On 8/1/2024 11:32 AM, Serge Semin wrote:

Hi Sagar

On Mon, Jul 08, 2024 at 02:30:01PM -0700, Sagar Cheluvegowda wrote:

quoted

Add interconnect support to vote for bus bandwidth based
on the current speed of the driver.
Adds support for two different paths - one from ethernet to
DDR and the other from CPU to ethernet, Vote from each
interconnect client is aggregated and the on-chip interconnect
hardware is configured to the most appropriate bandwidth profile.

Suggested-by: Andrew Halaney <redacted>
Signed-off-by: Sagar Cheluvegowda <redacted>
---
 drivers/net/ethernet/stmicro/stmmac/stmmac.h          |  1 +
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c     |  8 ++++++++
 drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c | 12 ++++++++++++
 include/linux/stmmac.h                                |  2 ++
 4 files changed, 23 insertions(+)

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac.h b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
index b23b920eedb1..56a282d2b8cd 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac.h
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac.h

@@ -21,6 +21,7 @@
 #include <linux/ptp_clock_kernel.h>
 #include <linux/net_tstamp.h>
 #include <linux/reset.h>
+#include <linux/interconnect.h>
 #include <net/page_pool/types.h>
 #include <net/xdp.h>
 #include <uapi/linux/bpf.h>

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index b3afc7cb7d72..ec7c61ee44d4 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c

@@ -985,6 +985,12 @@ static void stmmac_fpe_link_state_handle(struct stmmac_priv *priv, bool is_up)
 	}
 }
 
+static void stmmac_set_icc_bw(struct stmmac_priv *priv, unsigned int speed)
+{

quoted

+	icc_set_bw(priv->plat->axi_icc_path, Mbps_to_icc(speed), Mbps_to_icc(speed));
+	icc_set_bw(priv->plat->ahb_icc_path, Mbps_to_icc(speed), Mbps_to_icc(speed));

I've got two questions in this regard:

1. Don't we need to call icc_enable()/icc_disable() in someplace in
the driver? For instance the CPU-MEM path must be enabled before even
the stmmac_dvr_probe() is called, otherwise the CSR won't be
accessible. Right? For the same reason the CPU-MEM bandwidth should be
set in sync with that.

2. Why is the CPU-MAC speed is specified to match the Ethernet link
speed? It doesn't seem reasonable. It's the CSR's access speed and
should be done as fast as possible. Shouldn't it?

quoted

+}

I am having internal discussions with clocks team, I will revert back soon with answers.

quoted

+
 static void stmmac_mac_link_down(struct phylink_config *config,
 				 unsigned int mode, phy_interface_t interface)
 {

@@ -1080,6 +1086,8 @@ static void stmmac_mac_link_up(struct phylink_config *config,
 	if (priv->plat->fix_mac_speed)
 		priv->plat->fix_mac_speed(priv->plat->bsp_priv, speed, mode);
 
+	stmmac_set_icc_bw(priv, speed);
+
 	if (!duplex)
 		ctrl &= ~priv->hw->link.duplex;
 	else

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
index 54797edc9b38..201f9dea6da9 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c

@@ -642,6 +642,18 @@ stmmac_probe_config_dt(struct platform_device *pdev, u8 *mac)
 		dev_dbg(&pdev->dev, "PTP rate %d\n", plat->clk_ptp_rate);
 	}
 
+	plat->axi_icc_path = devm_of_icc_get(&pdev->dev, "mac-mem");
+	if (IS_ERR(plat->axi_icc_path)) {
+		ret = ERR_CAST(plat->axi_icc_path);
+		goto error_hw_init;
+	}
+
+	plat->ahb_icc_path = devm_of_icc_get(&pdev->dev, "cpu-mac");
+	if (IS_ERR(plat->ahb_icc_path)) {
+		ret = ERR_CAST(plat->ahb_icc_path);
+		goto error_hw_init;
+	}
+
 	plat->stmmac_rst = devm_reset_control_get_optional(&pdev->dev,
 							   STMMAC_RESOURCE_NAME);
 	if (IS_ERR(plat->stmmac_rst)) {

diff --git a/include/linux/stmmac.h b/include/linux/stmmac.h
index f92c195c76ed..385f352a0c23 100644
--- a/include/linux/stmmac.h
+++ b/include/linux/stmmac.h

@@ -283,6 +283,8 @@ struct plat_stmmacenet_data {
 	struct reset_control *stmmac_rst;
 	struct reset_control *stmmac_ahb_rst;
 	struct stmmac_axi *axi;

quoted

+	struct icc_path *axi_icc_path;

The MAC<->MEM interface isn't always AXI (it can be AHB or custom) and

quoted

+	struct icc_path *ahb_icc_path;

the CPU<->MAC isn't always AHB (it can also be APB, AXI, custom). So
the more generic naming would be:

axi_icc_path -> dma_icc_path
and
ahb_icc_path -> csr_icc_path

-Serge(y)

quoted

 	int has_gmac4;
 	int rss_en;
 	int mac_port_sel_speed;

-- 
2.34.1