Re: x86 ptep_get_and_clear question

[Manfred Spraul]

Jamie Lokier wrote:
> 
> /* mprotect.c */
>         entry = ptep_get_and_clear(pte);
>         set_pte(pte, pte_modify(entry, newprot));
> 
> I.e. the only code with the race condition is code which explicitly
> clears the dirty bit, in vmscan.c.
> 
> Do you see any possibility of losing a dirty bit here?
>
Of course.
Just check the output after preprocessing.
It's 
	int entry;
	entry = *pte;
	entry &= ~_PAGE_CHG_MASK;
	entry |= pgprot_val(newprot)
	*pte = entry;

We need
	atomic_clear_mask (_PAGE_CHG_MASK, pte);
	atomic_set_mask (pgprot_val(newprot), *pte);

for multi threaded apps.

> If not, there's no need for the intricate "gather" or "double scan"
> schemes for mprotect() and it can stay as fast as possible.
>
Correct, but we need a platform specific "update_pte", and perhaps
update_begin, update_end hooks (empty on i386) for other archs.

--
	Manfred