Re: SCSI hotplug issues with UEFI VM with guest kernel >= 6.5

[Igor Mammedov <imammedo@redhat.com>]

On Tue, 12 Dec 2023 16:25:29 +0100
Igor Mammedov <imammedo@redhat.com> wrote:

> On Tue, 12 Dec 2023 13:50:20 +0100
> Fiona Ebner <f.ebner@proxmox.com> wrote:
> 
> > Am 12.12.23 um 12:26 schrieb Igor Mammedov:  
> > > 
> > > it's not necessary, but it would help to find out what's going wrong faster.
> > > Otherwise we would need to fallback to debugging over email.
> > > Are you willing to help with testing/providing debug logs to track down
> > > the cause?.
> > >     
> > 
> > I submitted the dmesg logs in bugzilla:
> > https://bugzilla.kernel.org/show_bug.cgi?id=218255
> >   
> > > Though debug over email would be slow, so our best option is to revert
> > > offending patches until the cause if found/fixed.
> > >     
> > >>>>> Do you have to revert both cc22522fd55e2 and 40613da52b13f to make it
> > >>>>> work reliably?  If we have to revert something, reverting one would be
> > >>>>> better than reverting both.        
> > >>>>      
> > >>
> > >> Just reverting cc22522fd55e2 is not enough (and cc22522fd55e2 fixes
> > >> 40613da52b13f so I can't revert just 40613da52b13f).    
> > > 
> > > With UEFI setup, it still works for me fine with current master.
> > > 
> > > Kernel 6.7.0-rc5-00014-g26aff849438c on an x86_64 (ttyS0)
> > >     
> > 
> > I also built from current master (still 26aff849438c) to verify and it's
> > still broken for me.
> >   
> > > 
> > > it still doesn't work with Fedora's 6.7.0-0.rc2.20231125git0f5cc96c367f.26.fc40.x86_64 kernel.
> > > However it's necessary to have -smp 4 for it to break,
> > > with -smp 1 it works fine as well.
> > >     
> > 
> > For me it's (always with build from current master):
> > 
> > -smp 1 -> it worked 5 times out of 5
> > -smp 2 -> it worked 3 times out of 5
> > -smp 4 -> it worked 0 times out of 5
> > -smp 8 -> it worked 0 times out of 5  
> 
> 
> I managed to reproduce it with upstream using fedora 40 config as is
> (without converting it to mod2yesconfig).
> So give me a couple of days to debug it before reverting.

Actually here is another report + analysis explaining where the race is happening:
https://www.spinics.net/lists/kernel/msg5033061.html

That's the reason why my minimal config worked
(based on defconfig where CONFIG_SCSI_SCAN_ASYNC in not enabled by default for x86)

While distros (at least some) do enable it.

> 
> > 
> > Best Regards,
> > Fiona
> >   
>