Re: [linux-sh:03188] cache info

Paul Mundt <[email protected]> Sat, 20 Mar 2004 01:16:54 -0500
Newsgroups gmane.linux.ports.sh.devel,gmane.linux.ports.sh.general
Message-ID <[email protected]>
On Sat, Mar 20, 2004 at 12:54:00PM +0900, Kaz Kojima wrote:
> (1)  SH specific system call to get the cache info,
> (2)  interface (including system call) to read hurmless CPU
>      registers,
> (3)  special device,
> (4)  ...
> 
> Perhaps (1) is cleaner as an interface than another methods though
> I'm a bit concerned with SH specific system call number.  We have
> "holes" in system call numbers but there might be no guarantee
> that they won't be used for another purpose.
> 
> Thoughts?  Implementations will be more welcome :-)
> 
Since the system call interface is likely going to be the cleanest approach
for this sort of thing, why not implement this as a vsyscall? There was a
recent thread on l-k about using vsyscalls for optimized routines on ppc/ppc64,
so exposing this sort of thing through a vsyscall doesn't seem particularly
evil (especially if glibc is going to need this) -- note that x86_64 already
does a number of things in this fashion, gettimeofday(), time(), etc.

The interface on the kernel side would be relatively simple, since would just
populate the cache info by way of detect_cpu_and_cache_system() as normal,
and then hand this data back by way of a vsyscall page.
signature.asc (application/pgp-signature, 189 B)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)

iD8DBQFAW+HW1K+teJFxZ9wRAh6yAKCCJyq/015oBg5zIpMzQnSaB/p93gCfcVz+
vQgi5bBG17s6CFHfZ+zCX/0=
=iG+k
-----END PGP SIGNATURE-----