Re: [PATCH v2 3/9] remoteproc: qcom_q6v5_mss: Handle platforms with one power domain

[Luca Weiss <luca@lucaweiss.eu>]

On dinsdag 28 januari 2025 08:30:25 Midden-Europese standaardtijd Stephan Gerhold wrote:
> On Mon, Jan 27, 2025 at 11:21:04PM +0100, Luca Weiss wrote:
> > On maandag 27 januari 2025 09:58:45 Midden-Europese standaardtijd Stephan Gerhold wrote:
> > > On Sun, Jan 26, 2025 at 09:57:22PM +0100, Luca Weiss wrote:
> > > > For example MSM8974 has mx voltage rail exposed as regulator and only cx
> > > > voltage rail is exposed as power domain. This power domain (cx) is
> > > > attached internally in power domain and cannot be attached in this driver.
> > > > 
> > > > Fixes: 8750cf392394 ("remoteproc: qcom_q6v5_mss: Allow replacing regulators with power domains")
> > > > Co-developed-by: Matti Lehtimäki <matti.lehtimaki@gmail.com>
> > > > Signed-off-by: Matti Lehtimäki <matti.lehtimaki@gmail.com>
> > > > Signed-off-by: Luca Weiss <luca@lucaweiss.eu>
> > > > ---
> > > > Changes in v2:
> > > >   - Move MSM8974 mx-supply from "fallback_proxy_supply" to
> > > >     "proxy_supply" to match updated DT schema
> > > >   - Add fixes tag
> > > > ---
> > > >  drivers/remoteproc/qcom_q6v5_mss.c | 20 +++++++++++++++++---
> > > >  1 file changed, 17 insertions(+), 3 deletions(-)
> > > > 
> > > > diff --git a/drivers/remoteproc/qcom_q6v5_mss.c b/drivers/remoteproc/qcom_q6v5_mss.c
> > > > index e78bd986dc3f256effce4470222c0a5faeea86ec..e2523b01febf393abfe50740a68b85a04011293c 100644
> > > > --- a/drivers/remoteproc/qcom_q6v5_mss.c
> > > > +++ b/drivers/remoteproc/qcom_q6v5_mss.c
> > > > @@ -1828,6 +1828,13 @@ static int q6v5_pds_attach(struct device *dev, struct device **devs,
> > > >  	if (!pd_names)
> > > >  		return 0;
> > > >  
> > > > +	/* Handle single power domain */
> > > > +	if (dev->pm_domain) {
> > > > +		devs[0] = dev;
> > > > +		pm_runtime_enable(dev);
> > > > +		return 1;
> > > > +	}
> > > > +
> > > >  	while (pd_names[num_pds])
> > > >  		num_pds++;
> > > 
> > > Hmm, I think you should put the above if condition below this loop and
> > > verify that num_pds == 1. Otherwise this would hide the error condition
> > > if the platform needs multple PDs, but someone only specifies one of
> > > them in the DT. i.e.
> > > 
> > > 	if (num_pds == 1 && dev->pm_domain) {
> > > 		// ...
> > > 	}
> > > 
> > > >  
> > > > @@ -1851,8 +1858,15 @@ static int q6v5_pds_attach(struct device *dev, struct device **devs,
> > > >  static void q6v5_pds_detach(struct q6v5 *qproc, struct device **pds,
> > > >  			    size_t pd_count)
> > > >  {
> > > > +	struct device *dev = qproc->dev;
> > > >  	int i;
> > > >  
> > > > +	/* Handle single power domain */
> > > > +	if (dev->pm_domain && pd_count) {
> > > 
> > > Maybe if (pd_count == 1 && dev->pm_domain) for consistency with the
> > > above then.
> > 
> > Good suggestions, thanks!
> > 
> > > 
> > > > +		pm_runtime_disable(dev);
> > > 
> > > I'm guessing it does, but just to make sure: Have you verified that the
> > > power domain is indeed voted for off after this call to
> > > pm_runtime_disable()? Start the remoteproc and when it's booted inspect
> > > /sys/kernel/debug/pm_genpd/pm_genpd_summary to see if the PD/remoteproc
> > > dev is suspended.
> > 
> > Looks sane I think (modem: remoteproc@fc880000)
> > 
> 
> The modem does look sane yeah...
> 
> > htc-memul:~$ sudo cat /sys/kernel/debug/pm_genpd/pm_genpd_summary
> > domain                          status          children        performance
> >     /device                         runtime status                  managed by
> > ------------------------------------------------------------------------------
> > oxili_cx                        off-0                           0
> >     fdb00000.gpu                    suspended                   0           SW
> > camss_vfe                       off-0                           0
> > camss_jpeg                      off-0                           0
> > mdss                            on                              0
> >     fd900000.display-subsystem      active                      0           SW
> > venus0                          off-0                           0
> > cx_vfc                          off-0                           0
> > cx_ao                           off-0                           0
> > cx                              on                              0
> >     fc880000.remoteproc             suspended                   0           SW
> >     fe200000.remoteproc             unsupported                 0           SW
> >     fb204000.remoteproc             suspended                   0           SW
> 
> ... but "unsupported" for ADSP and the end result for cx ("on") does
> look suspicious. Looking at qcom_q6v5_pas.c, it uses almost the same
> code for single power domain support as here...
> 
> What state is the ADSP in here at this point? Did it boot sucessfully?
> 
> If I'm reading the code correctly, the pm_runtime_enable() in
> adsp_pds_attach() should get rid of the "unsupported" at probe time,
> so I'm a bit confused how this can happen..
> 
> [5 minutes of staring later...]
> 
> Um, qcom,msm8226-adsp-pil uses &adsp_resource_init, which doesn't have
> any "proxy_pd_names". It should use &msm8996_adsp_resource so that it
> actually manages the CX power domain. Same for MSM8974, but that was
> never converted to use power domains... 🙈

Oh wow, thanks for catching this!

> 
> Can you submit a patch that changes at least msm8226 to use
> &msm8996_adsp_resource? Would be also good to make the same changes
> I suggested here (check num_pds == 1 / pd_count == 1).

Sent a series to fix the above and this:
https://lore.kernel.org/linux-arm-msm/20250128-pas-singlepd-v1-0-85d9ae4b0093@lucaweiss.eu/T/

Regards
Luca

> 
> Thanks,
> Stephan
>