Messages in this thread | | | Subject | Re: [RFC 05/13] x86/mm: Add barriers and document switch_mm-vs-flush synchronization | From | Nadav Amit <> | Date | Thu, 9 Jun 2016 12:45:55 -0700 |
| |
Andy Lutomirski <luto@amacapital.net> wrote:
> On Fri, Jun 3, 2016 at 10:42 AM, Nadav Amit <nadav.amit@gmail.com> wrote: >> Following this patch, if (current->active_mm != mm), flush_tlb_page() still >> doesn’t call smp_mb() before checking mm_cpumask(mm). >> >> In contrast, flush_tlb_mm_range() does call smp_mb(). >> >> Is there a reason for this discrepancy? > > Not that I can remember. Is the remote flush case likely to be racy?
You replied separately on another email that included a patch to fix this case. It turns out smp_mb is not needed on flush_tlb_page, since the PTE is always updated using an atomic operation. Yet, a compiler barrier is still needed, so I added smp_mb__after_atomic instead.
Nadav
| |