@@ -2793,6 +2793,9 @@ _base_free_irq(struct MPT3SAS_ADAPTER *ioc)
list_for_each_entry_safe(reply_q, next, &ioc->reply_queue_list, list) {
list_del(&reply_q->list);
+ if (smp_affinity_enable)
+ irq_set_affinity_hint(pci_irq_vector(ioc->pdev,
+ reply_q->msix_index), NULL);
free_irq(pci_irq_vector(ioc->pdev, reply_q->msix_index),
reply_q);
kfree(reply_q);
@@ -2857,6 +2860,7 @@ _base_assign_reply_queues(struct MPT3SAS_ADAPTER *ioc)
{
unsigned int cpu, nr_cpus, nr_msix, index = 0;
struct adapter_reply_queue *reply_q;
+ int local_numa_node;
if (!_base_is_controller_msix_enabled(ioc))
return;
@@ -2875,13 +2879,32 @@ _base_assign_reply_queues(struct MPT3SAS_ADAPTER *ioc)
return;
if (smp_affinity_enable) {
+
+ /*
+ * set irq affinity to local numa node for those irqs
+ * corresponding to high iops queues.
+ */
+ if (ioc->high_iops_queues) {
+ local_numa_node = dev_to_node(&ioc->pdev->dev);
+ for (index = 0; index < ioc->high_iops_queues;
+ index++) {
+ irq_set_affinity_hint(pci_irq_vector(ioc->pdev,
+ index), cpumask_of_node(local_numa_node));
+ }
+ }
+
list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
- const cpumask_t *mask = pci_irq_get_affinity(ioc->pdev,
- reply_q->msix_index);
+ const cpumask_t *mask;
+
+ if (reply_q->msix_index < ioc->high_iops_queues)
+ continue;
+
+ mask = pci_irq_get_affinity(ioc->pdev,
+ reply_q->msix_index);
if (!mask) {
ioc_warn(ioc, "no affinity for msi %x\n",
reply_q->msix_index);
- continue;
+ goto fall_back;
}
for_each_cpu_and(cpu, mask, cpu_online_mask) {
@@ -2892,12 +2915,18 @@ _base_assign_reply_queues(struct MPT3SAS_ADAPTER *ioc)
}
return;
}
+
+fall_back:
cpu = cpumask_first(cpu_online_mask);
+ nr_msix -= ioc->high_iops_queues;
+ index = 0;
list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
-
unsigned int i, group = nr_cpus / nr_msix;
+ if (reply_q->msix_index < ioc->high_iops_queues)
+ continue;
+
if (cpu >= nr_cpus)
break;
@@ -2950,10 +2979,37 @@ _base_disable_msix(struct MPT3SAS_ADAPTER *ioc)
{
if (!ioc->msix_enable)
return;
- pci_disable_msix(ioc->pdev);
+ pci_free_irq_vectors(ioc->pdev);
ioc->msix_enable = 0;
}
+/**
+ * _base_alloc_irq_vectors - allocate msix vectors
+ * @ioc: per adapter object
+ *
+ */
+static int
+_base_alloc_irq_vectors(struct MPT3SAS_ADAPTER *ioc)
+{
+ int i, irq_flags = PCI_IRQ_MSIX;
+ struct irq_affinity desc = { .pre_vectors = ioc->high_iops_queues };
+ struct irq_affinity *descp = &desc;
+
+ if (smp_affinity_enable)
+ irq_flags |= PCI_IRQ_AFFINITY;
+ else
+ descp = NULL;
+
+ ioc_info(ioc, " %d %d\n", ioc->high_iops_queues,
+ ioc->msix_vector_count);
+
+ i = pci_alloc_irq_vectors_affinity(ioc->pdev,
+ ioc->high_iops_queues,
+ ioc->msix_vector_count, irq_flags, descp);
+
+ return i;
+}
+
/**
* _base_enable_msix - enables msix, failback to io_apic
* @ioc: per adapter object
@@ -2965,7 +3021,6 @@ _base_enable_msix(struct MPT3SAS_ADAPTER *ioc)
int r;
int i, local_max_msix_vectors;
u8 try_msix = 0;
- unsigned int irq_flags = PCI_IRQ_MSIX;
if (msix_disable == -1 || msix_disable == 0)
try_msix = 1;
@@ -2999,11 +3054,7 @@ _base_enable_msix(struct MPT3SAS_ADAPTER *ioc)
if (ioc->msix_vector_count < ioc->cpu_count)
smp_affinity_enable = 0;
- if (smp_affinity_enable)
- irq_flags |= PCI_IRQ_AFFINITY;
-
- r = pci_alloc_irq_vectors(ioc->pdev, 1, ioc->reply_queue_count,
- irq_flags);
+ r = _base_alloc_irq_vectors(ioc);
if (r < 0) {
dfailprintk(ioc,
ioc_info(ioc, "pci_alloc_irq_vectors failed (r=%d) !!!\n",
High iops queues are mapped to non-managed irqs. Set affinity of non-managed irqs to local numa node. Low latency queues are mapped to managed irq. Driver reserves some reply queues (pci_alloc_irq_vectors_affinity and .pre_vectors interface is used to meet the goal) for max iops and rest of queues for low latency. Based on io workload in io submission path driver will decide which group of reply queues (either high iops queues or low latency queues) to be used. High iops queues will be mapped to local numa node of controller and low latency queues will be mapped to cpus across numa nodes. In general, high iops queue and low latency queue together should fit into 128 reply queue (max reply queue supported by Aero/Sea). Signed-off-by: Suganath Prabu S <suganath-prabu.subramani@broadcom.com> --- drivers/scsi/mpt3sas/mpt3sas_base.c | 73 ++++++++++++++++++++++++----- 1 file changed, 62 insertions(+), 11 deletions(-)