Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Jacopo Mondi <jacopo@jmondi.org>
To:     Laurent Pinchart <laurent.pinchart@ideasonboard.com>,
        David Airlie <airlied@linux.ie>
Cc:     Jacopo Mondi <jacopo@jmondi.org>,
        dri-devel@lists.freedesktop.org (open list:DRM DRIVERS FOR RENESAS),
        linux-renesas-soc@vger.kernel.org (open list:DRM DRIVERS FOR RENESAS),
        linux-kernel@vger.kernel.org (open list)
Subject: [PATCH 3/3] drm: rcar-du: Improve non-DPLL clock selection
Date:   Mon, 30 Jul 2018 19:20:14 +0200
Message-Id: <1532971214-17962-4-git-send-email-jacopo@jmondi.org>
In-Reply-To: <1532971214-17962-1-git-send-email-jacopo@jmondi.org>
References: <1532971214-17962-1-git-send-email-jacopo@jmondi.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

DU channels not equipped with a DPLL use an internal (aka SoC provided) or
external clock source combined with an internal divider to generate the
desired output dot clock frequency.

The current clock selection procedure does not fully exploit the ability
of external clock sources to generate the exact dot clock frequency by
themselves, but relies instead on tuning the internal DU clock divider
only, resulting in a less precise clock generation process.

When possible, and desirable, ask the external clock source for the exact
output dot clock frequency, and test all available clock sources (internally
generated clock, externally generated clock rounded to the closest possible
frequency, and not rounded externally generated clock) to better approximate
the desired output dot clock.

This patch specifically targets platforms (like Salvator-X[S] and ULCBs)
where the DU's input dotclock.in is generated by the versaclock VC5 clock
source, which is capable of generating the exact rate the DU needs as pixel
clock output.

This patch fixes higher resolution modes wich requires an high pixel clock
output currently not working on non-HDMI DU channel (as VGA 1920x1080@60Hz).

Fixes: 1b30dbde8 "drm: rcar-du: Add support for external pixel clock"

Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
---
 drivers/gpu/drm/rcar-du/rcar_du_crtc.c | 96 ++++++++++++++++++++++++++--------
 1 file changed, 73 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/rcar-du/rcar_du_crtc.c b/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
index 4d7907c..0dfb28ff 100644
--- a/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
+++ b/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
@@ -207,12 +207,12 @@ static void rcar_du_crtc_set_display_timing(struct rcar_du_crtc *rcrtc)
 {
 	const struct drm_display_mode *mode = &rcrtc->crtc.state->adjusted_mode;
 	struct rcar_du_device *rcdu = rcrtc->group->dev;
-	unsigned long mode_clock = mode->clock * 1000;
+	unsigned long mode_pixelclock = mode->clock * 1000;
 	u32 dsmr;
 	u32 escr;
 
 	if (rcdu->info->dpll_ch & (1 << rcrtc->index)) {
-		unsigned long target = mode_clock;
+		unsigned long target = mode_pixelclock;
 		struct dpll_info dpll = { 0 };
 		unsigned long extclk;
 		u32 dpllcr;
@@ -258,42 +258,92 @@ static void rcar_du_crtc_set_display_timing(struct rcar_du_crtc *rcrtc)
 
 		escr = ESCR_DCLKSEL_DCLKIN | div;
 	} else {
-		unsigned long clk;
+		struct du_clkin_desc {
+			unsigned long dist;
+			unsigned long rate;
+			struct clk *clk;
+			u32 escr;
+		} du_clkins[3];
+		struct du_clkin_desc *du_clkin = du_clkins;
+		struct du_clkin_desc *clkin_best;
+		unsigned long best = -1UL;
+		unsigned long pixelclock;
+		unsigned long cpgrate;
 		u32 div;
 
 		/*
 		 * Compute the clock divisor and select the internal or external
 		 * dot clock based on the requested frequency.
 		 */
-		clk = clk_get_rate(rcrtc->clock);
-		div = DIV_ROUND_CLOSEST(clk, mode_clock);
-		div = clamp(div, 1U, 64U) - 1;
+		cpgrate = clk_get_rate(rcrtc->clock);
+		div = clamp(DIV_ROUND_CLOSEST(cpgrate, mode_pixelclock),
+			    1UL, 64UL) - 1;
+		pixelclock = cpgrate / (div + 1);
 
-		escr = ESCR_DCLKSEL_CLKS | div;
+		du_clkin->dist = abs(pixelclock - mode_pixelclock);
+		du_clkin->escr = ESCR_DCLKSEL_CLKS | div;
+		du_clkin->clk = rcrtc->clock;
+		du_clkin->rate = cpgrate;
+		clkin_best = du_clkin;
 
 		if (rcrtc->extclock) {
-			unsigned long extclk;
+			unsigned long roundrate;
 			unsigned long extrate;
-			unsigned long rate;
-			u32 extdiv;
-
-			extclk = clk_get_rate(rcrtc->extclock);
-			extdiv = DIV_ROUND_CLOSEST(extclk, mode_clock);
-			extdiv = clamp(extdiv, 1U, 64U) - 1;
 
-			extrate = extclk / (extdiv + 1);
-			rate = clk / (div + 1);
-
-			if (abs((long)extrate - (long)mode_clock) <
-			    abs((long)rate - (long)mode_clock))
-				escr = ESCR_DCLKSEL_DCLKIN | extdiv;
+			/*
+			 * If an external clock source is present ask it
+			 * for the exact dot clock rate, and test all possible
+			 * combinations (internal, external rounded, external
+			 * not rounded) to achieve the best possible clock
+			 * accuracy.
+			 */
+			extrate = clk_get_rate(rcrtc->extclock);
+			roundrate = clk_round_rate(rcrtc->extclock,
+						   mode_pixelclock);
+
+			if (roundrate == mode_pixelclock) {
+				/* We can't do better than this... */
+				clk_set_rate(rcrtc->extclock, roundrate);
+				escr = ESCR_DCLKSEL_DCLKIN;
+				goto set_escr;
+			}
 
-			dev_dbg(rcrtc->group->dev->dev,
-				"mode clock %lu extrate %lu rate %lu ESCR 0x%08x\n",
-				mode_clock, extrate, rate, escr);
+			/* Test the external clock 'plain' rate. */
+			du_clkin++;
+			div = clamp(DIV_ROUND_CLOSEST(extrate, mode_pixelclock),
+				    1UL, 64UL) - 1;
+			pixelclock = extrate / (div + 1);
+			du_clkin->dist = abs(pixelclock - mode_pixelclock);
+			du_clkin->escr = ESCR_DCLKSEL_DCLKIN | div;
+			du_clkin->clk = rcrtc->extclock;
+			du_clkin->rate = extrate;
+
+			/* Test the external clock 'rounded' rate. */
+			du_clkin++;
+			div = clamp(DIV_ROUND_CLOSEST(roundrate, mode_pixelclock),
+				    1UL, 64UL) - 1;
+			pixelclock = roundrate / (div + 1);
+			du_clkin->dist = abs(pixelclock - mode_pixelclock);
+			du_clkin->escr = ESCR_DCLKSEL_DCLKIN | div;
+			du_clkin->clk = rcrtc->extclock;
+			du_clkin->rate = roundrate;
+
+			/* Find out the best approximation we got. */
+			for (; du_clkin >= du_clkins; --du_clkin) {
+				if (du_clkin->dist < best) {
+					best = du_clkin->dist;
+					clkin_best = du_clkin;
+				}
+			}
+			clk_set_rate(clkin_best->clk, clkin_best->rate);
 		}
+
+		escr = clkin_best->escr;
 	}
 
+set_escr:
+	dev_err(rcrtc->group->dev->dev, "%s: ESCR 0x%08x\n", __func__, escr);
+
 	rcar_du_group_write(rcrtc->group, rcrtc->index % 2 ? ESCR2 : ESCR,
 			    escr);
 	rcar_du_group_write(rcrtc->group, rcrtc->index % 2 ? OTAR2 : OTAR, 0);
-- 
2.7.4