Hi James,

On Sun, Jan 03, 2021 at 09:11:22AM -0800, James Smart wrote:

+void
+efct_hw_map_wq_cpu(struct efct_hw *hw)
+{
+	struct efct *efct = hw->os;
+	u32 cpu = 0, i;
+
+	/* Init cpu_map array */
+	for_each_possible_cpu(cpu) {
+		hw->wq_cpu_array[cpu] = NULL;

This is asking for troubles:

	struct hw_wq		*wq_cpu_array[128];

+	for (i = 0; i < hw->config.n_eq; i++) {
+		const struct cpumask *maskp;
+
+		/* Get a CPU mask for all CPUs affinitized to this vector */
+		maskp = pci_irq_get_affinity(efct->pci, i);

Would it not make more sense to ask from the block layer which queue is
mapped to with CPU? blk_mq_map_queues or blk_mq_pci_map_queues? I am
still a bit confused a lot about all these queue/cpu mappings :)

I assuming the code for onlining/offlining CPUs is coming later. Are you
planing to use managed IRQs?

Thanks,
Daniel