On 02/20/2019 06:16 PM, Ming Lei wrote:

On Thu, Feb 21, 2019 at 2:16 AM Chaitanya Kulkarni
[off-list ref] wrote:

quoted

On 02/20/2019 06:17 AM, Christoph Hellwig wrote:

quoted

We shouldn't be allocating a scatterlist for a command that doesn't
have a payload.

The blk_rq_payload_bytes check in nvme_rdma_map_data is supposed to
prevent that.

Chaitanya, can you try to debug why this is not working?  I'm on
vacation and don't have much time right now unfortunately.

Hi Ming,

Can you please test following patch on your system ?

With this patch, the test can pass.

[root@ktest-04 blktests]# ./check nvmeof-mp/002
nvmeof-mp/002 (File I/O on top of multipath concurrently with logout
and login (mq)) [passed]
     runtime  75.478s  ...  54.317s

quoted

  From 79e178a53a9f101d1f5f6a4923298bb1ffe936ef Mon Sep 17 00:00:00 2001
From: Chaitanya Kulkarni <redacted>
Date: Wed, 20 Feb 2019 09:16:58 -0800
Subject: [PATCH] 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.

Signed-off-by: Chaitanya Kulkarni <redacted>
---
   drivers/nvme/host/rdma.c | 4 ++--
   1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c
index ac365366c2ec..c6a489049fd5 100644
--- a/drivers/nvme/host/rdma.c
+++ b/drivers/nvme/host/rdma.c

@@ -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;
--
2.17.0


Also can you please share all the QEMU config/setup information that
you have used to run the tests ? I want to add this test to the blktests
with QEMU platform.

Just enable the required kernel config options and install
device-mapper-multipath,
then run './check nvmeof-mp/002', it will be started.

Thanks,
Ming Lei

Thanks Ming for testing, I'll send an official patch to the mailing list.