LWN: Comments on "How to speed up system calls"

User space gettimeofday()

jmshh — Fri, 20 Dec 2002 20:54:47 +0000

None. But there were two conditions, the other being "infrequently
changing information". Updates require cache invalidation, and this is
better done when the data is really needed only.

Does anyone know...

proski — Fri, 20 Dec 2002 16:05:37 +0000

According to AMD Processor Recognition Application Note, the SYSENTER and SYSEXIT instructions are supported on AMD Athlon and Duron, but not on K6 (see tables at the end).

Does anyone know...

hjernemadsen — Fri, 20 Dec 2002 11:53:30 +0000

If I remember correctly, the sysenter instruction is not present on AMD
processors. But it really doesn't matter as the normal int $0x80 approach
on an AMD is faster than sysenter on a P4...

Does anyone know...

wolfrider — Fri, 20 Dec 2002 08:52:37 +0000

If this will affect AMD, etc processors?

User space gettimeofday()

acristianb — Fri, 20 Dec 2002 00:19:35 +0000

What are the side effects of gettimeofday?

User space getpid()

ncm — Thu, 19 Dec 2002 05:53:25 +0000

While getpid() could certainly be put in user space, the
number of programs that it would speed up noticeably
could probably be counted on a pig's hoof. It would be
pretty surprising if any syscall used frequently enough
to make a difference could be handled in user space.

gettimeofday() is called so frequently in real programs
that any speedup matters, at least for some programs.
Few other syscalls have that property.

User space gettimeofday()

kbob — Thu, 19 Dec 2002 03:54:45 +0000

Gettimeofday() may not be practical in user space, but
getpid() certainly is. Other syscalls without side effects
that return infrequently changing information could also be
put into user space.