From: Klaus Jensen <k.jen...@samsung.com> Add support for the Dataset Management command and the Deallocate attribute. Deallocation results in discards being sent to the underlying block device. Whether of not the blocks are actually deallocated is affected by the same factors as Write Zeroes (see previous commit).
format | discard | dsm (512B) dsm (4KiB) dsm (64KiB) -------------------------------------------------------- qcow2 ignore n n n qcow2 unmap n n y raw ignore n n n raw unmap n y y Again, a raw format and 4KiB LBAs are preferable. In order to set the Namespace Preferred Deallocate Granularity and Alignment fields (NPDG and NPDA), choose a sane minimum discard granularity of 4KiB. If we are using a passthru device supporting discard at a 512B granularity, user should set the discard_granularity property explicitly. NPDG and NPDA will also account for the cluster_size of the block driver if required (i.e. for QCOW2). See NVM Express 1.3d, Section 6.7 ("Dataset Management command"). Signed-off-by: Klaus Jensen <k.jen...@samsung.com> Reviewed-by: Keith Busch <kbu...@kernel.org> --- hw/block/nvme.h | 2 + hw/block/nvme-ns.c | 30 ++++++++++++-- hw/block/nvme.c | 98 +++++++++++++++++++++++++++++++++++++++++++++- 3 files changed, 125 insertions(+), 5 deletions(-) diff --git a/hw/block/nvme.h b/hw/block/nvme.h index e080a2318a50..574333caa3f9 100644 --- a/hw/block/nvme.h +++ b/hw/block/nvme.h @@ -28,6 +28,7 @@ typedef struct NvmeRequest { struct NvmeNamespace *ns; BlockAIOCB *aiocb; uint16_t status; + void *opaque; NvmeCqe cqe; NvmeCmd cmd; BlockAcctCookie acct; @@ -60,6 +61,7 @@ static inline const char *nvme_io_opc_str(uint8_t opc) case NVME_CMD_WRITE: return "NVME_NVM_CMD_WRITE"; case NVME_CMD_READ: return "NVME_NVM_CMD_READ"; case NVME_CMD_WRITE_ZEROES: return "NVME_NVM_CMD_WRITE_ZEROES"; + case NVME_CMD_DSM: return "NVME_NVM_CMD_DSM"; default: return "NVME_NVM_CMD_UNKNOWN"; } } diff --git a/hw/block/nvme-ns.c b/hw/block/nvme-ns.c index 53ded460348e..37f95951a6b8 100644 --- a/hw/block/nvme-ns.c +++ b/hw/block/nvme-ns.c @@ -28,10 +28,14 @@ #include "nvme.h" #include "nvme-ns.h" -static void nvme_ns_init(NvmeNamespace *ns) +#define MIN_DISCARD_GRANULARITY (4 * KiB) + +static int nvme_ns_init(NvmeNamespace *ns, Error **errp) { + BlockDriverInfo bdi; NvmeIdNs *id_ns = &ns->id_ns; int lba_index = NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas); + int npdg; ns->id_ns.dlfeat = 0x9; @@ -43,8 +47,19 @@ static void nvme_ns_init(NvmeNamespace *ns) id_ns->ncap = id_ns->nsze; id_ns->nuse = id_ns->ncap; - /* support DULBE */ - id_ns->nsfeat |= 0x4; + /* support DULBE and I/O optimization fields */ + id_ns->nsfeat |= (0x4 | 0x10); + + npdg = ns->blkconf.discard_granularity / ns->blkconf.logical_block_size; + + if (bdrv_get_info(blk_bs(ns->blkconf.blk), &bdi) >= 0 && + bdi.cluster_size > ns->blkconf.discard_granularity) { + npdg = bdi.cluster_size / ns->blkconf.logical_block_size; + } + + id_ns->npda = id_ns->npdg = npdg - 1; + + return 0; } static int nvme_ns_init_blk(NvmeCtrl *n, NvmeNamespace *ns, Error **errp) @@ -60,6 +75,11 @@ static int nvme_ns_init_blk(NvmeCtrl *n, NvmeNamespace *ns, Error **errp) return -1; } + if (ns->blkconf.discard_granularity == -1) { + ns->blkconf.discard_granularity = + MAX(ns->blkconf.logical_block_size, MIN_DISCARD_GRANULARITY); + } + ns->size = blk_getlength(ns->blkconf.blk); if (ns->size < 0) { error_setg_errno(errp, -ns->size, "could not get blockdev size"); @@ -93,7 +113,9 @@ int nvme_ns_setup(NvmeCtrl *n, NvmeNamespace *ns, Error **errp) return -1; } - nvme_ns_init(ns); + if (nvme_ns_init(ns, errp)) { + return -1; + } if (nvme_register_namespace(n, ns, errp)) { return -1; diff --git a/hw/block/nvme.c b/hw/block/nvme.c index 6e6bdb338ad7..f019d43788ac 100644 --- a/hw/block/nvme.c +++ b/hw/block/nvme.c @@ -242,6 +242,7 @@ static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq) static void nvme_req_clear(NvmeRequest *req) { req->ns = NULL; + req->opaque = NULL; memset(&req->cqe, 0x0, sizeof(req->cqe)); req->status = NVME_SUCCESS; } @@ -978,6 +979,99 @@ static void nvme_rw_cb(void *opaque, int ret) nvme_enqueue_req_completion(nvme_cq(req), req); } +static void nvme_aio_discard_cb(void *opaque, int ret) +{ + NvmeRequest *req = opaque; + uintptr_t *discards = (uintptr_t *)&req->opaque; + + trace_pci_nvme_aio_discard_cb(nvme_cid(req)); + + if (ret) { + nvme_aio_err(req, ret); + } + + (*discards)--; + + if (*discards) { + return; + } + + nvme_enqueue_req_completion(nvme_cq(req), req); +} + +static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req) +{ + NvmeNamespace *ns = req->ns; + NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd; + + uint32_t attr = le32_to_cpu(dsm->attributes); + uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1; + + uint16_t status = NVME_SUCCESS; + + trace_pci_nvme_dsm(nvme_cid(req), nvme_nsid(ns), nr, attr); + + if (attr & NVME_DSMGMT_AD) { + int64_t offset; + size_t len; + NvmeDsmRange range[nr]; + uintptr_t *discards = (uintptr_t *)&req->opaque; + + status = nvme_dma(n, (uint8_t *)range, sizeof(range), + DMA_DIRECTION_TO_DEVICE, req); + if (status) { + return status; + } + + /* + * AIO callbacks may be called immediately, so initialize discards to 1 + * to make sure the the callback does not complete the request before + * all discards have been issued. + */ + *discards = 1; + + for (int i = 0; i < nr; i++) { + uint64_t slba = le64_to_cpu(range[i].slba); + uint32_t nlb = le32_to_cpu(range[i].nlb); + + if (nvme_check_bounds(ns, slba, nlb)) { + trace_pci_nvme_err_invalid_lba_range(slba, nlb, + ns->id_ns.nsze); + continue; + } + + trace_pci_nvme_dsm_deallocate(nvme_cid(req), nvme_nsid(ns), slba, + nlb); + + offset = nvme_l2b(ns, slba); + len = nvme_l2b(ns, nlb); + + while (len) { + size_t bytes = MIN(BDRV_REQUEST_MAX_BYTES, len); + + (*discards)++; + + blk_aio_pdiscard(ns->blkconf.blk, offset, bytes, + nvme_aio_discard_cb, req); + + offset += bytes; + len -= bytes; + } + } + + /* account for the 1-initialization */ + (*discards)--; + + if (*discards) { + status = NVME_NO_COMPLETE; + } else { + status = req->status; + } + } + + return status; +} + static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req) { block_acct_start(blk_get_stats(req->ns->blkconf.blk), &req->acct, 0, @@ -1107,6 +1201,8 @@ static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req) case NVME_CMD_WRITE: case NVME_CMD_READ: return nvme_rw(n, req); + case NVME_CMD_DSM: + return nvme_dsm(n, req); default: trace_pci_nvme_err_invalid_opc(req->cmd.opcode); return NVME_INVALID_OPCODE | NVME_DNR; @@ -2829,7 +2925,7 @@ static void nvme_init_ctrl(NvmeCtrl *n, PCIDevice *pci_dev) id->cqes = (0x4 << 4) | 0x4; 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_FEATURES | NVME_ONCS_DSM); id->vwc = 0x1; id->sgls = cpu_to_le32(NVME_CTRL_SGLS_SUPPORT_NO_ALIGN | -- 2.30.0