Re: [PATCH] bus: mhi: pci_generic: increase timeout value for operations to 24000ms

From: Loic Poulain <loic.poulain@linaro.org>
To: Thomas Perrot <thomas.perrot@bootlin.com>
Cc: Aleksander Morgado <aleksander@aleksander.es>,
	Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>,
	linux-arm-msm <linux-arm-msm@vger.kernel.org>,
	Hemant Kumar <hemantk@codeaurora.org>,
	stable <stable@vger.kernel.org>
Subject: Re: [PATCH] bus: mhi: pci_generic: increase timeout value for operations to 24000ms
Date: Thu, 30 Sep 2021 12:19:48 +0200	[thread overview]
Message-ID: <CAMZdPi9AJcdVC2AS08ZtHU6sv1CMLhudgJHVpQF2zLfF+OV=Og@mail.gmail.com> (raw)
In-Reply-To: <54c063613fe63282a1c26b312c772e89b662eae6.camel@bootlin.com>

On Thu, 30 Sept 2021 at 10:08, Thomas Perrot <thomas.perrot@bootlin.com> wrote:
>
> Hello,
>
> On Wed, 2021-09-29 at 15:17 +0200, Aleksander Morgado wrote:
> > Hey Mani,
> >
> > > > > diff --git a/drivers/bus/mhi/pci_generic.c
> > > > > b/drivers/bus/mhi/pci_generic.c
> > > > > index 4dd1077354af..e08ed6e5031b 100644
> > > > > --- a/drivers/bus/mhi/pci_generic.c
> > > > > +++ b/drivers/bus/mhi/pci_generic.c
> > > > > @@ -248,7 +248,7 @@ static struct mhi_event_config
> > > > > modem_qcom_v1_mhi_events[] = {
> > > > >
> > > > >   static const struct mhi_controller_config
> > > > > modem_qcom_v1_mhiv_config = {
> > > > >      .max_channels = 128,
> > > > > -    .timeout_ms = 8000,
> > > > > +    .timeout_ms = 24000,
> > > >
> > > >
> > > > This modem_qcom_v1_mhiv_config config applies to all generic SDX24,
> > > > SDX55 and SDX65 modules.
> > > > Other vendor-branded SDX55 based modules in this same file (Foxconn
> > > > SDX55, MV31), have 20000ms as timeout.
> > > > Other vendor-branded SDX24 based modules in this same file (Quectel
> > > > EM12xx), have also 20000ms as timeout.
> > > > Maybe it makes sense to have a common timeout for all?
> > > >
> > >
> > > Eventhough the baseport coming from Qualcomm for the modem chipsets
> > > are same, it is possible that the vendors might have customized the
> > > firmware for their own usecase. That could be the cause of the delay
> > > for modem booting.
> > >
> > > So I don't think we should use the same timeout of 2400ms for all
> > > modems.
> > >
> >
> > Please note it's 24000ms what's being suggested here, not 2400ms.
> >
> > > > Thomas, is the 24000ms value taken from experimentation, or is it a
> > > > safe enough value? Maybe 20000ms as in other modules would have
> > > > been enough?
> > > >
>
> I made experimentation on a Sierra EM9190 (SDX55) engineering sample,
> using a old development firmware.
>
> So, I agree that setting the same timeout of 24000ms for all modems, is
> not necessarily relevant.
> However, the current default value seems too low, in view of timeouts
> used on vendor-branded, then using a higher value seems relevant.

I agree, let's set a conservative high value for generic SDX55, so
have wider support, that can be tuned per vendor IDs if necessary
anyway.

Regards,
Loic

>
> Moreover, Sierra EM919x modems use a custom controller configuration,
> we are currently working on it. As our tests not being sufficiently
> conclusive, so we have not yet submitted.
>
> Best regards,
> Thomas
>
> > >
> > > It was derived from testing I believe.
> >
> > Following your reasoning above, shouldn't this 24000ms timeout be
> > applied only to the Sierra Wireless EM91xx devices (which may have
> > custom firmware bits delaying the initialization a bit longer), and
> > not to the generic SDX24, SDX55 and SDX65?
> >
> > If I'm not mistaken, Thomas is testing with a custom mhi_pci_generic
> > entry for the EM91xx; as in
> > https://forum.sierrawireless.com/t/sierra-wireless-airprime-em919x-pcie-support/24927
> > .
> > I'm also playing with that same entry on my own setup, but have other
> > problems of my own :)
> >
> >
> > --
> > Aleksander
> > https://aleksander.es
>
> --
> Thomas Perrot, Bootlin
> Embedded Linux and kernel engineering
> https://bootlin.com
>