Re: ASoC: Intel: sst: Missing IRQ at index 5 on BYT-T device

[Stephan Gerhold <stephan@gerhold.net>]

On Mon, Dec 17, 2018 at 08:13:36PM -0600, Pierre-Louis Bossart wrote:
> 
> > > > > The quirks to get sound working with bytcr-rt5640 on that device are:
> > > > > BYT_RT5640_SSP0_AIF1 | BYT_RT5640_IN1_MAP | BYT_RT5640_MCLK_EN
> > > > > 
> > > > > I guess this means that SSP0 is being used?
> > > > Yes indeed, and that makes me think we should force this device to look like
> > > > Baytrail-CR.
> > > > 
> > > > You can do this with a DMI-based quirk (preferably in is_byt_cr directly so
> > > > that I can reuse the code when I move it to a helper at some point).
> > > Okay - thanks! One last question:
> > > I was looking at the ACPI DSDT tables of some similar devices and have
> > > found two others that look the same (only one IRQ listed). In this case,
> > > the BYT-T acpi_ipc_irq_index = 5 will never work, and we will definitely
> > > have a better chances with trying Baytrail-CR.
> > > 
> > > One of them actually had a similar patch proposed at [1] (although they
> > > did it in a weird way and also need an extra machine driver).
> > > 
> > > We could also detect this situation in a generic way with something like
> > > 
> > >    if (platform_irq_count(pdev) == 1) {
> > >    	*bytcr = true;
> > > 	return 0;
> > >    }
> > > 
> > > ... instead of a DMI quirk. What do you think?
> > > 
> > To avoid confusion: The existing PMIC-type based is_byt_cr() detection
> > would be used in all other cases (i.e. if irq_count != 1), so it won't
> > make any difference for the devices that are already working fine.
> > (Most BYT-CR devices seem to have 5 IRQs listed)
> > 
> > So it's more like
> > 
> >    if (platform_irq_count(pdev) == 1) {
> >    	*bytcr = true;
> >    } else {
> >    	// pmic-type based detection...
> >    }
> > 
> > with platform_irq_count == 1 as condition for the "quirk".
> 
> The solution seems appealing but
> 
> 1) does it really work? I am not sure an index=5 means there are 5
> interrupts.

Yes, I believe that it means that there need to be (at least) 5 
interrupts.

I have checked the source code of platform_get_irq.
When you do platform_get_irq(pdev, /* index = */ 5) (as done for BYT-T) 
it first calls

    platform_get_resource(/* type = */ IORESOURCE_IRQ, /* num = */ 5)

to lookup the resource. That method is really simple and looks like

	for (i = 0; i < dev->num_resources; i++) {
		struct resource *r = &dev->resource[i];

		if (type == resource_type(r) && num-- == 0)
			return r;
	}

So when you request an IRQ at index=5, it loops through all resources,
skips the first 5 IRQs and returns the 6th one (on my device, it
returns NULL because there are not enough IRQs listed).

There is a bit more magic in platform_irq_count (it looks up all IRQs 
and does not count invalid ones), so to be absolutely safe we could
just check platform_get_resource(IRQ, 5) == NULL early.
If it returns NULL, then platform_get_irq(pdev, 5) will also return 
-ENXIO, so treating the device as BYT-T is guaranteed to fail.
In this case, we have better chances when we assume BYT-CR.

Example patch: (I have added it in probe instead of is_byt_cr() because 
it requires the platform device, plus I think it might be better to 
differentiate the messages in the kernel log..)

--- a/sound/soc/intel/atom/sst/sst_acpi.c
+++ b/sound/soc/intel/atom/sst/sst_acpi.c
@@ -337,6 +337,16 @@ static int sst_acpi_probe(struct platform_device *pdev)
 	if (!((ret < 0) || (bytcr == false))) {
 		dev_info(dev, "Detected Baytrail-CR platform\n");

 		/* override resource info */
 		byt_rvp_platform_data.res_info = &bytcr_res_info;
+	} else if (platform_get_resource(pdev, IORESOURCE_IRQ, 5) == NULL) {
+		/*
+		 * Some devices detected as BYT-T have only a single IRQ listed.
+		 * In this case, platform_get_irq with index 5 will return -ENXIO.
+		 * Fall back to the BYT-CR resource info to use the correct IRQ.
+		 */
+		dev_info(dev, "Falling back to Baytrail-CR platform\n");
+
+		/* override resource info */
+		byt_rvp_platform_data.res_info = &bytcr_res_info;
 	}

> 
> 2) the test would affect all existing devices, and there's so much hardware
> proliferation that proving this change in harmless might be difficult. I
> personally only have one BYT-T (ASus T100) device left and it's not very
> reliable. Hans seems to have a ton of devices but they are mostly Byt-Cr?
> 

With the updated patch above I believe there is literally no way this 
can break sound on any device. The condition only evaluates to true if 
SST would normally fail to probe later anyway.

I have tested the patch above on my device with:
 - as-is, without any modifications:
    -> "Falling back to Baytrail-CR platform", sound now working
 - a simulated "BYT-T" device: (copied the IRQs from the DSDT of the T100TA)
    -> "BYT-CR not detected" - uses 5th IRQ, sound working
 - a simulated "BYT-CR" device (made is_byt_cr() return "true" and 
   copied the IRQs from the DSDT of the T100TAF)
    -> "Detected Baytrail-CR platform" - uses IRQ at index 0, sound working

Let me know what you think!