On Sun 08-03-15 16:38:38, Michal Hocko wrote:
> On Sat 07-03-15 13:38:08, Sasha Levin wrote:
> [...]
> > [ 1573.730097] ? kasan_free_pages (mm/kasan/kasan.c:301)
> > [ 1573.788680] free_pages_prepare (mm/page_alloc.c:791)
> > [ 1573.788680] ? free_hot_cold_page (./arch/x86/include/asm/paravirt.h:809 (discriminator 2) mm/page_alloc.c:1579 (discriminator 2))
> > [ 1573.788680] free_hot_cold_page (mm/page_alloc.c:1543)
> > [ 1573.788680] __free_pages (mm/page_alloc.c:2957)
> > [ 1573.788680] ? __vunmap (mm/vmalloc.c:1460 (discriminator 2))
> > [ 1573.788680] __vunmap (mm/vmalloc.c:1460 (discriminator 2))
>
> __vunmap is doing:
> for (i = 0; i < area->nr_pages; i++) {
> struct page *page = area->pages[i];
>
> BUG_ON(!page);
> __free_page(page);
> }
>
> is it possible that nr_pages is a huge number (a large vmalloc area)? I
> do not see any cond_resched down __free_page path at least. vfree
> delayes the call to workqueue when called from irq context and vunmap is
> marked as might_sleep). So to me it looks like it would be safe. Something
> for vmalloc familiar people, though.
Hmm, just looked into the git log and it seems that there are/were
some callers of vfree with spinlock held (e.g. 9265f1d0c759 (GFS2:
gfs2_dir_get_hash_table(): avoiding deferred vfree() is easy here...))
and who knows how many others like that we have so cond_resched here is
no-no.
--
Michal Hocko
SUSE Labs
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>