Re: [PATCH] x86, irq: Keep IRQ assignment for PCI devices during suspend/hibernation, bisected

[Jiang Liu]

On 2014/8/1 18:56, Borislav Petkov wrote:
> Ok,
> 
> so I went and looked at tip and which branches could be the usual
> suspects. And whaddya know, tip/x86/apic was the first to pick and it
> contained the culprit:
> 
> 6a38fa0e3c94 ("x86, irq, ACPI: Release IOAPIC pin when PCI device is disabled")
Hi Borislav,
	The above commit may cause failure of suspend/hiberrnation.
The reason is:
1) With recent changers, we dynamically allocate irq number for IOAPIC
   pins.
2) The allocated irq will be released when suspending/hibernating.
	pci_disable_device()->pcibios_disable_irq()
3) When resuming, a different irq may be assigned to the PCI device.
	pci_enable_device()->pcibios_enable_irq()
4) Now the hardware will deliver interrupt to the new allocated irq
   but the interrupt handler is still registered on old irq, so it
   breaks the driver and causes failure of hibernation.

The patch sent out by me is to fix this issue by keeping the allocated
irq when suspend/hibernate. And seems it works as expected.

But I still don't know why it causes IOMMU related warnings as:
AMD-Vi: Event logged [IO_PAGE_FAULT device=00:12.0 domain=0x0009
address=0x0000000000000000 flags=0x0000]

Regards!
Gerry
> 
> It seems like my machine doesn't like releasing the IOAPIC pin. Or,
> maybe it is IOMMU-related and the IOMMU doesn't like this releasing of
> the pin *after* the device gets suspended and *before* the IOMMU.
> 
> Strange.
> 
> Oh well, a well deserved lunch now :)
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/