On Fri, Jul 27, 2018 at 05:14:18PM +0000, Bart Van Assche wrote:

On Fri, 2018-07-27 at 11:04 -0600, Keith Busch wrote:

quoted

On Fri, Jul 27, 2018 at 04:59:34PM +0000, Bart Van Assche wrote:

quoted

On Fri, 2018-07-27 at 10:57 -0600, Keith Busch wrote:

quoted

You skip that code if the driver returns BLK_EH_DONT_RESET_TIMER.

How about applying the following patch on top of this series?

That works for me if you, but it breaks scsi again when
scmd_eh_abort_handler completes the command a second time.

How about introducing a new request queue flag that chooses between the
behavior with or without the patch in my previous e-mail? I don't think that
it is possible to come up with a single implementation that covers the needs
of NVMe and SCSI without introducing such a flag. If a SCSI request times out
then request ownership is transferred from the LLD to the error handler. For
the NVMe driver however there is no such transfer of ownership.

Instead of PATCH 1/5, how about creating a new timeout return code like
"BLK_EH_DONT_COMPLETE"?