net: macb: ignore tx_clk if MII is used
Subsystems (3)
PATCH 1/1 net: macb: ignore tx_clk if MII is used
If the MII interface is used, the PHY is the clock master, thus don't set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency: 25000000 Hz
--- drivers/net/ethernet/cadence/macb_main.c | 4 ++++ 1 file changed, 4 insertions(+)
drivers/net/ethernet/cadence/macb_main.c +4 -0
@@ -470,6 +470,10 @@ static void macb_set_tx_clk(struct macb *bp, int speed)if (!bp->tx_clk || (bp->caps &MACB_CAPS_CLK_HW_CHG ))return ; +/* In case of MII the PHY is the clock master */ +if (bp->phy_interface ==PHY_INTERFACE_MODE_MII ) +return ; +switch (speed) {case SPEED_10 : rate =2500000 ;
Signature · 1 line
2.20.1
Quoted 6 lines · Michael Walle
If the MII interface is used, the PHY is the clock master, thus don't set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency: 25000000 HzSigned-off-by Michael Walle <[email protected]>
Andrew
Hi Michael,
Quoted 7 lines · Michael Walle
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
If the MII interface is used, the PHY is the clock master, thus don't set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency: 25000000 Hz
Since in this case the PHY provides the TX clock and it provides the proper rate based on link speed, the MACB driver should not handle the bp->tx_clk at all (MACB driver uses this clock only for setting the proper rate on it based on link speed). So, I believe the proper fix would be to not pass the tx_clk at all in device tree. This clock is optional for MACB driver.
Thank you,
Claudiu Beznea
Quoted 23 lines
Signed-off-by Michael Walle <[email protected]>--- drivers/net/ethernet/cadence/macb_main.c | 4 ++++ 1 file changed, 4 insertions(+)drivers/net/ethernet/cadence/macb_main.c +4
-1@@ -470,6 +470,10 @@ static void macb_set_tx_clk(struct macb *bp, int speed)if (!bp->tx_clk || (bp->caps &MACB_CAPS_CLK_HW_CHG ))return ; +/* In case of MII the PHY is the clock master */ +if (bp->phy_interface ==PHY_INTERFACE_MODE_MII ) +return ; +switch (speed) {case SPEED_10 : rate =2500000 ; --2.20.1
Hi Claudiu,
Am 2021-01-21 10:19, schrieb [email protected]:
Quoted 10 lines · Michael Walle
Quoted 8 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
If the MII interface is used, the PHY is the clock master, thus don't set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency:
25000000 Hz
Quoted 5 lines
Since in this case the PHY provides the TX clock and it provides the proper rate based on link speed, the MACB driver should not handle the
bp->tx_clkat all (MACB driver uses this clock only for setting the proper rate on it based on link speed). So, I believe the proper fix would be to not pass thetx_clkat all in device tree. This clock is optional for MACB driver.
Thanks for looking into this.
I had the same thought. But shouldn't the driver handle this case gracefully? I mean it does know that the clock isn't needed at all. Ususually that clock is defined in a device tree include. So you'd have to redefine that node in an actual board file which means duplicating the other clocks.
-michael
Quoted 27 lines · Michael Walle
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
Hi Claudiu,
Am 2021-01-21 10:19, schrieb [email protected]:
Quoted 11 lines · Michael Walle
Quoted 9 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
If the MII interface is used, the PHY is the clock master, thus don't set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency:
25000000 HzQuoted 9 lines
Since in this case the PHY provides the TX clock and it provides the proper
rate based on link speed, the MACB driver should not handle thebp->tx_clk
at all (MACB driver uses this clock only for setting the proper rate on it
based on link speed). So, I believe the proper fix would be to not pass thetx_clkat all in device tree. This clock is optional for MACB driver.
Quoted 5 lines
Thanks for looking into this.
I had the same thought. But shouldn't the driver handle this case gracefully?
I mean it does know that the clock isn't needed at all.
Currently it may knows that by checking the bp->tx_clk. Moreover the clock could be provided by PHY not only for MII interface.
Moreover the IP has the bit "refclk" of register at offset 0xc (userio) that tells it to use the clock provided by PHY or to use one internal to the SoC. If a SoC generated clock would be used the IP logic may have the option to do the proper division based on link speed (if IP has this option enabled then this should be selected in driver with capability MACB_CAPS_CLK_HW_CHG).
If the clock provided by the PHY is the one to be used then this is selected with capability MACB_CAPS_USRIO_HAS_CLKEN. So, if the change you proposed in this patch is still imperative then checking for this capability would be the best as the clock could be provided by PHY not only for MII interface.
Quoted 7 lines
Ususually that
clock
is defined in a device tree include. So you'd have to redefine that node in
an actual board file which means duplicating the other clocks.-michael
Am 2021-01-22 10:10, schrieb [email protected]:
Quoted 28 lines · Michael Walle
Quoted 27 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
Hi Claudiu,
Am 2021-01-21 10:19, schrieb [email protected]:
Quoted 20 lines · Michael Walle
Quoted 9 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
If the MII interface is used, the PHY is the clock master, thus don't set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency:
25000000 HzSince in this case the PHY provides the TX clock and it provides the proper
rate based on link speed, the MACB driver should not handle thebp->tx_clk
at all (MACB driver uses this clock only for setting the proper rate on it
based on link speed). So, I believe the proper fix would be to not pass thetx_clkat all in device tree. This clock is optional for MACB driver.
Quoted 8 lines
Quoted 5 lines
Thanks for looking into this.
I had the same thought. But shouldn't the driver handle this case gracefully?
I mean it does know that the clock isn't needed at all.Currently it may knows that by checking the
bp->tx_clk. Moreover the clock could be provided by PHY not only for MII interface.
That doesn't make this patch wrong, does it? It just doesn't cover all use cases (which also wasn't covered before).
Quoted 7 lines
Moreover the IP has the bit "refclk" of register at offset
0xc(userio) that tells it to use the clock provided by PHY or to use one internal to the SoC. If a SoC generated clock would be used the IP logic may have the option to do the proper division based on link speed (if IP has this option enabled then this should be selected in driver with capabilityMACB_CAPS_CLK_HW_CHG).
Quoted 5 lines
If the clock provided by the PHY is the one to be used then this is selected with capability
MACB_CAPS_USRIO_HAS_CLKEN. So, if the change you proposed in this patch is still imperative then checking for this capability would be the best as the clock could be provided by PHY not only for MII interface.
Fair enough, but this register doesn't seem to be implemented on Zynq-7000. Albeit MACB_CAPS_USRIO_DISABLED isn't defined for the Zynq MACB. It isn't defined in the Zynq-7000 reference manual and you cannot set any bits:
=> mw 0xE000B00C 0xFFFFFFFF
=> md 0xE000B00C 1e000b00c: 00000000
Also please note, that tx_clk may be an arbitrary clock which doesn't necessarily need to be the clock which is controlled by CLK_EN. Or am I missing something here?
-michael
Quoted 7 lines
Quoted 7 lines
Ususually that
clock
is defined in a device tree include. So you'd have to redefine that node in
an actual board file which means duplicating the other clocks.-michael
Quoted 43 lines · Michael Walle
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
Am 2021-01-22 10:10, schrieb [email protected]:
Quoted 39 lines · Michael Walle
Quoted 37 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know the
content is safeHi Claudiu,
Am 2021-01-21 10:19, schrieb [email protected]:
Quoted 13 lines · Michael Walle
Quoted 11 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know
the content is safeIf the MII interface is used, the PHY is the clock master, thus don't
set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency:
25000000 HzQuoted 11 lines
Since in this case the PHY provides the TX clock and it provides the proper
rate based on link speed, the MACB driver should not handle thebp->tx_clk
at all (MACB driver uses this clock only for setting the proper rate on
it
based on link speed). So, I believe the proper fix would be to not pass
thetx_clkat all in device tree. This clock is optional for MACB driver.Thanks for looking into this.
I had the same thought. But shouldn't the driver handle this case gracefully?
I mean it does know that the clock isn't needed at all.
Quoted 6 lines
Quoted 3 lines
Currently it may knows that by checking the
bp->tx_clk. Moreover the clock
could be provided by PHY not only for MII interface.That doesn't make this patch wrong, does it? It just doesn't cover all use cases (which also wasn't covered before).
I would say that it breaks setups using MII interface and with clock provided via DT that need to be handled by macb_set_tx_clk().
Quoted 19 lines
Quoted 11 lines
Moreover the IP has the bit "refclk" of register at offset
0xc(userio) that tells it to use the clock provided by PHY or to use one internal to
the SoC. If a SoC generated clock would be used the IP logic may have the
option to do the proper division based on link speed (if IP has this option
enabled then this should be selected in driver with capabilityMACB_CAPS_CLK_HW_CHG).Quoted 7 lines
If the clock provided by the PHY is the one to be used then this is selected with capability
MACB_CAPS_USRIO_HAS_CLKEN. So, if the change you
proposed in this patch is still imperative then checking for this capability would be the best as the clock could be provided by PHY not only
for MII interface.
Quoted 8 lines
Fair enough, but this register doesn't seem to be implemented on Zynq-7000. Albeit
MACB_CAPS_USRIO_DISABLEDisn't defined for the Zynq MACB. It isn't defined in the Zynq-7000 reference manual and you cannot set any bits:=> mw
0xE000B00C0xFFFFFFFF
=> md0xE000B00C1e000b00c: 00000000
I wasn't aware of this. In this case, maybe adding the MACB_CAPS_USRIO_DISABLED to the Zync-7000 capability list and checking this one plus MACB_CAPS_USRIO_HAS_CLKEN would be better instead of checking the MAC-PHY interface?
Quoted 4 lines
Also please note, that
tx_clkmay be an arbitrary clock which doesn't necessarily need to be the clock which is controlled byCLK_EN. Or am I missing something here?
I suppose that whoever creates the device tree knows what is doing and it passes the proper clock to macb driver.
Thank you,
Claudiu Beznea
Quoted 11 lines
-michael
Quoted 8 lines
Quoted 8 lines
Ususually that
clock
is defined in a device tree include. So you'd have to redefine that node
in
an actual board file which means duplicating the other clocks.-michael
Am 2021-01-22 12:38, schrieb [email protected]:
Quoted 48 lines · Michael Walle
Quoted 47 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
Am 2021-01-22 10:10, schrieb [email protected]:
Quoted 42 lines · Michael Walle
Quoted 32 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know the
content is safeHi Claudiu,
Am 2021-01-21 10:19, schrieb [email protected]:
Quoted 13 lines · Michael Walle
Quoted 11 lines
EXTERNAL EMAIL: Do not click links or open attachments unless you know
the content is safeIf the MII interface is used, the PHY is the clock master, thus don't
set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency:
25000000 HzQuoted 11 lines
Since in this case the PHY provides the TX clock and it provides the proper
rate based on link speed, the MACB driver should not handle thebp->tx_clk
at all (MACB driver uses this clock only for setting the proper rate on
it
based on link speed). So, I believe the proper fix would be to not pass
thetx_clkat all in device tree. This clock is optional for MACB driver.Quoted 5 lines
Thanks for looking into this.
I had the same thought. But shouldn't the driver handle this case gracefully?
I mean it does know that the clock isn't needed at all.Currently it may knows that by checking the
bp->tx_clk. Moreover the clock
could be provided by PHY not only for MII interface.
Quoted 5 lines
Quoted 2 lines
That doesn't make this patch wrong, does it? It just doesn't cover all use cases (which also wasn't covered before).
I would say that it breaks setups using MII interface and with clock provided via DT that need to be handled by
macb_set_tx_clk().
But the MII interface has by definition no tx clock? At the moment tx_clk is set to the correct frequency for RGMII interfaces, right? How would it break boards then?
And if you use tx_clk for like driving the PHY refclk input, that would be an abuse of this clock (and shouldn't even work, because it just sets the correct RGMII frequencies).
Quoted 23 lines
Quoted 23 lines
Quoted 10 lines
Moreover the IP has the bit "refclk" of register at offset
0xc(userio) that tells it to use the clock provided by PHY or to use one internal to
the SoC. If a SoC generated clock would be used the IP logic may have the
option to do the proper division based on link speed (if IP has this option
enabled then this should be selected in driver with capabilityMACB_CAPS_CLK_HW_CHG).Quoted 7 lines
If the clock provided by the PHY is the one to be used then this is selected with capability
MACB_CAPS_USRIO_HAS_CLKEN. So, if the change you
proposed in this patch is still imperative then checking for this capability would be the best as the clock could be provided by PHY not only
for MII interface.Fair enough, but this register doesn't seem to be implemented on Zynq-7000. Albeit
MACB_CAPS_USRIO_DISABLEDisn't defined for the Zynq MACB. It isn't defined in the Zynq-7000 reference manual and you cannot set any bits:
Quoted 8 lines
Quoted 3 lines
=> mw
0xE000B00C0xFFFFFFFF
=> md0xE000B00C1e000b00c: 00000000I wasn't aware of this. In this case, maybe adding the
MACB_CAPS_USRIO_DISABLEDto the Zync-7000 capability list and checking this one plusMACB_CAPS_USRIO_HAS_CLKENwould be better instead of checking the MAC-PHY interface?
But then RGMII would be broken. Zynq and ZynqMP are the only users of the tx_clk as far as I can see. There, it will set the clock generated by the clock controller which is driving the TX_CLK for RGMII and the corresponding input of the GEM/MACB.
I mean you have the same SoC, thus the same caps and usrio settings, but you have two possible interfaces: MII and RGMII. (At least, because there might also be other interfaces like GMII). Therefore, you cannot check this bit, right? It will be the same for both MII and RGMII mode.
In any case, it should be correct to add MACB_CAPS_USRIO_DISABLED to caps for the zynq family (I'd need to double check the ZynqMP RM).
Quoted 6 lines
Quoted 3 lines
Also please note, that
tx_clkmay be an arbitrary clock which doesn't necessarily need to be the clock which is controlled byCLK_EN. Or am I missing something here?I suppose that whoever creates the device tree knows what is doing and it passes the proper clock to macb driver.
Mh. If CLK_EN is supported this might be the case. Unfortunalty, Zynq is the only user and doesn't have this bit.
-michael
PS. dont' get me wrong, I'm all for fixing this "the right way".
This patch was applied to netdev/net-next.git (refs/heads/master):
Hello:
This patch was applied to netdev/net-next.git (refs/heads/master):
Quoted 8 lines · On Wed, 20 Jan 2021 20:43:03 +0100 you wrote:
If the MII interface is used, the PHY is the clock master, thus don't set the clock rate. On Zynq-7000, this will prevent the following warning:
macb e000b000.ethernet eth0: unable to generate target frequency: 25000000 HzSigned-off-by Michael Walle <[email protected]>[...]
Here is the summary with links:
- net: macb: ignore tx_clk if MII is used
https://git.kernel.org/netdev/net-next/c/43e5763152e2
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html