On Fri, 16 Jun 2017, Thomas Gleixner wrote:

On Sat, 3 Jun 2017, Christoph Hellwig wrote:

quoted

+bool irq_affinity_set(int irq, struct irq_desc *desc, const cpumask_t *mask)
+{
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct irq_chip *chip = irq_data_get_irq_chip(data);
+	bool ret = false;
+
+	if (!irq_can_move_pcntxt(data) && chip->irq_mask)
+		chip->irq_mask(data);
+
+	if (chip->irq_set_affinity) {
+		if (chip->irq_set_affinity(data, mask, true) == -ENOSPC)
+			pr_crit("IRQ %d set affinity failed because there are no available vectors.  The device assigned to this IRQ is unstable.\n", irq);
+		ret = true;
+	}
+
+	/*
+	 * We unmask if the irq was not marked masked by the core code.
+	 * That respects the lazy irq disable behaviour.
+	 */
+	if (!irq_can_move_pcntxt(data) &&
+	    !irqd_irq_masked(data) && chip->irq_unmask)
+		chip->irq_unmask(data);

There is another issue with this. Nothing updates the affinity mask in
irq_desc, when we just invoke the chip callback. Let me have a look.

Indeed. So that magic you do in the next patches (the hotplug callbacks)
only work proper for affinity masks with a single cpu set.

The problem is that we don't have a distinction between the 'possible'
(e.g. set by /proc/irq/affinity) and the effective affinity mask.

Needs more thought.

Thanks,

	tglx