Re: [Bugfix 0/2] Fix bugs caused by "use irqdomain to dynamically allocate IRQ for IOAPIC"

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Wed, Aug 06, 2014 at 06:27:49PM +0800, Jiang Liu wrote:
> 
> 
> On 2014/8/6 1:58, Konrad Rzeszutek Wilk wrote:
> > On Wed, Aug 06, 2014 at 12:07:18AM +0800, Jiang Liu wrote:
> >>
> >>
> >> On 2014/8/5 21:04, Konrad Rzeszutek Wilk wrote:
> >>> On Tue, Aug 05, 2014 at 11:26:16AM +0800, Jiang Liu wrote:
> >>>> Two issues have been reported against patch set "use irqdomain to
> >>>> dynamically allocate IRQ for IOAPIC" at https://lkml.org/lkml/2014/6/9/44.
> >>>>
> >>>> This first one causes failure of suspend/hibernation, please refer to
> >>>> https://lkml.org/lkml/2014/7/28/822 for more information. And we have
> >>>> worked out a patch to fix it (https://lkml.org/lkml/2014/7/30/725) and
> >>>> Borislav has tested it. But with more testing and analysis, I found the
> >>>> provided patch still has some issues:
> >>>> 1) It may cause regression to Xen
> >>>
> >>> Could you elaborate please?
> >>>
> >>> Is there a git tree with all of these patches to test it?
> >> Hi Konrad,
> >> 	The patch at https://lkml.org/lkml/2014/7/30/725 skips invoking
> >> xen_pcifront_enable_irq() on resume from suspend or restore from
> >> hibernation. I'm not sure whether that will affect suspend/hibernation
> >> with Xen. This patch series won't affect Xen anymore.
> > 
> > Ah, it looks like:
> > 
> > 415         pcibios_enable_irq = xen_pcifront_enable_irq;                           
> > 416         pcibios_disable_irq = NULL;                                             
> > 
> > Is the culprit right? If there was an pcibios_disable_irq set, then said
> > patch would not been needed?
> > 
> > 
> >> 	I have prepared a tree for you at
> >> https://github.com/jiangliu/linux.git suspend
> > 
> > Thank you. Will give it a spin tomorrow.
> Hi Konrad,
> 	Thanks for review.
> 	I think no need to test it anymore. We are trying to fix an
> issue caused by IOAPIC related work. The previous version of patch
> is too coarse grain and may affect Xen. And now we have reworked it,
> so it shouldn't affect Xen anymore.

OK. Thanks for the heads up!
> Regards!
> Gerry
> 
> >>
> >> Thanks for help!
> >> Gerry
> >>
> >>>> 2) Flag dev->dev.power.is_prepared has already been cleared when
> >>>>    pcibios_enable_device() gets called, so it will cause IOAPIC pin
> >>>>    reference count leak.
> >>>>
> >>>> So I reworked the patch to fix above issues. The first patch fixes issue
> >>>> 1 by moving check of dev->dev.power.is_prepared pcibios_enable_irq, so
> >>>> it won't affect Xen. The second patch fixes the IOAPIC pin reference
> >>>> count leakage issue. It also solves the issue we have discussed at
> >>>> http://www.spinics.net/lists/linux-pci/msg32902.html
> >>>>
> >>>> Regards!
> >>>> Gerry
> >>>>
> >>>> Jiang Liu (2):
> >>>>   x86, irq, PCI: Keep IRQ assignment for PCI devices during
> >>>>     suspend/hibernation
> >>>>   x86, irq: Keep balance of IOAPIC pin reference count
> >>>>
> >>>>  arch/x86/pci/intel_mid_pci.c |    9 ++++++++-
> >>>>  arch/x86/pci/irq.c           |    8 +++++++-
> >>>>  drivers/acpi/pci_irq.c       |   15 +++++++++++++--
> >>>>  include/linux/pci.h          |    1 +
> >>>>  4 files changed, 29 insertions(+), 4 deletions(-)
> >>>>
> >>>> -- 
> >>>> 1.7.10.4
> >>>>