lkml.org 
[lkml]   [2009]   [Nov]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH RFC] x86: fix confusing name of /proc/cpuinfo "ht" flag

* H. Peter Anvin <hpa@zytor.com> wrote:

> On 11/12/2009 10:37 AM, Dave Jones wrote:
> > On Thu, Nov 12, 2009 at 06:59:08PM +0100, Ingo Molnar wrote:
> > > > It's an ABI. Keep it stable, please.
> > >
> > > That's generally true, but i'm not suggesting that: i'm suggesting to
> > > _clear_ the HT flag from the cpufeatures if there's only one sibling.
> > > It's meaningless in that case and as the link quoted by the original
> > > patch shows many people are confused by that.
> > >
> > > I have such a box so i can test it. (but i dont expect any problems)
> >
> > I agree that it's an ABI change, but any software depending on its current
> > state has to implement a fallback for the case where 'ht' isn't present anyway
> > unless there's some program that only runs on ht capable hardware, which
> > sounds just crazy.
> >
> > The only potential for breakage that I can see is that code that is tuned
> > to be run in the HT case will stop running in cases where it shouldn't.
> > Which sounds like a positive thing to me.
>
> The most likely breakage would be some stupid licensing scheme.
>
> The other aspect is that we as much as possible have tried to stay to
> the hardware-documented names of these strings. Inventing new strings
> is generally a bad idea.

Agreed - and we rejected such patches a couple of times in the past and
for good reasons. Some /proc details are rarely used by apps (so they
are no real ABIs) but cpuinfo is frequently parsed.

Clearing the ht flag on non-hyperthreading CPUs would be a limited
quirk/fix in essence applicable to a relatively narrow range of CPUs -
and easily undone, should it cause any problems. So if Bart wants to
take a stab at that it would be a nice solution to the problem at hand
...

Ingo


\
 
 \ /
  Last update: 2009-11-13 08:45    [W:0.137 / U:0.156 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site