2018-09-24 09:12:15

by Maxime Chevallier

[permalink] [raw]
Subject: [PATCH net-next 0/2] net: mvpp2: Add txq to CPU mapping

Hi everyone,

This short series adds XPS support to the mvpp2 driver, by mapping
txqs and CPUs. This comes with a patch using round-robin scheduling
for the HW to pick the next txq to transmit from, instead of the default
fixed-priority scheduling.

Maxime Chevallier (2):
net: mvpp2: support XPS by mapping TX queues to CPUs
net: mvpp2: use round-robin scheduling for TX queues on the same CPU

drivers/net/ethernet/marvell/mvpp2/mvpp2.h | 1 +
drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c | 9 ++++++++-
2 files changed, 9 insertions(+), 1 deletion(-)

--
2.11.0



2018-09-24 09:11:58

by Maxime Chevallier

[permalink] [raw]
Subject: [PATCH net-next 2/2] net: mvpp2: use round-robin scheduling for TX queues on the same CPU

This commit allows each TXQ to be picked in a round-robin fashion by
the PPv2 transmit scheduling mechanism. This is opposed to the default
behaviour that prioritizes the highest numbered queues.

Suggested-by: Yan Markman <[email protected]>
Signed-off-by: Maxime Chevallier <[email protected]>
---
drivers/net/ethernet/marvell/mvpp2/mvpp2.h | 1 +
drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c | 3 +++
2 files changed, 4 insertions(+)

diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
index f5dceef60b0e..176c6b56fdcc 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
@@ -331,6 +331,7 @@
#define MVPP2_TXP_SCHED_ENQ_MASK 0xff
#define MVPP2_TXP_SCHED_DISQ_OFFSET 8
#define MVPP2_TXP_SCHED_CMD_1_REG 0x8010
+#define MVPP2_TXP_SCHED_FIXED_PRIO_REG 0x8014
#define MVPP2_TXP_SCHED_PERIOD_REG 0x8018
#define MVPP2_TXP_SCHED_MTU_REG 0x801c
#define MVPP2_TXP_MTU_MAX 0x7FFFF
diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
index bdacb9577216..c2ed71788e4f 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
@@ -1448,6 +1448,9 @@ static void mvpp2_defaults_set(struct mvpp2_port *port)
tx_port_num);
mvpp2_write(port->priv, MVPP2_TXP_SCHED_CMD_1_REG, 0);

+ /* Set TXQ scheduling to Round-Robin */
+ mvpp2_write(port->priv, MVPP2_TXP_SCHED_FIXED_PRIO_REG, 0);
+
/* Close bandwidth for all queues */
for (queue = 0; queue < MVPP2_MAX_TXQ; queue++) {
ptxq = mvpp2_txq_phys(port->id, queue);
--
2.11.0


2018-09-24 09:12:04

by Maxime Chevallier

[permalink] [raw]
Subject: [PATCH net-next 1/2] net: mvpp2: support XPS by mapping TX queues to CPUs

Since the PPv2 controller has multiple TX queues, we can spread traffic
by assining TX queues to CPUs, allowing to use XPS to balance egress
traffic between CPUs.

Suggested-by : Yan Markman <[email protected]>
Signed-off-by: Maxime Chevallier <[email protected]>
---
drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
index d30ccc515bb7..bdacb9577216 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
@@ -2423,13 +2423,17 @@ static int mvpp2_setup_rxqs(struct mvpp2_port *port)
static int mvpp2_setup_txqs(struct mvpp2_port *port)
{
struct mvpp2_tx_queue *txq;
- int queue, err;
+ int queue, err, cpu;

for (queue = 0; queue < port->ntxqs; queue++) {
txq = port->txqs[queue];
err = mvpp2_txq_init(port, txq);
if (err)
goto err_cleanup;
+
+ /* Assign this queue to a CPU */
+ cpu = queue % num_present_cpus();
+ netif_set_xps_queue(port->dev, cpumask_of(cpu), queue);
}

if (port->has_tx_irqs) {
--
2.11.0


2018-09-24 17:01:53

by David Miller

[permalink] [raw]
Subject: Re: [PATCH net-next 0/2] net: mvpp2: Add txq to CPU mapping

From: Maxime Chevallier <[email protected]>
Date: Mon, 24 Sep 2018 11:11:04 +0200

> This short series adds XPS support to the mvpp2 driver, by mapping
> txqs and CPUs. This comes with a patch using round-robin scheduling
> for the HW to pick the next txq to transmit from, instead of the default
> fixed-priority scheduling.

Series applied, thanks Maxime.