2003-09-10 17:25:45

by Petr Vandrovec

[permalink] [raw]
Subject: Random memory corruption with 2.6.0-test4-almost-test5

Hi,
yesterday evening, I run 'bk pull', and it hapilly run, and updated by
bk tree to v2.6.0-test5 tag. Today I tried to do 'bk pull' again, and BK insisted
on s.ChangeSet being corrupted. After help and hints from Larry I've found
that s.ChangeSet contained bytes 0xD1 0x80 instead of original 0x31 0x2E
at file offset 0x378C24 (whole string should look "d D 1.889.117.48", but
it looked "d D \xD1\x80889.117.48"). Corruption disappeared when I copied
enough data for s.ChangeSet being kicked from pagecache. So I know that

(1) Yesterday evening there was correct 0x31 0x2E value somewhere in memory,
as it hit disk.

(2) Today at noon there was 0xD1 0x80 (like if couple of bits were set to
different value, masking it with 0x001F mask?), but page was never
marked dirty.

(3) Corruption did not hit disk.

Any clue what's going on? System is built with DEBUG_SLAB & so on, and it
does not complain about bitflips, just these two changed bytes in an innocent
file... And surprisingly I do not have loaded vmware modules, as I did not
build them for this kernel yet... lsmod + recent msg + bk output is below,
.config + boot dmesg attached.
Thanks,
Petr Vandrovec
[email protected]

vana:/var/log# lsmod
Module Size Used by
lp 9216 0
tvaudio 20876 0
bttv 151268 1
tuner 14732 0
video_buf 19332 1 bttv
btcx_risc 4232 1 bttv
v4l2_common 4224 1 bttv
videodev 7936 2 bttv
parport_pc 26068 1
parport 28512 2 lp,parport_pc
i810_audio 35860 1
ac97_codec 17420 1 i810_audio
soundcore 8640 3 bttv,i810_audio
nfsd 156448 8
exportfs 7680 1 nfsd
lockd 61680 2 nfsd
sunrpc 150344 2 nfsd,lockd
nls_cp852 4992 1
nls_iso8859_2 4736 1
ncpfs 59424 1
vana:/var/log# tail -12500 alllog | grep 'kernel:'
Sep 9 04:01:39 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 9 04:03:04 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 9 04:27:15 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 9 04:52:16 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 9 05:20:22 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 10 04:04:43 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 10 04:41:12 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 10 05:03:08 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 10 05:23:10 vana kernel: bttv1: skipped frame. no signal? high irq latency?
Sep 10 06:05:01 vana kernel: bttv1: skipped frame. no signal? high irq latency?

> vana:/mnt2/usr/src/bk/linus/linux-2.5$ bk pull
> Pull bk://linux.bkbits.net/linux-2.5
> -> file://mnt2/usr/src/bk/linus/linux-2.5
> ---------------------- Receiving the following csets -----------------------
> 1.1238 1.1237 1.1232.1.1 1.1236 1.1235 1.1234 1.1233
> ----------------------------------------------------------------------------
> ChangeSet: 8 deltas
> drivers/block/noop-iosched.c: 1 deltas
> include/sound/hdsp.h: 1 deltas
> include/sound/sscape_ioctl.h: 1 deltas
> drivers/serial/sa1100.c: 1 deltas
> Makefile: 1 deltas
> drivers/net/wan/sdla.c: 1 deltas
> include/asm-i386/ioctl.h: 1 deltas
> ---------------------------------------------------------------------------
> takepatch: saved entire patch in PENDING/2003-09-09.01
> ---------------------------------------------------------------------------
> Applying 8 revisions to ChangeSet
> Applying 1 revisions to drivers/block/noop-iosched.c
> Applying 1 revisions to include/sound/hdsp.h
> Applying 1 revisions to include/sound/sscape_ioctl.h
> Applying 1 revisions to drivers/serial/sa1100.c
> Applying 1 revisions to Makefile
> Applying 1 revisions to drivers/net/wan/sdla.c
> Applying 1 revisions to include/asm-i386/ioctl.h
> takepatch: 15 new revisions, 0 conflicts in 8 files
> 3285 bytes uncompressed to 11670, 3.55X expansion
> Running resolve to apply new work ...
> resolve: applied 8 files in pass 4
> resolve: running consistency check, please wait...
> 100% |=================================================================| OK
> Consistency check passed, resolve complete.

[go home, return back]

> vana:/mnt2/usr/src/bk/linus/linux-2.5$ bk pull
> Pull bk://linux.bkbits.net/linux-2.5
> -> file://mnt2/usr/src/bk/linus/linux-2.5
> Bad old style checksum for SCCS/s.ChangeSet, got 17365, wanted 29155
> Bad checksum for SCCS/s.ChangeSet, got 29397, wanted 29155
> Can't init changeset


Attachments:
(No filename) (4.64 kB)
.config (26.07 kB)
dmesg (11.91 kB)
Download all attachments