| Message ID | 201312082352.45265.sergei.shtylyov@cogentembedded.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Sergei, On Sun, Dec 8, 2013 at 9:52 PM, Sergei Shtylyov <sergei.shtylyov@cogentembedded.com> wrote: > Register Ether platform device and pin data on the Koelsch board. > Register platform fixup for Micrel KSZ8041 PHY, just like on the Lager board. > > Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com> > --- renesas.orig/arch/arm/mach-shmobile/board-koelsch.c > +++ renesas/arch/arm/mach-shmobile/board-koelsch.c > +/* Ether */ > +static const struct sh_eth_plat_data ether_pdata __initconst = { > + .phy = 0x1, > + .edmac_endian = EDMAC_LITTLE_ENDIAN, > + .phy_interface = PHY_INTERFACE_MODE_RMII, > + .ether_link_active_low = 1, > +}; With the above, NFS root is terrible slow, and ping indicates more than 10% packet loss. I was trying to debootstrap Debian jessie, but gave up after a few hours. Based on earlier discussions I found in the mailing list archives, I replaced the ".ether_link_active_low = 1" by ".no_ether_link = 1", and the problem went away. Debootstrap completed in about one hour. No more noticable packet loss during normal operation. I do still get "net eth0: Receive FIFO Overflow" once in a while, though (e.g. when running nuttcp). # nuttcp -r amd64box 112.8022 MB / 10.14 sec = 93.3131 Mbps 2 %TX 15 %RX 0 retrans 0.28 msRTT # nuttcp -t amd64box 94.1981 MB / 10.04 sec = 78.7000 Mbps 3 %TX 5 %RX 0 retrans With ".ether_link_active_low = 1", performance is halved: # nuttcp -r amd64box 56.0736 MB / 10.11 sec = 46.5290 Mbps 0 %TX 8 %RX 0 retrans 996.39 msRTT # nuttcp -t amd64box 47.7191 MB / 10.03 sec = 39.8991 Mbps 1 %TX 2 %RX 0 retrans BTW, I'm using renesas-devel-v3.13-rc2-20131205, with the following 4 patches from you applied on top: sh_eth: add R8A7791 support ARM: shmobile: r8a7791: add Ether clock ARM: shmobile: Koelsch: add Ether support ARM: shmobile: Koelsch: enable Ether in defconfig Anything I'm missing? Thanks! Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
On Mon, Dec 9, 2013 at 9:25 PM, Sergei Shtylyov <sergei.shtylyov@cogentembedded.com> wrote: >>> --- renesas.orig/arch/arm/mach-shmobile/board-koelsch.c >>> +++ renesas/arch/arm/mach-shmobile/board-koelsch.c >> >> >>> +/* Ether */ >>> +static const struct sh_eth_plat_data ether_pdata __initconst = { >>> + .phy = 0x1, >>> + .edmac_endian = EDMAC_LITTLE_ENDIAN, >>> + .phy_interface = PHY_INTERFACE_MODE_RMII, >>> + .ether_link_active_low = 1, >>> +}; > >> With the above, NFS root is terrible slow, > > Do you get NFS timeout(s) on booting? I do. It's a known issue. I only saw "NFS server not responding" once. > You can try my simplistic 'sh_eth' driver patch: > > http://marc.info/?l=linux-sh&m=137911743723032 Thanks, that also fixes the problem! Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
Hello. On 12/09/2013 09:52 PM, Geert Uytterhoeven wrote: >> Register Ether platform device and pin data on the Koelsch board. >> Register platform fixup for Micrel KSZ8041 PHY, just like on the Lager board. >> Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com> >> --- renesas.orig/arch/arm/mach-shmobile/board-koelsch.c >> +++ renesas/arch/arm/mach-shmobile/board-koelsch.c > >> +/* Ether */ >> +static const struct sh_eth_plat_data ether_pdata __initconst = { >> + .phy = 0x1, >> + .edmac_endian = EDMAC_LITTLE_ENDIAN, >> + .phy_interface = PHY_INTERFACE_MODE_RMII, >> + .ether_link_active_low = 1, >> +}; > With the above, NFS root is terrible slow, Do you get NFS timeout(s) on booting? I do. It's a known issue. > and ping indicates more than 10% packet loss. Hm, that I didn't see... > I was trying to debootstrap Debian jessie, but gave up > after a few hours. > Based on earlier discussions I found in the mailing list archives, I > replaced the > ".ether_link_active_low = 1" by ".no_ether_link = 1", and the problem went away. That's one way to deal with the issue, but it's not quite correct because it'll make Ether LEDs function in a way not matching to the intended one. I've added PHY platform fixup to force the LED mode to the intended one after PHY reset that the driver does. I've also submitted a somewhat naive patch to call the fixups after that PHY reset but it got refused by David Miller. That's why we have what we have now. Hopefully, this should get resolved in 3.14 if Florian Fainelli's patch series gets accepted: http://marc.info/?l=linux-netdev&m=138636384109141 > Debootstrap completed in about one hour. No more noticable packet loss > during normal operation. > I do still get "net eth0: Receive FIFO Overflow" once in a while, though (e.g. > when running nuttcp). > # nuttcp -r amd64box > 112.8022 MB / 10.14 sec = 93.3131 Mbps 2 %TX 15 %RX 0 retrans 0.28 msRTT > # nuttcp -t amd64box > 94.1981 MB / 10.04 sec = 78.7000 Mbps 3 %TX 5 %RX 0 retrans > With ".ether_link_active_low = 1", performance is halved: > # nuttcp -r amd64box > 56.0736 MB / 10.11 sec = 46.5290 Mbps 0 %TX 8 %RX 0 retrans 996.39 msRTT > # nuttcp -t amd64box > 47.7191 MB / 10.03 sec = 39.8991 Mbps 1 %TX 2 %RX 0 retrans > BTW, I'm using renesas-devel-v3.13-rc2-20131205, with the following 4 patches > from you applied on top: > sh_eth: add R8A7791 support > ARM: shmobile: r8a7791: add Ether clock > ARM: shmobile: Koelsch: add Ether support > ARM: shmobile: Koelsch: enable Ether in defconfig > Anything I'm missing? You can try my simplistic 'sh_eth' driver patch: http://marc.info/?l=linux-sh&m=137911743723032 > Thanks! > Gr{oetje,eeting}s, > Geert WBR, Sergei
On 12/09/2013 10:40 PM, Geert Uytterhoeven wrote: >>>> --- renesas.orig/arch/arm/mach-shmobile/board-koelsch.c >>>> +++ renesas/arch/arm/mach-shmobile/board-koelsch.c >>>> +/* Ether */ >>>> +static const struct sh_eth_plat_data ether_pdata __initconst = { >>>> + .phy = 0x1, >>>> + .edmac_endian = EDMAC_LITTLE_ENDIAN, >>>> + .phy_interface = PHY_INTERFACE_MODE_RMII, >>>> + .ether_link_active_low = 1, >>>> +}; >>> With the above, NFS root is terrible slow, >> Do you get NFS timeout(s) on booting? I do. It's a known issue. > I only saw "NFS server not responding" once. Yes, I meant that by NFS timeout. Sometimes I see 2 of them in a row. >> You can try my simplistic 'sh_eth' driver patch: >> http://marc.info/?l=linux-sh&m=137911743723032 > Thanks, that also fixes the problem! Let's hope Florian's patches get queued for 3.14. You can also test them meanwhile... > Gr{oetje,eeting}s, > > Geert WBR, Sergei
On Tue, Dec 10, 2013 at 12:10 AM, Sergei Shtylyov <sergei.shtylyov@cogentembedded.com> wrote: >>> You can try my simplistic 'sh_eth' driver patch: > >>> http://marc.info/?l=linux-sh&m=137911743723032 > >> Thanks, that also fixes the problem! > > Let's hope Florian's patches get queued for 3.14. You can also test them > meanwhile... I applied the whole series "net: phy: consolidate PHY reset", and sh_eth now works fine on Koelsch. Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
2013/12/10 Geert Uytterhoeven <geert@linux-m68k.org>: > On Tue, Dec 10, 2013 at 12:10 AM, Sergei Shtylyov > <sergei.shtylyov@cogentembedded.com> wrote: >>>> You can try my simplistic 'sh_eth' driver patch: >> >>>> http://marc.info/?l=linux-sh&m=137911743723032 >> >>> Thanks, that also fixes the problem! >> >> Let's hope Florian's patches get queued for 3.14. You can also test them >> meanwhile... > > I applied the whole series "net: phy: consolidate PHY reset", and sh_eth > now works fine on Koelsch. Great! Thanks Geert. -- Florian
Index: renesas/arch/arm/mach-shmobile/board-koelsch.c =================================================================== --- renesas.orig/arch/arm/mach-shmobile/board-koelsch.c +++ renesas/arch/arm/mach-shmobile/board-koelsch.c @@ -24,15 +24,31 @@ #include <linux/input.h> #include <linux/kernel.h> #include <linux/leds.h> +#include <linux/phy.h> #include <linux/pinctrl/machine.h> #include <linux/platform_data/gpio-rcar.h> #include <linux/platform_device.h> +#include <linux/sh_eth.h> #include <mach/common.h> +#include <mach/irqs.h> #include <mach/r8a7791.h> #include <mach/rcar-gen2.h> #include <asm/mach-types.h> #include <asm/mach/arch.h> +/* Ether */ +static const struct sh_eth_plat_data ether_pdata __initconst = { + .phy = 0x1, + .edmac_endian = EDMAC_LITTLE_ENDIAN, + .phy_interface = PHY_INTERFACE_MODE_RMII, + .ether_link_active_low = 1, +}; + +static const struct resource ether_resources[] __initconst = { + DEFINE_RES_MEM(0xee700000, 0x400), + DEFINE_RES_IRQ(gic_spi(162)), +}; + /* LEDS */ static struct gpio_led koelsch_leds[] = { { @@ -80,6 +96,15 @@ static const struct gpio_keys_platform_d }; static const struct pinctrl_map koelsch_pinctrl_map[] = { + /* Ether */ + PIN_MAP_MUX_GROUP_DEFAULT("r8a7791-ether", "pfc-r8a7791", + "eth_link", "eth"), + PIN_MAP_MUX_GROUP_DEFAULT("r8a7791-ether", "pfc-r8a7791", + "eth_mdio", "eth"), + PIN_MAP_MUX_GROUP_DEFAULT("r8a7791-ether", "pfc-r8a7791", + "eth_rmii", "eth"), + PIN_MAP_MUX_GROUP_DEFAULT("r8a7791-ether", "pfc-r8a7791", + "intc_irq0", "intc"), /* SCIF0 (CN19: DEBUG SERIAL0) */ PIN_MAP_MUX_GROUP_DEFAULT("sh-sci.6", "pfc-r8a7791", "scif0_data_d", "scif0"), @@ -95,6 +120,10 @@ static void __init koelsch_add_standard_ ARRAY_SIZE(koelsch_pinctrl_map)); r8a7791_pinmux_init(); r8a7791_add_standard_devices(); + platform_device_register_resndata(&platform_bus, "r8a7791-ether", -1, + ether_resources, + ARRAY_SIZE(ether_resources), + ðer_pdata, sizeof(ether_pdata)); platform_device_register_data(&platform_bus, "leds-gpio", -1, &koelsch_leds_pdata, sizeof(koelsch_leds_pdata)); @@ -103,6 +132,32 @@ static void __init koelsch_add_standard_ sizeof(koelsch_keys_pdata)); } +/* + * Ether LEDs on the Koelsch board are named LINK and ACTIVE which corresponds + * to non-default 01 setting of the Micrel KSZ8041 PHY control register 1 bits + * 14-15. We have to set them back to 01 from the default 00 value each time + * the PHY is reset. It's also important because the PHY's LED0 signal is + * connected to SoC's ETH_LINK signal and in the PHY's default mode it will + * bounce on and off after each packet, which we apparently want to avoid. + */ +static int koelsch_ksz8041_fixup(struct phy_device *phydev) +{ + u16 phyctrl1 = phy_read(phydev, 0x1e); + + phyctrl1 &= ~0xc000; + phyctrl1 |= 0x4000; + return phy_write(phydev, 0x1e, phyctrl1); +} + +static void __init koelsch_init(void) +{ + koelsch_add_standard_devices(); + + if (IS_ENABLED(CONFIG_PHYLIB)) + phy_register_fixup_for_id("r8a7791-ether-ff:01", + koelsch_ksz8041_fixup); +} + static const char * const koelsch_boards_compat_dt[] __initconst = { "renesas,koelsch", NULL, @@ -112,7 +167,7 @@ DT_MACHINE_START(KOELSCH_DT, "koelsch") .smp = smp_ops(r8a7791_smp_ops), .init_early = r8a7791_init_early, .init_time = rcar_gen2_timer_init, - .init_machine = koelsch_add_standard_devices, + .init_machine = koelsch_init, .init_late = shmobile_init_late, .dt_compat = koelsch_boards_compat_dt, MACHINE_END
Register Ether platform device and pin data on the Koelsch board. Register platform fixup for Micrel KSZ8041 PHY, just like on the Lager board. Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com> --- Changes in version 3: - changed "r8a779x-ether" wildcard device name to "r8a7791-ether"; - added PFC data for IRQ0 pin used by KSZ8041 PHY; - resolved rejects, refreshed the patch. Changes in version 2: - added *if* (IS_ENABLED(CONFIG_PHYLIB)) around phy_register_fixup_for_id() call; - changed Ether device name to "r8a779x-ether". arch/arm/mach-shmobile/board-koelsch.c | 57 ++++++++++++++++++++++++++++++++- 1 file changed, 56 insertions(+), 1 deletion(-)