_CSTS_SHST_COMPLETE;
640 } else if (!NVME_CC_SHN(data) && NVME_CC_SHN(n->bar.cc)) {
641 n->bar.csts &= ~NVME_CSTS_SHST_COMPLETE;
642 n->bar.cc = data;
644 break;
645 case 0x24:
646 n->bar.aqa = data & 0xffffffff;
647 break;
648 case 0x28:
649 n->bar.asq = data;
650 break;
651 case 0x2c:
652 n->bar.asq |= data << 32;
653 break;
654 case 0x30:
655 n->bar.acq = data;
656 break;
657 case 0x34:
658 n->bar.acq |= data << 32;
659 break;
660 default:
661 break;
665 static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
667 NvmeCtrl *n = (NvmeCtrl *)opaque;
668 uint8_t *ptr = (uint8_t *)&n->bar;
669 uint64_t val = 0;
671 if (addr < sizeof(n->bar)) {
672 memcpy(&val, ptr + addr, size);
674 return val;
677 static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
679 uint32_t qid;
681 if (addr & ((1 << 2) - 1)) {
682 return;
685 if (((addr - 0x1000) >> 2) & 1) {
686 uint16_t new_head = val & 0xffff;
687 int start_sqs;
688 NvmeCQueue *cq;
690 qid = (addr - (0x1000 + (1 << 2))) >> 3;
691 if (nvme_check_cqid(n, qid)) {
692 return;
695 cq = n->cq[qid];
696 if (new_head >= cq->size) {
697 return;
700 start_sqs = nvme_cq_full(cq) ? 1 : 0;
701 cq->head = new_head;
702 if (start_sqs) {
703 NvmeSQueue *sq;
704 QTAILQ_FOREACH(sq, &cq->sq_list, entry) {
705 timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
707 timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
710 if (cq->tail != cq->head) {
711 nvme_isr_notify(n, cq);
713 } else {
714 uint16_t new_tail = val & 0xffff;
715 NvmeSQueue *sq;
717 qid = (addr - 0x1000) >> 3;
718 if (nvme_check_sqid(n, qid)) {
719 return;
722 sq = n->sq[qid];
723 if (new_tail >= sq->size) {
724 return;
727 sq->tail = new_tail;
728 timer_mod(sq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
732 static void nvme_mmio_write(void *opaque, hwaddr addr, uint64_t data,
733 unsigned size)
735 NvmeCtrl *n = (NvmeCtrl *)opaque;
736 if (addr < sizeof(n->bar)) {
737 nvme_write_bar(n, addr, data, size);
738 } else if (addr >= 0x1000) {
739 nvme_process_db(n, addr, data);
743 static const MemoryRegionOps nvme_mmio_ops = {
744 .read = nvme_mmio_read,
745 .write = nvme_mmio_write,
746 .endianness = DEVICE_LITTLE_ENDIAN,
747 .impl = {
748 .min_access_size = 2,
749 .max_access_size = 8,
753 static int nvme_init(PCIDevice *pci_dev)
755 NvmeCtrl *n = NVME(pci_dev);
756 NvmeIdCtrl *id = &n->id_ctrl;
758 int i;
759 int64_t bs_size;
760 uint8_t *pci_conf;
762 if (!n->conf.blk) {
763 return -1;
766 bs_size = blk_getlength(n->conf.blk);
767 if (bs_size < 0) {
768 return -1;
771 blkconf_serial(&n->conf, &n->serial);
772 if (!n->serial) {
773 return -1;
775 blkconf_blocksizes(&n->conf);
777 pci_conf = pci_dev->config;
778 pci_conf[PCI_INTERRUPT_PIN] = 1;
779 pci_config_set_prog_interface(pci_dev->config, 0x2);
780 pci_config_set_class(pci_dev->config, PCI_CLASS_STORAGE_EXPRESS);
781 pcie_endpoint_cap_init(&n->parent_obj, 0x80);
783 n->num_namespaces = 1;
784 n->num_queues = 64;
785 n->reg_size = 1 << qemu_fls(0x1004 + 2 * (n->num_queues + 1) * 4);
786 n->ns_size = bs_size / (uint64_t)n->num_namespaces;
788 n->namespaces = g_new0(NvmeNamespace, n->num_namespaces);
789 n->sq = g_new0(NvmeSQueue *, n->num_queues);
790 n->cq = g_new0(NvmeCQueue *, n->num_queues);
792 memory_region_init_io(&n->iomem, OBJECT(n), &nvme_mmio_ops, n,
793 "nvme", n->reg_size);
794 pci_register_bar(&n->parent_obj, 0,
795 PCI_BASE_ADDRESS_SPACE_MEMORY | PCI_BASE_ADDRESS_MEM_TYPE_64,
796 &n->iomem);
797 msix_init_exclusive_bar(&n->parent_obj, n->num_queues, 4);
799 id->vid = cpu_to_le16(pci_get_word(pci_conf + PCI_VENDOR_ID));
800 id->ssvid = cpu_to_le16(pci_get_word(pci_conf + PCI_SUBSYSTEM_VENDOR_ID));
801 strpadcpy((char *)id->mn, sizeof(id->mn), "QEMU NVMe Ctrl", ' ');
802 strpadcpy((char *)id->fr, sizeof(id->fr), "1.0", ' ');
803 strpadcpy((char *)id->sn, sizeof(id->sn), n->serial, ' ');
804 id->rab = 6;
805 id->ieee[0] = 0x00;
806 id->ieee[1] = 0x02;
807 id->ieee[2] = 0xb3;
808 id->oacs = cpu_to_le16(0);
809 id->frmw = 7 << 1;
810 id->lpa = 1 << 0;
811 id->sqes = (0x6 << 4) | 0x6;
812 id->cqes = (0x4 << 4) | 0x4;
813 id->nn = cpu_to_le32(n->num_namespaces);
814 id->psd[0].mp = cpu_to_le16(0x9c4);
815 id->psd[0].enlat = cpu_to_le32(0x10);
816 id->psd[0].exlat = cpu_to_le32(0x4);
818 n->bar.cap = 0;
819 NVME_CAP_SET_MQES(n->bar.cap, 0x7ff);
820 NVME_CAP_SET_CQR(n->bar.cap, 1);
821 NVME_CAP_SET_AMS(n->bar.cap, 1);
822 NVME_CAP_SET_TO(n->bar.cap, 0xf);
823 NVME_CAP_SET_CSS(n->bar.cap, 1);
824 NVME_CAP_SET_MPSMAX(n->bar.cap, 4);
826 n->bar.vs = 0x00010100;
827 n->bar.intmc = n->bar.intms = 0;
829 for (i = 0; i < n->num_namespaces; i++) {
830 NvmeNamespace *ns = &n->namespaces[i];
831 NvmeIdNs *id_ns = &ns->id_ns;
832 id_ns->nsfeat = 0;
833 id_ns->nlbaf = 0;
834 id_ns->flbas = 0;
835 id_ns->mc = 0;
836 id_ns->dpc = 0;
837 id_ns->dps = 0;
838 id_ns->lbaf[0].ds = BDRV_SECTOR_BITS;
839 id_ns->ncap = id_ns->nuse = id_ns->nsze =
840 cpu_to_le64(n->ns_size >>
841 id_ns->lbaf[NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas)].ds);
843 return 0;
846 static void nvme_exit(PCIDevice *pci_dev)
848 NvmeCtrl *n = NVME(pci_dev);
850 nvme_clear_ctrl(n);
851 g_free(n->namespaces);
852 g_free(n->cq);
853 g_free(n->sq);
854 msix_uninit_exclusive_bar(pci_dev);
857 static Property nvme_props[] = {
858 DEFINE_BLOCK_PROPERTIES(NvmeCtrl, conf),
859 DEFINE_PROP_STRING("serial", NvmeCtrl, serial),
860 DEFINE_PROP_END_OF_LIST(),
863 static const VMStateDescription nvme_vmstate = {
864 .name = "nvme",
865 .unmigratable = 1,
868 static void nvme_class_init(ObjectClass *oc, void *data)
870 DeviceClass *dc = DEVICE_CLASS(oc);
871 PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
873 pc->init = nvme_init;
874 pc->exit = nvme_exit;
875 pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
876 pc->vendor_id = PCI_VENDOR_ID_INTEL;
877 pc->device_id = 0x5845;
878 pc->revision = 1;
879 pc->is_express = 1;
881 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
882 dc->desc = "Non-Volatile Memory Express";
883 dc->props = nvme_props;
884 dc->vmsd = &nvme_vmstate;
887 static void nvme_get_bootindex(Object *obj, Visitor *v, void *opaque,
888 const char *name, Error **errp)
890 NvmeCtrl *s = NVME(obj);
892 visit_type_int32(v, &s->conf.bootindex, name, errp);
895 static void nvme_set_bootindex(Object *obj, Visitor *v, void *opaque,
896 const char *name, Error **errp)
898 NvmeCtrl *s = NVME(obj);
899 int32_t boot_index;
900 Error *local_err = NULL;
902 visit_type_int32(v, &boot_index, name, &local_err);
903 if (local_err) {
904 goto out;
906 /* check whether bootindex is present in fw_boot_order list */
907 check_boot_index(boot_index, &local_err);
908 if (local_err) {
909 goto out;
911 /* change bootindex to a new one */
912 s->conf.bootindex = boot_index;
914 out:
915 if (local_err) {
916 error_propagate(errp, local_err);
920 static void nvme_instance_init(Object *obj)
922 object_property_add(obj, "bootindex", "int32",
923 nvme_get_bootindex,
924 nvme_set_bootindex, NULL, NULL, NULL);
925 object_property_set_int(obj, -1, "bootindex", NULL);
928 static const TypeInfo nvme_info = {
929 .name = "nvme",
930 .parent = TYPE_PCI_DEVICE,
931 .instance_size = sizeof(NvmeCtrl),
932 .class_init = nvme_class_init,
933 .instance_init = nvme_instance_init,
936 static void nvme_register_types(void)
938 type_register_static(&nvme_info);
941 type_init(nvme_register_types)