[Intel-gfx] [PATCH V4] drm/i915/skl: SKL CDCLK change on modeset tracking VCO
Ville Syrjälä
ville.syrjala at linux.intel.com
Fri Feb 12 11:18:19 UTC 2016
On Thu, Feb 11, 2016 at 03:22:08PM -0800, clinton.a.taylor at intel.com wrote:
> From: Clint Taylor <clinton.a.taylor at intel.com>
>
> Track VCO frequency of SKL instead of the boot CDCLK and allow modeset
> to set cdclk based on the max required pixel clock based on VCO
> selected.
>
> The vco should be tracked at the atomic level and all CRTCs updated if
> the required vco is changed. At this time the eDP pll is configured
> inside the encoder which has no visibility into the atomic state. When
> eDP v1.4 panel that require the 8640 vco are available this may need
> to be investigated.
>
> V1: initial version
> V2: add vco tracking in intel_dp_compute_config(), rename
> skl_boot_cdclk.
> V3: rebase, V2 feedback not possible as encoders are not aware of
> atomic.
> V4: track target vco is atomic state. modeset all CRTCs if vco changes
>
> Signed-off-by: Clint Taylor <clinton.a.taylor at intel.com>
> Cc: Ville =?iso-8859-1?Q?Syrj=E4l=E4?= <ville.syrjala at linux.intel.com>
> ---
> drivers/gpu/drm/i915/i915_drv.h | 2 +-
> drivers/gpu/drm/i915/intel_ddi.c | 2 +-
> drivers/gpu/drm/i915/intel_display.c | 97 +++++++++++++++++++++++++++++-----
> drivers/gpu/drm/i915/intel_dp.c | 10 ++--
> drivers/gpu/drm/i915/intel_drv.h | 4 ++
> 5 files changed, 97 insertions(+), 18 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 8216665..f65dd1a 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1822,7 +1822,7 @@ struct drm_i915_private {
> int num_fence_regs; /* 8 on pre-965, 16 otherwise */
>
> unsigned int fsb_freq, mem_freq, is_ddr3;
> - unsigned int skl_boot_cdclk;
> + unsigned int skl_vco_freq;
> unsigned int cdclk_freq, max_cdclk_freq, atomic_cdclk_freq;
> unsigned int max_dotclk_freq;
> unsigned int hpll_freq;
> diff --git a/drivers/gpu/drm/i915/intel_ddi.c b/drivers/gpu/drm/i915/intel_ddi.c
> index 6d5b09f..285adab 100644
> --- a/drivers/gpu/drm/i915/intel_ddi.c
> +++ b/drivers/gpu/drm/i915/intel_ddi.c
> @@ -2958,7 +2958,7 @@ void intel_ddi_pll_init(struct drm_device *dev)
> int cdclk_freq;
>
> cdclk_freq = dev_priv->display.get_display_clock_speed(dev);
> - dev_priv->skl_boot_cdclk = cdclk_freq;
> + dev_priv->skl_vco_freq = skl_cdclk_get_vco(cdclk_freq);
This should really read out the vco from the hardware. But I think we
can leave that for later. The problem really is the 540MHz case since it
can be using either vco frequency (IIRC).
> if (skl_sanitize_cdclk(dev_priv))
> DRM_DEBUG_KMS("Sanitized cdclk programmed by pre-os\n");
> if (!(I915_READ(LCPLL1_CTL) & LCPLL_PLL_ENABLE))
> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index 9e2273b..ef4ac34 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -5663,7 +5663,7 @@ static unsigned int skl_cdclk_decimal(unsigned int freq)
> return (freq - 1000) / 500;
> }
>
> -static unsigned int skl_cdclk_get_vco(unsigned int freq)
> +unsigned int skl_cdclk_get_vco(unsigned int freq)
> {
> unsigned int i;
>
> @@ -5821,17 +5821,17 @@ void skl_uninit_cdclk(struct drm_i915_private *dev_priv)
>
> void skl_init_cdclk(struct drm_i915_private *dev_priv)
> {
> - unsigned int required_vco;
> -
> /* DPLL0 not enabled (happens on early BIOS versions) */
> if (!(I915_READ(LCPLL1_CTL) & LCPLL_PLL_ENABLE)) {
> /* enable DPLL0 */
> - required_vco = skl_cdclk_get_vco(dev_priv->skl_boot_cdclk);
> - skl_dpll0_enable(dev_priv, required_vco);
> + if (dev_priv->skl_vco_freq != 8640) {
> + dev_priv->skl_vco_freq = 8100;
> + }
> + skl_dpll0_enable(dev_priv, dev_priv->skl_vco_freq);
> }
>
> /* set CDCLK to the frequency the BIOS chose */
> - skl_set_cdclk(dev_priv, dev_priv->skl_boot_cdclk);
> + skl_set_cdclk(dev_priv, (dev_priv->skl_vco_freq == 8100) ? 337500 : 308570 );
We really shouldn't change the cdclk if there are active outputs. This
whole area definitely needs more work, for BXT too I already have
some BXT patches lined up in some branch that frob around these parts
a bit, so maybe I should extend that stuff to SKL as well.
In the meantime we don't want to cause a regression at least, so
maybe we can just do something like:
if (!LCPLL_ENABLE) {
...
cdclk = vco == 8100 ? ...;
} else {
cdclk = dev_priv->cdclk_freq;
}
set_cdclk(cdclk);
>
> /* enable DBUF power */
> I915_WRITE(DBUF_CTL, I915_READ(DBUF_CTL) | DBUF_POWER_REQUEST);
> @@ -5847,7 +5847,7 @@ int skl_sanitize_cdclk(struct drm_i915_private *dev_priv)
> {
> uint32_t lcpll1 = I915_READ(LCPLL1_CTL);
> uint32_t cdctl = I915_READ(CDCLK_CTL);
> - int freq = dev_priv->skl_boot_cdclk;
> + int freq = dev_priv->cdclk_freq;
>
> /*
> * check if the pre-os intialized the display
> @@ -5871,11 +5871,7 @@ int skl_sanitize_cdclk(struct drm_i915_private *dev_priv)
> /* All well; nothing to sanitize */
> return false;
> sanitize:
> - /*
> - * As of now initialize with max cdclk till
> - * we get dynamic cdclk support
> - * */
> - dev_priv->skl_boot_cdclk = dev_priv->max_cdclk_freq;
> +
> skl_init_cdclk(dev_priv);
>
> /* we did have to sanitize */
> @@ -9845,6 +9841,68 @@ static void broadwell_modeset_commit_cdclk(struct drm_atomic_state *old_state)
> broadwell_set_cdclk(dev, req_cdclk);
> }
>
> +static int skl_modeset_calc_cdclk(struct drm_atomic_state *state)
> +{
> + struct drm_i915_private *dev_priv = to_i915(state->dev);
> + int max_pixclk = ilk_max_pixel_rate(state);
> + int cdclk;
> +
> + /*
> + * FIXME should also account for plane ratio
> + * once 64bpp pixel formats are supported.
> + */
A bit of formatting issue with the comment. The stars should be lining up :)
> +
> + if (to_intel_atomic_state(state)->vco_target == 8640) {
> + /* vco 8640 */
> + if (max_pixclk > 540000)
> + cdclk = 617140;
> + else if (max_pixclk > 432000)
> + cdclk = 540000;
> + else if (max_pixclk > 308570)
> + cdclk = 432000;
> + else
> + cdclk = 308570;
> + }
> + else {
> + /* VCO 8100 */
> + if (max_pixclk > 540000)
> + cdclk = 675000;
> + else if (max_pixclk > 450000)
> + cdclk = 540000;
> + else if (max_pixclk > 337500)
> + cdclk = 450000;
> + else
> + cdclk = 337500;
> + }
> +
> + /*
> + * FIXME move the cdclk caclulation to
> + * compute_config() so we can fail gracegully.
> + */
> + if (cdclk > dev_priv->max_cdclk_freq) {
> + DRM_ERROR("requested cdclk (%d kHz) exceeds max (%d kHz)\n",
> + cdclk, dev_priv->max_cdclk_freq);
> + cdclk = dev_priv->max_cdclk_freq;
> + }
> +
> + to_intel_atomic_state(state)->cdclk = cdclk;
> +
> + return 0;
> +}
> +
> +static void skl_modeset_commit_cdclk(struct drm_atomic_state *old_state)
> +{
> + struct drm_device *dev = old_state->dev;
> + struct drm_i915_private *dev_priv = dev->dev_private;
> + unsigned int req_cdclk = to_intel_atomic_state(old_state)->cdclk;
> +
> + skl_set_cdclk(dev_priv, req_cdclk);
> +
> + if (to_intel_atomic_state(old_state)->vco_target) {
> + dev_priv->skl_vco_freq = to_intel_atomic_state(old_state)->vco_target;
> + }
skl_set_cdclk() really needs to be taught to do the 'disable+enable pll'
dance so that it can actually change the vco frequency. But I think we
can leave that for a followup so that we can move this forward. A FIXME
comment here would be good though so we don't forget.
> +}
> +
> static int haswell_crtc_compute_clock(struct intel_crtc *crtc,
> struct intel_crtc_state *crtc_state)
> {
> @@ -13219,11 +13277,13 @@ static int intel_modeset_all_pipes(struct drm_atomic_state *state)
>
> static int intel_modeset_checks(struct drm_atomic_state *state)
> {
> + struct drm_device *dev = state->dev;
> struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
> struct drm_i915_private *dev_priv = state->dev->dev_private;
> struct drm_crtc *crtc;
> struct drm_crtc_state *crtc_state;
> int ret = 0, i;
> + unsigned int target_vco;
>
> if (!check_digital_port_conflicts(state)) {
> DRM_DEBUG_KMS("rejecting conflicting digital port configuration\n");
> @@ -13249,8 +13309,14 @@ static int intel_modeset_checks(struct drm_atomic_state *state)
> */
> if (dev_priv->display.modeset_calc_cdclk) {
> ret = dev_priv->display.modeset_calc_cdclk(state);
> + target_vco = to_intel_atomic_state(state)->vco_target;
>
> - if (!ret && intel_state->dev_cdclk != dev_priv->cdclk_freq)
> + if (IS_SKYLAKE(dev) || (IS_KABYLAKE(dev))) {
> + if (((target_vco) && (dev_priv->skl_vco_freq != target_vco)) ||
Hmm. If target_vco is 0, then skl_modeset_calc_cdclk() has already
done something wrong. So I think if the target_vco wasn't set by the
encoder(s) you'll need to just set '->vco_target = dev_priv->vco' before
calling .modeset_calc_cdclk().
BTW the vco_target vs. target_vco is bothering me a bit. In fact I think
it's pretty clear that it's a target value simply due to the fact that
it's part of the atomic state. So we can probably just call it something
like cdclk_pll_vco (keeping the name somewhat generic in case we need
the same treatment for a future platform, and we all know the hardware
folks like to rename things for fun and profit).
> + (!ret && intel_state->dev_cdclk != dev_priv->cdclk_freq)) {
> + ret = intel_modeset_all_pipes(state);
> + }
> + } else if (!ret && intel_state->dev_cdclk != dev_priv->cdclk_freq)
> ret = intel_modeset_all_pipes(state);
>
> if (ret < 0)
> @@ -15002,6 +15068,11 @@ static void intel_init_display(struct drm_device *dev)
> broxton_modeset_commit_cdclk;
> dev_priv->display.modeset_calc_cdclk =
> broxton_modeset_calc_cdclk;
> + } else if (IS_SKYLAKE(dev) || IS_KABYLAKE(dev)) {
> + dev_priv->display.modeset_commit_cdclk =
> + skl_modeset_commit_cdclk;
> + dev_priv->display.modeset_calc_cdclk =
> + skl_modeset_calc_cdclk;
> }
>
> switch (INTEL_INFO(dev)->gen) {
> diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c
> index a073f04..fbc37fe 100644
> --- a/drivers/gpu/drm/i915/intel_dp.c
> +++ b/drivers/gpu/drm/i915/intel_dp.c
> @@ -1242,9 +1242,10 @@ intel_dp_connector_unregister(struct intel_connector *intel_connector)
> }
>
> static void
> -skl_edp_set_pll_config(struct intel_crtc_state *pipe_config)
> +skl_edp_set_pll_config(struct drm_i915_private *dev_priv, struct intel_crtc_state *pipe_config)
dev_priv could be dug out via the pipe_config too, but this way is fine
too IMO.
> {
> u32 ctrl1;
> + u32 vco = 8100;
>
> memset(&pipe_config->dpll_hw_state, 0,
> sizeof(pipe_config->dpll_hw_state));
> @@ -1277,13 +1278,16 @@ skl_edp_set_pll_config(struct intel_crtc_state *pipe_config)
> case 108000:
> ctrl1 |= DPLL_CTRL1_LINK_RATE(DPLL_CTRL1_LINK_RATE_1080,
> SKL_DPLL0);
> + vco = 8640;
> break;
> case 216000:
> ctrl1 |= DPLL_CTRL1_LINK_RATE(DPLL_CTRL1_LINK_RATE_2160,
> SKL_DPLL0);
> + vco = 8640;
> break;
> -
> }
> +
> + to_intel_atomic_state(pipe_config->base.state)->vco_target = vco;
> pipe_config->dpll_hw_state.ctrl1 = ctrl1;
> }
>
> @@ -1664,7 +1668,7 @@ found:
> }
>
> if ((IS_SKYLAKE(dev) || IS_KABYLAKE(dev)) && is_edp(intel_dp))
> - skl_edp_set_pll_config(pipe_config);
> + skl_edp_set_pll_config(dev_priv, pipe_config);
> else if (IS_BROXTON(dev))
> /* handled in ddi */;
> else if (IS_HASWELL(dev) || IS_BROADWELL(dev))
> diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h
> index 878172a..005e036 100644
> --- a/drivers/gpu/drm/i915/intel_drv.h
> +++ b/drivers/gpu/drm/i915/intel_drv.h
> @@ -260,6 +260,9 @@ struct intel_atomic_state {
>
> struct intel_shared_dpll_config shared_dpll[I915_NUM_PLLS];
> struct intel_wm_config wm_config;
> +
> + /* SKL/KBL Only */
> + unsigned int vco_target;
> };
>
> struct intel_plane_state {
> @@ -1191,6 +1194,7 @@ void bxt_disable_dc9(struct drm_i915_private *dev_priv);
> void skl_init_cdclk(struct drm_i915_private *dev_priv);
> int skl_sanitize_cdclk(struct drm_i915_private *dev_priv);
> void skl_uninit_cdclk(struct drm_i915_private *dev_priv);
> +unsigned int skl_cdclk_get_vco(unsigned int freq);
> void skl_enable_dc6(struct drm_i915_private *dev_priv);
> void skl_disable_dc6(struct drm_i915_private *dev_priv);
> void intel_dp_get_m_n(struct intel_crtc *crtc,
> --
> 1.7.9.5
--
Ville Syrjälä
Intel OTC
More information about the Intel-gfx
mailing list