liquidio: optimize DMA in NUMA systems
Optimize DMA in NUMA systems by allocating memory from NUMA node that NIC
is plugged in to; DMA will no longer cross NUMA nodes. If NIC IRQs are
pinned to a local CPU, that CPU's access to the DMA'd data is also
optimized.
Signed-off-by: VSR Burru <veerasenareddy.burru@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/ethernet/cavium/liquidio/lio_main.c b/drivers/net/ethernet/cavium/liquidio/lio_main.c
index dffed43..acfd848 100644
--- a/drivers/net/ethernet/cavium/liquidio/lio_main.c
+++ b/drivers/net/ethernet/cavium/liquidio/lio_main.c
@@ -782,7 +782,7 @@ static int setup_glists(struct octeon_device *oct, struct lio *lio, int num_iqs)
}
for (i = 0; i < num_iqs; i++) {
- int numa_node = cpu_to_node(i % num_online_cpus());
+ int numa_node = dev_to_node(&oct->pci_dev->dev);
spin_lock_init(&lio->glist_lock[i]);