[PATCH] drm/i915/watermark: Modify latency programmed into PKG_C_LATENCY
Kandpal, Suraj
suraj.kandpal at intel.com
Mon Nov 11 11:39:19 UTC 2024
> -----Original Message-----
> From: Govindapillai, Vinod <vinod.govindapillai at intel.com>
> Sent: Monday, November 11, 2024 4:29 PM
> To: Kandpal, Suraj <suraj.kandpal at intel.com>; intel-xe at lists.freedesktop.org;
> intel-gfx at lists.freedesktop.org
> Cc: Pottumuttu, Sai Teja <sai.teja.pottumuttu at intel.com>
> Subject: Re: [PATCH] drm/i915/watermark: Modify latency programmed into
> PKG_C_LATENCY
>
> Hi Suraj,
>
> Some comments, please have a look..
>
> On Fri, 2024-11-08 at 14:20 +0530, Suraj Kandpal wrote:
> > > Increase the latency programmed into PKG_C_LATENCY latency to be a
> > > multiple of line time which is written into WM_LINETIME.
> > >
> > > --v2
> > > -Fix commit subject line [Sai Teja]
> > > -Use individual DISPLAY_VER checks instead of range [Sai Teja]
> > > -Initialize max_linetime [Sai Teja]
> > >
> > > WA: 22020299601
> > > Signed-off-by: Suraj Kandpal <suraj.kandpal at intel.com>
> > > ---
> > > drivers/gpu/drm/i915/display/skl_watermark.c | 26
> > > ++++++++++++++------
> > > 1 file changed, 18 insertions(+), 8 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/display/skl_watermark.c >
> > > b/drivers/gpu/drm/i915/display/skl_watermark.c
> > > index d3bbf335c749..9e208db55abb 100644
> > > --- a/drivers/gpu/drm/i915/display/skl_watermark.c
> > > +++ b/drivers/gpu/drm/i915/display/skl_watermark.c
> > > @@ -2848,9 +2848,11 @@ static int skl_wm_add_affected_planes(struct
> > > intel_atomic_state *state,
> > > * Program PKG_C_LATENCY Added Wake Time = 0
> > > */
> > > static void
> > > -skl_program_dpkgc_latency(struct drm_i915_private *i915, bool
> > > enable_dpkgc)
> > > +skl_program_dpkgc_latency(struct drm_i915_private *i915,
> > > + bool enable_dpkgc,
> > > + u32 max_linetime)
> > > {
> > > - u32 max_latency = 0;
> > > + u32 adjusted_latency = 0;
> I guess, if you initialize adjusted_latency to LNL_PKG_C_LATENCY_MASK, you
> could get rid of the else branch below. May be you could have a preparatory
> patch which get rid of this else branch?
You're right will get that done.
>
> > > u32 clear = 0, val = 0;
> > > u32 added_wake_time = 0;
> > >
> > > @@ -2858,18 +2860,23 @@ skl_program_dpkgc_latency(struct
> > > drm_i915_private *i915, bool >
> > > enable_dpkgc)
> > > return;
> > >
> > > if (enable_dpkgc) {
> > > - max_latency = skl_watermark_max_latency(i915, 1);
> > > - if (max_latency == 0)
> > > - max_latency = LNL_PKG_C_LATENCY_MASK;
> > > + adjusted_latency = skl_watermark_max_latency(i915,
> > > +1);
> > > + if (adjusted_latency == 0)
> > > + adjusted_latency = LNL_PKG_C_LATENCY_MASK;
> > > +
> Also, here I guess you need to avoid this WA if adjusted_latency is 0 ie. invalid
> level 1 latency.
Sure will get this fixed
>
> Also use intel_display instead of i915 wherever possible related to this
> change.
Sure you are right ill update the function itself to use intel_de_rmw instead on uncore_rmw too will have a separate patch for it.
Regards,
Suraj Kandpal
>
> > > + /* Wa_22020299601 */
> > > + if (DISPLAY_VER(i915) == 20 || DISPLAY_VER(i915) ==
> > > +30)
> > > + adjusted_latency = max_linetime *
> > > + DIV_ROUND_UP(adjusted_latency,
> > > +max_linetime);
> > > added_wake_time = DSB_EXE_TIME +
> > > i915->display.sagv.block_time_us;
> > > } else {
> > > - max_latency = LNL_PKG_C_LATENCY_MASK;
> > > + adjusted_latency = LNL_PKG_C_LATENCY_MASK;
> > > added_wake_time = 0;
> > > }
> > >
> > > clear |= LNL_ADDED_WAKE_TIME_MASK |
> LNL_PKG_C_LATENCY_MASK;
> > > - val |= REG_FIELD_PREP(LNL_PKG_C_LATENCY_MASK, max_latency);
> > > + val |= REG_FIELD_PREP(LNL_PKG_C_LATENCY_MASK,
> > > +adjusted_latency);
> > > val |= REG_FIELD_PREP(LNL_ADDED_WAKE_TIME_MASK,
> > > added_wake_time);
> > >
> > > intel_uncore_rmw(&i915->uncore, LNL_PKG_C_LATENCY, clear,
> > > val); @@ -2882,6 +2889,7 @@ skl_compute_wm(struct intel_atomic_state
> > > *state)
> > > struct intel_crtc_state __maybe_unused *new_crtc_state;
> > > int ret, i;
> > > bool enable_dpkgc = false;
> > > + u32 max_linetime = 0;
> > >
> > > for_each_new_intel_crtc_in_state(state, crtc,
> > > new_crtc_state, i) {
> > > ret = skl_build_pipe_wm(state, crtc); @@ -2911,9
> > > +2919,11 @@ skl_compute_wm(struct intel_atomic_state *state)
> > > new_crtc_state->vrr.vmin ==
> > > new_crtc_state->vrr.flipline) ||
> > > !new_crtc_state->vrr.enable)
> > > enable_dpkgc = true;
> > > +
> > > + max_linetime = max(new_crtc_state->linetime,
> > > +max_linetime);
> > > }
> > >
> > > - skl_program_dpkgc_latency(to_i915(state->base.dev),
> > > enable_dpkgc);
> > > + skl_program_dpkgc_latency(to_i915(state->base.dev),
> > > +enable_dpkgc, max_linetime);
> > >
> > > skl_print_wm_changes(state);
> > >
More information about the Intel-xe
mailing list