[PATCH 08/11] drm/radeon: add set_uvd_clocks callback for SI

Christian König deathsimple at vodafone.de
Sat Apr 6 06:24:58 PDT 2013


Signed-off-by: Christian König <christian.koenig at amd.com>
Reviewed-by: Jerome Glisse <jglisse at redhat.com>
---
 drivers/gpu/drm/radeon/radeon_asic.c |    1 +
 drivers/gpu/drm/radeon/radeon_asic.h |    1 +
 drivers/gpu/drm/radeon/si.c          |  167 ++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/radeon/sid.h         |   29 ++++++
 4 files changed, 198 insertions(+)

diff --git a/drivers/gpu/drm/radeon/radeon_asic.c b/drivers/gpu/drm/radeon/radeon_asic.c
index a65312c..03228cb 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.c
+++ b/drivers/gpu/drm/radeon/radeon_asic.c
@@ -1887,6 +1887,7 @@ static struct radeon_asic si_asic = {
 		.get_pcie_lanes = NULL,
 		.set_pcie_lanes = NULL,
 		.set_clock_gating = NULL,
+		.set_uvd_clocks = &si_set_uvd_clocks,
 	},
 	.pflip = {
 		.pre_page_flip = &evergreen_pre_page_flip,
diff --git a/drivers/gpu/drm/radeon/radeon_asic.h b/drivers/gpu/drm/radeon/radeon_asic.h
index 54a7ef7..365c964 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.h
+++ b/drivers/gpu/drm/radeon/radeon_asic.h
@@ -545,5 +545,6 @@ int si_copy_dma(struct radeon_device *rdev,
 void si_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm);
 u32 si_get_xclk(struct radeon_device *rdev);
 uint64_t si_get_gpu_clock_counter(struct radeon_device *rdev);
+int si_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk);
 
 #endif
diff --git a/drivers/gpu/drm/radeon/si.c b/drivers/gpu/drm/radeon/si.c
index cc9fe39..472d9fb 100644
--- a/drivers/gpu/drm/radeon/si.c
+++ b/drivers/gpu/drm/radeon/si.c
@@ -4666,3 +4666,170 @@ uint64_t si_get_gpu_clock_counter(struct radeon_device *rdev)
 	mutex_unlock(&rdev->gpu_clock_mutex);
 	return clock;
 }
+
+static int si_uvd_calc_post_div(unsigned target_freq,
+				unsigned vco_freq,
+				unsigned *div)
+{
+	/* target larger than vco frequency ? */
+	if (vco_freq < target_freq)
+		return -1; /* forget it */
+
+	/* Fclk = Fvco / PDIV */
+	*div = vco_freq / target_freq;
+
+	/* we alway need a frequency less than or equal the target */
+	if ((vco_freq / *div) > target_freq)
+		*div += 1;
+
+	/* dividers above 5 must be even */
+	if (*div > 5 && *div % 2)
+		*div += 1;
+
+	/* out of range ? */
+	if (*div >= 128)
+		return -1; /* forget it */
+
+	return vco_freq / *div;
+}
+
+static int si_uvd_send_upll_ctlreq(struct radeon_device *rdev)
+{
+	unsigned i;
+
+	/* assert UPLL_CTLREQ */
+	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK);
+
+	/* wait for CTLACK and CTLACK2 to get asserted */
+	for (i = 0; i < 100; ++i) {
+		uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK;
+		if ((RREG32(CG_UPLL_FUNC_CNTL) & mask) == mask)
+			break;
+		mdelay(10);
+	}
+	if (i == 100)
+		return -ETIMEDOUT;
+
+	/* deassert UPLL_CTLREQ */
+	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_CTLREQ_MASK);
+
+	return 0;
+}
+
+int si_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk)
+{
+	/* start off with something large */
+	int optimal_diff_score = 0x7FFFFFF;
+	unsigned optimal_fb_div = 0, optimal_vclk_div = 0;
+	unsigned optimal_dclk_div = 0, optimal_vco_freq = 0;
+	unsigned vco_freq;
+	int r;
+
+	/* loop through vco from low to high */
+	for (vco_freq = 125000; vco_freq <= 250000; vco_freq += 100) {
+		unsigned fb_div = vco_freq / rdev->clock.spll.reference_freq * 16384;
+		int calc_clk, diff_score, diff_vclk, diff_dclk;
+		unsigned vclk_div, dclk_div;
+
+		/* fb div out of range ? */
+		if (fb_div > 0x03FFFFFF)
+			break; /* it can oly get worse */
+
+		/* calc vclk with current vco freq. */
+		calc_clk = si_uvd_calc_post_div(vclk, vco_freq, &vclk_div);
+		if (calc_clk == -1)
+			break; /* vco is too big, it has to stop. */
+		diff_vclk = vclk - calc_clk;
+
+		/* calc dclk with current vco freq. */
+		calc_clk = si_uvd_calc_post_div(dclk, vco_freq, &dclk_div);
+		if (calc_clk == -1)
+			break; /* vco is too big, it has to stop. */
+		diff_dclk = dclk - calc_clk;
+
+		/* determine if this vco setting is better than current optimal settings */
+		diff_score = abs(diff_vclk) + abs(diff_dclk);
+		if (diff_score < optimal_diff_score) {
+			optimal_fb_div = fb_div;
+			optimal_vclk_div = vclk_div;
+			optimal_dclk_div = dclk_div;
+			optimal_vco_freq = vco_freq;
+			optimal_diff_score = diff_score;
+			if (optimal_diff_score == 0)
+				break; /* it can't get better than this */
+		}
+	}
+
+	/* set RESET_ANTI_MUX to 0 */
+	WREG32_P(CG_UPLL_FUNC_CNTL_5, 0, ~RESET_ANTI_MUX_MASK);
+
+	/* set VCO_MODE to 1 */
+	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_VCO_MODE_MASK, ~UPLL_VCO_MODE_MASK);
+
+	/* toggle UPLL_SLEEP to 1 then back to 0 */
+	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_SLEEP_MASK, ~UPLL_SLEEP_MASK);
+	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_SLEEP_MASK);
+
+	/* deassert UPLL_RESET */
+	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_RESET_MASK);
+
+	mdelay(1);
+
+	/* bypass vclk and dclk with bclk */
+	WREG32_P(CG_UPLL_FUNC_CNTL_2,
+		VCLK_SRC_SEL(1) | DCLK_SRC_SEL(1),
+		~(VCLK_SRC_SEL_MASK | DCLK_SRC_SEL_MASK));
+
+	/* put PLL in bypass mode */
+	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_BYPASS_EN_MASK, ~UPLL_BYPASS_EN_MASK);
+
+	r = si_uvd_send_upll_ctlreq(rdev);
+	if (r)
+		return r;
+
+	/* assert UPLL_RESET again */
+	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_RESET_MASK, ~UPLL_RESET_MASK);
+
+	/* disable spread spectrum. */
+	WREG32_P(CG_UPLL_SPREAD_SPECTRUM, 0, ~SSEN_MASK);
+
+	/* set feedback divider */
+	WREG32_P(CG_UPLL_FUNC_CNTL_3, UPLL_FB_DIV(optimal_fb_div), ~UPLL_FB_DIV_MASK);
+
+	/* set ref divider to 0 */
+	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_REF_DIV_MASK);
+
+	if (optimal_vco_freq < 187500)
+		WREG32_P(CG_UPLL_FUNC_CNTL_4, 0, ~UPLL_SPARE_ISPARE9);
+	else
+		WREG32_P(CG_UPLL_FUNC_CNTL_4, UPLL_SPARE_ISPARE9, ~UPLL_SPARE_ISPARE9);
+
+	/* set PDIV_A and PDIV_B */
+	WREG32_P(CG_UPLL_FUNC_CNTL_2,
+		UPLL_PDIV_A(optimal_vclk_div) | UPLL_PDIV_B(optimal_dclk_div),
+		~(UPLL_PDIV_A_MASK | UPLL_PDIV_B_MASK));
+
+	/* give the PLL some time to settle */
+	mdelay(15);
+
+	/* deassert PLL_RESET */
+	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_RESET_MASK);
+
+	mdelay(15);
+
+	/* switch from bypass mode to normal mode */
+	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_BYPASS_EN_MASK);
+
+	r = si_uvd_send_upll_ctlreq(rdev);
+	if (r)
+		return r;
+
+	/* switch VCLK and DCLK selection */
+	WREG32_P(CG_UPLL_FUNC_CNTL_2,
+		VCLK_SRC_SEL(2) | DCLK_SRC_SEL(2),
+		~(VCLK_SRC_SEL_MASK | DCLK_SRC_SEL_MASK));
+
+	mdelay(100);
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/radeon/sid.h b/drivers/gpu/drm/radeon/sid.h
index 759f682..c84e09b 100644
--- a/drivers/gpu/drm/radeon/sid.h
+++ b/drivers/gpu/drm/radeon/sid.h
@@ -29,6 +29,35 @@
 #define TAHITI_GB_ADDR_CONFIG_GOLDEN        0x12011003
 #define VERDE_GB_ADDR_CONFIG_GOLDEN         0x12010002
 
+/* discrete uvd clocks */
+#define	CG_UPLL_FUNC_CNTL				0x634
+#	define UPLL_RESET_MASK				0x00000001
+#	define UPLL_SLEEP_MASK				0x00000002
+#	define UPLL_BYPASS_EN_MASK			0x00000004
+#	define UPLL_CTLREQ_MASK				0x00000008
+#	define UPLL_VCO_MODE_MASK			0x00000600
+#	define UPLL_REF_DIV_MASK			0x001F0000
+#	define UPLL_CTLACK_MASK				0x40000000
+#	define UPLL_CTLACK2_MASK			0x80000000
+#define	CG_UPLL_FUNC_CNTL_2				0x638
+#	define UPLL_PDIV_A(x)				((x) << 0)
+#	define UPLL_PDIV_A_MASK				0x0000007F
+#	define UPLL_PDIV_B(x)				((x) << 8)
+#	define UPLL_PDIV_B_MASK				0x00007F00
+#	define VCLK_SRC_SEL(x)				((x) << 20)
+#	define VCLK_SRC_SEL_MASK			0x01F00000
+#	define DCLK_SRC_SEL(x)				((x) << 25)
+#	define DCLK_SRC_SEL_MASK			0x3E000000
+#define	CG_UPLL_FUNC_CNTL_3				0x63C
+#	define UPLL_FB_DIV(x)				((x) << 0)
+#	define UPLL_FB_DIV_MASK				0x01FFFFFF
+#define	CG_UPLL_FUNC_CNTL_4                             0x644
+#	define UPLL_SPARE_ISPARE9			0x00020000
+#define	CG_UPLL_FUNC_CNTL_5				0x648
+#	define RESET_ANTI_MUX_MASK			0x00000200
+#define	CG_UPLL_SPREAD_SPECTRUM				0x650
+#	define SSEN_MASK				0x00000001
+
 #define	CG_MULT_THERMAL_STATUS					0x714
 #define		ASIC_MAX_TEMP(x)				((x) << 0)
 #define		ASIC_MAX_TEMP_MASK				0x000001ff
-- 
1.7.9.5



More information about the dri-devel mailing list