[PATCH 18/18] linux: Only use 64-bit syscall if required for clock_nanosleep

[Adhemerval Zanella]

On 17/06/2021 12:11, Lukasz Majewski wrote:
> Hi Adhemerval,
> 
>> For !__ASSUME_TIME64_SYSCALLS there is no need to issue a 64-bit
>> syscall if the provided timeout fits in a 32-bit one.  The 64-bit
>> usage should be rare since the timeout is a relative one.
> 
> I'm not sure if the above description is correct.
> 
> As fair as I understand this patch set - we check if timeout fits into
> 32-bit if:
> 
> - We want to use legacy (32 bit) syscall when __ASSUME_TIME64_SYSCALLS
>   is not defined to avoid calling syscalls twice.
> 
> The problem with relative 64 bit timeouts will be apparent after we
> pass Y2038 year threshold. 

Yes, but 64-bit relative timeout means that caller is passing a *very*
large timeout.  It is possible indeed, but I don't think it the usual
way.

> 
>>
>> Checked on i686-linux-gnu on a 4.15 kernel and on a 5.11 kernel
>> (with and without --enable-kernel=5.1) and on x86_64-linux-gnu.
>> ---
>>  sysdeps/unix/sysv/linux/clock_nanosleep.c | 47
>> +++++++++++++---------- time/Makefile                             |
>> 9 +++++ time/tst-clock_nanosleep.c                | 40
>> +++++++++++-------- 3 files changed, 60 insertions(+), 36 deletions(-)
>>
>> diff --git a/sysdeps/unix/sysv/linux/clock_nanosleep.c
>> b/sysdeps/unix/sysv/linux/clock_nanosleep.c index
>> 007f1736cb..46b0f1e269 100644 ---
>> a/sysdeps/unix/sysv/linux/clock_nanosleep.c +++
>> b/sysdeps/unix/sysv/linux/clock_nanosleep.c @@ -27,8 +27,9 @@
>>  /* We can simply use the syscall.  The CPU clocks are not supported
>>     with this function.  */
>>  int
>> -__clock_nanosleep_time64 (clockid_t clock_id, int flags, const
>> struct __timespec64 *req,
>> -                          struct __timespec64 *rem)
>> +__clock_nanosleep_time64 (clockid_t clock_id, int flags,
>> +			  const struct __timespec64 *req,
>> +			  struct __timespec64 *rem)
>>  {
>>    if (clock_id == CLOCK_THREAD_CPUTIME_ID)
>>      return EINVAL;
>> @@ -37,33 +38,37 @@ __clock_nanosleep_time64 (clockid_t clock_id, int
>> flags, const struct __timespec 
>>    /* If the call is interrupted by a signal handler or encounters an
>> error, it returns a positive value similar to errno.  */
>> +
>>  #ifndef __NR_clock_nanosleep_time64
>>  # define __NR_clock_nanosleep_time64 __NR_clock_nanosleep
>>  #endif
>> -  int r = INTERNAL_SYSCALL_CANCEL (clock_nanosleep_time64, clock_id,
>> -				   flags, req, rem);
>> -
>> -#ifndef __ASSUME_TIME64_SYSCALLS
>> -  if (r == 0 || r != -ENOSYS)
>> -    return -r;
>>  
>> -  if (! in_time_t_range (req->tv_sec))
>> +  int r;
>> +#ifdef __ASSUME_TIME64_SYSCALLS
>> +  r = INTERNAL_SYSCALL_CANCEL (clock_nanosleep_time64, clock_id,
>> flags, req,
>> +			       rem);
>> +#else
>> +  bool is32bit = in_time_t_range (req->tv_sec);
>> +  if (!is32bit)
>>      {
>> -      __set_errno (EOVERFLOW);
>> -      return -1;
>> +      r = INTERNAL_SYSCALL_CANCEL (clock_nanosleep_time64, clock_id,
>> flags,
>> +				   req, rem);
>> +      if (r == -ENOSYS)
>> +	r = -EOVERFLOW;
>>      }
>> -
>> -  struct timespec tr32;
>> -  struct timespec ts32 = valid_timespec64_to_timespec (*req);
>> -  r = INTERNAL_SYSCALL_CANCEL (clock_nanosleep, clock_id, flags,
>> -                               &ts32, &tr32);
>> -  if (INTERNAL_SYSCALL_ERROR_P (r))
>> +  else
>>      {
>> -      if (r == -EINTR && rem != NULL && (flags & TIMER_ABSTIME) == 0)
>> -	*rem = valid_timespec_to_timespec64 (tr32);
>> +      struct timespec tr32;
>> +      struct timespec ts32 = valid_timespec64_to_timespec (*req);
>> +      r = INTERNAL_SYSCALL_CANCEL (clock_nanosleep, clock_id, flags,
>> &ts32,
>> +				   &tr32);
>> +      if (INTERNAL_SYSCALL_ERROR_P (r))
>> +	{
>> +	  if (r == -EINTR && rem != NULL && (flags & TIMER_ABSTIME)
>> == 0)
>> +	    *rem = valid_timespec_to_timespec64 (tr32);
>> +	}
>>      }
>> -#endif /* __ASSUME_TIME64_SYSCALLS */
>> -
>> +#endif
>>    return -r;
>>  }
>>  
>> diff --git a/time/Makefile b/time/Makefile
>> index c84bd5d3ec..0bea84966c 100644
>> --- a/time/Makefile
>> +++ b/time/Makefile
>> @@ -86,6 +86,15 @@ $(objpfx)tst-strftime2.out: $(gen-locales)
>>  $(objpfx)tst-strftime3.out: $(gen-locales)
>>  endif
>>  
>> +ifeq (yes,$(build-shared))
>> +librt = $(common-objpfx)rt/librt.so
>> +else
>> +librt = $(common-objpfx)rt/librt.a
>> +endif
>> +
>> +$(objpfx)tst-clock_nanosleep: $(librt)
>> +$(objpfx)tst-clock_nanosleep-time64: $(librt)
>> +
>>  tz-cflags = -DTZDIR='"$(zonedir)"' \
>>  	    -DTZDEFAULT='"$(localtime-file)"' \
>>  	    -DTZDEFRULES='"$(posixrules-file)"'
>> diff --git a/time/tst-clock_nanosleep.c b/time/tst-clock_nanosleep.c
>> index 47537435c1..a5a7f9430a 100644
>> --- a/time/tst-clock_nanosleep.c
>> +++ b/time/tst-clock_nanosleep.c
>> @@ -20,38 +20,48 @@
>>  #include <unistd.h>
>>  #include <sys/time.h>
>>  #include <time.h>
>> -
>> +#include <intprops.h>
>> +#include <support/support.h>
>> +#include <support/check.h>
>>  
>>  /* Test that clock_nanosleep() does sleep.  */
>> -static int
>> -do_test (void)
>> +static void
>> +clock_nanosleep_test (void)
>>  {
>>    /* Current time.  */
>>    struct timeval tv1;
>> -  (void) gettimeofday (&tv1, NULL);
>> +  gettimeofday (&tv1, NULL);
>>  
>> -  struct timespec ts;
>> -  ts.tv_sec = 1;
>> -  ts.tv_nsec = 0;
>> +  struct timespec ts = { 1, 0 };
>>    TEMP_FAILURE_RETRY (clock_nanosleep (CLOCK_REALTIME, 0, &ts, &ts));
>>  
>>    /* At least one second must have passed.  */
>>    struct timeval tv2;
>> -  (void) gettimeofday (&tv2, NULL);
>> +  gettimeofday (&tv2, NULL);
>>  
>>    tv2.tv_sec -= tv1.tv_sec;
>>    tv2.tv_usec -= tv1.tv_usec;
>>    if (tv2.tv_usec < 0)
>>      --tv2.tv_sec;
>>  
>> -  if (tv2.tv_sec < 1)
>> -    {
>> -      puts ("clock_nanosleep didn't sleep long enough");
>> -      return 1;
>> -    }
>> +  TEST_VERIFY (tv2.tv_sec >= 1);
>> +}
>> +
>> +static void
>> +clock_nanosleep_large_timeout (void)
>> +{
> 
> Please correct me if I'm wrong:
> 
>> +  support_create_timer (0, 100000000, false, NULL);
> 
> You create timer with 100ms timeout
> 
>> +  struct timespec ts = { TYPE_MAXIMUM (time_t), 0 };
> 
> ts is created to have maximal value for time_t seconds.
> 
>> +  int r = clock_nanosleep (CLOCK_REALTIME, 0, &ts, NULL);
> 
> The clock_nanosleep is called (64 or 32 bit version - with proper
> in-glibc aliasing).
> 
>> +  TEST_VERIFY (r == EINTR || r == EOVERFLOW);
> 
> We check if:
> 
> - return is EOVERFLOW - this means that we want to use 64 bit time on
>   machine which is only supporting 64 bit time.

We want to use a 64-bit relative timeout on a machine where 
__NR_clock_nanosleep_time64 failed with ENOSYS, meaning a pre v5.1 kernel
without 64-bit time_t support.

> 
> - return is EINTR - as the timer's (introduced in the earlier patch in
>   this set) alarm triggers. Am I right that this check is to find out
>   if the syscall is correctly executed by the kernel?

Yes, the EINTR means that syscall was interrupted by the alarm.

> 
>   Or is there any other issue?

I think these are the two possible situations. 

> 
>> +}
>>  
>> +static int
>> +do_test (void)
>> +{
>> +  clock_nanosleep_test ();
>> +  clock_nanosleep_large_timeout ();
>>    return 0;
>>  }
>>  
>> -#define TEST_FUNCTION do_test ()
>> -#include "../test-skeleton.c"
>> +#include <support/test-driver.c>
> 
> 
> 
> Best regards,
> 
> Lukasz Majewski
> 
> --
> 
> DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
> HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
> Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de
>