Skip to contentSkip to conversation

ETIMEOUT with ehci_setup()'s ehci_halt()

ehci-orion Valentin Longchamp 3 participants 7 messages Jun 5, 2015 – Jun 18, 2015

lore
Share & tools
lore
#0 lore raw

Hello,

I am currently bringing up the USB 2.0 Host controller of the Bobcat's (98DX4122 Marvell switch) internal kirkwood CPU on a variation of Keymile's km_kirwood hardware (kirkwood-km_kirkwood.dts).

When the driver registers its hcd, it fails with a timemout as it can be seen in the below log:

Quoted 6 lines

root at kmcoge5un:~# dmesg -n 8
root at kmcoge5un:~# insmod ehci-orion.ko
ehci-orion: EHCI orion driver
Initializing Orion-SoC USB Host Controller
orion-ehci f1050000.ehci: EHCI Host Controller
orion-ehci f1050000.ehci: new USB bus registered, assigned bus number 1

Quoted 8 lines

orion-ehci f1050000.ehci: reset hcs_params 0x10011 dbg=0 ind cc=0 pcc=0 ordered ports=1 orion-ehci f1050000.ehci: reset hcc_params 0006 thresh 0 uframes 256/512/1024 park orion-ehci f1050000.ehci: park 0
orion-ehci f1050000.ehci: reset command 0080002 (park)=0 ithresh=8 period=1024 Reset HALT orion-ehci f1050000.ehci: can't setup
orion-ehci f1050000.ehci: USB bus 1 deregistered
orion-ehci f1050000.ehci: init f1050000.ehci fail, -110
orion-ehci: probe of f1050000.ehci failed with error -110

This is very similar to this problem, except that it always happens:
https://lkml.org/lkml/2012/6/4/381

I have cornered it out to the last handshake() call of the ehci_halt() call, that should set the controller in the HALT state. If I comment this handshake() call out, the registration is fine and the controller then looks OK (I don't have a hardware with a real USB bus to further test it yet).

My current idea is that maybe a reset or an initialization may be missing, but I wanted to ask if anybody already has seen a similar behavior with the various hardware platform supported by ehci-orion ?

Best regards

Valentin

#1 lore raw
Quoted 27 lines · Valentin Longchamp

Hello,

I am currently bringing up the USB 2.0 Host controller of the Bobcat's (98DX4122 Marvell switch) internal kirkwood CPU on a variation of Keymile's km_kirwood hardware (kirkwood-km_kirkwood.dts).

When the driver registers its hcd, it fails with a timemout as it can be seen in the below log:

Quoted 6 lines

root at kmcoge5un:~# dmesg -n 8
root at kmcoge5un:~# insmod ehci-orion.ko
ehci-orion: EHCI orion driver
Initializing Orion-SoC USB Host Controller
orion-ehci f1050000.ehci: EHCI Host Controller
orion-ehci f1050000.ehci: new USB bus registered, assigned bus number 1

Quoted 8 lines

orion-ehci f1050000.ehci: reset hcs_params 0x10011 dbg=0 ind cc=0 pcc=0 ordered ports=1 orion-ehci f1050000.ehci: reset hcc_params 0006 thresh 0 uframes 256/512/1024 park orion-ehci f1050000.ehci: park 0
orion-ehci f1050000.ehci: reset command 0080002 (park)=0 ithresh=8 period=1024 Reset HALT orion-ehci f1050000.ehci: can't setup
orion-ehci f1050000.ehci: USB bus 1 deregistered
orion-ehci f1050000.ehci: init f1050000.ehci fail, -110
orion-ehci: probe of f1050000.ehci failed with error -110

This is very similar to this problem, except that it always happens:
https://lkml.org/lkml/2012/6/4/381

Quoted 8 lines

I have cornered it out to the last handshake() call of the ehci_halt() call, that should set the controller in the HALT state. If I comment this handshake() call out, the registration is fine and the controller then looks OK (I don't have a hardware with a real USB bus to further test it yet).

My current idea is that maybe a reset or an initialization may be missing, but I wanted to ask if anybody already has seen a similar behavior with the various hardware platform supported by ehci-orion ?

Hi Valentin

I don't remember seeing any problems like this on any kirkwood devices.

Do you know what phy the 98DX4122 uses? The ehci-orion.c has some code for handling the Orion5X USB phy. I also think u-boot might also have some code. Sebastian might know more, i think he implemented USB support for barebox.

Andrew

#2 lore raw
Quoted 23 lines · Andrew Lunn
Quoted 22 lines · Valentin Longchamp

I am currently bringing up the USB 2.0 Host controller of the Bobcat's (98DX4122 Marvell switch) internal kirkwood CPU on a variation of Keymile's km_kirwood hardware (kirkwood-km_kirkwood.dts).

When the driver registers its hcd, it fails with a timemout as it can be seen in the below log:

Quoted 14 lines

root at kmcoge5un:~# dmesg -n 8
root at kmcoge5un:~# insmod ehci-orion.ko
ehci-orion: EHCI orion driver
Initializing Orion-SoC USB Host Controller
orion-ehci f1050000.ehci: EHCI Host Controller
orion-ehci f1050000.ehci: new USB bus registered, assigned bus number 1 orion-ehci f1050000.ehci: reset hcs_params 0x10011 dbg=0 ind cc=0 pcc=0 ordered ports=1 orion-ehci f1050000.ehci: reset hcc_params 0006 thresh 0 uframes 256/512/1024 park orion-ehci f1050000.ehci: park 0
orion-ehci f1050000.ehci: reset command 0080002 (park)=0 ithresh=8 period=1024 Reset HALT orion-ehci f1050000.ehci: can't setup
orion-ehci f1050000.ehci: USB bus 1 deregistered
orion-ehci f1050000.ehci: init f1050000.ehci fail, -110
orion-ehci: probe of f1050000.ehci failed with error -110

Quoted 7 lines
Quoted 7 lines

This is very similar to this problem, except that it always happens:
https://lkml.org/lkml/2012/6/4/381

I have cornered it out to the last handshake() call of the ehci_halt() call, that should set the controller in the HALT state. If I comment this handshake() call out, the registration is fine and the controller then looks OK (I don't have a hardware with a real USB bus to further test it yet).

Valentin,

can you specify what "real USB bus" means? Does your current hardware support USB host or device with anything connected to it?

Quoted 3 lines
Quoted 3 lines

My current idea is that maybe a reset or an initialization may be missing, but I wanted to ask if anybody already has seen a similar behavior with the various hardware platform supported by ehci-orion ?

You might want to tryout chipidea DRC too. We do need some mvebu boiler plate code to setup mbus registers but the IP definitely is chipidea.

Quoted 4 lines

Do you know what phy the 98DX4122 uses? The ehci-orion.c has some code for handling the Orion5X USB phy. I also think u-boot might also have some code. Sebastian might know more, i think he implemented USB support for barebox.

From my experiments with MVEBU usb phy's I recall that (at least on
Armada 370/XP) ehci-orion will hang the system if PHYs are not setup by bootloader. I cannot recall what happened on KW.

Sebastian

#3 lore raw
Quoted 32 lines · Sebastian Hesselbarth
Quoted 23 lines · Andrew Lunn
Quoted 22 lines · Valentin Longchamp

I am currently bringing up the USB 2.0 Host controller of the Bobcat's (98DX4122 Marvell switch) internal kirkwood CPU on a variation of Keymile's km_kirwood hardware (kirkwood-km_kirkwood.dts).

When the driver registers its hcd, it fails with a timemout as it can be seen in the below log:

Quoted 14 lines

root at kmcoge5un:~# dmesg -n 8
root at kmcoge5un:~# insmod ehci-orion.ko
ehci-orion: EHCI orion driver
Initializing Orion-SoC USB Host Controller
orion-ehci f1050000.ehci: EHCI Host Controller
orion-ehci f1050000.ehci: new USB bus registered, assigned bus number 1 orion-ehci f1050000.ehci: reset hcs_params 0x10011 dbg=0 ind cc=0 pcc=0 ordered ports=1 orion-ehci f1050000.ehci: reset hcc_params 0006 thresh 0 uframes 256/512/1024 park orion-ehci f1050000.ehci: park 0
orion-ehci f1050000.ehci: reset command 0080002 (park)=0 ithresh=8 period=1024 Reset HALT orion-ehci f1050000.ehci: can't setup
orion-ehci f1050000.ehci: USB bus 1 deregistered
orion-ehci f1050000.ehci: init f1050000.ehci fail, -110
orion-ehci: probe of f1050000.ehci failed with error -110

Quoted 7 lines
Quoted 7 lines

This is very similar to this problem, except that it always happens:
https://lkml.org/lkml/2012/6/4/381

I have cornered it out to the last handshake() call of the ehci_halt() call, that should set the controller in the HALT state. If I comment this handshake() call out, the registration is fine and the controller then looks OK (I don't have a hardware with a real USB bus to further test it yet).

Quoted 4 lines

Valentin,

can you specify what "real USB bus" means? Does your current hardware support USB host or device with anything connected to it?

No. The current hardware I am testing on does not support USB since the USB bus is not "wired" at all, there is no phy present (this maybe causes a problem ?). What I test is if the ehci-orion USB host driver is able to register with and configure the USB host controller present in the SoC, in order to prepare things for when the hardware is available.

Hopefully I get some new hardware prototypes where the USB bus is complete this week so that I can test on the real setup.

Quoted 7 lines
Quoted 4 lines
Quoted 4 lines

My current idea is that maybe a reset or an initialization may be missing, but I wanted to ask if anybody already has seen a similar behavior with the various hardware platform supported by ehci-orion ?

You might want to tryout chipidea DRC too. We do need some mvebu boiler plate code to setup mbus registers but the IP definitely is chipidea.

OK, that's good to know, I will have a look at this driver and maybe I can find a few differences there. ehci-orion is however stil the driver to use with this platform, right ?

Quoted 10 lines
Quoted 5 lines

Do you know what phy the 98DX4122 uses? The ehci-orion.c has some code for handling the Orion5X USB phy. I also think u-boot might also have some code. Sebastian might know more, i think he implemented USB support for barebox.

From my experiments with MVEBU usb phy's I recall that (at least on
Armada 370/XP) ehci-orion will hang the system if PHYs are not setup by bootloader. I cannot recall what happened on KW.

I tried to look if I found something in u-boot regarding USB and Kirkwood and found nothing else that the USB_EHCI_MARVELL. With it enabled, u-boot is able to init the controller and scan the "empty" bus, but it then does not help the kernel to later probe the ehci-orion driver: it does not more that ehci_orion_conf_mbus_windows() from ehci-orion).

Valentin

#4 lore raw
Quoted 38 lines · Valentin Longchamp
Quoted 32 lines · Sebastian Hesselbarth
Quoted 23 lines · Andrew Lunn
Quoted 22 lines · Valentin Longchamp

I am currently bringing up the USB 2.0 Host controller of the Bobcat's (98DX4122 Marvell switch) internal kirkwood CPU on a variation of Keymile's km_kirwood hardware (kirkwood-km_kirkwood.dts).

When the driver registers its hcd, it fails with a timemout as it can be seen in the below log:

Quoted 14 lines

root at kmcoge5un:~# dmesg -n 8
root at kmcoge5un:~# insmod ehci-orion.ko
ehci-orion: EHCI orion driver
Initializing Orion-SoC USB Host Controller
orion-ehci f1050000.ehci: EHCI Host Controller
orion-ehci f1050000.ehci: new USB bus registered, assigned bus number 1 orion-ehci f1050000.ehci: reset hcs_params 0x10011 dbg=0 ind cc=0 pcc=0 ordered ports=1 orion-ehci f1050000.ehci: reset hcc_params 0006 thresh 0 uframes 256/512/1024 park orion-ehci f1050000.ehci: park 0
orion-ehci f1050000.ehci: reset command 0080002 (park)=0 ithresh=8 period=1024 Reset HALT orion-ehci f1050000.ehci: can't setup
orion-ehci f1050000.ehci: USB bus 1 deregistered
orion-ehci f1050000.ehci: init f1050000.ehci fail, -110
orion-ehci: probe of f1050000.ehci failed with error -110

Quoted 7 lines
Quoted 7 lines

This is very similar to this problem, except that it always happens:
https://lkml.org/lkml/2012/6/4/381

I have cornered it out to the last handshake() call of the ehci_halt() call, that should set the controller in the HALT state. If I comment this handshake() call out, the registration is fine and the controller then looks OK (I don't have a hardware with a real USB bus to further test it yet).

Quoted 4 lines

Valentin,

can you specify what "real USB bus" means? Does your current hardware support USB host or device with anything connected to it?

Quoted 9 lines

No. The current hardware I am testing on does not support USB since the USB bus is not "wired" at all, there is no phy present (this maybe causes a problem ?). What I test is if the ehci-orion USB host driver is able to register with and configure the USB host controller present in the SoC, in order to prepare things for when the hardware is available.

Hopefully I get some new hardware prototypes where the USB bus is complete this week so that I can test on the real setup.

We have now received the hardware prototype and the behavior is exactly the
same: ETIMEOUT in ehci_halt(). If the last handshake() call is removed, the EHCI host controller is correctly initialized and the device on the bus is detected and works correctly.

I thus tend to think that this is an IP (or Bobcat) internal problem and is not really related with the hardware nor a PHY.

Valentin

#5 lore raw
Quoted 49 lines · Valentin Longchamp
Quoted 38 lines · Valentin Longchamp
Quoted 32 lines · Sebastian Hesselbarth
Quoted 23 lines · Andrew Lunn
Quoted 22 lines · Valentin Longchamp

I am currently bringing up the USB 2.0 Host controller of the Bobcat's (98DX4122 Marvell switch) internal kirkwood CPU on a variation of Keymile's km_kirwood hardware (kirkwood-km_kirkwood.dts).

When the driver registers its hcd, it fails with a timemout as it can be seen in the below log:

Quoted 14 lines

root at kmcoge5un:~# dmesg -n 8
root at kmcoge5un:~# insmod ehci-orion.ko
ehci-orion: EHCI orion driver
Initializing Orion-SoC USB Host Controller
orion-ehci f1050000.ehci: EHCI Host Controller
orion-ehci f1050000.ehci: new USB bus registered, assigned bus number 1 orion-ehci f1050000.ehci: reset hcs_params 0x10011 dbg=0 ind cc=0 pcc=0 ordered ports=1 orion-ehci f1050000.ehci: reset hcc_params 0006 thresh 0 uframes 256/512/1024 park orion-ehci f1050000.ehci: park 0
orion-ehci f1050000.ehci: reset command 0080002 (park)=0 ithresh=8 period=1024 Reset HALT orion-ehci f1050000.ehci: can't setup
orion-ehci f1050000.ehci: USB bus 1 deregistered
orion-ehci f1050000.ehci: init f1050000.ehci fail, -110
orion-ehci: probe of f1050000.ehci failed with error -110

Quoted 7 lines
Quoted 7 lines

This is very similar to this problem, except that it always happens:
https://lkml.org/lkml/2012/6/4/381

I have cornered it out to the last handshake() call of the ehci_halt() call, that should set the controller in the HALT state. If I comment this handshake() call out, the registration is fine and the controller then looks OK (I don't have a hardware with a real USB bus to further test it yet).

Quoted 4 lines

Valentin,

can you specify what "real USB bus" means? Does your current hardware support USB host or device with anything connected to it?

Quoted 9 lines

No. The current hardware I am testing on does not support USB since the USB bus is not "wired" at all, there is no phy present (this maybe causes a problem ?). What I test is if the ehci-orion USB host driver is able to register with and configure the USB host controller present in the SoC, in order to prepare things for when the hardware is available.

Hopefully I get some new hardware prototypes where the USB bus is complete this week so that I can test on the real setup.

Quoted 8 lines

We have now received the hardware prototype and the behavior is exactly the
same: ETIMEOUT in ehci_halt(). If the last handshake() call is removed, the EHCI host controller is correctly initialized and the device on the bus is detected and works correctly.

I thus tend to think that this is an IP (or Bobcat) internal problem and is not really related with the hardware nor a PHY.

One final update on this topic: the kernel config was missing the CONFIG_USB_EHCI_ROOT_HUB_TT option. It turns out that the USB IP embeded a Transaction Translator and there is a check in ehci_halt() that aborts the routine in case of a Transaction Translator in the ehci controller. The check however only works in the above option is enabled.

Thanks for the support

Valentin

#6 lore raw
Quoted 5 lines

One final update on this topic: the kernel config was missing the CONFIG_USB_EHCI_ROOT_HUB_TT option. It turns out that the USB IP embeded a Transaction Translator and there is a check in ehci_halt() that aborts the routine in case of a Transaction Translator in the ehci controller. The check however only works in the above option is enabled.

Maybe it would be worth detecting this and issuing a warning?

Andrew