Re: [PATCH v3 2/2] perf bench: Add support for 32-bit systems with 64-bit time_t

[André Almeida]

Às 01:34 de 24/09/21, Alistair Francis escreveu:
> On Tue, Sep 21, 2021 at 8:47 AM André Almeida <andrealmeid@xxxxxxxxxxxxx> wrote:
>>
>> Hi Alistair,
>>
>> Às 03:10 de 17/09/21, Alistair Francis escreveu:
>>> From: Alistair Francis <alistair.francis@xxxxxxx>
>>>
>>> Some 32-bit architectures (such are 32-bit RISC-V) only have a 64-bit
>>> time_t and as such don't have the SYS_futex syscall. This patch will
>>> allow us to use the SYS_futex_time64 syscall on those platforms.
>>>
>>
>> Thanks for your patch! However, I don't think that any futex operation
>> at perf has timeout. Do you plan to implement a test that use it? Or the
>> idea is to get this ready for it in case someone want to do so in the
>> future?
> 
> I don't have plans to implement any new tests (although I'm happy to
> add one if need be).
> 
> My goal was just to get this to build for RISC-V 32-bit. The timeout
> was already exposed by the old futex macro, so I was just following
> that.
> 

I see, thanks for working on that.

>>
>>
>> Also, I faced a similar problem with the new futex2 syscalls, that
>> supports exclusively 64bit timespec. But I took a different approach: I
>> called __NR_clock_gettime64 for 32bit architectures so it wouldn't
>> require to convert the struct:
>>
>> #if defined(__i386__) || __TIMESIZE == 32
>> # define NR_gettime64 __NR_clock_gettime64
>> #else
>> # define NR_gettime64 __NR_clock_gettime
>> #endif
>>
>> struct timespec64 {
>>         long long tv_sec;       /* seconds */
>>         long long tv_nsec;      /* nanoseconds */
>> };
>>
>> int gettime64(clock_t clockid, struct timespec64 *tv)
>> {
>>         return syscall(NR_gettime64, clockid, tv);
>> }
>>
>> Then we can just use &timeout at __NR_futex_time64 for 32bit arch and at
>> __NR_futex for 64bit arch.
> 
> So the idea is to use 64-bit time_t everywhere and only work on 5.1+ kernels.
> 
> If that's the favoured approach I can convert this series to your idea.
> 

Yes, this is what I think it will be the best approach. I believe the
code will be less complex, it's more future proof (it's ready for y2038)
and when glibc supports time64, we can make this code even simpler using
`-D__USE_TIME_BITS64` to compile it. Thanks again for working on that!

> Alistair
> 
>>
>> This might be a simpler solution to the problem that you are facing but
>> I'm not entirely sure. Also, futex's selftests do use the timeout
>> argument and I think that they also won't compile in 32-bit RISC-V, so
>> maybe we can start from there so we can actually test the timeout
>> argument and check if it's working.
>>
>> Thanks,
>>         André