Re: 2.6.28-rc2: Unable to handle kernel paging request at iov_iter_copy_from_user_atomic

["Alexander Beregalov" <a.beregalov@gmail.com>]

2008/11/1 Hugh Dickins <hugh@veritas.com>:
> On Sat, 1 Nov 2008, Alexander Beregalov wrote:
>>  2.6.28-rc2-00452-gf891caf on sparc64
>>
>> How to reproduce: run dbench on tmpfs
>>
>>
>> Unable to handle kernel paging request at virtual address fffff80037c1c000
>> tsk->{mm,active_mm}->context = 0000000000001ae7
>> tsk->{mm,active_mm}->pgd = fffff8000ec8c000
>>               \|/ ____ \|/
>>               "@'/ .. \`@"
>>               /_| \__/ |_\
>>                  \__U_/
>> dbench(5007): Oops [#1]
>> TSTATE: 0000000011009604 TPC: 00000000005acbac TNPC: 00000000005acbb0
>> Y: 00000000    Not tainted
>> TPC: <__bzero+0x20/0xc0>
>> g0: 0000000000000016 g1: 0000000000000000 g2: 0000000000000000 g3:
>> 0000000000033ae7
>> g4: fffff8000ec9c380 g5: 0000000000000020 g6: fffff8003b834000 g7:
>> ffffffffffffe8b1
>> o0: fffff80037c1c8b1 o1: 00000000000008b1 o2: 0000000000000000 o3:
>> fffff80037c1c8b1
>> o4: 0000000000000000 o5: 0000000000034398 sp: fffff8003b836e41 ret_pc:
>> 00000000005ae73c
>> RPC: <copy_from_user_fixup+0x4c/0x70>
>> l0: 0000000000852800 l1: 0000000011009603 l2: 0000000000827ff4 l3:
>> 0000000000000400
>> l4: 0000000000000000 l5: 0000000000000001 l6: 0000000000000000 l7:
>> 0000000000000008
>> i0: fffff80037c1e000 i1: 0000000000032398 i2: 00000000000008b1 i3:
>> fffff80037c3e398
>> i4: fffff80037c1e000 i5: 0000000000000000 i6: fffff8003b836f01 i7:
>> 0000000000486f28
>> I7: <iov_iter_copy_from_user_atomic+0x90/0xe0>
>> Caller[0000000000486f28]: iov_iter_copy_from_user_atomic+0x90/0xe0
>> Caller[0000000000488a58]: generic_file_buffered_write+0x108/0x2a8
>> Caller[0000000000489140]: __generic_file_aio_write_nolock+0x35c/0x380
>> Caller[00000000004899b4]: generic_file_aio_write+0x58/0xc8
>> Caller[00000000004b23d4]: do_sync_write+0x90/0xe0
>> Caller[00000000004b2ca4]: vfs_write+0x7c/0x11c
>> Caller[00000000004b2d98]: sys_pwrite64+0x54/0x80
>> Caller[000000000043efa4]: sys32_pwrite64+0x20/0x34
>> Caller[0000000000406154]: linux_sparc_syscall32+0x34/0x40
>> Caller[00000000f7e8df80]: 0xf7e8df80
>> Instruction DUMP: c56a2000  808a2003  02480006 <d42a2000> 90022001
>> 808a2003  1247fffd  92226001  808a2007
>
> [ Snipped the rest of it, which is just a consequence of oopsing
>  there, then a repeat of this one on a different address ]
>
> Good find!  Though surprising it's not been found before: something
> for -stable I suspect.  Please don't take my signoff too seriously,
> I cannot test this myself and it needs confirmation from DaveM...
>
>
> Alexander Beregalov reports oops in __bzero() called from
> copy_from_user_fixup() called from iov_iter_copy_from_user_atomic(),
> when running dbench on tmpfs on sparc64: its __copy_from_user_inatomic
> and __copy_to_user_inatomic should be avoiding, not calling, the fixups.
>
> Signed-off-by: Hugh Dickins <hugh@veritas.com>
> ---
>
>  arch/sparc/include/asm/uaccess_64.h |    4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> --- 2.6.28-rc2/arch/sparc/include/asm/uaccess_64.h      2008-10-09 23:13:53.000000000 +0100
> +++ linux/arch/sparc/include/asm/uaccess_64.h   2008-11-01 18:33:59.000000000 +0000
> @@ -265,8 +265,8 @@ extern long __strnlen_user(const char __
>
>  #define strlen_user __strlen_user
>  #define strnlen_user __strnlen_user
> -#define __copy_to_user_inatomic __copy_to_user
> -#define __copy_from_user_inatomic __copy_from_user
> +#define __copy_to_user_inatomic ___copy_to_user
> +#define __copy_from_user_inatomic ___copy_from_user
>
>  #endif  /* __ASSEMBLY__ */
>
>
It helps, thanks.

Should we fix also sparc32?

diff --git a/arch/sparc/include/asm/uaccess_32.h
b/arch/sparc/include/asm/uaccess_32.h
index 47d5619..e878eb4 100644
--- a/arch/sparc/include/asm/uaccess_32.h
+++ b/arch/sparc/include/asm/uaccess_32.h
@@ -270,8 +270,8 @@ static inline unsigned long __copy_from_user(void
*to, const void __user *from,
        return __copy_user((__force void __user *) to, from, n);
 }

-#define __copy_to_user_inatomic __copy_to_user
-#define __copy_from_user_inatomic __copy_from_user
+#define __copy_to_user_inatomic ___copy_to_user
+#define __copy_from_user_inatomic ___copy_from_user

 static inline unsigned long __clear_user(void __user *addr, unsigned long size)
 {

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>