Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
In-Reply-To: <CAE-0n52xQ+SbwpDfzuW-tMLJP17Z2qog2s_pS=XBTULLNpUkLg@mail.gmail.com>
References: <20211125174751.25317-1-djakov@kernel.org> <CAE-0n51xeigKFS9Zek44HZGD9cdc4Em91aQ5HHzuy7P1FBmfFg@mail.gmail.com>
 <CAE-0n52xQ+SbwpDfzuW-tMLJP17Z2qog2s_pS=XBTULLNpUkLg@mail.gmail.com>
From:   Stephen Boyd <swboyd@chromium.org>
User-Agent: alot/0.10
Date:   Fri, 8 Apr 2022 11:52:12 -0700
Message-ID: <CAE-0n53b3_mjXDP=W6XHQ9_5NEh7ee1GV6jQUSxtgE864A1ONw@mail.gmail.com>
Subject: Re: [PATCH v3] interconnect: qcom: icc-rpmh: Add BCMs to commit list
 in pre_aggregate
To:     Alex Elder <elder@linaro.org>, djakov@kernel.org,
        quic_mdtipton@quicinc.com
Cc:     linux-arm-msm@vger.kernel.org, linux-pm@vger.kernel.org,
        linux-kernel@vger.kernel.org, mka@chromium.org,
        dianders@chromium.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Quoting Stephen Boyd (2022-04-05 21:47:07)
> Quoting Stephen Boyd (2022-04-05 16:00:55)
> >
> > It would be good to pick this back to stable kernels so we have a
> > working suspend/resume on LTS kernels. I tried picking it back to
> > 5.10.109 (latest 5.10 LTS) and booting it on my Lazor w/ LTE device but
> > it crashes at boot pretty reliably in the IPA driver. Interestingly I
> > can't get it to crash on 5.15.32 when I pick it back, so maybe something
> > has changed between 5.10 and 5.15 for IPA? I'll try to bisect it.
>
> Bisecting pointed to commit 1aac309d3207 ("net: ipa: use autosuspend")
> as fixing it. I think before that commit we weren't enabling some
> interconnect, but now we're booting, runtime suspending, and then
> runtime resuming again. With the sync state patch I suspect the
> interconnect bandwidth is dropped and IPA needs to use runtime PM to
> actually turn resources back on because it assumed that resources are on
> when it probes.

I also found that when I make CONFIG_QCOM_IPA=y (and subsequently
CONFIG_QCOM_Q6V5_MSS=y) I can reproduce a different crash in IPA on
5.15.32 and 5.17.1 LTS kernels. I suppose there is some missing
interconnect bandwidth request somewhere and the runtime PM patch half
fixed it, except for when the modem and IPA drivers are builtin. When
the two drivers are builtin they drop bandwidth requests earlier because
they probe earlier. My guess is that the IPA driver is missing a
runtime_pm_get_sync() call somewhere and accessing a register that isn't
clocked. More digging...