Re: [Alsa-devel] Slab corruptions & Re: 2.6.17-rc1: Oops in sound applications

From: Jan Niehusmann
Date: Tue Apr 04 2006 - 19:20:41 EST

On Tue, Apr 04, 2006 at 09:12:18PM +0200, Takashi Iwai wrote:
> At Tue, 4 Apr 2006 21:06:31 +0200,
> Jan Niehusmann wrote:
> >
> > On Tue, Apr 04, 2006 at 05:56:36PM +0200, Takashi Iwai wrote:
> > > Could you try the patch below by OGAWA Hirofumi
> > > <hirofumi@xxxxxxxxxxxxxxxxxx>?
> >
> > Sorry, no success with that patch applied:
> >
> > Apr 4 20:18:00 knautsch kernel: [17179569.184000] Linux version 2.6.17-rc1-g231a1569-dirty (root@knautsch) (gcc version 4.0.3 (Debian 4.0.3-1)) #1 Tue Apr 4 19:37:29 CEST 2006
> >
> > (This is 2.6.17-rc1 with the patch you sent applied)
> >
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] BUG: unable to handle kernel paging request at virtual address 6b6b6b6b
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] printing eip:
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] e04a2eb6
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] *pde = 00000000
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] Oops: 0000 [#1]
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] Modules linked in: tun ipv6 lp i915 drm rfcomm l2cap bluetooth thermal processor fan button battery asus_acpi ac usb_storage usbmouse sbp2 usbkbd autofs4 af_packet nls_iso8859_1 nls_cp437 usbhid cm4000_cs 8250_pci 8250 serial_core eth1394 pcmcia snd_intel8x0 snd_pcm_oss snd_mixer_oss snd_intel8x0m snd_ac97_codec snd_ac97_bus ipw2200 ieee80211 ieee80211_crypt 8139too mii nsc_ircc irda snd_pcm snd_timer ohci1394 ieee1394 evdev joydev crc_ccitt parport_pc parport pcspkr firmware_class yenta_socket rsrc_nonstatic pcmcia_core rtc snd soundcore snd_page_alloc ehci_hcd uhci_hcd usbcore intel_agp agpgart
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] CPU: 0
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] EIP: 0060:[pg0+537210550/1069212672] Not tainted VLI
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] EFLAGS: 00210002 (2.6.17-rc1-g231a1569-dirty #1)
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] EIP is at snd_pcm_lib_write1+0x196/0x4a0 [snd_pcm]
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] eax: 000009c4 ebx: cc785a84 ecx: de505e54 edx: 6b6b6b6b
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] esi: 0000062c edi: 00000000 ebp: dede51a4 esp: de505e78
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] ds: 007b es: 007b ss: 0068
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] Process twinkle (pid: 4936, threadinfo=de505000 task=cae0f550)
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] Stack: <0>dede51a4 000015cd e07a5000 00000b42 0000003c 00000001 cc785b28 c84ef000
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] e07a5000 00000b7e 00000b7e 00000000 00000000 0001c639 00000000 cae0f550
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] c0114d60 cc785b28 cc785b28 c0000000 de505000 dede51a4 cc785a84 e0547727
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] Call Trace:
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] <c0114d60> default_wake_function+0x0/0x20 <e0547727> snd_pcm_oss_write3+0x67/0xe0 [snd_pcm_oss]
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] <e04a0250> snd_pcm_lib_write_transfer+0x0/0xc0 [snd_pcm] <e054949f> snd_pcm_plug_write_transfer+0x8f/0xe0 [snd_pcm_oss]
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] <e054781a> snd_pcm_oss_write2+0x7a/0x110 [snd_pcm_oss] <e054892e> snd_pcm_oss_write+0x10e/0x1f0 [snd_pcm_oss]
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] <c0158f85> vfs_write+0xb5/0x190 <c0159a1b> sys_write+0x4b/0x80
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] <c0102e97> syscall_call+0x7/0xb
> > Apr 4 20:25:51 knautsch kernel: [17180076.984000] Code: b8 c4 09 00 00 e8 db fb e5 df fa 85 c0 0f 85 05 02 00 00 8b 93 98 00 00 00 8b 02 83 f8 02 74 0b 8b 02 83 f8 06 0f 85 da 02 00 00 <8b> 02 83 f8 05 0f 8f ed 01 00 00 83 f8 04 0f 8d 1e 02 00 00 48
> >
> > This oops is different from the others in that it contains the 6b6b6b6b
> > magic from slab poisoning. But, again, this is clearly sound related:
> > EIP points to snd_pcm_lib_write, and the trigger procedure was the
> > following:
> >
> > 1) Use twinke to make a VoIP-Call. Sound was a little bit choppy, so
> > there is already something wrong here. No oops or other error message
> > yet, though.
> >
> > 2) From a second xterm, call "yes >/dev/dsp"
> >
> > This lead to a beeping sound (as expected), so opening the device and
> > sending sound was successful - and to the oops quoted above.
>
> What happens if you copy the whole subtree linux/sound and
> linux/include/sound from 2.6.16?
>
> I don't figure out yet what could break except for that typo.

After the sound tree from 2.6.16 did work fine, I used git-bisect to
find the offending commit:

$git-bisect good
3bf75f9b90c981f18f27a0d35a44f488ab68c8ea is first bad commit
diff-tree 3bf75f9b90c981f18f27a0d35a44f488ab68c8ea (from bf1bbb5a49eec51c30d341606885507b501b37e8)
Author: Takashi Iwai <tiwai@xxxxxxx>
Date: Mon Mar 27 16:40:49 2006 +0200

[ALSA] Clean up PCM codes (take 2)

- Clean up initialization and destruction of substream instance
Now snd_pcm_open_substream() alone does most initialization jobs.
Add pcm_release callback for cleaning up at snd_pcm_release_substream()
- Tidy up PCM oss code

Signed-off-by: Takashi Iwai <tiwai@xxxxxxx>

:040000 040000 df14b16c98da02d8aea1e5ee18baae5caae6245b a6779060f9e41f5033376735f9b290ba5db98b6e M include
:040000 040000 7578b410668bcbfc2ed1f3f31011270e6e44f716 c70f96dd906c74d559385ef08ac6e518132936d1 M sound

...which happens to be the commit which contains the bug you already
mentioned. I wonder if there is a second one hidden somewhere in that
commit, or if git-bisect led me to that bug while the second one is
hidden in a different commit...

Jan

-
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/