| 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: <CALCETrU5jaennr5ziS9NzNA6KpK204acdroJpuc6yYy3PGvpHQ@mail.gmail.com>
Date: Wed, 21 Feb 2018 16:11:16 +0000
From: Andy Lutomirski <luto@...nel.org>
To: Matthew Wilcox <willy@...radead.org>
Cc: Konstantin Khlebnikov <khlebnikov@...dex-team.ru>,
Dave Hansen <dave.hansen@...el.com>,
LKML <linux-kernel@...r.kernel.org>,
Christoph Hellwig <hch@...radead.org>,
Linux-MM <linux-mm@...ck.org>, Andy Lutomirski <luto@...nel.org>,
Andrew Morton <akpm@...ux-foundation.org>,
"Kirill A. Shutemov" <kirill@...temov.name>
Subject: Re: Use higher-order pages in vmalloc
On Wed, Feb 21, 2018 at 3:42 PM, Matthew Wilcox <willy@...radead.org> wrote:
> On Tue, Jan 23, 2018 at 01:55:32PM +0300, Konstantin Khlebnikov wrote:
>> Virtually mapped stack have two bonuses: it eats order-0 pages and
>> adds guard page at the end. But it slightly slower if system have
>> plenty free high-order pages.
>>
>> This patch adds option to use virtually bapped stack as fallback for
>> atomic allocation of traditional high-order page.
>
> This prompted me to write a patch I've been meaning to do for a while,
> allocating large pages if they're available to satisfy vmalloc. I thought
> it would save on touching multiple struct pages, but it turns out that
> the checking code we currently have in the free_pages path requires you
> to have initialised all of the tail pages (maybe we can make that code
> conditional ...)
>
> It does save the buddy allocator the trouble of breaking down higher-order
> pages into order-0 pages, only to allocate them again immediately.
>
> (um, i seem to have broken the patch while cleaning it up for submission.
> since it probably won't be accepted anyway, I'm not going to try to debug it)
>
> diff --git a/kernel/fork.c b/kernel/fork.c
> index be8aa5b98666..2bc01071b6ae 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -319,12 +319,12 @@ static void account_kernel_stack(struct task_struct *tsk, int account)
> if (vm) {
> int i;
>
> - BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
> -
> - for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
> - mod_zone_page_state(page_zone(vm->pages[i]),
> + for (i = 0; i < vm->nr_pages; i++) {
> + struct page *page = vm->pages[i];
> + unsigned int size = PAGE_SIZE << compound_order(page);
> + mod_zone_page_state(page_zone(page),
> NR_KERNEL_STACK_KB,
> - PAGE_SIZE / 1024 * account);
> + size / 1024 * account);
> }
>
> /* All stack pages belong to the same memcg. */
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index b728c98f49cd..4bfc29b21bc1 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -134,6 +134,7 @@ static void vunmap_page_range(unsigned long addr, unsigned long end)
> static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
> unsigned long end, pgprot_t prot, struct page **pages, int *nr)
> {
> + unsigned int i;
> pte_t *pte;
>
> /*
> @@ -151,9 +152,13 @@ static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
> return -EBUSY;
> if (WARN_ON(!page))
> return -ENOMEM;
> - set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
> + for (i = 0; i < (1UL << compound_order(page)); i++) {
> + set_pte_at(&init_mm, addr, pte++,
> + mk_pte(page + i, prot));
> + addr += PAGE_SIZE;
> + }
> (*nr)++;
> - } while (pte++, addr += PAGE_SIZE, addr != end);
> + } while (addr != end);
> return 0;
> }
>
> @@ -1530,14 +1535,14 @@ static void __vunmap(const void *addr, int deallocate_pages)
> debug_check_no_obj_freed(addr, get_vm_area_size(area));
>
> if (deallocate_pages) {
> - int i;
> + unsigned int i;
>
> for (i = 0; i < area->nr_pages; i++) {
> struct page *page = area->pages[i];
>
> BUG_ON(!page);
> __ClearPageVmalloc(page);
> - __free_pages(page, 0);
> + __free_pages(page, compound_order(page));
> }
>
> kvfree(area->pages);
> @@ -1696,11 +1701,20 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
>
> for (i = 0; i < area->nr_pages; i++) {
> struct page *page;
> -
> - if (node == NUMA_NO_NODE)
> - page = alloc_page(alloc_mask);
> - else
> - page = alloc_pages_node(node, alloc_mask, 0);
> + unsigned int j = ilog2(area->nr_pages - i) + 1;
> +
> + do {
> + j--;
> + if (node == NUMA_NO_NODE)
> + page = alloc_pages(alloc_mask, j);
> + else
> + page = alloc_pages_node(node, alloc_mask, j);
> + } while (!page && j);
> +
> + if (j) {
> + area->nr_pages -= (1UL << j) - 1;
Is there any code that expects area->nr_pages to be the size of the
area in pages? I don't know of any such code.
> + prep_compound_page(page, j);
> + }
>
> if (unlikely(!page)) {
> /* Successfully allocated i pages, free them in __vunmap() */
> @@ -1719,8 +1733,8 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
>
> fail:
> warn_alloc(gfp_mask, NULL,
> - "vmalloc: allocation failure, allocated %ld of %ld bytes",
> - (area->nr_pages*PAGE_SIZE), area->size);
> + "vmalloc: allocation failure, allocated %ld of %ld bytes",
> + (nr_pages * PAGE_SIZE), get_vm_area_size(area));
> vfree(area->addr);
> return NULL;
> }
Powered by blists - more mailing lists