On 09/05/2012 06:07 PM, Michael S. Tsirkin wrote: > On Wed, Sep 05, 2012 at 05:46:17PM +0300, Avi Kivity wrote: >> On 08/21/2012 10:29 PM, Alex Williamson wrote: >> > KVM_IRQFD currently uses the reserved KVM_USERSPACE_IRQ_SOURCE_ID >> > which is also shared with userspace injection methods like >> > KVM_IRQ_LINE. This can cause a conflict if an irqfd triggers on >> > a GSI asserted through KVM_IRQ_LINE. Move irqfd to it's own >> > reserved IRQ source ID. Add a capability for userspace to test >> > for this fix. >> >> I don't think we need a cap, rather a backport if we identify real cases >> where an edge gsi is shared among several devices. Otherwise it is just >> a theoretical bug before level irqfd is introduced. > > In that case, I think it's safer to preserve the "bug" as is: we are > changing userspace-visible behaviour for edge interrupts otherwise. > For example if userspace uses kvm_irq_line for an edge > interrupt, set it to 1, previously it could then > continue to send any number of interrupts with irqfd, > now it can't. If anyone did that, they should have reported a bug, since they surely didn't expect edges if the line was held high. > > Basically the logical OR functionality of source IDs > does not make sense for edge. Edge is only interpreted at the ioapic or pic input; the line is just a line (an open collector line that ORs anything connected to it, or an equivalent). > > How about we do > if (flags&RESAMPLE) > source_id = USERSPACE > else > source_id = IRQFD Okay if we identify something that depends on the bug, otherwise not. -- error compiling committee.c: too many arguments to function -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
