Hi,
On Tue, Jul 7, 2020 at 10:29 PM Maulik Shah <mkshah@xxxxxxxxxxxxxx> wrote:
>
> Hi,
>
> On 7/8/2020 4:33 AM, Doug Anderson wrote:
> > Hi,
> >
> > On Mon, Jul 6, 2020 at 9:52 PM Rajendra Nayak <rnayak@xxxxxxxxxxxxxx> wrote:
> >>
> >> []..
> >>
> >>>>> @@ -1151,6 +1168,10 @@ static const struct msm_pinctrl_soc_data sc7180_pinctrl = {
> >>>>>
> >>>>> static int sc7180_pinctrl_probe(struct platform_device *pdev)
> >>>>> {
> >>>>> + if (of_machine_is_compatible("google,lazor")) {
> >>>>> + sc7180_pinctrl.wakeirq_map = sc7180_lazor_pdc_map;
> >>>>> + sc7180_pinctrl.nwakeirq_map = ARRAY_SIZE(sc7180_lazor_pdc_map);
> >>>>> + }
> >>>> As much as I want patches landed and things working, the above just
> >>>> doesn't feel like a viable solution. I guess it could work as a short
> >>>> term hack but it's going to become untenable pretty quickly.
> >>> I second that.
> >>>
> >>>> As we
> >>>> have more variants of this we're going to have to just keep piling
> >>>> more machines in here, right? ...this is also already broken for us
> >>>> because not all boards will have the "google,lazor" compatible. From
> >>>> the current Chrome OS here are the compatibles for various revs/SKUs
> >>>>
> >>>> compatible = "google,lazor-rev0", "qcom,sc7180";
> >>>> compatible = "google,lazor-rev0-sku0", "qcom,sc7180";
> >>>> compatible = "google,lazor", "qcom,sc7180";
> >>>> compatible = "google,lazor-sku0", "qcom,sc7180";
> >>>> compatible = "google,lazor-rev2", "qcom,sc7180";
> >>>>
> >>>> ...so of the 5 boards you'll only match one of them.
> >>>>
> >>>>
> >>>> Maybe I'm jumping into a situation again where I'm ignorant since I
> >>>> haven't followed all the prior conversation, but is it really that
> >>>> hard to just add dual edge support to the PDC irqchip driver? ...or
> >> FWIK, this is really a PDC hardware issue (with the specific IP rev that exists
> >> on sc7180) so working it around in SW could get ugly.
> > Ugh. I guess it's ugly because the workaround would need to be in the
> > PDC driver but to properly do the workaround you need to be able to
> > read the state of the pin from the PDC driver? ...and I guess you
> > can't do that with the PDC register space so you'd either need to
> > violate a layer or 3 of abstraction and snarf into the GPIO register
> > space from the PDC driver or you'd have to provide some sort of API
> > access from the PDC back down to the GPIO driver?
> >
> > --
> >
> > Actually, though, I'm still not sure why this would need to be in the
> > PDC driver. Sure, you can't just magically re-use the existing
> > dual-edge emulation in pinctrl-msm.c, but you can add some new
> > dual-edge emulation for when your parent handles your interrupts,
> > can't you? As per usually, I'm talking out of my rear end, but I
> > sorta imagine:
> >
> > 1. At the head of msm_gpio_irq_set_type() if you detect that
> > "skip_wake_irqs" is set and you're on an SoC with this hardware errata
> > then you do a loop much like the one in
> > msm_gpio_update_dual_edge_pos() except that instead of changing the
> > polarity with msm_writel_intr_cfg() you change the polarity with
> > "irq_chip_set_type_parent()".
> >
> > 2. At the head of msm_gpio_irq_ack() you make the same function call
> > if "skip_wake_irqs" is set and you're on an SoC with this hardware
> > errata.
> >
> > It doesn't feel all that ugly to me, assuming I'm understanding it
> > correctly. ...or maybe you can tell me why it'd be harder than that?
> >
> >
> >>>> maybe it's just easier to change the pinctrl driver to emulate dual
> >>>> edge itself and that can work around the problem in the PDC? There
> >>>> seem to be a few samples you could copy from:
> >>>>
> >>>> $ git log --oneline --no-merges --grep=emulate drivers/pinctrl/
> >>>> 3221f40b7631 pinctrl: mediatek: emulate GPIO interrupt on both-edges
> >>>> 5a92750133ff pinctrl: rockchip: emulate both edge triggered interrupts
> >>>>
> >>> pinctrl-msm already supports emulating dual edge, but my understanding
> >>> was that the problem lies in that somehow this emulation would have to
> >>> be tied to or affect the PDC driver?
> >> yes, thats correct, pinctrl-msm already supports it, the problem lies
> >> in the fact that PDC does not. This patch, infact was trying to fix the
> >> issue by removing all PDC involvement for gpio28 and making pinctrl-msm
> >> in charge of it.
> > If we're going to try to do this, I think we're stuck with one of these:
> >
> > 1. A really really long list that we keep jamming more boards into.
> >
> > 2. Add an entry at the top-level device tree compatible to all
> > affected boards _just_ for this purpose. Seems ugly since we don't
> > need it for any other reasons.
> >
> > 3. Add some sort of property to the pinctrl node on these boards.
> > Seems ugly since conceivably this _could_ be worked around in
> > software.
> >
> > I don't really like any of those options, so I'm really hoping we can
> > find out how to get a workaround in...
> Hi Doug,
>
> The client driver here never uses/needs both the edges at a given time.
> Another option (clean & correct IMO) is to update the driver to request
> proper irq type its expecting.
>
> Lets take SD card detect GPIO for example, which uses dual edge interrupt.
> one edge (rising type) can be used as a card insert detect interrupt and
> another edge (falling type) may be used for card removal detect.
>
> The sequence of operations, IMO should be..
> 1. Driver request a rising type irq to start with (the one that detects
> card insertion)
> 2. once card insertion irq comes in, the driver should change the type
> to falling which is expected (to detect the card removal)
> 3. once card removal irq comes in, it can change type to rising edge (to
> detect another insertion)
> 4. above steps (2,3) continues
>
> if above sequence is followed from drivers using dual edge IRQ, we don't
> need any workaround.
You're saying that we should make the rest of Linux change to work
around the bug in your pinctrl driver? I don't think that'll fly too
well with the maintainers of the other subsystems. The
IRQ_TYPE_EDGE_BOTH define exists and we should support it.
Was there something wrong with my proof of concept patch? Unless I
get swamped with other things, my plan was to try to make that more
real and post it unless someone told me why it was wrong...
-Doug
