On Thu, 23 Oct 2025, Tvrtko Ursulin <[email protected]> wrote: > On 23/10/2025 09:55, Jani Nikula wrote: >> On Thu, 23 Oct 2025, Tvrtko Ursulin <[email protected]> wrote: >>> On 23/10/2025 08:45, Jani Nikula wrote: >>>> Reduce the display dependency on struct drm_i915_private and i915_drv.h >>>> by converting the rps interface to struct drm_device. >>>> >>>> Signed-off-by: Jani Nikula <[email protected]> >>>> --- >>>> .../gpu/drm/i915/display/intel_display_rps.c | 12 +++++------- >>>> drivers/gpu/drm/i915/gt/intel_rps.c | 18 ++++++++++++++++-- >>>> drivers/gpu/drm/i915/gt/intel_rps.h | 7 ++++--- >>>> 3 files changed, 25 insertions(+), 12 deletions(-) >>>> >>>> diff --git a/drivers/gpu/drm/i915/display/intel_display_rps.c >>>> b/drivers/gpu/drm/i915/display/intel_display_rps.c >>>> index 82ea1ec482e4..8bf0f8cb6574 100644 >>>> --- a/drivers/gpu/drm/i915/display/intel_display_rps.c >>>> +++ b/drivers/gpu/drm/i915/display/intel_display_rps.c >>>> @@ -3,12 +3,14 @@ >>>> * Copyright © 2023 Intel Corporation >>>> */ >>>> >>>> +#include <linux/dma-fence.h> >>>> + >>>> #include <drm/drm_crtc.h> >>>> #include <drm/drm_vblank.h> >>>> >>>> #include "gt/intel_rps.h" >>>> -#include "i915_drv.h" >>>> #include "i915_reg.h" >>>> +#include "i915_request.h" >>>> #include "intel_display_core.h" >>>> #include "intel_display_irq.h" >>>> #include "intel_display_rps.h" >>>> @@ -77,12 +79,10 @@ void intel_display_rps_mark_interactive(struct >>>> intel_display *display, >>>> struct intel_atomic_state >>>> *state, >>>> bool interactive) >>>> { >>>> - struct drm_i915_private *i915 = to_i915(display->drm); >>>> - >>>> if (state->rps_interactive == interactive) >>>> return; >>>> >>>> - intel_rps_mark_interactive(&to_gt(i915)->rps, interactive); >>>> + intel_rps_mark_interactive(display->drm, interactive); >>>> state->rps_interactive = interactive; >>>> } >>>> >>>> @@ -102,7 +102,5 @@ void ilk_display_rps_disable(struct intel_display >>>> *display) >>>> >>>> void ilk_display_rps_irq_handler(struct intel_display *display) >>>> { >>>> - struct drm_i915_private *i915 = to_i915(display->drm); >>>> - >>>> - gen5_rps_irq_handler(&to_gt(i915)->rps); >>>> + gen5_rps_irq_handler(display->drm); >>>> } >>>> diff --git a/drivers/gpu/drm/i915/gt/intel_rps.c >>>> b/drivers/gpu/drm/i915/gt/intel_rps.c >>>> index b01c837ab646..a2c502609d96 100644 >>>> --- a/drivers/gpu/drm/i915/gt/intel_rps.c >>>> +++ b/drivers/gpu/drm/i915/gt/intel_rps.c >>>> @@ -782,7 +782,7 @@ static void gen6_rps_set_thresholds(struct intel_rps >>>> *rps, u8 val) >>>> mutex_unlock(&rps->power.mutex); >>>> } >>>> >>>> -void intel_rps_mark_interactive(struct intel_rps *rps, bool interactive) >>>> +static void _intel_rps_mark_interactive(struct intel_rps *rps, bool >>>> interactive) >>>> { >>>> GT_TRACE(rps_to_gt(rps), "mark interactive: %s\n", >>>> str_yes_no(interactive)); >>>> @@ -798,6 +798,13 @@ void intel_rps_mark_interactive(struct intel_rps >>>> *rps, bool interactive) >>>> mutex_unlock(&rps->power.mutex); >>>> } >>>> >>>> +void intel_rps_mark_interactive(struct drm_device *drm, bool interactive) >>>> +{ >>>> + struct drm_i915_private *i915 = to_i915(drm); >>>> + >>>> + _intel_rps_mark_interactive(&to_gt(i915)->rps, interactive); >>>> +} >>>> + >>>> static int gen6_rps_set(struct intel_rps *rps, u8 val) >>>> { >>>> struct intel_uncore *uncore = rps_to_uncore(rps); >>>> @@ -1953,7 +1960,7 @@ void gen6_rps_irq_handler(struct intel_rps *rps, u32 >>>> pm_iir) >>>> "Command parser error, pm_iir 0x%08x\n", >>>> pm_iir); >>>> } >>>> >>>> -void gen5_rps_irq_handler(struct intel_rps *rps) >>>> +static void _gen5_rps_irq_handler(struct intel_rps *rps) >>>> { >>>> struct intel_uncore *uncore = rps_to_uncore(rps); >>>> u32 busy_up, busy_down, max_avg, min_avg; >>>> @@ -1987,6 +1994,13 @@ void gen5_rps_irq_handler(struct intel_rps *rps) >>>> spin_unlock(&mchdev_lock); >>>> } >>>> >>>> +void gen5_rps_irq_handler(struct drm_device *drm) >>>> +{ >>>> + struct drm_i915_private *i915 = to_i915(drm); >>>> + >>>> + _gen5_rps_irq_handler(&to_gt(i915)->rps); >>>> +} >>>> + >>>> void intel_rps_init_early(struct intel_rps *rps) >>>> { >>>> mutex_init(&rps->lock); >>>> diff --git a/drivers/gpu/drm/i915/gt/intel_rps.h >>>> b/drivers/gpu/drm/i915/gt/intel_rps.h >>>> index 92fb01f5a452..c817a70fb3aa 100644 >>>> --- a/drivers/gpu/drm/i915/gt/intel_rps.h >>>> +++ b/drivers/gpu/drm/i915/gt/intel_rps.h >>>> @@ -9,8 +9,9 @@ >>>> #include "intel_rps_types.h" >>>> #include "i915_reg_defs.h" >>>> >>>> -struct i915_request; >>>> +struct drm_device; >>>> struct drm_printer; >>>> +struct i915_request; >>>> >>>> #define GT_FREQUENCY_MULTIPLIER 50 >>>> #define GEN9_FREQ_SCALER 3 >>>> @@ -33,7 +34,7 @@ u32 intel_rps_get_boost_frequency(struct intel_rps *rps); >>>> int intel_rps_set_boost_frequency(struct intel_rps *rps, u32 freq); >>>> >>>> int intel_rps_set(struct intel_rps *rps, u8 val); >>>> -void intel_rps_mark_interactive(struct intel_rps *rps, bool interactive); >>>> +void intel_rps_mark_interactive(struct drm_device *drm, bool interactive); >>> >>> This one breaks the design a bit since RPS is supposed to be per GT. On >>> the other hand intel_display_rps_mark_interactive is the only caller so >>> if it only wants to care about the primary GT thats probably okay. >>> >>> Presumably you don't want a for_each_gt in the display code either. >>> >>> Would it work to put a helper which did it for you somewhere one level >>> up from per gt (gt/) components? Sounds like for the end goal of proper >>> abstraction that would be the way. Getting rid of the #ifdef from >>> intel_display_rps.h and each driver would then implement the required >>> hooks to do what they want. >> >> See [1]. We might add display->parent->rps, and call the hooks via >> that. But even so, they'll need interfaces that are independent of gt, >> so something like the patch at hand will be needed. I don't particularly >> care if the functions on i915 core side are in gt/ or somewhere else. > > Okay, but from my point of view intel_rps_mark_interactive() should > remain taking rps. This is the design of all components under gt/ and I > do not think we should break it for this case. So for me a new helper > somewhere one level above gt/ sounds like the way to go. That one is > then perfectly fine to operate on the device.
Okay. > Pull out existing > intel_display_rps_mark_interactive() out from display and rename to > something like intel_display_mark_interactive(). As a first step. Xe can > stub it out in compat headers rather than #ifdefs in the display code. intel_display_rps_mark_interactive() needs to remain in display, because it handles display structures. struct drm_device is the abstraction between display and i915 core. Looks like the whole thing is going to have to wait for [1] to land, and we'll add the function pointers there, which will then have some functions that do exactly the same thing as the wrappers I added here do, but will have to find a new location for them somewhere in i915 core that is not gt/. BR, Jani. > > Regards, > > Tvrtko >> [1] >> https://lore.kernel.org/r/[email protected] >> >>> >>> Regards, >>> >>> Tvrtko >>>> int intel_gpu_freq(struct intel_rps *rps, int val); >>>> int intel_freq_opcode(struct intel_rps *rps, int val); >>>> @@ -64,7 +65,7 @@ bool rps_read_mask_mmio(struct intel_rps *rps, >>>> i915_reg_t reg32, u32 mask); >>>> >>>> void gen6_rps_frequency_dump(struct intel_rps *rps, struct drm_printer >>>> *p); >>>> >>>> -void gen5_rps_irq_handler(struct intel_rps *rps); >>>> +void gen5_rps_irq_handler(struct drm_device *drm); >>>> void gen6_rps_irq_handler(struct intel_rps *rps, u32 pm_iir); >>>> void gen11_rps_irq_handler(struct intel_rps *rps, u32 pm_iir); >>>> >>> >> > -- Jani Nikula, Intel
