[lkml]   [2011]   [Mar]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [RFC v2 8/8] x86, xsave: remove lazy allocation of xstate area
On Thu, Mar 24, 2011 at 07:39:13AM -0400, Brian Gerst wrote:
> On Wed, Mar 23, 2011 at 11:27 AM, Hans Rosenfeld <> wrote:
> > This patch completely removes lazy allocation of the xstate area. All
> > tasks will always have an xstate area preallocated, just like they
> > already do when non-lazy features are present. The size of the xsave
> > area ranges from 112 to 960 bytes, depending on the xstates present and
> > enabled. Since it is common to use SSE etc. for optimization, the actual
> > overhead is expected to negligible.
> >
> > This removes some of the special-case handling of non-lazy xstates. It
> > also greatly simplifies init_fpu() by removing the allocation code, the
> > check for presence of the xstate area or init_fpu() return value.
> >
> > Signed-off-by: Hans Rosenfeld <>
> I'm not sure I like this. I did a quick test on 64-bit, and found
> that while most if not all user processes allocated the fpu save area
> (probably because of glibc blindly initializing the fpu), kernel
> threads did not. This patch would force kernel threads to allocate
> memory they would never use.

Yes, up to a few kilobytes would be wasted by kernel threads. The
related code gets much simpler. I think that is a good thing.

Anyway, the patch is not essential for the rework and LWP support, so I
don't really care that much about it.

Did you take a look at the other patches? I haven't yet received a
single comment on them.


%SYSTEM-F-ANARCHISM, The operating system has been overthrown

 \ /
  Last update: 2011-03-29 16:19    [W:0.160 / U:0.024 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site