On 10/20/2021 3:38 AM, Max Gurtovoy wrote:

This structure is duplicated for RDMA/TCP fabric controllers. Move it to
common code.

FC controllers might use this attribute in the future instead of a local
ioerr_work attribute.

Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Israel Rukshin <redacted>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Max Gurtovoy <mgurtovoy@nvidia.com>
---
  drivers/nvme/host/nvme.h |  1 +
  drivers/nvme/host/rdma.c |  9 ++++-----
  drivers/nvme/host/tcp.c  | 12 +++++-------
  3 files changed, 10 insertions(+), 12 deletions(-)

This looks fine for commonizing rdma and tcp.

However, when I look at the calling sequences, such as 
rdma_error_recovery_work(), I do think rdma/tcp should look a lot like 
fc with it resulting in a call to nvme_reset_ctrl as that's what it 
really is. Gets rid of yet-another-independent-copy of stopping keep 
alive, queue teardown, queue state change, etc.  Just use the reset routine.

-- james