On 06/30/2016 06:50 PM, Nadav Amit wrote:
> Dave Hansen <d...@sr71.net> wrote:
>> +pte_t ptep_clear_flush(struct vm_area_struct *vma, unsigned long address,
>> + pte_t *ptep)
>> +{
>> + struct mm_struct *mm = vma->vm_mm;
>> + pte_t pte;
>> +
>> + pte = ptep_get_and_clear(mm, address, ptep);
>> + if (pte_accessible(mm, pte)) {
>> + flush_tlb_page(vma, address);
>> + /*
>> + * Ensure that the compiler orders our set_pte()
>> + * after the flush_tlb_page() no matter what.
>> + */
>> + barrier();
>
> I don’t think such a barrier (after remote TLB flush) is needed.
> Eventually, if a remote flush takes place, you get csd_lock_wait() to be
> called, and then smp_rmb() is called (which is essentially a barrier()
> call on x86).
Andi really wanted to make sure this got in here. He said there was a
bug that bit him really badly once where a function got reordered.
Granted, a call _should_ be sufficient to keep the compiler from
reordering things, but this makes double sure.
