Re: Build failures due to missing 'posix_timer_event'

From: Stafford Horne
Date: Wed Dec 14 2016 - 21:56:02 EST


Hello,

On Wed, Dec 14, 2016 at 10:14:39AM -0500, Nicolas Pitre wrote:
> On Wed, 14 Dec 2016, Guenter Roeck wrote:
>
> > avr32:allnoconfig:
> >
> > kernel/built-in.o: In function `do_adjtimex':
> > (.text+0x1d748): undefined reference to `posix_timer_event'
> > make[1]: *** [vmlinux] Error 1
> >
> > metag:allnoconfig:
> >
> > kernel/built-in.o: In function `alarm_handle_timer':
> > alarmtimer.c:(.text.alarm_handle_timer+0x38): undefined reference to
> > `posix_timer_event'
> >
> > openrisc:allnoconfig:
> >
> > kernel/built-in.o: In function `alarm_handle_timer':
> > alarmtimer.c:(.text+0x3fca4): undefined reference to `posix_timer_event'
> >
> > unicore32:allnoconfig:
> >
> > kernel/built-in.o: In function `alarm_handle_timer':
> > memremap.c:(.text+0x31c1c): undefined reference to `posix_timer_event'
> >
> > Since I don't run allnoconfig builds for all architectures, I strongly suspect
> > that more are affected.
>
> This is a known problem. Oddly enough only those "less popular"
> architectures are affected.
>
> > Looking into alarmtimer.c, with seems to call posix_timer_event()
> > unconditionally, I don't entirely understand how this can work in the
> > first place. Maybe I am missing something, though.
>
> The magic is in alarmtimer_init():
>
> if (IS_ENABLED(CONFIG_POSIX_TIMERS)) {
> posix_timers_register_clock(CLOCK_REALTIME_ALARM, &alarm_clock);
> posix_timers_register_clock(CLOCK_BOOTTIME_ALARM, &alarm_clock);
> }
>
> If CONFIG_POSIX_TIMERS is not enabled, the condition becomes a constant
> zero. The posix_timers_register_clock calls are then optimized away by
> the compiler. In turn the alarm_clock structure has no longer any
> reference to it and therefore is also optimized away, including its
> method functions such as alarm_timer_create() and finally
> alarm_handle_timer().
>
> But for some reasons, some gcc versions (mainly for "secondary"
> architectures) fail to optimize away all that code, and a dead reference
> to posix_timer_event() remains in the compiled code.
>
> There is a fix already queued in the mm tree. It goes like this:
>
> posix-timers: give lazy compilers some help optimizing code away
>
> The OpenRISC compiler (so far) fails to optimize away a large portion
> of code containing a reference to posix_timer_event in alarmtimer.c when
> CONFIG_POSIX_TIMERS is unset. Let's give it a direct clue to let the build
> succeed.

To be fair, this might be related to compiler version only. I guess its
not an excuse, but when i compile with or1k-elf-gcc (version 5.3.0) I
dont get this issue.

I guess other archs are not breaking because they are being tested is a
newer compiler?

Tested will allnoconfig

Compiler version:

Using built-in specs.
COLLECT_GCC=or1k-elf-gcc
COLLECT_LTO_WRAPPER=/opt/shorne/software/or1k/libexec/gcc/or1k-elf/5.3.0/lto-wrapper
Target: or1k-elf
Configured with: ../or1k-gcc/configure --target=or1k-elf --prefix=/opt/shorne/software/or1k --enable-languages=c,c++ --disable-shared --disable-libssp --with-newlib
Thread model: single
gcc version 5.3.0 (GCC)

-Stafford

> This fixes
> [linux-next:master 6682/7183] alarmtimer.c:undefined reference to `posix_timer_event'
> reported by kbuild test robot.
>
> Signed-off-by: Nicolas Pitre <nico@xxxxxxxxxx>
> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Cc: Josh Triplett <josh@xxxxxxxxxxxxxxxx>
> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>
> diff --git a/kernel/time/alarmtimer.c b/kernel/time/alarmtimer.c
> index 9b08ca391a..3921cf7fea 100644
> --- a/kernel/time/alarmtimer.c
> +++ b/kernel/time/alarmtimer.c
> @@ -516,7 +516,8 @@ static enum alarmtimer_restart alarm_handle_timer(struct alarm *alarm,
>
> spin_lock_irqsave(&ptr->it_lock, flags);
> if ((ptr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) {
> - if (posix_timer_event(ptr, 0) != 0)
> + if (IS_ENABLED(CONFIG_POSIX_TIMERS) &&
> + posix_timer_event(ptr, 0) != 0)
> ptr->it_overrun++;
> }
>