[Intel-gfx] [PATCH] drm/i915: Implement PSF GV point support
Matt Roper
matthew.d.roper at intel.com
Thu May 27 04:09:11 UTC 2021
On Fri, May 21, 2021 at 04:43:23PM +0300, Stanislav Lisovskiy wrote:
> PSF GV points are an additional factor that can limit the
> bandwidth available to display, separate from the traditional
> QGV points. Whereas traditional QGV points represent possible
> memory clock frequencies, PSF GV points reflect possible
> frequencies of the memory fabric.
>
> Switching between PSF GV points has the advantage of incurring
> almost no memory access block time and thus does not need to be
> accounted for in watermark calculations.
>
> This patch adds support for those on top of regular QGV points.
> Those are supposed to be used simultaneously, i.e we are always
> at some QGV and some PSF GV point, based on the current video
> mode requirements.
> Bspec: 64631, 53998
>
> v2: Seems that initial assumption made during ml conversation
> was wrong, PCode rejects any masks containing points beyond
> the ones returned, so even though BSpec says we have around
> 8 points theoretically, we can mask/unmask only those which
> are returned, trying to manipulate those beyond causes a
> failure from PCode. So switched back to generating mask
> from 1 - num_qgv_points, where num_qgv_points is the actual
> amount of points, advertised by PCode.
>
> Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy at intel.com>
> Cc: Matt Roper <matthew.d.roper at intel.com>
> ---
> drivers/gpu/drm/i915/display/intel_bw.c | 113 +++++++++++++++++++++++-
> drivers/gpu/drm/i915/i915_drv.h | 7 ++
> drivers/gpu/drm/i915/i915_reg.h | 4 +
> drivers/gpu/drm/i915/intel_dram.c | 1 +
> 4 files changed, 121 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/display/intel_bw.c b/drivers/gpu/drm/i915/display/intel_bw.c
> index 3a1ba52266a7..fd23a4818c19 100644
> --- a/drivers/gpu/drm/i915/display/intel_bw.c
> +++ b/drivers/gpu/drm/i915/display/intel_bw.c
> @@ -17,9 +17,15 @@ struct intel_qgv_point {
> u16 dclk, t_rp, t_rdpre, t_rc, t_ras, t_rcd;
> };
>
> +struct intel_psf_gv_point {
> + u8 clk; /* clock in multiples of 16.6666 MHz */
> +};
> +
> struct intel_qgv_info {
> struct intel_qgv_point points[I915_NUM_QGV_POINTS];
> + struct intel_psf_gv_point psf_points[I915_NUM_PSF_GV_POINTS];
> u8 num_points;
> + u8 num_psf_points;
> u8 t_bl;
> };
>
> @@ -49,6 +55,28 @@ static int icl_pcode_read_qgv_point_info(struct drm_i915_private *dev_priv,
> return 0;
> }
>
> +static int adls_pcode_read_psf_gv_point_info(struct drm_i915_private *dev_priv,
> + struct intel_psf_gv_point *points)
> +{
> + u32 val = 0, val2 = 0;
> + int ret;
> + int i;
> +
> + ret = sandybridge_pcode_read(dev_priv,
> + ICL_PCODE_MEM_SUBSYSYSTEM_INFO |
> + ADL_PCODE_MEM_SS_READ_PSF_GV_INFO,
> + &val, &val2);
We never use val2, so I believe we could drop that variable and just
pass NULL here if we wanted. Not a huge deal either way; I'll leave it
up to you as to whether you want to change it or not.
> + if (ret)
> + return ret;
> +
> + for (i = 0; i < I915_NUM_PSF_GV_POINTS; i++) {
> + points[i].clk = val & 0xff;
> + val >>= 8;
> + }
> +
> + return 0;
> +}
> +
> int icl_pcode_restrict_qgv_points(struct drm_i915_private *dev_priv,
> u32 points_mask)
> {
> @@ -62,7 +90,7 @@ int icl_pcode_restrict_qgv_points(struct drm_i915_private *dev_priv,
> 1);
>
> if (ret < 0) {
> - drm_err(&dev_priv->drm, "Failed to disable qgv points (%d)\n", ret);
> + drm_err(&dev_priv->drm, "Failed to disable qgv points (%d) points: %x\n", ret, points_mask);
Personally I like to do "0x%x" for hex values in the format strings just
so there's no ambiguity about how to interpret the value when someone
sees it show up in the dmesg. But again, this is a minor nitpick; up to
you whether it's worth changing.
One other thing I think we might have overlooked in this patch is that
we should be checking some additional bits (3:2) in the response value
of the DE_MEM_SS_CONFIG pcode command to see whether the requested PSF
programming was successful. Right now we're only checking bit 0
(ICL_PCODE_POINTS_RESTRICTED_MASK) to determine whether the masking was
successful and that bit only relates to the main QGV points.
And on a related note, it looks like we might have missed something
previously in our handling of QGV point masking -- about a year ago (Apr
6th, 2020) there was a bspec update indicating that bits 1:0 can return
a value of 0x2 as another kind of QGV point masking error rather than
just the 0x1 we're expecting today. If we get an 0x2 error back today
for QGV points, or any kind of PSF error, we're not going to notice and
will still consider it a success.
So maybe we want an initial patch to change
ICL_PCODE_POINTS_RESTRICTED_MASK from 0x1 to 0x3 so that we'll catch the
QGV point errors, and then in this PSF patch we'll update it again to
0xF so that we catch any errors in either the QGV or PSF fields? Does
that match your understanding of the spec?
Matt
> return ret;
> }
>
> @@ -76,6 +104,7 @@ static int icl_get_qgv_points(struct drm_i915_private *dev_priv,
> int i, ret;
>
> qi->num_points = dram_info->num_qgv_points;
> + qi->num_psf_points = dram_info->num_psf_gv_points;
>
> if (DISPLAY_VER(dev_priv) == 12)
> switch (dram_info->type) {
> @@ -109,6 +138,19 @@ static int icl_get_qgv_points(struct drm_i915_private *dev_priv,
> sp->t_rcd, sp->t_rc);
> }
>
> + if (qi->num_psf_points > 0) {
> + ret = adls_pcode_read_psf_gv_point_info(dev_priv, qi->psf_points);
> + if (ret) {
> + drm_err(&dev_priv->drm, "Failed to read PSF point data; PSF points will not be considered in bandwidth calculations.\n");
> + qi->num_psf_points = 0;
> + }
> +
> + for (i = 0; i < qi->num_psf_points; i++)
> + drm_dbg_kms(&dev_priv->drm,
> + "PSF GV %d: CLK=%d \n",
> + i, qi->psf_points[i].clk);
> + }
> +
> return 0;
> }
>
> @@ -118,6 +160,16 @@ static int icl_calc_bw(int dclk, int num, int den)
> return DIV_ROUND_CLOSEST(num * dclk * 100, den * 6);
> }
>
> +static int adl_calc_psf_bw(int clk)
> +{
> + /*
> + * clk is multiples of 16.666MHz (100/6)
> + * According to BSpec PSF GV bandwidth is
> + * calculated as BW = 64 * clk * 16.666Mhz
> + */
> + return DIV_ROUND_CLOSEST(64 * clk * 100, 6);
> +}
> +
> static int icl_sagv_max_dclk(const struct intel_qgv_info *qi)
> {
> u16 dclk = 0;
> @@ -194,6 +246,7 @@ static int icl_get_bw_info(struct drm_i915_private *dev_priv, const struct intel
> bi->num_planes = (ipqdepth - clpchgroup) / clpchgroup + 1;
>
> bi->num_qgv_points = qi.num_points;
> + bi->num_psf_gv_points = qi.num_psf_points;
>
> for (j = 0; j < qi.num_points; j++) {
> const struct intel_qgv_point *sp = &qi.points[j];
> @@ -217,6 +270,16 @@ static int icl_get_bw_info(struct drm_i915_private *dev_priv, const struct intel
> i, j, bi->num_planes, bi->deratedbw[j]);
> }
>
> + for (j = 0; j < qi.num_psf_points; j++) {
> + const struct intel_psf_gv_point *sp = &qi.psf_points[j];
> +
> + bi->psf_bw[j] = adl_calc_psf_bw(sp->clk);
> +
> + drm_dbg_kms(&dev_priv->drm,
> + "BW%d / PSF GV %d: num_planes=%d bw=%u\n",
> + i, j, bi->num_planes, bi->psf_bw[j]);
> + }
> +
> if (bi->num_planes == 1)
> break;
> }
> @@ -262,6 +325,15 @@ static unsigned int icl_max_bw(struct drm_i915_private *dev_priv,
> return 0;
> }
>
> +static unsigned int adl_psf_bw(struct drm_i915_private *dev_priv,
> + int psf_gv_point)
> +{
> + const struct intel_bw_info *bi =
> + &dev_priv->max_bw[0];
> +
> + return bi->psf_bw[psf_gv_point];
> +}
> +
> void intel_bw_init_hw(struct drm_i915_private *dev_priv)
> {
> if (!HAS_DISPLAY(dev_priv))
> @@ -534,12 +606,24 @@ int intel_bw_atomic_check(struct intel_atomic_state *state)
> u32 allowed_points = 0;
> unsigned int max_bw_point = 0, max_bw = 0;
> unsigned int num_qgv_points = dev_priv->max_bw[0].num_qgv_points;
> - u32 mask = (1 << num_qgv_points) - 1;
> + unsigned int num_psf_gv_points = dev_priv->max_bw[0].num_psf_gv_points;
> + u32 mask = 0;
>
> /* FIXME earlier gens need some checks too */
> if (DISPLAY_VER(dev_priv) < 11)
> return 0;
>
> + /*
> + * We can _not_ use the whole ADLS_QGV_PT_MASK here, as PCode rejects
> + * it with failure if we try masking any unadvertised points.
> + * So need to operate only with those returned from PCode.
> + */
> + if (num_qgv_points > 0)
> + mask |= REG_GENMASK(num_qgv_points - 1, 0);
> +
> + if (num_psf_gv_points > 0)
> + mask |= REG_GENMASK(num_psf_gv_points - 1, 0) << ADLS_PSF_PT_SHIFT;
> +
> for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
> new_crtc_state, i) {
> unsigned int old_data_rate =
> @@ -602,23 +686,44 @@ int intel_bw_atomic_check(struct intel_atomic_state *state)
> max_bw = max_data_rate;
> }
> if (max_data_rate >= data_rate)
> - allowed_points |= BIT(i);
> + allowed_points |= REG_FIELD_PREP(ADLS_QGV_PT_MASK, BIT(i));
> +
> drm_dbg_kms(&dev_priv->drm, "QGV point %d: max bw %d required %d\n",
> i, max_data_rate, data_rate);
> }
>
> + for (i = 0; i < num_psf_gv_points; i++) {
> + unsigned int max_data_rate = adl_psf_bw(dev_priv, i);
> +
> + if (max_data_rate >= data_rate)
> + allowed_points |= REG_FIELD_PREP(ADLS_PSF_PT_MASK, BIT(i));
> +
> + drm_dbg_kms(&dev_priv->drm, "PSF GV point %d: max bw %d"
> + " required %d\n",
> + i, max_data_rate, data_rate);
> + }
> +
> /*
> * BSpec states that we always should have at least one allowed point
> * left, so if we couldn't - simply reject the configuration for obvious
> * reasons.
> */
> - if (allowed_points == 0) {
> + if ((allowed_points & ADLS_QGV_PT_MASK) == 0) {
> drm_dbg_kms(&dev_priv->drm, "No QGV points provide sufficient memory"
> " bandwidth %d for display configuration(%d active planes).\n",
> data_rate, num_active_planes);
> return -EINVAL;
> }
>
> + if (num_psf_gv_points > 0) {
> + if ((allowed_points & ADLS_PSF_PT_MASK) == 0) {
> + drm_dbg_kms(&dev_priv->drm, "No PSF GV points provide sufficient memory"
> + " bandwidth %d for display configuration(%d active planes).\n",
> + data_rate, num_active_planes);
> + return -EINVAL;
> + }
> + }
> +
> /*
> * Leave only single point with highest bandwidth, if
> * we can't enable SAGV due to the increased memory latency it may
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 9cb02618ba15..16938e268e25 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -591,6 +591,9 @@ i915_fence_timeout(const struct drm_i915_private *i915)
>
> #define HAS_HW_SAGV_WM(i915) (DISPLAY_VER(i915) >= 13 && !IS_DGFX(i915))
>
> +/* Amount of PSF GV points, BSpec precisely defines this */
> +#define I915_NUM_PSF_GV_POINTS 3
> +
> struct ddi_vbt_port_info {
> /* Non-NULL if port present. */
> struct intel_bios_encoder_data *devdata;
> @@ -1103,12 +1106,16 @@ struct drm_i915_private {
> INTEL_DRAM_LPDDR5,
> } type;
> u8 num_qgv_points;
> + u8 num_psf_gv_points;
> } dram_info;
>
> struct intel_bw_info {
> /* for each QGV point */
> unsigned int deratedbw[I915_NUM_QGV_POINTS];
> + /* for each PSF GV point */
> + unsigned int psf_bw[I915_NUM_PSF_GV_POINTS];
> u8 num_qgv_points;
> + u8 num_psf_gv_points;
> u8 num_planes;
> } max_bw[6];
>
> diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> index 4dbe79009c0e..668f41476e40 100644
> --- a/drivers/gpu/drm/i915/i915_reg.h
> +++ b/drivers/gpu/drm/i915/i915_reg.h
> @@ -9336,9 +9336,13 @@ enum {
> #define ICL_PCODE_MEM_SUBSYSYSTEM_INFO 0xd
> #define ICL_PCODE_MEM_SS_READ_GLOBAL_INFO (0x0 << 8)
> #define ICL_PCODE_MEM_SS_READ_QGV_POINT_INFO(point) (((point) << 16) | (0x1 << 8))
> +#define ADL_PCODE_MEM_SS_READ_PSF_GV_INFO ((0) | (0x2 << 8))
> #define ICL_PCODE_SAGV_DE_MEM_SS_CONFIG 0xe
> #define ICL_PCODE_POINTS_RESTRICTED 0x0
> #define ICL_PCODE_POINTS_RESTRICTED_MASK 0x1
> +#define ADLS_PSF_PT_SHIFT 8
> +#define ADLS_QGV_PT_MASK REG_GENMASK(7, 0)
> +#define ADLS_PSF_PT_MASK REG_GENMASK(10, 8)
> #define GEN6_PCODE_READ_D_COMP 0x10
> #define GEN6_PCODE_WRITE_D_COMP 0x11
> #define ICL_PCODE_EXIT_TCCOLD 0x12
> diff --git a/drivers/gpu/drm/i915/intel_dram.c b/drivers/gpu/drm/i915/intel_dram.c
> index 1e53c017c30d..53ccd106785f 100644
> --- a/drivers/gpu/drm/i915/intel_dram.c
> +++ b/drivers/gpu/drm/i915/intel_dram.c
> @@ -468,6 +468,7 @@ static int icl_pcode_read_mem_global_info(struct drm_i915_private *dev_priv)
>
> dram_info->num_channels = (val & 0xf0) >> 4;
> dram_info->num_qgv_points = (val & 0xf00) >> 8;
> + dram_info->num_psf_gv_points = (val & 0x3000) >> 12;
>
> return 0;
> }
> --
> 2.24.1.485.gad05a3d8e5
>
--
Matt Roper
Graphics Software Engineer
VTT-OSGC Platform Enablement
Intel Corporation
(916) 356-2795
More information about the Intel-gfx
mailing list