nvme-rdma/nvmet-rdma: Add metadata/T10-PI support
Subsystems (3)
v2 cover nvme-rdma/nvmet-rdma: Add metadata/T10-PI support
Hello Sagi, Christoph, Keith, Martin and Co
This patchset adds metadata (T10-PI) support for NVMeoF/RDMA host side and target side, using signature verbs API. This set starts with a few preparation commits to the NVMe host core layer. It continues with NVMeoF/RDMA host implementation + BIP flag patch (block) + few preparation commits to the NVMe target core layer. The patchset ends with NVMeoF/RDMA target implementation.
Configuration:
Host:
- nvme connect --pi_enable --transport=rdma --traddr=10.0.1.1 --nqn=test-nvme
Target:
- echo 1 > /config/nvmet/subsystems/${NAME}/attr_pi_enable
- echo 1 > /config/nvmet/ports/${PORT_NUM}/param_pi_enable
The code was tested using Mellanox's ConnectX-4/ConnectX-5 HCAs. This series applies on top of nvme_5.5 branch cleanly.
Changes from v1:
- Added Reviewed-by signatures
- Added namespace features flag (Patch 01/16)
- Remove nvme_ns_has_pi function (Patch 01/16)
- Added has_pi field to struct nvme_request (Patch 01/16)
- Subject change for patch 02/16
- Fix comment for PCI metadata (Patch 03/16)
- Rebase over "nvme: Avoid preallocating big SGL for data" patchset
- Introduce NVME_INLINE_PROT_SG_CNT flag (Patch 05/16)
- Introduce nvme_rdma_sgl structure (Patch 06/16)
- Remove first_sgl pointer from struct nvme_rdma_request (Patch 06/16)
- Split nvme-rdma patches (Patches 06/16, 07/16)
- Rename is_protected to use_pi (Patch 07/16)
- Refactor nvme_rdma_get_max_fr_pages function (Patch 07/16)
- Added ifdef CONFIG_BLK_DEV_INTEGRITY (Patches 07/16, 10/16, 14/16, 15/16, 16/16)
- Use existing BIP_MAPPED_INTEGRITY flag (Patches 08/16, 15/16)
- Added port configfs pi_enable (Patch 14/16)
Israel Rukshin (13):
nvme: Introduce namespace features flag
nvme-fabrics: Allow user enabling metadata/T10-PI support
nvme: Introduce NVME_INLINE_PROT_SG_CNT
nvme-rdma: Introduce nvme_rdma_sgl structure
block: Add a BIP_MAPPED_INTEGRITY check to complete_fn
nvmet: Prepare metadata request
nvmet: Add metadata characteristics for a namespace
nvmet: Rename nvmet_rw_len to nvmet_rw_data_len
nvmet: Rename nvmet_check_data_len to nvmet_check_transfer_len
nvme: Add Metadata Capabilities enumerations
nvmet: Add metadata/T10-PI support
nvmet: Add metadata support for block devices
nvmet-rdma: Add metadata/T10-PI support
Max Gurtovoy (3):
nvme: Enforce extended LBA format for fabrics metadata
nvme: Introduce max_integrity_segments ctrl attribute
nvme-rdma: Add metadata/T10-PI support
block/t10-pi.c | 4 + drivers/nvme/host/core.c | 75 +++++--- drivers/nvme/host/fabrics.c | 11 ++ drivers/nvme/host/fabrics.h | 3 + drivers/nvme/host/nvme.h | 9 +- drivers/nvme/host/pci.c | 7 + drivers/nvme/host/rdma.c | 368 +++++++++++++++++++++++++++++++++----- drivers/nvme/target/admin-cmd.c | 37 ++-- drivers/nvme/target/configfs.c | 61 +++++++ drivers/nvme/target/core.c | 54 ++++-- drivers/nvme/target/discovery.c | 8 +- drivers/nvme/target/fabrics-cmd.c | 19 +- drivers/nvme/target/io-cmd-bdev.c | 115 +++++++++++- drivers/nvme/target/io-cmd-file.c | 8 +- drivers/nvme/target/nvmet.h | 40 ++++- drivers/nvme/target/rdma.c | 235 ++++++++++++++++++++++-- include/linux/nvme.h | 2 + 17 files changed, 929 insertions(+), 127 deletions(-)
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 1/16 nvme: Introduce namespace features flag
From: Israel Rukshin <[email protected]>
This patch is a preparation for adding PI support for fabrics drivers that will use an extended LBA format for metadata.
--- drivers/nvme/host/core.c | 40 +++++++++++++++++++++++++++------------- drivers/nvme/host/nvme.h | 6 +++++- 2 files changed, 32 insertions(+), 14 deletions(-)
drivers/nvme/host/core.c +27 -13
@@ -209,11 +209,6 @@ static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)return ret; } -static inline bool nvme_ns_has_pi (struct nvme_ns *ns) -{ -return ns->pi_type && ns->ms ==sizeof (struct t10_pi_tuple ); -} -static blk_status_t nvme_error_status (u16 status) {switch (status &0x7ff ) {
@@ -473,6 +468,7 @@ static inline void nvme_clear_nvme_request(struct request *req)if (!(req->rq_flags &RQF_DONTPREP )) {nvme_req (req)->retries =0 ;nvme_req (req)->flags =0 ; +nvme_req (req)->has_pi = false; req->rq_flags |=RQF_DONTPREP ; } }
@@ -706,6 +702,7 @@ static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,nvme_assign_write_stream (ctrl, req, &control, &dsmgmt);if (ns->ms) { +nvme_req (req)->has_pi = ns->features &NVME_NS_DIF_SUPPORTED ;/* * If formated with metadata, the block layer always provides a * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
@@ -713,7 +710,7 @@ static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,* namespace capacity to zero to prevent any I/O. */ if (!blk_integrity_rq (req)) { -if (WARN_ON_ONCE (!nvme_ns_has_pi (ns))) +if (WARN_ON_ONCE (!nvme_req (req)->has_pi))return BLK_STS_NOTSUPP ; control |=NVME_RW_PRINFO_PRACT ; }
@@ -1271,7 +1268,7 @@ static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) meta_len = (io.nblocks +1 ) * ns->ms; metadata = (void __user *)(uintptr_t )io.metadata; -if (ns->ext) { +if (ns->features &NVME_NS_EXT_LBAS ) { length += meta_len; meta_len =0 ; }else if (meta_len) {
@@ -1799,10 +1796,10 @@ static void nvme_update_disk_info(struct gendisk *disk,blk_queue_io_min (disk->queue, phys_bs);blk_queue_io_opt (disk->queue, io_opt); -if (ns->ms && !ns->ext && - (ns->ctrl->ops->flags &NVME_F_METADATA_SUPPORTED )) +if (ns->features &NVME_NS_DIX_SUPPORTED )nvme_init_integrity (disk, ns->ms, ns->pi_type); -if ((ns->ms && !nvme_ns_has_pi (ns) && !blk_get_integrity (disk)) || +if ((ns->ms && !(ns->features &NVME_NS_DIF_SUPPORTED ) && + !blk_get_integrity (disk)) || ns->lba_shift >PAGE_SHIFT ) capacity =0 ;
@@ -1832,12 +1829,29 @@ static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id) ns->lba_shift =9 ; ns->noiob =le16_to_cpu (id->noiob); ns->ms =le16_to_cpu (id->lbaf[id->flbas &NVME_NS_FLBAS_LBA_MASK ].ms); - ns->ext = ns->ms && (id->flbas &NVME_NS_FLBAS_META_EXT ); + ns->features =0 ;/* the PI implementation requires metadata equal t10 pi tuple size */ -if (ns->ms ==sizeof (struct t10_pi_tuple )) +if (ns->ms ==sizeof (struct t10_pi_tuple )) { ns->pi_type = id->dps &NVME_NS_DPS_PI_MASK ; -else +if (ns->pi_type) + ns->features |=NVME_NS_DIF_SUPPORTED ; + }else { ns->pi_type =0 ; + } + +if (ns->ms) { +if (id->flbas &NVME_NS_FLBAS_META_EXT ) + ns->features |=NVME_NS_EXT_LBAS ; + +/* +* For PCI, Extended logical block will be generated by the +* controller. +*/ +if (ns->ctrl->ops->flags &NVME_F_METADATA_SUPPORTED ) { +if (!(ns->features &NVME_NS_EXT_LBAS )) + ns->features |=NVME_NS_DIX_SUPPORTED ; + } + }if (ns->noiob)nvme_set_chunk_size (ns);
drivers/nvme/host/nvme.h +5 -1
@@ -139,6 +139,7 @@ struct nvme_request {u8 flags;u16 status;struct nvme_ctrl *ctrl; +bool has_pi; };/*
@@ -381,8 +382,11 @@ struct nvme_ns {u16 ms;u16 sgs;u32 sws; -bool ext;u8 pi_type; +unsigned long features; +#define NVME_NS_EXT_LBAS (1 << 0) +#define NVME_NS_DIX_SUPPORTED (1 << 1) +#define NVME_NS_DIF_SUPPORTED (1 << 2)unsigned long flags;#define NVME_NS_REMOVING 0#define NVME_NS_DEAD 1
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
Christoph Hellwig
maintainerRe: [PATCH 01/16] nvme: Introduce namespace features flag
Quoted 3 lines · Max Gurtovoy
From: Israel Rukshin <[email protected]>
This patch is a preparation for adding PI support for fabrics drivers
No need to ever say "This patch..". Better explain why it is added.
Also the patch adds a field to the nvme_req structure as well.
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 2/16 nvme: Enforce extended LBA format for fabrics metadata
In reply to Max GurtovoyAn extended LBA is a larger LBA that is created when metadata associated with the LBA is transferred contiguously with the LBA data (AKA interleaved). The metadata may be either transferred as part of the LBA (creating an extended LBA) or it may be transferred as a separate contiguous buffer of data. According to the NVMeoF spec, a fabrics ctrl supports only an Extended LBA format. Fail revalidation in case we have a spec violation. Also initialize the integrity profile for the block device for fabrics ctrl.
--- drivers/nvme/host/core.c | 25 +++++++++++++++++++++---- 1 file changed, 21 insertions(+), 4 deletions(-)
drivers/nvme/host/core.c +21 -4
@@ -1816,7 +1816,7 @@ static void nvme_update_disk_info(struct gendisk *disk,blk_mq_unfreeze_queue (disk->queue); } -static void __nvme_revalidate_disk (struct gendisk *disk,struct nvme_id_ns *id) +static int __nvme_revalidate_disk (struct gendisk *disk,struct nvme_id_ns *id) {struct nvme_ns *ns = disk->private_data;
@@ -1843,12 +1843,22 @@ static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)if (id->flbas &NVME_NS_FLBAS_META_EXT ) ns->features |=NVME_NS_EXT_LBAS ; +/* +* For Fabrics, only metadata as part of extended data LBA is +* supported. Fail in case of a spec violation. +*/ +if (ns->ctrl->ops->flags &NVME_F_FABRICS ) { +if (WARN_ON_ONCE (!(ns->features &NVME_NS_EXT_LBAS ))) +return -EINVAL ; + } +/* * For PCI, Extended logical block will be generated by the * controller. */ if (ns->ctrl->ops->flags &NVME_F_METADATA_SUPPORTED ) { -if (!(ns->features &NVME_NS_EXT_LBAS )) +if (ns->ctrl->ops->flags &NVME_F_FABRICS || + !(ns->features &NVME_NS_EXT_LBAS )) ns->features |=NVME_NS_DIX_SUPPORTED ; } }
@@ -1863,6 +1873,7 @@ static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id) revalidate_disk(ns->head->disk); }#endif +return 0 ; }static int nvme_revalidate_disk (struct gendisk *disk)
@@ -1887,7 +1898,10 @@ static int nvme_revalidate_disk(struct gendisk *disk) goto free_id; } -__nvme_revalidate_disk (disk, id); + ret =__nvme_revalidate_disk (disk, id); +if (ret) + goto free_id; + ret =nvme_report_ns_ids (ctrl, ns->head->ns_id, id, &ids);if (ret) goto free_id;
@@ -3563,7 +3577,8 @@ static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)memcpy (disk->disk_name, disk_name,DISK_NAME_LEN ); ns->disk = disk; -__nvme_revalidate_disk (disk, id); +if (__nvme_revalidate_disk (disk, id)) + goto out_free_disk;if ((ctrl->quirks &NVME_QUIRK_LIGHTNVM ) && id->vs[0 ] ==0x1 ) { ret =nvme_nvm_register (ns, disk_name, node);
@@ -3588,6 +3603,8 @@ static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)return 0 ; out_put_disk:put_disk (ns->disk); + out_free_disk: +del_gendisk (ns->disk); out_unlink_ns:mutex_lock (&ctrl->subsys->lock);list_del_rcu (&ns->siblings);
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 3/16 nvme: Introduce max_integrity_segments ctrl attribute
In reply to Max GurtovoyThis patch doesn't change any logic, and is needed as a preparation for adding PI support for fabrics drivers that will use an extended LBA format for metadata.
--- drivers/nvme/host/core.c | 11 +++++++---- drivers/nvme/host/nvme.h | 1 + drivers/nvme/host/pci.c | 7 +++++++ 3 files changed, 15 insertions(+), 4 deletions(-)
drivers/nvme/host/core.c +7 -4
@@ -1619,7 +1619,8 @@ static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo) }#ifdef CONFIG_BLK_DEV_INTEGRITY -static void nvme_init_integrity (struct gendisk *disk,u16 ms,u8 pi_type) +static void nvme_init_integrity (struct gendisk *disk,u16 ms,u8 pi_type, +u32 max_integrity_segments) {struct blk_integrity integrity;
@@ -1642,10 +1643,11 @@ static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type) } integrity.tuple_size = ms;blk_integrity_register (disk, &integrity); -blk_queue_max_integrity_segments (disk->queue,1 ); +blk_queue_max_integrity_segments (disk->queue, max_integrity_segments); }#else -static void nvme_init_integrity (struct gendisk *disk,u16 ms,u8 pi_type) +static void nvme_init_integrity (struct gendisk *disk,u16 ms,u8 pi_type, +u32 max_integrity_segments) { }#endif /* CONFIG_BLK_DEV_INTEGRITY */
@@ -1797,7 +1799,8 @@ static void nvme_update_disk_info(struct gendisk *disk,blk_queue_io_opt (disk->queue, io_opt);if (ns->features &NVME_NS_DIX_SUPPORTED ) -nvme_init_integrity (disk, ns->ms, ns->pi_type); +nvme_init_integrity (disk, ns->ms, ns->pi_type, + ns->ctrl->max_integrity_segments);if ((ns->ms && !(ns->features &NVME_NS_DIF_SUPPORTED ) && !blk_get_integrity (disk)) || ns->lba_shift >PAGE_SHIFT )
drivers/nvme/host/nvme.h +1 -0
@@ -229,6 +229,7 @@ struct nvme_ctrl {u32 page_size;u32 max_hw_sectors;u32 max_segments; +u32 max_integrity_segments;u16 crdt[3 ];u16 oncs;u16 oacs;
drivers/nvme/host/pci.c +7 -0
@@ -2588,6 +2588,13 @@ static void nvme_reset_work(struct work_struct *work) goto out; } +/* +* NVMe PCI driver doesn't support Extended LBA format and supports +* only a single integrity segment for a separate contiguous buffer +* of metadata. +*/ + dev->ctrl.max_integrity_segments =1 ; + result =nvme_init_identify (&dev->ctrl);if (result) goto out;
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 4/16 nvme-fabrics: Allow user enabling metadata/T10-PI support
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
Preparation for adding metadata (T10-PI) over fabric support. This will allow end-to-end protection information passthrough and validation for NVMe over Fabric.
--- drivers/nvme/host/core.c | 3 ++- drivers/nvme/host/fabrics.c | 11 +++++++++++ drivers/nvme/host/fabrics.h | 3 +++ 3 files changed, 16 insertions(+), 1 deletion(-)
drivers/nvme/host/core.c +2 -1
@@ -1860,7 +1860,8 @@ static int __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)* controller. */ if (ns->ctrl->ops->flags &NVME_F_METADATA_SUPPORTED ) { -if (ns->ctrl->ops->flags &NVME_F_FABRICS || +if ((ns->ctrl->ops->flags &NVME_F_FABRICS && + ns->ctrl->opts->pi_enable) || !(ns->features &NVME_NS_EXT_LBAS )) ns->features |=NVME_NS_DIX_SUPPORTED ; }
drivers/nvme/host/fabrics.c +11 -0
@@ -612,6 +612,7 @@ static const match_table_t opt_tokens = { { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" }, { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" }, { NVMF_OPT_TOS, "tos=%d" }, + { NVMF_OPT_PI_ENABLE, "pi_enable" }, { NVMF_OPT_ERR, NULL } };
@@ -634,6 +635,7 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts, opts->hdr_digest = false; opts->data_digest = false; opts->tos =-1 ;/* < 0 == use transport default */ + opts->pi_enable = false; options = o =kstrdup (buf,GFP_KERNEL );if (!options)
@@ -867,6 +869,15 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts, } opts->tos = token;break ; +#ifdef CONFIG_BLK_DEV_INTEGRITY +case NVMF_OPT_PI_ENABLE : +if (opts->discovery_nqn) { +pr_debug ("Ignoring pi_enable value for discovery controller\n" ); +break ; + } + opts->pi_enable = true; +break ; +#endif default :pr_warn ("unknown parameter or missing value '%s' in ctrl creation request\n" , p);
drivers/nvme/host/fabrics.h +3 -0
@@ -56,6 +56,7 @@ enum {NVMF_OPT_NR_WRITE_QUEUES =1 <<17 ,NVMF_OPT_NR_POLL_QUEUES =1 <<18 ,NVMF_OPT_TOS =1 <<19 , + NVMF_OPT_PI_ENABLE = 1 << 20, };/**
@@ -89,6 +90,7 @@ enum {* @nr_write_queues: number of queues for write I/O * @nr_poll_queues: number of queues for polling I/O * @tos: type of service +* @pi_enable: Enable metadata (T10-PI) support */ struct nvmf_ctrl_options {unsigned mask;
@@ -111,6 +113,7 @@ struct nvmf_ctrl_options {unsigned int nr_write_queues;unsigned int nr_poll_queues;int tos; +bool pi_enable; };/*
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 5/16 nvme: Introduce NVME_INLINE_PROT_SG_CNT
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
SGL size of PI metadata is usually small. Thus, 1 inline sg should cover most cases. The macro will be used for pre-allocate a single SGL entry for protection information. The preallocation of small inline SGLs depends on SG_CHAIN capability so if the ARCH doesn't support SG_CHAIN, use the runtime allocation for the SGL. This patch is a preparation for adding metadata (T10-PI) over fabric support.
--- drivers/nvme/host/nvme.h | 2 ++ 1 file changed, 2 insertions(+)
drivers/nvme/host/nvme.h +2 -0
@@ -30,8 +30,10 @@ extern unsigned int admin_timeout;#ifdef CONFIG_ARCH_NO_SG_CHAIN #define NVME_INLINE_SG_CNT 0 +#define NVME_INLINE_PROT_SG_CNT 0#else #define NVME_INLINE_SG_CNT 2 +#define NVME_INLINE_PROT_SG_CNT 1#endif extern struct workqueue_struct *nvme_wq;
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 6/16 nvme-rdma: Introduce nvme_rdma_sgl structure
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
Remove first_sgl pointer from struct nvme_rdma_request and use pointer arithmetic instead. The inline scatterlist, if exists, will be located right after the nvme_rdma_request. This patch is needed as a preparation for adding PI support.
--- drivers/nvme/host/rdma.c | 41 ++++++++++++++++++++++++----------------- 1 file changed, 24 insertions(+), 17 deletions(-)
drivers/nvme/host/rdma.c +24 -17
@@ -48,6 +48,11 @@ struct nvme_rdma_qe {u64 dma; }; +struct nvme_rdma_sgl { +int nents; +struct sg_table sg_table; +}; +struct nvme_rdma_queue ;struct nvme_rdma_request {struct nvme_request req;
@@ -58,12 +63,10 @@ struct nvme_rdma_request {refcount_t ref;struct ib_sge sge[1 +NVME_RDMA_MAX_INLINE_SEGMENTS ];u32 num_sge; -int nents;struct ib_reg_wr reg_wr;struct ib_cqe reg_cqe;struct nvme_rdma_queue *queue; -struct sg_table sg_table; -struct scatterlist first_sgl[]; +struct nvme_rdma_sgl data_sgl; };enum nvme_rdma_queue_flags {
@@ -1159,8 +1162,9 @@ static void nvme_rdma_unmap_data(struct nvme_rdma_queue *queue, req->mr =NULL ; } -ib_dma_unmap_sg (ibdev, req->sg_table.sgl, req->nents,rq_dma_dir (rq)); -sg_free_table_chained (&req->sg_table,NVME_INLINE_SG_CNT ); +ib_dma_unmap_sg (ibdev, req->data_sgl.sg_table.sgl, req->data_sgl.nents, +rq_dma_dir (rq)); +sg_free_table_chained (&req->data_sgl.sg_table,NVME_INLINE_SG_CNT ); }static int nvme_rdma_set_sg_null (struct nvme_command *c)
@@ -1179,7 +1183,7 @@ static int nvme_rdma_map_sg_inline(struct nvme_rdma_queue *queue, int count) {struct nvme_sgl_desc *sg = &c->common.dptr.sgl; -struct scatterlist *sgl = req->sg_table.sgl; +struct scatterlist *sgl = req->data_sgl.sg_table.sgl;struct ib_sge *sge = &req->sge[1 ];u32 len =0 ;int i;
@@ -1204,8 +1208,8 @@ static int nvme_rdma_map_sg_single(struct nvme_rdma_queue *queue, {struct nvme_keyed_sgl_desc *sg = &c->common.dptr.ksgl; - sg->addr =cpu_to_le64 (sg_dma_address (req->sg_table.sgl)); -put_unaligned_le24 (sg_dma_len (req->sg_table.sgl), sg->length); + sg->addr =cpu_to_le64 (sg_dma_address (req->data_sgl.sg_table.sgl)); +put_unaligned_le24 (sg_dma_len (req->data_sgl.sg_table.sgl), sg->length);put_unaligned_le32 (queue->device->pd->unsafe_global_rkey, sg->key); sg->type =NVME_KEY_SGL_FMT_DATA_DESC <<4 ;return 0 ;
@@ -1226,7 +1230,8 @@ static int nvme_rdma_map_sg_fr(struct nvme_rdma_queue *queue,* Align the MR to a 4K page size to match the ctrl page size and * the block virtual boundary. */ - nr =ib_map_mr_sg (req->mr, req->sg_table.sgl, count,NULL ,SZ_4K ); + nr =ib_map_mr_sg (req->mr, req->data_sgl.sg_table.sgl, count,NULL , +SZ_4K );if (unlikely (nr < count)) {ib_mr_pool_put (queue->qp, &queue->qp->rdma_mrs, req->mr); req->mr =NULL ;
@@ -1273,17 +1278,18 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue,if (!blk_rq_nr_phys_segments (rq))return nvme_rdma_set_sg_null (c); - req->sg_table.sgl = req->first_sgl; - ret =sg_alloc_table_chained (&req->sg_table, -blk_rq_nr_phys_segments (rq), req->sg_table.sgl, + req->data_sgl.sg_table.sgl = (struct scatterlist *)(req +1 ); + ret =sg_alloc_table_chained (&req->data_sgl.sg_table, +blk_rq_nr_phys_segments (rq), req->data_sgl.sg_table.sgl,NVME_INLINE_SG_CNT );if (ret)return -ENOMEM ; - req->nents =blk_rq_map_sg (rq->q, rq, req->sg_table.sgl); + req->data_sgl.nents =blk_rq_map_sg (rq->q, rq, + req->data_sgl.sg_table.sgl); - count =ib_dma_map_sg (ibdev, req->sg_table.sgl, req->nents, -rq_dma_dir (rq)); + count =ib_dma_map_sg (ibdev, req->data_sgl.sg_table.sgl, + req->data_sgl.nents,rq_dma_dir (rq));if (unlikely (count <=0 )) { ret = -EIO ; goto out_free_table;
@@ -1312,9 +1318,10 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue,return 0 ; out_unmap_sg: -ib_dma_unmap_sg (ibdev, req->sg_table.sgl, req->nents,rq_dma_dir (rq)); +ib_dma_unmap_sg (ibdev, req->data_sgl.sg_table.sgl, req->data_sgl.nents, +rq_dma_dir (rq)); out_free_table: -sg_free_table_chained (&req->sg_table,NVME_INLINE_SG_CNT ); +sg_free_table_chained (&req->data_sgl.sg_table,NVME_INLINE_SG_CNT );return ret; }
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 7/16 nvme-rdma: Add metadata/T10-PI support
In reply to Max GurtovoyFor capable HCAs (e.g. ConnectX-4/ConnectX-5) this will allow end-to-end protection information passthrough and validation for NVMe over RDMA transport. Metadata offload support was implemented over the new RDMA signature verbs API and it is enabled per controller by using nvme-cli.
usage example:
nvme connect --pi_enable --transport=rdma --traddr=10.0.1.1 --nqn=test-nvme
--- drivers/nvme/host/rdma.c | 331 ++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 298 insertions(+), 33 deletions(-)
drivers/nvme/host/rdma.c +298 -33
@@ -67,6 +67,9 @@ struct nvme_rdma_request {struct ib_cqe reg_cqe;struct nvme_rdma_queue *queue;struct nvme_rdma_sgl data_sgl; +/* Metadata (T10-PI) support */ +struct nvme_rdma_sgl *pi_sgl; +bool use_pi; };enum nvme_rdma_queue_flags {
@@ -88,6 +91,7 @@ struct nvme_rdma_queue {struct rdma_cm_id *cm_id;int cm_error;struct completion cm_done; +bool pi_support; };struct nvme_rdma_ctrl {
@@ -115,6 +119,7 @@ struct nvme_rdma_ctrl {struct nvme_ctrl ctrl;bool use_inline_data;u32 io_queues[HCTX_MAX_TYPES ]; +bool pi_support; };static inline struct nvme_rdma_ctrl *to_rdma_ctrl (struct nvme_ctrl *ctrl)
@@ -272,6 +277,8 @@ static int nvme_rdma_create_qp(struct nvme_rdma_queue *queue, const int factor) init_attr.qp_type =IB_QPT_RC ; init_attr.send_cq = queue->ib_cq; init_attr.recv_cq = queue->ib_cq; +if (queue->pi_support) + init_attr.create_flags |=IB_QP_CREATE_INTEGRITY_EN ; ret =rdma_create_qp (queue->cm_id, dev->pd, &init_attr);
@@ -287,6 +294,18 @@ static void nvme_rdma_exit_request(struct blk_mq_tag_set *set,kfree (req->sqe.data); } +static unsigned int nvme_rdma_cmd_size (bool has_pi) +{ +/* +* nvme rdma command consists of struct nvme_rdma_request, data SGL, +* PI nvme_rdma_sgl struct and then PI SGL. +*/ +return sizeof (struct nvme_rdma_request ) + +sizeof (struct scatterlist ) *NVME_INLINE_SG_CNT + + has_pi * (sizeof (struct nvme_rdma_sgl ) + +sizeof (struct scatterlist ) *NVME_INLINE_PROT_SG_CNT ); +} +static int nvme_rdma_init_request (struct blk_mq_tag_set *set,struct request *rq,unsigned int hctx_idx,unsigned int numa_node)
@@ -301,6 +320,10 @@ static int nvme_rdma_init_request(struct blk_mq_tag_set *set,if (!req->sqe.data)return -ENOMEM ; +/* PI nvme_rdma_sgl struct is located after command's data SGL */ +if (queue->pi_support) + req->pi_sgl = (void *)nvme_req (rq) +nvme_rdma_cmd_size (false); + req->queue = queue;return 0 ;
@@ -411,6 +434,8 @@ static void nvme_rdma_destroy_queue_ib(struct nvme_rdma_queue *queue) dev = queue->device; ibdev = dev->dev; +if (queue->pi_support) +ib_mr_pool_destroy (queue->qp, &queue->qp->sig_mrs);ib_mr_pool_destroy (queue->qp, &queue->qp->rdma_mrs);/*
@@ -427,10 +452,13 @@ static void nvme_rdma_destroy_queue_ib(struct nvme_rdma_queue *queue)nvme_rdma_dev_put (dev); } -static int nvme_rdma_get_max_fr_pages (struct ib_device *ibdev) +static int nvme_rdma_get_max_fr_pages (struct ib_device *ibdev,bool pi_support) { -return min_t (u32,NVME_RDMA_MAX_SEGMENTS , - ibdev->attrs.max_fast_reg_page_list_len -1 ); +u32 max_page_list_len = + pi_support ? ibdev->attrs.max_pi_fast_reg_page_list_len : + ibdev->attrs.max_fast_reg_page_list_len; + +return min_t (u32,NVME_RDMA_MAX_SEGMENTS , max_page_list_len -1 ); }static int nvme_rdma_create_queue_ib (struct nvme_rdma_queue *queue)
@@ -487,7 +515,7 @@ static int nvme_rdma_create_queue_ib(struct nvme_rdma_queue *queue)* misaligned we'll end up using two entries for a single data page, * so one additional entry is required. */ - pages_per_mr =nvme_rdma_get_max_fr_pages (ibdev) +1 ; + pages_per_mr =nvme_rdma_get_max_fr_pages (ibdev, queue->pi_support) +1 ; ret = ib_mr_pool_init(queue->qp, &queue->qp->rdma_mrs, queue->queue_size, IB_MR_TYPE_MEM_REG,
@@ -499,10 +527,24 @@ static int nvme_rdma_create_queue_ib(struct nvme_rdma_queue *queue) goto out_destroy_ring; } +if (queue->pi_support) { + ret =ib_mr_pool_init (queue->qp, &queue->qp->sig_mrs, + queue->queue_size,IB_MR_TYPE_INTEGRITY , + pages_per_mr, pages_per_mr); +if (ret) { +dev_err (queue->ctrl->ctrl.device, +"failed to initialize PI MR pool sized %d for QID %d\n" , + queue->queue_size, idx); + goto out_destroy_mr_pool; + } + } +set_bit (NVME_RDMA_Q_TR_READY , &queue->flags);return 0 ; +out_destroy_mr_pool: +ib_mr_pool_destroy (queue->qp, &queue->qp->rdma_mrs); out_destroy_ring:nvme_rdma_free_ring (ibdev, queue->rsp_ring, queue->queue_size,sizeof (struct nvme_completion ),DMA_FROM_DEVICE );
@@ -524,6 +566,7 @@ static int nvme_rdma_alloc_queue(struct nvme_rdma_ctrl *ctrl, queue = &ctrl->queues[idx]; queue->ctrl = ctrl; + queue->pi_support = idx && ctrl->pi_support;init_completion (&queue->cm_done);if (idx >0 )
@@ -733,8 +776,7 @@ static struct blk_mq_tag_set *nvme_rdma_alloc_tagset(struct nvme_ctrl *nctrl, set->queue_depth =NVME_AQ_MQ_TAG_DEPTH ; set->reserved_tags =2 ;/* connect + keep-alive */ set->numa_node = nctrl->numa_node; - set->cmd_size =sizeof (struct nvme_rdma_request ) + -NVME_INLINE_SG_CNT *sizeof (struct scatterlist ); + set->cmd_size =nvme_rdma_cmd_size (false); set->driver_data = ctrl; set->nr_hw_queues =1 ; set->timeout =ADMIN_TIMEOUT ;
@@ -747,8 +789,7 @@ static struct blk_mq_tag_set *nvme_rdma_alloc_tagset(struct nvme_ctrl *nctrl, set->reserved_tags =1 ;/* fabric connect */ set->numa_node = nctrl->numa_node; set->flags =BLK_MQ_F_SHOULD_MERGE ; - set->cmd_size =sizeof (struct nvme_rdma_request ) + -NVME_INLINE_SG_CNT *sizeof (struct scatterlist ); + set->cmd_size =nvme_rdma_cmd_size (ctrl->pi_support); set->driver_data = ctrl; set->nr_hw_queues = nctrl->queue_count -1 ; set->timeout =NVME_IO_TIMEOUT ;
@@ -790,7 +831,21 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl, ctrl->device = ctrl->queues[0 ].device; ctrl->ctrl.numa_node =dev_to_node (ctrl->device->dev->dma_device); - ctrl->max_fr_pages =nvme_rdma_get_max_fr_pages (ctrl->device->dev); +/* T10-PI support */ +if (ctrl->ctrl.opts->pi_enable) { +if (!(ctrl->device->dev->attrs.device_cap_flags & +IB_DEVICE_INTEGRITY_HANDOVER )) { +dev_warn (ctrl->ctrl.device, +"T10-PI requested but not supported on %s, continue without T10-PI\n" , + ctrl->device->dev->name); + ctrl->ctrl.opts->pi_enable = false; + }else { + ctrl->pi_support = true; + } + } + + ctrl->max_fr_pages =nvme_rdma_get_max_fr_pages (ctrl->device->dev, + ctrl->pi_support);/* * Bind the async event SQE DMA mapping to the admin queue lifetime.
@@ -832,6 +887,8 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl, ctrl->ctrl.max_segments = ctrl->max_fr_pages; ctrl->ctrl.max_hw_sectors = ctrl->max_fr_pages << (ilog2 (SZ_4K ) -9 ); +if (ctrl->pi_support) + ctrl->ctrl.max_integrity_segments = ctrl->max_fr_pages;blk_mq_unquiesce_queue (ctrl->ctrl.admin_q);
@@ -1157,8 +1214,19 @@ static void nvme_rdma_unmap_data(struct nvme_rdma_queue *queue,if (!blk_rq_nr_phys_segments (rq))return ; +if (blk_integrity_rq (rq)) { +ib_dma_unmap_sg (ibdev, req->pi_sgl->sg_table.sgl, + req->pi_sgl->nents,rq_dma_dir (rq)); +sg_free_table_chained (&req->pi_sgl->sg_table, +NVME_INLINE_PROT_SG_CNT ); + } +if (req->mr) { -ib_mr_pool_put (queue->qp, &queue->qp->rdma_mrs, req->mr); +if (req->use_pi) +ib_mr_pool_put (queue->qp, &queue->qp->sig_mrs, req->mr); +else +ib_mr_pool_put (queue->qp, &queue->qp->rdma_mrs, + req->mr); req->mr =NULL ; }
@@ -1215,34 +1283,115 @@ static int nvme_rdma_map_sg_single(struct nvme_rdma_queue *queue,return 0 ; } -static int nvme_rdma_map_sg_fr (struct nvme_rdma_queue *queue, -struct nvme_rdma_request *req,struct nvme_command *c, -int count) +#ifdef CONFIG_BLK_DEV_INTEGRITY +static void nvme_rdma_set_diff_domain (struct nvme_command *cmd,struct bio *bio, +struct ib_sig_domain *domain,struct request *rq) { -struct nvme_keyed_sgl_desc *sg = &c->common.dptr.ksgl; -int nr; +struct blk_integrity *bi =blk_get_integrity (bio->bi_disk); +struct nvme_ns *ns = rq->rq_disk->private_data; + +WARN_ON (bi ==NULL ); - req->mr =ib_mr_pool_get (queue->qp, &queue->qp->rdma_mrs); -if (WARN_ON_ONCE (!req->mr)) -return -EAGAIN ; + domain->sig_type =IB_SIG_TYPE_T10_DIF ; + domain->sig.dif.bg_type =IB_T10DIF_CRC ; + domain->sig.dif.pi_interval =1 << bi->interval_exp; + domain->sig.dif.ref_tag =le32_to_cpu (cmd->rw.reftag);/* -* Align the MR to a 4K page size to match the ctrl page size and -* the block virtual boundary. +* At the moment we hard code those, but in the future +* we will take them from cmd. */ - nr =ib_map_mr_sg (req->mr, req->data_sgl.sg_table.sgl, count,NULL , -SZ_4K ); -if (unlikely (nr < count)) { -ib_mr_pool_put (queue->qp, &queue->qp->rdma_mrs, req->mr); - req->mr =NULL ; -if (nr <0 ) -return nr; -return -EINVAL ; + domain->sig.dif.apptag_check_mask =0xffff ; + domain->sig.dif.app_escape = true; + domain->sig.dif.ref_escape = true; +if (ns->pi_type !=NVME_NS_DPS_PI_TYPE3 ) + domain->sig.dif.ref_remap = true; +} + +static void nvme_rdma_set_sig_attrs (struct bio *bio, +struct ib_sig_attrs *sig_attrs,struct nvme_command *c, +struct request *rq) +{ +u16 control =le16_to_cpu (c->rw.control); + +memset (sig_attrs,0 ,sizeof (*sig_attrs)); + +if (control &NVME_RW_PRINFO_PRACT ) { +/* for WRITE_INSERT/READ_STRIP no memory domain */ + sig_attrs->mem.sig_type =IB_SIG_TYPE_NONE ; +nvme_rdma_set_diff_domain (c, bio, &sig_attrs->wire, rq); +/* Clear the PRACT bit since HCA will generate/verify the PI */ + control &= ~NVME_RW_PRINFO_PRACT ; + c->rw.control =cpu_to_le16 (control); + }else { +/* for WRITE_PASS/READ_PASS both wire/memory domains exist */ +nvme_rdma_set_diff_domain (c, bio, &sig_attrs->wire, rq); +nvme_rdma_set_diff_domain (c, bio, &sig_attrs->mem, rq); } +} + +static void nvme_rdma_set_prot_checks (struct nvme_command *cmd,u8 *mask) +{ + *mask =0 ; +if (le16_to_cpu (cmd->rw.control) &NVME_RW_PRINFO_PRCHK_REF ) + *mask |=IB_SIG_CHECK_REFTAG ; +if (le16_to_cpu (cmd->rw.control) &NVME_RW_PRINFO_PRCHK_GUARD ) + *mask |=IB_SIG_CHECK_GUARD ; +} + +static void nvme_rdma_sig_done (struct ib_cq *cq,struct ib_wc *wc) +{ +if (unlikely (wc->status !=IB_WC_SUCCESS )) +nvme_rdma_wr_error (cq, wc,"SIG" ); +} + +static void nvme_rdma_set_pi_wr (struct nvme_rdma_request *req, +struct nvme_command *c) +{ +struct ib_sig_attrs *sig_attrs = req->mr->sig_attrs; +struct ib_reg_wr *wr = &req->reg_wr; +struct request *rq =blk_mq_rq_from_pdu (req); +struct bio *bio = rq->bio; +struct nvme_keyed_sgl_desc *sg = &c->common.dptr.ksgl; + +nvme_rdma_set_sig_attrs (bio, sig_attrs, c, rq); + +nvme_rdma_set_prot_checks (c, &sig_attrs->check_mask);ib_update_fast_reg_key (req->mr,ib_inc_rkey (req->mr->rkey)); + req->reg_cqe.done = nvme_rdma_sig_done; + +memset (wr,0 ,sizeof (*wr)); + wr->wr.opcode =IB_WR_REG_MR_INTEGRITY ; + wr->wr.wr_cqe = &req->reg_cqe; + wr->wr.num_sge =0 ; + wr->wr.send_flags =0 ; + wr->mr = req->mr; + wr->key = req->mr->rkey; + wr->access =IB_ACCESS_LOCAL_WRITE | +IB_ACCESS_REMOTE_READ | +IB_ACCESS_REMOTE_WRITE ; + + sg->addr =cpu_to_le64 (req->mr->iova); +put_unaligned_le24 (req->mr->length, sg->length); +put_unaligned_le32 (req->mr->rkey, sg->key); + sg->type = (NVME_KEY_SGL_FMT_DATA_DESC <<4 ) |NVME_SGL_FMT_INVALIDATE ; +} +#else +static void nvme_rdma_set_pi_wr (struct nvme_rdma_request *req, +struct nvme_command *c) +{ +} +#endif /* CONFIG_BLK_DEV_INTEGRITY */ + +static void nvme_rdma_set_reg_wr (struct nvme_rdma_request *req, +struct nvme_command *c) +{ +struct nvme_keyed_sgl_desc *sg = &c->common.dptr.ksgl; + req->reg_cqe.done = nvme_rdma_memreg_done; +memset (&req->reg_wr,0 ,sizeof (req->reg_wr)); req->reg_wr.wr.opcode =IB_WR_REG_MR ; req->reg_wr.wr.wr_cqe = &req->reg_cqe;
@@ -1258,8 +1407,52 @@ static int nvme_rdma_map_sg_fr(struct nvme_rdma_queue *queue,put_unaligned_le32 (req->mr->rkey, sg->key); sg->type = (NVME_KEY_SGL_FMT_DATA_DESC <<4 ) |NVME_SGL_FMT_INVALIDATE ; +} + +static int nvme_rdma_map_sg_fr (struct nvme_rdma_queue *queue, +struct nvme_rdma_request *req,struct nvme_command *c, +int count,int pi_count) +{ +struct nvme_rdma_sgl *sgl = &req->data_sgl; +int nr; + +if (req->use_pi) { + req->mr =ib_mr_pool_get (queue->qp, &queue->qp->sig_mrs); +if (WARN_ON_ONCE (!req->mr)) +return -EAGAIN ; + + nr =ib_map_mr_sg_pi (req->mr, sgl->sg_table.sgl, count,0 , + req->pi_sgl->sg_table.sgl, pi_count,0 , +SZ_4K ); +if (unlikely (nr)) + goto mr_put; + +nvme_rdma_set_pi_wr (req, c); + }else { + req->mr =ib_mr_pool_get (queue->qp, &queue->qp->rdma_mrs); +if (WARN_ON_ONCE (!req->mr)) +return -EAGAIN ; +/* +* Align the MR to a 4K page size to match the ctrl page size +* and the block virtual boundary. +*/ + nr =ib_map_mr_sg (req->mr, sgl->sg_table.sgl, count,0 ,SZ_4K ); +if (unlikely (nr < count)) + goto mr_put; + +nvme_rdma_set_reg_wr (req, c); + }return 0 ; +mr_put: +if (req->use_pi) +ib_mr_pool_put (queue->qp, &queue->qp->sig_mrs, req->mr); +else +ib_mr_pool_put (queue->qp, &queue->qp->rdma_mrs, req->mr); + req->mr =NULL ; +if (nr <0 ) +return nr; +return -EINVAL ; } static int nvme_rdma_map_data(struct nvme_rdma_queue *queue,
@@ -1268,6 +1461,7 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue,struct nvme_rdma_request *req =blk_mq_rq_to_pdu (rq);struct nvme_rdma_device *dev = queue->device;struct ib_device *ibdev = dev->dev; +int pi_count =0 ;int count, ret; req->num_sge =1 ;
@@ -1295,7 +1489,30 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue, goto out_free_table; } - if (count <= dev->num_inline_segments) { +if (blk_integrity_rq (rq)) { +memset (req->pi_sgl,0 ,sizeof (struct nvme_rdma_sgl )); + req->pi_sgl->sg_table.sgl = + (struct scatterlist *)(req->pi_sgl +1 ); + ret =sg_alloc_table_chained (&req->pi_sgl->sg_table, +blk_rq_count_integrity_sg (rq->q, rq->bio), + req->pi_sgl->sg_table.sgl, +NVME_INLINE_PROT_SG_CNT ); +if (unlikely (ret)) { + ret = -ENOMEM ; + goto out_unmap_sg; + } + + req->pi_sgl->nents =blk_rq_map_integrity_sg (rq->q, rq->bio, + req->pi_sgl->sg_table.sgl); + pi_count =ib_dma_map_sg (ibdev, req->pi_sgl->sg_table.sgl, + req->pi_sgl->nents,rq_dma_dir (rq)); +if (unlikely (pi_count <=0 )) { + ret = -EIO ; + goto out_free_pi_table; + } + } + + if (count <= dev->num_inline_segments && !req->use_pi) { if (rq_data_dir(rq) == WRITE && nvme_rdma_queue_idx(queue) && queue->ctrl->use_inline_data && blk_rq_payload_bytes(rq) <=
@@ -1310,13 +1527,21 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue, } } - ret =nvme_rdma_map_sg_fr (queue, req, c, count); + ret =nvme_rdma_map_sg_fr (queue, req, c, count, pi_count); out:if (unlikely (ret)) - goto out_unmap_sg; + goto out_unmap_pi_sg;return 0 ; +out_unmap_pi_sg: +if (blk_integrity_rq (rq)) +ib_dma_unmap_sg (ibdev, req->pi_sgl->sg_table.sgl, + req->pi_sgl->nents,rq_dma_dir (rq)); +out_free_pi_table: +if (blk_integrity_rq (rq)) +sg_free_table_chained (&req->pi_sgl->sg_table, +NVME_INLINE_PROT_SG_CNT ); out_unmap_sg:ib_dma_unmap_sg (ibdev, req->data_sgl.sg_table.sgl, req->data_sgl.nents,rq_dma_dir (rq));
@@ -1769,6 +1994,8 @@ static blk_status_t nvme_rdma_queue_rq(struct blk_mq_hw_ctx *hctx,blk_mq_start_request (rq); + req->use_pi = queue->pi_support &&nvme_req (rq)->has_pi; + err =nvme_rdma_map_data (queue, rq, c);if (unlikely (err <0 )) { dev_err(queue->ctrl->ctrl.device,
@@ -1809,12 +2036,50 @@ static int nvme_rdma_poll(struct blk_mq_hw_ctx *hctx)return ib_process_cq_direct (queue->ib_cq,-1 ); } +#ifdef CONFIG_BLK_DEV_INTEGRITY +static void nvme_rdma_check_pi_status (struct nvme_rdma_request *req) +{ +struct request *rq =blk_mq_rq_from_pdu (req); +struct ib_mr_status mr_status; +int ret; + + ret =ib_check_mr_status (req->mr,IB_MR_CHECK_SIG_STATUS , &mr_status); +if (ret) { +pr_err ("ib_check_mr_status failed, ret %d\n" , ret); +nvme_req (rq)->status =NVME_SC_INVALID_PI ; +return ; + } + +if (mr_status.fail_status &IB_MR_CHECK_SIG_STATUS ) { +switch (mr_status.sig_err.err_type) { +case IB_SIG_BAD_GUARD : +nvme_req (rq)->status =NVME_SC_GUARD_CHECK ; +break ; +case IB_SIG_BAD_REFTAG : +nvme_req (rq)->status =NVME_SC_REFTAG_CHECK ; +break ; +case IB_SIG_BAD_APPTAG : +nvme_req (rq)->status =NVME_SC_APPTAG_CHECK ; +break ; + } +pr_err ("PI error found type %d expected 0x%x vs actual 0x%x\n" , + mr_status.sig_err.err_type, mr_status.sig_err.expected, + mr_status.sig_err.actual); + } +} +#endif +static void nvme_rdma_complete_rq (struct request *rq) {struct nvme_rdma_request *req =blk_mq_rq_to_pdu (rq);struct nvme_rdma_queue *queue = req->queue;struct ib_device *ibdev = queue->device->dev; +#ifdef CONFIG_BLK_DEV_INTEGRITY +if (req->use_pi) +nvme_rdma_check_pi_status (req); +#endif +nvme_rdma_unmap_data (queue, rq);ib_dma_unmap_single (ibdev, req->sqe.dma,sizeof (struct nvme_command ),DMA_TO_DEVICE );
@@ -1934,7 +2199,7 @@ static void nvme_rdma_reset_ctrl_work(struct work_struct *work)static const struct nvme_ctrl_ops nvme_rdma_ctrl_ops = { .name ="rdma" , .module =THIS_MODULE , - .flags =NVME_F_FABRICS , + .flags =NVME_F_FABRICS |NVME_F_METADATA_SUPPORTED , .reg_read32 = nvmf_reg_read32, .reg_read64 = nvmf_reg_read64, .reg_write32 = nvmf_reg_write32,
@@ -2078,7 +2343,7 @@ static struct nvmf_transport_ops nvme_rdma_transport = { .allowed_opts = NVMF_OPT_TRSVCID | NVMF_OPT_RECONNECT_DELAY |NVMF_OPT_HOST_TRADDR |NVMF_OPT_CTRL_LOSS_TMO |NVMF_OPT_NR_WRITE_QUEUES |NVMF_OPT_NR_POLL_QUEUES | - NVMF_OPT_TOS, + NVMF_OPT_TOS | NVMF_OPT_PI_ENABLE, .create_ctrl = nvme_rdma_create_ctrl, };
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 07/16] nvme-rdma: Add metadata/T10-PI support
Quoted 38 lines · Max Gurtovoy
drivers/nvme/host/rdma.c +13
-16@@ -1215,34 +1283,115 @@ static int nvme_rdma_map_sg_single(struct nvme_rdma_queue *queue,return 0 ; } -static int nvme_rdma_map_sg_fr (struct nvme_rdma_queue *queue, -struct nvme_rdma_request *req,struct nvme_command *c, -int count) +#ifdef CONFIG_BLK_DEV_INTEGRITY +static void nvme_rdma_set_diff_domain (struct nvme_command *cmd,struct bio *bio, +struct ib_sig_domain *domain,struct request *rq) { -struct nvme_keyed_sgl_desc *sg = &c->common.dptr.ksgl; -int nr; +struct blk_integrity *bi =blk_get_integrity (bio->bi_disk); +struct nvme_ns *ns = rq->rq_disk->private_data; + +WARN_ON (bi ==NULL ); - req->mr =ib_mr_pool_get (queue->qp, &queue->qp->rdma_mrs); -if (WARN_ON_ONCE (!req->mr)) -return -EAGAIN ; + domain->sig_type =IB_SIG_TYPE_T10_DIF ; + domain->sig.dif.bg_type =IB_T10DIF_CRC ; + domain->sig.dif.pi_interval =1 << bi->interval_exp; + domain->sig.dif.ref_tag =le32_to_cpu (cmd->rw.reftag);/* -* Align the MR to a 4K page size to match the ctrl page size and -* the block virtual boundary. +* At the moment we hard code those, but in the future +* we will take them from cmd. */ - nr =ib_map_mr_sg (req->mr, req->data_sgl.sg_table.sgl, count,NULL , -SZ_4K ); -if (unlikely (nr < count)) { -ib_mr_pool_put (queue->qp, &queue->qp->rdma_mrs, req->mr); - req->mr =NULL ; -if (nr <0 )
drivers/nvme/host/rdma.c static int nvme_rdma_map_sg_single(struct nvme_rdma_queue *queue, · line 1311
Quoted 10 lines
drivers/nvme/host/rdma.c +7
-2-return nr; -return -EINVAL ; + domain->sig.dif.apptag_check_mask =0xffff ; + domain->sig.dif.app_escape = true; + domain->sig.dif.ref_escape = true; +if (ns->pi_type !=NVME_NS_DPS_PI_TYPE3 ) + domain->sig.dif.ref_remap = true; +} +
On a per-io basis, there needs to be specific descriptions of the DIF information to program the port hardware. Things such as block size, type, and so. I see this routine using a mix of the bio that is associated with the original request as well as the namespace pointer to get this info. To me the reaching into the bio, as well as the locating of the ns structures are reaching into the other layers too much.
Wouldn't we be better off with with the core layer doing all the reaching and setting up a pi structure in the nvme_request with this information ? replace has_pi with this pi struct and "nvme_req(rq)->pi.pi_type == 0" is equivalent to has_pi ? If we didn't want to replicate the PI info, then nvme_request can simply add a pointer to the ns, and the ns can be looked at explicitly to gather the attributes.
Thoughts ?
-- james
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 07/16] nvme-rdma: Add metadata/T10-PI support
Quoted 53 lines · James Smart
Quoted 37 lines · Max Gurtovoy
drivers/nvme/host/rdma.c +13
-14@@ -1215,34 +1283,115 @@ static int nvme_rdma_map_sg_single(structnvme_rdma_queue *queue, return 0; } -static int nvme_rdma_map_sg_fr(struct nvme_rdma_queue *queue, - struct nvme_rdma_request *req, struct nvme_command *c, - int count) +#ifdef CONFIG_BLK_DEV_INTEGRITY +static void nvme_rdma_set_diff_domain(struct nvme_command *cmd, struct bio *bio, + struct ib_sig_domain *domain, struct request *rq) { - struct nvme_keyed_sgl_desc *sg = &c->common.dptr.ksgl; - int nr; + struct blk_integrity *bi = blk_get_integrity(bio->bi_disk); + struct nvme_ns *ns = rq->rq_disk->private_data; + + WARN_ON(bi == NULL); - req->mr = ib_mr_pool_get(queue->qp, &queue->qp->rdma_mrs); - if (WARN_ON_ONCE(!req->mr)) - return -EAGAIN; + domain->sig_type = IB_SIG_TYPE_T10_DIF; + domain->sig.dif.bg_type = IB_T10DIF_CRC; + domain->sig.dif.pi_interval = 1 << bi->interval_exp; + domain->sig.dif.ref_tag = le32_to_cpu(cmd->rw.reftag); /* - * Align the MR to a 4K page size to match the ctrl page size and - * the block virtual boundary. +* At the moment we hard code those, but in the future +* we will take them from cmd. */ - nr = ib_map_mr_sg(req->mr, req->data_sgl.sg_table.sgl, count, NULL, - SZ_4K); - if (unlikely(nr < count)) { - ib_mr_pool_put(queue->qp, &queue->qp->rdma_mrs, req->mr); - req->mr = NULL; - if (nr < 0)drivers/nvme/host/rdma.c static int nvme_rdma_map_sg_single(struct · line 1312
Quoted 10 lines
drivers/nvme/host/rdma.c +7
-2- return nr; - return -EINVAL; + domain->sig.dif.apptag_check_mask = 0xffff; + domain->sig.dif.app_escape = true; + domain->sig.dif.ref_escape = true; + if (ns->pi_type != NVME_NS_DPS_PI_TYPE3) + domain->sig.dif.ref_remap = true; +} +On a per-io basis, there needs to be specific descriptions of the DIF information to program the port hardware. Things such as block size, type, and so. I see this routine using a mix of the bio that is associated with the original request as well as the namespace pointer to get this info. To me the reaching into the bio, as well as the locating of the ns structures are reaching into the other layers too much.
Quoted 4 lines
Wouldn't we be better off with with the core layer doing all the reaching and setting up a pi structure in the
nvme_requestwith this information ? replacehas_piwith this pi struct and "nvme_req(rq)->pi.pi_type == 0" is equivalent tohas_pi? If we didn't want to replicate the PI info, thennvme_requestcan simply add a pointer to the ns, and the ns can be looked at explicitly to gather the attributes.Thoughts ?
NVMe namespace is used by RDMA/TCP/FC transport drivers in each queue_rq implementation. We can pass it instead of reaching it from the rq, if this looks better.
NVMe PI attributes (e.g reftag/check_flags/action_flags) are all set in the core layer.
PI attributes for configuring the HW of each transport should be done in the transport driver.
Quoted 2 lines
-- james
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 07/16] nvme-rdma: Add metadata/T10-PI support
Quoted 9 lines · Max Gurtovoy
Quoted 6 lines · James Smart
On a per-io basis, there needs to be specific descriptions of the DIF information to program the port hardware. Things such as block size, type, and so. I see this routine using a mix of the bio that is associated with the original request as well as the namespace pointer to get this info. To me the reaching into the bio, as well as the locating of the ns structures are reaching into the other layers too much.
Wouldn't we be better off with with the core layer doing all the reaching and setting up a pi structure in the
nvme_requestwith this information ? replacehas_piwith this pi struct and "nvme_req(rq)->pi.pi_type == 0" is equivalent tohas_pi? If we didn't want to replicate the PI info, thennvme_requestcan simply add a pointer to the ns, and the ns can be looked at explicitly to gather the attributes.Thoughts ?
NVMe namespace is used by RDMA/TCP/FC transport drivers in each
queue_rqimplementation. We can pass it instead of reaching it from the rq, if this looks better.
?? there hasn't been any reason to look at the namespace by the transports to date.
But yes, I much prefer passing the ns and using it for the dif info. This removes what I didn't like - encoding the relationship of how to get from the rq to ns in each transport.
Quoted 6 lines
NVMe PI attributes (e.g reftag/check_flags/action_flags) are all set in the core layer.
PI attributes for configuring the HW of each transport should be done in the transport driver.
agreed.
-- james
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 07/16] nvme-rdma: Add metadata/T10-PI support
Quoted 9 lines · James Smart
Quoted 6 lines · Max Gurtovoy
Quoted 4 lines · James Smart
On a per-io basis, there needs to be specific descriptions of the DIF information to program the port hardware. Things such as block size, type, and so. I see this routine using a mix of the bio that is associated with the original request as well as the namespace pointer to get this info. To me the reaching into the bio, as well as the locating of the ns structures are reaching into the other layers too much.
Wouldn't we be better off with with the core layer doing all the reaching and setting up a pi structure in the
nvme_requestwith this information ? replacehas_piwith this pi struct and "nvme_req(rq)->pi.pi_type == 0" is equivalent tohas_pi? If we didn't want to replicate the PI info, thennvme_requestcan simply add a pointer to the ns, and the ns can be looked at explicitly to gather the attributes.
Quoted 8 lines
Quoted 3 lines
Quoted 2 lines
Thoughts ?
NVMe namespace is used by RDMA/TCP/FC transport drivers in each
queue_rqimplementation. We can pass it instead of reaching it from the rq, if this looks better.?? there hasn't been any reason to look at the namespace by the transports to date.
But yes, I much prefer passing the ns and using it for the dif info. This removes what I didn't like - encoding the relationship of how to get from the rq to ns in each transport.
also means the bio isn't necessary in most of the other routines until you call blk_rq_count_integrity_sg(). blksz can come from (1 << ns->lba_shift)
-- james
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 8/16 block: Add a BIP_MAPPED_INTEGRITY check to complete_fn
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
This check is needed in case some other layer did the reftag remapping (e.g. in NVMe/RDMA the initiator controller performs the remapping so target side shouldn't map it again).
--- block/t10-pi.c | 4 ++++ 1 file changed, 4 insertions(+)
block/t10-pi.c +4 -0
@@ -193,6 +193,10 @@ static void t10_pi_type1_complete(struct request *rq, unsigned int nr_bytes)struct bio_vec iv;struct bvec_iter iter; +/* Already remapped? */ +if (bip->bip_flags &BIP_MAPPED_INTEGRITY ) +break ; +bip_for_each_vec (iv, bip, iter) {void *p, *pmap;unsigned int j;
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH 1/1 nvme-cli/fabrics: Add pi_enable param to connect cmd
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
Added 'pi_enable' to 'connect' command so users can enable metadata support.
usage examples:
nvme connect --pi_enable --transport=rdma --traddr=10.0.1.1 --nqn=test-nvme nvme connect -p -t rdma -a 10.0.1.1 -n test_nvme
--- fabrics.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
fabrics.c +13 -1
@@ -72,6 +72,7 @@ static struct config {int disable_sqflow;int hdr_digest;int data_digest; +int pi_enable;bool persistent;bool quiet; } cfg = { NULL };
@@ -756,7 +757,9 @@ static int build_options(char *argstr, int max_len, bool discover)add_bool_argument (&argstr, &max_len,"disable_sqflow" , cfg.disable_sqflow) ||add_bool_argument (&argstr, &max_len,"hdr_digest" , cfg.hdr_digest) || - add_bool_argument(&argstr, &max_len, "data_digest", cfg.data_digest)) +add_bool_argument (&argstr, &max_len,"data_digest" , + cfg.data_digest) || + add_bool_argument(&argstr, &max_len, "pi_enable", cfg.pi_enable)) return -EINVAL;return 0 ;
@@ -885,6 +888,13 @@ retry: p += len; } +if (cfg.pi_enable) { + len =sprintf (p,",pi_enable" ); +if (len <0 ) +return -EINVAL ; + p += len; + } +switch (e->trtype) {case NVMF_TRTYPE_RDMA :case NVMF_TRTYPE_TCP :
@@ -1171,6 +1181,7 @@ int discover(const char *desc, int argc, char **argv, bool connect)OPT_INT ("tos" ,'T' , &cfg.tos,"type of service" ),OPT_FLAG ("hdr_digest" ,'g' , &cfg.hdr_digest,"enable transport protocol header digest (TCP transport)" ),OPT_FLAG ("data_digest" ,'G' , &cfg.data_digest,"enable transport protocol data digest (TCP transport)" ), +OPT_FLAG ("pi_enable" ,'p' , &cfg.pi_enable,"enable metadata (T10-PI) support (default false)" ),OPT_INT ("nr-io-queues" ,'i' , &cfg.nr_io_queues,"number of io queues to use (default is core count)" ),OPT_INT ("nr-write-queues" ,'W' , &cfg.nr_write_queues,"number of write queues to use (default 0)" ), OPT_INT("nr-poll-queues", 'P', &cfg.nr_poll_queues, "number of poll queues to use (default 0)"),
@@ -1231,6 +1242,7 @@ int connect(const char *desc, int argc, char **argv)OPT_FLAG ("disable-sqflow" ,'d' , &cfg.disable_sqflow,"disable controller sq flow control (default false)" ),OPT_FLAG ("hdr-digest" ,'g' , &cfg.hdr_digest,"enable transport protocol header digest (TCP transport)" ),OPT_FLAG ("data-digest" ,'G' , &cfg.data_digest,"enable transport protocol data digest (TCP transport)" ), +OPT_FLAG ("pi_enable" ,'p' , &cfg.pi_enable,"enable metadata (T10-PI) support (default false)" ), OPT_END() };
Signature · 7 lines
1.8.3.1 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 9/16 nvmet: Prepare metadata request
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
Allocate the metadata SGL buffers and add metadata fields for the request. This is a preparation for adding metadata support over the fabrics.
--- drivers/nvme/target/core.c | 48 ++++++++++++++++++++++++++++++++++++++------- drivers/nvme/target/nvmet.h | 5 +++++ 2 files changed, 46 insertions(+), 7 deletions(-)
drivers/nvme/target/core.c +41 -7
@@ -863,13 +863,17 @@ bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq, req->sq = sq; req->ops = ops; req->sg =NULL ; + req->prot_sg =NULL ; req->sg_cnt =0 ; + req->prot_sg_cnt =0 ; req->transfer_len =0 ; + req->prot_len =0 ; req->cqe->status =0 ; req->cqe->sq_head =0 ; req->ns =NULL ; req->error_loc =NVMET_NO_ERROR_LOC ; req->error_slba =0 ; + req->use_pi = false;trace_nvmet_req_init (req, req->cmd);
@@ -941,6 +945,7 @@ EXPORT_SYMBOL_GPL(nvmet_check_data_len);int nvmet_req_alloc_sgl (struct nvmet_req *req) {struct pci_dev *p2p_dev =NULL ; +int data_len = req->transfer_len - req->prot_len;if (IS_ENABLED (CONFIG_PCI_P2PDMA )) {if (req->sq->ctrl && req->ns)
@@ -950,11 +955,23 @@ int nvmet_req_alloc_sgl(struct nvmet_req *req) req->p2p_dev =NULL ;if (req->sq->qid && p2p_dev) { req->sg =pci_p2pmem_alloc_sgl (p2p_dev, &req->sg_cnt, - req->transfer_len); -if (req->sg) { - req->p2p_dev = p2p_dev; -return 0 ; + data_len); +if (!req->sg) + goto fallback; + +if (req->prot_len) { + req->prot_sg = +pci_p2pmem_alloc_sgl (p2p_dev, + &req->prot_sg_cnt, + req->prot_len); +if (!req->prot_sg) { +pci_p2pmem_free_sgl (req->p2p_dev, + req->sg); + goto fallback; + } } + req->p2p_dev = p2p_dev; +return 0 ; }/*
@@ -963,23 +980,40 @@ int nvmet_req_alloc_sgl(struct nvmet_req *req)*/ } - req->sg =sgl_alloc (req->transfer_len,GFP_KERNEL , &req->sg_cnt); +fallback: + req->sg =sgl_alloc (data_len,GFP_KERNEL , &req->sg_cnt);if (unlikely (!req->sg))return -ENOMEM ; +if (req->prot_len) { + req->prot_sg =sgl_alloc (req->prot_len,GFP_KERNEL , + &req->prot_sg_cnt); +if (unlikely (!req->prot_sg)) { +sgl_free (req->sg); +return -ENOMEM ; + } + } +return 0 ; }EXPORT_SYMBOL_GPL (nvmet_req_alloc_sgl);void nvmet_req_free_sgl (struct nvmet_req *req) { -if (req->p2p_dev) +if (req->p2p_dev) {pci_p2pmem_free_sgl (req->p2p_dev, req->sg); -else +if (req->prot_sg) +pci_p2pmem_free_sgl (req->p2p_dev, req->prot_sg); + }else {sgl_free (req->sg); +if (req->prot_sg) +sgl_free (req->prot_sg); + } req->sg =NULL ; + req->prot_sg =NULL ; req->sg_cnt =0 ; + req->prot_sg_cnt =0 ; }EXPORT_SYMBOL_GPL (nvmet_req_free_sgl);
drivers/nvme/target/nvmet.h +5 -0
@@ -291,6 +291,7 @@ struct nvmet_req {struct nvmet_cq *cq;struct nvmet_ns *ns;struct scatterlist *sg; +struct scatterlist *prot_sg;struct bio_vec inline_bvec[NVMET_MAX_INLINE_BIOVEC ]; union { struct {
@@ -304,8 +305,10 @@ struct nvmet_req { } f; };int sg_cnt; +int prot_sg_cnt;/* data length as parsed from the SGL descriptor: */ size_t transfer_len; +size_t prot_len;struct nvmet_port *port;
@@ -316,6 +319,8 @@ struct nvmet_req {struct device *p2p_client;u16 error_loc;u64 error_slba; +/* Metadata (T10-PI) support */ +bool use_pi; };extern struct workqueue_struct *buffered_io_wq;
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 10/16 nvmet: Add metadata characteristics for a namespace
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
This is a preparation for adding metadata (T10-PI) over fabric support.
--- drivers/nvme/target/io-cmd-bdev.c | 22 ++++++++++++++++++++++ drivers/nvme/target/nvmet.h | 3 +++ 2 files changed, 25 insertions(+)
drivers/nvme/target/io-cmd-bdev.c +22 -0
@@ -50,6 +50,9 @@ void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id)int nvmet_bdev_ns_enable (struct nvmet_ns *ns) {int ret; +#ifdef CONFIG_BLK_DEV_INTEGRITY +struct blk_integrity *bi; +#endif ns->bdev =blkdev_get_by_path (ns->device_path,FMODE_READ |FMODE_WRITE ,NULL );
@@ -64,6 +67,25 @@ int nvmet_bdev_ns_enable(struct nvmet_ns *ns) } ns->size =i_size_read (ns->bdev->bd_inode); ns->blksize_shift =blksize_bits (bdev_logical_block_size (ns->bdev)); + + ns->prot_type =0 ; + ns->ms =0 ; +#ifdef CONFIG_BLK_DEV_INTEGRITY + bi =bdev_get_integrity (ns->bdev); +if (bi) { + ns->ms = bi->tuple_size; +if (bi->profile == &t10_pi_type1_crc) + ns->prot_type =NVME_NS_DPS_PI_TYPE1 ; +else if (bi->profile == &t10_pi_type3_crc) + ns->prot_type =NVME_NS_DPS_PI_TYPE3 ; +else +/* Unsupported metadata type */ + ns->ms =0 ; + } + +pr_debug ("ms %d pi_type %d\n" , ns->ms, ns->prot_type); +#endif +return 0 ; }
drivers/nvme/target/nvmet.h +3 -0
@@ -19,6 +19,7 @@#include <linux/rcupdate.h> #include <linux/blkdev.h> #include <linux/radix-tree.h> +#include <linux/t10-pi.h> #define NVMET_ASYNC_EVENTS 4#define NVMET_ERROR_LOG_SLOTS 128
@@ -76,6 +77,8 @@ struct nvmet_ns {int use_p2pmem;struct pci_dev *p2p_dev; +int prot_type; +int ms; };static inline struct nvmet_ns *to_nvmet_ns (struct config_item *item)
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 11/16 nvmet: Rename nvmet_rw_len to nvmet_rw_data_len
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
The function doesn't add the metadata length (only data length is calculated). This is preparation for adding metadata (T10-PI) support.
--- drivers/nvme/target/io-cmd-bdev.c | 2 +- drivers/nvme/target/io-cmd-file.c | 2 +- drivers/nvme/target/nvmet.h | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-)
drivers/nvme/target/io-cmd-bdev.c +1 -1
@@ -173,7 +173,7 @@ static void nvmet_bdev_execute_rw(struct nvmet_req *req)sector_t sector;int op, i; -if (!nvmet_check_data_len (req,nvmet_rw_len (req))) +if (!nvmet_check_data_len (req,nvmet_rw_data_len (req)))return ;if (!req->sg_cnt) {
drivers/nvme/target/io-cmd-file.c +1 -1
@@ -232,7 +232,7 @@ static void nvmet_file_execute_rw(struct nvmet_req *req) {ssize_t nr_bvec = req->sg_cnt; -if (!nvmet_check_data_len (req,nvmet_rw_len (req))) +if (!nvmet_check_data_len (req,nvmet_rw_data_len (req)))return ;if (!req->sg_cnt || !nr_bvec) {
drivers/nvme/target/nvmet.h +1 -1
@@ -495,7 +495,7 @@ u16 nvmet_bdev_flush(struct nvmet_req *req);u16 nvmet_file_flush (struct nvmet_req *req);void nvmet_ns_changed (struct nvmet_subsys *subsys,u32 nsid); -static inline u32 nvmet_rw_len (struct nvmet_req *req) +static inline u32 nvmet_rw_data_len (struct nvmet_req *req) {return ((u32 )le16_to_cpu (req->cmd->rw.length) +1 ) << req->ns->blksize_shift;
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 12/16 nvmet: Rename nvmet_check_data_len to nvmet_check_transfer_len
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
The function doesn't checks only the data length, because the transfer length includes also the metadata length in some cases. This is preparation for adding metadata (T10-PI) support.
--- drivers/nvme/target/admin-cmd.c | 14 +++++++------- drivers/nvme/target/core.c | 6 +++--- drivers/nvme/target/discovery.c | 8 ++++---- drivers/nvme/target/fabrics-cmd.c | 8 ++++---- drivers/nvme/target/io-cmd-bdev.c | 8 ++++---- drivers/nvme/target/io-cmd-file.c | 8 ++++---- drivers/nvme/target/nvmet.h | 2 +- 7 files changed, 27 insertions(+), 27 deletions(-)
drivers/nvme/target/admin-cmd.c +7 -7
@@ -284,7 +284,7 @@ static void nvmet_execute_get_log_page_ana(struct nvmet_req *req)static void nvmet_execute_get_log_page (struct nvmet_req *req) { -if (!nvmet_check_data_len (req,nvmet_get_log_page_len (req->cmd))) +if (!nvmet_check_transfer_len (req,nvmet_get_log_page_len (req->cmd)))return ;switch (req->cmd->get_log_page.lid) {
@@ -597,7 +597,7 @@ static void nvmet_execute_identify_desclist(struct nvmet_req *req)static void nvmet_execute_identify (struct nvmet_req *req) { -if (!nvmet_check_data_len (req,NVME_IDENTIFY_DATA_SIZE )) +if (!nvmet_check_transfer_len (req,NVME_IDENTIFY_DATA_SIZE ))return ;switch (req->cmd->identify.cns) {
@@ -626,7 +626,7 @@ static void nvmet_execute_identify(struct nvmet_req *req)*/ static void nvmet_execute_abort (struct nvmet_req *req) { -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;nvmet_set_result (req,1 );nvmet_req_complete (req,0 );
@@ -712,7 +712,7 @@ static void nvmet_execute_set_features(struct nvmet_req *req)u32 cdw10 =le32_to_cpu (req->cmd->common.cdw10);u16 status =0 ; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;switch (cdw10 &0xff ) {
@@ -778,7 +778,7 @@ static void nvmet_execute_get_features(struct nvmet_req *req)u32 cdw10 =le32_to_cpu (req->cmd->common.cdw10);u16 status =0 ; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;switch (cdw10 &0xff ) {
@@ -845,7 +845,7 @@ void nvmet_execute_async_event(struct nvmet_req *req) {struct nvmet_ctrl *ctrl = req->sq->ctrl; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;mutex_lock (&ctrl->lock);
@@ -864,7 +864,7 @@ void nvmet_execute_keep_alive(struct nvmet_req *req) {struct nvmet_ctrl *ctrl = req->sq->ctrl; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ; pr_debug("ctrl %d update keep-alive timer for %d secs\n",
drivers/nvme/target/core.c +3 -3
@@ -930,9 +930,9 @@ void nvmet_req_uninit(struct nvmet_req *req) }EXPORT_SYMBOL_GPL (nvmet_req_uninit); -bool nvmet_check_data_len (struct nvmet_req *req,size_t data_len) +bool nvmet_check_transfer_len (struct nvmet_req *req,size_t len) { -if (unlikely (data_len != req->transfer_len)) { +if (unlikely (len != req->transfer_len)) { req->error_loc = offsetof(struct nvme_common_command , dptr);nvmet_req_complete (req,NVME_SC_SGL_INVALID_DATA |NVME_SC_DNR );return false;
@@ -940,7 +940,7 @@ bool nvmet_check_data_len(struct nvmet_req *req, size_t data_len)return true; } -EXPORT_SYMBOL_GPL (nvmet_check_data_len); +EXPORT_SYMBOL_GPL (nvmet_check_transfer_len);int nvmet_req_alloc_sgl (struct nvmet_req *req) {
drivers/nvme/target/discovery.c +4 -4
@@ -171,7 +171,7 @@ static void nvmet_execute_disc_get_log_page(struct nvmet_req *req)u16 status =0 ;void *buffer; -if (!nvmet_check_data_len (req, data_len)) +if (!nvmet_check_transfer_len (req, data_len))return ;if (req->cmd->get_log_page.lid !=NVME_LOG_DISC ) {
@@ -244,7 +244,7 @@ static void nvmet_execute_disc_identify(struct nvmet_req *req)const char model[] ="Linux" ;u16 status =0 ; -if (!nvmet_check_data_len (req,NVME_IDENTIFY_DATA_SIZE )) +if (!nvmet_check_transfer_len (req,NVME_IDENTIFY_DATA_SIZE ))return ;if (req->cmd->identify.cns !=NVME_ID_CNS_CTRL ) {
@@ -298,7 +298,7 @@ static void nvmet_execute_disc_set_features(struct nvmet_req *req)u32 cdw10 =le32_to_cpu (req->cmd->common.cdw10);u16 stat; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;switch (cdw10 &0xff ) {
@@ -324,7 +324,7 @@ static void nvmet_execute_disc_get_features(struct nvmet_req *req)u32 cdw10 =le32_to_cpu (req->cmd->common.cdw10);u16 stat =0 ; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;switch (cdw10 &0xff ) {
drivers/nvme/target/fabrics-cmd.c +4 -4
@@ -12,7 +12,7 @@ static void nvmet_execute_prop_set(struct nvmet_req *req)u64 val =le64_to_cpu (req->cmd->prop_set.value);u16 status =0 ; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;if (req->cmd->prop_set.attrib &1 ) {
@@ -41,7 +41,7 @@ static void nvmet_execute_prop_get(struct nvmet_req *req)u16 status =0 ;u64 val =0 ; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;if (req->cmd->prop_get.attrib &1 ) {
@@ -151,7 +151,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)struct nvmet_ctrl *ctrl =NULL ;u16 status =0 ; -if (!nvmet_check_data_len (req,sizeof (struct nvmf_connect_data ))) +if (!nvmet_check_transfer_len (req,sizeof (struct nvmf_connect_data )))return ; d =kmalloc (sizeof (*d),GFP_KERNEL );
@@ -218,7 +218,7 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)u16 qid =le16_to_cpu (c->qid);u16 status =0 ; -if (!nvmet_check_data_len (req,sizeof (struct nvmf_connect_data ))) +if (!nvmet_check_transfer_len (req,sizeof (struct nvmf_connect_data )))return ; d =kmalloc (sizeof (*d),GFP_KERNEL );
drivers/nvme/target/io-cmd-bdev.c +4 -4
@@ -173,7 +173,7 @@ static void nvmet_bdev_execute_rw(struct nvmet_req *req)sector_t sector;int op, i; -if (!nvmet_check_data_len (req,nvmet_rw_data_len (req))) +if (!nvmet_check_transfer_len (req,nvmet_rw_data_len (req)))return ;if (!req->sg_cnt) {
@@ -234,7 +234,7 @@ static void nvmet_bdev_execute_flush(struct nvmet_req *req) {struct bio *bio = &req->b.inline_bio; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;bio_init (bio, req->inline_bvec,ARRAY_SIZE (req->inline_bvec));
@@ -302,7 +302,7 @@ static void nvmet_bdev_execute_discard(struct nvmet_req *req)static void nvmet_bdev_execute_dsm (struct nvmet_req *req) { -if (!nvmet_check_data_len (req,nvmet_dsm_len (req))) +if (!nvmet_check_transfer_len (req,nvmet_dsm_len (req)))return ;switch (le32_to_cpu (req->cmd->dsm.attributes)) {
@@ -326,7 +326,7 @@ static void nvmet_bdev_execute_write_zeroes(struct nvmet_req *req)sector_t nr_sector;int ret; -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ; sector = le64_to_cpu(write_zeroes->slba) <<
drivers/nvme/target/io-cmd-file.c +4 -4
@@ -232,7 +232,7 @@ static void nvmet_file_execute_rw(struct nvmet_req *req) {ssize_t nr_bvec = req->sg_cnt; -if (!nvmet_check_data_len (req,nvmet_rw_data_len (req))) +if (!nvmet_check_transfer_len (req,nvmet_rw_data_len (req)))return ;if (!req->sg_cnt || !nr_bvec) {
@@ -276,7 +276,7 @@ static void nvmet_file_flush_work(struct work_struct *w)static void nvmet_file_execute_flush (struct nvmet_req *req) { -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;INIT_WORK (&req->f.work, nvmet_file_flush_work);schedule_work (&req->f.work);
@@ -336,7 +336,7 @@ static void nvmet_file_dsm_work(struct work_struct *w)static void nvmet_file_execute_dsm (struct nvmet_req *req) { -if (!nvmet_check_data_len (req,nvmet_dsm_len (req))) +if (!nvmet_check_transfer_len (req,nvmet_dsm_len (req)))return ;INIT_WORK (&req->f.work, nvmet_file_dsm_work);schedule_work (&req->f.work);
@@ -366,7 +366,7 @@ static void nvmet_file_write_zeroes_work(struct work_struct *w)static void nvmet_file_execute_write_zeroes (struct nvmet_req *req) { -if (!nvmet_check_data_len (req,0 )) +if (!nvmet_check_transfer_len (req,0 ))return ;INIT_WORK (&req->f.work, nvmet_file_write_zeroes_work);schedule_work (&req->f.work);
drivers/nvme/target/nvmet.h +1 -1
@@ -381,7 +381,7 @@ u16 nvmet_parse_fabrics_cmd(struct nvmet_req *req);bool nvmet_req_init (struct nvmet_req *req,struct nvmet_cq *cq,struct nvmet_sq *sq,const struct nvmet_fabrics_ops *ops);void nvmet_req_uninit (struct nvmet_req *req); -bool nvmet_check_data_len (struct nvmet_req *req,size_t data_len); +bool nvmet_check_transfer_len (struct nvmet_req *req,size_t len);void nvmet_req_complete (struct nvmet_req *req,u16 status);int nvmet_req_alloc_sgl (struct nvmet_req *req);void nvmet_req_free_sgl (struct nvmet_req *req);
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 13/16 nvme: Add Metadata Capabilities enumerations
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
This is preparation for adding metadata (T10-PI) over fabric support.
--- include/linux/nvme.h | 2 ++ 1 file changed, 2 insertions(+)
include/linux/nvme.h +2 -0
@@ -396,6 +396,8 @@ enum {NVME_NS_FEAT_THIN =1 <<0 ,NVME_NS_FLBAS_LBA_MASK =0xf ,NVME_NS_FLBAS_META_EXT =0x10 , +NVME_NS_MC_META_EXT =1 <<0 , +NVME_NS_MC_META_BUF =1 <<1 ,NVME_LBAF_RP_BEST =0 ,NVME_LBAF_RP_BETTER =1 , NVME_LBAF_RP_GOOD = 2,
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 14/16 nvmet: Add metadata/T10-PI support
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
Expose the namespace metadata format when PI is enabled. The user needs to enable the capability per subsystem and per port. The other metadata properties are taken from the namespace/bdev.
Usage example:
echo 1 > /config/nvmet/subsystems/${NAME}/attr_pi_enable echo 1 > /config/nvmet/ports/${PORT_NUM}/param_pi_enable
--- drivers/nvme/target/admin-cmd.c | 23 ++++++++++++--- drivers/nvme/target/configfs.c | 61 +++++++++++++++++++++++++++++++++++++++ drivers/nvme/target/fabrics-cmd.c | 11 +++++++ drivers/nvme/target/nvmet.h | 28 ++++++++++++++++++ 4 files changed, 119 insertions(+), 4 deletions(-)
drivers/nvme/target/admin-cmd.c +19 -4
@@ -346,8 +346,8 @@ static void nvmet_execute_identify_ctrl(struct nvmet_req *req)/* we support multiple ports, multiples hosts and ANA: */ id->cmic = (1 <<0 ) | (1 <<1 ) | (1 <<3 ); -/* no limit on data transfer sizes for now */ - id->mdts =0 ; +/* Limit MDTS for metadata capable controllers (assume mpsmin is 4k) */ + id->mdts = ctrl->pi_support ?ilog2 (NVMET_T10_PI_MAX_KB_SZ >>2 ) :0 ; id->cntlid =cpu_to_le16 (ctrl->cntlid); id->ver =cpu_to_le32 (ctrl->subsys->ver);
@@ -405,9 +405,13 @@ static void nvmet_execute_identify_ctrl(struct nvmet_req *req)strlcpy (id->subnqn, ctrl->subsys->subsysnqn,sizeof (id->subnqn)); -/* Max command capsule size is sqe + single page of in-capsule data */ +/* +* Max command capsule size is sqe + single page of in-capsule data. +* Disable inline data for Metadata capable controllers. +*/ id->ioccsz =cpu_to_le32 ((sizeof (struct nvme_command ) + - req->port->inline_data_size) /16 ); + req->port->inline_data_size * + !ctrl->pi_support) /16 );/* Max response capsule size is cqe */ id->iorcsz =cpu_to_le32 (sizeof (struct nvme_completion ) /16 );
@@ -437,6 +441,7 @@ static void nvmet_execute_identify_ctrl(struct nvmet_req *req)static void nvmet_execute_identify_ns (struct nvmet_req *req) { +struct nvmet_ctrl *ctrl = req->sq->ctrl;struct nvmet_ns *ns;struct nvme_id_ns *id;u16 status =0 ;
@@ -493,6 +498,16 @@ static void nvmet_execute_identify_ns(struct nvmet_req *req) id->lbaf[0 ].ds = ns->blksize_shift; +if (ctrl->pi_support &&nvmet_ns_has_pi (ns)) { + id->dpc =NVME_NS_DPC_PI_FIRST |NVME_NS_DPC_PI_LAST | +NVME_NS_DPC_PI_TYPE1 |NVME_NS_DPC_PI_TYPE2 | +NVME_NS_DPC_PI_TYPE3 ; + id->mc =NVME_NS_MC_META_EXT ; + id->dps = ns->prot_type; + id->flbas =NVME_NS_FLBAS_META_EXT ; + id->lbaf[0 ].ms = ns->ms; + } +if (ns->readonly) id->nsattr |= (1 <<0 );nvmet_put_namespace (ns);
drivers/nvme/target/configfs.c +61 -0
83 lines not on this page · raw message
drivers/nvme/target/fabrics-cmd.c +11 -0
@@ -192,6 +192,17 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req) goto out; } +if (ctrl->subsys->pi_support && ctrl->port->pi_enable) { +if (ctrl->port->pi_capable) { + ctrl->pi_support = true; +pr_info ("controller %d T10-PI enabled\n" , ctrl->cntlid); + }else { + ctrl->pi_support = false; +pr_warn ("T10-PI is not supported on controller %d\n" , + ctrl->cntlid); + } + } +uuid_copy (&ctrl->hostid, &d->hostid); status =nvmet_install_queue (ctrl, req);
drivers/nvme/target/nvmet.h +28 -0
@@ -144,6 +144,8 @@ struct nvmet_port {bool enabled;int inline_data_size;const struct nvmet_fabrics_ops *tr_ops; +bool pi_capable; +bool pi_enable; };static inline struct nvmet_port *to_nvmet_port (struct config_item *item)
@@ -203,6 +205,7 @@ struct nvmet_ctrl {spinlock_t error_lock;u64 err_counter;struct nvme_error_slot slots[NVMET_ERROR_LOG_SLOTS ]; +bool pi_support; };struct nvmet_subsys {
@@ -225,6 +228,7 @@ struct nvmet_subsys {u64 ver;u64 serial;char *subsysnqn; +bool pi_support;struct config_group group;
@@ -501,6 +505,30 @@ static inline u32 nvmet_rw_data_len(struct nvmet_req *req) req->ns->blksize_shift; } +#ifdef CONFIG_BLK_DEV_INTEGRITY +static inline u32 nvmet_rw_prot_len (struct nvmet_req *req) +{ +return ((u32 )le16_to_cpu (req->cmd->rw.length) +1 ) * req->ns->ms; +} + +static inline bool nvmet_ns_has_pi (struct nvmet_ns *ns) +{ +return ns->prot_type && ns->ms ==sizeof (struct t10_pi_tuple ); +} +#else +static inline u32 nvmet_rw_prot_len (struct nvmet_req *req) +{ +return 0 ; +} + +static inline bool nvmet_ns_has_pi (struct nvmet_ns *ns) +{ +return false; +} +#endif /* CONFIG_BLK_DEV_INTEGRITY */ + +#define NVMET_T10_PI_MAX_KB_SZ 128 +static inline u32 nvmet_dsm_len (struct nvmet_req *req) { return (le32_to_cpu(req->cmd->dsm.nr) + 1) *
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 15/16 nvmet: Add metadata support for block devices
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
Create a block IO request for the metadata from the protection SG list.
--- drivers/nvme/target/io-cmd-bdev.c | 87 ++++++++++++++++++++++++++++++++++++++- 1 file changed, 85 insertions(+), 2 deletions(-)
drivers/nvme/target/io-cmd-bdev.c +85 -2
@@ -164,6 +164,61 @@ static void nvmet_bio_done(struct bio *bio)bio_put (bio); } +#ifdef CONFIG_BLK_DEV_INTEGRITY +static int nvmet_bdev_alloc_bip (struct nvmet_req *req,struct bio *bio, +struct sg_mapping_iter *miter) +{ +struct blk_integrity *bi; +struct bio_integrity_payload *bip; +struct block_device *bdev = req->ns->bdev; +int rc; +size_t resid, len; + + bi =bdev_get_integrity (bdev); +if (unlikely (!bi)) { +pr_err ("Unable to locate bio_integrity\n" ); +return -ENODEV ; + } + + bip =bio_integrity_alloc (bio,GFP_NOIO , + min_t(unsigned int, req->prot_sg_cnt, BIO_MAX_PAGES)); +if (IS_ERR (bip)) { +pr_err ("Unable to allocate bio_integrity_payload\n" ); +return PTR_ERR (bip); + } + +/* Payload is mapped by the fabrics host side */ + bip->bip_flags |=BIP_MAPPED_INTEGRITY ; + bip->bip_iter.bi_size =bio_integrity_bytes (bi,bio_sectors (bio)); +bip_set_seed (bip, bio->bi_iter.bi_sector); + + resid = bip->bip_iter.bi_size; +while (resid >0 &&sg_miter_next (miter)) { + len =min_t (size_t, miter->length, resid); + rc =bio_integrity_add_page (bio, miter->page, len, +offset_in_page (miter->addr)); +if (unlikely (rc != len)) { +pr_err ("bio_integrity_add_page() failed; %d\n" , rc); +sg_miter_stop (miter); +return -ENOMEM ; + } + + resid -= len; +if (len < miter->length) + miter->consumed -= miter->length - len; + } +sg_miter_stop (miter); + +return 0 ; +} +#else +static int nvmet_bdev_alloc_bip (struct nvmet_req *req,struct bio *bio, +struct sg_mapping_iter *miter) +{ +return 0 ; +} +#endif /* CONFIG_BLK_DEV_INTEGRITY */ +static void nvmet_bdev_execute_rw (struct nvmet_req *req) {int sg_cnt = req->sg_cnt;
@@ -171,9 +226,11 @@ static void nvmet_bdev_execute_rw(struct nvmet_req *req)struct scatterlist *sg;struct blk_plug plug;sector_t sector; -int op, i; +int op, i, rc; +struct sg_mapping_iter prot_miter; -if (!nvmet_check_transfer_len (req,nvmet_rw_data_len (req))) +if (!nvmet_check_transfer_len (req, +nvmet_rw_data_len (req) + req->prot_len))return ;if (!req->sg_cnt) {
@@ -208,11 +265,25 @@ static void nvmet_bdev_execute_rw(struct nvmet_req *req) bio->bi_opf = op;blk_start_plug (&plug); +if (req->use_pi) +sg_miter_start (&prot_miter, req->prot_sg, req->prot_sg_cnt, + op ==REQ_OP_READ ?SG_MITER_FROM_SG : +SG_MITER_TO_SG ); +for_each_sg (req->sg, sg, req->sg_cnt, i) {while (bio_add_page (bio,sg_page (sg), sg->length, sg->offset) != sg->length) {struct bio *prev = bio; +if (req->use_pi) { + rc =nvmet_bdev_alloc_bip (req, bio, + &prot_miter); +if (unlikely (rc)) { +bio_io_error (bio); +return ; + } + } + bio =bio_alloc (GFP_KERNEL ,min (sg_cnt,BIO_MAX_PAGES ));bio_set_dev (bio, req->ns->bdev); bio->bi_iter.bi_sector = sector;
@@ -226,6 +297,14 @@ static void nvmet_bdev_execute_rw(struct nvmet_req *req) sg_cnt--; } +if (req->use_pi) { + rc =nvmet_bdev_alloc_bip (req, bio, &prot_miter); +if (unlikely (rc)) { +bio_io_error (bio); +return ; + } + } +submit_bio (bio);blk_finish_plug (&plug); }
@@ -353,6 +432,10 @@ u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req)case nvme_cmd_read:case nvme_cmd_write: req->execute = nvmet_bdev_execute_rw; +if (req->sq->ctrl->pi_support &&nvmet_ns_has_pi (req->ns)) { + req->use_pi = true; + req->prot_len =nvmet_rw_prot_len (req); + }return 0 ;case nvme_cmd_flush: req->execute = nvmet_bdev_execute_flush;
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
PATCH v2 16/16 nvmet-rdma: Add metadata/T10-PI support
In reply to Max GurtovoyFrom: Israel Rukshin <[email protected]>
For capable HCAs (e.g. ConnectX-4/ConnectX-5) this will allow end-to-end protection information passthrough and validation for NVMe over RDMA transport. Metadata support was implemented over the new RDMA signature verbs API.
--- drivers/nvme/target/rdma.c | 235 +++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 217 insertions(+), 18 deletions(-)
drivers/nvme/target/rdma.c +217 -18
@@ -54,6 +54,7 @@ struct nvmet_rdma_rsp {struct nvmet_rdma_queue *queue;struct ib_cqe read_cqe; +struct ib_cqe write_cqe;struct rdma_rw_ctx rw;struct nvmet_req req;
@@ -129,6 +130,9 @@ static bool nvmet_rdma_execute_command(struct nvmet_rdma_rsp *rsp);static void nvmet_rdma_send_done (struct ib_cq *cq,struct ib_wc *wc);static void nvmet_rdma_recv_done (struct ib_cq *cq,struct ib_wc *wc);static void nvmet_rdma_read_data_done (struct ib_cq *cq,struct ib_wc *wc); +#ifdef CONFIG_BLK_DEV_INTEGRITY +static void nvmet_rdma_write_data_done (struct ib_cq *cq,struct ib_wc *wc); +#endif static void nvmet_rdma_qp_event (struct ib_event *event,void *priv);static void nvmet_rdma_queue_disconnect (struct nvmet_rdma_queue *queue); static void nvmet_rdma_free_rsp(struct nvmet_rdma_device *ndev,
@@ -386,6 +390,10 @@ static int nvmet_rdma_alloc_rsp(struct nvmet_rdma_device *ndev,/* Data In / RDMA READ */ r->read_cqe.done = nvmet_rdma_read_data_done; +#ifdef CONFIG_BLK_DEV_INTEGRITY +/* Data Out / RDMA WRITE */ + r->write_cqe.done = nvmet_rdma_write_data_done; +#endif return 0 ; out_free_rsp:
@@ -495,6 +503,138 @@ static void nvmet_rdma_process_wr_wait_list(struct nvmet_rdma_queue *queue)spin_unlock (&queue->rsp_wr_wait_lock); } +#ifdef CONFIG_BLK_DEV_INTEGRITY +static u16 nvmet_rdma_check_pi_status (struct ib_mr *sig_mr) +{ +struct ib_mr_status mr_status; +int ret; +u16 status =0 ; + + ret =ib_check_mr_status (sig_mr,IB_MR_CHECK_SIG_STATUS , &mr_status); +if (ret) { +pr_err ("ib_check_mr_status failed, ret %d\n" , ret); +return NVME_SC_INVALID_PI ; + } + +if (mr_status.fail_status &IB_MR_CHECK_SIG_STATUS ) { +switch (mr_status.sig_err.err_type) { +case IB_SIG_BAD_GUARD : + status =NVME_SC_GUARD_CHECK ; +break ; +case IB_SIG_BAD_REFTAG : + status =NVME_SC_REFTAG_CHECK ; +break ; +case IB_SIG_BAD_APPTAG : + status =NVME_SC_APPTAG_CHECK ; +break ; + } +pr_err ("PI error found type %d expected 0x%x vs actual 0x%x\n" , + mr_status.sig_err.err_type, + mr_status.sig_err.expected, + mr_status.sig_err.actual); + } + +return status; +} + +static void nvmet_rdma_set_diff_domain (struct nvme_command *cmd, +struct ib_sig_domain *domain,struct nvmet_ns *ns) +{ +struct blk_integrity *bi =bdev_get_integrity (ns->bdev); + +WARN_ON (bi ==NULL ); + + domain->sig_type =IB_SIG_TYPE_T10_DIF ; + domain->sig.dif.bg_type =IB_T10DIF_CRC ; + domain->sig.dif.pi_interval =1 << bi->interval_exp; + domain->sig.dif.ref_tag =le32_to_cpu (cmd->rw.reftag); + +/* +* At the moment we hard code those, but in the future +* we will take them from cmd. +*/ + domain->sig.dif.apptag_check_mask =0xffff ; + domain->sig.dif.app_escape = true; + domain->sig.dif.ref_escape = true; +if (ns->prot_type !=NVME_NS_DPS_PI_TYPE3 ) + domain->sig.dif.ref_remap = true; +} + +static void nvmet_rdma_set_sig_attrs (struct nvmet_req *req, +struct ib_sig_attrs *sig_attrs) +{ +struct nvme_command *cmd = req->cmd; +u16 control =le16_to_cpu (cmd->rw.control); + +memset (sig_attrs,0 ,sizeof (*sig_attrs)); + +if (control &NVME_RW_PRINFO_PRACT ) { +/* for WRITE_INSERT/READ_STRIP no wire domain */ + sig_attrs->wire.sig_type =IB_SIG_TYPE_NONE ; +nvmet_rdma_set_diff_domain (cmd, &sig_attrs->mem, req->ns); +/* Clear the PRACT bit since HCA will generate/verify the PI */ + control &= ~NVME_RW_PRINFO_PRACT ; + cmd->rw.control =cpu_to_le16 (control); +/* PI is added by the HW */ + req->transfer_len += req->prot_len; + }else { +/* for WRITE_PASS/READ_PASS both wire/memory domains exist */ +nvmet_rdma_set_diff_domain (cmd, &sig_attrs->wire, req->ns); +nvmet_rdma_set_diff_domain (cmd, &sig_attrs->mem, req->ns); + } + +if (le16_to_cpu (cmd->rw.control) &NVME_RW_PRINFO_PRCHK_REF ) + sig_attrs->check_mask |=IB_SIG_CHECK_REFTAG ; +if (le16_to_cpu (cmd->rw.control) &NVME_RW_PRINFO_PRCHK_GUARD ) + sig_attrs->check_mask |=IB_SIG_CHECK_GUARD ; +if (le16_to_cpu (cmd->rw.control) &NVME_RW_PRINFO_PRCHK_APP ) + sig_attrs->check_mask |=IB_SIG_CHECK_APPTAG ; +} +#else +static u16 nvmet_rdma_check_pi_status (struct ib_mr *sig_mr) +{ +return 0 ; +} + +static void nvmet_rdma_set_sig_attrs (struct nvmet_req *req, +struct ib_sig_attrs *sig_attrs) +{ +} +#endif /* CONFIG_BLK_DEV_INTEGRITY */ + +static int nvmet_rdma_rw_ctx_init (struct nvmet_rdma_rsp *rsp,u64 addr,u32 key, +struct ib_sig_attrs *sig_attrs) +{ +struct rdma_cm_id *cm_id = rsp->queue->cm_id; +struct nvmet_req *req = &rsp->req; +int ret; + +if (req->use_pi) + ret =rdma_rw_ctx_signature_init (&rsp->rw, cm_id->qp, + cm_id->port_num, req->sg, req->sg_cnt, req->prot_sg, + req->prot_sg_cnt, sig_attrs, addr, key, +nvmet_data_dir (req)); +else + ret =rdma_rw_ctx_init (&rsp->rw, cm_id->qp, cm_id->port_num, + req->sg, req->sg_cnt,0 , addr, key, +nvmet_data_dir (req)); + +return ret; +} + +static void nvmet_rdma_rw_ctx_destroy (struct nvmet_rdma_rsp *rsp) +{ +struct rdma_cm_id *cm_id = rsp->queue->cm_id; +struct nvmet_req *req = &rsp->req; + +if (req->use_pi) +rdma_rw_ctx_destroy_signature (&rsp->rw, cm_id->qp, + cm_id->port_num, req->sg, req->sg_cnt, req->prot_sg, + req->prot_sg_cnt,nvmet_data_dir (req)); +else +rdma_rw_ctx_destroy (&rsp->rw, cm_id->qp, cm_id->port_num, + req->sg, req->sg_cnt,nvmet_data_dir (req)); +}static void nvmet_rdma_release_rsp (struct nvmet_rdma_rsp *rsp) {
@@ -502,11 +642,8 @@ static void nvmet_rdma_release_rsp(struct nvmet_rdma_rsp *rsp)atomic_add (1 + rsp->n_rdma, &queue->sq_wr_avail); -if (rsp->n_rdma) { -rdma_rw_ctx_destroy (&rsp->rw, queue->cm_id->qp, - queue->cm_id->port_num, rsp->req.sg, - rsp->req.sg_cnt,nvmet_data_dir (&rsp->req)); - } +if (rsp->n_rdma) +nvmet_rdma_rw_ctx_destroy (rsp);if (rsp->req.sg != rsp->cmd->inline_sg)nvmet_req_free_sgl (&rsp->req);
@@ -561,11 +698,16 @@ static void nvmet_rdma_queue_response(struct nvmet_req *req) rsp->send_wr.opcode =IB_WR_SEND ; } -if (nvmet_rdma_need_data_out (rsp)) - first_wr =rdma_rw_ctx_wrs (&rsp->rw, cm_id->qp, - cm_id->port_num,NULL , &rsp->send_wr); -else +if (nvmet_rdma_need_data_out (rsp)) { +if (rsp->req.use_pi) + first_wr =rdma_rw_ctx_wrs (&rsp->rw, cm_id->qp, + cm_id->port_num, &rsp->write_cqe,NULL ); +else + first_wr =rdma_rw_ctx_wrs (&rsp->rw, cm_id->qp, + cm_id->port_num,NULL , &rsp->send_wr); + }else { first_wr = &rsp->send_wr; + }nvmet_rdma_post_recv (rsp->queue->dev, rsp->cmd);
@@ -584,15 +726,14 @@ static void nvmet_rdma_read_data_done(struct ib_cq *cq, struct ib_wc *wc)struct nvmet_rdma_rsp *rsp = container_of(wc->wr_cqe, struct nvmet_rdma_rsp, read_cqe);struct nvmet_rdma_queue *queue = cq->cq_context; +u16 status =0 ;WARN_ON (rsp->n_rdma <=0 );atomic_add (rsp->n_rdma, &queue->sq_wr_avail); -rdma_rw_ctx_destroy (&rsp->rw, queue->cm_id->qp, - queue->cm_id->port_num, rsp->req.sg, - rsp->req.sg_cnt,nvmet_data_dir (&rsp->req)); rsp->n_rdma =0 ;if (unlikely (wc->status !=IB_WC_SUCCESS )) { +nvmet_rdma_rw_ctx_destroy (rsp);nvmet_req_uninit (&rsp->req);nvmet_rdma_release_rsp (rsp);if (wc->status !=IB_WC_WR_FLUSH_ERR ) {
@@ -603,9 +744,59 @@ static void nvmet_rdma_read_data_done(struct ib_cq *cq, struct ib_wc *wc)return ; } - rsp->req.execute (&rsp->req); +if (rsp->req.use_pi) + status =nvmet_rdma_check_pi_status (rsp->rw.reg->mr); +nvmet_rdma_rw_ctx_destroy (rsp); + +if (unlikely (status)) +nvmet_req_complete (&rsp->req, status); +else + rsp->req.execute (&rsp->req); } +#ifdef CONFIG_BLK_DEV_INTEGRITY +static void nvmet_rdma_write_data_done (struct ib_cq *cq,struct ib_wc *wc) +{ +struct nvmet_rdma_rsp *rsp = + container_of(wc->wr_cqe, struct nvmet_rdma_rsp, write_cqe); +struct nvmet_rdma_queue *queue = cq->cq_context; +struct rdma_cm_id *cm_id = rsp->queue->cm_id; +u16 status; + +WARN_ON (rsp->n_rdma <=0 ); +atomic_add (rsp->n_rdma, &queue->sq_wr_avail); + rsp->n_rdma =0 ; + +if (unlikely (wc->status !=IB_WC_SUCCESS )) { +nvmet_rdma_rw_ctx_destroy (rsp); +nvmet_req_uninit (&rsp->req); +nvmet_rdma_release_rsp (rsp); +if (wc->status !=IB_WC_WR_FLUSH_ERR ) { +pr_info ("RDMA WRITE for CQE 0x%p failed with status %s (%d).\n" , + wc->wr_cqe,ib_wc_status_msg (wc->status), + wc->status); +nvmet_rdma_error_comp (queue); + } +return ; + } + +/* +* Upon RDMA completion check the signature status +* - if succeeded send good NVMe response +* - if failed send bad NVMe response with appropriate error +*/ + status =nvmet_rdma_check_pi_status (rsp->rw.reg->mr); +if (unlikely (status)) + rsp->req.cqe->status =cpu_to_le16 (status <<1 ); +nvmet_rdma_rw_ctx_destroy (rsp); + +if (unlikely (ib_post_send (cm_id->qp, &rsp->send_wr,NULL ))) { +pr_err ("sending cmd response failed\n" ); +nvmet_rdma_release_rsp (rsp); + } +} +#endif /* CONFIG_BLK_DEV_INTEGRITY */ +static void nvmet_rdma_use_inline_sg (struct nvmet_rdma_rsp *rsp,u32 len,u64 off) {
@@ -660,9 +851,9 @@ static u16 nvmet_rdma_map_sgl_inline(struct nvmet_rdma_rsp *rsp)static u16 nvmet_rdma_map_sgl_keyed (struct nvmet_rdma_rsp *rsp,struct nvme_keyed_sgl_desc *sgl,bool invalidate) { -struct rdma_cm_id *cm_id = rsp->queue->cm_id;u64 addr =le64_to_cpu (sgl->addr);u32 key =get_unaligned_le32 (sgl->key); +struct ib_sig_attrs sig_attrs;int ret; rsp->req.transfer_len =get_unaligned_le24 (sgl->length);
@@ -671,13 +862,14 @@ static u16 nvmet_rdma_map_sgl_keyed(struct nvmet_rdma_rsp *rsp,if (!rsp->req.transfer_len)return 0 ; +if (rsp->req.use_pi) +nvmet_rdma_set_sig_attrs (&rsp->req, &sig_attrs); + ret =nvmet_req_alloc_sgl (&rsp->req);if (unlikely (ret <0 )) goto error_out; - ret =rdma_rw_ctx_init (&rsp->rw, cm_id->qp, cm_id->port_num, - rsp->req.sg, rsp->req.sg_cnt,0 , addr, key, -nvmet_data_dir (&rsp->req)); + ret =nvmet_rdma_rw_ctx_init (rsp, addr, key, &sig_attrs);if (unlikely (ret <0 )) goto error_out; rsp->n_rdma += ret;
@@ -956,6 +1148,9 @@ nvmet_rdma_find_get_device(struct rdma_cm_id *cm_id) goto out_free_pd; } + port->pi_capable = ndev->device->attrs.device_cap_flags & +IB_DEVICE_INTEGRITY_HANDOVER ? true : false; +list_add (&ndev->entry, &device_list); out_unlock:mutex_unlock (&device_list_mutex);
@@ -1020,6 +1215,10 @@ static int nvmet_rdma_create_queue_ib(struct nvmet_rdma_queue *queue) qp_attr.cap.max_recv_sge =1 + ndev->inline_page_count; } +if (queue->port->pi_capable && queue->port->pi_enable && + queue->host_qid) + qp_attr.create_flags |=IB_QP_CREATE_INTEGRITY_EN ; + ret =rdma_create_qp (queue->cm_id, ndev->pd, &qp_attr);if (ret) {pr_err ("failed to create_qp ret= %d\n" , ret);
@@ -1164,6 +1363,7 @@ nvmet_rdma_alloc_queue(struct nvmet_rdma_device *ndev,INIT_WORK (&queue->release_work, nvmet_rdma_release_queue_work); queue->dev = ndev; queue->cm_id = cm_id; + queue->port = cm_id->context;spin_lock_init (&queue->state_lock); queue->state =NVMET_RDMA_Q_CONNECTING ;
@@ -1282,7 +1482,6 @@ static int nvmet_rdma_queue_connect(struct rdma_cm_id *cm_id, ret = -ENOMEM ; goto put_device; } - queue->port = cm_id->context;if (queue->host_qid ==0 ) {/* Let inflight controller teardown complete */
Signature · 7 lines
2.16.3 _______________________________________________ linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme