[Xenomai-help] BUG fs/buffer.c:1245

[Xenomai-help] BUG fs/buffer.c:1245

[Steven Seeger]

[-- Attachment #1: Type: text/plain, Size: 3054 bytes --]

I get the following bug when sending data over tcp/ip from a linux
thread with xenomai apps running:

 

The reason I ask the list is because the call that triggers this is
irqs_disabled().

 

What I'm doing is sending data over an RT_PIPE from a realtime task,
while a linux thread wakes up and receives a message from the pipe and
then sends it out over tcp/ip.

 

Steven

 

kernel BUG at fs/buffer.c:1245!

invalid opcode: 0000 [#1]

PREEMPT

Modules linked in: nfs lockd sunrpc st_dacsnd(P) eepro100 xeno_native
xeno_nucleus

CPU:    0

EIP:    0060:[<c017e9a2>]    Tainted: P        VLI

EFLAGS: 00010046   (2.6.23.12-target-xenomai #3)

EIP is at __find_get_block+0x182/0x1a0

eax: 00000000   ebx: 00000000   ecx: 00000400   edx: 00033a01

esi: c10551a0   edi: 00033a01   ebp: 00000000   esp: c187dc08

ds: 007b   es: 007b   fs: 0000  gs: 0033  ss: 0068

Process rtlogd (pid: 881, ti=c187c000 task=c13b6aa0 task.ti=c187c000)<0>

I-pipe domain Linux

Stack: c1111380 c11114d0 00000000 00000001 00000400 00000000 c017e9d6
c1c69700

       00000000 00000001 00000400 00000000 c017e9d6 00000001 c10ee400
00000000

       00033a01 c10551a0 00000000 00000001 c1b92090 00000000 c01a124b
00000001

Call Trace:

 [<c017e9d6>] __getblk+0x16/0x210

 [<c017e9d6>] __getblk+0x16/0x210

 [<c01a124b>] ext3_getblk+0xfb/0x230

 [<c01a4a79>] dx_probe+0x89/0x2f0

 [<c01a5745>] ext3_find_entry+0x235/0x620

 [<c017e8c4>] __find_get_block+0xa4/0x1a0

 [<c01a008a>] __ext3_get_inode_loc+0xea/0x350

 [<c017e9d6>] __getblk+0x16/0x210

 [<c0117afc>] do_softirq+0x4c/0x60

 [<c01a33f2>] ext3_read_inode+0x1f2/0x320

 [<c0159a33>] kmem_cache_alloc+0x53/0xc0

 [<c016e3b5>] d_rehash+0x15/0x50

 [<c01a718c>] ext3_lookup+0x3c/0x120

 [<c016f42c>] d_alloc+0x1c/0x1a0

 [<c0164d02>] do_lookup+0x122/0x1a0

 [<c0166879>] __link_path_walk+0x7f9/0xeb0

 [<c010bdce>] __ipipe_pin_range_globally+0xfe/0x110

 [<c0166f73>] link_path_walk+0x43/0xc0

 [<c2950b78>] xnarch_trap_fault+0x18/0x20 [xeno_nucleus]

 [<c0275e6a>] exception_event+0x2a/0x40

 [<c015b72f>] get_unused_fd_flags+0x4f/0xe0

 [<c0167225>] do_path_lookup+0x85/0x200

 [<c015e6a3>] get_empty_filp+0x33/0x120

 [<c0167e81>] __path_lookup_intent_open+0x51/0xa0

 [<c0167f60>] path_lookup_open+0x20/0x30

 [<c016806f>] open_namei+0x6f/0x640

 [<c014b636>] acct_stack_growth+0xa6/0xc0

 [<c015baa3>] do_filp_open+0x33/0x60

 [<c2950b78>] xnarch_trap_fault+0x18/0x20 [xeno_nucleus]

 [<c0275e6a>] exception_event+0x2a/0x40

 [<c015b72f>] get_unused_fd_flags+0x4f/0xe0

 [<c015bb20>] do_sys_open+0x50/0xf0

 [<c015bbfc>] sys_open+0x1c/0x20

 [<c0102b81>] syscall_call+0x7/0xb

 =======================

Code: ff ff 83 68 14 01 8b 40 08 a8 04 75 2c 85 ed 0f 84 30 ff ff ff 89
e8 e8 7d f6 ff ff e9 24 ff ff ff 89 d8 e8 71 f6 ff ff 90 eb 81 <0f> 0b
eb fe e8 35 36 18 00 e9 08 ff ff ff e8 2b 36 18 00 eb cd

EIP: [<c017e9a2>] __find_get_block+0x182/0x1a0 SS:ESP 0068:c187dc08

/ #

 


[-- Attachment #2: Type: text/html, Size: 12746 bytes --]

Re: [Xenomai-help] BUG fs/buffer.c:1245

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 780 bytes --]

Steven Seeger wrote:
> I get the following bug when sending data over tcp/ip from a linux
> thread with xenomai apps running:
> 
>  
> 
> The reason I ask the list is because the call that triggers this is
> irqs_disabled().
> 
>  
> 
> What I'm doing is sending data over an RT_PIPE from a realtime task,
> while a linux thread wakes up and receives a message from the pipe and
> then sends it out over tcp/ip.

Sporadic or deterministic bug? Specifically in case of the latter I 
would be interested in your .config.

Unless the slowdown papers over the actual issue, letting the ipipe 
tracer run, so that it reports a nice backtrace on the BUG, may help us 
tracking down the point where irq_enable is missing (also in 
non-Xenomai/Ipipe code).

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]

Re: [Xenomai-help] BUG fs/buffer.c:1245

[Steven Seeger]

It's very hard to reproduce. I found via google a bug report about the
same bug, but it turned out to be a patch in the e1000 driver. I'm using
the eepro100 driver. I wonder if the issue is related? I should see if
it's fixed in 2.6.24.

Steven
> 
> > -----Original Message-----
> > From: jan.kiszka@domain.hid [mailto:jan.kiszka@domain.hid]
> > Sent: Friday, February 08, 2008 11:30 AM
> > To: Steven Seeger
> > Cc: xenomai@xenomai.org
> > Subject: Re: [Xenomai-help] BUG fs/buffer.c:1245
> >
> > Sporadic or deterministic bug? Specifically in case of the latter I
> > would be interested in your .config.
> >
> > Unless the slowdown papers over the actual issue, letting the ipipe
> > tracer run, so that it reports a nice backtrace on the BUG, may help
us
> > tracking down the point where irq_enable is missing (also in
> > non-Xenomai/Ipipe code).
> >
> > Jan