Re: [RFC PATCH 2/2] zap_pte_range: fix partial TLB flushing in response to a dirty pte

[Linus Torvalds]

On Tue, Oct 28, 2014 at 4:44 AM, Will Deacon <will.deacon@xxxxxxx> wrote:
> @@ -1194,11 +1194,10 @@ again:
>                  * then update the range to be the remaining
>                  * TLB range.
>                  */
> -               old_end = tlb->end;
> -               tlb->end = addr;
> +               tlb->end = old_end = min(tlb->end, addr);
>                 tlb_flush_mmu_tlbonly(tlb);
> -               tlb->start = addr;
> -               tlb->end = old_end;
> +               tlb->start = old_end;
> +               tlb->end = end;

I don't think this is right. Setting "tlb->end = end" looks very wrong
indeed, because "end" here inside zap_pte_range() is *not* the final
end of the zap range, it is just the end of the current set of pte's.

There's a reason the old code *saved* the old end value. You've now
ripped that out, and use the "old_end" for something else entirely.

Your arm64 version of tlb_add_flush() then hides the bug you just
introduced by updating the end range for each page you encounter. But
quite frankly, I think your problems are all fundamental to that very
issue. You're playing games with start/end during the TLB flush
itself, which is not how those things were designed to work.

So now you break everything that *doesn't* do your arm games.

                   Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/