[RFC -mm] more cpu_relax() places?

[Andreas Mohr <andi@rhlx01.fht-esslingen.de>]

Hi all,

while reviewing 2.6.17-rc6-mm1, I found some places that might
want to make use of cpu_relax() in order to not block secondary
pipelines while busy-polling (probably especially useful on SMT CPUs):

arch/i386/kernel/smp.c/flush_tlb_others():

        while (!cpus_empty(flush_cpumask))
                /* nothing. lockup detection does not belong here */
                mb();

should probably be made

        while (!cpus_empty(flush_cpumask)) {
		cpu_relax();
                /* nothing. lockup detection does not belong here */
                mb();
	}

(to have memory barrier directly before flush_cpumask is read).


Second,
include/asm-i386/apic.h/apic_wait_icr_idle() does use cpu_relax(),
but the version in asm-x86_64/ NOT!
Is this because there's not much use doing cpu_relax() on SMP non-SMT
machines, and x86_64 are always non-SMT? Or rather because someone
screwed up?


And what about include/asm-i386/acpi.h/__acpi_acquire/release_global_lock()?

static inline int
__acpi_acquire_global_lock (unsigned int *lock)
{
        unsigned int old, new, val;
        do {
                old = *lock;
                new = (((old & ~0x3) + 2) + ((old >> 1) & 0x1));
                val = cmpxchg(lock, old, new);
        } while (unlikely (val != old));
        return (new < 3) ? -1 : 0;
}

could probably be made

static inline int
__acpi_acquire_global_lock (unsigned int *lock)
{
        unsigned int old, new, val;
        while (1) {
                old = *lock;
                new = (((old & ~0x3) + 2) + ((old >> 1) & 0x1));
                val = cmpxchg(lock, old, new);
		if (likely(val == old))
			break;
		cpu_relax();
        }
        return (new < 3) ? -1 : 0;
}

Andreas Mohr