Re: [Intel-gfx] [RFC 1/3] drm/i915/power: move dc state members to struct i915_power_domains

From: Jani Nikula <jani.nikula@intel.com>
To: imre.deak@intel.com
Cc: intel-gfx@lists.freedesktop.org
Subject: Re: [Intel-gfx] [RFC 1/3] drm/i915/power: move dc state members to struct i915_power_domains
Date: Tue, 07 Feb 2023 12:05:43 +0200	[thread overview]
Message-ID: <87ttzxyemg.fsf@intel.com> (raw)
In-Reply-To: <Y+Exp5Gk+JEVFvcq@ideak-desk.fi.intel.com>

On Mon, 06 Feb 2023, Imre Deak <imre.deak@intel.com> wrote:
> On Mon, Feb 06, 2023 at 06:21:11PM +0200, Jani Nikula wrote:
>> On Fri, 03 Feb 2023, Imre Deak <imre.deak@intel.com> wrote:
>> > On Thu, Feb 02, 2023 at 10:47:46PM +0200, Jani Nikula wrote:
>> >> There's only one reference to the struct intel_dmc members dc_state,
>> >> target_dc_state, and allowed_dc_mask within intel_dmc.c, begging the
>> >> question why they are under struct intel_dmc to begin with.
>> >> 
>> >> Moreover, the only references to i915->display.dmc outside of
>> >> intel_dmc.c are to these members.
>> >> 
>> >> They don't belong. Move them from struct intel_dmc to struct
>> >> i915_power_domains, which seems like a more suitable place.
>> >> 
>> >> Cc: Imre Deak <imre.deak@intel.com>
>> >> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
>> >> ---
>> >> [...]
>> >>
>> >> @@ -481,7 +482,7 @@ void intel_dmc_load_program(struct drm_i915_private *dev_priv)
>> >>  		}
>> >>  	}
>> >>  
>> >> -	dev_priv->display.dmc.dc_state = 0;
>> >> +	power_domains->dc_state = 0;
>> >
>> > This could be dropped as well, as it's already inited by the time the
>> > function is called.
>> 
>> The whole dc_state thing originates to 832dba889e27 ("drm/i915/gen9:
>> Check for DC state mismatch"), and it's all about detecting
>> mismatches. I'm not sure how that works and if it's still useful.
>
> On SKL writing the DC_STATE_EN register didn't take effect, so a
> workaround was added to reread/verify the write. Yes, on newer platforms
> we should probably revisit this and try to remove at least the part
> rereading the register 5 times.
>
>> > I agree with making struct intel_dmc internal to intel_dmc.c. Since DC
>> > state is a functionality provided by the firmware (except of DC9), an
>> > alternative would be to move/add get/set/assert etc. DC state functions
>> > to intel_dmc.c instead and call these from intel_display_power*.c.
>> 
>> That was actually the first patch I wrote but didn't send. There were a
>> few reasons I switched to this one:
>> 
>> First, it adds a dependency between power well and dmc initialization.
>
> Do you mean the dependency that is there already?: taking some power ref
> and keeping the whole runtime PM functionality disabled until the
> firmware load completes. This is based on an earlier decision not to
> support runtime PM w/o DMC.

intel_power_domains_init() is called very early. It adjusts
allowed_dc_mask and target_dc_state.

intel_dmc_ucode_init() is called later, and depends on
intel_power_domains_init() in the way you describe above.

If intel_dmc_ucode_init() starts allocating intel_dmc dynamically, it
needs to exist before intel_power_domains_init() modifies
allowed_dc_mask and target_dc_state.

It's an interdependency that would need to be broken up somehow.

>
>> Second, it's a bunch of boilerplate, six get/set functions altogether,
>> and all of them checking for dmc init in patch 3.
>> 
>> Third, it still seems funny to have these members stored in intel_dmc,
>> accessed via intel_dmc.c, but intel_dmc.c not actually using them for
>> anything internally. It's only the power well code that uses
>> them. Should more of the DC state handling be moved to intel_dmc.c then?
>
> I thought that any functions accessing the DC_STATE_EN register would be
> moved as well (besides the state checkers you refer to). I wasn't sure
> if my suggestion makes sense without actually seeing the end result; I
> think we can also regard the DC_STATE_EN register as a more general
> display PM HW interface leaving that in intel_display_power* (since DC9
> which isn't a DMC thing is also toggled via it) and leave only the
> firmware loading/initialization in intel_dmc.c as in your RFC.

Yeah, I don't know. *lots of shrugging* :)

>
>> BR,
>> Jani.
>> 
>> >>  	gen9_set_dc_state_debugmask(dev_priv);
>> >>  
>> >> diff --git a/drivers/gpu/drm/i915/display/intel_dmc.h b/drivers/gpu/drm/i915/display/intel_dmc.h
>> >> index 88eae74dbcf2..da8ba246013e 100644
>> >> --- a/drivers/gpu/drm/i915/display/intel_dmc.h
>> >> +++ b/drivers/gpu/drm/i915/display/intel_dmc.h
>> >> @@ -40,9 +40,6 @@ struct intel_dmc {
>> >>  		bool present;
>> >>  	} dmc_info[DMC_FW_MAX];
>> >>  
>> >> -	u32 dc_state;
>> >> -	u32 target_dc_state;
>> >> -	u32 allowed_dc_mask;
>> >>  	intel_wakeref_t wakeref;
>> >>  };
>> >>  
>> >> diff --git a/drivers/gpu/drm/i915/display/intel_psr.c b/drivers/gpu/drm/i915/display/intel_psr.c
>> >> index 2954759e9d12..cf13580af34a 100644
>> >> --- a/drivers/gpu/drm/i915/display/intel_psr.c
>> >> +++ b/drivers/gpu/drm/i915/display/intel_psr.c
>> >> @@ -702,6 +702,7 @@ tgl_dc3co_exitline_compute_config(struct intel_dp *intel_dp,
>> >>  {
>> >>  	const u32 crtc_vdisplay = crtc_state->uapi.adjusted_mode.crtc_vdisplay;
>> >>  	struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
>> >> +	struct i915_power_domains *power_domains = &dev_priv->display.power.domains;
>> >>  	u32 exit_scanlines;
>> >>  
>> >>  	/*
>> >> @@ -718,7 +719,7 @@ tgl_dc3co_exitline_compute_config(struct intel_dp *intel_dp,
>> >>  	if (crtc_state->enable_psr2_sel_fetch)
>> >>  		return;
>> >>  
>> >> -	if (!(dev_priv->display.dmc.allowed_dc_mask & DC_STATE_EN_DC3CO))
>> >> +	if (!(power_domains->allowed_dc_mask & DC_STATE_EN_DC3CO))
>> >>  		return;
>> >>  
>> >>  	if (!dc3co_is_pipe_port_compatible(intel_dp, crtc_state))
>> >> -- 
>> >> 2.34.1
>> >> 
>> 
>> -- 
>> Jani Nikula, Intel Open Source Graphics Center

-- 
Jani Nikula, Intel Open Source Graphics Center