mips64 SMP patch of kexec

[wilbur.chan]

Hi, Maxim Syrchin

I've just  found your  kernel patch for mips64 SMP kexec:

http://lists.infradead.org/pipermail/kexec/2008-June/001909.html

 I was porting  your kexec part  code(excluding crash_dump )  on
xlr408 ,which is of  8 cores SMP mips64 .

I found that in machine_kexec.c ,
flush_icache_range(reboot_code_buffer,  reboot_code_buffer +
KEXEC_CONTROL_CODE_SIZE);

caused an error, so I delete this line directly,although I don't know
if it would bring problem after I del it...


 It is strange that , in relocate_kernel.S, the operator 'REG_L	s5,
(s2)' caused an  invalid access on kernel space.

If I change REG_L to ld , everything went well.   So I change all the
macro such as  REG_L to  ld ,REG_S to sd , PTR_ADD to dadd

LONG_SUB to dsub, and SZREG to 8


At the same time ,  I changed the kexec_smp_wait function to a simple
loop, without checking the kexec_flag:


.globl kexec_smp_wait
kexec_smp_wait:
3: b 3b

at last , the a0,a1,a3,a4 were set to address of fw_arg0,
fw_arg1,fw_arg2,fw_arg3 respectively.


As you see, I just let one of the cpus jumps into new kernel,while the
others be  in an infinite loop.

After jumped into second kernel , the code seemed to be crazy.


Firstly,  the code seemd to stop before calling 'start_kernel--->
calibrate_delay ' .


 However,  if I added a 'printk' in any 'if  branches'  in
'calibrate_delay '  function  , the code would stop before calling

'j  start_kernel'  in head.S. (This time, because printk seemed
unavailable, I  wrote directlly to serial address to trace the
kernel).


I thougt it might be the problem of irq , so I added 'disable_irq(17)'
 in machine_kexec before invoking reboot_code_buffer();



However, the problem was still there.

I don't know why this happedn can you give me some advice on how to
solve this , or if there were some mistake in my code ?


Thank you

regards,

wilbur