Skip to contentSkip to conversation

net: ethernet: xilinx_emaclite: keep protocol selector bits by reading ANAR

PATCH In review Jens Renner (EFE) 3 participants 8 messages May 28, 2013 – May 29, 2013

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net: ethernet: xilinx_emaclite: keep protocol selector bits by reading ANAR

This patch reads the PHY's MII_ADVERTISE register (ANAR) before modifying it. Hence, the protocol selector bits (4:0) which indicate IEEE 803.3u support are prevented from being cleared. While the selector bits are fixed / read-only on some PHYs, not setting them correctly on others (like TI DP83630) makes the PHY fall back to 10/half mode which should be avoided.

---
drivers/net/ethernet/xilinx/xilinx_emaclite.c | 7 +++++--

 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/net/ethernet/xilinx/xilinx_emaclite.c +5 -2
index 919b983..4a67af1 100644
@@ -922,7 +922,7 @@ void xemaclite_adjust_link(struct net_device *ndev)
 static int xemaclite_open(struct net_device *dev)
 {
        struct net_local *lp = netdev_priv(dev);
-       int retval;
+       int retval, adv;
 
        /* Just to be safe, stop the device first */
        xemaclite_disable_interrupts(lp);
@@ -946,7 +946,10 @@ static int xemaclite_open(struct net_device *dev)
                phy_write(lp->phy_dev, MII_CTRL1000, 0);
 
                /* Advertise only 10 and 100mbps full/half duplex speeds */
-               phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL);
+               adv = phy_read(lp->phy_dev, MII_ADVERTISE);
+               if (adv < 0)
+                       return adv;
+               phy_write(lp->phy_dev, MII_ADVERTISE, adv | ADVERTISE_ALL);
 
                /* Restart auto negotiation */
                bmcr = phy_read(lp->phy_dev, MII_BMCR);
#1 lore raw

drivers/net/ethernet/xilinx/xilinx_emaclite.c static int xemaclite_open(struct net_device *dev) · line 952

Quoted 10 lines

This patch reads the PHY's MII_ADVERTISE register (ANAR) before modifying it. Hence, the protocol selector bits (4:0) which indicate IEEE 803.3u support are prevented from being cleared.
...

                /* Advertise only 10 and 100mbps full/half duplex speeds */
drivers/net/ethernet/xilinx/xilinx_emaclite.c +4 -1
-               phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL);
+               adv = phy_read(lp->phy_dev, MII_ADVERTISE);
+               if (adv < 0)
+                       return adv;
+               phy_write(lp->phy_dev, MII_ADVERTISE, adv | ADVERTISE_ALL);

Given the comment shouldn't this be removing some bits?

It is a long time since I read the definitions of the ANAR (etc). But I remember it having at least 5 bits defined for 10M and 100M operation (including 100T4), so the reference to keeping bits (4:0) seems confusing.

David

#2 lore raw
Quoted 17 lines

David Laight <[email protected]> hat am 28. Mai 2013 um 18:22 geschrieben:

drivers/net/ethernet/xilinx/xilinx_emaclite.c static int xemaclite_open(struct net_device *dev) · line 952

Quoted 12 lines

This patch reads the PHY's MII_ADVERTISE register (ANAR) before modifying it.
Hence, the protocol selector bits (4:0) which indicate IEEE 803.3u support are
prevented from being cleared.
...

                /* Advertise only 10 and 100mbps full/half duplex speeds */
drivers/net/ethernet/xilinx/xilinx_emaclite.c +4 -1
-               phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL);
+               adv = phy_read(lp->phy_dev, MII_ADVERTISE);
+               if (adv < 0)
+                       return adv;
+               phy_write(lp->phy_dev, MII_ADVERTISE, adv | ADVERTISE_ALL);
Quoted 9 lines

Given the comment shouldn't this be removing some bits?

It is a long time since I read the definitions of the ANAR (etc). But I remember it having at least 5 bits defined for 10M and 100M operation (including 100T4), so the reference to keeping bits (4:0) seems confusing.

David

David, thanks for your feedback!
The bits I mentioned (the 5 least bits in ANAR) are the so-called protocol selection bits (at least that's what they call it at Intel, TI, Marvell, Maxim, etc.). The only value that seems actually to be used is 1 (0b00001) which means 802.3 / fast ethernet capability (maybe 0 for slower devices?). 1 is the default value for all the 10/100(/1000) PHYs I know of, in some PHYs it is even hardcoded (i.e. read-only). It must not be confused with ANAR bits 9:5 where the actual speed and duplex mode can be selected.
ADVERTISE_ALL as defined in mii.h lacks this bit, in contrast to ADVERTISE_FULL where it is set through ADVERTISE_CSMA. Whether this is correct or not; at least it makes some sense to me to read what's actually written in these register bits and not to clear them. Otherwise some PHYs might fall back to 10-half as seen for the TI DP83640 which has been used to test this patch. I hope this explains the reason for this patch.

Jens.

#3 lore raw
Quoted 27 lines
Quoted 17 lines

David Laight <[email protected]> hat am 28. Mai 2013 um 18:22 geschrieben:

drivers/net/ethernet/xilinx/xilinx_emaclite.c static int xemaclite_open(struct net_device *dev) · line 952

Quoted 12 lines

This patch reads the PHY's MII_ADVERTISE register (ANAR) before modifying it.
Hence, the protocol selector bits (4:0) which indicate IEEE 803.3u support are
prevented from being cleared.
...

                /* Advertise only 10 and 100mbps full/half duplex speeds */
drivers/net/ethernet/xilinx/xilinx_emaclite.c +4 -1
-               phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL);
+               adv = phy_read(lp->phy_dev, MII_ADVERTISE);
+               if (adv < 0)
+                       return adv;
+               phy_write(lp->phy_dev, MII_ADVERTISE, adv | ADVERTISE_ALL);
Quoted 9 lines

Given the comment shouldn't this be removing some bits?

It is a long time since I read the definitions of the ANAR (etc). But I remember it having at least 5 bits defined for 10M and 100M operation (including 100T4), so the reference to keeping bits (4:0) seems confusing.

David

Quoted 14 lines

David, thanks for your feedback!
The bits I mentioned (the 5 least bits in ANAR) are the so-called protocol selection bits (at least that's what they call it at Intel, TI, Marvell, Maxim, etc.). The only value that seems actually to be used is 1 (0b00001) which means 802.3 / fast ethernet capability (maybe 0 for slower devices?). 1 is the default value for all the 10/100(/1000) PHYs I know of, in some PHYs it is even hardcoded (i.e. read-only). It must not be confused with ANAR bits 9:5 where the actual speed and duplex mode can be selected.
ADVERTISE_ALL as defined in mii.h lacks this bit, in contrast to ADVERTISE_FULL where it is set through ADVERTISE_CSMA. Whether this is correct or not; at least it makes some sense to me to read what's actually written in these register bits and not to clear them. Otherwise some PHYs might fall back to 10-half as seen for the TI DP83640 which has been used to test this patch. I hope this explains the reason for this patch.

Part of my confusion was that I'd forgotten that the MII registers are 16bit! (I said it had been a long time!)

However if you want to advertise 10M/100M HDX and FDX you need to unset the other speed bits (100T4 was assigned a bit, and I'm sure there were extra bits reserved for higher speeds - which also need clearing). Given that the ANAR is defined by 803.x you shouldn't really have a problem!

My real bugbear is that none of the standard MII registers actually tells you the mode the PHY is working in! Not sure how that got through the standards body.

David

#4 lore raw

Am 29.05.2013 10:51, schrieb David Laight:

Quoted 42 lines
Quoted 41 lines
Quoted 17 lines

David Laight <[email protected]> hat am 28. Mai 2013 um 18:22 geschrieben:

drivers/net/ethernet/xilinx/xilinx_emaclite.c static int xemaclite_open(struct net_device *dev) · line 952

Quoted 12 lines

This patch reads the PHY's MII_ADVERTISE register (ANAR) before modifying it.
Hence, the protocol selector bits (4:0) which indicate IEEE 803.3u support are
prevented from being cleared.
...

                /* Advertise only 10 and 100mbps full/half duplex speeds */
drivers/net/ethernet/xilinx/xilinx_emaclite.c +4 -1
-               phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL);
+               adv = phy_read(lp->phy_dev, MII_ADVERTISE);
+               if (adv < 0)
+                       return adv;
+               phy_write(lp->phy_dev, MII_ADVERTISE, adv | ADVERTISE_ALL);
Quoted 9 lines

Given the comment shouldn't this be removing some bits?

It is a long time since I read the definitions of the ANAR (etc). But I remember it having at least 5 bits defined for 10M and 100M operation (including 100T4), so the reference to keeping bits (4:0) seems confusing.

David

David, thanks for your feedback!
The bits I mentioned (the 5 least bits in ANAR) are the so-called protocol selection bits (at least that's what they call it at Intel, TI, Marvell, Maxim, etc.). The only value that seems actually to be used is 1 (0b00001) which means 802.3 / fast ethernet capability (maybe 0 for slower devices?). 1 is the default value for all the 10/100(/1000) PHYs I know of, in some PHYs it is even hardcoded (i.e. read-only). It must not be confused with ANAR bits 9:5 where the actual speed and duplex mode can be selected.
ADVERTISE_ALL as defined in mii.h lacks this bit, in contrast to ADVERTISE_FULL where it is set through ADVERTISE_CSMA. Whether this is correct or not; at least it makes some sense to me to read what's actually written in these register bits and not to clear them. Otherwise some PHYs might fall back to 10-half as seen for the TI DP83640 which has been used to test this patch. I hope this explains the reason for this patch.

[...]

Quoted 4 lines

However if you want to advertise 10M/100M HDX and FDX you need to unset the other speed bits (100T4 was assigned a bit, and I'm sure there were extra bits reserved for higher speeds - which also need clearing). Given that the ANAR is defined by 803.x you shouldn't really have a problem!

[...]

You are right. The 1000BASE-T capability is advertised in another register and already gets cleared one code line before (see original list mail), while the 100BASE-T4 bit now remains unchanged with my patch which is not intended. Consequently the new patch would look like this:
- phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL);
+ phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL |
+ ADVERTISE_CSMA);
This only advertises HDX & FDX for both 10M / 100M, sets the 802.3 selector bits as defined in ADVERTISE_CSMA, and clears all other bits which should be safe. It is done a similar way in some other drivers.
I will prepare a patch resend then.

Jens.

Acked-by on v1 1/1
#5 lore raw
In reply to Jens Renner (EFE)
Quoted 26 lines · Jens Renner (EFE)

This patch reads the PHY's MII_ADVERTISE register (ANAR) before modifying it. Hence, the protocol selector bits (4:0) which indicate IEEE 803.3u support are prevented from being cleared. While the selector bits are fixed / read-only on some PHYs, not setting them correctly on others (like TI DP83630) makes the PHY fall back to 10/half mode which should be avoided.

---
drivers/net/ethernet/xilinx/xilinx_emaclite.c | 7 +++++--

 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/net/ethernet/xilinx/xilinx_emaclite.c +1 -1
index 919b983..4a67af1 100644
@@ -922,7 +922,7 @@ void xemaclite_adjust_link(struct net_device *ndev)
 static int xemaclite_open(struct net_device *dev)
 {
        struct net_local *lp = netdev_priv(dev);
-       int retval;
+       int retval, adv;
 
        /* Just to be safe, stop the device first */
        xemaclite_disable_interrupts(lp);
@@ -946,7 +946,10 @@ static int xemaclite_open(struct net_device *dev)
                phy_write(lp->phy_dev, MII_CTRL1000, 0);

drivers/net/ethernet/xilinx/xilinx_emaclite.c static int xemaclite_open(struct net_device *dev) · line 955

Quoted 10 lines
drivers/net/ethernet/xilinx/xilinx_emaclite.c +4 -1
 
                /* Advertise only 10 and 100mbps full/half duplex speeds */
-               phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL);
+               adv = phy_read(lp->phy_dev, MII_ADVERTISE);
+               if (adv < 0)
+                       return adv;
+               phy_write(lp->phy_dev, MII_ADVERTISE, adv | ADVERTISE_ALL);
 
                /* Restart auto negotiation */
                bmcr = phy_read(lp->phy_dev, MII_BMCR);

Thanks,
Michal

Signature · 5 lines
Michal Simek, Ing. (M.Eng), OpenPGP -> KeyID: FE3D1F91
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/
Maintainer of Linux kernel - Xilinx Zynq ARM architecture
Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
#6 lore raw

Am 29.05.2013 17:23, schrieb Michal Simek:

Quoted 31 lines · Jens Renner (EFE)
Quoted 30 lines

This patch reads the PHY's MII_ADVERTISE register (ANAR) before modifying it. Hence, the protocol selector bits (4:0) which indicate IEEE 803.3u support are prevented from being cleared. While the selector bits are fixed / read-only on some PHYs, not setting them correctly on others (like TI DP83630) makes the PHY fall back to 10/half mode which should be avoided.

drivers/net/ethernet/xilinx/xilinx_emaclite.c | 7
+++++-- 1 file changed, 5 insertions(+), 2 deletions(-)

(quoted diff) +0 -0

b/drivers/net/ethernet/xilinx/xilinx_emaclite.c index 919b983..4a67af1 100644 --- a/drivers/net/ethernet/xilinx/xilinx_emaclite.c +++ b/drivers/net/ethernet/xilinx/xilinx_emaclite.c @@ -922,7 +922,7 @@ void xemaclite_adjust_link(struct net_device *ndev) static int xemaclite_open(struct net_device *dev) { struct net_local *lp =
netdev_priv(dev); - int retval; + int retval, adv;

/* Just to be safe, stop the device first */ xemaclite_disable_interrupts(lp); @@ -946,7 +946,10 @@ static int xemaclite_open(struct net_device *dev) phy_write(lp->phy_dev, MII_CTRL1000, 0);

/* Advertise only 10 and 100mbps full/half duplex speeds */ -
phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL); + adv
= phy_read(lp->phy_dev, MII_ADVERTISE); + if (adv < 0) +
return adv; + phy_write(lp->phy_dev, MII_ADVERTISE, adv |
ADVERTISE_ALL);

Quoted 7 lines
Quoted 1 line

/* Restart auto negotiation */ bmcr = phy_read(lp->phy_dev, MII_BMCR);

Thanks, Michal

Michal, please mind the follow-ups to my original list mail:
http://lists.openwall.net/netdev/2013/05/28/126
http://lists.openwall.net/netdev/2013/05/29/140
There will be a patch v2 later this day (I will put you in CC). Thanks,
Jens.

#7 lore raw
Quoted 36 lines · Jens Renner (EFE)

Am 29.05.2013 17:23, schrieb Michal Simek:

Quoted 35 lines · Jens Renner (EFE)
Quoted 24 lines

This patch reads the PHY's MII_ADVERTISE register (ANAR) before modifying it. Hence, the protocol selector bits (4:0) which indicate IEEE 803.3u support are prevented from being cleared. While the selector bits are fixed / read-only on some PHYs, not setting them correctly on others (like TI DP83630) makes the PHY fall back to 10/half mode which should be avoided.

drivers/net/ethernet/xilinx/xilinx_emaclite.c | 7
+++++-- 1 file changed, 5 insertions(+), 2 deletions(-)

(quoted diff) +0 -0

b/drivers/net/ethernet/xilinx/xilinx_emaclite.c index 919b983..4a67af1 100644 --- a/drivers/net/ethernet/xilinx/xilinx_emaclite.c +++ b/drivers/net/ethernet/xilinx/xilinx_emaclite.c @@ -922,7 +922,7 @@ void xemaclite_adjust_link(struct net_device *ndev) static int xemaclite_open(struct net_device *dev) { struct net_local *lp =
netdev_priv(dev); - int retval; + int retval, adv;

/* Just to be safe, stop the device first */ xemaclite_disable_interrupts(lp); @@ -946,7 +946,10 @@ static int xemaclite_open(struct net_device *dev) phy_write(lp->phy_dev, MII_CTRL1000, 0);

Quoted 7 lines

/* Advertise only 10 and 100mbps full/half duplex speeds */ -
phy_write(lp->phy_dev, MII_ADVERTISE, ADVERTISE_ALL); + adv
= phy_read(lp->phy_dev, MII_ADVERTISE); + if (adv < 0) +
return adv; + phy_write(lp->phy_dev, MII_ADVERTISE, adv |
ADVERTISE_ALL);

/* Restart auto negotiation */ bmcr = phy_read(lp->phy_dev, MII_BMCR);

Quoted 8 lines
Quoted 3 lines

Thanks, Michal

Michal, please mind the follow-ups to my original list mail:
http://lists.openwall.net/netdev/2013/05/28/126
http://lists.openwall.net/netdev/2013/05/29/140
There will be a patch v2 later this day (I will put you in CC).

Ok. I have tested it on xilinx qemu model and your patch doesn't break our existing driver usage that's why I gave you my ACK. And yes, please CC me I will retest your v2.

Thanks,
Michal

Signature · 5 lines
Michal Simek, Ing. (M.Eng), OpenPGP -> KeyID: FE3D1F91
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/
Maintainer of Linux kernel - Xilinx Zynq ARM architecture
Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform