Re: [PATCH v2] linux: Avoid shifting a negative signed on POSIX timer interface

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

On 17/10/22 19:33, Arjun Shankar wrote:
> Hi Adhemerval,
> 
>> The current macros uses pid as signed value, which triggers a compiler
>> warning for process and thread timers.  Replace MAKE_PROCESS_CPUCLOCK
>> with static inline function that expects the pid as unsigned.  These
>> are similar to what Linux does internally.
>>
>> Checked on x86_64-linux-gnu.
> 
> I tried building master with -Wshift-negative-value and ran into a
> couple of other left-shifts on negative values in
> sysdeps/x86/dl-cacheinfo.h:

That's a good experiment, I see that clang enables it as default.

> 
>> count_mask = ~(-1 << (count_mask + 1));
> 
> Perhaps a cast of the -1 to unsigned here would make the intent even
> more clear, and get glibc building cleanly with
> -Wshift-negative-value. What do you think?

I think we can also make the mask unsigned to avoid a potential UB
and remove the inline assembly since we have a proper builtin for it:

 unsigned int count_mask = __builtin_clz (threads_l2);
 count_mask = ~(-1U << (count_mask + 1));

> 
> Anyway, this patch itself looks good to me.
> 
> Reviewed-by: Arjun Shankar <arjun@redhat.com>

Thanks.

> 
>>  nptl/pthread_getcpuclockid.c                  |  2 +-
>>  sysdeps/unix/sysv/linux/clock_getcpuclockid.c |  2 +-
>>  sysdeps/unix/sysv/linux/clock_nanosleep.c     |  2 +-
>>  .../unix/sysv/linux/kernel-posix-cpu-timers.h | 28 +++++++++++++++----
>>  sysdeps/unix/sysv/linux/timer_create.c        |  4 +--
>>  5 files changed, 28 insertions(+), 10 deletions(-)
>>
>> diff --git a/nptl/pthread_getcpuclockid.c b/nptl/pthread_getcpuclockid.c
>> index 344bd6560e..b8bf09f550 100644
>> --- a/nptl/pthread_getcpuclockid.c
>> +++ b/nptl/pthread_getcpuclockid.c
>> @@ -35,7 +35,7 @@ __pthread_getcpuclockid (pthread_t threadid, clockid_t *clockid)
>>
>>    /* The clockid_t value is a simple computation from the TID.  */
>>
>> -  const clockid_t tidclock = MAKE_THREAD_CPUCLOCK (pd->tid, CPUCLOCK_SCHED);
>> +  const clockid_t tidclock = make_thread_cpuclock (pd->tid, CPUCLOCK_SCHED);
>>
>>    *clockid = tidclock;
>>    return 0;
> 
> OK. Use the new function instead of the old macro.
> 
>> diff --git a/sysdeps/unix/sysv/linux/clock_getcpuclockid.c b/sysdeps/unix/sysv/linux/clock_getcpuclockid.c
>> index 5534127ed7..355d3c86af 100644
>> --- a/sysdeps/unix/sysv/linux/clock_getcpuclockid.c
>> +++ b/sysdeps/unix/sysv/linux/clock_getcpuclockid.c
>> @@ -29,7 +29,7 @@ __clock_getcpuclockid (pid_t pid, clockid_t *clock_id)
>>    /* The clockid_t value is a simple computation from the PID.
>>       But we do a clock_getres call to validate it.  */
>>
>> -  const clockid_t pidclock = MAKE_PROCESS_CPUCLOCK (pid, CPUCLOCK_SCHED);
>> +  const clockid_t pidclock = make_process_cpuclock (pid, CPUCLOCK_SCHED);
>>
>>  #ifndef __NR_clock_getres_time64
>>  # define __NR_clock_getres_time64 __NR_clock_getres
> 
> OK. Same.
> 
>> diff --git a/sysdeps/unix/sysv/linux/clock_nanosleep.c b/sysdeps/unix/sysv/linux/clock_nanosleep.c
>> index befe6ecb8c..e610fd4e8d 100644
>> --- a/sysdeps/unix/sysv/linux/clock_nanosleep.c
>> +++ b/sysdeps/unix/sysv/linux/clock_nanosleep.c
>> @@ -34,7 +34,7 @@ __clock_nanosleep_time64 (clockid_t clock_id, int flags,
>>    if (clock_id == CLOCK_THREAD_CPUTIME_ID)
>>      return EINVAL;
>>    if (clock_id == CLOCK_PROCESS_CPUTIME_ID)
>> -    clock_id = MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED);
>> +    clock_id = PROCESS_CLOCK;
>>
>>    /* If the call is interrupted by a signal handler or encounters an error,
>>       it returns a positive value similar to errno.  */
> 
> OK. PROCESS_CLOCK expands to the appropriate call to make_process_cpuclock.
> 
>> diff --git a/sysdeps/unix/sysv/linux/kernel-posix-cpu-timers.h b/sysdeps/unix/sysv/linux/kernel-posix-cpu-timers.h
>> index 164a90ddeb..bea1e0e62d 100644
>> --- a/sysdeps/unix/sysv/linux/kernel-posix-cpu-timers.h
>> +++ b/sysdeps/unix/sysv/linux/kernel-posix-cpu-timers.h
>> @@ -1,4 +1,12 @@
>> -/* Parameters for the Linux kernel ABI for CPU clocks.  */
>> +/*
>> +  Parameters for the Linux kernel ABI for CPU clocks, the bit fields within
>> +  a clockid:
>> +
>> +  - The most significant 29 bits hold either a pid or a file descriptor.
>> +  - Bit 2 indicates whether a cpu clock refers to a thread or a process.
>> +  - Bits 1 and 0 give the type: PROF=0, VIRT=1, SCHED=2, or FD=3.
>> +  - A clockid is invalid if bits 2, 1, and 0 are all set.
>> + */
>>
>>  #define CPUCLOCK_PID(clock)            ((pid_t) ~((clock) >> 3))
>>  #define CPUCLOCK_PERTHREAD(clock) \
> 
> OK. Comment lines up with the one in Linux's "include/linux/posix-timers.h".
> 
>> @@ -12,7 +20,17 @@
>>  #define CPUCLOCK_SCHED         2
>>  #define CPUCLOCK_MAX           3
>>
>> -#define MAKE_PROCESS_CPUCLOCK(pid, clock) \
>> -       ((~(clockid_t) (pid) << 3) | (clockid_t) (clock))
> 
> OK. clockid_t is a signed type, thus expanding with pid = 0 leads to a
> left-shift on a negative value.
> 
>> -#define MAKE_THREAD_CPUCLOCK(tid, clock) \
>> -       MAKE_PROCESS_CPUCLOCK((tid), (clock) | CPUCLOCK_PERTHREAD_MASK)
> 
> OK. Same macro used. Same issue.
> 
>> +static inline clockid_t
>> +make_process_cpuclock (unsigned int pid, clockid_t clock)
>> +{
>> +  return ((~pid) << 3) | clock;
>> +}
>> +
> 
> OK. Replacement function that uses unsigned. Lines up with the version
> in Linux's sources.
> 
>> +static inline clockid_t
>> +make_thread_cpuclock (unsigned int tid, clockid_t clock)
>> +{
>> +  return make_process_cpuclock (tid, clock | CPUCLOCK_PERTHREAD_MASK);
>> +}
>> +
> 
> OK. Replaces the second macro for tid.
> 
>> +#define PROCESS_CLOCK  make_process_cpuclock (0, CPUCLOCK_SCHED)
>> +#define THREAD_CLOCK   make_thread_cpuclock (0, CPUCLOCK_SCHED)
> 
> OK. Macros for convenience.
> 
>> diff --git a/sysdeps/unix/sysv/linux/timer_create.c b/sysdeps/unix/sysv/linux/timer_create.c
>> index a8b2a41d9e..290324a7ea 100644
>> --- a/sysdeps/unix/sysv/linux/timer_create.c
>> +++ b/sysdeps/unix/sysv/linux/timer_create.c
>> @@ -33,9 +33,9 @@ ___timer_create (clockid_t clock_id, struct sigevent *evp, timer_t *timerid)
>>  {
>>    {
>>      clockid_t syscall_clockid = (clock_id == CLOCK_PROCESS_CPUTIME_ID
>> -                                ? MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED)
>> +                                ? PROCESS_CLOCK
>>                                  : clock_id == CLOCK_THREAD_CPUTIME_ID
>> -                                ? MAKE_THREAD_CPUCLOCK (0, CPUCLOCK_SCHED)
>> +                                ? THREAD_CLOCK
>>                                  : clock_id);
>>
>>      /* If the user wants notification via a thread we need to handle
> 
> OK. Use the new macros.
>