smp/up: Make smp_call_function_single() match SMP semantics
authorPaul E. McKenney <paulmck@kernel.org>
Wed, 5 Feb 2020 14:34:09 +0000 (06:34 -0800)
committerThomas Gleixner <tglx@linutronix.de>
Fri, 7 Feb 2020 14:34:12 +0000 (15:34 +0100)
In CONFIG_SMP=y kernels, smp_call_function_single() returns -ENXIO when
invoked for a non-existent CPU.  In contrast, in CONFIG_SMP=n kernels,
a splat is emitted and smp_call_function_single() otherwise silently
ignores its "cpu" argument, instead pretending that the caller intended
to have something happen on CPU 0.  Given that there is now code that
expects smp_call_function_single() to return an error if a bad CPU was
specified, this difference in semantics needs to be addressed.

Bring the semantics of the CONFIG_SMP=n version of
smp_call_function_single() into alignment with its CONFIG_SMP=y
counterpart.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20200205143409.GA7021@paulmck-ThinkPad-P72
kernel/up.c

index 53144d0562522e6d00824073f65ffa9926ae6261..c6f323dcd45bb9efe1401b1b7893350667305493 100644 (file)
@@ -14,7 +14,8 @@ int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
 {
        unsigned long flags;
 
-       WARN_ON(cpu != 0);
+       if (cpu != 0)
+               return -ENXIO;
 
        local_irq_save(flags);
        func(info);