On Wed, 2023-06-21 at 09:43 +0200, Peter Zijlstra wrote:

On Tue, Jun 20, 2023 at 05:46:16PM +0300, Yair Podemsky wrote:

quoted

Currently the tlb_remove_table_smp_sync IPI is sent to all CPUs
indiscriminately, this causes unnecessary work and delays notable
in
real-time use-cases and isolated cpus.
By limiting the IPI to only be sent to cpus referencing the
effected
mm.
a config to differentiate architectures that support mm_cpumask
from
those that don't will allow safe usage of this feature.

changes from -v1:
- Previous version included a patch to only send the IPI to CPU's
with
context_tracking in the kernel space, this was removed due to race 
condition concerns.
- for archs that do not maintain mm_cpumask the mask used should be
 cpu_online_mask (Peter Zijlstra).
 

Would it not be much better to fix the root cause? As per the last
time,
there's patches that cure the thp abuse of this.

Hi Peter,
Thanks for your reply.
There are two code paths leading to this IPI, one is the thp,
But the other is the failure to allocate page in tlb_remove_table,
It is the the second path that we are most interested in as it was
found
to cause interference in a real time process for a client (That system
did
 not have thp).
So while curing thp abuses is a good thing, it will not unfortunately
solve
our root cause.
If you have any idea of how to remove the tlb_remove_table_sync_one()
usage
in the tlb_remove_table()->tlb_remove_table_one() call path -- the
usage 
that's relevant for us -- that would be great. As long as we can't
remove
that, I'm afraid all we can do is optimize for it to not broadcast an
IPI
to all CPUs in the system, as done in this patch.
Thanks,
Yair