nvme-fc/rdma: use blk_rq_nr_phys_segments
Subsystems (2)
cover nvme-fc/rdma: use blk_rq_nr_phys_segments
Hi,
Here are two patches to fix the issue reported by Ming :-
(https://marc.info/?l=linux-block&m=155071174416909&w=2).
Ming has tested rdma patch. I've added the code changes for the host/fc also but I've *not* tested the 2nd patch due to availability of hardware.
James can you please take a look ?
Regards,
Chaitanya
Chaitanya Kulkarni (2):
nvme-rdma: use nr_phys_segments when map rq to sgl
nvme-fc: use nr_phys_segments when map rq to sgl
drivers/nvme/host/fc.c | 4 ++-- drivers/nvme/host/rdma.c | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-)
Signature · 1 line
2.17.0
PATCH 1/2 nvme-rdma: use nr_phys_segments when map rq to sgl
Use blk_rq_nr_phys_segments() instead of blk_rq_payload_bytes() to check if a command contains data to be mapped. This fixes the case where a struct request contains LBAs, but it has no payload, such as Write Zeroes support.
--- drivers/nvme/host/rdma.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/nvme/host/rdma.c +2 -2
@@ -1150,7 +1150,7 @@ static void nvme_rdma_unmap_data(struct nvme_rdma_queue *queue,struct nvme_rdma_device *dev = queue->device;struct ib_device *ibdev = dev->dev; -if (!blk_rq_payload_bytes (rq)) +if (!blk_rq_nr_phys_segments (rq))return ;if (req->mr) {
@@ -1273,7 +1273,7 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue, c->common.flags |=NVME_CMD_SGL_METABUF ; -if (!blk_rq_payload_bytes (rq)) +if (!blk_rq_nr_phys_segments (rq))return nvme_rdma_set_sg_null (c); req->sg_table.sgl = req->first_sgl;
Signature · 1 line
2.17.0
PATCH 2/2 nvme-fc: use nr_phys_segments when map rq to sgl
In reply to Chaitanya KulkarniUse blk_rq_nr_phys_segments() instead of blk_rq_payload_bytes() to check if a command contains data to be mapped. This fixes the case where a struct request contains LBAs, but it has no payload, such as Write Zeroes support.
--- drivers/nvme/host/fc.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/nvme/host/fc.c +2 -2
@@ -2119,7 +2119,7 @@ nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq, freq->sg_cnt =0 ; -if (!blk_rq_payload_bytes (rq)) +if (!blk_rq_nr_phys_segments (req))return 0 ; freq->sg_table.sgl = freq->first_sgl;
@@ -2316,7 +2316,7 @@ nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx,if (ret)return ret; - data_len =blk_rq_payload_bytes (rq); + data_len =blk_rq_nr_phys_segments (rq);if (data_len) io_dir = ((rq_data_dir (rq) ==WRITE ) ?NVMEFC_FCP_WRITE :NVMEFC_FCP_READ );
Signature · 1 line
2.17.0
PATCH 2/2 nvme-fc: use nr_phys_segments when map rq to sgl
Quoted 15 lines · Chaitanya Kulkarni
Use
blk_rq_nr_phys_segments()instead ofblk_rq_payload_bytes()to check if a command contains data to be mapped. This fixes the case where a struct request contains LBAs, but it has no payload, such as Write Zeroes support.Signed-off-by Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>--- drivers/nvme/host/fc.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)drivers/nvme/host/fc.c +0
-0
drivers/nvme/host/fc.c nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq, · line 2122
Quoted 5 lines
drivers/nvme/host/fc.c +1
-1@@ -2119,7 +2119,7 @@ nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq, freq->sg_cnt =0 ; -if (!blk_rq_payload_bytes (rq)) +if (!blk_rq_nr_phys_segments (req))
This is broken - needs to be "rq" not "req". Not sure how it passed a compile test.
drivers/nvme/host/fc.c nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx, · line 2322
Quoted 12 lines
drivers/nvme/host/fc.c +1
-1return 0 ; freq->sg_table.sgl = freq->first_sgl;@@ -2316,7 +2316,7 @@ nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx,if (ret)return ret; - data_len =blk_rq_payload_bytes (rq); + data_len = (rq);if (data_len) io_dir = ((rq_data_dir (rq) ==WRITE ) ?NVMEFC_FCP_WRITE :NVMEFC_FCP_READ );
This is also broken as data_len is used elsewhere to specify the size of the payload (data outside of sqe) that is to be transferred. So although it was ok for the local check, you switched things that cared about a real length from a bytecount to a segment count. Needless to say it killed things.
So what does blk_rq_payload_bytes(rq) on a Write Zeros command returns ?? Why isn't returning 0 valid for Write Zeros ?
Can you describe what the issue is that prompted this change.
-- james
Christoph Hellwig
maintainercover nvme-fc/rdma: use blk_rq_nr_phys_segments
In reply to Chaitanya KulkarniThanks,
I'm going to pick up patch 1 for the pull request I'm about to send to Jens. 2 looks fine to me, but I'd rather wait for James to review it.