Re: Regression due to "Workaround for uPD72020x USB3 chips"

[Ard Biesheuvel]

On 2 May 2018 at 11:18, Domenico Andreoli <domenico.andreoli@xxxxxxxxx> wrote:
> On Wed, May 02, 2018 at 10:22:05AM +0200, Ard Biesheuvel wrote:
>> On 2 May 2018 at 10:06, Domenico Andreoli <domenico.andreoli@xxxxxxxxx> wrote:
>> > Dear all,
>> >
>> >   my home machine stopped to boot starting from kernel version 4.12.7.
>> >
>> > The last message I read is about resetting some USB3 bus. It's 100%
>> > reproducible also with any recent kernel up to 4.17.0-rc3.
>> >
>> > I bisected down to the following commit:
>> >
>> > commit 0e1f0eaed6c20db41ff61e024b361ee3ec9d686c (tag: my_broken_xhci)
>> > Author: Marc Zyngier <marc.zyngier@xxxxxxx>
>> > Date:   Tue Aug 1 20:11:08 2017 -0500
>> >
>> >     xhci: Reset Renesas uPD72020x USB controller for 32-bit DMA issue
>> >
>>
>> Hello Domenico,
>
> Sir..
>
>>
>> Long time no see :-)
>
> Instead I see you pretty much everywhere :P
>
>>
>> Please refer to the following thread
>>
>> https://marc.info/?l=linux-usb&m=151872295419486
>>
>> for some discussion on this topic, and a possible workaround.
>
> Interesting, thanks for the pointer!
>
> I'm happy to learn I've this nice piece of HW in my machine. It might
> have at least some other issue while trying to hibernate (there is a
> fair chance for it to abort the process because of some "business").
>
> I tried the usb/uPD720202-reset branch from Marc (with some massaging
> due to quirks now being >32) and my system boots again. I think it's
> more because of the PCI reset is removed, I am not affected by the
> original issue it was meant to fix.
>

Indeed. The majority of x86 machines never perform DMA above 4 GB in
UEFI, so the issue does not occur, although it could still affect a
kernel with IOMMU on kexec'ed from a kernel with IOMMU off.