[2.6.37-rc1] /proc/vmstat failure.

[2.6.37-rc1] /proc/vmstat failure.

[Tetsuo Handa]

Hello.

"cat /proc/vmstat" triggered below failure.

BUG: unable to handle kernel paging request at 07c06d16
IP: [<c050c336>] strnlen+0x6/0x20
*pde = 00000000
Oops: 0000 [#2] SMP
last sysfs file: /sys/devices/pci0000:00/0000:00:11.0/0000:02:02.0/irq
Modules linked in: ipv6 video output sbs sbshc battery ac button pcnet32 serio_raw mii floppy i2c_piix4 i2c_core rtc_cmos rtc_core rtc_lib mptspi mptscsih mptbase scsi_transport_spi ext3 jbd mbcache

Pid: 2982, comm: cat Tainted: G      D     2.6.36+ #12 440BX Desktop Reference Platform/VMware Virtual Platform
EIP: 0060:[<c050c336>] EFLAGS: 00010297 CPU: 0
EIP is at strnlen+0x6/0x20
EAX: 07c06d16 EBX: 07c06d16 ECX: 07c06d16 EDX: fffffffe
ESI: df0ce000 EDI: de769f0c EBP: df0cf000 ESP: de769e80
 DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
Process cat (pid: 2982, ti=de768000 task=df0e8e00 task.ti=de768000)
Stack:
 c050a249 ffff9f10 dea43000 00000000 00000001 c042be6b c06d1514 00000002
<0> c06d1513 de769f0c df0ce000 c050afcc ff0a0004 ffffffff 00000000 00000004
<0> 00001000 df0ce000 df0cf000 c06d1515 de769f10 ff0a0004 ffffffff df0ce000
Call Trace:
 [<c050a249>] ? string+0x39/0xe0
 [<c042be6b>] ? __wake_up_common+0x4b/0x80
 [<c050afcc>] ? vsnprintf+0x1ec/0x380
 [<c04b380e>] ? seq_printf+0x2e/0x60
 [<c04829a6>] ? vmstat_show+0x26/0x30
 [<c04b3bb6>] ? seq_read+0xa6/0x380
 [<c04b3b10>] ? seq_read+0x0/0x380
 [<c04d5d2f>] ? proc_reg_read+0x5f/0x90
 [<c049c4a1>] ? vfs_read+0xa1/0x140
 [<c04d5cd0>] ? proc_reg_read+0x0/0x90
 [<c049c981>] ? sys_read+0x41/0x70
 [<c0402bd0>] ? sysenter_do_call+0x12/0x26
Code: 90 8d 74 26 00 85 c9 57 89 c7 89 d0 74 05 f2 ae 75 01 4f 89 f8 5f c3 8d b4 26 00 00 00 00 8d bc 27 00 00 00 00 89 c1 89 c8 eb 06 <80> 38 00 74 07 40 4a 83 fa ff 75 f4 29 c8 c3 90 90 90 90 90 90
EIP: [<c050c336>] strnlen+0x6/0x20 SS:ESP 0068:de769e80
CR2: 0000000007c06d16
---[ end trace 051c52adf97c7465 ]---

git bisect reported that commit 79da826aee6a10902ef411bc65864bd02102fa83
"writeback: report dirty thresholds in /proc/vmstat" is first bad commit.

My environment:
  CentOS 5.5 on i686

Steps to reproduce:
  Boot and login via SSH and do "cat /proc/vmstat".

  What is strange, this failure occurs when tty is /dev/pts/0 (i.e.
  devpts filesystem / interactive ssh login session) and does not occur
  when tty is /dev/tty1 (i.e. console login session).

Regards.

Re: [2.6.37-rc1] /proc/vmstat failure.

[Linus Torvalds]

On Wed, Nov 3, 2010 at 7:11 AM, Tetsuo Handa
<penguin-kernel@i-love.sakura.ne.jp> wrote:
>
> git bisect reported that commit 79da826aee6a10902ef411bc65864bd02102fa83
> "writeback: report dirty thresholds in /proc/vmstat" is first bad commit.

Have you verified that reverting that on top of current -git also
fixes the problem (outside of just the bisected kernels)?

> Steps to reproduce:
>  Boot and login via SSH and do "cat /proc/vmstat".
>
>  What is strange, this failure occurs when tty is /dev/pts/0 (i.e.
>  devpts filesystem / interactive ssh login session) and does not occur
>  when tty is /dev/tty1 (i.e. console login session).

It really sounds like memory corruption, likely because something has
stepped on somebody elses allocation. One thing that patch does is to
change the size of the allocation - perhaps it now ends up being in
the same bucket size with some tty data structure?

Have you tried enabling SLUB/SLAB debugging?

               Linus

Re: [2.6.37-rc1] /proc/vmstat failure.

[Wu Fengguang]

On Wed, Nov 03, 2010 at 07:11:11PM +0800, Tetsuo Handa wrote:
> Hello.
> 
> "cat /proc/vmstat" triggered below failure.
> 
> BUG: unable to handle kernel paging request at 07c06d16
> IP: [<c050c336>] strnlen+0x6/0x20

It seems that m->private is of type (void *) while the original v is
(unsigned long *). Can be fixed by the following patch.

Thanks,
Fengguang
---
diff --git a/mm/vmstat.c b/mm/vmstat.c
index cd2e42b..42eac4d 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -949,7 +949,7 @@ static void *vmstat_start(struct seq_file *m, loff_t *pos)
 	v[PGPGIN] /= 2;		/* sectors -> kbytes */
 	v[PGPGOUT] /= 2;
 #endif
-	return m->private + *pos;
+	return (unsigned long *)m->private + *pos;
 }
 
 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)

Re: [2.6.37-rc1] /proc/vmstat failure.

[Linus Torvalds]

On Wed, Nov 3, 2010 at 1:46 PM, Wu Fengguang <fengguang.wu@intel.com> wrote:
>
> It seems that m->private is of type (void *) while the original v is
> (unsigned long *). Can be fixed by the following patch.

Good catch.  Tetsuo, does that fix it for you?

And Wu - sign-off/changelog, please?

               Linus

Re: [2.6.37-rc1] /proc/vmstat failure.

[Tetsuo Handa]

Wu Fengguang wrote:
> On Wed, Nov 03, 2010 at 07:11:11PM +0800, Tetsuo Handa wrote:
> > Hello.
> > 
> > "cat /proc/vmstat" triggered below failure.
> > 
> > BUG: unable to handle kernel paging request at 07c06d16
> > IP: [<c050c336>] strnlen+0x6/0x20
> 
> It seems that m->private is of type (void *) while the original v is
> (unsigned long *). Can be fixed by the following patch.

Fixed by this patch. Thank you.

By the way, GCC treats "(void *) + offset" as "(char *) + offset"
but it is GCC specific extention. Maybe we should avoid such usage
in order to avoid build failure for other compilers.

Re: [2.6.37-rc1] /proc/vmstat failure.

[Wu Fengguang]

On Thu, Nov 04, 2010 at 05:54:48AM +0800, Tetsuo Handa wrote:
> Wu Fengguang wrote:
> > On Wed, Nov 03, 2010 at 07:11:11PM +0800, Tetsuo Handa wrote:
> > > Hello.
> > > 
> > > "cat /proc/vmstat" triggered below failure.
> > > 
> > > BUG: unable to handle kernel paging request at 07c06d16
> > > IP: [<c050c336>] strnlen+0x6/0x20
> > 
> > It seems that m->private is of type (void *) while the original v is
> > (unsigned long *). Can be fixed by the following patch.
> 
> Fixed by this patch. Thank you.

Great, thanks!

> By the way, GCC treats "(void *) + offset" as "(char *) + offset"
> but it is GCC specific extention. Maybe we should avoid such usage
> in order to avoid build failure for other compilers.

"(void *) + offset" seems to be used quite often.

I tried enable this gcc option, only to see huge amount of warnings.

       -Wpointer-arith
           Warn about anything that depends on the "size of" a
           function type or of "void".  GNU C assigns these types a
           size of 1, for convenience in calculations with "void *"
           pointers and pointers to functions.

/cc/linux-next/scripts/mod/file2alias.c: In function ‘device_id_check’:
/cc/linux-next/scripts/mod/file2alias.c:88: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c:88: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c:88: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c:94: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c:94: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c:94: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c: In function ‘do_usb_table’:
/cc/linux-next/scripts/mod/file2alias.c:275: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c: In function ‘do_table’:
/cc/linux-next/scripts/mod/file2alias.c:870: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c: In function ‘handle_moddevtable’:
/cc/linux-next/scripts/mod/file2alias.c:897: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/file2alias.c:897: warning: pointer of type ‘void *’ used in arithmetic
  HOSTCC  scripts/mod/modpost.o
/cc/linux-next/scripts/mod/modpost.c: In function ‘parse_elf’:
/cc/linux-next/scripts/mod/modpost.c:412: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:456: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:472: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:489: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:491: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:491: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:494: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:501: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:503: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:503: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c: In function ‘sec_name’:
/cc/linux-next/scripts/mod/modpost.c:716: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:717: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c: In function ‘sech_name’:
/cc/linux-next/scripts/mod/modpost.c:723: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:724: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c: In function ‘reloc_location’:
/cc/linux-next/scripts/mod/modpost.c:1443: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:1444: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:1444: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c: In function ‘section_rela’:
/cc/linux-next/scripts/mod/modpost.c:1520: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:1521: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c: In function ‘section_rel’:
/cc/linux-next/scripts/mod/modpost.c:1564: warning: pointer of type ‘void *’ used in arithmetic
/cc/linux-next/scripts/mod/modpost.c:1565: warning: pointer of type ‘void *’ used in arithmetic

Thanks,
Fengguang

Re: [2.6.37-rc1] /proc/vmstat failure.

[Américo Wang]

On Thu, Nov 04, 2010 at 10:20:52AM +0800, Wu Fengguang wrote:
>On Thu, Nov 04, 2010 at 05:54:48AM +0800, Tetsuo Handa wrote:
>> Wu Fengguang wrote:
>> > On Wed, Nov 03, 2010 at 07:11:11PM +0800, Tetsuo Handa wrote:
>> > > Hello.
>> > > 
>> > > "cat /proc/vmstat" triggered below failure.
>> > > 
>> > > BUG: unable to handle kernel paging request at 07c06d16
>> > > IP: [<c050c336>] strnlen+0x6/0x20
>> > 
>> > It seems that m->private is of type (void *) while the original v is
>> > (unsigned long *). Can be fixed by the following patch.
>> 
>> Fixed by this patch. Thank you.
>
>Great, thanks!
>
>> By the way, GCC treats "(void *) + offset" as "(char *) + offset"
>> but it is GCC specific extention. Maybe we should avoid such usage
>> in order to avoid build failure for other compilers.
>
>"(void *) + offset" seems to be used quite often.
>
>I tried enable this gcc option, only to see huge amount of warnings.
>
>       -Wpointer-arith
>           Warn about anything that depends on the "size of" a
>           function type or of "void".  GNU C assigns these types a
>           size of 1, for convenience in calculations with "void *"
>           pointers and pointers to functions.
>

Yup! I think it is good to have this.

>/cc/linux-next/scripts/mod/file2alias.c:88: warning: pointer of type ‘void *’ used in arithmetic
...
>/cc/linux-next/scripts/mod/modpost.c:1565: warning: pointer of type ‘void *’ used in arithmetic
>

No problem, this is from kbuild and most of them that I checked are actually fine,
we can send patches to remove these warnings.

Re: [2.6.37-rc1] /proc/vmstat failure.

[Wu Fengguang]

On Thu, Nov 04, 2010 at 10:52:51AM +0800, Américo Wang wrote:
> On Thu, Nov 04, 2010 at 10:20:52AM +0800, Wu Fengguang wrote:
> >On Thu, Nov 04, 2010 at 05:54:48AM +0800, Tetsuo Handa wrote:
> >> Wu Fengguang wrote:
> >> > On Wed, Nov 03, 2010 at 07:11:11PM +0800, Tetsuo Handa wrote:
> >> > > Hello.
> >> > > 
> >> > > "cat /proc/vmstat" triggered below failure.
> >> > > 
> >> > > BUG: unable to handle kernel paging request at 07c06d16
> >> > > IP: [<c050c336>] strnlen+0x6/0x20
> >> > 
> >> > It seems that m->private is of type (void *) while the original v is
> >> > (unsigned long *). Can be fixed by the following patch.
> >> 
> >> Fixed by this patch. Thank you.
> >
> >Great, thanks!
> >
> >> By the way, GCC treats "(void *) + offset" as "(char *) + offset"
> >> but it is GCC specific extention. Maybe we should avoid such usage
> >> in order to avoid build failure for other compilers.
> >
> >"(void *) + offset" seems to be used quite often.
> >
> >I tried enable this gcc option, only to see huge amount of warnings.
> >
> >       -Wpointer-arith
> >           Warn about anything that depends on the "size of" a
> >           function type or of "void".  GNU C assigns these types a
> >           size of 1, for convenience in calculations with "void *"
> >           pointers and pointers to functions.
> >
> 
> Yup! I think it is good to have this.
> 
> >/cc/linux-next/scripts/mod/file2alias.c:88: warning: pointer of type ‘void *’ used in arithmetic
> ...
> >/cc/linux-next/scripts/mod/modpost.c:1565: warning: pointer of type ‘void *’ used in arithmetic
> >
> 
> No problem, this is from kbuild and most of them that I checked are actually fine,
> we can send patches to remove these warnings.

Confirmed. Now my full kernel build is complete and the above are the
only warnings I see. So it should be easy to add the -Wpointer-arith
check.

Thanks,
Fengguang

Re: [2.6.37-rc1] /proc/vmstat failure.

[Américo Wang]

On Thu, Nov 04, 2010 at 10:53:41AM +0800, Wu Fengguang wrote:
>On Thu, Nov 04, 2010 at 10:52:51AM +0800, Américo Wang wrote:
>> On Thu, Nov 04, 2010 at 10:20:52AM +0800, Wu Fengguang wrote:
>> >On Thu, Nov 04, 2010 at 05:54:48AM +0800, Tetsuo Handa wrote:
>> >> Wu Fengguang wrote:
>> >> > On Wed, Nov 03, 2010 at 07:11:11PM +0800, Tetsuo Handa wrote:
>> >> > > Hello.
>> >> > > 
>> >> > > "cat /proc/vmstat" triggered below failure.
>> >> > > 
>> >> > > BUG: unable to handle kernel paging request at 07c06d16
>> >> > > IP: [<c050c336>] strnlen+0x6/0x20
>> >> > 
>> >> > It seems that m->private is of type (void *) while the original v is
>> >> > (unsigned long *). Can be fixed by the following patch.
>> >> 
>> >> Fixed by this patch. Thank you.
>> >
>> >Great, thanks!
>> >
>> >> By the way, GCC treats "(void *) + offset" as "(char *) + offset"
>> >> but it is GCC specific extention. Maybe we should avoid such usage
>> >> in order to avoid build failure for other compilers.
>> >
>> >"(void *) + offset" seems to be used quite often.
>> >
>> >I tried enable this gcc option, only to see huge amount of warnings.
>> >
>> >       -Wpointer-arith
>> >           Warn about anything that depends on the "size of" a
>> >           function type or of "void".  GNU C assigns these types a
>> >           size of 1, for convenience in calculations with "void *"
>> >           pointers and pointers to functions.
>> >
>> 
>> Yup! I think it is good to have this.
>> 
>> >/cc/linux-next/scripts/mod/file2alias.c:88: warning: pointer of type ‘void *’ used in arithmetic
>> ...
>> >/cc/linux-next/scripts/mod/modpost.c:1565: warning: pointer of type ‘void *’ used in arithmetic
>> >
>> 
>> No problem, this is from kbuild and most of them that I checked are actually fine,
>> we can send patches to remove these warnings.
>
>Confirmed. Now my full kernel build is complete and the above are the
>only warnings I see. So it should be easy to add the -Wpointer-arith
>check.
>

Well, after I tried to compile with it, I got lots of warnings.
Most of them that I checked is something like:

 (void *) CONST_PTR + OFFSET

in macros. In order to fix this, we need to:

 (void *) ((char *) CONST_PTR + OFFSET)

So, we have to make some efforts to kill all the warnings. :-/

Re: [2.6.37-rc1] /proc/vmstat failure.

[Wu Fengguang]

On Thu, Nov 04, 2010 at 11:31:48AM +0800, Américo Wang wrote:
> On Thu, Nov 04, 2010 at 10:53:41AM +0800, Wu Fengguang wrote:
> >On Thu, Nov 04, 2010 at 10:52:51AM +0800, Américo Wang wrote:
> >> On Thu, Nov 04, 2010 at 10:20:52AM +0800, Wu Fengguang wrote:
> >> >On Thu, Nov 04, 2010 at 05:54:48AM +0800, Tetsuo Handa wrote:
> >> >> Wu Fengguang wrote:
> >> >> > On Wed, Nov 03, 2010 at 07:11:11PM +0800, Tetsuo Handa wrote:
> >> >> > > Hello.
> >> >> > > 
> >> >> > > "cat /proc/vmstat" triggered below failure.
> >> >> > > 
> >> >> > > BUG: unable to handle kernel paging request at 07c06d16
> >> >> > > IP: [<c050c336>] strnlen+0x6/0x20
> >> >> > 
> >> >> > It seems that m->private is of type (void *) while the original v is
> >> >> > (unsigned long *). Can be fixed by the following patch.
> >> >> 
> >> >> Fixed by this patch. Thank you.
> >> >
> >> >Great, thanks!
> >> >
> >> >> By the way, GCC treats "(void *) + offset" as "(char *) + offset"
> >> >> but it is GCC specific extention. Maybe we should avoid such usage
> >> >> in order to avoid build failure for other compilers.
> >> >
> >> >"(void *) + offset" seems to be used quite often.
> >> >
> >> >I tried enable this gcc option, only to see huge amount of warnings.
> >> >
> >> >       -Wpointer-arith
> >> >           Warn about anything that depends on the "size of" a
> >> >           function type or of "void".  GNU C assigns these types a
> >> >           size of 1, for convenience in calculations with "void *"
> >> >           pointers and pointers to functions.
> >> >
> >> 
> >> Yup! I think it is good to have this.
> >> 
> >> >/cc/linux-next/scripts/mod/file2alias.c:88: warning: pointer of type ‘void *’ used in arithmetic
> >> ...
> >> >/cc/linux-next/scripts/mod/modpost.c:1565: warning: pointer of type ‘void *’ used in arithmetic
> >> >
> >> 
> >> No problem, this is from kbuild and most of them that I checked are actually fine,
> >> we can send patches to remove these warnings.
> >
> >Confirmed. Now my full kernel build is complete and the above are the
> >only warnings I see. So it should be easy to add the -Wpointer-arith
> >check.
> >
> 
> Well, after I tried to compile with it, I got lots of warnings.
> Most of them that I checked is something like:
> 
>  (void *) CONST_PTR + OFFSET
> 
> in macros. In order to fix this, we need to:
> 
>  (void *) ((char *) CONST_PTR + OFFSET)
> 
> So, we have to make some efforts to kill all the warnings. :-/

Sorry I didn't run allyesconfig/allmodconfig as suggested by Tetsuo.
So expect more warnings..

Thanks,
Fengguang