[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH 4/4] AMD IOMMU: untie remap and vector maps

On 28/03/13 13:09, Jan Beulich wrote:
On 28.03.13 at 13:37, George Dunlap <George.Dunlap@xxxxxxxxxxxxx> wrote:
On Tue, Mar 26, 2013 at 9:05 AM, Jan Beulich <JBeulich@xxxxxxxx> wrote:
With the specific IRTEs used for an interrupt no longer depending on
the vector, there's no need to tie the remap sharing model to the
vector sharing one.

Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
Unfortunately I'm having a bit of trouble "paging in" the technical
details of why this change was necessary.  I thought that the
limitation had to do with the remapping table itself.  Are you saying
that if interrupt X maps to pcpu 0 vector 5, and interrupt Y maps to
pcpu 2 vector 5, that even if X and Y are in the same table, they will
still go to the right place?  Could you give me a brief explanation of
why that is?  (I don't know what IRTE is, and Google is not helpful.)
Previously the remapping table was indexed by vector (combined
with delivery mode, for whatever reason). That way, a particular
vector, no matter which CPU it would have been meant to go to
caused the same IRTE (Interrupt Remapping Table Entry) to be
selected, and with it the same destination ID.

By not using the vector for indexing anymore, a vector allocated
in CPU A's space connects with IRTE x with destination ID set to
(or including, when in cluster or flat mode) whereas CPU B's
identical vector connects to a different IRTE (even when using a
single global remapping table), properly set to the destination ID
for (or including) that CPU.

OK -- well assuming that to be the case:

Acked-by: George Dunlap <george.dunlap@xxxxxxxxxxxxx>

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.