Re: [RFC v3 2/3] dt-bindings: sdm845-pinctrl: add wakeup interrupt parent for GPIO

From: Stephen Boyd <swboyd@chromium.org>
To: Bjorn Andersson <bjorn.andersson@linaro.org>,
	Lina Iyer <ilina@codeaurora.org>
Cc: evgreen@chromium.org, marc.zyngier@arm.com,
	linux-kernel@vger.kernel.org, rplsssn@codeaurora.org,
	linux-arm-msm@vger.kernel.org, devicetree@vger.kernel.org,
	thierry.reding@gmail.com
Subject: Re: [RFC v3 2/3] dt-bindings: sdm845-pinctrl: add wakeup interrupt parent for GPIO
Date: Mon, 03 Dec 2018 15:48:29 -0800	[thread overview]
Message-ID: <154388090959.88331.13819513007141877197@swboyd.mtv.corp.google.com> (raw)
In-Reply-To: <20181130183317.GL18262@codeaurora.org>

Quoting Lina Iyer (2018-11-30 10:33:17)
> On Thu, Nov 29 2018 at 14:45 -0700, Lina Iyer wrote:
> >On Wed, Nov 28 2018 at 17:25 -0700, Bjorn Andersson wrote:
> >>On Wed 28 Nov 09:39 PST 2018, Lina Iyer wrote:
> >>
> >>>On Tue, Nov 27 2018 at 14:45 -0700, Stephen Boyd wrote:
> >>>> Quoting Lina Iyer (2018-11-27 10:21:23)
> >>>> > On Tue, Nov 27 2018 at 02:12 -0700, Stephen Boyd wrote:
> 
> [...]
> >BTW, I am discussing with the internal folks here on if we need to mask
> >TLMM when the wakeup-parent is MPM. If we don't have to, we should be
> >able to follow the same model as we done in this patch and don't even
> >have to check the compatible or use the approach suggested by Stephen.
> >
> The TLMM and the MPM are not active at the same time. However, there is
> a small chance they might be (a few clock cycles) when the system is
> going down, but even then, since we replay the interrupt from the MPM
> driver before the interrupts are serviced by Linux, we would not see
> multiple GPIO interrupts.
> 
> The way we have MPM working downstream, for a wakeup GPIO IRQ -
> 
> a. Application cores gets a wakeup interrupt either from RPM or GIC (if
> TLMM was not powered down) while still in the interrupt locked context.
> 
> b. In the hardware, apps core handshakes with the RPM and then starts
> resuming from the platform's system idle driver.
> 
> c. The first CPU to wake up calls MPM driver from the idle driver, which
> reads the shared memory to find the MPM pins that are set. Converts the
> MPM pins to their corresponding linux interrupt and replays the
> interrupt.
> 
> d. Idle driver exits and wakeup GPIO interrupt is handled.
> 
> The MPM pins are not updated after the RPM lets the application core to
> run. Since TLMM is functional after the RPM handshake, it takes over.
> 

Thanks for the background info. I don't think it really changes anything
that we've discussed though. We still need to mask the IRQ in TLMM all
the time when we're using the PDC and we need to leave it unmasked and
replay edges that the MPM sees when we use the MPM. Should I clean up my
RFC patch and post it to the list? I'd like to see hierarchical gpio
irqs work in general for this problem and also the SSBI/SPMI gpio irq
problem that Linus pointed out last week.