From: David Woodhouse <dwmw@xxxxxxxxxxxx>
There is often significant latency in the early stages of CPU bringup,
and time is wasted by waking each CPU (e.g. with SIPI/INIT/INIT on x86)
and then waiting for it to respond before moving on to the next.
Allow a platform to register a set of pre-bringup CPUHP states to which
each CPU can be stepped in parallel, thus absorbing some of that latency.
There is a subtlety here: even with an empty CPUHP_BP_PARALLEL_DYN step,
this means that *all* CPUs are brought through the prepare states and to
CPUHP_BP_PREPARE_DYN before any of them are taken to CPUHP_BRINGUP_CPU
and then are allowed to run for themselves to CPUHP_ONLINE.
So any combination of prepare/start calls which depend on A-B ordering
for each CPU in turn, such as the X2APIC code which used to allocate a
cluster mask 'just in case' and store it in a global variable in the
prep stage, then potentially consume that preallocated structure from
the AP and set the global pointer to NULL to be reallocated in
CPUHP_X2APIC_PREPARE for the next CPU... would explode horribly.
Any platform enabling the CPUHP_BP_PARALLEL_DYN steps must be reviewed
and tested to ensure that such issues do not exist, and the existing
behaviour of bringing CPUs to CPUHP_BP_PREPARE_DYN and then immediately
to CPUHP_BRINGUP_CPU and CPUHP_ONLINE only one at a time does not change
unless such a state is registered.
Note that the new parallel stages do *not* yet bring each AP to the
CPUHP_BRINGUP_CPU state at the same time, only to the new states which
exist before it. The final loop in bringup_nonboot_cpus() is untouched,
bringing each AP in turn from the final PARALLEL_DYN state (or all the
way from CPUHP_OFFLINE) to CPUHP_BRINGUP_CPU and then waiting for that
AP to do its own processing and reach CPUHP_ONLINE before releasing the
next.
Parallelising that part by bringing them all to CPUHP_BRINGUP_CPU
and then waiting for them all is an exercise for the future.
Signed-off-by: David Woodhouse <dwmw@xxxxxxxxxxxx>
Signed-off-by: Usama Arif <usama.arif@xxxxxxxxxxxxx>
Tested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
Tested-by: Kim Phillips <kim.phillips@xxxxxxx>
Tested-by: Oleksandr Natalenko <oleksandr@xxxxxxxxxxxxxx>
Tested-by: Guilherme G. Piccoli <gpiccoli@xxxxxxxxxx>
---
include/linux/cpuhotplug.h | 2 ++
kernel/cpu.c | 49 ++++++++++++++++++++++++++++++++++----
2 files changed, 47 insertions(+), 4 deletions(-)
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index c6fab004104a..ef3cf69a3d5b 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -133,6 +133,8 @@ enum cpuhp_state {
CPUHP_MIPS_SOC_PREPARE,
CPUHP_BP_PREPARE_DYN,
CPUHP_BP_PREPARE_DYN_END = CPUHP_BP_PREPARE_DYN + 20,
+ CPUHP_BP_PARALLEL_DYN,
+ CPUHP_BP_PARALLEL_DYN_END = CPUHP_BP_PARALLEL_DYN + 4,
CPUHP_BRINGUP_CPU,
/*
diff --git a/kernel/cpu.c b/kernel/cpu.c
index 43e0a77f21e8..cf3c1c6f0710 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -1504,13 +1504,49 @@ int bringup_hibernate_cpu(unsigned int sleep_cpu)
void bringup_nonboot_cpus(unsigned int setup_max_cpus)
{
+ unsigned int n = setup_max_cpus - num_online_cpus();
unsigned int cpu;
+ /*
+ * An architecture may have registered parallel pre-bringup states to
+ * which each CPU may be brought in parallel. For each such state,
+ * bring N CPUs to it in turn before the final round of bringing them
+ * online.
+ */
+ if (n > 0) {
+ enum cpuhp_state st = CPUHP_BP_PARALLEL_DYN;
+
+ while (st <= CPUHP_BP_PARALLEL_DYN_END && cpuhp_hp_states[st].name) {
+ int i = n;
+
+ for_each_present_cpu(cpu) {
+ cpu_up(cpu, st);
+ if (!--i)
+ break;
+ }
+ st++;
+ }
+ }
+
for_each_present_cpu(cpu) {
if (num_online_cpus() >= setup_max_cpus)
break;
- if (!cpu_online(cpu))
- cpu_up(cpu, CPUHP_ONLINE);
+ if (!cpu_online(cpu)) {
+ int ret = cpu_up(cpu, CPUHP_ONLINE);
+
+ /*
+ * For the parallel bringup case, roll all the way back
+ * to CPUHP_OFFLINE on failure; don't leave them in the
+ * parallel stages. This happens in the nosmt case for
+ * non-primary threads.
+ */
+ if (ret && cpuhp_hp_states[CPUHP_BP_PARALLEL_DYN].name) {
+ struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
+ if (can_rollback_cpu(st))
+ WARN_ON(cpuhp_invoke_callback_range(false, cpu, st,
+ CPUHP_OFFLINE));
+ }
+ }
}
}
@@ -1882,6 +1918,10 @@ static int cpuhp_reserve_state(enum cpuhp_state state)
step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
end = CPUHP_BP_PREPARE_DYN_END;
break;
+ case CPUHP_BP_PARALLEL_DYN:
+ step = cpuhp_hp_states + CPUHP_BP_PARALLEL_DYN;
+ end = CPUHP_BP_PARALLEL_DYN_END;
+ break;
default:
return -EINVAL;
}
@@ -1906,14 +1946,15 @@ static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
/*
* If name is NULL, then the state gets removed.
*
- * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
+ * CPUHP_AP_ONLINE_DYN and CPUHP_BP_P*_DYN are handed out on
* the first allocation from these dynamic ranges, so the removal
* would trigger a new allocation and clear the wrong (already
* empty) state, leaving the callbacks of the to be cleared state
* dangling, which causes wreckage on the next hotplug operation.
*/
if (name && (state == CPUHP_AP_ONLINE_DYN ||
- state == CPUHP_BP_PREPARE_DYN)) {
+ state == CPUHP_BP_PREPARE_DYN ||
+ state == CPUHP_BP_PARALLEL_DYN)) {
ret = cpuhp_reserve_state(state);
if (ret < 0)
return ret;
--
2.25.1
