The branch main has been updated by cperciva: URL: https://cgit.FreeBSD.org/src/commit/?id=2e21f7e8140447c2dcbc964ff3482f12f9bd6683
commit 2e21f7e8140447c2dcbc964ff3482f12f9bd6683 Author: Colin Percival <[email protected]> AuthorDate: 2026-06-27 23:33:46 +0000 Commit: Colin Percival <[email protected]> CommitDate: 2026-07-14 01:19:48 +0000 ena: Put taskqueues into correct domain if !RSS When compiled without 'options RSS', the ena driver created taskqueues using taskqueue_start_threads_cpuset passing a mask value of NULL, both in the ena_setup_tx_resources path (for enqueues) and in the ena_create_io_queues path (for the completion-processing). In the default configuration, on most EC2 instances, this results in taskqueues running in the right NUMA domain, but only by accident; in non-default configurations (e.g. with with multiple EBS volumes attached and associated NVMe taskqueues) the taskqueues may land in the wrong NUMA domain even on instance types where the one-EBS-one-ENA case produces the desired results. Set (struct ena_que)->domain and use that to inform the choice of CPU sets. On a c8gn.48xlarge EC2 instance this doubles throughput on a 32-TCP-stream benchmark. Reviewed by: akiyano MFC after: 7 days Sponsored by: Amazon Differential Revision: https://reviews.freebsd.org/D57918 --- sys/dev/ena/ena.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/sys/dev/ena/ena.c b/sys/dev/ena/ena.c index 4f45d9103d76..79b9937991c1 100644 --- a/sys/dev/ena/ena.c +++ b/sys/dev/ena/ena.c @@ -726,6 +726,8 @@ ena_setup_tx_resources(struct ena_adapter *adapter, int qid) snprintf(thread_name, sizeof(thread_name), "%s txeq %d", device_get_nameunit(adapter->pdev), que->cpu); #else + if (que->domain >= 0) + cpu_mask = &cpuset_domain[que->domain]; snprintf(thread_name, sizeof(thread_name), "%s txeq %d", device_get_nameunit(adapter->pdev), que->id); #endif @@ -1671,6 +1673,9 @@ ena_create_io_queues(struct ena_adapter *adapter) #ifdef RSS cpu_mask = &queue->cpu_mask; +#else + if (queue->domain >= 0) + cpu_mask = &cpuset_domain[queue->domain]; #endif taskqueue_start_threads_cpuset(&queue->cleanup_tq, 1, PI_NET, cpu_mask, "%s queue %d cleanup", @@ -1815,6 +1820,8 @@ ena_setup_io_intr(struct ena_adapter *adapter) static int last_bind = 0; int cur_bind; int idx; +#else + int domain; #endif int irq_idx; @@ -1827,6 +1834,9 @@ ena_setup_io_intr(struct ena_adapter *adapter) last_bind = (last_bind + adapter->num_io_queues) % num_buckets; } cur_bind = adapter->first_bind; +#else + if (bus_get_domain(adapter->pdev, &domain)) + domain = -1; #endif for (int i = 0; i < adapter->num_io_queues; i++) { @@ -1860,7 +1870,7 @@ ena_setup_io_intr(struct ena_adapter *adapter) } adapter->que[i].domain = idx; #else - adapter->que[i].domain = -1; + adapter->que[i].domain = domain; #endif /* RSS */ }
