From: Souradeep Chakrabarti <schakrabarti@linux.microsoft.com>
To: kys@microsoft.com, haiyangz@microsoft.com, wei.liu@kernel.org,
decui@microsoft.com, davem@davemloft.net, edumazet@google.com,
kuba@kernel.org, pabeni@redhat.com, longli@microsoft.com,
sharmaajay@microsoft.com, leon@kernel.org, cai.huoqing@linux.dev,
ssengar@linux.microsoft.com, vkuznets@redhat.com,
tglx@linutronix.de, linux-hyperv@vger.kernel.org,
netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
linux-rdma@vger.kernel.org
Cc: schakrabarti@microsoft.com, paulros@microsoft.com,
Souradeep Chakrabarti <schakrabarti@linux.microsoft.com>
Subject: [PATCH V2 net-next] net: mana: Assigning IRQ affinity on HT cores
Date: Tue, 21 Nov 2023 05:54:37 -0800 [thread overview]
Message-ID: <1700574877-6037-1-git-send-email-schakrabarti@linux.microsoft.com> (raw)
Existing MANA design assigns IRQ to every CPUs, including sibling hyper-threads
in a core. This causes multiple IRQs to work on same CPU and may reduce the network
performance with RSS.
Improve the performance by adhering the configuration for RSS, which assigns IRQ
on HT cores.
Signed-off-by: Souradeep Chakrabarti <schakrabarti@linux.microsoft.com>
---
V1 -> V2:
* Simplified the code by removing filter_mask_list and using avail_cpus.
* Addressed infinite loop issue when there are numa nodes with no CPUs.
* Addressed uses of local numa node instead of 0 to start.
* Removed uses of BUG_ON.
* Placed cpus_read_lock in parent function to avoid num_online_cpus
to get changed before function finishes the affinity assignment.
---
.../net/ethernet/microsoft/mana/gdma_main.c | 134 ++++++++++++++++--
1 file changed, 123 insertions(+), 11 deletions(-)
diff --git a/drivers/net/ethernet/microsoft/mana/gdma_main.c b/drivers/net/ethernet/microsoft/mana/gdma_main.c
index 6367de0c2c2e..8177502ffbd9 100644
--- a/drivers/net/ethernet/microsoft/mana/gdma_main.c
+++ b/drivers/net/ethernet/microsoft/mana/gdma_main.c
@@ -1243,15 +1243,120 @@ void mana_gd_free_res_map(struct gdma_resource *r)
r->size = 0;
}
+static int irq_setup(int *irqs, int nvec, int start_numa_node)
+{
+ unsigned int *core_id_list;
+ cpumask_var_t filter_mask, avail_cpus;
+ int i, core_count = 0, cpu_count = 0, err = 0, node_count = 0;
+ unsigned int cpu_first, cpu, irq_start, cores = 0, numa_node = start_numa_node;
+
+ if(!alloc_cpumask_var(&filter_mask, GFP_KERNEL)
+ || !alloc_cpumask_var(&avail_cpus, GFP_KERNEL)) {
+ err = -ENOMEM;
+ goto free_irq;
+ }
+ cpumask_copy(filter_mask, cpu_online_mask);
+ cpumask_copy(avail_cpus, cpu_online_mask);
+ /* count the number of cores
+ */
+ for_each_cpu(cpu, filter_mask) {
+ cpumask_andnot(filter_mask, filter_mask, topology_sibling_cpumask(cpu));
+ cores++;
+ }
+ core_id_list = kcalloc(cores, sizeof(unsigned int), GFP_KERNEL);
+ cpumask_copy(filter_mask, cpu_online_mask);
+ /* initialize core_id_list array */
+ for_each_cpu(cpu, filter_mask) {
+ core_id_list[core_count] = cpu;
+ cpumask_andnot(filter_mask, filter_mask, topology_sibling_cpumask(cpu));
+ core_count++;
+ }
+
+ /* if number of cpus are equal to max_queues per port, then
+ * one extra interrupt for the hardware channel communication.
+ */
+ if (nvec - 1 == num_online_cpus()) {
+ irq_start = 1;
+ cpu_first = cpumask_first(cpu_online_mask);
+ irq_set_affinity_and_hint(irqs[0], cpumask_of(cpu_first));
+ } else {
+ irq_start = 0;
+ }
+
+ /* reset the core_count and num_node to 0.
+ */
+ core_count = 0;
+
+ /* for each interrupt find the cpu of a particular
+ * sibling set and if it belongs to the specific numa
+ * then assign irq to it and clear the cpu bit from
+ * the corresponding avail_cpus.
+ * Increase the cpu_count for that node.
+ * Once all cpus for a numa node is assigned, then
+ * move to different numa node and continue the same.
+ */
+ for (i = irq_start; i < nvec; ) {
+
+ /* check if the numa node has cpu or not
+ * to avoid infinite loop.
+ */
+ if (cpumask_empty(cpumask_of_node(numa_node))) {
+ numa_node++;
+ if (++node_count == num_online_nodes()) {
+ err = -EAGAIN;
+ goto free_irq;
+ }
+ }
+ cpu_first = cpumask_first_and(avail_cpus,
+ topology_sibling_cpumask(core_id_list[core_count]));
+ if (cpu_first < nr_cpu_ids && cpu_to_node(cpu_first) == numa_node) {
+ irq_set_affinity_and_hint(irqs[i], cpumask_of(cpu_first));
+ cpumask_clear_cpu(cpu_first, avail_cpus);
+ cpu_count = cpu_count + 1;
+ i = i + 1;
+
+ /* checking if all the cpus are used from the
+ * particular node.
+ */
+ if (cpu_count == nr_cpus_node(numa_node)) {
+ numa_node = numa_node + 1;
+ if (numa_node == num_online_nodes())
+ numa_node = 0;
+
+ /* wrap around once numa nodes
+ * are traversed.
+ */
+ if (numa_node == start_numa_node) {
+ node_count = 0;
+ cpumask_copy(avail_cpus, cpu_online_mask);
+ }
+ cpu_count = 0;
+ core_count = 0;
+ continue;
+ }
+ }
+ if (++core_count == cores)
+ core_count = 0;
+ }
+free_irq:
+ free_cpumask_var(filter_mask);
+ free_cpumask_var(avail_cpus);
+ if (core_id_list)
+ kfree(core_id_list);
+ return err;
+}
+
static int mana_gd_setup_irqs(struct pci_dev *pdev)
{
- unsigned int max_queues_per_port = num_online_cpus();
+ unsigned int max_queues_per_port;
struct gdma_context *gc = pci_get_drvdata(pdev);
struct gdma_irq_context *gic;
- unsigned int max_irqs, cpu;
- int nvec, irq;
+ unsigned int max_irqs;
+ int nvec, *irqs, irq;
int err, i = 0, j;
+ cpus_read_lock();
+ max_queues_per_port = num_online_cpus();
if (max_queues_per_port > MANA_MAX_NUM_QUEUES)
max_queues_per_port = MANA_MAX_NUM_QUEUES;
@@ -1261,6 +1366,11 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
nvec = pci_alloc_irq_vectors(pdev, 2, max_irqs, PCI_IRQ_MSIX);
if (nvec < 0)
return nvec;
+ irqs = kmalloc_array(nvec, sizeof(int), GFP_KERNEL);
+ if (!irqs) {
+ err = -ENOMEM;
+ goto free_irq_vector;
+ }
gc->irq_contexts = kcalloc(nvec, sizeof(struct gdma_irq_context),
GFP_KERNEL);
@@ -1281,27 +1391,27 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
snprintf(gic->name, MANA_IRQ_NAME_SZ, "mana_q%d@pci:%s",
i - 1, pci_name(pdev));
- irq = pci_irq_vector(pdev, i);
- if (irq < 0) {
- err = irq;
+ irqs[i] = pci_irq_vector(pdev, i);
+ if (irqs[i] < 0) {
+ err = irqs[i];
goto free_irq;
}
- err = request_irq(irq, mana_gd_intr, 0, gic->name, gic);
+ err = request_irq(irqs[i], mana_gd_intr, 0, gic->name, gic);
if (err)
goto free_irq;
-
- cpu = cpumask_local_spread(i, gc->numa_node);
- irq_set_affinity_and_hint(irq, cpumask_of(cpu));
}
+ err = irq_setup(irqs, nvec, gc->numa_node);
+ if (err)
+ goto free_irq;
err = mana_gd_alloc_res_map(nvec, &gc->msix_resource);
if (err)
goto free_irq;
gc->max_num_msix = nvec;
gc->num_msix_usable = nvec;
-
+ cpus_read_unlock();
return 0;
free_irq:
@@ -1314,8 +1424,10 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
}
kfree(gc->irq_contexts);
+ kfree(irqs);
gc->irq_contexts = NULL;
free_irq_vector:
+ cpus_read_unlock();
pci_free_irq_vectors(pdev);
return err;
}
--
2.34.1
next reply other threads:[~2023-11-21 13:54 UTC|newest]
Thread overview: 16+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-11-21 13:54 Souradeep Chakrabarti [this message]
2023-11-21 17:37 ` [PATCH V2 net-next] net: mana: Assigning IRQ affinity on HT cores Haiyang Zhang
2023-11-21 18:51 ` Michael Kelley
2023-11-21 22:51 ` kernel test robot
2023-11-21 23:48 ` Jakub Kicinski
2023-11-27 9:36 ` [EXTERNAL] " Souradeep Chakrabarti
2023-11-27 14:32 ` Zhu Yanjun
2023-11-27 18:06 ` Jakub Kicinski
2023-11-29 22:17 ` Souradeep Chakrabarti
2023-11-30 0:02 ` Jakub Kicinski
2023-11-27 19:07 ` Florian Fainelli
2023-11-29 22:24 ` Souradeep Chakrabarti
2023-11-30 2:16 ` Yury Norov
2023-11-30 12:05 ` Souradeep Chakrabarti
2023-11-30 16:57 ` Yury Norov
2023-12-04 9:02 ` Souradeep Chakrabarti
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1700574877-6037-1-git-send-email-schakrabarti@linux.microsoft.com \
--to=schakrabarti@linux.microsoft.com \
--cc=cai.huoqing@linux.dev \
--cc=davem@davemloft.net \
--cc=decui@microsoft.com \
--cc=edumazet@google.com \
--cc=haiyangz@microsoft.com \
--cc=kuba@kernel.org \
--cc=kys@microsoft.com \
--cc=leon@kernel.org \
--cc=linux-hyperv@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-rdma@vger.kernel.org \
--cc=longli@microsoft.com \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=paulros@microsoft.com \
--cc=schakrabarti@microsoft.com \
--cc=sharmaajay@microsoft.com \
--cc=ssengar@linux.microsoft.com \
--cc=tglx@linutronix.de \
--cc=vkuznets@redhat.com \
--cc=wei.liu@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).