On 5 July 2018 at 09:46, Luc Michel <luc.mic...@greensocs.com> wrote:
> Yes I encountered the same issue with Xen in SMP (see my cover letter).
> Re-reading the GICv2 specs, it's now clear to me that a read to
> ITARGETSR0 to ITARGETSR7 should return "the number of the processor
> performing the read". Reading the message of commit 2bb3135 in Linux, it
> seems that older versions of the GIC exposed this value in IRQs 29, 30,
> 31, hence the
> if (irq >= 29 && irq <= 31) { res = cm; }
> in the current QEMU implementation.
>
> I should probably add a patch to fix that. I'll have to dig in specs of
> older GIC revisions to see when this behaviour actually appeared.
>
> Maybe I wait for some reviews before sending a new re-roll?
> Peter, any thoughts?
I'm probably not going to have time to look at any of this
GICv2 stuff for a bit (due to softfreeze and other for-3.0
work), so don't wait for my responses if you think a reroll
makes sense.
thanks
-- PMM
