[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Xen-devel] Re: [PATCH] IRQ: manually EOI migrating line interrupts
On 30/08/11 15:38, Keir Fraser wrote: > On 30/08/2011 15:28, "Andrew Cooper" <andrew.cooper3@xxxxxxxxxx> wrote: > >>> @@ -1739,6 +1739,14 @@ static void end_level_ioapic_irq (unsign >>> */ >>> i = IO_APIC_VECTOR(irq); >>> >>> + /* Manually EOI the old vector if we are moving to the new */ >>> + if ( vector && i != vector ) >>> + { >>> + int ioapic; >>> + for (ioapic = 0; ioapic < nr_ioapics; ioapic++) >>> + io_apic_eoi(ioapic, i); >>> + } >>> + > I don't know whether it's worth the effort, but we ought to be able to do > better than this and send EOI to exactly the correct IO-APIC. I think > irq=gsi here? And we should know the gsi_base of every IO-APIC, so we can > work out in fact which pin of which IO-APIC needs clobbering? > > -- Keir irq does (or really should) equal gsi. I had not noticed gsi_base and gsi_end when making this fix. io_apic_eoi does not require a pin, but it is using IO-APIC registers which I can not find references to. The Local APIC document implies that you just write the vector to the EOI register, and the IO-APIC will work out which pin to clear. However, because Xen currently might assign the same vector to two pins in the same IO-APIC, changing the code at this point will not fix the problem - just make it rarer. Therefore, I would suggest that it is not worth the effort, as the problem is already very rare, and unlikely to be a problem with any sane hardware which avoids PCI INTx interrupts where possible. P.S. If anyone knows which manual contains the specificaion/programming guide for the IO-APIC, I would be very gratefull. Google always points to 82093AA datasheet which is very out of date. -- Andrew Cooper - Dom0 Kernel Engineer, Citrix XenServer T: +44 (0)1223 225 900, http://www.citrix.com _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |