ehci support on old VIA vt82xxx disappeared around kernel 3.8.

ehci support on old VIA vt82xxx disappeared around kernel 3.8.

[Geoff Winkless]

Hi

Apologies if this is the incorrect place to post this, if so please
feel free to call me names and then suggest somewhere more appropriate
:)

We have an embedded device on an old EPIA Mini-ITX board that runs
Linux 2.6. There are features in more recent (>3.10) kernels that
would be useful to us, so I tried to build them for it; however while
the kernel runs perfectly fine, ehci support simply fails, which is
catastrophic for the device's use - we need USB2 speeds.

I worked backwards and found that it runs normally on 3.7.9, but on
3.7.10 it starts throwing up errors:

usb 1-5: new high-speed USB device number 3 using ehci_hcd
usb 1-5: device descriptor read/8, error -110
usb 1-5: device descriptor read/8, error -110
usb 1-5: new high-speed USB device number 4 using ehci_hcd
usb 1-5: device not accepting address 4, error -110
usb 1-5: new high-speed USB device number 5 using ehci_hcd

By 3.8.0 ehci simply doesn't work, as if someone decided these errors
were too hard to deal with and just disabled the device support.

Chipset is VIA vt82xxx; the ID of the offending bus is 1106:3104

I tried every combination of loading ehci before and after etc,
setting the old_scheme_first value, disabling acpi/apm, basically
anything I could find on the web that seems related to ehci problems,
but nothing seems to make the ehci driver even recognise the device.

/proc/bus/pci/devices shows

0000 11063123 0  e6000008        0   0   0    0 0     0  400000
0 0 0  0 0     0 agpgart-via
0008 1106b091 0         0        0   0   0    0 0     0       0
0 0 0  0 0     0
0080 11063038 c         0        0   0   0 d001 0     0       0
0 0 0 20 0     0 uhci_hcd
0081 11063038 c         0        0   0   0 d401 0     0       0
0 0 0 20 0     0 uhci_hcd
0082 11063038 5         0        0   0   0 d801 0     0       0
0 0 0 20 0     0 uhci_hcd
0083 11063104 9  e6400000        0   0   0    0 0     0     100
0 0 0  0 0     0
0088 11063177 0         0        0   0   0    0 0     0       0
0 0 0  0 0     0
0089 11060571 0       1f0      3f6 170 376 dc01 0     0       8
0 8 0 10 0     0 VIA_IDE
008d 11063059 5      e001        0   0   0    0 0     0     100
0 0 0  0 0     0 snd_via82xx
0090 11063065 c      e801 e6401000   0   0    0 0     0     100
100 0 0  0 0     0 via-rhine
0100 11063122 c  e0000008 e4000000   0   0    0 0 c0002 4000000
1000000 0 0  0 0 20000

(apologies for the long lines, I cut down the spacing as much as I could).

So you can see that the 11063104 line doesn't even have the ehci-hcd
driver associated; on earlier kernel versions the line reads

0083 11063104 9  e6400000 0 0 0 0 0 0 100 0 0 0 0 0 0 ehci_hcd

Output from lsmod, just in case you're thinking I just haven't
inserted the ehci driver...

usbnet 10726 0 - Live 0xcfb4f000
ohci_hcd 15520 0 - Live 0xcfad9000
uhci_hcd 15679 0 - Live 0xcfa94000
ehci_hcd 30853 0 - Live 0xcfa49000
pl2303 6016 0 - Live 0xcf979000
ftdi_sio 25410 0 - Live 0xcf940000
option 18882 0 - Live 0xcf8e7000
usb_wwan 4082 1 option, Live 0xcf8a2000

I'm happy to dig for myself - I appreciate this is a fairly niche
problem; I have getting on for 30 years' development experience in
various platforms, including low-level hardware access in assembly
when I was young, but all I've ever done with the kernel is modify in
a very small way a few device drivers and I don't even know where to
start with this so could do with some pointers.

I tried running a diff on drivers/usb between 3.7.9 and 3.7.10, but
other than a few changes around usbserial there doesn't seem to be
much, which seems odd given that the behaviour clearly changed. The
3.7.10 changelog only lists some usb-audio changes, a change for
memory allocation for some urb blocks, and some fixes for usb-serial,
so I guess something else changing has modified the way the USB core
interacts with the hardware.

Did we intentionally simply drop support for this chipset in 3.8?

Thanks!

Geoff

Re: ehci support on old VIA vt82xxx disappeared around kernel 3.8.

[Greg KH]

On Fri, May 31, 2019 at 05:40:23PM +0100, Geoff Winkless wrote:
> Hi
> 
> Apologies if this is the incorrect place to post this, if so please
> feel free to call me names and then suggest somewhere more appropriate
> :)
> 
> We have an embedded device on an old EPIA Mini-ITX board that runs
> Linux 2.6. There are features in more recent (>3.10) kernels that
> would be useful to us, so I tried to build them for it; however while
> the kernel runs perfectly fine, ehci support simply fails, which is
> catastrophic for the device's use - we need USB2 speeds.
> 
> I worked backwards and found that it runs normally on 3.7.9, but on
> 3.7.10 it starts throwing up errors:
> 
> usb 1-5: new high-speed USB device number 3 using ehci_hcd
> usb 1-5: device descriptor read/8, error -110
> usb 1-5: device descriptor read/8, error -110
> usb 1-5: new high-speed USB device number 4 using ehci_hcd
> usb 1-5: device not accepting address 4, error -110
> usb 1-5: new high-speed USB device number 5 using ehci_hcd
> 
> By 3.8.0 ehci simply doesn't work, as if someone decided these errors
> were too hard to deal with and just disabled the device support.
> 
> Chipset is VIA vt82xxx; the ID of the offending bus is 1106:3104
> 
> I tried every combination of loading ehci before and after etc,
> setting the old_scheme_first value, disabling acpi/apm, basically
> anything I could find on the web that seems related to ehci problems,
> but nothing seems to make the ehci driver even recognise the device.
> 
> /proc/bus/pci/devices shows
> 
> 0000 11063123 0  e6000008        0   0   0    0 0     0  400000
> 0 0 0  0 0     0 agpgart-via
> 0008 1106b091 0         0        0   0   0    0 0     0       0
> 0 0 0  0 0     0
> 0080 11063038 c         0        0   0   0 d001 0     0       0
> 0 0 0 20 0     0 uhci_hcd
> 0081 11063038 c         0        0   0   0 d401 0     0       0
> 0 0 0 20 0     0 uhci_hcd
> 0082 11063038 5         0        0   0   0 d801 0     0       0
> 0 0 0 20 0     0 uhci_hcd
> 0083 11063104 9  e6400000        0   0   0    0 0     0     100
> 0 0 0  0 0     0
> 0088 11063177 0         0        0   0   0    0 0     0       0
> 0 0 0  0 0     0
> 0089 11060571 0       1f0      3f6 170 376 dc01 0     0       8
> 0 8 0 10 0     0 VIA_IDE
> 008d 11063059 5      e001        0   0   0    0 0     0     100
> 0 0 0  0 0     0 snd_via82xx
> 0090 11063065 c      e801 e6401000   0   0    0 0     0     100
> 100 0 0  0 0     0 via-rhine
> 0100 11063122 c  e0000008 e4000000   0   0    0 0 c0002 4000000
> 1000000 0 0  0 0 20000
> 
> (apologies for the long lines, I cut down the spacing as much as I could).
> 
> So you can see that the 11063104 line doesn't even have the ehci-hcd
> driver associated; on earlier kernel versions the line reads
> 
> 0083 11063104 9  e6400000 0 0 0 0 0 0 100 0 0 0 0 0 0 ehci_hcd
> 
> Output from lsmod, just in case you're thinking I just haven't
> inserted the ehci driver...
> 
> usbnet 10726 0 - Live 0xcfb4f000
> ohci_hcd 15520 0 - Live 0xcfad9000
> uhci_hcd 15679 0 - Live 0xcfa94000
> ehci_hcd 30853 0 - Live 0xcfa49000
> pl2303 6016 0 - Live 0xcf979000
> ftdi_sio 25410 0 - Live 0xcf940000
> option 18882 0 - Live 0xcf8e7000
> usb_wwan 4082 1 option, Live 0xcf8a2000
> 
> I'm happy to dig for myself - I appreciate this is a fairly niche
> problem; I have getting on for 30 years' development experience in
> various platforms, including low-level hardware access in assembly
> when I was young, but all I've ever done with the kernel is modify in
> a very small way a few device drivers and I don't even know where to
> start with this so could do with some pointers.
> 
> I tried running a diff on drivers/usb between 3.7.9 and 3.7.10, but
> other than a few changes around usbserial there doesn't seem to be
> much, which seems odd given that the behaviour clearly changed. The
> 3.7.10 changelog only lists some usb-audio changes, a change for
> memory allocation for some urb blocks, and some fixes for usb-serial,
> so I guess something else changing has modified the way the USB core
> interacts with the hardware.

Can you run 'git bisect' to determine the exact commit that caused this
problem?  That would be most helpful.

> Did we intentionally simply drop support for this chipset in 3.8?

That kernel was released in 2013, that's a long time ago to try to
remember something as specific as that :(

greg k-h

Re: ehci support on old VIA vt82xxx disappeared around kernel 3.8.

[Geoff Winkless]

On Fri, 31 May 2019 at 20:11, Greg KH <greg@kroah.com> wrote:

> Can you run 'git bisect' to determine the exact commit that caused this
> problem?  That would be most helpful.

Oh for goodness sake. Sorry, I was being an idiot. After half a day of
building bisect scripts and another 3 hours of waiting for
compile-reboot-test I finally get to

[adfa79d1c06a32650332930ca4c488ca570b3407] USB: EHCI: make ehci-pci a
separate driver

I guess that should have been fairly obvious, really.

modprobe ehci-pci and the bus entry reappears.

Thanks (and also to Alan) for the help, apologies for wasting your time.

Geoff (going to hide his blushes in a corner somewhere)

Re: ehci support on old VIA vt82xxx disappeared around kernel 3.8.

[Alan Stern]

On Mon, 3 Jun 2019, Geoff Winkless wrote:

> On Fri, 31 May 2019 at 20:11, Greg KH <greg@kroah.com> wrote:
> 
> > Can you run 'git bisect' to determine the exact commit that caused this
> > problem?  That would be most helpful.
> 
> Oh for goodness sake. Sorry, I was being an idiot. After half a day of
> building bisect scripts and another 3 hours of waiting for
> compile-reboot-test I finally get to
> 
> [adfa79d1c06a32650332930ca4c488ca570b3407] USB: EHCI: make ehci-pci a
> separate driver
> 
> I guess that should have been fairly obvious, really.
> 
> modprobe ehci-pci and the bus entry reappears.
> 
> Thanks (and also to Alan) for the help, apologies for wasting your time.
> 
> Geoff (going to hide his blushes in a corner somewhere)

Just goes to show we all have our blind spots.  I didn't realize what 
was going on either, and I should have.

On the other hand, you mentioned that it was getting some errors even
under earlier kernels such as 3.7.10 (before ehci-pci was split off
into a separate module, or it would have failed completely).  Any idea
about that?

Alan Stern

Re: ehci support on old VIA vt82xxx disappeared around kernel 3.8.

[Geoff Winkless]

On Mon, 3 Jun 2019 at 16:29, Alan Stern <stern@rowland.harvard.edu> wrote:
> On Mon, 3 Jun 2019, Geoff Winkless wrote:
>
> > Thanks (and also to Alan) for the help, apologies for wasting your time.
>
> Just goes to show we all have our blind spots.  I didn't realize what
> was going on either, and I should have.

You're too kind :)

> On the other hand, you mentioned that it was getting some errors even
> under earlier kernels such as 3.7.10 (before ehci-pci was split off
> into a separate module, or it would have failed completely).  Any idea
> about that?

Well I was going to go after that next (test script would be a bit
more complicated), but given that those errors don't seem to be
showing up on 4.x I'll probably just drop it, unless you care.

Geoff

Re: ehci support on old VIA vt82xxx disappeared around kernel 3.8.

[Alan Stern]

On Mon, 3 Jun 2019, Geoff Winkless wrote:

> On Mon, 3 Jun 2019 at 16:29, Alan Stern <stern@rowland.harvard.edu> wrote:
> > On Mon, 3 Jun 2019, Geoff Winkless wrote:
> >
> > > Thanks (and also to Alan) for the help, apologies for wasting your time.
> >
> > Just goes to show we all have our blind spots.  I didn't realize what
> > was going on either, and I should have.
> 
> You're too kind :)
> 
> > On the other hand, you mentioned that it was getting some errors even
> > under earlier kernels such as 3.7.10 (before ehci-pci was split off
> > into a separate module, or it would have failed completely).  Any idea
> > about that?
> 
> Well I was going to go after that next (test script would be a bit
> more complicated), but given that those errors don't seem to be
> showing up on 4.x I'll probably just drop it, unless you care.

No, occasional errors don't seem to be worth tracking down, given that 
they don't occur with later kernels.

Alan Stern

Re: ehci support on old VIA vt82xxx disappeared around kernel 3.8.

[Alan Stern]

On Fri, 31 May 2019, Geoff Winkless wrote:

> Hi
> 
> Apologies if this is the incorrect place to post this, if so please
> feel free to call me names and then suggest somewhere more appropriate
> :)
> 
> We have an embedded device on an old EPIA Mini-ITX board that runs
> Linux 2.6. There are features in more recent (>3.10) kernels that
> would be useful to us, so I tried to build them for it; however while
> the kernel runs perfectly fine, ehci support simply fails, which is
> catastrophic for the device's use - we need USB2 speeds.
> 
> I worked backwards and found that it runs normally on 3.7.9, but on
> 3.7.10 it starts throwing up errors:
> 
> usb 1-5: new high-speed USB device number 3 using ehci_hcd
> usb 1-5: device descriptor read/8, error -110
> usb 1-5: device descriptor read/8, error -110
> usb 1-5: new high-speed USB device number 4 using ehci_hcd
> usb 1-5: device not accepting address 4, error -110
> usb 1-5: new high-speed USB device number 5 using ehci_hcd
> 
> By 3.8.0 ehci simply doesn't work, as if someone decided these errors
> were too hard to deal with and just disabled the device support.
> 
> Chipset is VIA vt82xxx; the ID of the offending bus is 1106:3104
> 
> I tried every combination of loading ehci before and after etc,
> setting the old_scheme_first value, disabling acpi/apm, basically
> anything I could find on the web that seems related to ehci problems,
> but nothing seems to make the ehci driver even recognise the device.
> 
> /proc/bus/pci/devices shows
> 
> 0000 11063123 0  e6000008        0   0   0    0 0     0  400000
> 0 0 0  0 0     0 agpgart-via
> 0008 1106b091 0         0        0   0   0    0 0     0       0
> 0 0 0  0 0     0
> 0080 11063038 c         0        0   0   0 d001 0     0       0
> 0 0 0 20 0     0 uhci_hcd
> 0081 11063038 c         0        0   0   0 d401 0     0       0
> 0 0 0 20 0     0 uhci_hcd
> 0082 11063038 5         0        0   0   0 d801 0     0       0
> 0 0 0 20 0     0 uhci_hcd
> 0083 11063104 9  e6400000        0   0   0    0 0     0     100
> 0 0 0  0 0     0
> 0088 11063177 0         0        0   0   0    0 0     0       0
> 0 0 0  0 0     0
> 0089 11060571 0       1f0      3f6 170 376 dc01 0     0       8
> 0 8 0 10 0     0 VIA_IDE
> 008d 11063059 5      e001        0   0   0    0 0     0     100
> 0 0 0  0 0     0 snd_via82xx
> 0090 11063065 c      e801 e6401000   0   0    0 0     0     100
> 100 0 0  0 0     0 via-rhine
> 0100 11063122 c  e0000008 e4000000   0   0    0 0 c0002 4000000
> 1000000 0 0  0 0 20000
> 
> (apologies for the long lines, I cut down the spacing as much as I could).
> 
> So you can see that the 11063104 line doesn't even have the ehci-hcd
> driver associated; on earlier kernel versions the line reads
> 
> 0083 11063104 9  e6400000 0 0 0 0 0 0 100 0 0 0 0 0 0 ehci_hcd
> 
> Output from lsmod, just in case you're thinking I just haven't
> inserted the ehci driver...
> 
> usbnet 10726 0 - Live 0xcfb4f000
> ohci_hcd 15520 0 - Live 0xcfad9000
> uhci_hcd 15679 0 - Live 0xcfa94000
> ehci_hcd 30853 0 - Live 0xcfa49000
> pl2303 6016 0 - Live 0xcf979000
> ftdi_sio 25410 0 - Live 0xcf940000
> option 18882 0 - Live 0xcf8e7000
> usb_wwan 4082 1 option, Live 0xcf8a2000
> 
> I'm happy to dig for myself - I appreciate this is a fairly niche
> problem; I have getting on for 30 years' development experience in
> various platforms, including low-level hardware access in assembly
> when I was young, but all I've ever done with the kernel is modify in
> a very small way a few device drivers and I don't even know where to
> start with this so could do with some pointers.
> 
> I tried running a diff on drivers/usb between 3.7.9 and 3.7.10, but
> other than a few changes around usbserial there doesn't seem to be
> much, which seems odd given that the behaviour clearly changed. The
> 3.7.10 changelog only lists some usb-audio changes, a change for
> memory allocation for some urb blocks, and some fixes for usb-serial,
> so I guess something else changing has modified the way the USB core
> interacts with the hardware.

It is quite likely that the changes responsible for your problem are 
not located under drivers/usb.  I would guess they are in the 
platform-specific code somewhere.  In any case, you should broaden your 
approach.  Greg's suggestion about using git bisect, say between 3.7.9 
and 3.7.10, would be an excellent start.

> Did we intentionally simply drop support for this chipset in 3.8?

Not as far as I'm aware.  But I don't pay much attention to the
platform-specific chipset details, so I wouldn't really know.  You'd be 
better off asking the maintainers responsible for your platform.

Alan Stern