[parisc-linux] xchg/cmpxchg defined in wrong place?

[parisc-linux] xchg/cmpxchg defined in wrong place?

[Grant Grundler]

Hello Linux Kernel Guru's,

After surveying all the arches that define __HAVE_ARCH_CMPXCHG:

./include/asm-alpha/system.h:#define __HAVE_ARCH_CMPXCHG 1
./include/asm-i386/system.h:#define __HAVE_ARCH_CMPXCHG 1
./include/asm-ia64/system.h:#define __HAVE_ARCH_CMPXCHG 1
./include/asm-ppc/system.h:#define __HAVE_ARCH_CMPXCHG  1
./include/asm-sparc64/system.h:#define __HAVE_ARCH_CMPXCHG 1

I've come to the conclusion xchg/cmpxchg definitions do NOT
belong in system.h.  AFAICT, all the above use Load Linked semantics
(or in the i386 case, operation is atomic). In other words, xchg/cmpxchg
are atomic operations.  Shouldn't xchg/cmpxchg definitions live
with other atomic operations - asm/atomic.h?

Currently, parisc __xchg lives in arch/parisc/kernel/setup.c. There is no
way to inline xchg or cmpxchg for parisc at the moment. And given the
dependendencies between asm/system.h, asm/spinlock.h, and asm/atomic.h,
I'm not trying to inline them. Perhaps when the definitions formally move
to atomic.h (or other more appropriate place).

And to add insult to injury, drivers/char/drm/drmP.h (the only visible
consumer of __HAVE_ARCH_CMPXCHG) doesn't even explicitly include
asm/system.h!

thanks,
grant

Re: [parisc-linux] xchg/cmpxchg defined in wrong place?

[Grant Grundler]

Grant Grundler wrote:
...
> Currently, parisc __xchg lives in arch/parisc/kernel/setup.c.

Sorry. I got this fact wrong: __xchg lives in arch/parisc/lib/bitops.c

__xchg_called_with_bad_pointer *is* defined in arch/parisc/kernel/setup.c.
My understanding is this fuction in fact should NOT be defined.
It's a tool to catch problems for architectures which can inline __xchg.

Neither item is material to my arguments for defining them in atomic.h.

grant

Grant Grundler
Unix Systems Enablement Lab
+1.408.447.7253

Re: [parisc-linux] xchg/cmpxchg defined in wrong place?

[Alan Cox]

> way to inline xchg or cmpxchg for parisc at the moment. And given the
> dependendencies between asm/system.h, asm/spinlock.h, and asm/atomic.h,
> I'm not trying to inline them. Perhaps when the definitions formally move
> to atomic.h (or other more appropriate place).
> 
> And to add insult to injury, drivers/char/drm/drmP.h (the only visible
> consumer of __HAVE_ARCH_CMPXCHG) doesn't even explicitly include
> asm/system.h!

Seems a reasonable thing to try and move to atomic.h yes

> 

Re: [parisc-linux] xchg/cmpxchg defined in wrong place?

[Grant Grundler]

Alan Cox wrote:
> Seems a reasonable thing to try and move to atomic.h yes

ok. I'll take a whack at parisc arch and see what happens.
If it's easy, I'll try the i386 arch as well - mostly
to build more drivers as a sanity check.

thanks,
grant

Re: [parisc-linux] xchg/cmpxchg defined in wrong place?

[Grant Grundler]

Grant Grundler wrote:
> ok. I'll take a whack at parisc arch and see what happens.

This was easy. Was even able to build/link with DRM and all the
related drivers. Code change committed to parisc-linux CVS:
o removes bogus definition of __xchg_called_with_bad_pointer from setup.c
o definitions follow sparc64 model except I didn't try to inline all
  of this mess.
o added __xchgNN() and __cmpxchg_uNN() functions to bitops.c and
  declared them in parisc_ksyms.c for CONFIG_MODULES support.
o added "include <asm/atomic.h>" to arch/parisc/kernel/semaphore.c
  to make sure it gets the xchg() macro definition.

TODO: DRM should explicitly include asm/atomic.h to pick up
  #define __HAVE_ARCH_CMPXCHG directly.

> If it's easy, I'll try the i386 arch as well - mostly
> to build more drivers as a sanity check.

I moved all the xchg/cmpxchg related code from include/asm-i386/system.h
to include/asm-i386/atomic.h and built an x86 kernel (natively) but didn't
boot it. Booting the kernel shouldn't be necessary since mis-use will
result in a linker error (xchg undefined).

I used 2.4.0-test10 since it was handy. Once I get newer Linus
tree, I can cut/paste again and submit the patch to l-k.
Should be able to do this in the next couple of days unless someone
tells me they've done it already.

thanks,
grant

Re: [parisc-linux] xchg/cmpxchg defined in wrong place?

[R P Herrold]

On Sat, 30 Dec 2000, Grant Grundler wrote:

> I moved all the xchg/cmpxchg related code from include/asm-i386/system.h
> to include/asm-i386/atomic.h and built an x86 kernel (natively) but didn't
> boot it. Booting the kernel shouldn't be necessary since mis-use will
> result in a linker error (xchg undefined).

I've always liked this one:

Theory and Practice, distinguished:  In Theory, there is no
difference between Theory and Practice; in Practice, there
always is ...

I'm going to try spinning up a G50, and sniffing at a K220 for
parisc linux ... 735/125 on the way in ... I'd like to leave
HP-UX on the K220 and 'dual boot' -- but it is such a pain to
hook a console up -- Any thoughts on an OS/2 style Boot
Manager leaving 'next boot' semaphore flags behind ...

-- Russ

Re: Dual Boot

[Grant Grundler]

R P Herrold wrote:
...
> I'm going to try spinning up a G50, and sniffing at a K220 for
> parisc linux ... 735/125 on the way in ... I'd like to leave
> HP-UX on the K220 and 'dual boot' -- but it is such a pain to
> hook a console up -- Any thoughts on an OS/2 style Boot
> Manager leaving 'next boot' semaphore flags behind ...

It's hard to hook up a serial console on K220?!

Under HPUX, "setboot" command can set the primary boot path
so it can reboot from any disk/lan you want. You would have to
write a similar program for parisc-linux to poke stuff into NVRAM.
It might be a PDC call to do it; I don't know. But it's doable.

grant

Grant Grundler
Unix Systems Enablement Lab
+1.408.447.7253