CONFIG_SLUB/USE_SPLIT_PTLOCKS compatibility

[Max Filippov]

Hello,

I'm reliably getting kernel crash on xtensa when CONFIG_SLUB
is selected and USE_SPLIT_PTLOCKS appears to be true (SMP=y,
NR_CPUS=4, DEBUG_SPINLOCK=n, DEBUG_LOCK_ALLOC=n).
This happens because spinlock_t ptl and struct page *first_page overlap
in the struct page. The following call chain makes allocation of order
3 and initializes first_page pointer in its 7 tail pages:

 do_page_fault
  handle_mm_fault
   __pte_alloc
    kmem_cache_alloc
     __slab_alloc
      new_slab
       __alloc_pages_nodemask
        get_page_from_freelist
         prep_compound_page

Later pte_offset_map_lock is called with one of these tail pages
overwriting its first_page pointer:

 do_fork
  copy_process
   dup_mm
    copy_page_range
     copy_pte_range
      pte_alloc_map_lock
       pte_offset_map_lock

Finally kmem_cache_free is called for that tail page, which calls
slab_free(s, virt_to_head_page(x),... but virt_to_head_page here
returns NULL, because the page's first_page pointer was overwritten
earlier:

exit_mmap
 free_pgtables
  free_pgd_range
   free_pud_range
    free_pmd_range
     free_pte_range
      pte_free
       kmem_cache_free
        slab_free
         __slab_free

__slab_free touches NULL struct page, that's it.

Changing allocator to SLAB or enabling DEBUG_SPINLOCK
fixes that crash.

My question is, is CONFIG_SLUB supposed to work with
USE_SPLIT_PTLOCKS (and if yes what's wrong in my case)?

-- 
Thanks.
-- Max
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/