Re: [PATCH] mm/mmu_gather: send tlb_remove_table_smp_sync IPI only to MM CPUs

From: ypodemsk@redhat.com
To: Will Deacon <will@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>,
	aneesh.kumar@linux.ibm.com, akpm@linux-foundation.org,
	npiggin@gmail.com, arnd@arndb.de, linux-arch@vger.kernel.org,
	linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	mtosatti@redhat.com, ppandit@redhat.com, alougovs@redhat.com,
	David Hildenbrand <david@redhat.com>
Subject: Re: [PATCH] mm/mmu_gather: send tlb_remove_table_smp_sync IPI only to MM CPUs
Date: Sun, 26 Mar 2023 10:40:08 +0300	[thread overview]
Message-ID: <b88b37de44222a67bc5cf22798e84fb178b58725.camel@redhat.com> (raw)
In-Reply-To: <20230324151259.GC27199@willie-the-truck>

On Fri, 2023-03-24 at 15:13 +0000, Will Deacon wrote:
> On Wed, Mar 22, 2023 at 04:11:44PM +0200, ypodemsk@redhat.com wrote:
> > On Mon, 2023-03-20 at 09:49 +0100, Peter Zijlstra wrote:
> > > On Sun, Mar 12, 2023 at 10:09:45AM +0200, Yair Podemsky wrote:
> > > > 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, this patch will limit
> > > > this
> > > > IPI to
> > > > only be sent to cpus referencing the effected mm and are
> > > > currently
> > > > in
> > > > kernel space.
> > > 
> > > Did you validate that all architectures for which this is
> > > relevant
> > > actually set bits in mm_cpumask() ?
> > > 
> > Hi Peter,
> > Thank you for bringing this to my attention.
> > I reviewed the architectures using the MMU_GATHER_RCU_TABLE_FREE:
> > arm, powerpc, s390, sparc and x86 set the bit when switching
> > process
> > in.
> > for arm64 removed set/clear bit in 38d96287504a ("arm64: mm: kill
> > mm_cpumask usage")
> > The reason given was that mm_cpumask was not used.
> > Given that we now have a use for it, I will add a patch to revert.
> 
> Maintaining the mask is also not free, so I'm not keen on adding it
> back
> unless there's a net win.
> 
> Will
> 
How about adding a Kconfig to mark which architectures set/use the
mm_cpumask?
This will allow us to use the mm_cpumask on architectures that support
it, and use acpu_online_mask on those that don't.
Also make it clear which architectures set the bit for the future.
Yair