2019-04-17 07:45:39

by Jose Abreu

[permalink] [raw]
Subject: [PATCH net-next 0/2] net: stmmac: Enable Flow Control

I don't know of any specific reason why Flow Control is off by default but
do let me know if there is any.

Tested in B2B between XGMAC2 and GMAC5.

Cc: Joao Pinto <[email protected]>
Cc: David S. Miller <[email protected]>
Cc: Giuseppe Cavallaro <[email protected]>
Cc: Alexandre Torgue <[email protected]>

Jose Abreu (2):
net: stmmac: dwxgmac: Finish the Flow Control implementation
net: stmmac: Set Flow Control to automatic mode in the driver

drivers/net/ethernet/stmicro/stmmac/dwxgmac2.h | 5 +++
drivers/net/ethernet/stmicro/stmmac/dwxgmac2_dma.c | 46 ++++++++++++++++++++++
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 2 +-
3 files changed, 52 insertions(+), 1 deletion(-)

--
2.7.4


2019-04-17 07:45:02

by Jose Abreu

[permalink] [raw]
Subject: [PATCH net-next 2/2] net: stmmac: Set Flow Control to automatic mode in the driver

By default Flow Control feature is not being enabled in stmmac.

This is a useful feature that can prevent loss of packets and now that
XGMAC already supports it (along with GMAC and QoS) it makes sense to
activate it.

Switch the module parameter to FLOW_AUTO so that Flow Control is
activated.

Signed-off-by: Jose Abreu <[email protected]>
Cc: Joao Pinto <[email protected]>
Cc: David S. Miller <[email protected]>
Cc: Giuseppe Cavallaro <[email protected]>
Cc: Alexandre Torgue <[email protected]>
---
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index a26e36dbb5df..7a895a2889e3 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -74,7 +74,7 @@ MODULE_PARM_DESC(phyaddr, "Physical device address");
#define STMMAC_TX_THRESH (DMA_TX_SIZE / 4)
#define STMMAC_RX_THRESH (DMA_RX_SIZE / 4)

-static int flow_ctrl = FLOW_OFF;
+static int flow_ctrl = FLOW_AUTO;
module_param(flow_ctrl, int, 0644);
MODULE_PARM_DESC(flow_ctrl, "Flow control ability [on/off]");

--
2.7.4

2019-04-17 07:45:09

by Jose Abreu

[permalink] [raw]
Subject: [PATCH net-next 1/2] net: stmmac: dwxgmac: Finish the Flow Control implementation

Finish the implementation of Flow Control feature. In order for it to
work correctly we need to set EHFC bit and the correct threshold values
for activating and deactivating it.

Signed-off-by: Jose Abreu <[email protected]>
Cc: Joao Pinto <[email protected]>
Cc: David S. Miller <[email protected]>
Cc: Giuseppe Cavallaro <[email protected]>
Cc: Alexandre Torgue <[email protected]>
---
drivers/net/ethernet/stmicro/stmmac/dwxgmac2.h | 5 +++
drivers/net/ethernet/stmicro/stmmac/dwxgmac2_dma.c | 46 ++++++++++++++++++++++
2 files changed, 51 insertions(+)

diff --git a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2.h b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2.h
index 37d5e6fe7473..085b700a4994 100644
--- a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2.h
+++ b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2.h
@@ -143,6 +143,11 @@
#define XGMAC_RSF BIT(5)
#define XGMAC_RTC GENMASK(1, 0)
#define XGMAC_RTC_SHIFT 0
+#define XGMAC_MTL_RXQ_FLOW_CONTROL(x) (0x00001150 + (0x80 * (x)))
+#define XGMAC_RFD GENMASK(31, 17)
+#define XGMAC_RFD_SHIFT 17
+#define XGMAC_RFA GENMASK(15, 1)
+#define XGMAC_RFA_SHIFT 1
#define XGMAC_MTL_QINTEN(x) (0x00001170 + (0x80 * (x)))
#define XGMAC_RXOIE BIT(16)
#define XGMAC_MTL_QINT_STATUS(x) (0x00001174 + (0x80 * (x)))
diff --git a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_dma.c b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_dma.c
index 2ba712b48a89..e79037f511e1 100644
--- a/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_dma.c
+++ b/drivers/net/ethernet/stmicro/stmmac/dwxgmac2_dma.c
@@ -147,6 +147,52 @@ static void dwxgmac2_dma_rx_mode(void __iomem *ioaddr, int mode,
value &= ~XGMAC_RQS;
value |= (rqs << XGMAC_RQS_SHIFT) & XGMAC_RQS;

+ if ((fifosz >= 4096) && (qmode != MTL_QUEUE_AVB)) {
+ u32 flow = readl(ioaddr + XGMAC_MTL_RXQ_FLOW_CONTROL(channel));
+ unsigned int rfd, rfa;
+
+ value |= XGMAC_EHFC;
+
+ /* Set Threshold for Activating Flow Control to min 2 frames,
+ * i.e. 1500 * 2 = 3000 bytes.
+ *
+ * Set Threshold for Deactivating Flow Control to min 1 frame,
+ * i.e. 1500 bytes.
+ */
+ switch (fifosz) {
+ case 4096:
+ /* This violates the above formula because of FIFO size
+ * limit therefore overflow may occur in spite of this.
+ */
+ rfd = 0x03; /* Full-2.5K */
+ rfa = 0x01; /* Full-1.5K */
+ break;
+
+ case 8192:
+ rfd = 0x06; /* Full-4K */
+ rfa = 0x0a; /* Full-6K */
+ break;
+
+ case 16384:
+ rfd = 0x06; /* Full-4K */
+ rfa = 0x12; /* Full-10K */
+ break;
+
+ default:
+ rfd = 0x06; /* Full-4K */
+ rfa = 0x1e; /* Full-16K */
+ break;
+ }
+
+ flow &= ~XGMAC_RFD;
+ flow |= rfd << XGMAC_RFD_SHIFT;
+
+ flow &= ~XGMAC_RFA;
+ flow |= rfa << XGMAC_RFA_SHIFT;
+
+ writel(flow, ioaddr + XGMAC_MTL_RXQ_FLOW_CONTROL(channel));
+ }
+
writel(value, ioaddr + XGMAC_MTL_RXQ_OPMODE(channel));

/* Enable MTL RX overflow */
--
2.7.4

2019-04-17 17:17:24

by David Miller

[permalink] [raw]
Subject: Re: [PATCH net-next 0/2] net: stmmac: Enable Flow Control

From: Jose Abreu <[email protected]>
Date: Wed, 17 Apr 2019 09:33:03 +0200

> I don't know of any specific reason why Flow Control is off by default but
> do let me know if there is any.
>
> Tested in B2B between XGMAC2 and GMAC5.
>
> Cc: Joao Pinto <[email protected]>
> Cc: David S. Miller <[email protected]>
> Cc: Giuseppe Cavallaro <[email protected]>
> Cc: Alexandre Torgue <[email protected]>

I'm just going to apply this series now, it's easy to revert if necessary.

Thanks.

2019-04-17 18:02:13

by Aaro Koskinen

[permalink] [raw]
Subject: Re: [PATCH net-next 0/2] net: stmmac: Enable Flow Control

Hi,

On Wed, Apr 17, 2019 at 09:33:03AM +0200, Jose Abreu wrote:
> I don't know of any specific reason why Flow Control is off by default but
> do let me know if there is any.

It seems dwmac1000 assumes 4K FIFO size when configuring threshold
values. This could be non-optimal for 8K/16K FIFOs, but that can be
fixed later on with incremental patches.

A.