[PATCH 3/4] x86/apic: Introduce the per vector cpumask array

[Chen Yu <yu.c.chen@intel.com>]

This array is a bitmap for sorting the number of vectors
assigned on each CPU, thus to quickly retrieve the CPU
which have assigned the least amount of vectors, then
choose that CPU as a hint to spread the vectors among
multiple CPUs.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Len Brown <lenb@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
---
 arch/x86/kernel/apic/vector.c | 33 +++++++++++++++++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/arch/x86/kernel/apic/vector.c b/arch/x86/kernel/apic/vector.c
index 4ff84c0..b60cc66 100644
--- a/arch/x86/kernel/apic/vector.c
+++ b/arch/x86/kernel/apic/vector.c
@@ -32,6 +32,16 @@ struct irq_domain *x86_vector_domain;
 EXPORT_SYMBOL_GPL(x86_vector_domain);
 static DEFINE_RAW_SPINLOCK(vector_lock);
 static cpumask_var_t vector_cpumask, vector_searchmask, searched_cpumask;
+/*
+ * vectors_alloc_mask[i] records the CPUs which have assigned i vectors each.
+ * For example, let vectors_alloc_mask[3] = cpumask_or(cpumask_of(8), cpumask_of(9))
+ * which means that:
+ * On CPU8 and CPU9, the number of vectors been assigned is 3.
+ * This bitmap can be used to quickly find out the CPUs who has allocated the least
+ * number of vectors, by checking from vectors_alloc_mask[0] to vectors_alloc_mask[255]
+ * until a non-zero value is found. These CPUs are chosen for vector spreading.
+ */
+static cpumask_var_t vectors_alloc_mask[NR_VECTORS];
 static struct irq_chip lapic_controller;
 #ifdef	CONFIG_X86_IO_APIC
 static struct apic_chip_data *legacy_irq_data[NR_IRQS_LEGACY];
@@ -65,6 +75,9 @@ static void update_vectors_alloc(const struct cpumask *mask,
 			per_cpu(vector_irq, cpu).alloc -= count;
 		else
 			continue;
+		/* Update the position for this CPU. */
+		cpumask_clear_cpu(cpu, vectors_alloc_mask[cur_alloc]);
+		cpumask_set_cpu(cpu, vectors_alloc_mask[per_cpu(vector_irq, cpu).alloc]);
 	}
 }
 
@@ -479,6 +492,25 @@ static void __init init_legacy_irqs(void)
 static inline void init_legacy_irqs(void) { }
 #endif
 
+static int __init alloc_assigned_vectors_mask(void)
+{
+	int i, ret;
+
+	for (i = 0; i < NR_VECTORS; i++) {
+		ret = alloc_cpumask_var(&vectors_alloc_mask[i], GFP_KERNEL);
+		if (!ret)
+			goto free_mask;
+	}
+	/* Initially all the CPUs have 0 vector assigned. */
+	cpumask_setall(vectors_alloc_mask[0]);
+	return 0;
+
+ free_mask:
+	while (i--)
+		free_cpumask_var(vectors_alloc_mask[i]);
+	return -ENOMEM;
+}
+
 int __init arch_early_irq_init(void)
 {
 	struct fwnode_handle *fn;
@@ -499,6 +531,7 @@ int __init arch_early_irq_init(void)
 	BUG_ON(!alloc_cpumask_var(&vector_cpumask, GFP_KERNEL));
 	BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
 	BUG_ON(!alloc_cpumask_var(&searched_cpumask, GFP_KERNEL));
+	BUG_ON(alloc_assigned_vectors_mask());
 
 	return arch_early_ioapic_init();
 }
-- 
2.7.4