Re: [PATCH] drm/panthor: attach the driver's multiple power domains

[Boris Brezillon <boris.brezillon@collabora.com>]

Hello Rain,

On Fri, 10 Oct 2025 17:14:52 +0800
Rain Yang <jiyu.yang@oss.nxp.com> wrote:

> On Thu, Oct 09, 2025 at 05:23:20PM +0200, Boris Brezillon wrote:
> >On Thu, 9 Oct 2025 23:06:17 +0800
> >Rain Yang <jiyu.yang@oss.nxp.com> wrote:
> >  
> >> On Thu, Oct 09, 2025 at 04:09:29PM +0200, Boris Brezillon wrote:  
> >> >On Thu, 9 Oct 2025 16:08:20 +0200
> >> >Boris Brezillon <boris.brezillon@collabora.com> wrote:
> >> >    
> >> >> On Thu,  9 Oct 2025 22:00:39 +0800
> >> >> Rain Yang <jiyu.yang@oss.nxp.com> wrote:
> >> >>     
> >> >> > From: Rain Yang <jiyu.yang@nxp.com>
> >> >> > 
> >> >> > Some platforms, such as i.MX95, utilize multiple power domains that need
> >> >> > to be attached explicitly. This patch ensures that the driver properly
> >> >> > attaches all available power domains using devm_pm_domain_attach_list().
> >> >> > 
> >> >> > Signed-off-by: Prabhu Sundararaj <prabhu.sundararaj@nxp.com>
> >> >> > Signed-off-by: Rain Yang <jiyu.yang@nxp.com>
> >> >> > ---
> >> >> >  drivers/gpu/drm/panthor/panthor_device.c | 6 ++++++
> >> >> >  drivers/gpu/drm/panthor/panthor_device.h | 2 ++
> >> >> >  2 files changed, 8 insertions(+)
> >> >> > 
> >> >> > diff --git a/drivers/gpu/drm/panthor/panthor_device.c b/drivers/gpu/drm/panthor/panthor_device.c
> >> >> > index f0b2da5b2b96..6f40d053b16c 100644
> >> >> > --- a/drivers/gpu/drm/panthor/panthor_device.c
> >> >> > +++ b/drivers/gpu/drm/panthor/panthor_device.c
> >> >> > @@ -218,6 +218,12 @@ int panthor_device_init(struct panthor_device *ptdev)
> >> >> >  	if (ret)
> >> >> >  		return ret;
> >> >> >  
> >> >> > +	ret = devm_pm_domain_attach_list(ptdev->base.dev, NULL, &ptdev->pd_list);
> >> >> > +	if (ret < 0) {
> >> >> > +		drm_err(&ptdev->base, "attach power domains failed, ret=%d", ret);
> >> >> > +		return ret;
> >> >> > +	}
> >> >> > +
> >> >> >  	ret = panthor_devfreq_init(ptdev);
> >> >> >  	if (ret)
> >> >> >  		return ret;
> >> >> > diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
> >> >> > index 4fc7cf2aeed5..5ecb541ec67b 100644
> >> >> > --- a/drivers/gpu/drm/panthor/panthor_device.h
> >> >> > +++ b/drivers/gpu/drm/panthor/panthor_device.h
> >> >> > @@ -196,6 +196,8 @@ struct panthor_device {
> >> >> >  		/** @recovery_needed: True when a resume attempt failed. */
> >> >> >  		atomic_t recovery_needed;
> >> >> >  	} pm;      
> >> >> 
> >> >> Add a blank line here.
> >> >>     
> >> >> > +	/** @pm: Power management related data. */    
> >> >
> >> >Also, the comment is wrong, and it would probably make sense to move
> >> >that to the pm sub-struct since this is PM related.    
> >> thanks, will fix it next version.  
> >> >    
> >> >> > +	struct dev_pm_domain_list  *pd_list;
> >> >> >        
> >> >> 
> >> >> Do we even need to keep the pd_list in panthor_device if we don't do
> >> >> anything with it?    
> >> The second power domain is typically used for frequency scaling. The driver
> >> works fine when there's only one single power domain. That said, I will update
> >> the implementation to attach multiple power domains only when the domain count
> >> is greater than one.  
> >
> >That's not what I meant. What I mean is that the
> >panthor_device::pd_list is never used, because the extra power domains
> >get attached to the struct device directly, and the PM core
> >automatically enables all PDs on resume and disable them on suspend.
> >Because it's a devm_ function, you don't need it to detach the pd_list
> >at ::remove() time either. TLDR; that means you can pass
> >devm_pm_domain_attach_list() a local pd_list instead of ptdev->pd_list
> >and you can thus drop panthor_device::pd_list altogether, I think.
> >
> >If you intend to use the pd_list for manual PD control in panthor in a
> >follow-up patchset, this should be mentioned in the commit message.  
> thanks for your suggestion, pd_list will be allocated by devm_kmalloc.

There's no extra allocation needed. Just do:

	struct dev_pm_domain_list  *pd_list;

	...
	
	ret = devm_pm_domain_attach_list(ptdev->base.dev, NULL, &pd_list);
	if (ret < 0) {
		drm_err(&ptdev->base, "attach power domains failed, ret=%d", ret);
		return ret;
	}

and that's it.

> 
> I have not found a general solution to modify the frequency via SCMI perf
> and OPP framework except the manual PD control, I'd appreciate it if
> there is any idea.

What's the manual PD control you're talking about? I don't see
anything using the pd_list in this patch. Is there another patchset
on top of this one that you haven't posted yet? Dunno if that's of
any help, but this patchset [1] might give you some ideas.

Regards,

Boris

[1]https://lwn.net/Articles/1040831