ath11k: QCA6390 on Dell XPS 13 and kernel crashes

[wi nk]

On Sat, Dec 12, 2020 at 6:37 AM Kalle Valo <kvalo at codeaurora.org> wrote:
>
> wi nk <wink at technolu.st> writes:
>
> >> > and the modification that disables m2 state:
> >> >
> >> > diff --git a/drivers/bus/mhi/core/pm.c b/drivers/bus/mhi/core/pm.c
> >> > index 3de7b1639ec6..20f670c8b129 100644
> >> > --- a/drivers/bus/mhi/core/pm.c
> >> > +++ b/drivers/bus/mhi/core/pm.c
> >> > @@ -55,12 +55,12 @@ static struct mhi_pm_transitions const
> >> > dev_state_transitions[] = {
> >> >      },
> >> >      {
> >> >          MHI_PM_M0,
> >> > -        MHI_PM_M0 | MHI_PM_M2 | MHI_PM_M3_ENTER |
> >> > +        MHI_PM_M0 | MHI_PM_M3_ENTER |
> >> >          MHI_PM_SYS_ERR_DETECT | MHI_PM_SHUTDOWN_PROCESS |
> >> >          MHI_PM_LD_ERR_FATAL_DETECT | MHI_PM_FW_DL_ERR
> >> >      },
> >> >      {
> >> > -        MHI_PM_M2,
> >> > +        MHI_PM_M0,
> >> >          MHI_PM_M0 | MHI_PM_SYS_ERR_DETECT | MHI_PM_SHUTDOWN_PROCESS |
> >> >          MHI_PM_LD_ERR_FATAL_DETECT
> >> >      },
> >>
> >> Adding one more data point.  The driver will not crash on
> >> initialization this way, but also with the M2 state transition
> >> disabled the system survives suspend and wake and the adapter
> >> successfully reassociates consistently.  As expected with my patch,
> >> the MHI driver shows everything stays in the M1 state instead of
> >> attempting to transition to M2 ever.  It also doesn't return back to
> >> M0 if I disconnect the power / replug it.  I'm not sure what things
> >> are affected by me hacking this state machine, but avoiding that M2
> >> transition has removed any obvious issues from my system.
> >
> > While waiting for someone else to confirm, I can report that I've
> > still not seen any instability since this patch.  The laptop has been
> > stable through reboots, power cycling, suspension, etc.
>
> Very interesting! Are you saying that with this patch the wireless
> connection using QCA6390 works fine on your Dell XPS 9310, you can
> connect to an AP and transfer data normally?
>

Precisely.  The machine is now over 24h of uptime, I can reboot/sleep
without any issues, and throughput seems to saturate my wifi link
(5-600mpbs).


> I would like to submit your patch to patchwork.kernel.org as RFC patch
> so that it's easier for everyone to download. But before I can do that I
> need your Signed-off-by, can you read Developer's Certificate of Origin:
>
> https://www.kernel.org/doc/html/latest/process/submitting-patches.html#sign-your-work-the-developer-s-certificate-of-origin
>
> And if you agree with the DCO please send your s-o-b by replying to this
> email. But you can also submit the RFC patch yourself, instructions
> here:
>
> https://wireless.wiki.kernel.org/en/users/drivers/ath11k/submittingpatches
>

Signed-off-by: Lee Smith <wink at technolu.st>

I'll get an email out later this afternoon, if you get there first,
please feel free :).

> > I'd be happy to continue to try to understand why this is this case.
> > It sounds like Stephen isn't seeing these issues on 5.10 rc6 with the
> > single msi patch+reverting that one commit. I can try to give that a
> > shot if it'd produce something useful.
>
> Yes, being able to give datapoints what affects this bug is very helpful
> to track down it.
>

Ok, I'll try to rebuild to that configuration later today and report back.

> > Kalle - a couple quick questions, in the driver comments the M2 state
> > is loosely documented as a low power mode.  Why would it transition to
> > that while on charger/plugging in, but stay in M0 while on battery
> > (you can see this behavior in the videos I linked previously)?
> > Naively I would've expected the opposite behavior.
>
> I would have expected the same as well, it does sound strange or we are
> misunderstanding something. I'll try to find out why it's so. But if you
> learn more, please do let me know.
>

Will do.

> > Also, is there any way to prevent that transition other than my brute
> > force? It seems on battery the 'nominal' state for it is M0, I'm not
> > sure what the effect of it being left in this M1 state really is even
> > though there's nothing observable. Lastly, any thoughts as to why it
> > seems that transition causes the EE state to become invalid?
>
> TBH I'm not very familiar with MHI, you seem to already know it much
> more better than I do :) I'll include more folks to the thread later,
> hopefully they can help.
>

Thanks!


> --
> https://patchwork.kernel.org/project/linux-wireless/list/
>
> https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches