Re: [PATCH] 2.6.5-aa1 arch updates

[Andrew Morton <akpm@osdl.org>]

Andrew Morton <akpm@osdl.org> wrote:
>
> Hugh Dickins <hugh@veritas.com> wrote:
>  >
>  > I notice that's a __GFP_REPEAT allocation, but even those fail when
>  >  OOM-killed - I find its alias __GFP_NOFAIL very misleading.
> 
>  #define __GFP_REPEAT	0x400	/* Retry the allocation.  Might fail */
>  #define __GFP_NOFAIL	0x800	/* Retry for ever.  Cannot fail */
> 
>  __GFP_REPEAT is mainly for higher-order allocations which would otherwise
>  have given up too early.

It all comes back to me now.  The reason there is a __GFP_REPEAT in the
pte_alloc_one() implementations is that lots of architectures used to do
stuff like this:

static inline pte_t *
pte_alloc_one_kernel(struct mm_struct *mm, unsigned long addr)
{
	int count = 0;
	pte_t *pte;

	do {
		pte = (pte_t *)__get_free_page(GFP_KERNEL);
		if (pte)
			clear_page(pte);
		else {
			current->state = TASK_UNINTERRUPTIBLE;
			schedule_timeout(HZ);
		}
	} while (!pte && (count++ < 10));

	return pte;
}

That was all removed and the __GFP_REPEAT flag was added instead, as a "try
really hard" hint to the page allocator.