Skip to contentSkip to conversation

Atom processor inclusion

DISCUSSION Kelly Bowa 5 participants 8 messages Aug 20, 2009 – Aug 21, 2009

lore
Share & tools
lore
#2 lore raw

Hi,

Am Donnerstag, 20. August 2009 12:50:29 schrieb Ingo Molnar:

Quoted 2 lines

Yep, it looked acceptable - Tobias, do you have any updates / latest version of that patch?

No - it's still the improved version I posted at the end of May [1]. The question is what to do with MODULE_PROC_FAMILY (CORE2 or ATOM) and the mtune- fallback (generic, i686, ...)?

Regards,

Tobias

[1] http://lkml.org/lkml/2009/5/29/374

#3 lore raw
Quoted 9 lines · Tobias Doerffel

Hi,

Am Donnerstag, 20. August 2009 12:50:29 schrieb Ingo Molnar:

Quoted 2 lines

Yep, it looked acceptable - Tobias, do you have any updates / latest version of that patch?

No - it's still the improved version I posted at the end of May [1]. The question is what to do with MODULE_PROC_FAMILY (CORE2 or ATOM) and the mtune- fallback (generic, i686, ...)?

Without benchmarks, we're flying blind on that one... although in general, "generic" is probably best in the sense that it doesn't imply that anything else has been done to it.

As far as MODULE_PROC_FAMILY it really comes down to if we use movbe or not, which I don't believe your patch does. On the other hand, I really think it's extremely unlikely that anyone will use modules compiled for a different CPU, so I'm personally fine with changing that string.

That whole mechanism is kind of broken, anyway.

-hpa

#4 lore raw
Quoted 14 lines · H. Peter Anvin
Quoted 8 lines · Tobias Doerffel

Hi,

Am Donnerstag, 20. August 2009 12:50:29 schrieb Ingo Molnar:

Quoted 2 lines

Yep, it looked acceptable - Tobias, do you have any updates / latest version of that patch?

No - it's still the improved version I posted at the end of May [1]. The question is what to do with MODULE_PROC_FAMILY (CORE2 or ATOM) and the mtune- fallback (generic, i686, ...)?

Without benchmarks, we're flying blind on that one... although in general, "generic" is probably best in the sense that it doesn't imply that anything else has been done to it.

Quoted 7 lines

As far as MODULE_PROC_FAMILY it really comes down to if we use movbe or not, which I don't believe your patch does. On the other hand, I really think it's extremely unlikely that anyone will use modules compiled for a different CPU, so I'm personally fine with changing that string.

That whole mechanism is kind of broken, anyway.

personally, I would prefer it if we did a simple hash of the WHOLE cflags, and put that into the module version string.
Anything else is just a weak, and useless, substitute for that.

Using different CFLAGS in any shape or form should disqualify the module as "incompatible".. and a simple hash is sufficient for that.....

#5 lore raw
Quoted 25 lines · Arjan van de Ven
Quoted 16 lines · H. Peter Anvin
Quoted 10 lines · Tobias Doerffel

Hi,

Am Donnerstag, 20. August 2009 12:50:29 schrieb Ingo Molnar:

Quoted 2 lines

Yep, it looked acceptable - Tobias, do you have any updates / latest version of that patch?

No - it's still the improved version I posted at the end of May [1]. The question is what to do with MODULE_PROC_FAMILY (CORE2 or ATOM) and the mtune-
fallback (generic, i686, ...)?

Without benchmarks, we're flying blind on that one... although in general, "generic" is probably best in the sense that it doesn't imply that anything else has been done to it.

Quoted 7 lines

As far as MODULE_PROC_FAMILY it really comes down to if we use movbe or not, which I don't believe your patch does. On the other hand, I really think it's extremely unlikely that anyone will use modules compiled for a different CPU, so I'm personally fine with changing that string.

That whole mechanism is kind of broken, anyway.

Quoted 7 lines

personally, I would prefer it if we did a simple hash of the WHOLE cflags, and put that into the module version string. Anything else is just a weak, and useless, substitute for that.

Using different CFLAGS in any shape or form should disqualify the module as "incompatible".. and a simple hash is sufficient for that.....

makes sense.

Ingo

#6 lore raw
Quoted 3 lines · Ingo Molnar

makes sense.

OK, but I think that's a separate project. For now, I'll put the patch as-is into tip:x86/cpu.

-hpa

#7 lore raw
Quoted 7 lines · H. Peter Anvin
Quoted 3 lines · Ingo Molnar

makes sense.

OK, but I think that's a separate project. For now, I'll put the patch as-is into tip:x86/cpu.

I agree with the review feedback given - we can make merge of that dependent on measurements, etc.

Ingo