Re: [PATCH] bcache: add separate workqueue for journal_write to avoid deadlock

[Eddie Chapman]

On 28/09/18 03:32, Coly Li wrote:
> On 9/27/18 11:53 PM, Eddie Chapman wrote:
> > On 27/09/18 16:23, Coly Li wrote:
> > > On 9/27/18 9:45 PM, guoju wrote:
> > > > After write SSD completed, bcache schedule journal_write work to
> > > > system_wq, that is a public workqueue in system, without WQ_MEM_RECLAIM>>>> flag. system_wq is also a bound wq, and there may be no idle kworker on
> > > > current processor. Creating a new kworker may unfortunately need to
> > > > reclaim memory first, by shrinking cache and slab used by vfs, which
> > > > depends on bcache device. That's a deadlock.
> > > >
> > > > This patch create a new workqueue for journal_write with WQ_MEM_RECLAIM
> > > > flag. It's rescuer thread will work to avoid the deadlock.
> > > >
> > > > Signed-off-by: guoju <fangguoju@xxxxxxxxx>
> > >
> > > Nice catch, this fix is quite important. I will try to submit to Jens 
> > > ASAP.
> > >
> > > Thanks.
> > >
> > > Coly Li
> >
> > Once this goes into 4.19, would this be a candidate for backporting to 
> > any stable kernels, or does it only fix something introduced in this 
> > cycle?
> This bug exists in upstream for quite long time, it should be applied to 
> all stable kernels which it can be applied. And it is Cced to 
> stable@xxxxxxxxxxxxxxx already.
>
> Coly Li

Thanks Coly! :-)

Just to let you know, I applied this (and couple of other cherry picks) 
to a couple of 4.14 boxes last night, so far so good, running without 
issues. However, this one needed this recent commit upstream as a 
pre-requisite:

16c1fdf4cfd6c0091e59b93ec2cb7e99973f8244
bcache: do not assign in if condition in bcache_init()

in order to be able to apply it.

This is because the context of the second hunk for 
drivers/md/bcache/super.c (in this journal_write workqueue patch) 
contains code added by that commit 16c1fdf4cfd6c0091e59b93ec2cb7e99973f8244.

So I guess either 16c1fdf4cfd6c0091e59b93ec2cb7e99973f8244 also needs 
tagging for stable, or perhaps a backport of this journal_write 
workqueue will have to be created for earlier kernels, with different 
context for that hunk?

Eddie