Re: 2.6.25-rc7-git2: Reported regressions from 2.6.24

[Christoph Lameter <clameter@sgi.com>]

On Thu, 27 Mar 2008, Linus Torvalds wrote:

> Totally irrelevant.
> 
> The page allocation path does
> 
>         if (gfp_flags & __GFP_ZERO)
>                 prep_zero_page(page, order, gfp_flags);
> 
> and that will cause a warning REGARDLESS of whether the page is a HIGHMEM 
> page or not.


prep_zero_page does:

static inline void prep_zero_page(struct page *page, int order, gfp_t 
gfp_flags)
{
        int i;

        /*
         * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
         * and __GFP_HIGHMEM from hard or soft interrupt context.
         */
        VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
        for (i = 0; i < (1 << order); i++)
                clear_highpage(page + i);
}

So we forbit __GFP_HIGHMEM and in_interrupt which makes sense. The simple 
forwarding of large kmallocs to the page allocator as done by SLUB / SLOB 
is fine.

Then clear_highpage calls additional checking functions that have 
the effect of generally forbiding zeroing in interrupt context if 
CONFIG_HIGHMEM is set. This is wrong and needs to be fixed.

> And the fact is, passing in GFP_ZERO from the SLUB code is a bug 
> regardless, because it unnecessarily does the dual memset().

Well that is only the fallback path of __slab_alloc which is not triggered 
here and not performance sensitive. We could clear the flag there but 
that is irrevelant for this issue.