This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: Improve build-id checking when the task we're interested in isn't 'current'. git commit causing problems on ARM and IA64
- From: William Cohen <wcohen at redhat dot com>
- To: David Smith <dsmith at redhat dot com>
- Cc: systemtap at sourceware dot org
- Date: Tue, 03 Apr 2012 10:17:46 -0400
- Subject: Re: Improve build-id checking when the task we're interested in isn't 'current'. git commit causing problems on ARM and IA64
- References: <4F76225B.8010600@redhat.com> <4F79FE43.9010902@redhat.com> <4F7A091E.6040208@redhat.com> <4F7A23DB.5000800@redhat.com>
On 04/02/2012 06:10 PM, David Smith wrote:
> On 04/02/2012 03:16 PM, William Cohen wrote:
>
>> On 04/02/2012 03:30 PM, David Smith wrote:
>>> ../install/bin/stap -k ../systemtap/testsuite/systemtap.base/add.stp
>>
>> Hi David,
>>
>> Thanks for taking a look at this problem.
>>
>> The git checkin fixes the problem for ia64.
>>
>> The arm machine is using a kernel from the linus torvald's git repo. This particular kernel doesn't
>
>> have utrace support in it, so it doesn't have CONFIG_UTRACE set.
>> Thus, things still fail in the same way on the ARM machine.
>
>>
>> Does this code work with stock x86_64 kernel? I am wondering why this problem wasn't seen on the x86
>
>> machine. Does the x86 not need to do the explicit flushes unlike the
>> ia64 and ARM?
>
> Right. From what I've been looking at on the stock x86_64 kernel,
> copy_to_user_page() boils down to a memcpy() call.
>
> Since x86_64 doesn't have an arch-specific cacheflush.h file, it
> inherits the following from asm-generic/cacheflush.h:
>
> ====
> ....
> #define flush_icache_user_range(vma,pg,adr,len) do { } while (0)
> ....
> #define copy_to_user_page(vma, page, vaddr, dst, src, len) \
> do { \
> memcpy(dst, src, len); \
> flush_icache_user_range(vma, page, vaddr, len); \
> } while (0)
> ====
>
> Arm has a arch-specific copy_to_user_page() (defined in
> arch/arm/include/cacheflush.h) which isn't exported. Sigh.
>
> It looks like on current kernels access_process_vm() is exported, which
> means we could use the real function instead of our copy (can you check
> and make sure this is exported on arm?). However, we've added the
> __access_process_vm_noflush() variant which isn't present upstream.
>
> I'm not sure there are easy answers here.
>
I looked through Module.symvers for the currently running kernel on the ARM machine and didn't see access_process_vm listed. Also seems that all the uses of access_process_vm() were for things that are built into the kernel rather than modules.
-Will