Fwd: Linux - AMD SB950 USB Regression

Fwd: Linux - AMD SB950 USB Regression

[dAgeCKo]

Hello,

Some months ago I noticed that my USB controler (and also my network
controler) does not work anymore on newer kernels from my debian wheezy
distribution.

I then reported a bug for the former one here:

https://bugzilla.kernel.org/show_bug.cgi?id=47791

With some helps and time, I could locate the source of the regression,
that is an old commit from the year 2010 you seem to commit. Git
besection identified the commit as:

git bisect bad 5c80cc78de46aef6cd5e714208da05c3f7f548f8

Someone from the usb kernel mailing list (Alan Stern) then asked me to
write an email to you, in my hope the bug to be solved.

Some informations were said on the bug tracker, but if you need any
other information that could help you in any manner, please let me know.

Please note that the regression only involves USB2 ports of the
mainboard, not the USB3 ports.

Regards.

---
Jean-Dominique Frattini

Re: Fwd: Linux - AMD SB950 USB Regression

[dAgeCKo]

Hi,

the bug has been resolved after a BIOS upgrade. Since the mainboard 
constructor said me that it's impossible that the BIOS was an issue 
since their Windows users do not face the same issue, I beleived it 
could have come from the Linux kernel.

Bye.

Le 04/01/2013 11:22, dAgeCKo a écrit :
>
> Hello,
>
> Some months ago I noticed that my USB controler (and also my network
> controler) does not work anymore on newer kernels from my debian wheezy
> distribution.
>
> I then reported a bug for the former one here:
>
> https://bugzilla.kernel.org/show_bug.cgi?id=47791
>
> With some helps and time, I could locate the source of the regression,
> that is an old commit from the year 2010 you seem to commit. Git
> besection identified the commit as:
>
> git bisect bad 5c80cc78de46aef6cd5e714208da05c3f7f548f8
>
> Someone from the usb kernel mailing list (Alan Stern) then asked me to
> write an email to you, in my hope the bug to be solved.
>
> Some informations were said on the bug tracker, but if you need any
> other information that could help you in any manner, please let me know.
>
> Please note that the regression only involves USB2 ports of the
> mainboard, not the USB3 ports.
>
> Regards.
>
> ---
> Jean-Dominique Frattini

Re: Fwd: Linux - AMD SB950 USB Regression

[Borislav Petkov]

Hi,

(leaving in the rest for reference, answer below, adding Alan, Joerg,
Shane and linux-usb to CC).

On Fri, Jan 11, 2013 at 04:59:11PM +0100, dAgeCKo wrote:
> the bug has been resolved after a BIOS upgrade. Since the mainboard
> constructor said me that it's impossible that the BIOS was an issue
> since their Windows users do not face the same issue, I beleived it
> could have come from the Linux kernel.
> 
> Bye.
> 
> Le 04/01/2013 11:22, dAgeCKo a écrit :
> >
> >Hello,
> >
> >Some months ago I noticed that my USB controler (and also my network
> >controler) does not work anymore on newer kernels from my debian wheezy
> >distribution.
> >
> >I then reported a bug for the former one here:
> >
> >https://bugzilla.kernel.org/show_bug.cgi?id=47791
> >
> >With some helps and time, I could locate the source of the regression,
> >that is an old commit from the year 2010 you seem to commit. Git
> >besection identified the commit as:
> >
> >git bisect bad 5c80cc78de46aef6cd5e714208da05c3f7f548f8
> >
> >Someone from the usb kernel mailing list (Alan Stern) then asked me to
> >write an email to you, in my hope the bug to be solved.
> >
> >Some informations were said on the bug tracker, but if you need any
> >other information that could help you in any manner, please let me know.
> >
> >Please note that the regression only involves USB2 ports of the
> >mainboard, not the USB3 ports.

So, there seems to be some serious breakage with GART on SB9xx. I don't
know whether this is the platform BIOS or the vendor BIOS causing it
because the original bug reporter says he observes the issue on an MSI
board and I'm experiencing this on my favourite bunch ASUS.

So, if you use the GART as an IOMMU, i.e.:

Jan 17 22:08:15 pd kernel: [    0.924063] PCI-DMA: using GART IOMMU.
Jan 17 22:08:15 pd kernel: [    0.924120] PCI-DMA: Reserving 64MB of IOMMU area in the AGP aperture

USB ports start choking like this:

Jan 17 22:08:15 pd kernel: [    3.229909] usb 4-1: device descriptor read/64, error -32
Jan 17 22:08:15 pd kernel: [    3.432786] usb 4-1: new high-speed USB device number 3 using ehci-pci
Jan 17 22:08:15 pd kernel: [    3.534684] usb 4-1: device descriptor read/64, error -32
Jan 17 22:08:15 pd kernel: [    3.737642] usb 4-1: device descriptor read/64, error -32

I'd go and venture a guess here since I don't have an idea that DMA
somehow gets busted with the GART and thus the errors.

Now, those boards normally have an IOMMU too so if you go and enable

CONFIG_AMD_IOMMU=y

the problem is gone (we're using the real IOMMU for DMA mapping, etc,
etc). So dAgeCKo, that would be another thing you could do: try enabling
the IOMMU in the BIOS and the above CONFIG option and the issue would be
fixed too.

At least this fixes it on my box.

To the question "how do we fix the GART issue?" I have no answer and
would expect more informed opinions from someone else.

Thanks and HTH.

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--

Re: Fwd: Linux - AMD SB950 USB Regression

[dAgeCKo]

Le 17/01/2013 22:36, Borislav Petkov a écrit :

>
> So, there seems to be some serious breakage with GART on SB9xx. I don't
> know whether this is the platform BIOS or the vendor BIOS causing it
> because the original bug reporter says he observes the issue on an MSI
> board and I'm experiencing this on my favourite bunch ASUS.
>
> So, if you use the GART as an IOMMU, i.e.:
>
> Jan 17 22:08:15 pd kernel: [    0.924063] PCI-DMA: using GART IOMMU.
> Jan 17 22:08:15 pd kernel: [    0.924120] PCI-DMA: Reserving 64MB of IOMMU area in the AGP aperture
>
> USB ports start choking like this:
>
> Jan 17 22:08:15 pd kernel: [    3.229909] usb 4-1: device descriptor read/64, error -32
> Jan 17 22:08:15 pd kernel: [    3.432786] usb 4-1: new high-speed USB device number 3 using ehci-pci
> Jan 17 22:08:15 pd kernel: [    3.534684] usb 4-1: device descriptor read/64, error -32
> Jan 17 22:08:15 pd kernel: [    3.737642] usb 4-1: device descriptor read/64, error -32

Yes, this is exactly what I faced.

>
> I'd go and venture a guess here since I don't have an idea that DMA
> somehow gets busted with the GART and thus the errors.
>
> Now, those boards normally have an IOMMU too so if you go and enable
>
> CONFIG_AMD_IOMMU=y
>
> the problem is gone (we're using the real IOMMU for DMA mapping, etc,
> etc). So dAgeCKo, that would be another thing you could do: try enabling
> the IOMMU in the BIOS and the above CONFIG option and the issue would be
> fixed too.

Here is below, exactly what the guy from MSI said me after I said him a 
BIOS upgrade did resolve my bug:

"The  reacts differently to Win/Linux, but it relies on updates by AMD. 
The new AGESA in the BIOS might had fixes for Linux."

After few researches, it seems that this "BIOS-level" software has 
effectively a special implementation wrapper for Linux (something I 
can't understand why actually). For more reference you can find little 
information here 
(http://www.coreboot.org/data/LinuxBIOS%20AMD%202006%20Final_10-02-2006.pdf).

I can't try the solution you are giving since I can't (and don't want to 
try to) downgrade my BIOS.

Another important thing you might be interested in is that not only my 
USB 2 weren't working. My mainboard ethernet wasn't working to (and was 
resolved with the BIOS upgrade too). The ethernet chip is:

03:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. 
RTL8111/8168B PCI Express Gigabit Ethernet controller (rev 06)

I don't know if that can be related to what you are saying.

>
> At least this fixes it on my box.
>
> To the question "how do we fix the GART issue?" I have no answer and
> would expect more informed opinions from someone else.

Actually I can't tell if I have any GART issues. The only thing I can 
say is that 3D looks to work nicely. However I think I never consumed 
more memory than my graphic card has.

>
> Thanks and HTH.
>

Thanks you too.

Re: Fwd: Linux - AMD SB950 USB Regression

[Borislav Petkov]

On Sat, Jan 19, 2013 at 03:47:53PM +0100, dAgeCKo wrote:
> I can't try the solution you are giving since I can't (and don't want
> to try to) downgrade my BIOS.

I don't think I said that. Here's what I actually said:

>> So dAgeCKo, that would be another thing you could do: try enabling
>> the IOMMU in the BIOS and the above CONFIG option and the issue would
>> be fixed too.

IOW, you enter the BIOS, go to "NB settings" or similar and set IOMMU
to Auto/Enabled. Provided you have an IOMMU but you should have that on
this chipset.

Then, you enable CONFIG_AMD_IOMMU in your kernel and done: the kernel
uses the IOMMU hardware instead of the GART.

That's it, no need for any BIOS up-/downgrading.

> Another important thing you might be interested in is that not only
> my USB 2 weren't working. My mainboard ethernet wasn't working to
> (and was resolved with the BIOS upgrade too). The ethernet chip is:
> 
> 03:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd.
> RTL8111/8168B PCI Express Gigabit Ethernet controller (rev 06)
> 
> I don't know if that can be related to what you are saying.

It could be, my nic kept resetting and once got an xmit queue timeout so
it could very well be related.

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--