Use NUMA node cpu mask in irq affinity [v3]

[Resending ... Seems to have died on the vine here
http://marc.info/?l=linux-kernel&m=132914231706596&q=raw ]

The irq affinity files (/proc/irq/.../smp_affinity) contain a mask that is used to "pin" an irq to a set of cpus. On boot this set is currently all cpus. This can be incorrect as ACPI SRAT may tell us that a specific device or bus is attached to a particular node and it's cpus.

When setting up the irq affinity we should take into account the NUMA node cpu mask by and'ing it into the irq's affinity mask.

[v2]: Thomas and Yinghai pointed out that it is possible that the mask is empty after and'ing in the NUMA node's cpumask. The code now checks to see if the mask will be empty before applying the NUMA node's cpumask.

[v3]: David requests NUMA_NO_NODE instead of -1

---
 kernel/irq/manage.c |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)

index b0ccd1a..96a18f1 100644
@@ -301,6 +301,13 @@ setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
 	}
 
 	cpumask_and(mask, cpu_online_mask, set);
+	if (desc->irq_data.node != NUMA_NO_NODE) {
+		const struct cpumask *nodemask =
+					   cpumask_of_node(desc->irq_data.node);
+		/* make sure at least one of the cpus in nodemask is online */
+		if (cpumask_intersects(mask, nodemask))
+			cpumask_and(mask, mask, nodemask);
+	}
 	ret = chip->irq_set_affinity(&desc->irq_data, mask, false);
 	switch (ret) {
 	case IRQ_SET_MASK_OK:

1.7.1

Commit-ID:  241fc640be783f903e74b6d9c68481c01873f758
Gitweb:     http://git.kernel.org/tip/241fc640be783f903e74b6d9c68481c01873f758
Author:     Prarit Bhargava <[email protected]>
AuthorDate: Mon, 26 Mar 2012 15:02:18 -0400
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Thu, 29 Mar 2012 11:31:53 +0200

genirq: Respect NUMA node affinity in setup_irq_irq affinity()

We respect node affinity of devices already in the irq descriptor allocation, but we ignore it for the initial interrupt affinity setup, so the interrupt might be routed to a different node.

Restrict the default affinity mask to the node on which the irq descriptor is allocated.

[ tglx: Massaged changelog ]

---
 kernel/irq/manage.c |    9 ++++++++-
 1 files changed, 8 insertions(+), 1 deletions(-)

index bf606a5..89a3ea8 100644
@@ -282,7 +282,7 @@ setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
 {
 	struct irq_chip *chip = irq_desc_get_chip(desc);
 	struct cpumask *set = irq_default_affinity;
-	int ret;
+	int ret, node = desc->irq_data.node;
 
 	/* Excludes PER_CPU and NO_BALANCE interrupts */
 	if (!irq_can_set_affinity(irq))

@@ -301,6 +301,13 @@ setup_affinity(unsigned int irq, struct irq_desc *desc, struct cpumask *mask)
 	}
 
 	cpumask_and(mask, cpu_online_mask, set);
+	if (node != NUMA_NO_NODE) {
+		const struct cpumask *nodemask = cpumask_of_node(node);
+
+		/* make sure at least one of the cpus in nodemask is online */
+		if (cpumask_intersects(mask, nodemask))
+			cpumask_and(mask, mask, nodemask);
+	}
 	ret = chip->irq_set_affinity(&desc->irq_data, mask, false);
 	switch (ret) {
 	case IRQ_SET_MASK_OK: