On 11/19/2019 5:28 AM, Daniel Wagner wrote:
On Tue, Nov 12, 2019 at 10:15:00PM -0500, Martin K. Petersen wrote:
While trying to understand what's going on in the Oops below I figured
that it could be the result of the invalid pointer access. The patch
still needs testing by our customer but indepent of this I think the
patch fixes a real bug.
I was able to reproduce the same stack trace with this patch
applied... That is obviously bad. The good news, I have access to this
machine, so maybe I able to figure out what's the root cause of this
crash.
fyi - Dick and I were taking our time reviewing your patch as the major
concern we had with it was that the splice and then the re-add of the
work-list could have an abort complete before the IO, which could lead
to DC.
We'll look anew at your repro.
-- james
