GIC v3 and later support SGI broadcast, i.e., the mode that routes
interrupts to all PEs in the system excluding the local CPU.

Supporting this mode can avoid looping through all the remote CPUs
when broadcasting SGIs, especially for systems with 200+ CPUs. The
performance improvement can be measured with the rest of this series
booted with "hugetlb_free_vmemmap=on irqchip.gicv3_pseudo_nmi=1":

  cd /sys/kernel/mm/hugepages/
  echo 600 >hugepages-1048576kB/nr_hugepages
  echo 2048kB >hugepages-1048576kB/demote_size
  perf record -g time echo 600 >hugepages-1048576kB/demote"

With 80 CPUs:
           gic_ipi_send_mask()  bash sys time
  Before:  38.14%               0m10.513s
  After:    0.20%               0m5.132s

Signed-off-by: Yu Zhao <yuzhao@google.com>
---
 drivers/irqchip/irq-gic-v3.c | 31 ++++++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)

diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
index ce87205e3e82..7ebe870e4608 100644
--- a/drivers/irqchip/irq-gic-v3.c
+++ b/drivers/irqchip/irq-gic-v3.c
@@ -1322,6 +1322,7 @@ static void gic_cpu_init(void)
 
 #define MPIDR_TO_SGI_RS(mpidr)	(MPIDR_RS(mpidr) << ICC_SGI1R_RS_SHIFT)
 #define MPIDR_TO_SGI_CLUSTER_ID(mpidr)	((mpidr) & ~0xFUL)
+#define MPIDR_TO_SGI_TARGET_LIST(mpidr)	(1 << ((mpidr) & 0xf))
 
 /*
  * gic_starting_cpu() is called after the last point where cpuhp is allowed
@@ -1356,7 +1357,7 @@ static u16 gic_compute_target_list(int *base_cpu, const struct cpumask *mask,
 	mpidr = gic_cpu_to_affinity(cpu);
 
 	while (cpu < nr_cpu_ids) {
-		tlist |= 1 << (mpidr & 0xf);
+		tlist |= MPIDR_TO_SGI_TARGET_LIST(mpidr);
 
 		next_cpu = cpumask_next(cpu, mask);
 		if (next_cpu >= nr_cpu_ids)
@@ -1394,9 +1395,20 @@ static void gic_send_sgi(u64 cluster_id, u16 tlist, unsigned int irq)
 	gic_write_sgi1r(val);
 }
 
+static void gic_broadcast_sgi(unsigned int irq)
+{
+	u64 val;
+
+	val = BIT_ULL(ICC_SGI1R_IRQ_ROUTING_MODE_BIT) | (irq << ICC_SGI1R_SGI_ID_SHIFT);
+
+	pr_devel("CPU %d: broadcasting SGI %u\n", smp_processor_id(), irq);
+	gic_write_sgi1r(val);
+}
+
 static void gic_ipi_send_mask(struct irq_data *d, const struct cpumask *mask)
 {
-	int cpu;
+	int cpu = smp_processor_id();
+	bool self = cpumask_test_cpu(cpu, mask);
 
 	if (WARN_ON(d->hwirq >= 16))
 		return;
@@ -1407,6 +1419,19 @@ static void gic_ipi_send_mask(struct irq_data *d, const struct cpumask *mask)
 	 */
 	dsb(ishst);
 
+	if (cpumask_weight(mask) + !self == num_online_cpus()) {
+		/* Broadcast to all but self */
+		gic_broadcast_sgi(d->hwirq);
+		if (self) {
+			unsigned long mpidr = gic_cpu_to_affinity(cpu);
+
+			/* Send to self */
+			gic_send_sgi(MPIDR_TO_SGI_CLUSTER_ID(mpidr),
+				     MPIDR_TO_SGI_TARGET_LIST(mpidr), d->hwirq);
+		}
+		goto done;
+	}
+
 	for_each_cpu(cpu, mask) {
 		u64 cluster_id = MPIDR_TO_SGI_CLUSTER_ID(gic_cpu_to_affinity(cpu));
 		u16 tlist;
@@ -1414,7 +1439,7 @@ static void gic_ipi_send_mask(struct irq_data *d, const struct cpumask *mask)
 		tlist = gic_compute_target_list(&cpu, mask, cluster_id);
 		gic_send_sgi(cluster_id, tlist, d->hwirq);
 	}
-
+done:
 	/* Force the above writes to ICC_SGI1R_EL1 to be executed */
 	isb();
 }
-- 
2.47.0.277.g8800431eea-goog