On Wed, 2020-10-07 at 16:05 +0200, Thomas Gleixner wrote:

On Wed, Oct 07 2020 at 14:08, David Woodhouse wrote:

quoted

On 7 October 2020 13:59:00 BST, Thomas Gleixner [off-list ref] wrote:

quoted

On Wed, Oct 07 2020 at 08:48, David Woodhouse wrote:

quoted

To fix *that* case, we really do need the whole series giving us per-
domain restricted affinity, and to use it for those MSIs/IOAPICs that
the IRQ remapping doesn't cover.

Which do not exist today.

Sure. But at patch 10/13 into this particular patch series, it *does*
exist.

As I told you before: Your ordering is wrong. We do not introduce bugs
first and then fix them later ....

I didn't introduce that bug; it's been there for years. Fixing it
properly requires per-irqdomain affinity limits.

There's a cute little TODO at least in the Intel irq-remapping driver,
noting that we should probably check if there are any IOAPICs that
aren't in the scope of any DRHD at all. But that's all.

If it happens, I think we'll end up doing the right thing and
instantiating a non-IR IOAPIC domain for it, and if we *don't* have any
CPU with an APIC ID above... (checks notes)... 7 ... then it'll just
about work out. Over 7 and we're screwed (because logical mode; see
also https://git.infradead.org/users/dwmw2/linux.git/commit/429f0c33f
for a straw man but that's getting even further down the rabbit hole)