Re: A kernel warning when entering suspend

[Maxim Levitsky <mlevitsk@redhat.com>]

On Mon, 2019-12-23 at 16:52 +0200, Maxim Levitsky wrote:
> On Thu, 2019-04-04 at 13:14 +0200, Thomas Gleixner wrote:
> > On Thu, 4 Apr 2019, Ming Lei wrote:
> > 
> > > On Thu, Apr 04, 2019 at 08:23:59AM +0000, fin4478 fin4478 wrote:
> > > > Hi,
> > > > 
> > > > I do not use suspend/resume but noticed this kernel warning when testing it. This warning is present in earlier kernels too. My system works fine after resume.
> > > > If there is a patch to fix this, I can test it.
> > > > 
> > > > [   53.403033] PM: suspend entry (deep)
> > > > [   53.403034] PM: Syncing filesystems ... done.
> > > > [   53.404775] Freezing user space processes ... (elapsed 0.001 seconds) done.
> > > > [   53.405972] OOM killer disabled.
> > > > [   53.405973] Freezing remaining freezable tasks ... (elapsed 0.001 seconds) done.
> > > > [   53.407036] printk: Suspending console(s) (use no_console_suspend to debug)
> > > > [   53.407491] ACPI Debug:  "RRIO"
> > > > [   53.407505] serial 00:03: disabled
> > > > [   53.407560] r8169 0000:07:00.0 enp7s0: Link is Down
> > > > [   53.415042] sd 5:0:0:0: [sda] Synchronizing SCSI cache
> > > > [   53.415065] sd 5:0:0:0: [sda] Stopping disk
> > > > [   53.428943] WARNING: CPU: 10 PID: 3127 at kernel/irq/chip.c:210 irq_startup+0xd6/0xe0
> > > 
> > > Looks the 'WARN_ON_ONCE(force)' in irq_startup() is a bit too strict.
> > 
> > Why?
> > 
> > > irq_build_affinity_masks() doesn't guarantee that each IRQ's affinity
> > > can include at least one online CPU.
> > 
> > Then why is it tried to start up an interrupt which has no online cpu in
> > the mask? I'm missing context obviously.
> 
> Hi!
> 
> The problem here is that recently an optimization was added to avoid taking a spinlock in
> the interrupt handler in expense of enable_irq/disable_irq around the place where the race can happen.
> 
> The patch is https://patchwork.kernel.org/patch/10708055/
> 
> And the problem here is that indeed if nvme driver allocates more interrupts that number
> of currently online cpus (which happens if number of possible cpus is larger that number of cpus
> the system was booted with), then some of the nvme queues exist but effectively disabled (till relevant cpus
> are plugged which can happen any time).
> 
> I think that this is a valid use case, and since other than that warning, enable_irq in this case doesn't
> really enable it, maybe the warning should be removed?
> 
> We have a bug report about this issue, and I was told that the warning breaks up some QE scripts.
> https://bugzilla.redhat.com/show_bug.cgi?id=1779247
> 

Ping.

Best regards,
	Maxim Levitsky


_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme