| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20110411152223.3fb91a62.akpm@linux-foundation.org>
Date: Mon, 11 Apr 2011 15:22:23 -0700
From: Andrew Morton <akpm@...ux-foundation.org>
To: Dave Hansen <dave@...ux.vnet.ibm.com>
Cc: linux-mm@...ck.org, linux-kernel@...r.kernel.org,
Timur Tabi <timur@...escale.com>,
Andi Kleen <andi@...stfloor.org>, Mel Gorman <mel@....ul.ie>,
Michal Nazarewicz <mina86@...a86.com>,
David Rientjes <rientjes@...gle.com>
Subject: Re: [PATCH 2/3] make new alloc_pages_exact()
On Mon, 11 Apr 2011 15:03:46 -0700
Dave Hansen <dave@...ux.vnet.ibm.com> wrote:
>
> What I really wanted in the end was a highmem-capable alloc_pages_exact(),
> so here it is. This function can be used to allocate unmapped (like
> highmem) non-power-of-two-sized areas of memory. This is in constast to
> get_free_pages_exact() which can only allocate from lowmem.
>
> My plan is to use this in the virtio_balloon driver to allocate large,
> oddly-sized contiguous areas.
>
> The new __alloc_pages_exact() now takes a size in numbers of pages,
> and returns a 'struct page', which means it can now address highmem.
>
> It's a bit unfortunate that this introduces __free_pages_exact()
> alongside free_pages_exact(). But that mess already exists with
> __free_pages() vs. free_pages_exact(). So, at worst, this mirrors
> the mess that we already have.
>
> I'm also a bit worried that I've not put in something named
> alloc_pages_exact(), but that behaves differently than it did before this
> set. I got all of the in-tree cases, but I'm a bit worried about
> stragglers elsewhere. So, I'm calling this __alloc_pages_exact() for
> the moment. We can take out the __ some day if it bothers people.
Yup, that's fair enough.
> Note that the __get_free_pages() has a !GFP_HIGHMEM check. Now that
> we are using alloc_pages_exact() instead of __get_free_pages() for
> get_free_pages_exact(), we had to add a new check in
> get_free_pages_exact().
>
> This has been compile and boot tested, and I checked that
>
> echo 2 > /sys/kernel/profiling
>
> still works, since it uses get_free_pages_exact().
>
> Signed-off-by: Dave Hansen <dave@...ux.vnet.ibm.com>
> ---
>
> linux-2.6.git-dave/include/linux/gfp.h | 4 +
> linux-2.6.git-dave/mm/page_alloc.c | 84 ++++++++++++++++++++++++---------
> 2 files changed, 67 insertions(+), 21 deletions(-)
>
> diff -puN include/linux/gfp.h~make_new_alloc_pages_exact include/linux/gfp.h
> --- linux-2.6.git/include/linux/gfp.h~make_new_alloc_pages_exact 2011-04-11 15:01:17.165822836 -0700
> +++ linux-2.6.git-dave/include/linux/gfp.h 2011-04-11 15:01:17.177822831 -0700
> @@ -351,6 +351,10 @@ extern struct page *alloc_pages_vma(gfp_
> extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
> extern unsigned long get_zeroed_page(gfp_t gfp_mask);
>
> +/* 'struct page' version */
> +struct page *__alloc_pages_exact(gfp_t gfp_mask, size_t size);
> +void __free_pages_exact(struct page *page, size_t size);
The declarations use "size", but the definitions use "nr_pages".
"nr_pages" is way better.
Should it really be size_t? size_t's units are "bytes", usually.
> -void *get_free_pages_exact(gfp_t gfp_mask, size_t size)
> +struct page *__alloc_pages_exact(gfp_t gfp_mask, size_t nr_pages)
Most allocation functions are of the form foo(size, gfp_t), but this
one has the args reversed. Was there a reason for that?
> {
> - unsigned int order = get_order(size);
> - unsigned long addr;
> + unsigned int order = get_order(nr_pages * PAGE_SIZE);
> + struct page *page;
>
> - addr = __get_free_pages(gfp_mask, order);
> - if (addr) {
> - unsigned long alloc_end = addr + (PAGE_SIZE << order);
> - unsigned long used = addr + PAGE_ALIGN(size);
> + page = alloc_pages(gfp_mask, order);
> + if (page) {
> + struct page *alloc_end = page + (1 << order);
> + struct page *used = page + nr_pages;
>
> - split_page(virt_to_page((void *)addr), order);
> + split_page(page, order);
> while (used < alloc_end) {
> - free_page(used);
> - used += PAGE_SIZE;
> + __free_page(used);
> + used++;
> }
> }
>
> - return (void *)addr;
> + return page;
> +}
> +EXPORT_SYMBOL(__alloc_pages_exact);
> +
> +/**
> + * __free_pages_exact - release memory allocated via __alloc_pages_exact()
> + * @virt: the value returned by get_free_pages_exact.
> + * @nr_pages: size in pages, same value as passed to __alloc_pages_exact().
> + *
> + * Release the memory allocated by a previous call to __alloc_pages_exact().
> + */
> +void __free_pages_exact(struct page *page, size_t nr_pages)
> +{
> + struct page *end = page + nr_pages;
> +
> + while (page < end) {
Hand-optimised. Old school. Doesn't trust the compiler :)
> + __free_page(page);
> + page++;
> + }
> +}
> +EXPORT_SYMBOL(__free_pages_exact);
Really, this function duplicates release_pages(). release_pages() is
big and fat and complex and is a crime against uniprocessor but it does
make some effort to reduce the spinlocking frequency and in many
situations, release_pages() will cause vastly less locked bus traffic
than your __free_pages_exact(). And who knows, smart use of
release_pages()'s "cold" hint may provide some benefits.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists