drm/radeon: implement si_set_vce_clocks v2

For setting clocks with VCE v1.0

v2: (chk) rebased on current tree

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

5 files changed, 157 insertions, 0 deletions

diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 5999560d4284..3d74fdcac4bf 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -2578,6 +2578,13 @@ static inline struct radeon_fence *to_radeon_fence(struct fence *f)
 		tmp_ |= ((val) & ~(mask));			\
 		WREG32_PLL(reg, tmp_);				\
 	} while (0)
+#define WREG32_SMC_P(reg, val, mask)				\
+	do {							\
+		uint32_t tmp_ = RREG32_SMC(reg);		\
+		tmp_ &= (mask);					\
+		tmp_ |= ((val) & ~(mask));			\
+		WREG32_SMC(reg, tmp_);				\
+	} while (0)
 #define DREG32_SYS(sqf, rdev, reg) seq_printf((sqf), #reg " : 0x%08X\n", r100_mm_rreg((rdev), (reg), false))
 #define RREG32_IO(reg) r100_io_rreg(rdev, (reg))
 #define WREG32_IO(reg, v) r100_io_wreg(rdev, (reg), (v))
diff --git a/drivers/gpu/drm/radeon/radeon_asic.c b/drivers/gpu/drm/radeon/radeon_asic.c
index 8dbf5083c4ff..4227b30b5ce6 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.c
+++ b/drivers/gpu/drm/radeon/radeon_asic.c
@@ -1973,6 +1973,7 @@ static struct radeon_asic si_asic = {
 		.set_pcie_lanes = &r600_set_pcie_lanes,
 		.set_clock_gating = NULL,
 		.set_uvd_clocks = &si_set_uvd_clocks,
+		.set_vce_clocks = &si_set_vce_clocks,
 		.get_temperature = &si_get_temp,
 	},
 	.dpm = {
diff --git a/drivers/gpu/drm/radeon/radeon_asic.h b/drivers/gpu/drm/radeon/radeon_asic.h
index cb7787fedf1b..0469d440764b 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.h
+++ b/drivers/gpu/drm/radeon/radeon_asic.h
@@ -745,6 +745,7 @@ void si_dma_vm_flush(struct radeon_device *rdev, struct radeon_ring *ring,
 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);
+int si_set_vce_clocks(struct radeon_device *rdev, u32 evclk, u32 ecclk);
 int si_get_temp(struct radeon_device *rdev);
 int si_get_allowed_info_register(struct radeon_device *rdev,
 				 u32 reg, u32 *val);
diff --git a/drivers/gpu/drm/radeon/si.c b/drivers/gpu/drm/radeon/si.c
index 5326f753e107..0c04f6d1faf8 100644
--- a/drivers/gpu/drm/radeon/si.c
+++ b/drivers/gpu/drm/radeon/si.c
@@ -7675,3 +7675,124 @@ static void si_program_aspm(struct radeon_device *rdev)
 		}
 	}
 }
+
+int si_vce_send_vcepll_ctlreq(struct radeon_device *rdev)
+{
+        unsigned i;
+
+        /* make sure VCEPLL_CTLREQ is deasserted */
+        WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~UPLL_CTLREQ_MASK);
+
+        mdelay(10);
+
+        /* assert UPLL_CTLREQ */
+        WREG32_SMC_P(CG_VCEPLL_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_SMC(CG_VCEPLL_FUNC_CNTL) & mask) == mask)
+                        break;
+                mdelay(10);
+        }
+
+        /* deassert UPLL_CTLREQ */
+        WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~UPLL_CTLREQ_MASK);
+
+        if (i == 100) {
+                DRM_ERROR("Timeout setting UVD clocks!\n");
+                return -ETIMEDOUT;
+        }
+
+        return 0;
+}
+
+int si_set_vce_clocks(struct radeon_device *rdev, u32 evclk, u32 ecclk)
+{
+	unsigned fb_div = 0, evclk_div = 0, ecclk_div = 0;
+	int r;
+
+	/* bypass evclk and ecclk with bclk */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_2,
+		     EVCLK_SRC_SEL(1) | ECCLK_SRC_SEL(1),
+		     ~(EVCLK_SRC_SEL_MASK | ECCLK_SRC_SEL_MASK));
+
+	/* put PLL in bypass mode */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_BYPASS_EN_MASK,
+		     ~VCEPLL_BYPASS_EN_MASK);
+
+	if (!evclk || !ecclk) {
+		/* keep the Bypass mode, put PLL to sleep */
+		WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_SLEEP_MASK,
+			     ~VCEPLL_SLEEP_MASK);
+		return 0;
+	}
+
+	r = radeon_uvd_calc_upll_dividers(rdev, evclk, ecclk, 125000, 250000,
+					  16384, 0x03FFFFFF, 0, 128, 5,
+					  &fb_div, &evclk_div, &ecclk_div);
+	if (r)
+		return r;
+
+	/* set RESET_ANTI_MUX to 0 */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_5, 0, ~RESET_ANTI_MUX_MASK);
+
+	/* set VCO_MODE to 1 */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_VCO_MODE_MASK,
+		     ~VCEPLL_VCO_MODE_MASK);
+
+	/* toggle VCEPLL_SLEEP to 1 then back to 0 */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_SLEEP_MASK,
+		     ~VCEPLL_SLEEP_MASK);
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_SLEEP_MASK);
+
+	/* deassert VCEPLL_RESET */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_RESET_MASK);
+
+	mdelay(1);
+
+	r = si_vce_send_vcepll_ctlreq(rdev);
+	if (r)
+		return r;
+
+	/* assert VCEPLL_RESET again */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_RESET_MASK, ~VCEPLL_RESET_MASK);
+
+	/* disable spread spectrum. */
+	WREG32_SMC_P(CG_VCEPLL_SPREAD_SPECTRUM, 0, ~SSEN_MASK);
+
+	/* set feedback divider */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_3, VCEPLL_FB_DIV(fb_div), ~VCEPLL_FB_DIV_MASK);
+
+	/* set ref divider to 0 */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_REF_DIV_MASK);
+
+	/* set PDIV_A and PDIV_B */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_2,
+		     VCEPLL_PDIV_A(evclk_div) | VCEPLL_PDIV_B(ecclk_div),
+		     ~(VCEPLL_PDIV_A_MASK | VCEPLL_PDIV_B_MASK));
+
+	/* give the PLL some time to settle */
+	mdelay(15);
+
+	/* deassert PLL_RESET */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_RESET_MASK);
+
+	mdelay(15);
+
+	/* switch from bypass mode to normal mode */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_BYPASS_EN_MASK);
+
+	r = si_vce_send_vcepll_ctlreq(rdev);
+	if (r)
+		return r;
+
+	/* switch VCLK and DCLK selection */
+	WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_2,
+		     EVCLK_SRC_SEL(16) | ECCLK_SRC_SEL(16),
+		     ~(EVCLK_SRC_SEL_MASK | ECCLK_SRC_SEL_MASK));
+
+	mdelay(100);
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/radeon/sid.h b/drivers/gpu/drm/radeon/sid.h
index 3afac3013983..1630440c5136 100644
--- a/drivers/gpu/drm/radeon/sid.h
+++ b/drivers/gpu/drm/radeon/sid.h
@@ -1917,4 +1917,31 @@
 #define VCE_CMD_IB_AUTO					0x00000005
 #define VCE_CMD_SEMAPHORE				0x00000006
 
+/* discrete vce clocks */
+#define	CG_VCEPLL_FUNC_CNTL				0xc0030600
+#	define VCEPLL_RESET_MASK			0x00000001
+#	define VCEPLL_SLEEP_MASK			0x00000002
+#	define VCEPLL_BYPASS_EN_MASK			0x00000004
+#	define VCEPLL_CTLREQ_MASK			0x00000008
+#	define VCEPLL_VCO_MODE_MASK			0x00000600
+#	define VCEPLL_REF_DIV_MASK			0x003F0000
+#	define VCEPLL_CTLACK_MASK			0x40000000
+#	define VCEPLL_CTLACK2_MASK			0x80000000
+#define	CG_VCEPLL_FUNC_CNTL_2				0xc0030601
+#	define VCEPLL_PDIV_A(x)				((x) << 0)
+#	define VCEPLL_PDIV_A_MASK			0x0000007F
+#	define VCEPLL_PDIV_B(x)				((x) << 8)
+#	define VCEPLL_PDIV_B_MASK			0x00007F00
+#	define EVCLK_SRC_SEL(x)				((x) << 20)
+#	define EVCLK_SRC_SEL_MASK			0x01F00000
+#	define ECCLK_SRC_SEL(x)				((x) << 25)
+#	define ECCLK_SRC_SEL_MASK			0x3E000000
+#define	CG_VCEPLL_FUNC_CNTL_3				0xc0030602
+#	define VCEPLL_FB_DIV(x)				((x) << 0)
+#	define VCEPLL_FB_DIV_MASK			0x01FFFFFF
+#define	CG_VCEPLL_FUNC_CNTL_4				0xc0030603
+#define	CG_VCEPLL_FUNC_CNTL_5				0xc0030604
+#define	CG_VCEPLL_SPREAD_SPECTRUM			0xc0030606
+#	define VCEPLL_SSEN_MASK				0x00000001
+
 #endif