Date:   Tue, 17 Dec 2019 08:11:27 +0000
From:   Lee Jones <lee.jones@...aro.org>
To:     Hans de Goede <hdegoede@...hat.com>
Cc:     Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
        Jani Nikula <jani.nikula@...ux.intel.com>,
        Joonas Lahtinen <joonas.lahtinen@...ux.intel.com>,
        Rodrigo Vivi <rodrigo.vivi@...el.com>,
        Ville Syrjälä 
        <ville.syrjala@...ux.intel.com>,
        "Rafael J . Wysocki" <rjw@...ysocki.net>,
        Len Brown <lenb@...nel.org>,
        Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
        linux-acpi@...r.kernel.org,
        intel-gfx <intel-gfx@...ts.freedesktop.org>,
        dri-devel@...ts.freedesktop.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/3] mfd: intel_soc_pmic: Rename pwm_backlight pwm-lookup
 to pwm_pmic_backlight

On Mon, 16 Dec 2019, Hans de Goede wrote:

> Hi,
> 
> On 16-12-2019 10:30, Lee Jones wrote:
> > [...]
> > 
> > > > > > > > > > > Which use a Crystal Cove PMIC, yet the LCD is connected to the SoC/LPSS
> > > > > > > > > > > PWM controller (and the VBT correctly indicates this), so here our old
> > > > > > > > > > > heuristics fail.
> > > > > > > > > > > 
> > > > > > > > > > > Since only the i915 driver has access to the VBT, this commit renames
> > > > > > > > > > > the "pwm_backlight" lookup entries for the Crystal Cove PMIC's PWM
> > > > > > > > > > > controller to "pwm_pmic_backlight" so that the i915 driver can do a
> > > > > > > > > > > pwm_get() for the right controller depending on the VBT bit, instead of
> > > > > > > > > > > the i915 driver relying on a "pwm_backlight" lookup getting registered
> > > > > > > > > > > which magically points to the right controller.
> > > > > > > > > > > 
> > > > > > > > > > > Signed-off-by: Hans de Goede <hdegoede@...hat.com>
> > > > > > > > > > > ---
> > > > > > > > > > >       drivers/mfd/intel_soc_pmic_core.c | 2 +-
> > > > > > > > > > >       1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > > > > > > 
> > > > > > > > > > For my own reference:
> > > > > > > > > >        Acked-for-MFD-by: Lee Jones <lee.jones@...aro.org>
> > > > > > > > > 
> > > > > > > > > As mentioned in the cover-letter, to avoid breaking bi-sectability
> > > > > > > > > as well as to avoid breaking the intel-gfx CI we need to merge this series
> > > > > > > > > in one go through one tree. Specifically through the drm-intel tree.
> > > > > > > > > Is that ok with you ?
> > > > > > > > > 
> > > > > > > > > If this is ok with you, then you do not have to do anything, I will just push
> > > > > > > > > the entire series to drm-intel. drivers/mfd/intel_soc_pmic_core.c
> > > > > > > > > does not see much changes so I do not expect this to lead to any conflicts.
> > > > > > > > 
> > > > > > > > It's fine, so long as a minimal immutable pull-request is provided.
> > > > > > > > Whether it's pulled or not will depend on a number of factors, but it
> > > > > > > > needs to be an option.
> > > > > > > 
> > > > > > > The way the drm subsys works that is not really a readily available
> > > > > > > option. The struct definition which this patch changes a single line in
> > > > > > > has not been touched since 2015-06-26 so I really doubt we will get a
> > > > > > > conflict from this.
> > > > > > 
> > > > > > Always with the exceptions ...
> > > > > > 
> > > > > > OOI, why does this *have* to go through the DRM tree?
> > > > > 
> > > > > This patch renames the name used to lookup the pwm controller from
> > > > > "pwm_backlight" to "pwm_pmic_backlight" because there are 2 possible
> > > > > pwm controllers which may be used, one in the SoC itself and one
> > > > > in the PMIC. Which controller should be used is described in a table
> > > > > in the Video BIOS, so another part of this series adds this code to
> > > > > the i915 driver:
> > > > > 
> > > > > -	panel->backlight.pwm = pwm_get(dev->dev, "pwm_backlight");
> > > > > +	/* Get the right PWM chip for DSI backlight according to VBT */
> > > > > +	if (dev_priv->vbt.dsi.config->pwm_blc == PPS_BLC_PMIC) {
> > > > > +		panel->backlight.pwm = pwm_get(dev->dev, "pwm_pmic_backlight");
> > > > > +		desc = "PMIC";
> > > > > +	} else {
> > > > > +		panel->backlight.pwm = pwm_get(dev->dev, "pwm_soc_backlight");
> > > > > +		desc = "SoC";
> > > > > +	}
> > > > > 
> > > > > So both not to break bisectability, but also so as to not break the extensive
> > > > > CI system which is used to test the i915 driver we need the MFD change doing
> > > > > the rename to go upstrream through the same tree as the i915 change.
> > > > > 
> > > > > I have even considered just squashing the 2 commits together as having only 1
> > > > > present, but not the other breaks stuff left and right.
> > > > 
> > > > That doesn't answer the question.
> > > > 
> > > > Why do they all *have* to go in via the DRM tree specifically?
> > > 
> > > 1. As explained these chanegs need to stay together
> > > 2. This change is primarily a drm/i915 change. Also the i915 code sees lots
> > > of changes every cycle, where as the change to the mfd code touches a block
> > > of code which has not been touched since 2015-06-26, so the chance of conflicts
> > > is much bigger if this goes on through another tree.
> > > 
> > > I honestly do not see the problem here? Let me reverse the question why should this
> > > NOT go in through the drm tree?
> > 
> > There isn't a problem with *this* patch.  I could say, "sure, take it"
> > and the chances are everything could be fine from a technical
> > perspective.
> > 
> > However, I'm taking exception to the fact you think this series is
> > *special* enough to warrant circumventing the usual way in which we
> > usually work when dealing with cross-subsystem patch-sets.  Something
> > I personally deal with a lot due to the inherent hierarchical nature
> > of Multi-Functional Devices.
> > 
> > I'm on the fence on this one.  Due to the circumstances surrounding
> > *this* patch alone, it would be so much easier (for both of us!) to
> > just Ack the patch and hope no further changes occur which could
> > potentially cause someone else (you, me, Linus) more work later on.
> > However, I'm very keen to prevent setting a precedent for this kind of
> > action, as it's clearly not the right path to take in a vast majority
> > of cases.
> > 
> > > 1. As explained these chanegs need to stay together
> > 
> > The patch-set would stay together regardless.  That's the point of an
> > immutable branch, it can be taken in by all relevant parties and Git
> > will just do-the-right-thing.
> > 
> > > 2. This change is primarily a drm/i915 change. Also the i915 code sees lots
> > > of changes every cycle, where as the change to the mfd code touches a block
> > > of code which has not been touched since 2015-06-26, so the chance of conflicts
> > > is much bigger if this goes on through another tree.
> > 
> > This too is irrelevant, since the patch-set could/would go though
> > both/all trees simultaneously.  The way in which we normally work with
> > other subsystems doesn't involve a gamble over which subsystem is most
> > likely going to be affected by a merge conflict as you suggest, it
> > eradicates conflicts for all.
> 
> I'm well aware of using immutable branches and that those are
> often used for patch-set's which touch multiple subsystems. But
> although immutable branches are used often they are about as often
> not used for various reasons, with people choosing to just merge
> through a single tree.
> > I'm not saying "no" by the way.  I just want to find out your
> > reasons/motivation as to why you're insisting this needs go through
> > a) a specific tree and b) just one tree.  Questions which I am yet to
> > see a compelling answer.
> 
> Doing immutable branches assumes that there is a base point,
> e.g. 5.5-rc1 where the immutable branch can then be based on and
> that the branch can then be merged without issues into both subsystems.
> 
> drm is constantly evolving to deal with and mostly catch up with new
> hardware as both GPUs and display-pipelines are evolving quite rapidly
> atm drm-intel-next has about 400 commits on top of 5.5-rc1 so for an
> immutable branch I can either base it on drm-intel-next which
> violates your request for a clean minimal branch to merge; or I can
> base it on 5.5-rc1 which leads to a big chance of problems when
> merging it given to large amount of churn in drm-intel-next.

This is a *slightly* more compelling reason than the ones you've
previously provided.

> So instead of the normal case of 2 subsystems seeing some changes
> on both side the case we have here is a part of a file which has
> not changed since 2015-06-26 in one subsys (and changing only
> a single line there!) and OTOH we have bigger changes to a subsys
> which see 400 patches land in the first week since rc1 .

This is not.

> I hope that you agree that in this case given the large amount of
> churn in drm-intel-next it makes since to just straight forward
> apply these patches on top of drm-intel-next.

I have Acked this patch, but remember *this* is the exception rather
than the rule.  If/when we have a case where a contributor works
cross-subsystem with DRM and the code/file adapted is live (more
likely to change), I will have to insist on an immutable branch
strategy.  DRM will have to deal with that appropriately.

-- 
Lee Jones [李琼斯]
Linaro Services Technical Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog

Powered by blists - more mailing lists