Re: [Xen-devel] [PATCH v3 1/7] xen-pciback: Document the various parameters and attributes in SysFS

[Konrad Rzeszutek Wilk]

On Wed, Jul 09, 2014 at 03:57:15PM +0100, David Vrabel wrote:
> On 09/07/14 15:47, Konrad Rzeszutek Wilk wrote:
> > On Wed, Jul 09, 2014 at 03:45:03PM +0100, David Vrabel wrote:
> >> On 09/07/14 15:25, Konrad Rzeszutek Wilk wrote:
> >>> On Wed, Jul 09, 2014 at 03:22:30PM +0100, Andrew Cooper wrote:
> >>>> On 09/07/14 15:13, Konrad Rzeszutek Wilk wrote:
> >>>>> On Wed, Jul 09, 2014 at 03:05:56PM +0100, Andrew Cooper wrote:
> >>>>>> On 09/07/14 14:59, Konrad Rzeszutek Wilk wrote:
> >>>>>>>>> +What:           /sys/bus/pci/drivers/pciback/irq_handler_state
> >>>>>>>>> +Date:           Oct 2011
> >>>>>>>>> +KernelVersion:  3.1
> >>>>>>>>> +Contact:        xen-devel@xxxxxxxxxxxxxxxxxxxx
> >>>>>>>>> +Description:
> >>>>>>>>> +                An option to toggle Xen PCI back to acknowledge (or stop)
> >>>>>>>>> +                interrupts for the specific device regardless of whether the
> >>>>>>>>> +                device is shared, enabled, or on a level interrupt line.
> >>>>>>>>> +                Writing a string of DDDD:BB:DD.F will toggle the state.
> >>>>>>>>> +                This is Domain:Bus:Device.Function where domain is optional.
> >>>>>>>> I do not understand under what circumstances this should be used in.
> >>>>>>> So that dom0 does not disable the IRQ line as it would be getting the IRQs
> >>>>>>> for the guest as well (because the IRQ line is level and another guest
> >>>>>>> uses an PCI device that is using the same line).
> >>>>>> Why is this relevant?  Xen (and Xen alone) actually controls this aspect
> >>>>>> of interrupts.  Xen manages passing line level interrupts to any domain
> >>>>>> which might have a device hanging off a particular line, and has to wait
> >>>>>> until all domains have EOI'd the line until it can clear the interrupt
> >>>>>> at the IO-APIC.
> >>>>> Because Linux will think there is an IRQ storm as the event->IRQ points
> >>>>> to the default one. And then it will mask the event, which means dom0
> >>>>> will mask the PIRQ, and Xen will then also mask the IRQ.
> >>>>
> >>>> Xen will (and by this I mean 'should', and this was the behaviour last
> >>>> time I delved in there) only mask the IRQ if dom0 is the only consumer
> >>>> of these interrupts.
> >>>>
> >>>> For any PCIPassthrough, dom0 will get line interrupts for passed-through
> >>>> devices, but in this case pci-back should always handle the line
> >>>> interrupts so Linux doesn't block them as an IRQ storm.
> >>>
> >>> And that is what it does - and this option provides the option to enable/disable
> >>> it the system admin wishes to do it.
> >>
> >> I still don't understand why someone would want to flip the handler to
> >> a broken mode.
> > 
> > The intent was to allow you to flip to the 'enable' mode in case Linux
> > did not detect it correctly.
> 
> We should not provide sysfs knobs to work around kernel bugs.

It is already in the code. This is just a documentation patch.
But I can drop this chunk from it if you would like.
> 
> >> The original commit isn't very enlightening either.
> > 
> > Thoughts then on what this documentation patch should say to make it
> > clear of its intent?
> 
> I think it should be removed.
> 
> It also has non-standard behaviour of /toggling/ with every write
> instead of using a write of a 1 or a 0 like every other sysfs file.

The 'unbind' and 'bind' expect an BDF as well and they are part
of the SysFS and in documentation.

> 
> David
--
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/