New: Marvell 88E1145 phy configured incorrectly in fiber mode
Subsystems (3)
(switched to email. Please respond via emailed reply-to-all, not via the bugzilla web interface).
Quoted 33 lines · On Mon, 11 Apr 2011 09:28:18 GMT [email protected] wrote:
https://bugzilla.kernel.org/show_bug.cgi?id=33042
Summary: Marvell 88E1145 phy configured incorrectly in fiber mode Product: Drivers Version: 2.5 Kernel Version: Linux-2.6.39-rc1-00191 Platform: All OS/Version: Linux Tree: Mainline Status: NEW Severity: normal Priority: P1 Component: Network AssignedTo: [email protected] ReportedBy: [email protected] Regression: NoOn my board, Marvell 88E1145 phy is attached to Freescale gianfar controller. Backplane connection is detected by u-boot as 1000/Full fiber mode.
The network works perfectly in u-boot (dhcp, tftp of large files, ntp). Upon booting, kernel detects the link as 100/Full and no data can be exchanged over the interface.
As a quick fix, I tried setting .read_settings method of the 1145 driver to
marvell_read_status(instead ofgenphy_read_status). Now the link is correctly detected as 1000/Full, but the data still can not be exchanged.I assume, there's a configuration bit missing on the kernel side, as there are no such problems in u-boot.
Quoted 40 lines · Andrew Morton
(switched to email. Please respond via emailed reply-to-all, not via the bugzilla web interface).
Quoted 33 lines · On Mon, 11 Apr 2011 09:28:18 GMT [email protected] wrote:
https://bugzilla.kernel.org/show_bug.cgi?id=33042
Summary: Marvell 88E1145 phy configured incorrectly in fiber mode Product: Drivers Version: 2.5 Kernel Version: Linux-2.6.39-rc1-00191 Platform: All OS/Version: Linux Tree: Mainline Status: NEW Severity: normal Priority: P1 Component: Network AssignedTo: [email protected] ReportedBy: [email protected] Regression: NoOn my board, Marvell 88E1145 phy is attached to Freescale gianfar controller. Backplane connection is detected by u-boot as 1000/Full fiber mode.
The network works perfectly in u-boot (dhcp, tftp of large files, ntp). Upon booting, kernel detects the link as 100/Full and no data can be exchanged over the interface.
As a quick fix, I tried setting .read_settings method of the 1145 driver to
marvell_read_status(instead ofgenphy_read_status). Now the link is correctly detected as 1000/Full, but the data still can not be exchanged.I assume, there's a configuration bit missing on the kernel side, as there are no such problems in u-boot.
How does your u-boot configure the part? Does it write any of the configuration registers, or is it just the default configuration set via the strapping pins?
In any event, you will probably have to read the configuration before the drivers/net/phy/marvel.c changes them. Then compare that to what the driver is trying to set. Then you will either have to override the configuration with the device tree "marvell,reg-init" property, or if you are not using the device tree, add a 88e1145 specific flag that you set when calling phy_connect().
David Daney
Quoted 6 lines
How does your u-boot configure the part? Does it write any of the
configuration registers, or is it just the default configuration set via
the strapping pins?
U-boot configures this phy just like any other phy - by running a set of register assignments from phy_info_M88E1145.
Unfortunately, I don't have a datasheet for this phy and kernel does quite a few things differently, so simply copying stuff from u-boot does not work well (in kernel, phy initialization is broken into 3 functions, if I'm not mistaken).
Otherwise, my problem seems to be identical to the one reported some time ago against 88E1111 phy (which resulted in the addition of "marvell_read_status" in the first place). The problem was, as it seems to be now, that phy is always configured in "copper" mode, instead of driver checking for the correct "fiber" mode bits.
Quoted 15 lines
In any event, you will probably have to read the configuration before
thedrivers/net/phy/marvel.cchanges them. Then compare that to what
the driver is trying to set. Then you will either have to override the
configuration with the device tree "marvell,reg-init" property, or if
you are not using the device tree, add a 88e1145 specific flag that you
set when callingphy_connect().David Daney
Quoted 11 lines · Alex Dubov
Quoted 6 lines
How does your u-boot configure the part? Does it write any of the
configuration registers, or is it just the default configuration set via
the strapping pins?U-boot configures this phy just like any other phy - by running a set of register assignments from
phy_info_M88E1145.Unfortunately, I don't have a datasheet for this phy
I would guess that the people who designed your board have it.
You do seem to have the uboot code though, so you know which registers are being set. Given this, you can fiddle around with the Linux driver until it works. Then send a patch.
To me it looks like you need to set register 22 (Page Select) to the value of 1 to access the register sets associated with fiber. I don't have any hardware with this PHY connected to fiber, so I can't really test it.
David Daney
Quoted 27 lines
and kernel does
quite a few things differently, so simply copying stuff from u-boot does not work well (in kernel, phy initialization is broken into 3 functions, if I'm not mistaken).Otherwise, my problem seems to be identical to the one reported some time ago against 88E1111 phy (which resulted in the addition of "
marvell_read_status" in the first place). The problem was, as it seems to be now, that phy is always configured in "copper" mode, instead of driver checking for the correct "fiber" mode bits.Quoted 15 lines
In any event, you will probably have to read the configuration before
thedrivers/net/phy/marvel.cchanges them. Then compare that to what
the driver is trying to set. Then you will either have to override the
configuration with the device tree "marvell,reg-init" property, or if
you are not using the device tree, add a 88e1145 specific flag that you
set when callingphy_connect().David Daney
Quoted 7 lines · Alex Dubov
Quoted 6 lines
How does your u-boot configure the part? Does it write any of the
configuration registers, or is it just the default configuration set via
the strapping pins?
Quoted 7 lines
U-boot configures this phy just like any other phy - by running a set of register assignments from
phy_info_M88E1145.Unfortunately, I don't have a datasheet for this phy and kernel does quite a few things differently, so simply copying stuff from u-boot does not work well (in kernel, phy initialization is broken into 3 functions, if I'm not mistaken).
I've just rewritten the U-Boot code for PHY management, so I'd be interested in hearing if this breaks your board. But what's interesting to me is that, in order for U-Boot to report that the link is a "fiber" link, something had to set the TSEC_FIBER flag, and only one PHY in the public source did. This implies to me that your board isn't supported by mainline U-Boot, and suggests that someone may have modified the 88e1145 driver. Otherwise, I don't see any fiber-related differences between the U-Boot 1145 driver, and the Linux one.
Quoted 14 lines
Quoted 14 lines
In any event, you will probably have to read the configuration before
thedrivers/net/phy/marvel.cchanges them. Then compare that to what
the driver is trying to set. Then you will either have to override the
configuration with the device tree "marvell,reg-init" property, or if
you are not using the device tree, add a 88e1145 specific flag that you
set when callingphy_connect().
Reading the configuration from U-Boot is straightforward. use the "mii" command to read the registers. But don't forget to set register 22 (16 - mii command only reads hex) to 1, and read all of the registers that way, too.
You will either need to add some code to detect when the PHY is using fiber, and change the phydev->port to PORT_FIBRE, or you will need to add a board-level "fixup" to change the port to PORT_FIBRE on your board.
Then the PHY driver should use that information to do the right configuration.
Andy
Quoted 17 lines · Andy Fleming
I've just rewritten the U-Boot code for PHY management, so I'd be
interested in hearing if this breaks your board. But what's
interesting to me is that, in order for U-Boot to report that the link
is a "fiber" link, something had to set theTSEC_FIBERflag, and only
one PHY in the public source did. This implies to me that your board
isn't supported by mainline U-Boot, and suggests that someone may have
modified the 88e1145 driver. Otherwise, I don't see any fiber-related
differences between the U-Boot 1145 driver, and the Linux one.
I had not seen any difference, that's true. But the problem somehow creeps in.
The u-boot is standard stock u-boot pulled from the recent git, no special configuration involved.
I actually managed to make kernel transmit stuff by playing with register values from other marvell phy varieties, but it keeps receiving garbage, so the link is still not operational.
I tried to prevent kernel from reconfiguring the phy, but to no avail. It seems very weird to me, because I did quite a lot of testing with u-boot and network just works on that interface. However, when kernel starts booting it suddenly looses the ability to talk to it.
I have a copper link attached to the same transceiver and it works fine all along.
Quoted 10 lines
Reading the configuration from U-Boot is straightforward. use the
"mii" command to read the registers. But don't forget to set register
22 (16 - mii command only reads hex) to 1, and read all of the
registers that way, too.
I have no recourse but to keep investigating.
Quoted 14 lines
You will either need to add some code to detect when the PHY is using
fiber, and change thephydev->porttoPORT_FIBRE, or you will need to
add a board-level "fixup" to change the port toPORT_FIBREon your
board.Then the PHY driver should use that information to do the right configuration.
Andy
Quoted 22 lines · Alex Dubov
Quoted 17 lines · Andy Fleming
I've just rewritten the U-Boot code for PHY management, so I'd be
interested in hearing if this breaks your board. But what's
interesting to me is that, in order for U-Boot to report that the link
is a "fiber" link, something had to set theTSEC_FIBERflag, and only
one PHY in the public source did. This implies to me that your board
isn't supported by mainline U-Boot, and suggests that someone may have
modified the 88e1145 driver. Otherwise, I don't see any fiber-related
differences between the U-Boot 1145 driver, and the Linux one.
Quoted 5 lines
I had not seen any difference, that's true. But the problem somehow creeps in.
The u-boot is standard stock u-boot pulled from the recent git, no special configuration involved.
Are you seeing this message when you run ethernet in u-boot?
"Speed: 1000, full duplex, fiber mode"
Because that last part only shows up if someone sets TSEC_FIBER in the tsec's "flags" field...
Quoted 4 lines
I tried to prevent kernel from reconfiguring the phy, but to no avail. It seems very weird to me, because I did quite a lot of testing with u-boot and network just works on that interface. However, when kernel starts booting it suddenly looses the ability to talk to it.
Believe me, I feel your pain. These devices are often remarkably fickle. The kernel tries to be
more robust, but sometimes the PHYs just don't like to be touched at all.
You could probably change to use a fixed link by removing the phy-handle property from your ethernet device node, and adding:
"fixed-link=<0 1000 1 0 0>". If that works, then the issue is that Linux is breaking something when it connects. It might be good enough for you to use fixed-link, though it would be good to actually find out what's going wrong with the PHY driver.
Andy
Quoted 7 lines
Quoted 2 lines
--- On Thu, 14/4/11, Andy Fleming <[email protected]>
Quoted 4 lines · wrote:
Quoted 2 lines
The u-boot is standard stock u-boot pulled from the
Quoted 8 lines
recent git,
Quoted 1 line
no special configuration involved.
Are you seeing this message when you run ethernet in u-boot?
"Speed: 1000, full duplex, fiber mode"
I have this weird problem whereupon I assume something and then keep believing it's true. So not, I'm not seeing this message. :)
There are two interfaces attached to that phy, one is normal RJ45, the other is an uTCA backplane (it has only two pairs, so I assumed it must be operating in fiber mode). Uboot works perfectly well with both and prints the same messages.
That's the example printout from u-boot (eTSEC1 is connected to backplane)
Quoted 1 line
dhcp
Speed: 1000, full duplex
BOOTP broadcast 1
DHCP client bound to address 192.168.1.202
Using eTSEC1 device
Quoted 13 lines
You could probably change to use a fixed link by removing the
phy-handle property from your ethernet device node, and adding:
"fixed-link=<0 1000 1 0 0>". If that works, then the issue is that
Linux is breaking something when it connects. It might be good enough
for you to use fixed-link, though it would be good to actually find
out what's going wrong with the PHY driver.
I'll try this out now.
I'm trying to obtain a data sheet for the device, but had not succeeded in it just yet.
Quoted 13 lines · Andy Fleming
You could probably change to use a fixed link by removing the
phy-handle property from your ethernet device node, and adding:
"fixed-link=<0 1000 1 0 0>". If that works, then the issue is that
Linux is breaking something when it connects. It might be good enough
for you to use fixed-link, though it would be good to actually find
out what's going wrong with the PHY driver.
It seems, I found my problem. On the linux side it purely amount to this:
drivers/net/phy/marvell.c +1 -1
@@ -803,7 +803,7 @@ static struct phy_driver marvell_drivers[] = { .flags = PHY_HAS_INTERRUPT, .config_init = &m88e1145_config_init, .config_aneg = &marvell_config_aneg, - .read_status = &genphy_read_status, + .read_status = &marvell_read_status, .ack_interrupt = &marvell_ack_interrupt, .config_intr = &marvell_config_intr, .driver = { .owner = THIS_MODULE },
But then I managed to accidentally put an incorrect value of the phy interrupt into dts file and it caused all kinds of weird problems.
Shall I send a one-liner patch for the above?