Re: [Xenomai-core] [Xenomai-commits] r4392 - /trunk/include/asm-x86/arith_64.h

Re: [Xenomai-core] [Xenomai-commits] r4392 - /trunk/include/asm-x86/arith_64.h

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 713 bytes --]

Gilles Chanteperdrix wrote:
> Author: gch
> Date: Sun Nov 16 03:30:40 2008
> New Revision: 4392
> 
> URL: http://svn.gna.org/viewcvs/xenomai?rev=4392&view=rev
> Log:
> Implement nodiv_ullimd on x86_64
> 
> Modified:
>     trunk/include/asm-x86/arith_64.h

Nice. This solves the user-triggerable kernel oops due to idiv overflows
in the original version (rt_task_sleep_until(1LL<<63);). I happened to
receive such a bug report on Friday and was about to consider
alternative solutions beyond limit checks.

But how accurate is this conversion? I mean how many bits can we lose
when doing xnarch_ns_to_tsc(xnarch_tsc_to_ns(x))?

Final question: Already prepared a 32-bit version? :)

Jan



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]

Re: [Xenomai-core] [Xenomai-commits] r4392 - /trunk/include/asm-x86/arith_64.h

[Gilles Chanteperdrix]

Jan Kiszka wrote:
> Gilles Chanteperdrix wrote:
>> Author: gch
>> Date: Sun Nov 16 03:30:40 2008
>> New Revision: 4392
>>
>> URL: http://svn.gna.org/viewcvs/xenomai?rev=4392&view=rev
>> Log:
>> Implement nodiv_ullimd on x86_64
>>
>> Modified:
>>     trunk/include/asm-x86/arith_64.h
> 
> Nice. This solves the user-triggerable kernel oops due to idiv overflows
> in the original version (rt_task_sleep_until(1LL<<63);). I happened to
> receive such a bug report on Friday and was about to consider
> alternative solutions beyond limit checks.
> 
> But how accurate is this conversion? I mean how many bits can we lose
> when doing xnarch_ns_to_tsc(xnarch_tsc_to_ns(x))?

Thanks to a little addition, nodiv_ullimd has the same precision as ullimd.

> 
> Final question: Already prepared a 32-bit version? :)

Yes, the generic vercion should work, provided we implement
__rthal_add96and64
which does an addl and two adcl
-- 
					    Gilles.