On Thu, Jul 22, 2021 at 08:06:47AM +0100, Christoph Hellwig wrote:

Due the combination of the code move and the actual change I had
a really hard time reviewing this.

quoted

-		trace_xlog_iclog_syncing(iclog, _RET_IP_);

This tracepoint got lost.

Ah, will fix that.

Otherwise this looks good.

As I had to split the move from the technical changes to follow
the change I might as well share my split, attached below as two
patches:

Yeah, I haven't got to that point yet - I'm still chasing a wild
goose that is now only showing up once every 3 or 4 hours of
cyclic testing, so I've been trying to capture that in sufficient
detail to be able to diagnose the failure. I wanted to get these out
there so new eyes might see what I'm missing or suggest a better
approach...

I'll add this split into my patchset tomorrow when I go looking at
the corpses overnight testing created...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com