On 08/23/2017 05:57 PM, Benjamin Block wrote:

Hello all,

This is the second try for fixing the regression in the BSG-interface that
exists since v4.11 (for more infos see the first series).

I separated my other changes from the bug-fix so that it is easier to apply
if judged good. I will rebase my cleanups I sent in v1 and send them when I
get a bit more time. But the regression-fix is more important, so here's
that.

I did some more tests on it than on v1, including some heavy parallel I/O
on the same blk-queue using both BSG and the normal SCSI-stack at the same
time (throwing some intentional bad commands in it too). That seemed to
work all well enough - i.e. it didn't crash and got the expected results. I
haven't done any external error-inject, but IMO that would be beyond the
scope right now.

The fix is based on Christoph's idea, I discussed this with him off-list
already.

I rebased the series on Jens' for-next.

Added for 4.13, thanks.

-- 
Jens Axboe