Re: [patch] Real-Time Preemption, -RT-2.6.12-rc6-V0.7.48-00

[Keith Owens <kaos@ocs.com.au>]

On Wed, 08 Jun 2005 17:04:23 +0200, 
Michal Schmidt <xschmi00@stud.feec.vutbr.cz> wrote:
>Paulo Marques wrote:
>> This is probably just bad luck and a known problem that I'm trying to 
>> fix (but hadn't have much time lately).
>> 
>> Can you try to change the line:
>> 
>> #define WORKING_SET        1024
>> 
>> in scripts/kallsyms.c to:
>> 
>> #define WORKING_SET        65536
>> 
>> and disable CONFIG_KALLSYMS_EXTRA_PASS, to see if the problem goes away?
>
>Yes, this helps.
>
>> It it does go away, then it is the same problem, and I'm working on it...

Not the same problem.  The significant difference in the maps is :-

--- .tmp_map1   2005-06-09 01:14:50.303658655 +1000
+++ .tmp_map2   2005-06-09 01:14:52.829274854 +1000
@@ -8326,8 +8326,8 @@
 c02b93b0 T ipv6_skip_exthdr
 c02b9500 T sha_transform
 c02b96e0 T sha_init
-c02b970f T __sched_text_start
 c02b9710 t __compat_down
+c02b9710 T __sched_text_start
 c02b9810 t __compat_down_interruptible
 c02b9948 T __compat_down_failed
 c02b9958 T __compat_down_failed_interruptible

__sched_text_start has moved up by 1 byte between pass 1 and 2.  Text
addresses are not allowed to move between kallsyms passes, kallsyms
only adds data, it never touches the text segment.  Paulo's change to
the working set hides this peculiarity, rather than fixing the real
cause.  This looks like a toolchain bug, it is moving symbols for no
good reason.