diff options
author | Klaus Jensen | 2020-11-06 10:46:01 +0100 |
---|---|---|
committer | Klaus Jensen | 2021-03-09 11:00:57 +0100 |
commit | e4e430b3d6baa1c908ba71ca37aad87edac98804 (patch) | |
tree | 726602000b64b5c5511a45955e71b9892e7d78fe /hw/block/nvme.c | |
parent | nvme: updated shared header for copy command (diff) | |
download | qemu-e4e430b3d6baa1c908ba71ca37aad87edac98804.tar.gz qemu-e4e430b3d6baa1c908ba71ca37aad87edac98804.tar.xz qemu-e4e430b3d6baa1c908ba71ca37aad87edac98804.zip |
hw/block/nvme: add simple copy command
Add support for TP 4065a ("Simple Copy Command"), v2020.05.04
("Ratified").
The implementation uses a bounce buffer to first read in the source
logical blocks, then issue a write of that bounce buffer. The default
maximum number of source logical blocks is 128, translating to 512 KiB
for 4k logical blocks which aligns with the default value of MDTS.
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Diffstat (limited to 'hw/block/nvme.c')
-rw-r--r-- | hw/block/nvme.c | 252 |
1 files changed, 251 insertions, 1 deletions
diff --git a/hw/block/nvme.c b/hw/block/nvme.c index 44129f8e8b..ab4723ff31 100644 --- a/hw/block/nvme.c +++ b/hw/block/nvme.c @@ -195,6 +195,7 @@ static const uint32_t nvme_cse_iocs_nvm[256] = { [NVME_CMD_WRITE] = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC, [NVME_CMD_READ] = NVME_CMD_EFF_CSUPP, [NVME_CMD_DSM] = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC, + [NVME_CMD_COPY] = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC, [NVME_CMD_COMPARE] = NVME_CMD_EFF_CSUPP, }; @@ -204,6 +205,7 @@ static const uint32_t nvme_cse_iocs_zoned[256] = { [NVME_CMD_WRITE] = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC, [NVME_CMD_READ] = NVME_CMD_EFF_CSUPP, [NVME_CMD_DSM] = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC, + [NVME_CMD_COPY] = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC, [NVME_CMD_COMPARE] = NVME_CMD_EFF_CSUPP, [NVME_CMD_ZONE_APPEND] = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC, [NVME_CMD_ZONE_MGMT_SEND] = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC, @@ -1532,6 +1534,136 @@ static void nvme_aio_zone_reset_cb(void *opaque, int ret) nvme_enqueue_req_completion(nvme_cq(req), req); } +struct nvme_copy_ctx { + int copies; + uint8_t *bounce; + uint32_t nlb; +}; + +struct nvme_copy_in_ctx { + NvmeRequest *req; + QEMUIOVector iov; +}; + +static void nvme_copy_cb(void *opaque, int ret) +{ + NvmeRequest *req = opaque; + NvmeNamespace *ns = req->ns; + struct nvme_copy_ctx *ctx = req->opaque; + + trace_pci_nvme_copy_cb(nvme_cid(req)); + + if (ns->params.zoned) { + NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd; + uint64_t sdlba = le64_to_cpu(copy->sdlba); + NvmeZone *zone = nvme_get_zone_by_slba(ns, sdlba); + + __nvme_advance_zone_wp(ns, zone, ctx->nlb); + } + + if (!ret) { + block_acct_done(blk_get_stats(ns->blkconf.blk), &req->acct); + } else { + block_acct_failed(blk_get_stats(ns->blkconf.blk), &req->acct); + nvme_aio_err(req, ret); + } + + g_free(ctx->bounce); + g_free(ctx); + + nvme_enqueue_req_completion(nvme_cq(req), req); +} + +static void nvme_copy_in_complete(NvmeRequest *req) +{ + NvmeNamespace *ns = req->ns; + NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd; + struct nvme_copy_ctx *ctx = req->opaque; + uint64_t sdlba = le64_to_cpu(copy->sdlba); + uint16_t status; + + trace_pci_nvme_copy_in_complete(nvme_cid(req)); + + block_acct_done(blk_get_stats(ns->blkconf.blk), &req->acct); + + status = nvme_check_bounds(ns, sdlba, ctx->nlb); + if (status) { + trace_pci_nvme_err_invalid_lba_range(sdlba, ctx->nlb, ns->id_ns.nsze); + goto invalid; + } + + if (ns->params.zoned) { + NvmeZone *zone = nvme_get_zone_by_slba(ns, sdlba); + + status = nvme_check_zone_write(ns, zone, sdlba, ctx->nlb); + if (status) { + goto invalid; + } + + status = nvme_zrm_auto(ns, zone); + if (status) { + goto invalid; + } + + zone->w_ptr += ctx->nlb; + } + + qemu_iovec_init(&req->iov, 1); + qemu_iovec_add(&req->iov, ctx->bounce, nvme_l2b(ns, ctx->nlb)); + + block_acct_start(blk_get_stats(ns->blkconf.blk), &req->acct, 0, + BLOCK_ACCT_WRITE); + + req->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, sdlba), + &req->iov, 0, nvme_copy_cb, req); + + return; + +invalid: + req->status = status; + + g_free(ctx->bounce); + g_free(ctx); + + nvme_enqueue_req_completion(nvme_cq(req), req); +} + +static void nvme_aio_copy_in_cb(void *opaque, int ret) +{ + struct nvme_copy_in_ctx *in_ctx = opaque; + NvmeRequest *req = in_ctx->req; + NvmeNamespace *ns = req->ns; + struct nvme_copy_ctx *ctx = req->opaque; + + qemu_iovec_destroy(&in_ctx->iov); + g_free(in_ctx); + + trace_pci_nvme_aio_copy_in_cb(nvme_cid(req)); + + if (ret) { + nvme_aio_err(req, ret); + } + + ctx->copies--; + + if (ctx->copies) { + return; + } + + if (req->status) { + block_acct_failed(blk_get_stats(ns->blkconf.blk), &req->acct); + + g_free(ctx->bounce); + g_free(ctx); + + nvme_enqueue_req_completion(nvme_cq(req), req); + + return; + } + + nvme_copy_in_complete(req); +} + struct nvme_compare_ctx { QEMUIOVector iov; uint8_t *bounce; @@ -1650,6 +1782,121 @@ static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req) return status; } +static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req) +{ + NvmeNamespace *ns = req->ns; + NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd; + g_autofree NvmeCopySourceRange *range = NULL; + + uint16_t nr = copy->nr + 1; + uint8_t format = copy->control[0] & 0xf; + uint32_t nlb = 0; + + uint8_t *bounce = NULL, *bouncep = NULL; + struct nvme_copy_ctx *ctx; + uint16_t status; + int i; + + trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format); + + if (!(n->id_ctrl.ocfs & (1 << format))) { + trace_pci_nvme_err_copy_invalid_format(format); + return NVME_INVALID_FIELD | NVME_DNR; + } + + if (nr > ns->id_ns.msrc + 1) { + return NVME_CMD_SIZE_LIMIT | NVME_DNR; + } + + range = g_new(NvmeCopySourceRange, nr); + + status = nvme_dma(n, (uint8_t *)range, nr * sizeof(NvmeCopySourceRange), + DMA_DIRECTION_TO_DEVICE, req); + if (status) { + return status; + } + + for (i = 0; i < nr; i++) { + uint64_t slba = le64_to_cpu(range[i].slba); + uint32_t _nlb = le16_to_cpu(range[i].nlb) + 1; + + if (_nlb > le16_to_cpu(ns->id_ns.mssrl)) { + return NVME_CMD_SIZE_LIMIT | NVME_DNR; + } + + status = nvme_check_bounds(ns, slba, _nlb); + if (status) { + trace_pci_nvme_err_invalid_lba_range(slba, _nlb, ns->id_ns.nsze); + return status; + } + + if (NVME_ERR_REC_DULBE(ns->features.err_rec)) { + status = nvme_check_dulbe(ns, slba, _nlb); + if (status) { + return status; + } + } + + if (ns->params.zoned) { + status = nvme_check_zone_read(ns, slba, _nlb); + if (status) { + return status; + } + } + + nlb += _nlb; + } + + if (nlb > le32_to_cpu(ns->id_ns.mcl)) { + return NVME_CMD_SIZE_LIMIT | NVME_DNR; + } + + bounce = bouncep = g_malloc(nvme_l2b(ns, nlb)); + + block_acct_start(blk_get_stats(ns->blkconf.blk), &req->acct, 0, + BLOCK_ACCT_READ); + + ctx = g_new(struct nvme_copy_ctx, 1); + + ctx->bounce = bounce; + ctx->nlb = nlb; + ctx->copies = 1; + + req->opaque = ctx; + + for (i = 0; i < nr; i++) { + uint64_t slba = le64_to_cpu(range[i].slba); + uint32_t nlb = le16_to_cpu(range[i].nlb) + 1; + + size_t len = nvme_l2b(ns, nlb); + int64_t offset = nvme_l2b(ns, slba); + + trace_pci_nvme_copy_source_range(slba, nlb); + + struct nvme_copy_in_ctx *in_ctx = g_new(struct nvme_copy_in_ctx, 1); + in_ctx->req = req; + + qemu_iovec_init(&in_ctx->iov, 1); + qemu_iovec_add(&in_ctx->iov, bouncep, len); + + ctx->copies++; + + blk_aio_preadv(ns->blkconf.blk, offset, &in_ctx->iov, 0, + nvme_aio_copy_in_cb, in_ctx); + + bouncep += len; + } + + /* account for the 1-initialization */ + ctx->copies--; + + if (!ctx->copies) { + nvme_copy_in_complete(req); + } + + return NVME_NO_COMPLETE; +} + static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req) { NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd; @@ -2387,6 +2634,8 @@ static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req) return nvme_compare(n, req); case NVME_CMD_DSM: return nvme_dsm(n, req); + case NVME_CMD_COPY: + return nvme_copy(n, req); case NVME_CMD_ZONE_MGMT_SEND: return nvme_zone_mgmt_send(n, req); case NVME_CMD_ZONE_MGMT_RECV: @@ -4484,9 +4733,10 @@ static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev) id->nn = cpu_to_le32(n->num_namespaces); id->oncs = cpu_to_le16(NVME_ONCS_WRITE_ZEROES | NVME_ONCS_TIMESTAMP | NVME_ONCS_FEATURES | NVME_ONCS_DSM | - NVME_ONCS_COMPARE); + NVME_ONCS_COMPARE | NVME_ONCS_COPY); id->vwc = (0x2 << 1) | 0x1; + id->ocfs = cpu_to_le16(NVME_OCFS_COPY_FORMAT_0); id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN | NVME_CTRL_SGLS_BITBUCKET); |