sfc: Set default parallelism to per-core by default
The previous default of per-package can be more CPU-efficient, but
users generally seem to prefer per-core. It should also allow
accelerated RFS to direct packets more precisely, if IRQ affinity
is properly spread out.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
diff --git a/drivers/net/ethernet/sfc/efx.c b/drivers/net/ethernet/sfc/efx.c
index 5fcc42f..d7301d2 100644
--- a/drivers/net/ethernet/sfc/efx.c
+++ b/drivers/net/ethernet/sfc/efx.c
@@ -162,7 +162,7 @@
* interrupt handling.
*
* Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
- * The default (0) means to assign an interrupt to each package (level II cache)
+ * The default (0) means to assign an interrupt to each core.
*/
static unsigned int rss_cpus;
module_param(rss_cpus, uint, 0444);
@@ -1148,14 +1148,14 @@
static int efx_wanted_parallelism(void)
{
- cpumask_var_t core_mask;
+ cpumask_var_t thread_mask;
int count;
int cpu;
if (rss_cpus)
return rss_cpus;
- if (unlikely(!zalloc_cpumask_var(&core_mask, GFP_KERNEL))) {
+ if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
printk(KERN_WARNING
"sfc: RSS disabled due to allocation failure\n");
return 1;
@@ -1163,14 +1163,14 @@
count = 0;
for_each_online_cpu(cpu) {
- if (!cpumask_test_cpu(cpu, core_mask)) {
+ if (!cpumask_test_cpu(cpu, thread_mask)) {
++count;
- cpumask_or(core_mask, core_mask,
- topology_core_cpumask(cpu));
+ cpumask_or(thread_mask, thread_mask,
+ topology_thread_cpumask(cpu));
}
}
- free_cpumask_var(core_mask);
+ free_cpumask_var(thread_mask);
return count;
}