Re: [RFC PATCH for 4.15 v3 15/22] rseq: selftests: Provide self-tests

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



----- On Nov 21, 2017, at 10:34 AM, shuah shuah@xxxxxxxxxx wrote:

[...]
>> ---
>>  MAINTAINERS                                        |    1 +
>>  tools/testing/selftests/Makefile                   |    1 +
>>  tools/testing/selftests/rseq/.gitignore            |    4 +
> 
> Thanks for the .gitignore files. It is commonly missed change, I end
> up adding one to clean things up after tests get in.

I'm used to receive patches where contributors forget to add new files
to gitignore within my own projects, which may contribute to my awareness
of this pain point. :)

[...]

>> +
>> +void *test_percpu_inc_thread(void *arg)
>> +{
>> +	struct inc_thread_test_data *thread_data = arg;
>> +	struct inc_test_data *data = thread_data->data;
>> +	long long i, reps;
>> +
>> +	if (!opt_disable_rseq && thread_data->reg
>> +			&& rseq_register_current_thread())
>> +		abort();
>> +	reps = thread_data->reps;
>> +	for (i = 0; i < reps; i++) {
>> +		int cpu, ret;
>> +
>> +#ifndef SKIP_FASTPATH
>> +		/* Try fast path. */
>> +		cpu = rseq_cpu_start();
>> +		ret = rseq_addv(&data->c[cpu].count, 1, cpu);
>> +		if (likely(!ret))
>> +			goto next;
>> +#endif
> 
> So the test needs to compiled with this enabled? I think it would be better
> to make this an argument to be abel to select at test start time as opposed
> to making this compile time option. Remember that these tests get run in
> automated test rings. Making this a compile time otpion pertty much ensures
> that this path will not be tested.
> 
> So I would reccommend adding a paratemer.
> 
>> +	slowpath:
>> +		__attribute__((unused));
>> +		for (;;) {
>> +			/* Fallback on cpu_opv system call. */
>> +			cpu = rseq_current_cpu();
>> +			ret = cpu_op_addv(&data->c[cpu].count, 1, cpu);
>> +			if (likely(!ret))
>> +				break;
>> +			assert(ret >= 0 || errno == EAGAIN);
>> +		}
>> +	next:
>> +		__attribute__((unused));
>> +#ifndef BENCHMARK
>> +		if (i != 0 && !(i % (reps / 10)))
>> +			printf_verbose("tid %d: count %lld\n", (int) gettid(), i);
>> +#endif
> 
> Same comment as before. Avoid compile time options.

The goal of those compiler define are to generate the altered code without
adding branches into the fast-paths.

Here is an alternative solution that should take care of your concern: I'll
build multiple targets for param_test.c:

param_test
param_test_skip_fastpath (built with -DSKIP_FASTPATH)
param_test_benchmark (build with -DBENCHMARK)

I'll update run_param_test.sh to run both param_test and param_test_skip_fastpath.

Note that "param_test_benchmark" is only useful for benchmarking,
so I don't plan to run it from run_param_test.sh which is meant
to track regressions.

Is that approach OK with you ?

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux