Re: hangs after merge window changes

[Jakub Kicinski <kuba@kernel.org>]

On Thu, 18 Jul 2024 05:45:18 +0900 Dominique Martinet wrote:
> Eric Van Hensbergen wrote on Wed, Jul 17, 2024 at 01:01:26PM -0500:
> > I held way back on adding anything new over the past cycle, so this
> > sounds likely some external change maybe perturbing an older issue.
> > What guest side mount options, qemu options and qemu version are you
> > using?  
> 
> Mount option if any would be much appreciated, yes.
> Also, what version you upgraded from to limit the limit the search a
> bit; the previous 6.9 tree?

Between networking pull requests. Basically
528dd46d0fc35c0176257a13a27d41e44fcc6cb3 to 
51835949dda3783d4639cfa74ce13a3c9829de00

> Also couldn't find much infos about the CI in general, given the traces
> I don't think the workload matters all that much but it can give
> something to start with e.g. running from initrd and mounting 9p or
> using 9p as rootfs; that shouldn't make a difference but might as well
> try to reproduce from something close to the actual netdev CI given the
> low reproducibility.

What CI does is described here:
https://github.com/linux-netdev/nipa/wiki/How-to-run-netdev-selftests-CI-style
it's fairly simple, virtme-ng just runs kselftests inside a VM with 
the host FS mounted via 9p-virtio.

I'd offer running experiments on the exact systems, but this week 
I'm partially AFK (at a conference).

> > Dominique, you aware of anything coming in that might have messed with
> > this?  Maybe the new common code refactoring underpinning
> > readahead/etc.?  
> 
> Given the trace it's probably stuck in __wait_on_freeing_inode (the only
> schedule() call in find_inode_fast), so we could be looking at
> fallbacks from the iget changes but that made it in 6.9 already so
> that's unlikely...
> 
> I have a feeling that adding cache will make the error go away if you
> just want to fix tests for now, so would be great if you could try
> adding -o cache=loose to the 9p mount options; hopefully the inodes
> won't be freed as often and that'll make the race harder to hit.

Poked that into where I think virtme-ng does the mounting, let's see.

> I had a quick look at other globals fs changes and nothing stood out,
> it's going to be fun to track down.
> 
> > On Wed, Jul 17, 2024 at 12:24 PM Jakub Kicinski <kuba@kernel.org> wrote:  
> > > Apologies for possible duplicate, I tried to search lore and haven't
> > > found anything, but also I can't find the exact 9p archive there :S  
> 
> The new list is here: https://lore.kernel.org/v9fs/
> You probably didn't miss anything :)
>