| Message ID | 20210421203409.40717-1-ezequiel@collabora.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | net: stmmac: RK3566 | expand |
Hi Ezequiel, 在 2021/4/22 上午4:34, Ezequiel Garcia 写道: > Now that RK3568 SoC devicetree upstreaming is happening [1], > here's another round for the RK3566 dwmac. There wasn't any clear > consensus on how to implement the two interfaces present > on RK3568, so I decided to drop that and just submit RK3566 for now. > > This has been tested on a Pine64 RK3566 Quartz64 Model B board, > DHCP and iperf are looking good. > > For all the people testing, here's Quartz 64 Model B device tree > snippet: > > gmac1: ethernet@fe010000 { > compatible = "rockchip,rk3566-gmac", "snps,dwmac-4.20a"; It is better to use "rockchip,rk3568-gmac" here, "rockchip,rk3566-gmac" is not compatible, 3568 has two gmacs, which are compatible with 3566. If there is no better way, using bus_id from alias is good, it is a fixed id, and U-Boot also use the id to write MAC address into kernel DTB. plat->bus_id = of_alias_get_id(np, "ethernet"); > reg = <0x0 0xfe010000 0x0 0x10000>; > interrupts = <GIC_SPI 32 IRQ_TYPE_LEVEL_HIGH>, > <GIC_SPI 29 IRQ_TYPE_LEVEL_HIGH>; > interrupt-names = "macirq", "eth_wake_irq"; > rockchip,grf = <&grf>; > clocks = <&cru SCLK_GMAC1>, <&cru SCLK_GMAC1_RX_TX>, > <&cru SCLK_GMAC1_RX_TX>, <&cru CLK_MAC1_REFOUT>, > <&cru ACLK_GMAC1>, <&cru PCLK_GMAC1>, > <&cru SCLK_GMAC1_RX_TX>, <&cru CLK_GMAC1_PTP_REF>; > clock-names = "stmmaceth", "mac_clk_rx", > "mac_clk_tx", "clk_mac_refout", > "aclk_mac", "pclk_mac", > "clk_mac_speed", "ptp_ref"; > resets = <&cru SRST_A_GMAC1>; > reset-names = "stmmaceth"; > > snps,mixed-burst; > snps,tso; > > snps,axi-config = <&gmac1_stmmac_axi_setup>; > snps,mtl-rx-config = <&gmac1_mtl_rx_setup>; > snps,mtl-tx-config = <&gmac1_mtl_tx_setup>; > status = "disabled"; > > mdio1: mdio { > compatible = "snps,dwmac-mdio"; > #address-cells = <0x1>; > #size-cells = <0x0>; > }; > > gmac1_stmmac_axi_setup: stmmac-axi-config { > snps,wr_osr_lmt = <4>; > snps,rd_osr_lmt = <8>; > snps,blen = <0 0 0 0 16 8 4>; > }; > > gmac1_mtl_rx_setup: rx-queues-config { > snps,rx-queues-to-use = <1>; > queue0 {}; > }; > > gmac1_mtl_tx_setup: tx-queues-config { > snps,tx-queues-to-use = <1>; > queue0 {}; > }; > }; > > While here, I'm adding a small patch from David Wu, for some > sanity checks for dwmac-rockchip-specific non-NULL ops. > > Thanks! > > [1] http://lore.kernel.org/r/20210421065921.23917-1-cl@rock-chips.com > > David Wu (2): > net: stmmac: dwmac-rk: Check platform-specific ops > net: stmmac: Add RK3566 SoC support > > Ezequiel Garcia (1): > net: stmmac: Don't set has_gmac if has_gmac4 is set > > .../bindings/net/rockchip-dwmac.txt | 1 + > .../net/ethernet/stmicro/stmmac/dwmac-rk.c | 126 +++++++++++++++++- > 2 files changed, 124 insertions(+), 3 deletions(-) >
Hi David, Thanks a lot for reviewing. On Fri, 2021-04-23 at 17:24 +0800, David Wu wrote: > Hi Ezequiel, > > 在 2021/4/22 上午4:34, Ezequiel Garcia 写道: > > Now that RK3568 SoC devicetree upstreaming is happening [1], > > here's another round for the RK3566 dwmac. There wasn't any clear > > consensus on how to implement the two interfaces present > > on RK3568, so I decided to drop that and just submit RK3566 for now. > > > > This has been tested on a Pine64 RK3566 Quartz64 Model B board, > > DHCP and iperf are looking good. > > > > For all the people testing, here's Quartz 64 Model B device tree > > snippet: > > > > gmac1: ethernet@fe010000 { > > compatible = "rockchip,rk3566-gmac", "snps,dwmac-4.20a"; > > It is better to use "rockchip,rk3568-gmac" here, "rockchip,rk3566-gmac" > is not compatible, 3568 has two gmacs, which are compatible with 3566. > > If there is no better way, using bus_id from alias is good, it is a > fixed id, and U-Boot also use the id to write MAC address into kernel DTB. > > plat->bus_id = of_alias_get_id(np, "ethernet"); > This was discussed, see ChenYu's reply. I think the idea is considered fragile: https://lore.kernel.org/netdev/CAGb2v67ZBR=XDFPeXQc429HNu_dbY__-KN50tvBW44fXMs78_w@mail.gmail.com/ However, I agree with you about going back to just "rockchip,rk3568-gmac". Adding rockchip,rk3566-gmac maybe wasn't a great idea :-) The most accepted way forward seems Heiko's proposal of hardcoding the mmio addresses to identify each block, as it's done in the DSI driver: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/rockchip/dw-mipi-dsi-rockchip.c#n1170 Would you agree with it? Thanks! Ezequiel > > reg = <0x0 0xfe010000 0x0 0x10000>; > > interrupts = <GIC_SPI 32 IRQ_TYPE_LEVEL_HIGH>, > > <GIC_SPI 29 IRQ_TYPE_LEVEL_HIGH>; > > interrupt-names = "macirq", "eth_wake_irq"; > > rockchip,grf = <&grf>; > > clocks = <&cru SCLK_GMAC1>, <&cru SCLK_GMAC1_RX_TX>, > > <&cru SCLK_GMAC1_RX_TX>, <&cru CLK_MAC1_REFOUT>, > > <&cru ACLK_GMAC1>, <&cru PCLK_GMAC1>, > > <&cru SCLK_GMAC1_RX_TX>, <&cru CLK_GMAC1_PTP_REF>; > > clock-names = "stmmaceth", "mac_clk_rx", > > "mac_clk_tx", "clk_mac_refout", > > "aclk_mac", "pclk_mac", > > "clk_mac_speed", "ptp_ref"; > > resets = <&cru SRST_A_GMAC1>; > > reset-names = "stmmaceth"; > > > > snps,mixed-burst; > > snps,tso; > > > > snps,axi-config = <&gmac1_stmmac_axi_setup>; > > snps,mtl-rx-config = <&gmac1_mtl_rx_setup>; > > snps,mtl-tx-config = <&gmac1_mtl_tx_setup>; > > status = "disabled"; > > > > mdio1: mdio { > > compatible = "snps,dwmac-mdio"; > > #address-cells = <0x1>; > > #size-cells = <0x0>; > > }; > > > > gmac1_stmmac_axi_setup: stmmac-axi-config { > > snps,wr_osr_lmt = <4>; > > snps,rd_osr_lmt = <8>; > > snps,blen = <0 0 0 0 16 8 4>; > > }; > > > > gmac1_mtl_rx_setup: rx-queues-config { > > snps,rx-queues-to-use = <1>; > > queue0 {}; > > }; > > > > gmac1_mtl_tx_setup: tx-queues-config { > > snps,tx-queues-to-use = <1>; > > queue0 {}; > > }; > > }; > > > > While here, I'm adding a small patch from David Wu, for some > > sanity checks for dwmac-rockchip-specific non-NULL ops. > > > > Thanks! > > > > [1] http://lore.kernel.org/r/20210421065921.23917-1-cl@rock-chips.com > > > > David Wu (2): > > net: stmmac: dwmac-rk: Check platform-specific ops > > net: stmmac: Add RK3566 SoC support > > > > Ezequiel Garcia (1): > > net: stmmac: Don't set has_gmac if has_gmac4 is set > > > > .../bindings/net/rockchip-dwmac.txt | 1 + > > .../net/ethernet/stmicro/stmmac/dwmac-rk.c | 126 +++++++++++++++++- > > 2 files changed, 124 insertions(+), 3 deletions(-) > > > >
Hi Ezequiel, 在 2021/4/23 下午9:57, Ezequiel Garcia 写道: > Hi David, > > Thanks a lot for reviewing. > > On Fri, 2021-04-23 at 17:24 +0800, David Wu wrote: >> Hi Ezequiel, >> >> 在 2021/4/22 上午4:34, Ezequiel Garcia 写道: >>> Now that RK3568 SoC devicetree upstreaming is happening [1], >>> here's another round for the RK3566 dwmac. There wasn't any clear >>> consensus on how to implement the two interfaces present >>> on RK3568, so I decided to drop that and just submit RK3566 for now. >>> >>> This has been tested on a Pine64 RK3566 Quartz64 Model B board, >>> DHCP and iperf are looking good. >>> >>> For all the people testing, here's Quartz 64 Model B device tree >>> snippet: >>> >>> gmac1: ethernet@fe010000 { >>> compatible = "rockchip,rk3566-gmac", "snps,dwmac-4.20a"; >> >> It is better to use "rockchip,rk3568-gmac" here, "rockchip,rk3566-gmac" >> is not compatible, 3568 has two gmacs, which are compatible with 3566. >> >> If there is no better way, using bus_id from alias is good, it is a >> fixed id, and U-Boot also use the id to write MAC address into kernel DTB. >> >> plat->bus_id = of_alias_get_id(np, "ethernet"); >> > > This was discussed, see ChenYu's reply. I think the idea > is considered fragile: > > https://lore.kernel.org/netdev/CAGb2v67ZBR=XDFPeXQc429HNu_dbY__-KN50tvBW44fXMs78_w@mail.gmail.com/ > > However, I agree with you about going back to just "rockchip,rk3568-gmac". > Adding rockchip,rk3566-gmac maybe wasn't a great idea :-) > > The most accepted way forward seems Heiko's proposal of hardcoding the mmio > addresses to identify each block, as it's done in the DSI driver: > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/rockchip/dw-mipi-dsi-rockchip.c#n1170 > > Would you agree with it? > Yes, I agree. I also think the alias should be rarely changed. The DSI mmio addresses example here can be used for reference, and it seems that it will be stronger than alias. > Thanks! > Ezequiel > >>> reg = <0x0 0xfe010000 0x0 0x10000>; >>> interrupts = <GIC_SPI 32 IRQ_TYPE_LEVEL_HIGH>, >>> <GIC_SPI 29 IRQ_TYPE_LEVEL_HIGH>; >>> interrupt-names = "macirq", "eth_wake_irq"; >>> rockchip,grf = <&grf>; >>> clocks = <&cru SCLK_GMAC1>, <&cru SCLK_GMAC1_RX_TX>, >>> <&cru SCLK_GMAC1_RX_TX>, <&cru CLK_MAC1_REFOUT>, >>> <&cru ACLK_GMAC1>, <&cru PCLK_GMAC1>, >>> <&cru SCLK_GMAC1_RX_TX>, <&cru CLK_GMAC1_PTP_REF>; >>> clock-names = "stmmaceth", "mac_clk_rx", >>> "mac_clk_tx", "clk_mac_refout", >>> "aclk_mac", "pclk_mac", >>> "clk_mac_speed", "ptp_ref"; >>> resets = <&cru SRST_A_GMAC1>; >>> reset-names = "stmmaceth"; >>> >>> snps,mixed-burst; >>> snps,tso; >>> >>> snps,axi-config = <&gmac1_stmmac_axi_setup>; >>> snps,mtl-rx-config = <&gmac1_mtl_rx_setup>; >>> snps,mtl-tx-config = <&gmac1_mtl_tx_setup>; >>> status = "disabled"; >>> >>> mdio1: mdio { >>> compatible = "snps,dwmac-mdio"; >>> #address-cells = <0x1>; >>> #size-cells = <0x0>; >>> }; >>> >>> gmac1_stmmac_axi_setup: stmmac-axi-config { >>> snps,wr_osr_lmt = <4>; >>> snps,rd_osr_lmt = <8>; >>> snps,blen = <0 0 0 0 16 8 4>; >>> }; >>> >>> gmac1_mtl_rx_setup: rx-queues-config { >>> snps,rx-queues-to-use = <1>; >>> queue0 {}; >>> }; >>> >>> gmac1_mtl_tx_setup: tx-queues-config { >>> snps,tx-queues-to-use = <1>; >>> queue0 {}; >>> }; >>> }; >>> >>> While here, I'm adding a small patch from David Wu, for some >>> sanity checks for dwmac-rockchip-specific non-NULL ops. >>> >>> Thanks! >>> >>> [1] http://lore.kernel.org/r/20210421065921.23917-1-cl@rock-chips.com >>> >>> David Wu (2): >>> net: stmmac: dwmac-rk: Check platform-specific ops >>> net: stmmac: Add RK3566 SoC support >>> >>> Ezequiel Garcia (1): >>> net: stmmac: Don't set has_gmac if has_gmac4 is set >>> >>> .../bindings/net/rockchip-dwmac.txt | 1 + >>> .../net/ethernet/stmicro/stmmac/dwmac-rk.c | 126 +++++++++++++++++- >>> 2 files changed, 124 insertions(+), 3 deletions(-) >>> >> >> > > > > >
Now that RK3568 SoC devicetree upstreaming is happening [1], here's another round for the RK3566 dwmac. There wasn't any clear consensus on how to implement the two interfaces present on RK3568, so I decided to drop that and just submit RK3566 for now. This has been tested on a Pine64 RK3566 Quartz64 Model B board, DHCP and iperf are looking good. For all the people testing, here's Quartz 64 Model B device tree snippet: gmac1: ethernet@fe010000 { compatible = "rockchip,rk3566-gmac", "snps,dwmac-4.20a"; reg = <0x0 0xfe010000 0x0 0x10000>; interrupts = <GIC_SPI 32 IRQ_TYPE_LEVEL_HIGH>, <GIC_SPI 29 IRQ_TYPE_LEVEL_HIGH>; interrupt-names = "macirq", "eth_wake_irq"; rockchip,grf = <&grf>; clocks = <&cru SCLK_GMAC1>, <&cru SCLK_GMAC1_RX_TX>, <&cru SCLK_GMAC1_RX_TX>, <&cru CLK_MAC1_REFOUT>, <&cru ACLK_GMAC1>, <&cru PCLK_GMAC1>, <&cru SCLK_GMAC1_RX_TX>, <&cru CLK_GMAC1_PTP_REF>; clock-names = "stmmaceth", "mac_clk_rx", "mac_clk_tx", "clk_mac_refout", "aclk_mac", "pclk_mac", "clk_mac_speed", "ptp_ref"; resets = <&cru SRST_A_GMAC1>; reset-names = "stmmaceth"; snps,mixed-burst; snps,tso; snps,axi-config = <&gmac1_stmmac_axi_setup>; snps,mtl-rx-config = <&gmac1_mtl_rx_setup>; snps,mtl-tx-config = <&gmac1_mtl_tx_setup>; status = "disabled"; mdio1: mdio { compatible = "snps,dwmac-mdio"; #address-cells = <0x1>; #size-cells = <0x0>; }; gmac1_stmmac_axi_setup: stmmac-axi-config { snps,wr_osr_lmt = <4>; snps,rd_osr_lmt = <8>; snps,blen = <0 0 0 0 16 8 4>; }; gmac1_mtl_rx_setup: rx-queues-config { snps,rx-queues-to-use = <1>; queue0 {}; }; gmac1_mtl_tx_setup: tx-queues-config { snps,tx-queues-to-use = <1>; queue0 {}; }; }; While here, I'm adding a small patch from David Wu, for some sanity checks for dwmac-rockchip-specific non-NULL ops. Thanks! [1] http://lore.kernel.org/r/20210421065921.23917-1-cl@rock-chips.com David Wu (2): net: stmmac: dwmac-rk: Check platform-specific ops net: stmmac: Add RK3566 SoC support Ezequiel Garcia (1): net: stmmac: Don't set has_gmac if has_gmac4 is set .../bindings/net/rockchip-dwmac.txt | 1 + .../net/ethernet/stmicro/stmmac/dwmac-rk.c | 126 +++++++++++++++++- 2 files changed, 124 insertions(+), 3 deletions(-)