Re: [PATCH v6 6/7] mm: fix MADV_[FREE|DONTNEED] TLB flush miss problem

[Peter Zijlstra]

On Sun, Aug 13, 2017 at 06:14:21AM +0000, Nadav Amit wrote:
> Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> 
> > On Tue, Aug 01, 2017 at 05:08:17PM -0700, Nadav Amit wrote:
> >> void tlb_finish_mmu(struct mmu_gather *tlb,
> >> 		unsigned long start, unsigned long end)
> >> {
> >> -	arch_tlb_finish_mmu(tlb, start, end);
> >> +	/*
> >> +	 * If there are parallel threads are doing PTE changes on same range
> >> +	 * under non-exclusive lock(e.g., mmap_sem read-side) but defer TLB
> >> +	 * flush by batching, a thread has stable TLB entry can fail to flush
> >> +	 * the TLB by observing pte_none|!pte_dirty, for example so flush TLB
> >> +	 * forcefully if we detect parallel PTE batching threads.
> >> +	 */
> >> +	bool force = mm_tlb_flush_nested(tlb->mm);
> >> +
> >> +	arch_tlb_finish_mmu(tlb, start, end, force);
> >> }
> > 
> > I don't understand the comment nor the ordering. What guarantees we see
> > the increment if we need to?
> 
> The comment regards the problem that is described in the change-log, and a
> long thread that is referenced in it. So the question is whether âI donât
> understandâ means âI donât understandâ or âit is not clear enoughâ. Iâll
> be glad to address either one - just say which.

I only read the comment, that _should_ be sufficient. Comments that rely
on Changelogs and random threads are useless.

The comment on its own simply doesn't make sense.

> As for the ordering - I tried to clarify it in the thread of the commit. Let
> me know if it is clear now.

Yeah, I'll do a new patch because if it only cares about _the_ PTL, we
can do away with that extra smp_mb__after_atomic().