Re: [RFC/PATCH] perf: Add sizeof operator support

[Jeremy Linton <jeremy.linton@arm.com>]

Hi Steven,

On 06/17/2016 11:17 AM, Steven Rostedt wrote:
> On Tue, 14 Jun 2016 11:38:32 -0500
> Jeremy Linton <jeremy.linton@arm.com> wrote:
>
>> There are a fair number of tracepoints in the kernel making
>> use of the sizeof operator. Allow perf to understand some of
>> those cases, and report a more informative error message for
>> the ones it cannot understand.
>>
>> Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
>> ---
>>
>> So this is as much a RFC as a patch because the use of sizeof
>> seems to extend to structures, pointers, etc that aren't easy
>> to deduce from userspace. I'm not sure what the correct solution
>> should be in those cases.
>>
>>   tools/lib/traceevent/event-parse.c | 46 ++++++++++++++++++++++++++++++++++++++

(trimming)

>> +
>> +	if (strcmp(token, "__u64") == 0) {
>> +		if (asprintf(&arg->atom.atom, "%zd", sizeof(__u64)) < 0)
>> +			goto out_free_atom;
>> +	} else if (strcmp(token, "__u32") == 0) {
>> +		if (asprintf(&arg->atom.atom, "%zd", sizeof(__u32)) < 0)
>> +			goto out_free_atom;
>
> What events are doing sizeof(__u64) and sizeof(__u32)?
>
> First, that's useless, as sizeof(__u64) will always be 8, and
> sizeof(__u32) will always be 4.
>
> What exactly is this fixing?

It starts to fix things like:

kmem:mm_page_alloc
   Warning: [kmem:mm_page_alloc] function sizeof not defined

or:

# perf stat -e kvm:kvm_arm_set_regset -- true
   Warning: [kvm:kvm_arm_set_regset] function sizeof not defined
   Warning: Error: expected type 5 but read 0
*** Error in `perf': double free or corruption (fasttop): 
0x00000000303f5930 ***

There is a RH bug about it (and the "~" operator, which has been fixed) 
here: https://bugzilla.redhat.com/show_bug.cgi?id=1298229

Thanks for taking a look at this,