Re: [bhavesh@avaya.com] libc/3259: semctl on PPC barfs expecting optional 4th argument

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

On Monday 15 April 2002 19:32, Ulrich Drepper wrote:
> On Mon, 2002-04-15 at 05:57, Franz Sirl wrote:
> > I guess, probably semctl.c should be
> > rewritten to only do va_arg if the IPC code needs the 4th argument.
>
> Contributions welcome.

Well, give me some time, I'm busy enough with gcc-3.1 in my spare time :-). 
BTW, is there a release pending? The 2.3cvs relabeled as 2.2.5 in rh skipjack 
puzzles me a bit... I would like to have the the gcc-3.1 warning fixes and 
the prelinking stuff so users can play with it, but I don't know yet if I'm 
brave enough to follow rh there...

> > Also there seem to be way to many semctl.c in the tree, why do we have a
> > lot of linux/<cpu>/semctl.c including linux/i386/semctl.c when there is a
> > generic linux/semctl.c?
>
> The kernel side is a mess with differences among the architectures which
> make it hard to have a single, optimized user-level implementation.  If
> you want to take a look at this you're welcome.  But be aware that each
> implementation should be at least as efficient at runtime as the current
> code.

Well, I know that, but in the semctl.c case I was puzzled to see nearly nobody 
uses linux/semctl.c, but a lot use the x86 semctl.c. For example I can't 
believe s390, arm or sh ever used 16-bit UID's, which seems to be the only 
legitimate reason to use the x86 version.

Franz.