[lkml]   [2005]   [Mar]   [31]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectStack usage tasks
On Thu, Mar 31, 2005 at 05:05:48PM +0200, Jörn Engel wrote:
> On Wed, 30 March 2005 23:39:40 -0800, Yum Rayan wrote:
> >
> > Before patch
> > ------------
> > check_free_space - 128
> > do_acct_process - 105
> >
> > After patch
> > -----------
> > check_free_space - 36
> > do_acct_process - 44
> It is always nice to see enthusiams, but in your case it might be a
> bit misguided. None of the functions you worked on appear to be real
> problems wrt. stack usage.
> But if you have time to tackle some of these functions, that may make
> a real difference:
> In principle, all recursive paths should consume as little stack as
> possible. Or the recursion itself could be avoided, even better. And

Sometimes it's easy to prove that the recursion can't occur more than

Especially with a moderate stack usage, such cases are not a problem.

But auditing the recursive paths for problematic ones is still an open

> some of the call chains with ~3k of stack consumption may be
> problematic on other platforms, like the x86-64. Taking care of those
> could result in smaller stacks for the respective platform.

There's also something different that can be done:

On i386, unit-at-a-time is disabled (the only currently released version
of GNU gcc with unit-at-a-time is gcc 3.4 [1]) since gcc's stack
handling isn't very good.

With unit-at-a-time, the highest stack usage within a single function is
over 3kB.

While this is technically gcc's fault, workarounds were IMHO worth it
since unit-at-a-time gives me kernel images that are smaller by 2% [2]
and I was surprised if the speed effect wasn't positive [3].

The task I'm suggesting was therefore:
- remove the -fno-unit-at-a-time in arch/i386/Makefile in your private
kernel sources
- use gcc 3.4
- reduce the stack usages in call paths > 3kB
Note that with unit-at-a-time, gcc inline several static functions, so
the stack usage you see for a function might be accumulated from several

It's IMHO the best doing this against -mm.

I do currently not have the time for doing this, but it was something
with a real advantage for many users.

> Jörn


[1] SuSE "gcc 3.3" also supports this
[2] with -O2
[3] I do not claim it has to be measurable positive, but at least not


"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-04-06 13:31    [from the cache]
©2003-2014 Jasper Spaans. hosted at Digital Ocean