[RFC PATCH 2/2] mm/mmu_gather: Avoid multiple page walk cache flush

[Peter Zijlstra]

On Tue, Dec 17, 2019 at 03:45:36PM +0530, Aneesh Kumar K.V wrote:
> On 12/17/19 2:28 PM, Peter Zijlstra wrote:
> > On Tue, Dec 17, 2019 at 12:47:13PM +0530, Aneesh Kumar K.V wrote:
> > > On tlb_finish_mmu() kernel does a tlb flush before  mmu gather table invalidate.
> > > The mmu gather table invalidate depending on kernel config also does another
> > > TLBI. Avoid the later on tlb_finish_mmu().
> > 
> > That is already avoided, if you look at tlb_flush_mmu_tlbonly() it does
> > __tlb_range_reset(), which results in ->end = 0, which then triggers the
> > early exit on the next invocation:
> > 
> > 	if (!tlb->end)
> > 		return;
> > 
> 
> Is that true for tlb->fulmm flush?

Hmm, no, but I'm thinking you patch is broken, even for that case. We
must issue the TLBI before call_rcu().

Perhaps if we replace !tlb->end with something like:

  !tlb->freed_tables && !tlb->cleared_p*

(which GCC should be able to do with a single load and mask)

I've not really thought too hard about it yet, I need to run some
errands, but I'll look at it more closely when I get back.