Page 'new' during MIGRATION can't be flushed by flush_cache_page().
Using flush_cache_page(vma, addr, pfn) is justified only if
page is already placed in process page table, and that is done right
after flush_cache_page(). But without it the arch function has
no knowledge of process PTE and does nothing.
Besides that, flush_cache_page() flushes an application cache,
kernel has a different page virtual address and dirtied it.
Replace it with flush_dcache_page(new) which is a proper usage.
Old page is flushed in try_to_unmap_one() before MIGRATION.
This bug takes place in Sead3 board with M14Kc MIPS CPU without
cache aliasing (but Harvard arch - separate I and D cache)
in tight memory environment (128MB) each 1-3days on SOAK test.
It fails in cc1 during kernel build (SIGILL, SIGBUS, SIGSEG) if
CONFIG_COMPACTION is switched ON.
Author: Leonid Yegoshin <[email protected]>
Signed-off-by: Leonid Yegoshin <[email protected]>
---
mm/migrate.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/mm/migrate.c b/mm/migrate.c
index 2fd8b4a..4c6250a 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -165,7 +165,7 @@ static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
pte = arch_make_huge_pte(pte, vma, new, 0);
}
#endif
- flush_cache_page(vma, addr, pte_pfn(pte));
+ flush_dcache_page(new);
set_pte_at(mm, addr, ptep, pte);
if (PageHuge(new)) {
On 05/08/2013 08:18 PM, Leonid Yegoshin wrote:
> Page 'new' during MIGRATION can't be flushed by flush_cache_page().
> Using flush_cache_page(vma, addr, pfn) is justified only if
> page is already placed in process page table, and that is done right
> after flush_cache_page(). But without it the arch function has
> no knowledge of process PTE and does nothing.
>
> Besides that, flush_cache_page() flushes an application cache,
> kernel has a different page virtual address and dirtied it.
>
> Replace it with flush_dcache_page(new) which is a proper usage.
>
> Old page is flushed in try_to_unmap_one() before MIGRATION.
>
> This bug takes place in Sead3 board with M14Kc MIPS CPU without
> cache aliasing (but Harvard arch - separate I and D cache)
> in tight memory environment (128MB) each 1-3days on SOAK test.
> It fails in cc1 during kernel build (SIGILL, SIGBUS, SIGSEG) if
> CONFIG_COMPACTION is switched ON.
Good catch!
> Author: Leonid Yegoshin <[email protected]>
> Signed-off-by: Leonid Yegoshin <[email protected]>
Acked-by: Rik van Riel <[email protected]>
On Wed, May 08, 2013 at 05:18:21PM -0700, Leonid Yegoshin wrote:
> Page 'new' during MIGRATION can't be flushed by flush_cache_page().
> Using flush_cache_page(vma, addr, pfn) is justified only if
> page is already placed in process page table, and that is done right
> after flush_cache_page(). But without it the arch function has
> no knowledge of process PTE and does nothing.
>
> Besides that, flush_cache_page() flushes an application cache,
> kernel has a different page virtual address and dirtied it.
>
> Replace it with flush_dcache_page(new) which is a proper usage.
>
> Old page is flushed in try_to_unmap_one() before MIGRATION.
>
> This bug takes place in Sead3 board with M14Kc MIPS CPU without
> cache aliasing (but Harvard arch - separate I and D cache)
> in tight memory environment (128MB) each 1-3days on SOAK test.
> It fails in cc1 during kernel build (SIGILL, SIGBUS, SIGSEG) if
> CONFIG_COMPACTION is switched ON.
>
> Author: Leonid Yegoshin <[email protected]>
> Signed-off-by: Leonid Yegoshin <[email protected]>
Acked-by: Mel Gorman <[email protected]>
--
Mel Gorman
SUSE Labs