ARM: mxs: Do not reconfigure FEC clock in FEC init
PATCH 1/1 ARM: mxs: Do not reconfigure FEC clock in FEC init
Do not reconfigure the FEC clock during board_eth_init(), otherwise the FEC might have stability issues, refuse to autonegotiate link entirely or even corrupt packets while indicating correct checksum on them. Instead, move the FEC clock init to board_early_init_f(), where all the other upstream clock are initialized and also make sure there is proper stabilization delay.
--- board/bluegiga/apx4devkit/apx4devkit.c | 10 ++++------ board/denx/m28evk/m28evk.c | 21 +++++++++++---------- board/freescale/mx28evk/mx28evk.c | 9 +++++---- board/schulercontrol/sc_sps_1/sc_sps_1.c | 19 +++++++++++-------- 4 files changed, 31 insertions(+), 28 deletions(-)
NOTE: I'd like to get this into current release as it fixes a serious
issue I observe here on the MX28EVK (packets being corrupted on long transfers initiated early after boot). Please test this and report back ASAP.
board/bluegiga/apx4devkit/apx4devkit.c +4 -6
@@ -39,6 +39,10 @@ int board_early_init_f(void)/* SSP0 clock at 96MHz */ mxs_set_sspclk (MXC_SSPCLK0 ,96000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +udelay (10000 ); +#endif return 0 ; }
@@ -79,12 +83,6 @@ int board_eth_init(bd_t *bis)int ret;struct eth_device *dev; - ret =cpu_eth_init (bis); -if (ret) { -printf ("FEC MXS: Unable to init FEC clocks\n" ); -return ret; - } - ret =fecmxc_initialize (bis);if (ret) {printf ("FEC MXS: Unable to init FEC\n" );
board/denx/m28evk/m28evk.c +11 -10
@@ -26,6 +26,9 @@ DECLARE_GLOBAL_DATA_PTR;*/ int board_early_init_f (void ) { +struct mxs_clkctrl_regs *clkctrl_regs = + (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ; +/* IO0 clock at 480MHz */ mxs_set_ioclk (MXC_IOCLK0 ,480000 );/* IO1 clock at 480MHz */
@@ -36,6 +39,14 @@ int board_early_init_f(void)/* SSP2 clock at 160MHz */ mxs_set_sspclk (MXC_SSPCLK2 ,160000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, +CLKCTRL_ENET_TIME_SEL_MASK |CLKCTRL_ENET_CLK_OUT_EN , +CLKCTRL_ENET_TIME_SEL_RMII_CLK ); +udelay (10000 ); +#endif +#ifdef CONFIG_CMD_USB mxs_iomux_setup_pad (MX28_PAD_SSP2_SS1__USB1_OVERCURRENT ); mxs_iomux_setup_pad(MX28_PAD_AUART3_TX__GPIO_3_13 |
@@ -110,19 +121,9 @@ int fecmxc_mii_postcall(int phy)int board_eth_init (bd_t *bis) { -struct mxs_clkctrl_regs *clkctrl_regs = - (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ;struct eth_device *dev;int ret; - ret =cpu_eth_init (bis); -if (ret) -return ret; - -clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, -CLKCTRL_ENET_TIME_SEL_MASK |CLKCTRL_ENET_CLK_OUT_EN , -CLKCTRL_ENET_TIME_SEL_RMII_CLK ); -#if !defined(CONFIG_DENX_M28_V11 ) && !defined(CONFIG_DENX_M28_V10 )/* Reset the new PHY */ gpio_direction_output (MX28_PAD_AUART2_RTS__GPIO_3_11 ,0 );
board/freescale/mx28evk/mx28evk.c +5 -4
@@ -41,6 +41,11 @@ int board_early_init_f(void)/* SSP2 clock at 160MHz */ mxs_set_sspclk (MXC_SSPCLK2 ,160000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +udelay (10000 ); +#endif +#ifdef CONFIG_CMD_USB mxs_iomux_setup_pad (MX28_PAD_SSP2_SS1__USB1_OVERCURRENT ); mxs_iomux_setup_pad(MX28_PAD_AUART2_RX__GPIO_3_8 |
@@ -102,10 +107,6 @@ int board_eth_init(bd_t *bis)struct eth_device *dev;int ret; - ret =cpu_eth_init (bis); -if (ret) -return ret; -/* MX28EVK uses ENET_CLK PAD to drive FEC clock */ writel (CLKCTRL_ENET_TIME_SEL_RMII_CLK |CLKCTRL_ENET_CLK_OUT_EN , &clkctrl_regs->hw_clkctrl_enet);
board/schulercontrol/sc_sps_1/sc_sps_1.c +11 -8
@@ -26,6 +26,9 @@ DECLARE_GLOBAL_DATA_PTR;*/ int board_early_init_f (void ) { +struct mxs_clkctrl_regs *clkctrl_regs = + (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ; +/* IO0 clock at 480MHz */ mxs_set_ioclk (MXC_IOCLK0 ,480000 );/* IO1 clock at 480MHz */
@@ -36,6 +39,14 @@ int board_early_init_f(void)/* SSP2 clock at 96MHz */ mxs_set_sspclk (MXC_SSPCLK2 ,96000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, +CLKCTRL_ENET_TIME_SEL_MASK , +CLKCTRL_ENET_TIME_SEL_RMII_CLK |CLKCTRL_ENET_CLK_OUT_EN ); +udelay (10000 ); +#endif +#ifdef CONFIG_CMD_USB mxs_iomux_setup_pad (MX28_PAD_AUART1_CTS__USB0_OVERCURRENT ); mxs_iomux_setup_pad(MX28_PAD_AUART2_TX__GPIO_3_9 |
@@ -69,16 +80,8 @@ int board_mmc_init(bd_t *bis)#ifdef CONFIG_CMD_NET int board_eth_init (bd_t *bis) { -struct mxs_clkctrl_regs *clkctrl_regs = - (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ;int ret; - ret =cpu_eth_init (bis); - -clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, -CLKCTRL_ENET_TIME_SEL_MASK , -CLKCTRL_ENET_TIME_SEL_RMII_CLK |CLKCTRL_ENET_CLK_OUT_EN ); - ret =fecmxc_initialize_multi (bis,0 ,0 ,MXS_ENET0_BASE );if (ret) {printf ("FEC MXS: Unable to init FEC0\n" );
Signature · 1 line
1.8.4.rc3
PATCH 1/1 ARM: mxs: Do not reconfigure FEC clock in FEC init
Hi,
CCing Alexandre and Hector. Can you test please?
Quoted 176 lines
Do not reconfigure the FEC clock during
board_eth_init(), otherwise the FEC might have stability issues, refuse to autonegotiate link entirely or even corrupt packets while indicating correct checksum on them. Instead, move the FEC clock init toboard_early_init_f(), where all the other upstream clock are initialized and also make sure there is proper stabilization delay.Signed-off-by Marek Vasut <[email protected]>Cc Fabio Estevam <[email protected]>Cc Hector Palacios <[email protected]>Cc Oliver Metz <[email protected]>Cc Otavio Salvador <[email protected]>Cc Stefano Babic <[email protected]>Cc Tom Rini <[email protected]>--- board/bluegiga/apx4devkit/apx4devkit.c | 10 ++++------ board/denx/m28evk/m28evk.c | 21 +++++++++++---------- board/freescale/mx28evk/mx28evk.c | 9 +++++---- board/schulercontrol/sc_sps_1/sc_sps_1.c | 19 +++++++++++-------- 4 files changed, 31 insertions(+), 28 deletions(-)NOTE: I'd like to get this into current release as it fixes a serious
issue I observe here on the MX28EVK (packets being corrupted on long transfers initiated early after boot). Please test this and report back ASAP.(quoted diff) +0
-0
b/board/bluegiga/apx4devkit/apx4devkit.cindex 08e79bd..044a182 100644board/bluegiga/apx4devkit/apx4devkit.c +4
-6@@ -39,6 +39,10 @@ int board_early_init_f(void)/* SSP0 clock at 96MHz */ mxs_set_sspclk (MXC_SSPCLK0 ,96000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +udelay (10000 ); +#endif return 0 ; }@@ -79,12 +83,6 @@ int board_eth_init(bd_t *bis)int ret;struct eth_device *dev; - ret =cpu_eth_init (bis); -if (ret) { -printf ("FEC MXS: Unable to init FEC clocks\n" ); -return ret; - } - ret =fecmxc_initialize (bis);if (ret) {printf ("FEC MXS: Unable to init FEC\n" );board/denx/m28evk/m28evk.c +11
-10@@ -26,6 +26,9 @@ DECLARE_GLOBAL_DATA_PTR;*/ int board_early_init_f (void ) { +struct mxs_clkctrl_regs *clkctrl_regs = + (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ; +/* IO0 clock at 480MHz */ mxs_set_ioclk (MXC_IOCLK0 ,480000 );/* IO1 clock at 480MHz */ @@ -36,6 +39,14 @@ int board_early_init_f(void)/* SSP2 clock at 160MHz */ mxs_set_sspclk (MXC_SSPCLK2 ,160000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, +CLKCTRL_ENET_TIME_SEL_MASK |CLKCTRL_ENET_CLK_OUT_EN , +CLKCTRL_ENET_TIME_SEL_RMII_CLK ); +udelay (10000 ); +#endif +#ifdef CONFIG_CMD_USB mxs_iomux_setup_pad (MX28_PAD_SSP2_SS1__USB1_OVERCURRENT ); mxs_iomux_setup_pad(MX28_PAD_AUART3_TX__GPIO_3_13 |@@ -110,19 +121,9 @@ int fecmxc_mii_postcall(int phy)int board_eth_init (bd_t *bis) { -struct mxs_clkctrl_regs *clkctrl_regs = - (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ;struct eth_device *dev;int ret; - ret =cpu_eth_init (bis); -if (ret) -return ret; - -clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, -CLKCTRL_ENET_TIME_SEL_MASK |CLKCTRL_ENET_CLK_OUT_EN , -CLKCTRL_ENET_TIME_SEL_RMII_CLK ); -#if !defined(CONFIG_DENX_M28_V11 ) && !defined(CONFIG_DENX_M28_V10 )/* Reset the new PHY */ gpio_direction_output (MX28_PAD_AUART2_RTS__GPIO_3_11 ,0 );(quoted diff) +0
-0
b/board/freescale/mx28evk/mx28evk.cindex 5005fe2..2c93c44 100644board/freescale/mx28evk/mx28evk.c +5
-4@@ -41,6 +41,11 @@ int board_early_init_f(void)/* SSP2 clock at 160MHz */ mxs_set_sspclk (MXC_SSPCLK2 ,160000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +udelay (10000 ); +#endif +#ifdef CONFIG_CMD_USB mxs_iomux_setup_pad (MX28_PAD_SSP2_SS1__USB1_OVERCURRENT ); mxs_iomux_setup_pad(MX28_PAD_AUART2_RX__GPIO_3_8 |@@ -102,10 +107,6 @@ int board_eth_init(bd_t *bis)struct eth_device *dev;int ret; - ret =cpu_eth_init (bis); -if (ret) -return ret; -/* MX28EVK uses ENET_CLK PAD to drive FEC clock */ writel (CLKCTRL_ENET_TIME_SEL_RMII_CLK |CLKCTRL_ENET_CLK_OUT_EN , &clkctrl_regs->hw_clkctrl_enet);(quoted diff) +0
-0
b/board/schulercontrol/sc_sps_1/sc_sps_1.cindex 7f0b591..9d3c970 100644board/schulercontrol/sc_sps_1/sc_sps_1.c +11
-8@@ -26,6 +26,9 @@ DECLARE_GLOBAL_DATA_PTR;*/ int board_early_init_f (void ) { +struct mxs_clkctrl_regs *clkctrl_regs = + (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ; +/* IO0 clock at 480MHz */ mxs_set_ioclk (MXC_IOCLK0 ,480000 );/* IO1 clock at 480MHz */ @@ -36,6 +39,14 @@ int board_early_init_f(void)/* SSP2 clock at 96MHz */ mxs_set_sspclk (MXC_SSPCLK2 ,96000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, +CLKCTRL_ENET_TIME_SEL_MASK , +CLKCTRL_ENET_TIME_SEL_RMII_CLK |CLKCTRL_ENET_CLK_OUT_EN ); +udelay (10000 ); +#endif +#ifdef CONFIG_CMD_USB mxs_iomux_setup_pad (MX28_PAD_AUART1_CTS__USB0_OVERCURRENT ); mxs_iomux_setup_pad(MX28_PAD_AUART2_TX__GPIO_3_9 |@@ -69,16 +80,8 @@ int board_mmc_init(bd_t *bis)#ifdef CONFIG_CMD_NET int board_eth_init (bd_t *bis) { -struct mxs_clkctrl_regs *clkctrl_regs = - (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ;int ret; - ret =cpu_eth_init (bis); - -clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, -CLKCTRL_ENET_TIME_SEL_MASK , -CLKCTRL_ENET_TIME_SEL_RMII_CLK |CLKCTRL_ENET_CLK_OUT_EN ); - ret =fecmxc_initialize_multi (bis,0 ,0 ,MXS_ENET0_BASE );if (ret) {printf ("FEC MXS: Unable to init FEC0\n" );
PATCH 1/1 ARM: mxs: Do not reconfigure FEC clock in FEC init
In reply to Marek VasutDear Marek,
Quoted 6 lines · Marek Vasut
Do not reconfigure the FEC clock during
board_eth_init(), otherwise the FEC might have stability issues, refuse to autonegotiate link entirely or even corrupt packets while indicating correct checksum on them. Instead, move the FEC clock init toboard_early_init_f(), where all the other upstream clock are initialized and also make sure there is proper stabilization delay.
Do you have any means to reproduce the problem?
I have seldom seen times when the Ethernet did not work, but it was so infrequent that it was impossible to know where it came from or how to reproduce it.
[...]
board/denx/m28evk/m28evk.c DECLARE_GLOBAL_DATA_PTR; · line 31
Quoted 11 lines
board/denx/m28evk/m28evk.c +3
-0@@ -26,6 +26,9 @@ DECLARE_GLOBAL_DATA_PTR;*/ int board_early_init_f (void ) { +struct mxs_clkctrl_regs *clkctrl_regs = + (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ; +
You may want to wrap these within #ifdef CONFIG_CMD_NET to avoid a warning if not defined.
board/schulercontrol/sc_sps_1/sc_sps_1.c DECLARE_GLOBAL_DATA_PTR; · line 31
Quoted 11 lines
board/schulercontrol/sc_sps_1/sc_sps_1.c +3
-0@@ -26,6 +26,9 @@ DECLARE_GLOBAL_DATA_PTR;*/ int board_early_init_f (void ) { +struct mxs_clkctrl_regs *clkctrl_regs = + (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ; +
And the same here.
It works ok, but I can't say if it fixes exactly those seldom initialization problems because I don't have a reliable way to reproduce them.
Best regards,
--
Hector Palacios
PATCH 1/1 ARM: mxs: Do not reconfigure FEC clock in FEC init
In reply to Marek VasutDear Marek,
Quoted 20 lines · Marek Vasut
Do not reconfigure the FEC clock during
board_eth_init(), otherwise the FEC might have stability issues, refuse to autonegotiate link entirely or even corrupt packets while indicating correct checksum on them. Instead, move the FEC clock init toboard_early_init_f(), where all the other upstream clock are initialized and also make sure there is proper stabilization delay.Signed-off-by Marek Vasut <[email protected]>Cc Fabio Estevam <[email protected]>Cc Hector Palacios <[email protected]>Cc Oliver Metz <[email protected]>Cc Otavio Salvador <[email protected]>Cc Stefano Babic <[email protected]>Cc Tom Rini <[email protected]>--- board/bluegiga/apx4devkit/apx4devkit.c | 10 ++++------ board/denx/m28evk/m28evk.c | 21 +++++++++++---------- board/freescale/mx28evk/mx28evk.c | 9 +++++---- board/schulercontrol/sc_sps_1/sc_sps_1.c | 19 +++++++++++-------- 4 files changed, 31 insertions(+), 28 deletions(-)
[...]
Quoted 17 lines
board/freescale/mx28evk/mx28evk.c +5
-0@@ -41,6 +41,11 @@ int board_early_init_f(void)/* SSP2 clock at 160MHz */ mxs_set_sspclk (MXC_SSPCLK2 ,160000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +udelay (10000 ); +#endif +#ifdef CONFIG_CMD_USB mxs_iomux_setup_pad (MX28_PAD_SSP2_SS1__USB1_OVERCURRENT ); mxs_iomux_setup_pad(MX28_PAD_AUART2_RX__GPIO_3_8 |
board/freescale/mx28evk/mx28evk.c int board_eth_init(bd_t *bis) · line 112
Quoted 10 lines
board/freescale/mx28evk/mx28evk.c +0
-4@@ -102,10 +107,6 @@ int board_eth_init(bd_t *bis)struct eth_device *dev;int ret; - ret =cpu_eth_init (bis); -if (ret) -return ret; -/* MX28EVK uses ENET_CLK PAD to drive FEC clock */ writel (CLKCTRL_ENET_TIME_SEL_RMII_CLK |CLKCTRL_ENET_CLK_OUT_EN , &clkctrl_regs->hw_clkctrl_enet);
Why didn't you move the clk enable instruction on mx28evk, as you did with the rest of platforms?
Best regards,
--
Hector Palacios
PATCH 1/1 ARM: mxs: Do not reconfigure FEC clock in FEC init
In reply to Marek VasutHi Marek,
Quoted 6 lines · Marek Vasut
Do not reconfigure the FEC clock during
board_eth_init(), otherwise the FEC might have stability issues, refuse to autonegotiate link entirely or even corrupt packets while indicating correct checksum on them. Instead, move the FEC clock init toboard_early_init_f(), where all the other upstream clock are initialized and also make sure there is proper stabilization delay.
However, it seems to me that moving the code remove the problem on the mx28evk without finding the cause. Really there is still a big chunk of time between the initialization of the clock and the first packet is sent to the network, when the autonegotiation should occur.
Quoted 25 lines
Signed-off-by Marek Vasut <[email protected]>Cc Fabio Estevam <[email protected]>Cc Hector Palacios <[email protected]>Cc Oliver Metz <[email protected]>Cc Otavio Salvador <[email protected]>Cc Stefano Babic <[email protected]>Cc Tom Rini <[email protected]>--- board/bluegiga/apx4devkit/apx4devkit.c | 10 ++++------ board/denx/m28evk/m28evk.c | 21 +++++++++++---------- board/freescale/mx28evk/mx28evk.c | 9 +++++---- board/schulercontrol/sc_sps_1/sc_sps_1.c | 19 +++++++++++-------- 4 files changed, 31 insertions(+), 28 deletions(-)NOTE: I'd like to get this into current release as it fixes a serious
issue I observe here on the MX28EVK (packets being corrupted on long transfers initiated early after boot). Please test this and report back ASAP.board/bluegiga/apx4devkit/apx4devkit.c +0
-0
board/bluegiga/apx4devkit/apx4devkit.c int board_early_init_f(void) · line 45
Quoted 7 lines
board/bluegiga/apx4devkit/apx4devkit.c +4
-0@@ -39,6 +39,10 @@ int board_early_init_f(void)/* SSP0 clock at 96MHz */ mxs_set_sspclk (MXC_SSPCLK0 ,96000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +udelay (10000 ); +#endif
This looks like a hack...
Calling cpu_eth_init() without parameters is absolutely an exception in u-boot, there is no other instance / SOC doing this.
Are we sure that the issue you see on mx28evk are reproducible on the other boards ? You are starting to fix all boards, but you report corruption only on one board.
Quoted 31 lines
board/bluegiga/apx4devkit/apx4devkit.c +0
-6return 0 ; }@@ -79,12 +83,6 @@ int board_eth_init(bd_t *bis)int ret;struct eth_device *dev; - ret =cpu_eth_init (bis); -if (ret) { -printf ("FEC MXS: Unable to init FEC clocks\n" ); -return ret; - } - ret =fecmxc_initialize (bis);if (ret) {printf ("FEC MXS: Unable to init FEC\n" );board/denx/m28evk/m28evk.c +3
-0@@ -26,6 +26,9 @@ DECLARE_GLOBAL_DATA_PTR;*/ int board_early_init_f (void ) { +struct mxs_clkctrl_regs *clkctrl_regs = + (struct mxs_clkctrl_regs *)MXS_CLKCTRL_BASE ; +/* IO0 clock at 480MHz */ mxs_set_ioclk (MXC_IOCLK0 ,480000 );/* IO1 clock at 480MHz */
board/denx/m28evk/m28evk.c int board_early_init_f(void) · line 47
Quoted 9 lines
board/denx/m28evk/m28evk.c +6
-0@@ -36,6 +39,14 @@ int board_early_init_f(void)/* SSP2 clock at 160MHz */ mxs_set_sspclk (MXC_SSPCLK2 ,160000 ,0 ); +#ifdef CONFIG_CMD_NET +cpu_eth_init (NULL ); +clrsetbits_le32 (&clkctrl_regs->hw_clkctrl_enet, +CLKCTRL_ENET_TIME_SEL_MASK |CLKCTRL_ENET_CLK_OUT_EN , +CLKCTRL_ENET_TIME_SEL_RMII_CLK ); +udelay (10000 );
It seems to me that the most obvious change is the 10mS delay you added after setting the clock. If we find that such delay is really necessary, should we not fix it insife the cpu_eth_init() else in all board initialization functions ?
Best regards,
Stefano Babic
Signature · 5 lines
===================================================================== DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: +49-8142-66989-53 Fax: +49-8142-66989-80 Email: sbabic at denx.de =====================================================================
PATCH 1/1 ARM: mxs: Do not reconfigure FEC clock in FEC init
Dear Stefano Babic,
Quoted 10 lines
Hi Marek,
Quoted 6 lines · Marek Vasut
Do not reconfigure the FEC clock during
board_eth_init(), otherwise the FEC might have stability issues, refuse to autonegotiate link entirely or even corrupt packets while indicating correct checksum on them. Instead, move the FEC clock init toboard_early_init_f(), where all the other upstream clock are initialized and also make sure there is proper stabilization delay.
Quoted 4 lines
However, it seems to me that moving the code remove the problem on the mx28evk without finding the cause. Really there is still a big chunk of time between the initialization of the clock and the first packet is sent to the network, when the autonegotiation should occur.
Forget this, the issue is still present. Damn!
It's at least a little clearer to me what happens now. The FEC fails to transmit an TFTP block ACK packet -> the server waits 5 seconds -> sends ARP packet -> FEC replies and resends the TFTP block ACK again. Why does the FEC swallow a TX packet, I don't know.
Best regards,
Marek Vasut
PATCH 1/1 ARM: mxs: Do not reconfigure FEC clock in FEC init
Hi Marek,
Quoted 7 lines · Marek Vasut
Forget this, the issue is still present. Damn!
It's at least a little clearer to me what happens now. The FEC fails to transmit an TFTP block ACK packet -> the server waits 5 seconds -> sends ARP packet -> FEC replies and resends the TFTP block ACK again. Why does the FEC swallow a TX packet, I don't know.
Anyway, it is strange that this happens only with this board. It appears as the SOC sends or thinks to have sent, but the packet is corrupted or not sent at all on the physical layer. Maybe you can check on the other side of the connection on the switch and/or on the server, if some errors are detected.
Best regards,
Stefano Babic
Signature · 5 lines
===================================================================== DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: +49-8142-66989-53 Fax: +49-8142-66989-80 Email: sbabic at denx.de =====================================================================
PATCH 1/1 ARM: mxs: Do not reconfigure FEC clock in FEC init
Dear Stefano Babic,
Quoted 10 lines
Hi Marek,
Quoted 6 lines · Marek Vasut
Forget this, the issue is still present. Damn!
It's at least a little clearer to me what happens now. The FEC fails to transmit an TFTP block ACK packet -> the server waits 5 seconds -> sends ARP packet -> FEC replies and resends the TFTP block ACK again. Why does the FEC swallow a TX packet, I don't know.
Quoted 5 lines
Anyway, it is strange that this happens only with this board. It appears as the SOC sends or thinks to have sent, but the packet is corrupted or not sent at all on the physical layer. Maybe you can check on the other side of the connection on the switch and/or on the server, if some errors are detected.
This happens on multiple boards and we're sometimes getting reports about this issue. So this is not a single board issue. Maybe we should figure out what do these boards have in common.
Best regards,
Marek Vasut