Fill in CPU info in mini-dump for Arm64.

Fill in CPU info in mini-dump for Arm64.

Adds Arm64 varients of CPUFillThreadInfo and CPUFillFromUContext and
WriteCPUInformation for the Linux/Android client.

BUG=354405,335641
R=mark@chromium.org

Committed: 1310

[rmcilroy]

PTAL.  AFAIK this should be all the changes required for Breakpad support of
ARM64 for Linux / Android.  Are there any more server-side changes required?

[Mark Mentovai]

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
File client/linux/minidump_writer/minidump_writer.cc (right):

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
client/linux/minidump_writer/minidump_writer.cc:1354: #if !defined(__mips__) &&
!defined(__aarch64__)
This is more properly

#if defined(__i386__) || defined(__x86_64__)

since these attributes are fairly x86-specific.

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
client/linux/minidump_writer/minidump_writer.cc:1430: #if !defined(__mips__) &&
!defined(__aarch64__)
Same here.

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
client/linux/minidump_writer/minidump_writer.cc:1443: bool
WriteCPUInformation(MDRawSystemInfo* sys_info) {
Does anything in here make sense for arm64? Is this function a better basis for
the arm64 implementation than the one above?

[rmcilroy]

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
File client/linux/minidump_writer/minidump_writer.cc (right):

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
client/linux/minidump_writer/minidump_writer.cc:1354: #if !defined(__mips__) &&
!defined(__aarch64__)
On 2014/04/09 16:42:54, Mark Mentovai wrote:
> This is more properly
> 
> #if defined(__i386__) || defined(__x86_64__)
> 
> since these attributes are fairly x86-specific.

Done.

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
client/linux/minidump_writer/minidump_writer.cc:1430: #if !defined(__mips__) &&
!defined(__aarch64__)
On 2014/04/09 16:42:54, Mark Mentovai wrote:
> Same here.

Done.

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
client/linux/minidump_writer/minidump_writer.cc:1430: #if !defined(__mips__) &&
!defined(__aarch64__)
On 2014/04/09 16:42:54, Mark Mentovai wrote:
> Same here.

Done.

https://breakpad.appspot.com/1464002/diff/1/client/linux/minidump_writer/mini...
client/linux/minidump_writer/minidump_writer.cc:1443: bool
WriteCPUInformation(MDRawSystemInfo* sys_info) {
On 2014/04/09 16:42:54, Mark Mentovai wrote:
> Does anything in here make sense for arm64? Is this function a better basis
for
> the arm64 implementation than the one above?

This is where I started, but it doesn't seem to be useful for arm64.  The only
things in the "features" field in proc/cpuinfo that I can see are "fp" and
"asimd", both of which are required features on ARMv8 AFAIK.  There is also no
MD_CPU_ARM_ELF_XXX equivalents for ARM64 as far as I can tell.

[Mark Mentovai]

Yeah, I figured that the baseline for ARM64 was way more feature-filled. I
thought that the CPU implementer/variant/part/revision and Processor fields
might still have useful information, though.

LGTM

[rmcilroy]

Having thought about it again I think you are right - it could be useful to have
the cpuid values for aarch64, and we may want to ad hwcaps in the future as
ARMv8 get's more fragmented in the future.  I've reworked the
WriteCPUInformation function to base aarch64 on arm's approach.  PTAL.

https://breakpad.appspot.com/1464002/diff/70001/client/linux/minidump_writer/...
File client/linux/minidump_writer/minidump_writer.cc (right):

https://breakpad.appspot.com/1464002/diff/70001/client/linux/minidump_writer/...
client/linux/minidump_writer/minidump_writer.cc:1574:
sys_info->cpu.arm_cpu_info.cpuid |=
Is this OK to share the cpu.arm_cpu_info stuct for both arm and aarch64?  it
seems to be the same situation for x86/x86_64/

[rmcilroy]

On 2014/04/10 16:38:18, rmcilroy wrote:
> Having thought about it again I think you are right - it could be useful to
have
> the cpuid values for aarch64, and we may want to ad hwcaps in the future as
> ARMv8 get's more fragmented in the future.  I've reworked the
> WriteCPUInformation function to base aarch64 on arm's approach.  PTAL.
> 
>
https://breakpad.appspot.com/1464002/diff/70001/client/linux/minidump_writer/...
> File client/linux/minidump_writer/minidump_writer.cc (right):
> 
>
https://breakpad.appspot.com/1464002/diff/70001/client/linux/minidump_writer/...
> client/linux/minidump_writer/minidump_writer.cc:1574:
> sys_info->cpu.arm_cpu_info.cpuid |=
> Is this OK to share the cpu.arm_cpu_info stuct for both arm and aarch64?  it
> seems to be the same situation for x86/x86_64/

I see you are OOO now - I will land this now based on your previous LGTM -
please let me know if there are any changes you would like to make once you get
back.

[rmcilroy]

Committed patchset #3 manually as r1310 (presubmit successful).

[Mark Mentovai]

LGTM.

https://breakpad.appspot.com/1464002/diff/70001/client/linux/minidump_writer/...
File client/linux/minidump_writer/minidump_writer.cc (right):

https://breakpad.appspot.com/1464002/diff/70001/client/linux/minidump_writer/...
client/linux/minidump_writer/minidump_writer.cc:1574:
sys_info->cpu.arm_cpu_info.cpuid |=
rmcilroy wrote:
> Is this OK to share the cpu.arm_cpu_info stuct for both arm and aarch64?  it
> seems to be the same situation for x86/x86_64/

Yes, it sure is.