Le 04/05/2017 à 09:39, Balbir Singh a écrit :
On Wed, 2017-05-03 at 16:29 +0200, Frederic Barrat wrote:
The PSL needs to see all TLBI pertinent to the memory contexts used on
the cxl adapter. For the hash memory model, it was done by making all
TLBIs global as soon as the cxl driver is in us. For radix, we need
something similar, but we can refine and only convert to global the
invalidations for contexts actually used by the device.
So mark the contexts being attached to the cxl adapter as requiring
global TLBIs.
<snip>
+#ifdef CONFIG_PPC_BOOK3S_64
+ if (ctx->mm)
+ mm_context_set_global_tlbi(&ctx->mm->context);
Just curious and wondering
Could we do mm_context_set_global_tlbi() before ->attach_process() that
way we won't need atomic tests (set_bit() and test_bit())? May be a memory
barrier would suffice. Not 100% sure, hence checking
You're right, I need to move mm_context_set_global_tlbi() before the
attach and have a write memory barrier.
If the attach fails, then the context will still be marked for global
TLBIs (some other driver may also set the bit for a different reason).
But I would expect the life expectancy of a process designed to use an
accelerator and failing to attach to be pretty short.
I still think we need the atomic set_bit() though, but that's really for
the future and if somebody introduces new bits in the context 'flags'.
Fred
Balbir Singh.
