All of lore.kernel.org
 help / color / mirror / Atom feed
* [Xenomai] Lacking xsave support in Xenomai
@ 2013-01-25 14:06 Jan Kiszka
  2013-01-25 18:44 ` Gilles Chanteperdrix
  0 siblings, 1 reply; 4+ messages in thread
From: Jan Kiszka @ 2013-01-25 14:06 UTC (permalink / raw)
  To: Xenomai

Hi all,

yesterday I pushed a fix to my for-upstream queue that properly disables
all xsave-dependent CPU features (AVX/AVX2 namely). However, this is no
real solution for us as it blocks a lot of acceleration potential on
current Intel CPUs.

Did anyone already look into this in more details? Is it just a matter
of implementing the necessary context switching bits? Are there known
pitfalls?

Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SDP-DE
Corporate Competence Center Embedded Linux


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [Xenomai] Lacking xsave support in Xenomai
  2013-01-25 14:06 [Xenomai] Lacking xsave support in Xenomai Jan Kiszka
@ 2013-01-25 18:44 ` Gilles Chanteperdrix
  2013-01-25 19:01   ` Jan Kiszka
  0 siblings, 1 reply; 4+ messages in thread
From: Gilles Chanteperdrix @ 2013-01-25 18:44 UTC (permalink / raw)
  To: Jan Kiszka; +Cc: Xenomai

On 01/25/2013 03:06 PM, Jan Kiszka wrote:

> Hi all,
> 
> yesterday I pushed a fix to my for-upstream queue that properly disables
> all xsave-dependent CPU features (AVX/AVX2 namely). However, this is no
> real solution for us as it blocks a lot of acceleration potential on
> current Intel CPUs.
> 
> Did anyone already look into this in more details? Is it just a matter
> of implementing the necessary context switching bits? Are there known

> pitfalls?


The known pitfalls of FPU support is that we never get it right the
first time. And debugging why it fails is hard.

-- 
                                                                Gilles.


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [Xenomai] Lacking xsave support in Xenomai
  2013-01-25 18:44 ` Gilles Chanteperdrix
@ 2013-01-25 19:01   ` Jan Kiszka
  2013-01-25 19:43     ` Gilles Chanteperdrix
  0 siblings, 1 reply; 4+ messages in thread
From: Jan Kiszka @ 2013-01-25 19:01 UTC (permalink / raw)
  To: Gilles Chanteperdrix; +Cc: Xenomai

On 2013-01-25 19:44, Gilles Chanteperdrix wrote:
> On 01/25/2013 03:06 PM, Jan Kiszka wrote:
> 
>> Hi all,
>>
>> yesterday I pushed a fix to my for-upstream queue that properly disables
>> all xsave-dependent CPU features (AVX/AVX2 namely). However, this is no
>> real solution for us as it blocks a lot of acceleration potential on
>> current Intel CPUs.
>>
>> Did anyone already look into this in more details? Is it just a matter
>> of implementing the necessary context switching bits? Are there known
> 
>> pitfalls?
> 
> The known pitfalls of FPU support is that we never get it right the
> first time. And debugging why it fails is hard.

OK, so no conceptual issues. And it will remain controllable via "noxsave".

Still need to discuss details here, but I bet we will look into this soon.

Thanks,
Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SDP-DE
Corporate Competence Center Embedded Linux


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [Xenomai] Lacking xsave support in Xenomai
  2013-01-25 19:01   ` Jan Kiszka
@ 2013-01-25 19:43     ` Gilles Chanteperdrix
  0 siblings, 0 replies; 4+ messages in thread
From: Gilles Chanteperdrix @ 2013-01-25 19:43 UTC (permalink / raw)
  To: Jan Kiszka; +Cc: Xenomai

On 01/25/2013 08:01 PM, Jan Kiszka wrote:

> On 2013-01-25 19:44, Gilles Chanteperdrix wrote:
>> On 01/25/2013 03:06 PM, Jan Kiszka wrote:
>>
>>> Hi all,
>>>
>>> yesterday I pushed a fix to my for-upstream queue that properly disables
>>> all xsave-dependent CPU features (AVX/AVX2 namely). However, this is no
>>> real solution for us as it blocks a lot of acceleration potential on
>>> current Intel CPUs.
>>>
>>> Did anyone already look into this in more details? Is it just a matter
>>> of implementing the necessary context switching bits? Are there known
>>
>>> pitfalls?
>>
>> The known pitfalls of FPU support is that we never get it right the
>> first time. And debugging why it fails is hard.
> 
> OK, so no conceptual issues. And it will remain controllable via "noxsave".
> 
> Still need to discuss details here, but I bet we will look into this soon.


You may want to use the "fpu tracer" patches:
http://git.xenomai.org/?p=ipipe-gch.git;a=shortlog;h=refs/heads/core-3.4-fpu_trace
http://git.xenomai.org/?p=xenomai-gch.git;a=shortlog;h=refs/heads/2.6-fpu_trace


-- 
                                                                Gilles.


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2013-01-25 19:43 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-01-25 14:06 [Xenomai] Lacking xsave support in Xenomai Jan Kiszka
2013-01-25 18:44 ` Gilles Chanteperdrix
2013-01-25 19:01   ` Jan Kiszka
2013-01-25 19:43     ` Gilles Chanteperdrix

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.