[PATCH 07/16] smp: optimize smp_call_function_many_cond()

[Yury Norov]

smp_call_function_many_cond() is often passed with cpu_online_mask.
If this is the case, we can use num_online_cpus(), which is O(1)
instead of cpumask_{first,next}(), which is O(N).

It can be optimized further: if cpu_online_mask has 0 or single bit
set (depending on cpu_online(this_cpu), we can return result without
AND'ing with user's mask.

Caught with CONFIG_DEBUG_BITMAP:
[    7.830337] Call trace:
[    7.830397]  __bitmap_check_params+0x1d8/0x260
[    7.830499]  smp_call_function_many_cond+0x1e8/0x45c
[    7.830607]  kick_all_cpus_sync+0x44/0x80
[    7.830698]  bpf_int_jit_compile+0x34c/0x5cc
[    7.830796]  bpf_prog_select_runtime+0x118/0x190
[    7.830900]  bpf_prepare_filter+0x3dc/0x51c
[    7.830995]  __get_filter+0xd4/0x170
[    7.831145]  sk_attach_filter+0x18/0xb0
[    7.831236]  sock_setsockopt+0x5b0/0x1214
[    7.831330]  __sys_setsockopt+0x144/0x170
[    7.831431]  __arm64_sys_setsockopt+0x2c/0x40
[    7.831541]  invoke_syscall+0x48/0x114
[    7.831634]  el0_svc_common.constprop.0+0x44/0xfc
[    7.831745]  do_el0_svc+0x30/0xc0
[    7.831825]  el0_svc+0x2c/0x84
[    7.831899]  el0t_64_sync_handler+0xbc/0x140
[    7.831999]  el0t_64_sync+0x18c/0x190
[    7.832086] ---[ end trace 0000000000000000 ]---
[    7.832375] b1:		ffff24d1ffd98a48
[    7.832385] b2:		ffffa65533a29a38
[    7.832393] b3:		ffffa65533a29a38
[    7.832400] nbits:	256
[    7.832407] start:	0
[    7.832412] off:	0
[    7.832418] smp: Bitmap: parameters check failed
[    7.832432] smp: include/linux/bitmap.h [363]: bitmap_and

Signed-off-by: Yury Norov <yury.norov@xxxxxxxxx>
---
 kernel/smp.c | 29 +++++++++++++++++++++++------
 1 file changed, 23 insertions(+), 6 deletions(-)

diff --git a/kernel/smp.c b/kernel/smp.c
index dd215f439426..7ed2b9b12f74 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -880,6 +880,28 @@ EXPORT_SYMBOL_GPL(smp_call_function_any);
 #define SCF_WAIT	(1U << 0)
 #define SCF_RUN_LOCAL	(1U << 1)
 
+/* Check if we need remote execution, i.e., any CPU excluding this one. */
+static inline bool __need_remote_exec(const struct cpumask *mask, unsigned int this_cpu)
+{
+	unsigned int cpu;
+
+	switch (num_online_cpus()) {
+	case 0:
+		return false;
+	case 1:
+		return cpu_online(this_cpu) ? false : true;
+	default:
+		if (mask == cpu_online_mask)
+			return true;
+	}
+
+	cpu = cpumask_first_and(mask, cpu_online_mask);
+	if (cpu == this_cpu)
+		cpu = cpumask_next_and(cpu, mask, cpu_online_mask);
+
+	return cpu < nr_cpu_ids;
+}
+
 static void smp_call_function_many_cond(const struct cpumask *mask,
 					smp_call_func_t func, void *info,
 					unsigned int scf_flags,
@@ -916,12 +938,7 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 	if ((scf_flags & SCF_RUN_LOCAL) && cpumask_test_cpu(this_cpu, mask))
 		run_local = true;
 
-	/* Check if we need remote execution, i.e., any CPU excluding this one. */
-	cpu = cpumask_first_and(mask, cpu_online_mask);
-	if (cpu == this_cpu)
-		cpu = cpumask_next_and(cpu, mask, cpu_online_mask);
-	if (cpu < nr_cpu_ids)
-		run_remote = true;
+	run_remote = __need_remote_exec(mask, this_cpu);
 
 	if (run_remote) {
 		cfd = this_cpu_ptr(&cfd_data);
-- 
2.34.1