Skip to contentSkip to conversation

net: ravb: Fix NULL pointer access

PATCH New Biju Das 3 participants 13 messages Sep 15, 2020 – Sep 21, 2020 Network drivers

lore
Share & tools
lore

Biju Das

author
#0 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

Some phy's like rtl8211e do not support writeext() callback. Add a check to avoid null pointer access before calling writeext() callback.

---
 drivers/net/ravb.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ravb.c +2 -1
index 886f53ee82..623d377897 100644
@@ -438,7 +438,8 @@ static int ravb_config(struct udevice *dev)
 
 	writel(mask, eth->iobase + RAVB_REG_ECMR);
 
-	phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
+	if (phy->drv->writeext)
+		phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
 
 	return 0;
 }
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

Quoted 15 lines · Biju Das

Some phy's like rtl8211e do not support writeext() callback. Add a check to avoid null pointer access before calling writeext() callback.

---
 drivers/net/ravb.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ravb.c +0 -0
index 886f53ee82..623d377897 100644

drivers/net/ravb.c static int ravb_config(struct udevice *dev) · line 445

Quoted 9 lines
drivers/net/ravb.c +2 -1
@@ -438,7 +438,8 @@ static int ravb_config(struct udevice *dev)
 
 	writel(mask, eth->iobase + RAVB_REG_ECMR);
 
-	phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
+	if (phy->drv->writeext)
+		phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
 
 	return 0;
 }

Make sense.

Thanks,
Michal

#2 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

In reply to Biju Das

On 9/15/20 4:10 PM, Biju Das wrote:
[...]

drivers/net/ravb.c static int ravb_config(struct udevice *dev) · line 442

Quoted 8 lines

+++ b/drivers/net/ravb.c

drivers/net/ravb.c +2 -1
@@ -438,7 +438,8 @@ static int ravb_config(struct udevice *dev)
 
 	writel(mask, eth->iobase + RAVB_REG_ECMR);
 
-	phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
+	if (phy->drv->writeext)
+		phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);

Shouldn't we rather move this into the PHY driver altogether ? I _think_ this might be specific to some Micrel PHY.

author
#3 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

Hi Marek,

Thanks for the review.

Quoted 7 lines

Subject: Re: [PATCH] net: ravb: Fix NULL pointer access

On 9/15/20 4:10 PM, Biju Das wrote:
[...]

Quoted 3 lines

+++ b/drivers/net/ravb.c

(quoted diff) +0 -0
@@ -438,7 +438,8 @@ static int ravb_config(struct udevice *dev)
 
Quoted 7 lines

drivers/net/ravb.c static int ravb_config(struct udevice *dev) · line 442

Quoted 5 lines
drivers/net/ravb.c +2 -1
 writel(mask, eth->iobase + RAVB_REG_ECMR);
 
-phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
+if (phy->drv->writeext)
+phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);

Shouldn't we rather move this into the PHY driver altogether ?

If I fix, the phydriver with empty function, compiler will complain about unused parameter right? What about other phy's which doesn't have this callback.

Quoted 1 line

I _think_ this might be specific to some Micrel PHY.

As mentioned in the commit message, it is seen with Realtek phy(rtl8211e) and crash seen on Hihope RZ/G2M board.

Thanks and regards,
biju

Renesas Electronics Europe GmbH, Geschaeftsfuehrer/President: Carsten Jauch, Sitz der Gesellschaft/Registered office: Duesseldorf, Arcadiastrasse 10, 40472 Duesseldorf, Germany, Handelsregister/Commercial Register: Duesseldorf, HRB 3708 USt-IDNr./Tax identification no.: DE 119353406 WEEE-Reg.-Nr./WEEE reg. no.: DE 14978647

#4 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

Quoted 1 line · Biju Das

Hi Marek,

Hi,

[...]

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

+++ b/drivers/net/ravb.c

drivers/net/ravb.c +0 -0
@@ -438,7 +438,8 @@ static int ravb_config(struct udevice *dev)
 
 writel(mask, eth->iobase + RAVB_REG_ECMR);
 
Quoted 8 lines
Quoted 5 lines

drivers/net/ravb.c static int ravb_config(struct udevice *dev) · line 442

Quoted 3 lines
drivers/net/ravb.c +2 -1
-phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
+if (phy->drv->writeext)
+phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);

Shouldn't we rather move this into the PHY driver altogether ?

If I fix, the phydriver with empty function, compiler will complain about unused parameter right? What about other phy's which doesn't have this callback.

I mean, should we not move this entire code which configures something in a Micrel PHY, and is specific to Micrel PHY, into the Micrel PHY driver and remove the whole call of writeext from this ethernet driver ?

Quoted 2 lines
Quoted 1 line

I _think_ this might be specific to some Micrel PHY.

As mentioned in the commit message, it is seen with Realtek phy(rtl8211e) and crash seen on Hihope RZ/G2M board.

Right, because that writeext seems to be micrel-phy specific.

author
#5 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

Hi Marek,

Thanks for the feedback.

Quoted 20 lines

Subject: Re: [PATCH] net: ravb: Fix NULL pointer access

Quoted 1 line · Biju Das

Hi Marek,

Hi,

[...]

Quoted 11 lines
Quoted 10 lines

drivers/net/ravb.c static int ravb_config(struct udevice *dev) · line 442

Quoted 8 lines

+++ b/drivers/net/ravb.c

drivers/net/ravb.c +2 -1
@@ -438,7 +438,8 @@ static int ravb_config(struct udevice *dev)
 
 writel(mask, eth->iobase + RAVB_REG_ECMR);
 
-phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
+if (phy->drv->writeext)
+phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);

Shouldn't we rather move this into the PHY driver altogether ?

Quoted 7 lines
Quoted 1 line

If I fix, the phydriver with empty function, compiler will complain about

unused parameter right?

Quoted 1 line

What about other phy's which doesn't have this callback.

I mean, should we not move this entire code which configures something in a Micrel PHY, and is specific to Micrel PHY, into the Micrel PHY driver and remove the whole call of writeext from this ethernet driver ?

Ok , this function is invoked during data transfer. We could remove this function and test on all RCar boards to see any issues present or not? By looking at [1], only this driver is using writeext. [1]https://elixir.bootlin.com/u-boot/v2020.10-rc4/A/ident/writeext

Cheers,
Biju

Renesas Electronics Europe GmbH, Geschaeftsfuehrer/President: Carsten Jauch, Sitz der Gesellschaft/Registered office: Duesseldorf, Arcadiastrasse 10, 40472 Duesseldorf, Germany, Handelsregister/Commercial Register: Duesseldorf, HRB 3708 USt-IDNr./Tax identification no.: DE 119353406 WEEE-Reg.-Nr./WEEE reg. no.: DE 14978647

#6 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

On 9/18/20 5:26 PM, Biju Das wrote:
[...]

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 10 lines

drivers/net/ravb.c static int ravb_config(struct udevice *dev) · line 442

Quoted 8 lines

+++ b/drivers/net/ravb.c

drivers/net/ravb.c +2 -1
@@ -438,7 +438,8 @@ static int ravb_config(struct udevice *dev)
 
 writel(mask, eth->iobase + RAVB_REG_ECMR);
 
-phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
+if (phy->drv->writeext)
+phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);

Shouldn't we rather move this into the PHY driver altogether ?

If I fix, the phydriver with empty function, compiler will complain about

Quoted 8 lines
Quoted 6 lines

unused parameter right?

Quoted 1 line

What about other phy's which doesn't have this callback.

I mean, should we not move this entire code which configures something in a Micrel PHY, and is specific to Micrel PHY, into the Micrel PHY driver and remove the whole call of writeext from this ethernet driver ?

Ok , this function is invoked during data transfer.

The ravb_config() is invoked when starting the interface, see ravb_start(), it is not repeatedly called during data transfer.

Quoted 1 line

We could remove this function and test on all RCar boards to see any issues present or not?

No, because calling the writeext configures something in the Micrel KSZ90x1 PHY, and by removing it, I suspect the Micrel PHY would stop working.

Quoted 2 lines

By looking at [1], only this driver is using writeext. [1]https://elixir.bootlin.com/u-boot/v2020.10-rc4/A/ident/writeext

git grep indicates a couple more sites where the writeext is called. But look into the KSZ9031 datasheet, that particular writeext call seems to be setting up RGMII Clock Pad Skew (MMD Address 2, Register 8), and I think there is a matching DT binding to set those up too, rxc-skew-ps and txc-skew-ps I think.

So I think if you set those two DT properties in rcar2/3 DTs accordingly (and that might already be the case, but please double-check), you can even drop this entire writeext altogether.

[...]

author
#7 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

Hi Marek,

Thanks for the feedback.

Quoted 24 lines

Subject: Re: [PATCH] net: ravb: Fix NULL pointer access

On 9/18/20 5:26 PM, Biju Das wrote:
[...]

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines

drivers/net/ravb.c static int ravb_config(struct udevice *dev) · line 442

Quoted 8 lines

+++ b/drivers/net/ravb.c

drivers/net/ravb.c +2 -1
@@ -438,7 +438,8 @@ static int ravb_config(struct udevice *dev)
 
 writel(mask, eth->iobase + RAVB_REG_ECMR);
 
-phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);
+if (phy->drv->writeext)
+phy->drv->writeext(phy, -1, 0x02, 0x08, (0x0f << 5) | 0x19);

Shouldn't we rather move this into the PHY driver altogether ?

Quoted 8 lines
Quoted 8 lines
Quoted 2 lines

If I fix, the phydriver with empty function, compiler will complain about

unused parameter right?

Quoted 1 line

What about other phy's which doesn't have this callback.

I mean, should we not move this entire code which configures something in a Micrel PHY, and is specific to Micrel PHY, into the Micrel PHY driver and remove the whole call of writeext from this

Quoted 6 lines

ethernet driver ?

Quoted 2 lines

Ok , this function is invoked during data transfer.

The ravb_config() is invoked when starting the interface, see ravb_start(), it is not repeatedly called during data transfer.

Yep you are correct , it will be called during starting of the interface. I have seen the crash when I started doing tftp or ping, which in turn calls ravb_start.

Quoted 6 lines
Quoted 1 line

We could remove this function and test on all RCar boards to see any issues

present or not?

No, because calling the writeext configures something in the Micrel KSZ90x1 PHY, and by removing it, I suspect the Micrel PHY would stop working.

I agree, otherwise it will use default values which may cause issues.

Quoted 8 lines
Quoted 2 lines

By looking at [1], only this driver is using writeext. [1]https://elixir.bootlin.com/u-boot/v2020.10-rc4/A/ident/writeext

git grep indicates a couple more sites where the writeext is called. But look into the KSZ9031 datasheet, that particular writeext call seems to be setting up RGMII Clock Pad Skew (MMD Address 2, Register 8), and I think there is a matching DT binding to set those up too, rxc-skew-ps and txc- skew-ps I think.

Thanks for the pointers. I checked the configs[2] which uses renesas ravb driver and found that we are defining only rxc-skew-ps in all dts.

since CONFIG DM_ETH is defined it is already picking the value corresponding to "rxc-skew-ps".

For txc-skew-ps anyway the value is default one. So we don't care.

Quoted 3 lines

So I think if you set those two DT properties in rcar2/3 DTs accordingly (and that might already be the case, but please double-check), you can even drop this entire writeext altogether.

I did a grep and the below configs enables renesas ravb driver .Only Gen3 uses renesas ravb driver and all of them have "rxc-skew-ps" in Devicetree. So I agree with you, it is safe to remove writephyext. I will post V2, removing this

[2] u-boot/configs/rcar3_ulcb_defconfig:CONFIG_RENESAS_RAVB=y u-boot/configs/r8a77990_ebisu_defconfig:CONFIG_RENESAS_RAVB=y u-boot/configs/r8a774a1_beacon_defconfig:CONFIG_RENESAS_RAVB=y u-boot/configs/hihope_rzg2_defconfig:CONFIG_RENESAS_RAVB=y u-boot/configs/r8a77995_draak_defconfig:CONFIG_RENESAS_RAVB=y u-boot/configs/r8a77970_eagle_defconfig:CONFIG_RENESAS_RAVB=y u-boot/configs/rcar3_salvator-x_defconfig:CONFIG_RENESAS_RAVB=y

Cheers,
Biju

Renesas Electronics Europe GmbH, Geschaeftsfuehrer/President: Carsten Jauch, Sitz der Gesellschaft/Registered office: Duesseldorf, Arcadiastrasse 10, 40472 Duesseldorf, Germany, Handelsregister/Commercial Register: Duesseldorf, HRB 3708 USt-IDNr./Tax identification no.: DE 119353406 WEEE-Reg.-Nr./WEEE reg. no.: DE 14978647

#8 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

On 9/19/20 1:14 PM, Biju Das wrote:
[...]

Quoted 9 lines
Quoted 8 lines
Quoted 2 lines

By looking at [1], only this driver is using writeext. [1]https://elixir.bootlin.com/u-boot/v2020.10-rc4/A/ident/writeext

git grep indicates a couple more sites where the writeext is called. But look into the KSZ9031 datasheet, that particular writeext call seems to be setting up RGMII Clock Pad Skew (MMD Address 2, Register 8), and I think there is a matching DT binding to set those up too, rxc-skew-ps and txc- skew-ps I think.

Quoted 6 lines

Thanks for the pointers. I checked the configs[2] which uses renesas ravb driver and found that we are defining only rxc-skew-ps in all dts.

since CONFIG DM_ETH is defined it is already picking the value corresponding to "rxc-skew-ps".

For txc-skew-ps anyway the value is default one. So we don't care.

Are you sure (0xf << 5) | 0x19 is the same as the default value of the clock pad skew register ?

[...]

author
#9 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

Hi Marek,

Thanks for the feedback.

Quoted 16 lines

Subject: Re: [PATCH] net: ravb: Fix NULL pointer access

On 9/19/20 1:14 PM, Biju Das wrote:
[...]

Quoted 12 lines
Quoted 8 lines
Quoted 2 lines

By looking at [1], only this driver is using writeext. [1]https://elixir.bootlin.com/u-boot/v2020.10-rc4/A/ident/writeext

git grep indicates a couple more sites where the writeext is called. But look into the KSZ9031 datasheet, that particular writeext call seems to be setting up RGMII Clock Pad Skew (MMD Address 2, Register 8), and I think there is a matching DT binding to set those up too, rxc-skew-ps and txc- skew-ps I think.

Thanks for the pointers. I checked the configs[2] which uses renesas ravb driver and found that we are defining only rxc-skew-ps in all dts.

Quoted 7 lines
Quoted 1 line

since CONFIG DM_ETH is defined it is already picking the value

corresponding to "rxc-skew-ps".

Quoted 2 lines

For txc-skew-ps anyway the value is default one. So we don't care.

Are you sure (0xf << 5) | 0x19 is the same as the default value of the clock pad skew register ?

No.
As per [1] & [2], the default values for this registers are 0xf
[1] https://elixir.bootlin.com/u-boot/v2020.10-rc4/source/drivers/net/phy/micrel_ksz90x1.c#L105
[2] http://ww1.microchip.com/downloads/en/devicedoc/00002117f.pdf

if we remove writephyext, by looking the code at [1], rxc-skew-ps will be taken from the device tree[3] and "txc-skew-pc" will be the default value(0xf). [3]https://elixir.bootlin.com/u-boot/v2020.10-rc4/source/arch/arm/dts/salvator-common.dtsi#L331

I will check this and let you know the results after checking on RCar board. Unfortunately currently I don't have RCar board.

Cheers,
Biju

Renesas Electronics Europe GmbH, Geschaeftsfuehrer/President: Carsten Jauch, Sitz der Gesellschaft/Registered office: Duesseldorf, Arcadiastrasse 10, 40472 Duesseldorf, Germany, Handelsregister/Commercial Register: Duesseldorf, HRB 3708 USt-IDNr./Tax identification no.: DE 119353406 WEEE-Reg.-Nr./WEEE reg. no.: DE 14978647

#10 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

On 9/19/20 8:14 PM, Biju Das wrote:

Hi,

[...]

Quoted 15 lines
Quoted 15 lines
Quoted 9 lines
Quoted 8 lines
Quoted 2 lines

By looking at [1], only this driver is using writeext. [1]https://elixir.bootlin.com/u-boot/v2020.10-rc4/A/ident/writeext

git grep indicates a couple more sites where the writeext is called. But look into the KSZ9031 datasheet, that particular writeext call seems to be setting up RGMII Clock Pad Skew (MMD Address 2, Register 8), and I think there is a matching DT binding to set those up too, rxc-skew-ps and txc- skew-ps I think.

Quoted 4 lines

Thanks for the pointers. I checked the configs[2] which uses renesas ravb driver and found that we are defining only rxc-skew-ps in all dts.

since CONFIG DM_ETH is defined it is already picking the value

corresponding to "rxc-skew-ps".

Quoted 1 line
Quoted 7 lines
Quoted 4 lines
Quoted 1 line

For txc-skew-ps anyway the value is default one. So we don't care.

Are you sure (0xf << 5) | 0x19 is the same as the default value of the clock pad skew register ?

No.
As per [1] & [2], the default values for this registers are 0xf

So the combined value of the MMD 2-8 register is (0xf << 5) | (0xf << 0) , right.

Quoted 5 lines

[1] https://elixir.bootlin.com/u-boot/v2020.10-rc4/source/drivers/net/phy/micrel_ksz90x1.c#L105
[2] http://ww1.microchip.com/downloads/en/devicedoc/00002117f.pdf

if we remove writephyext, by looking the code at [1], rxc-skew-ps will be taken from the device tree[3] and "txc-skew-pc" will be the default value(0xf). [3]https://elixir.bootlin.com/u-boot/v2020.10-rc4/source/arch/arm/dts/salvator-common.dtsi#L331

So you want to check whether each RCar3 DT contains a PHY node and that PHY node has rxc-skew-ps and txc-skew-ps , which combined then results into a register value (0xf << 5) | (0x19 << 0) .

Quoted 1 line

I will check this and let you know the results after checking on RCar board. Unfortunately currently I don't have RCar board.

It's enough to just check the DTs and verify that they set the matching correct values of rxc-skew-ps/txc-skew-ps . I can test it on the real hardware.

If you want, you can add the txc-skew-ps into the Linux R-Car3 DTs too.

btw unrelated, you seem to have rxc-skew-ps in your hihope-rzg2-ex.dtsi, but I think you don't have KSZ9031 PHY, so maybe you want to remove it form your DT too.

Thanks

author
#11 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

Hi Marek,

Thanks for the feedback.

Quoted 41 lines

Subject: Re: [PATCH] net: ravb: Fix NULL pointer access

On 9/19/20 8:14 PM, Biju Das wrote:

Hi,

[...]

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 8 lines
Quoted 2 lines

By looking at [1], only this driver is using writeext. [1]https://elixir.bootlin.com/u-boot/v2020.10-rc4/A/ident/writeext

git grep indicates a couple more sites where the writeext is called. But look into the KSZ9031 datasheet, that particular writeext call seems to be setting up RGMII Clock Pad Skew (MMD Address 2, Register 8), and I think there is a matching DT binding to set those up too, rxc-skew-ps and txc- skew-ps I think.

Thanks for the pointers. I checked the configs[2] which uses renesas ravb driver and found that we are defining only rxc-skew-ps in

all dts.

Quoted 11 lines
Quoted 8 lines
Quoted 2 lines

since CONFIG DM_ETH is defined it is already picking the value

corresponding to "rxc-skew-ps".

Quoted 2 lines

For txc-skew-ps anyway the value is default one. So we don't care.

Are you sure (0xf << 5) | 0x19 is the same as the default value of the clock pad skew register ?

No.
As per [1] & [2], the default values for this registers are 0xf

So the combined value of the MMD 2-8 register is (0xf << 5) | (0xf << 0) , right.

Quoted 6 lines

[1]
https://elixir.bootlin.com/u-boot/v2020.10-rc4/source/drivers/net/phy/
micrel_ksz90x1.c#L105 [2]
http://ww1.microchip.com/downloads/en/devicedoc/00002117f.pdf

if we remove writephyext, by looking the code at [1], rxc-skew-ps will be

Quoted 8 lines

taken from the device tree[3] and "txc-skew-pc" will be the default value(0xf).

Quoted 2 lines

[3]https://elixir.bootlin.com/u-boot/v2020.10-rc4/source/arch/arm/dts/ salvator-common.dtsi#L331

So you want to check whether each RCar3 DT contains a PHY node and that PHY node has rxc-skew-ps and txc-skew-ps , which combined then results into a register value (0xf << 5) | (0x19 << 0) .

rxc-skew-ps set in DTS is 1500. 1 step is 60 ps, so 1500/60 = 25 which is 0x19 and this value will be overridden and stored in ofcfg->grp's val[0].

Quoted 5 lines
Quoted 1 line

I will check this and let you know the results after checking on RCar board.

Unfortunately currently I don't have RCar board.

It's enough to just check the DTs and verify that they set the matching correct values of rxc-skew-ps/txc-skew-ps . I can test it on the real hardware.

Yes, that way we can make sure the mapping operation is correct for this phy. 1500 in dts after mapping operation should override ofcfg->grp's val[0] with 0x19.

Quoted 1 line

If you want, you can add the txc-skew-ps into the Linux R-Car3 DTs too.

We don't need to set txc-skew-ps in dts, since it is same as default value and is filled in ofcfg->grp's val[1]. We can avoid unnecessary mapping operations by not specifying in device tree, for default values.

Quoted 3 lines

btw unrelated, you seem to have rxc-skew-ps in your hihope-rzg2-ex.dtsi, but I think you don't have KSZ9031 PHY, so maybe you want to remove it form your DT too.

OK. Thanks for reporting this. Will fix this in kernel as well as here.

Cheers,
Biju

Renesas Electronics Europe GmbH, Geschaeftsfuehrer/President: Carsten Jauch, Sitz der Gesellschaft/Registered office: Duesseldorf, Arcadiastrasse 10, 40472 Duesseldorf, Germany, Handelsregister/Commercial Register: Duesseldorf, HRB 3708 USt-IDNr./Tax identification no.: DE 119353406 WEEE-Reg.-Nr./WEEE reg. no.: DE 14978647

#12 lore raw

PATCH 1/1 net: ravb: Fix NULL pointer access

On 9/20/20 9:34 AM, Biju Das wrote:

Hi,

[...]

Quoted 11 lines
Quoted 9 lines
Quoted 1 line

if we remove writephyext, by looking the code at [1], rxc-skew-ps will be

taken from the device tree[3] and "txc-skew-pc" will be the default value(0xf).

Quoted 2 lines

[3]https://elixir.bootlin.com/u-boot/v2020.10-rc4/source/arch/arm/dts/ salvator-common.dtsi#L331

So you want to check whether each RCar3 DT contains a PHY node and that PHY node has rxc-skew-ps and txc-skew-ps , which combined then results into a register value (0xf << 5) | (0x19 << 0) .

rxc-skew-ps set in DTS is 1500. 1 step is 60 ps, so 1500/60 = 25 which is 0x19 and this value will be overridden and stored in ofcfg->grp's val[0].

OK, good, thanks.

Quoted 6 lines
Quoted 5 lines
Quoted 1 line

I will check this and let you know the results after checking on RCar board.

Unfortunately currently I don't have RCar board.

It's enough to just check the DTs and verify that they set the matching correct values of rxc-skew-ps/txc-skew-ps . I can test it on the real hardware.

Quoted 7 lines

Yes, that way we can make sure the mapping operation is correct for this phy. 1500 in dts after mapping operation should override ofcfg->grp's val[0] with 0x19.

Quoted 1 line

If you want, you can add the txc-skew-ps into the Linux R-Car3 DTs too.

We don't need to set txc-skew-ps in dts, since it is same as default value and is filled in ofcfg->grp's val[1]. We can avoid unnecessary mapping operations by not specifying in device tree, for default values.

Hopefully the default setting will not change in some new revision of the PHY then.