Re: [PATCH v4 1/2] soc: qcom: smem: Register gunyah watchdog device

[Bjorn Andersson]

On Mon, Nov 03, 2025 at 03:23:27PM +0530, Hrishabh Rajput wrote:
> 
> On 11/2/2025 12:20 AM, Bjorn Andersson wrote:
> > On Fri, Oct 31, 2025 at 08:24:44AM -0700, Guenter Roeck wrote:
> > > On 10/31/25 03:18, Hrishabh Rajput via B4 Relay wrote:
> > > > From: Hrishabh Rajput <hrishabh.rajput@xxxxxxxxxxxxxxxx>
> > > > 
> > > > To restrict gunyah watchdog initialization to Qualcomm platforms,
> > > > register the watchdog device in the SMEM driver.
> > > > 
> > > > When Gunyah is not present or Gunyah emulates MMIO-based
> > > > watchdog, we expect Qualcomm watchdog or ARM SBSA watchdog device to be
> > > > present in the devicetree. If none of these device nodes are detected,
> > > > we register the SMC-based Gunyah watchdog device.
> > > > 
> > > There should also be an explanation why there is no "qcom,gunyah-wdt"
> > > devicetree node, both here and in the file.
> > > 
> > > > Signed-off-by: Hrishabh Rajput <hrishabh.rajput@xxxxxxxxxxxxxxxx>
> > > > ---
> > > >    drivers/soc/qcom/smem.c | 37 +++++++++++++++++++++++++++++++++++++
> > > >    1 file changed, 37 insertions(+)
> > > > 
> > > > diff --git a/drivers/soc/qcom/smem.c b/drivers/soc/qcom/smem.c
> > > > index cf425930539e..40e4749fab02 100644
> > > > --- a/drivers/soc/qcom/smem.c
> > > > +++ b/drivers/soc/qcom/smem.c
> > > > @@ -1118,6 +1118,34 @@ static int qcom_smem_resolve_mem(struct qcom_smem *smem, const char *name,
> > > >    	return 0;
> > > >    }
> > > > +static int register_gunyah_wdt_device(void)
> > > > +{
> > > > +	struct platform_device *gunyah_wdt_dev;
> > > > +	struct device_node *np;
> > > > +
> > > > +	/*
> > > > +	 * When Gunyah is not present or Gunyah is emulating a memory-mapped
> > > > +	 * watchdog, either of Qualcomm watchdog or ARM SBSA watchdog will be
> > > > +	 * present. Skip initialization of SMC-based Gunyah watchdog if that is
> > > > +	 * the case.
> > > > +	 */
> > > > +	np = of_find_compatible_node(NULL, NULL, "qcom,kpss-wdt");
> > > > +	if (np) {
> > > > +		of_node_put(np);
> > > > +		return 0;
> > > > +	}
> > > > +
> > > > +	np = of_find_compatible_node(NULL, NULL, "arm,sbsa-gwdt");
> > > > +	if (np) {
> > > > +		of_node_put(np);
> > > > +		return 0;
> > > > +	}
> > > > +
> > > > +	gunyah_wdt_dev = platform_device_register_simple("gunyah-wdt", -1,
> > > > +							 NULL, 0);
> > > > +	return PTR_ERR_OR_ZERO(gunyah_wdt_dev);
> > > > +}
> > > > +
> > > >    static int qcom_smem_probe(struct platform_device *pdev)
> > > >    {
> > > >    	struct smem_header *header;
> > > > @@ -1236,11 +1264,20 @@ static int qcom_smem_probe(struct platform_device *pdev)
> > > >    	if (IS_ERR(smem->socinfo))
> > > >    		dev_dbg(&pdev->dev, "failed to register socinfo device\n");
> > > > +	ret = register_gunyah_wdt_device();
> > > > +	if (ret)
> > > > +		dev_dbg(&pdev->dev, "failed to register watchdog device\n");
> > > > +
> > > >    	return 0;
> > > >    }
> > > >    static void qcom_smem_remove(struct platform_device *pdev)
> > > >    {
> > > > +	/*
> > > > +	 * Gunyah watchdog is intended to be a persistent module. Hence, the
> > > > +	 * watchdog device is not unregistered.
> > > > +	 */
> > > > +
> > > Odd explanation.
> > > I would assume that the smem device is supposed to be
> > > persistent as well.
> > Yes, but it's perfectly possible to run a modern Qualcomm device without
> > SMEM, with a fair amount of functionality. So, the reevaluation of this
> > decision is grinding in the back of my mind...
> 
> One option can be the SCM driver which was suggested by Neil in v3 [1].
> 
> Let us know if you have any suggestions where we can register the watchdog
> device?
> 

I think it makes more sense to tie it to the SCM driver, it's after all
related to the HVC-interface, which SMEM isn't.

But I don't think that answers our question of how do you determine if
Gunyah is there or not. Because qcom,scm-sc7280 is regardless of Linux
running under Gunyah, KVM, or directly in EL2.


I can't help feeling that this is a property of the hardware/firmware
interface that the OS finds itself in, which should be expressed in
DeviceTree - unless there's a bulletproof "Gunyah, are you there?"
check.

Regards,
Bjorn


> [1]:
> https://lore.kernel.org/lkml/321f5289-64c0-48f1-91b5-c45e82396ca9@xxxxxxxxxx/
> 
> Thanks,
> 
> Hrishabh
>