On Thu, Oct 07, 2010 at 06:09:11AM +0800, Andi Kleen wrote:
> From: Andi Kleen <ak@xxxxxxxxxxxxxxx>
>
> Hard offlining was always able to offline most kernel pages by setting the
> HWPoison flag on it and waiting for the next free. This works on all
> pages that get eventually freed.
>
> This didn't work for soft offlining unfortunately, because it cannot
> safely set the HWPoison flag on a still alive page. Handle this
> by introducing a second page flag HWPoisonOnFree that is handled
> by page free and enables HWPoison then. This does not add any
> overhead to page free because it can be handled as a bad flag
> which are already checked for.
>
> Since page flags are scarce on 32bit architectures this is only done on
> 64bit .
I have no problem with the 64bit only page flag.
> Signed-off-by: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> ---
> include/linux/mm.h | 1 +
> include/linux/page-flags.h | 15 +++++++++++++-
> mm/Kconfig | 4 +++
> mm/memory-failure.c | 46 +++++++++++++++++++++++++++++++++++++++++++-
> mm/page_alloc.c | 2 +
> 5 files changed, 66 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index 74949fb..f7da94d 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -1481,6 +1481,7 @@ enum mf_flags {
> extern void memory_failure(unsigned long pfn, int trapno);
> extern int __memory_failure(unsigned long pfn, int trapno, int flags);
> extern int unpoison_memory(unsigned long pfn);
> +extern void hwpoison_page_on_free(struct page *p);
> extern int sysctl_memory_failure_early_kill;
> extern int sysctl_memory_failure_recovery;
> extern void shake_page(struct page *p, int access);
> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> index 6fa3178..2d4196b 100644
> --- a/include/linux/page-flags.h
> +++ b/include/linux/page-flags.h
> @@ -107,6 +107,9 @@ enum pageflags {
> #endif
> #ifdef CONFIG_MEMORY_FAILURE
> PG_hwpoison, /* hardware poisoned page. Don't touch */
> +#ifdef CONFIG_HWPOISON_ON_FREE
> + PG_hwpoison_on_free, /* hwpoison on free */
> +#endif
> #endif
> __NR_PAGEFLAGS,
>
> @@ -280,6 +283,15 @@ PAGEFLAG_FALSE(HWPoison)
> #define __PG_HWPOISON 0
> #endif
>
> +#if defined(CONFIG_MEMORY_FAILURE) && BITS_PER_LONG == 64
We have the simpler CONFIG_HWPOISON_ON_FREE :)
> +PAGEFLAG(HWPoisonOnFree, hwpoison_on_free)
> +TESTSCFLAG(HWPoisonOnFree, hwpoison_on_free)
> +#define __PG_HWPOISON_ON_FREE (1UL << PG_hwpoison_on_free)
> +#else
> +PAGEFLAG_FALSE(HWPoisonOnFree)
Could define SETPAGEFLAG_NOOP(HWPoisonOnFree) too, for eliminating an
#ifdef in the .c file.
> +#define __PG_HWPOISON_ON_FREE 0
> +#endif
> +
> u64 stable_page_flags(struct page *page);
>
> static inline int PageUptodate(struct page *page)
> @@ -406,7 +418,8 @@ static inline void __ClearPageTail(struct page *page)
> 1 << PG_private | 1 << PG_private_2 | \
> 1 << PG_buddy | 1 << PG_writeback | 1 << PG_reserved | \
> 1 << PG_slab | 1 << PG_swapcache | 1 << PG_active | \
> - 1 << PG_unevictable | __PG_MLOCKED | __PG_HWPOISON)
> + 1 << PG_unevictable | __PG_MLOCKED | __PG_HWPOISON | \
> + __PG_HWPOISON_ON_FREE)
>
> /*
> * Flags checked when a page is prepped for return by the page allocator.
> diff --git a/mm/Kconfig b/mm/Kconfig
> index f0fb912..ede5444 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -269,6 +269,10 @@ config MEMORY_FAILURE
> even when some of its memory has uncorrected errors. This requires
> special hardware support and typically ECC memory.
>
> +config HWPOISON_ON_FREE
> + depends on MEMORY_FAILURE && 64BIT
> + def_bool y
> +
> config HWPOISON_INJECT
> tristate "HWPoison pages injector"
> depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index 9c26eec..34901f6 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -1268,6 +1268,14 @@ int soft_offline_page(struct page *page, int flags)
> if (ret == 0)
> goto done;
>
> +#ifdef CONFIG_HWPOISON_ON_FREE
> + /*
> + * When the page gets freed make sure to poison
> + * it immediately.
> + */
> + SetPageHWPoisonOnFree(page);
> +#endif
> +
> /*
> * Page cache page we can handle?
> */
> @@ -1279,7 +1287,15 @@ int soft_offline_page(struct page *page, int flags)
> shake_page(page, 1);
>
> /*
> - * Did it turn free?
> + * Did it get poisoned on free?
> + */
> + if (PageHWPoison(page)) {
> + pr_info("soft_offline: %#lx: Page freed through shaking\n", pfn);
> + return 0;
> + }
> +
> + /*
> + * Try to grab it as a free page again.
> */
> ret = get_any_page(page, pfn, 0);
> if (ret < 0)
> @@ -1287,12 +1303,23 @@ int soft_offline_page(struct page *page, int flags)
> if (ret == 0)
> goto done;
> }
> +
> + if (PageHWPoisonOnFree(page)) {
> + pr_info("soft_offline: %#lx: Delaying poision of unknown page %lx to free\n",
> + pfn, page->flags);
> + return -EIO; /* or 0? */
-EIO looks safer because HWPoisonOnFree does not guarantee success.
> + }
> +
> if (!PageLRU(page)) {
> pr_debug("soft_offline: %#lx: unknown non LRU page type %lx\n",
> pfn, page->flags);
> return -EIO;
> }
>
> + /*
> + * Normal page cache page here.
> + */
> +
> lock_page(page);
> wait_on_page_writeback(page);
>
> @@ -1389,3 +1416,20 @@ int is_hwpoison_address(unsigned long addr)
> return is_hwpoison_entry(entry);
> }
> EXPORT_SYMBOL_GPL(is_hwpoison_address);
> +
> +#ifdef CONFIG_HWPOISON_ON_FREE
> +
> +/*
> + * HWPoison a page after freeing.
> + * This is the fallback path for most pages we cannot free early.
> + */
> +void hwpoison_page_on_free(struct page *p)
> +{
> + get_page(p);
> + SetPageHWPoison(p);
> + ClearPageHWPoisonOnFree(p);
> + pr_info("MCE: %#lx: Delayed poisioning of page after free\n",
> + page_to_pfn(p));
> + atomic_long_inc(&mce_bad_pages);
> +}
> +#endif
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index a8cfa9c..519c24c 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -564,6 +564,8 @@ static inline int free_pages_check(struct page *page)
> (page->mapping != NULL) |
> (atomic_read(&page->_count) != 0) |
> (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
> + if (PageHWPoisonOnFree(page))
> + hwpoison_page_on_free(page);
hwpoison_page_on_free() seems to be undefined when
CONFIG_HWPOISON_ON_FREE is not defined.
> bad_page(page);
> return 1;
> }
Thanks,
Fengguang
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>