All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Ville Syrjälä" <ville.syrjala@linux.intel.com>
To: Hanson Wang <hanson.wang@ugreen.com>
Cc: intel-gfx@lists.freedesktop.org, jani.nikula@linux.intel.com,
	rodrigo.vivi@intel.com, joonas.lahtinen@linux.intel.com,
	tursulin@ursulin.net, intel-xe@lists.freedesktop.org,
	dri-devel@lists.freedesktop.org
Subject: Re: [PATCH] drm/i915/bw: sync global BW after non-atomic plane disable
Date: Wed, 18 Nov 2026 06:38:24 +0200	[thread overview]
Message-ID: <av0rwG6o5gzMXGDB@intel.com> (raw)
In-Reply-To: <20261118033312.902366-1-hanson.wang@ugreen.com>

On Wed, Nov 18, 2026 at 11:33:12AM +0800, Hanson Wang wrote:
> Hardware readout copies the firmware plane configuration into the
> global bandwidth state. skl_dbuf_sanitize() may then disable those
> planes through intel_plane_disable_noatomic(), which clears the
> CRTC-local active_planes, data_rate and DDB entries but does not
> update the global state.
> 
> A later atomic modeset sees both the old and new per-CRTC bandwidth
> as zero and skips the global update. The sanitized pipe therefore
> keeps contributing a phantom data rate and an active plane to the
> QGV check.
> 
> Do not force !hw.active to zero. A logically enabled but inactive CRTC
> must retain its bandwidth reservation so that reactivating it cannot
> fail due to insufficient bandwidth. After the non-atomic disable,
> recompute the global intel_bw_state and the separate dbuf bandwidth
> state from the updated CRTC-local plane and DDB state.
> 
> Reproduced on two N5105 systems from different board vendors running
> 6.18.15. After DBUF sanitization, pipe B retained 196000 kB/s and one
> plane in the global BW state despite its local plane state being
> cleared and the pipe subsequently being disabled.
> 
> With the corresponding fix backported to 6.18.15 on one system, the
> same sanitization path updated the global accounting to 0 kB/s and
> zero planes. Subsequent QGV checks no longer included the stale
> bandwidth.

I think what we might want to do is add a new
intel_display_update_hw_state() wrapper that calls all the
intel_foo_update_hw_state() functions. Then we just call that
after any disable_noatomic() stuff is done. Not sure if a single
call would be sufficient once we've sanitized everything, or if
we'd need to do a few extra calls during the sanitization
sequence.

And if we're missing any intel_foo_update_hw_state()
implementations then we should just add them.

> 
> Link: https://lore.kernel.org/intel-gfx/20261016014704.3876294-1-hanson.wang@ugreen.com
> 
> Signed-off-by: Hanson Wang <hanson.wang@ugreen.com>
> ---
> Testing:
> - The corresponding 6.18.15 backport was tested on one N5105 system.
> - This mainline version has not been runtime-tested.
> - The original high-bandwidth application failure was not replayed.
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_bw.c b/drivers/gpu/drm/i915/display/intel_bw.c
> index 99fd84a5ac13..84fe1f68e312 100644
> --- a/drivers/gpu/drm/i915/display/intel_bw.c
> +++ b/drivers/gpu/drm/i915/display/intel_bw.c
> @@ -1444,6 +1444,20 @@ static void intel_bw_crtc_update(struct intel_bw_state *bw_state,
>  		    bw_state->num_active_planes[crtc->pipe]);
>  }
>  
> +void intel_bw_crtc_update_noatomic(struct intel_crtc *crtc)
> +{
> +	struct intel_display *display = to_intel_display(crtc);
> +	struct intel_bw_state *bw_state =
> +		to_intel_bw_state(display->bw.obj.state);
> +	const struct intel_crtc_state *crtc_state =
> +		to_intel_crtc_state(crtc->base.state);
> +
> +	if (DISPLAY_VER(display) < 11)
> +		return;
> +
> +	intel_bw_crtc_update(bw_state, crtc_state);
> +}
> +
>  void intel_bw_update_hw_state(struct intel_display *display)
>  {
>  	struct intel_bw_state *bw_state =
> @@ -1456,12 +1470,9 @@ void intel_bw_update_hw_state(struct intel_display *display)
>  	bw_state->pipe_sagv_reject = 0;
>  
>  	for_each_intel_crtc(display, crtc) {
> -		const struct intel_crtc_state *crtc_state =
> -			to_intel_crtc_state(crtc->base.state);
>  		enum pipe pipe = crtc->pipe;
>  
> -		if (DISPLAY_VER(display) >= 11)
> -			intel_bw_crtc_update(bw_state, crtc_state);
> +		intel_bw_crtc_update_noatomic(crtc);
>  
>  		/* initially SAGV has been forced off */
>  		bw_state->pipe_sagv_reject |= BIT(pipe);
> diff --git a/drivers/gpu/drm/i915/display/intel_bw.h b/drivers/gpu/drm/i915/display/intel_bw.h
> index 99b447388245..41ace27608dc 100644
> --- a/drivers/gpu/drm/i915/display/intel_bw.h
> +++ b/drivers/gpu/drm/i915/display/intel_bw.h
> @@ -30,6 +30,7 @@ void intel_bw_init_hw(struct intel_display *display);
>  int intel_bw_init(struct intel_display *display);
>  int intel_bw_atomic_check(struct intel_atomic_state *state);
>  void intel_bw_update_hw_state(struct intel_display *display);
> +void intel_bw_crtc_update_noatomic(struct intel_crtc *crtc);
>  void intel_bw_crtc_disable_noatomic(struct intel_crtc *crtc);
>  
>  bool intel_bw_pmdemand_needs_update(struct intel_atomic_state *state);
> diff --git a/drivers/gpu/drm/i915/display/intel_dbuf_bw.c b/drivers/gpu/drm/i915/display/intel_dbuf_bw.c
> index 6cf674c586dc..453ceaa52e26 100644
> --- a/drivers/gpu/drm/i915/display/intel_dbuf_bw.c
> +++ b/drivers/gpu/drm/i915/display/intel_dbuf_bw.c
> @@ -226,21 +226,29 @@ int intel_dbuf_bw_calc_min_cdclk(struct intel_atomic_state *state,
>  	return 0;
>  }
>  
> -void intel_dbuf_bw_update_hw_state(struct intel_display *display)
> +void intel_dbuf_bw_crtc_update_noatomic(struct intel_crtc *crtc)
>  {
> +	struct intel_display *display = to_intel_display(crtc);
>  	struct intel_dbuf_bw_state *dbuf_bw_state =
>  		to_intel_dbuf_bw_state(display->dbuf_bw.obj.state);
> -	struct intel_crtc *crtc;
> +	const struct intel_crtc_state *crtc_state =
> +		to_intel_crtc_state(crtc->base.state);
>  
>  	if (DISPLAY_VER(display) < 9)
>  		return;
>  
> -	for_each_intel_crtc(display, crtc) {
> -		const struct intel_crtc_state *crtc_state =
> -			to_intel_crtc_state(crtc->base.state);
> +	skl_crtc_calc_dbuf_bw(&dbuf_bw_state->dbuf_bw[crtc->pipe], crtc_state);
> +}
>  
> -		skl_crtc_calc_dbuf_bw(&dbuf_bw_state->dbuf_bw[crtc->pipe], crtc_state);
> -	}
> +void intel_dbuf_bw_update_hw_state(struct intel_display *display)
> +{
> +	struct intel_crtc *crtc;
> +
> +	if (DISPLAY_VER(display) < 9)
> +		return;
> +
> +	for_each_intel_crtc(display, crtc)
> +		intel_dbuf_bw_crtc_update_noatomic(crtc);
>  }
>  
>  void intel_dbuf_bw_crtc_disable_noatomic(struct intel_crtc *crtc)
> diff --git a/drivers/gpu/drm/i915/display/intel_dbuf_bw.h b/drivers/gpu/drm/i915/display/intel_dbuf_bw.h
> index 61875b9d5969..434f3c605995 100644
> --- a/drivers/gpu/drm/i915/display/intel_dbuf_bw.h
> +++ b/drivers/gpu/drm/i915/display/intel_dbuf_bw.h
> @@ -32,6 +32,7 @@ int intel_dbuf_bw_calc_min_cdclk(struct intel_atomic_state *state,
>  int intel_dbuf_bw_min_cdclk(struct intel_display *display,
>  			    const struct intel_dbuf_bw_state *dbuf_bw_state);
>  void intel_dbuf_bw_update_hw_state(struct intel_display *display);
> +void intel_dbuf_bw_crtc_update_noatomic(struct intel_crtc *crtc);
>  void intel_dbuf_bw_crtc_disable_noatomic(struct intel_crtc *crtc);
>  
>  #endif /* __INTEL_DBUF_BW_H__ */
> diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> index b897678a93c0..b81d8f288810 100644
> --- a/drivers/gpu/drm/i915/display/intel_display.c
> +++ b/drivers/gpu/drm/i915/display/intel_display.c
> @@ -67,6 +67,7 @@
>  #include "intel_cursor.h"
>  #include "intel_cursor_regs.h"
>  #include "intel_cx0_phy.h"
> +#include "intel_dbuf_bw.h"
>  #include "intel_ddi.h"
>  #include "intel_de.h"
>  #include "intel_dip.h"
> @@ -679,6 +680,14 @@ void intel_plane_disable_noatomic(struct intel_crtc *crtc,
>  
>  	skl_wm_plane_disable_noatomic(crtc, plane);
>  
> +	/*
> +	 * Keep the global bandwidth states in sync with the updated
> +	 * CRTC-local plane and DDB state. Otherwise a later atomic check
> +	 * may skip the update when its old and new inputs are identical.
> +	 */
> +	intel_bw_crtc_update_noatomic(crtc);
> +	intel_dbuf_bw_crtc_update_noatomic(crtc);
> +
>  	if ((crtc_state->active_planes & ~BIT(PLANE_CURSOR)) == 0 &&
>  	    hsw_ips_disable(crtc_state)) {
>  		crtc_state->ips_enabled = false;
> -- 
> 2.34.1

-- 
Ville Syrjälä
Intel

      parent reply	other threads:[~2026-11-18  4:38 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-18  3:33 [PATCH] drm/i915/bw: sync global BW after non-atomic plane disable Hanson Wang
2026-11-18  4:00 ` sashiko-bot
2026-11-18  4:03 ` ✗ LGCI.VerificationFailed: failure for " Patchwork
2026-11-18  4:38 ` Ville Syrjälä [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=av0rwG6o5gzMXGDB@intel.com \
    --to=ville.syrjala@linux.intel.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=hanson.wang@ugreen.com \
    --cc=intel-gfx@lists.freedesktop.org \
    --cc=intel-xe@lists.freedesktop.org \
    --cc=jani.nikula@linux.intel.com \
    --cc=joonas.lahtinen@linux.intel.com \
    --cc=rodrigo.vivi@intel.com \
    --cc=tursulin@ursulin.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.