diff options
| author | Thomas Gleixner <tglx@linutronix.de> | 2019-07-09 16:23:40 +0200 |
|---|---|---|
| committer | Thomas Gleixner <tglx@linutronix.de> | 2019-07-25 15:48:01 +0200 |
| commit | 0c09ab96fc820109d63097a2adcbbd20836b655f (patch) | |
| tree | 813c75e34ed913bf7dbe7b8e7dfbda997ac27071 /kernel/cpu.c | |
| parent | b9fa6442f7043e2cdd247905d4f3b80f2e9605cb (diff) | |
cpu/hotplug: Cache number of online CPUs
Re-evaluating the bitmap wheight of the online cpus bitmap in every
invocation of num_online_cpus() over and over is a pretty useless
exercise. Especially when num_online_cpus() is used in code paths
like the IPI delivery of x86 or the membarrier code.
Cache the number of online CPUs in the core and just return the cached
variable. The accessor function provides only a snapshot when used without
protection against concurrent CPU hotplug.
The storage needs to use an atomic_t because the kexec and reboot code
(ab)use set_cpu_online() in their 'shutdown' handlers without any form of
serialization as pointed out by Mathieu. Regular CPU hotplug usage is
properly serialized.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1907091622590.1634@nanos.tec.linutronix.de
Diffstat (limited to 'kernel/cpu.c')
| -rw-r--r-- | kernel/cpu.c | 24 |
1 files changed, 24 insertions, 0 deletions
diff --git a/kernel/cpu.c b/kernel/cpu.c index 05778e32674a..e1967e9eddc2 100644 --- a/kernel/cpu.c +++ b/kernel/cpu.c @@ -2298,6 +2298,9 @@ EXPORT_SYMBOL(__cpu_present_mask); struct cpumask __cpu_active_mask __read_mostly; EXPORT_SYMBOL(__cpu_active_mask); +atomic_t __num_online_cpus __read_mostly; +EXPORT_SYMBOL(__num_online_cpus); + void init_cpu_present(const struct cpumask *src) { cpumask_copy(&__cpu_present_mask, src); @@ -2313,6 +2316,27 @@ void init_cpu_online(const struct cpumask *src) cpumask_copy(&__cpu_online_mask, src); } +void set_cpu_online(unsigned int cpu, bool online) +{ + /* + * atomic_inc/dec() is required to handle the horrid abuse of this + * function by the reboot and kexec code which invoke it from + * IPI/NMI broadcasts when shutting down CPUs. Invocation from + * regular CPU hotplug is properly serialized. + * + * Note, that the fact that __num_online_cpus is of type atomic_t + * does not protect readers which are not serialized against + * concurrent hotplug operations. + */ + if (online) { + if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask)) + atomic_inc(&__num_online_cpus); + } else { + if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask)) + atomic_dec(&__num_online_cpus); + } +} + /* * Activate the first processor. */ |