[Xenomai-core] [BUG] rtf_put broken due to kernel 2.4 bug

[Xenomai-core] [BUG] rtf_put broken due to kernel 2.4 bug

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 935 bytes --]

Hi,

this issue

http://www.ussg.iu.edu/hypermail/linux/kernel/0610.1/0211.html

was revealed by the rtai fifo code. We currently have the double
__test_and_set_bit combination at least here:

http://www.rts.uni-hannover.de/xenomai/lxr/source/ksrc/skins/rtai/fifo.c#L309
(resolves to __test_and_set_bit via ipipe_test_and_stall_pipeline_from)
http://www.rts.uni-hannover.de/xenomai/lxr/source/ksrc/skins/rtai/fifo.c#L311

The effect on this code is that multiple rtf_put on an unconnected fifo
corrupts its queue.

I'm now awaiting the comment of the 2.4 maintainer, but we will still
have to cope with this issue on current 2.4 kernels. There is also the
risk that more spots are affected or will be affected in the future
given the embedded use of __test_and_set_bit in pipeline stalling commands.

Anyone any smart idea how to handle this? Should we simply include the
bitops patch in 2.4-x86-ipipe?

Jan


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

Re: [Xenomai-core] [BUG] rtf_put broken due to kernel 2.4 bug

[Philippe Gerum]

On Mon, 2006-10-09 at 00:10 +0200, Jan Kiszka wrote:
> Hi,
> 
> this issue
> 
> http://www.ussg.iu.edu/hypermail/linux/kernel/0610.1/0211.html
> 
> was revealed by the rtai fifo code. We currently have the double
> __test_and_set_bit combination at least here:
> 
> http://www.rts.uni-hannover.de/xenomai/lxr/source/ksrc/skins/rtai/fifo.c#L309
> (resolves to __test_and_set_bit via ipipe_test_and_stall_pipeline_from)
> http://www.rts.uni-hannover.de/xenomai/lxr/source/ksrc/skins/rtai/fifo.c#L311
> 
> The effect on this code is that multiple rtf_put on an unconnected fifo
> corrupts its queue.
> 
> I'm now awaiting the comment of the 2.4 maintainer, but we will still
> have to cope with this issue on current 2.4 kernels. There is also the
> risk that more spots are affected or will be affected in the future
> given the embedded use of __test_and_set_bit in pipeline stalling commands.
> 
> Anyone any smart idea how to handle this? Should we simply include the
> bitops patch in 2.4-x86-ipipe?

Yes, I'm going to wire this patch into 1.3-01 for 2.4.33.

> 
> Jan
> 
> _______________________________________________
> Xenomai-core mailing list
> Xenomai-core@domain.hid
> https://mail.gna.org/listinfo/xenomai-core
-- 
Philippe.