[PATCH] makedumpfile: inhibit predefined macros when cross building

Wang Nan wangnan0 at huawei.com
Tue Apr 8 03:42:45 EDT 2014 

On 2014/4/8 10:30, Atsushi Kumagai wrote:
>>>> I'm working on cross analysising arm vmcore on x86. I know that crash support
>>>> cross analysing (at least for analysing little endian arm vmcore on x86). I also notice
>>>> that makedumpfile has been designed for cross analysing (see commit 96f24dc77dc48e4c8f4c3527f4f5b4d27845393c
>>>> by Suzuki K. Poulose <suzuki at in.ibm.com>), but it doesn't work correctly for arm vmcore.
>>>
>>> You may misunderstand, makedumpfile doesn't support cross analyzing.
>>> According to the patch, I'm sure that Suzuki sent the patch to enable
>>> *cross compile* (e.g. building arm binary on x86 machine) by setting
>>> a cross compiler to CC variable:
>>>
>>>   (README)
>>>     5.Build for a different architecture than the host :
>>>       # make TARGET=<arch> ; make install
>>>       where <arch> is the 'uname -m' of the target architecture.
>>>       The user has to set the environment variable CC to appropriate
>>>       compiler for the target architecture.
>>>
>>> When CC is the appropriate compiler for the target architecture,
>>> the mismatch predefining macros will not happen.
>>> So I think there is no problem in the current code.
>>>
>>>> What I'm trying to do is to enable the whole kdump analysing stuff crossly (I know it is hard).
>>>> It is useful in some special cases, for example, for some boards with strictly limited memory
>>>> and storage space.
>>>
>>> It sounds reasonable if you mentioned crash.
>>>
>>> The mission of makedumpfile is reducing dump size, so it's best to work
>>> during capturing dump. Otherwise, you have to save full dump on the
>>> *limited storage space*.
>>
>> I think I didn't explain my case very clearly.
>>
>> Our situation is that, our device has very short storage space which can hold busybox only
>> (but with large memory). Most of its work is done in kernel (so kdump is important). When crash
>> happens, What it can do is only scp /proc/vmcore to a x86 machine via network.
>> Although using crash on the full vmcore on a x86 machine is affordable, makedumpfile is also useful
>> because we want to backup those vmcore files for furture use, so its size must be reduced.
> 
> OK, I understand your case.
> "makedumpfile -F" can resolve the byte-order issue if you can use it
> to transport vmcore, but it seems impossible in your case.
> 
> Again, this patch doesn't introduce the cross analyzing feature legally,
> but I accept it since it will not affect other features.
> 
> 
> Thanks
> Atsushi Kumagai
> 

Thank you.

I have checked that, after applying my patch, crash (for arm, running on x86) can cross
analysis ELF dumpfile, but not for default output:


$ ./makedumpfile ../halogen-install/vmcore ./xxx.elf -d 31 -f -E
Copying data                       : [100.0 %] /

The dumpfile is saved to ./xxx.elf.

makedumpfile Completed.

$ ./crash  ~/makedumpfile/xxx.elf  ~/vmlinux

crash 7.0.4
Copyright (C) 2002-2013  Red Hat, Inc.
...
This program is free software, covered by the GNU General Public License,
...

GNU gdb (GDB) 7.6
Copyright (C) 2013 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
...
This GDB was configured as "--host=x86_64-unknown-linux-gnu --target=arm-elf-linux"...

      KERNEL: /xxx/xxx/build-qemu/vmlinux
    DUMPFILE: /xxx/xxx/xxx.elf
        CPUS: 1
        DATE: Sat Mar 29 17:22:47 2014
      UPTIME: 00:01:05
LOAD AVERAGE: 0.09, 0.04, 0.02
       TASKS: 32
    NODENAME: arma15el
     RELEASE: 3.10.34
     VERSION: #2 SMP Sat Mar 29 15:12:37 CST 2014
     MACHINE: armv7l  (unknown Mhz)
      MEMORY: 1 GB
       PANIC: "Internal error: Oops: 817 [#1] SMP ARM" (check log for details)
         PID: 835
     COMMAND: "bash"
        TASK: bf969680  [THREAD_INFO: bdd12000]
         CPU: 0
       STATE: TASK_RUNNING (PANIC)

crash> bt
PID: 835    TASK: bf969680  CPU: 0   COMMAND: "bash"
 #0 [<801c62b0>] (sysrq_handle_crash) from [<801c6964>]
 #1 [<801c6964>] (__handle_sysrq) from [<801c6a68>]
 #2 [<801c6a68>] (write_sysrq_trigger) from [<80103354>]
 #3 [<80103354>] (proc_reg_write) from [<800bc644>]
 #4 [<800bc644>] (vfs_write) from [<800bca04>]
 #5 [<800bca04>] (sys_write) from [<8000e140>]
    pc : [<76e4ba2c>]    lr : [<76df6f98>]    psr: 60070010
    sp : 7ebf68dc  ip : 00000000  fp : 00000000
    r10: 000e0408  r9 : 00000002  r8 : 00000002
    r7 : 00000004  r6 : 76ec1a90  r5 : 000e0408  r4 : 00000002
    r3 : 00000000  r2 : 00000002  r1 : 000e0408  r0 : 00000001
    Flags: nZCv  IRQs on  FIQs on  Mode USER_32  ISA ARM
crash> log
[    0.000000] Booting Linux on physical CPU 0x0
[    0.000000] Booting Linux on physical CPU 0x0
[    0.000000] Initializing cgroup subsys cpuset
...

>>> As for makedumpfile, I don't have any ideas about the advantage of the
>>> cross analyzing you said.
>>>
>>>> With this patch, at least makedumpfile can be compiled to x86_64 binary, and can extract
>>>> symbol information correctly for a vmcore file dumpped from a qemu virtual machine (little
>>>> endian).
>>>
>>> Your patch sounds inconsistent to me since makedumpfile doesn't support
>>> cross analyzing. The case you said will go well, but that's just a lucky
>>> situation.
>>>
>>>
>>> Thanks
>>> Atsushi Kumagai
>>>
>>>>>>> This patch changes Makefile: if host arch and target arch are different,
>>>>>>> passes "-U__$(HOST_ARCH)__" to gcc.
>>>>>>>
>>>>>>> I have tested by running following command on x86_64 machine:
>>>>>>>
>>>>>>> make ARCH=arm
>>>>>>
>>>>>> Sorry, should be TARGET=arm. I have tested and build okay.
>>>>>>>
>>>>>>> Signed-off-by: Wang Nan <wangnan0 at huawei.com>
>>>>>>> Cc: Atsushi Kumagai <kumagai-atsushi at mxc.nes.nec.co.jp>
>>>>>>> Cc: kexec at lists.infradead.org
>>>>>>> Cc: Geng Hui <hui.geng at huawei.com>
>>>>>>> ---
>>>>>>>  Makefile | 12 +++++++++---
>>>>>>>  1 file changed, 9 insertions(+), 3 deletions(-)
>>>>>>>
>>>>>>> diff --git a/Makefile b/Makefile
>>>>>>> index 2f4845c..f85cc21 100644
>>>>>>> --- a/Makefile
>>>>>>> +++ b/Makefile
>>>>>>> @@ -15,10 +15,11 @@ CFLAGS_ARCH	= -g -O2 -Wall -D_FILE_OFFSET_BITS=64 \
>>>>>>>  		    -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE
>>>>>>>  # LDFLAGS = -L/usr/local/lib -I/usr/local/include
>>>>>>>
>>>>>>> +HOST_ARCH := $(shell uname -m)
>>>>>>>  # Use TARGET as the target architecture if specified.
>>>>>>>  # Defaults to uname -m
>>>>>>>  ifeq ($(strip($TARGET)),)
>>>>>>> -TARGET := $(shell uname -m)
>>>>>>> +TARGET := $(HOST_ARCH)
>>>>>>>  endif
>>>>>>>
>>>>>>>  ARCH := $(shell echo ${TARGET}  | sed -e s/i.86/x86/ -e s/sun4u/sparc64/ \
>>>>>>> @@ -26,8 +27,13 @@ ARCH := $(shell echo ${TARGET}  | sed -e s/i.86/x86/ -e s/sun4u/sparc64/ \
>>>>>>>  			       -e s/s390x/s390/ -e s/parisc64/parisc/ \
>>>>>>>  			       -e s/ppc64/powerpc64/ -e s/ppc/powerpc32/)
>>>>>>>
>>>>>>> -CFLAGS += -D__$(ARCH)__
>>>>>>> -CFLAGS_ARCH += -D__$(ARCH)__
>>>>>>> +CROSS :=
>>>>>>> +ifneq ($(TARGET), $(HOST_ARCH))
>>>>>>> +CROSS := -U__$(HOST_ARCH)__
>>>>>>> +endif
>>>>>>> +
>>>>>>> +CFLAGS += -D__$(ARCH)__ $(CROSS)
>>>>>>> +CFLAGS_ARCH += -D__$(ARCH)__ $(CROSS)
>>>>>>>
>>>>>>>  ifeq ($(ARCH), powerpc64)
>>>>>>>  CFLAGS += -m64
>>>>>>>
>>>>>>
>>>>
>>>
>>

More information about the kexec mailing list