Hello Steven,
A while ago I already mentioned that 2.6.25 did not compile on ARM.
(e.g. AT91SAM9261-EK)
Today I looked at it again with the latest 2.6.25 kernel with latest RT-patch.
Several errors are gone, but I still I get these warnings on every
file that is being compiled:
-----------------------------------------------------------------
CC arch/arm/kernel/asm-offsets.s
In file included from include/linux/rt_lock.h:14,
from include/linux/spinlock.h:117,
from include/linux/seqlock.h:29,
from include/linux/time.h:8,
from include/linux/timex.h:57,
from include/linux/sched.h:54,
from arch/arm/kernel/asm-offsets.c:13:
include/asm/atomic.h:241:1: warning: "cmpxchg" redefined
In file included from include/asm/system.h:378,
from include/asm/bitops.h:27,
from include/linux/bitops.h:17,
from include/linux/kernel.h:15,
from include/linux/sched.h:52,
from arch/arm/kernel/asm-offsets.c:13:
include/asm-generic/cmpxchg.h:19:1: warning: this is the location of
the previous definition
...
and many more...
...
-----------------------------------------------------------------
I looked at the broken-out series which patches are causing these
warnings and these are the offending patches:
* arm-fix-atomic-cmpxchg.patch
* arm-cmpxchg.patch
When these patches are removed from the Preempt-RT patchset everything
compiles again, without warnings.
But these patches are probably there for a reason, and leaving them
out could break some things. (Although I believe that these patches
are just there because they apply on 2.6.25-RT)
But, Here I need some (RT-mutex-expert) review help.
Because this ARM architecture (ARM <= V5) does not have a real
cmpxchg() assembler
instruction, and the implementation is almost the same as the generic
definition, the generic definition can be used instead.
However, The generic implementation does NOT define __HAVE_ARCH_CMPXCHG.
My questions are:
* What are the results of NOT defining __HAVE_ARCH_CMPXCHG in
kernel/rtmutex.c ?
* Does the RT-mutex still behave correctly?
* Why is this check in rtmutex.c ? Is it really needed, because there
is always a generic implementation of cmpxchg() these days?
FURTHER: The patch arm-fix-atomic-cmpxchg.patch made the
local_irq_save() a raw_local_irq_save().
Looking at the generic implementation I believe that this code should
also use the raw_* types. I attached a new patch for this to this
mail.
So, attached 3 patches:
* revert-arm-fix-atomic-cmpxchg.patch (reverts the same named patch in
preempt-RT set)
* revert-arm-cmpxchg.patch (reverts the same named patch in preempt-RT set)
* make-generic-cmpxchg-use-raw-intlock-types.patch
Can you clear some things up here? And also repair the RT-patch for
ARM and 2.6.25 ;-) (and newer)
Kind regards,
Remy