Kdump on ARM64

Kdump on ARM64

[Anurup M]

Hi Akashi, Geoff,

	I tried to check kdump on arm64 on Foundation model. The/proc/vmcore is generated.
But I couldnot check if it is proper. Is there a makedumpfile binary for arm64? and also crash tool.
I used the master branch of https://git.linaro.org/people/geoff.levand/linux-kexec.git to verify.
 	Please provide some information about the current status or support for kdump on arm64?
and the ways to verify or start development on Foundation model.

Thanks,
Anurup M

_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

Re: Kdump on ARM64

[AKASHI Takahiro]

Hi Anurup,

On 01/08/2015 03:35 PM, Anurup M wrote:
> Hi Akashi, Geoff,
>
> 	I tried to check kdump on arm64 on Foundation model. The/proc/vmcore is generated.
> But I couldnot check if it is proper. Is there a makedumpfile binary for arm64? and also crash tool.
> I used the master branch of https://git.linaro.org/people/geoff.levand/linux-kexec.git to verify.
>   	Please provide some information about the current status or support for kdump on arm64?
> and the ways to verify or start development on Foundation model.

The story is that I used to work on kdump, but handed it off
to Geoff at some time. Since then, he merged it into his repo
but I don't think that he has tested (verified) it yet.
So please take it as a study work (or prototype), although I'm
pretty sure that it works well on model except some issues.

Regarding /proc/vmcore, if you can retrieve it to your host machine,
you can simply verify the contents as follows:
   $ aarch64-linux-gnu-gdb vmlinux core(=vmcore)
   (gdb) info threads
   (gdb) bt
    ...

I also have a trivial patch for crash tool to cross-build it, but
arm64 support (by RedHat folk) is already there.

Thanks,
-Takahiro AKASHI

> Thanks,
> Anurup M
>

_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

Re: Kdump on ARM64

[Anurup M]

Hi Akashi,

	Thanks for the reply.

Hi  Akashi, Geoff,

	In the code I could find assert(arm64_mem.memstart) in the function virt_to_phys. This address can be 0x0 when the
  segment start address is 0x0. so can we use this assert?

	The code flow is elf_arm64_load->load_crashdump_segments->add_buffer_phys_virt->add_segment_phys_virt->virt_to_phys
  and this call is also before the call to load_elf_exec_in_crashmem which init's arm64_mem.memstart to crash_reserved_mem.start.
  So arm64_mem.memstart is also not set before the call to virt_to_phys.

	Please share your comments.

Thanks,
Anurup


On 1/8/2015 1:40 PM, AKASHI Takahiro wrote:
> Hi Anurup,
> 
> On 01/08/2015 03:35 PM, Anurup M wrote:
>> Hi Akashi, Geoff,
>>
>>     I tried to check kdump on arm64 on Foundation model. The/proc/vmcore is generated.
>> But I couldnot check if it is proper. Is there a makedumpfile binary for arm64? and also crash tool.
>> I used the master branch of https://git.linaro.org/people/geoff.levand/linux-kexec.git to verify.
>>       Please provide some information about the current status or support for kdump on arm64?
>> and the ways to verify or start development on Foundation model.
> 
> The story is that I used to work on kdump, but handed it off
> to Geoff at some time. Since then, he merged it into his repo
> but I don't think that he has tested (verified) it yet.
> So please take it as a study work (or prototype), although I'm
> pretty sure that it works well on model except some issues.
> 
> Regarding /proc/vmcore, if you can retrieve it to your host machine,
> you can simply verify the contents as follows:
>   $ aarch64-linux-gnu-gdb vmlinux core(=vmcore)
>   (gdb) info threads
>   (gdb) bt
>    ...
> 
> I also have a trivial patch for crash tool to cross-build it, but
> arm64 support (by RedHat folk) is already there.
> 
> Thanks,
> -Takahiro AKASHI
> 
>> Thanks,
>> Anurup M
>>
> 

_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

Re: Kdump on ARM64

[AKASHI Takahiro]

On 01/13/2015 08:23 PM, Anurup M wrote:
>
> Hi Akashi,
>
> 	Thanks for the reply.
>
> Hi  Akashi, Geoff,
>
> 	In the code I could find assert(arm64_mem.memstart) in the function virt_to_phys. This address can be 0x0 when the
>    segment start address is 0x0. so can we use this assert?

Probably you're right.

> 	The code flow is elf_arm64_load->load_crashdump_segments->add_buffer_phys_virt->add_segment_phys_virt->virt_to_phys
>    and this call is also before the call to load_elf_exec_in_crashmem which init's arm64_mem.memstart to crash_reserved_mem.start.
>    So arm64_mem.memstart is also not set before the call to virt_to_phys.

I need refresh my memory about the flow, but in this particular case,
the last argument, phys, of add_buffer_phys_virt() is 0 and virt_to_phys()
will not be called.

Do you have any other evidence that the assertion above is broken?

Thanks,
-Takahiro AKASHI


> 	Please share your comments.
>
> Thanks,
> Anurup
>
>
> On 1/8/2015 1:40 PM, AKASHI Takahiro wrote:
>> Hi Anurup,
>>
>> On 01/08/2015 03:35 PM, Anurup M wrote:
>>> Hi Akashi, Geoff,
>>>
>>>      I tried to check kdump on arm64 on Foundation model. The/proc/vmcore is generated.
>>> But I couldnot check if it is proper. Is there a makedumpfile binary for arm64? and also crash tool.
>>> I used the master branch of https://git.linaro.org/people/geoff.levand/linux-kexec.git to verify.
>>>        Please provide some information about the current status or support for kdump on arm64?
>>> and the ways to verify or start development on Foundation model.
>>
>> The story is that I used to work on kdump, but handed it off
>> to Geoff at some time. Since then, he merged it into his repo
>> but I don't think that he has tested (verified) it yet.
>> So please take it as a study work (or prototype), although I'm
>> pretty sure that it works well on model except some issues.
>>
>> Regarding /proc/vmcore, if you can retrieve it to your host machine,
>> you can simply verify the contents as follows:
>>    $ aarch64-linux-gnu-gdb vmlinux core(=vmcore)
>>    (gdb) info threads
>>    (gdb) bt
>>     ...
>>
>> I also have a trivial patch for crash tool to cross-build it, but
>> arm64 support (by RedHat folk) is already there.
>>
>> Thanks,
>> -Takahiro AKASHI
>>
>>> Thanks,
>>> Anurup M
>>>
>>

_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

Re: Kdump on ARM64

[Anurup M]

On 1/14/2015 7:40 AM, AKASHI Takahiro wrote:
> On 01/13/2015 08:23 PM, Anurup M wrote:
>>
>> Hi Akashi,
>>
>>     Thanks for the reply.
>>
>> Hi  Akashi, Geoff,
>>
>>     In the code I could find assert(arm64_mem.memstart) in the function virt_to_phys. This address can be 0x0 when the
>>    segment start address is 0x0. so can we use this assert?
> 
> Probably you're right.
> 
>>     The code flow is elf_arm64_load->load_crashdump_segments->add_buffer_phys_virt->add_segment_phys_virt->virt_to_phys
>>    and this call is also before the call to load_elf_exec_in_crashmem which init's arm64_mem.memstart to crash_reserved_mem.start.
>>    So arm64_mem.memstart is also not set before the call to virt_to_phys.
> 
> I need refresh my memory about the flow, but in this particular case,
> the last argument, phys, of add_buffer_phys_virt() is 0 and virt_to_phys()
> will not be called.
> 

Yes, You are right. The code flow which I gave was wrong. It is observed  when I don't input the -p option (only kexec). The error flow is
elf_arm64_load->elf_exec_load->add_segment->add_buffer_phys_virt->add_segment_phys_virt->virt_to_phys

Thanks,
Anurup

> Do you have any other evidence that the assertion above is broken?
> 
> Thanks,
> -Takahiro AKASHI
> 
> 
>>     Please share your comments.
>>
>> Thanks,
>> Anurup
>>
>>
>> On 1/8/2015 1:40 PM, AKASHI Takahiro wrote:
>>> Hi Anurup,
>>>
>>> On 01/08/2015 03:35 PM, Anurup M wrote:
>>>> Hi Akashi, Geoff,
>>>>
>>>>      I tried to check kdump on arm64 on Foundation model. The/proc/vmcore is generated.
>>>> But I couldnot check if it is proper. Is there a makedumpfile binary for arm64? and also crash tool.
>>>> I used the master branch of https://git.linaro.org/people/geoff.levand/linux-kexec.git to verify.
>>>>        Please provide some information about the current status or support for kdump on arm64?
>>>> and the ways to verify or start development on Foundation model.
>>>
>>> The story is that I used to work on kdump, but handed it off
>>> to Geoff at some time. Since then, he merged it into his repo
>>> but I don't think that he has tested (verified) it yet.
>>> So please take it as a study work (or prototype), although I'm
>>> pretty sure that it works well on model except some issues.
>>>
>>> Regarding /proc/vmcore, if you can retrieve it to your host machine,
>>> you can simply verify the contents as follows:
>>>    $ aarch64-linux-gnu-gdb vmlinux core(=vmcore)
>>>    (gdb) info threads
>>>    (gdb) bt
>>>     ...
>>>
>>> I also have a trivial patch for crash tool to cross-build it, but
>>> arm64 support (by RedHat folk) is already there.
>>>
>>> Thanks,
>>> -Takahiro AKASHI
>>>
>>>> Thanks,
>>>> Anurup M
>>>>
>>>
> .
> 

_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec