drm/radeon/kms: fix up rs780/rs880 display watermark calc for dpm

calculate the low and high watermarks based on the low and high
clocks for the current power state.  The dynamic pm hw will select
the appropriate watermark based on the internal dpm state.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 167 insertions, 124 deletions

diff --git a/drivers/gpu/drm/radeon/rs690.c b/drivers/gpu/drm/radeon/rs690.c
index 55880d5962c3..d8ddfb34545d 100644
--- a/drivers/gpu/drm/radeon/rs690.c
+++ b/drivers/gpu/drm/radeon/rs690.c
@@ -248,13 +248,16 @@ struct rs690_watermark {
 };
 
 static void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
-				  struct radeon_crtc *crtc,
-				  struct rs690_watermark *wm)
+					 struct radeon_crtc *crtc,
+					 struct rs690_watermark *wm,
+					 bool low)
 {
 	struct drm_display_mode *mode = &crtc->base.mode;
 	fixed20_12 a, b, c;
 	fixed20_12 pclk, request_fifo_depth, tolerable_latency, estimated_width;
 	fixed20_12 consumption_time, line_time, chunk_time, read_delay_latency;
+	fixed20_12 sclk, core_bandwidth, max_bandwidth;
+	u32 selected_sclk;
 
 	if (!crtc->base.enabled) {
 		/* FIXME: wouldn't it better to set priority mark to maximum */
@@ -262,6 +265,21 @@ static void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
 		return;
 	}
 
+	if (((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880)) &&
+	    (rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled)
+		selected_sclk = radeon_dpm_get_sclk(rdev, low);
+	else
+		selected_sclk = rdev->pm.current_sclk;
+
+	/* sclk in Mhz */
+	a.full = dfixed_const(100);
+	sclk.full = dfixed_const(selected_sclk);
+	sclk.full = dfixed_div(sclk, a);
+
+	/* core_bandwidth = sclk(Mhz) * 16 */
+	a.full = dfixed_const(16);
+	core_bandwidth.full = dfixed_div(rdev->pm.sclk, a);
+
 	if (crtc->vsc.full > dfixed_const(2))
 		wm->num_line_pair.full = dfixed_const(2);
 	else
@@ -322,36 +340,36 @@ static void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
 	wm->active_time.full = dfixed_div(wm->active_time, a);
 
 	/* Maximun bandwidth is the minimun bandwidth of all component */
-	rdev->pm.max_bandwidth = rdev->pm.core_bandwidth;
+	max_bandwidth = core_bandwidth;
 	if (rdev->mc.igp_sideport_enabled) {
-		if (rdev->pm.max_bandwidth.full > rdev->pm.sideport_bandwidth.full &&
+		if (max_bandwidth.full > rdev->pm.sideport_bandwidth.full &&
 			rdev->pm.sideport_bandwidth.full)
-			rdev->pm.max_bandwidth = rdev->pm.sideport_bandwidth;
+			max_bandwidth = rdev->pm.sideport_bandwidth;
 		read_delay_latency.full = dfixed_const(370 * 800 * 1000);
 		read_delay_latency.full = dfixed_div(read_delay_latency,
 			rdev->pm.igp_sideport_mclk);
 	} else {
-		if (rdev->pm.max_bandwidth.full > rdev->pm.k8_bandwidth.full &&
+		if (max_bandwidth.full > rdev->pm.k8_bandwidth.full &&
 			rdev->pm.k8_bandwidth.full)
-			rdev->pm.max_bandwidth = rdev->pm.k8_bandwidth;
-		if (rdev->pm.max_bandwidth.full > rdev->pm.ht_bandwidth.full &&
+			max_bandwidth = rdev->pm.k8_bandwidth;
+		if (max_bandwidth.full > rdev->pm.ht_bandwidth.full &&
 			rdev->pm.ht_bandwidth.full)
-			rdev->pm.max_bandwidth = rdev->pm.ht_bandwidth;
+			max_bandwidth = rdev->pm.ht_bandwidth;
 		read_delay_latency.full = dfixed_const(5000);
 	}
 
 	/* sclk = system clocks(ns) = 1000 / max_bandwidth / 16 */
 	a.full = dfixed_const(16);
-	rdev->pm.sclk.full = dfixed_mul(rdev->pm.max_bandwidth, a);
+	sclk.full = dfixed_mul(max_bandwidth, a);
 	a.full = dfixed_const(1000);
-	rdev->pm.sclk.full = dfixed_div(a, rdev->pm.sclk);
+	sclk.full = dfixed_div(a, sclk);
 	/* Determine chunk time
 	 * ChunkTime = the time it takes the DCP to send one chunk of data
 	 * to the LB which consists of pipeline delay and inter chunk gap
 	 * sclk = system clock(ns)
 	 */
 	a.full = dfixed_const(256 * 13);
-	chunk_time.full = dfixed_mul(rdev->pm.sclk, a);
+	chunk_time.full = dfixed_mul(sclk, a);
 	a.full = dfixed_const(10);
 	chunk_time.full = dfixed_div(chunk_time, a);
 
@@ -415,175 +433,200 @@ static void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
 	}
 }
 
-void rs690_bandwidth_update(struct radeon_device *rdev)
+static void rs690_compute_mode_priority(struct radeon_device *rdev,
+					struct rs690_watermark *wm0,
+					struct rs690_watermark *wm1,
+					struct drm_display_mode *mode0,
+					struct drm_display_mode *mode1,
+					u32 *d1mode_priority_a_cnt,
+					u32 *d2mode_priority_a_cnt)
 {
-	struct drm_display_mode *mode0 = NULL;
-	struct drm_display_mode *mode1 = NULL;
-	struct rs690_watermark wm0;
-	struct rs690_watermark wm1;
-	u32 tmp;
-	u32 d1mode_priority_a_cnt = S_006548_D1MODE_PRIORITY_A_OFF(1);
-	u32 d2mode_priority_a_cnt = S_006548_D1MODE_PRIORITY_A_OFF(1);
 	fixed20_12 priority_mark02, priority_mark12, fill_rate;
 	fixed20_12 a, b;
 
-	radeon_update_display_priority(rdev);
-
-	if (rdev->mode_info.crtcs[0]->base.enabled)
-		mode0 = &rdev->mode_info.crtcs[0]->base.mode;
-	if (rdev->mode_info.crtcs[1]->base.enabled)
-		mode1 = &rdev->mode_info.crtcs[1]->base.mode;
-	/*
-	 * Set display0/1 priority up in the memory controller for
-	 * modes if the user specifies HIGH for displaypriority
-	 * option.
-	 */
-	if ((rdev->disp_priority == 2) &&
-	    ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))) {
-		tmp = RREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER);
-		tmp &= C_000104_MC_DISP0R_INIT_LAT;
-		tmp &= C_000104_MC_DISP1R_INIT_LAT;
-		if (mode0)
-			tmp |= S_000104_MC_DISP0R_INIT_LAT(1);
-		if (mode1)
-			tmp |= S_000104_MC_DISP1R_INIT_LAT(1);
-		WREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER, tmp);
-	}
-	rs690_line_buffer_adjust(rdev, mode0, mode1);
-
-	if ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))
-		WREG32(R_006C9C_DCP_CONTROL, 0);
-	if ((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880))
-		WREG32(R_006C9C_DCP_CONTROL, 2);
-
-	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0);
-	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1);
-
-	tmp = (wm0.lb_request_fifo_depth - 1);
-	tmp |= (wm1.lb_request_fifo_depth - 1) << 16;
-	WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp);
+	*d1mode_priority_a_cnt = S_006548_D1MODE_PRIORITY_A_OFF(1);
+	*d2mode_priority_a_cnt = S_006548_D1MODE_PRIORITY_A_OFF(1);
 
 	if (mode0 && mode1) {
-		if (dfixed_trunc(wm0.dbpp) > 64)
-			a.full = dfixed_mul(wm0.dbpp, wm0.num_line_pair);
+		if (dfixed_trunc(wm0->dbpp) > 64)
+			a.full = dfixed_mul(wm0->dbpp, wm0->num_line_pair);
 		else
-			a.full = wm0.num_line_pair.full;
-		if (dfixed_trunc(wm1.dbpp) > 64)
-			b.full = dfixed_mul(wm1.dbpp, wm1.num_line_pair);
+			a.full = wm0->num_line_pair.full;
+		if (dfixed_trunc(wm1->dbpp) > 64)
+			b.full = dfixed_mul(wm1->dbpp, wm1->num_line_pair);
 		else
-			b.full = wm1.num_line_pair.full;
+			b.full = wm1->num_line_pair.full;
 		a.full += b.full;
-		fill_rate.full = dfixed_div(wm0.sclk, a);
-		if (wm0.consumption_rate.full > fill_rate.full) {
-			b.full = wm0.consumption_rate.full - fill_rate.full;
-			b.full = dfixed_mul(b, wm0.active_time);
-			a.full = dfixed_mul(wm0.worst_case_latency,
-						wm0.consumption_rate);
+		fill_rate.full = dfixed_div(wm0->sclk, a);
+		if (wm0->consumption_rate.full > fill_rate.full) {
+			b.full = wm0->consumption_rate.full - fill_rate.full;
+			b.full = dfixed_mul(b, wm0->active_time);
+			a.full = dfixed_mul(wm0->worst_case_latency,
+						wm0->consumption_rate);
 			a.full = a.full + b.full;
 			b.full = dfixed_const(16 * 1000);
 			priority_mark02.full = dfixed_div(a, b);
 		} else {
-			a.full = dfixed_mul(wm0.worst_case_latency,
-						wm0.consumption_rate);
+			a.full = dfixed_mul(wm0->worst_case_latency,
+						wm0->consumption_rate);
 			b.full = dfixed_const(16 * 1000);
 			priority_mark02.full = dfixed_div(a, b);
 		}
-		if (wm1.consumption_rate.full > fill_rate.full) {
-			b.full = wm1.consumption_rate.full - fill_rate.full;
-			b.full = dfixed_mul(b, wm1.active_time);
-			a.full = dfixed_mul(wm1.worst_case_latency,
-						wm1.consumption_rate);
+		if (wm1->consumption_rate.full > fill_rate.full) {
+			b.full = wm1->consumption_rate.full - fill_rate.full;
+			b.full = dfixed_mul(b, wm1->active_time);
+			a.full = dfixed_mul(wm1->worst_case_latency,
+						wm1->consumption_rate);
 			a.full = a.full + b.full;
 			b.full = dfixed_const(16 * 1000);
 			priority_mark12.full = dfixed_div(a, b);
 		} else {
-			a.full = dfixed_mul(wm1.worst_case_latency,
-						wm1.consumption_rate);
+			a.full = dfixed_mul(wm1->worst_case_latency,
+						wm1->consumption_rate);
 			b.full = dfixed_const(16 * 1000);
 			priority_mark12.full = dfixed_div(a, b);
 		}
-		if (wm0.priority_mark.full > priority_mark02.full)
-			priority_mark02.full = wm0.priority_mark.full;
+		if (wm0->priority_mark.full > priority_mark02.full)
+			priority_mark02.full = wm0->priority_mark.full;
 		if (dfixed_trunc(priority_mark02) < 0)
 			priority_mark02.full = 0;
-		if (wm0.priority_mark_max.full > priority_mark02.full)
-			priority_mark02.full = wm0.priority_mark_max.full;
-		if (wm1.priority_mark.full > priority_mark12.full)
-			priority_mark12.full = wm1.priority_mark.full;
+		if (wm0->priority_mark_max.full > priority_mark02.full)
+			priority_mark02.full = wm0->priority_mark_max.full;
+		if (wm1->priority_mark.full > priority_mark12.full)
+			priority_mark12.full = wm1->priority_mark.full;
 		if (dfixed_trunc(priority_mark12) < 0)
 			priority_mark12.full = 0;
-		if (wm1.priority_mark_max.full > priority_mark12.full)
-			priority_mark12.full = wm1.priority_mark_max.full;
-		d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
-		d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
+		if (wm1->priority_mark_max.full > priority_mark12.full)
+			priority_mark12.full = wm1->priority_mark_max.full;
+		*d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
+		*d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
 		if (rdev->disp_priority == 2) {
-			d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
-			d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
+			*d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
+			*d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
 		}
 	} else if (mode0) {
-		if (dfixed_trunc(wm0.dbpp) > 64)
-			a.full = dfixed_mul(wm0.dbpp, wm0.num_line_pair);
+		if (dfixed_trunc(wm0->dbpp) > 64)
+			a.full = dfixed_mul(wm0->dbpp, wm0->num_line_pair);
 		else
-			a.full = wm0.num_line_pair.full;
-		fill_rate.full = dfixed_div(wm0.sclk, a);
-		if (wm0.consumption_rate.full > fill_rate.full) {
-			b.full = wm0.consumption_rate.full - fill_rate.full;
-			b.full = dfixed_mul(b, wm0.active_time);
-			a.full = dfixed_mul(wm0.worst_case_latency,
-						wm0.consumption_rate);
+			a.full = wm0->num_line_pair.full;
+		fill_rate.full = dfixed_div(wm0->sclk, a);
+		if (wm0->consumption_rate.full > fill_rate.full) {
+			b.full = wm0->consumption_rate.full - fill_rate.full;
+			b.full = dfixed_mul(b, wm0->active_time);
+			a.full = dfixed_mul(wm0->worst_case_latency,
+						wm0->consumption_rate);
 			a.full = a.full + b.full;
 			b.full = dfixed_const(16 * 1000);
 			priority_mark02.full = dfixed_div(a, b);
 		} else {
-			a.full = dfixed_mul(wm0.worst_case_latency,
-						wm0.consumption_rate);
+			a.full = dfixed_mul(wm0->worst_case_latency,
+						wm0->consumption_rate);
 			b.full = dfixed_const(16 * 1000);
 			priority_mark02.full = dfixed_div(a, b);
 		}
-		if (wm0.priority_mark.full > priority_mark02.full)
-			priority_mark02.full = wm0.priority_mark.full;
+		if (wm0->priority_mark.full > priority_mark02.full)
+			priority_mark02.full = wm0->priority_mark.full;
 		if (dfixed_trunc(priority_mark02) < 0)
 			priority_mark02.full = 0;
-		if (wm0.priority_mark_max.full > priority_mark02.full)
-			priority_mark02.full = wm0.priority_mark_max.full;
-		d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
+		if (wm0->priority_mark_max.full > priority_mark02.full)
+			priority_mark02.full = wm0->priority_mark_max.full;
+		*d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
 		if (rdev->disp_priority == 2)
-			d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
+			*d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
 	} else if (mode1) {
-		if (dfixed_trunc(wm1.dbpp) > 64)
-			a.full = dfixed_mul(wm1.dbpp, wm1.num_line_pair);
+		if (dfixed_trunc(wm1->dbpp) > 64)
+			a.full = dfixed_mul(wm1->dbpp, wm1->num_line_pair);
 		else
-			a.full = wm1.num_line_pair.full;
-		fill_rate.full = dfixed_div(wm1.sclk, a);
-		if (wm1.consumption_rate.full > fill_rate.full) {
-			b.full = wm1.consumption_rate.full - fill_rate.full;
-			b.full = dfixed_mul(b, wm1.active_time);
-			a.full = dfixed_mul(wm1.worst_case_latency,
-						wm1.consumption_rate);
+			a.full = wm1->num_line_pair.full;
+		fill_rate.full = dfixed_div(wm1->sclk, a);
+		if (wm1->consumption_rate.full > fill_rate.full) {
+			b.full = wm1->consumption_rate.full - fill_rate.full;
+			b.full = dfixed_mul(b, wm1->active_time);
+			a.full = dfixed_mul(wm1->worst_case_latency,
+						wm1->consumption_rate);
 			a.full = a.full + b.full;
 			b.full = dfixed_const(16 * 1000);
 			priority_mark12.full = dfixed_div(a, b);
 		} else {
-			a.full = dfixed_mul(wm1.worst_case_latency,
-						wm1.consumption_rate);
+			a.full = dfixed_mul(wm1->worst_case_latency,
+						wm1->consumption_rate);
 			b.full = dfixed_const(16 * 1000);
 			priority_mark12.full = dfixed_div(a, b);
 		}
-		if (wm1.priority_mark.full > priority_mark12.full)
-			priority_mark12.full = wm1.priority_mark.full;
+		if (wm1->priority_mark.full > priority_mark12.full)
+			priority_mark12.full = wm1->priority_mark.full;
 		if (dfixed_trunc(priority_mark12) < 0)
 			priority_mark12.full = 0;
-		if (wm1.priority_mark_max.full > priority_mark12.full)
-			priority_mark12.full = wm1.priority_mark_max.full;
-		d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
+		if (wm1->priority_mark_max.full > priority_mark12.full)
+			priority_mark12.full = wm1->priority_mark_max.full;
+		*d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
 		if (rdev->disp_priority == 2)
-			d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
+			*d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
 	}
+}
+
+void rs690_bandwidth_update(struct radeon_device *rdev)
+{
+	struct drm_display_mode *mode0 = NULL;
+	struct drm_display_mode *mode1 = NULL;
+	struct rs690_watermark wm0_high, wm0_low;
+	struct rs690_watermark wm1_high, wm1_low;
+	u32 tmp;
+	u32 d1mode_priority_a_cnt, d1mode_priority_b_cnt;
+	u32 d2mode_priority_a_cnt, d2mode_priority_b_cnt;
+
+	radeon_update_display_priority(rdev);
+
+	if (rdev->mode_info.crtcs[0]->base.enabled)
+		mode0 = &rdev->mode_info.crtcs[0]->base.mode;
+	if (rdev->mode_info.crtcs[1]->base.enabled)
+		mode1 = &rdev->mode_info.crtcs[1]->base.mode;
+	/*
+	 * Set display0/1 priority up in the memory controller for
+	 * modes if the user specifies HIGH for displaypriority
+	 * option.
+	 */
+	if ((rdev->disp_priority == 2) &&
+	    ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))) {
+		tmp = RREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER);
+		tmp &= C_000104_MC_DISP0R_INIT_LAT;
+		tmp &= C_000104_MC_DISP1R_INIT_LAT;
+		if (mode0)
+			tmp |= S_000104_MC_DISP0R_INIT_LAT(1);
+		if (mode1)
+			tmp |= S_000104_MC_DISP1R_INIT_LAT(1);
+		WREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER, tmp);
+	}
+	rs690_line_buffer_adjust(rdev, mode0, mode1);
+
+	if ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))
+		WREG32(R_006C9C_DCP_CONTROL, 0);
+	if ((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880))
+		WREG32(R_006C9C_DCP_CONTROL, 2);
+
+	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0_high, false);
+	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1_high, false);
+
+	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0_low, true);
+	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1_low, true);
+
+	tmp = (wm0_high.lb_request_fifo_depth - 1);
+	tmp |= (wm1_high.lb_request_fifo_depth - 1) << 16;
+	WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp);
+
+	rs690_compute_mode_priority(rdev,
+				    &wm0_high, &wm1_high,
+				    mode0, mode1,
+				    &d1mode_priority_a_cnt, &d2mode_priority_a_cnt);
+	rs690_compute_mode_priority(rdev,
+				    &wm0_low, &wm1_low,
+				    mode0, mode1,
+				    &d1mode_priority_b_cnt, &d2mode_priority_b_cnt);
 
 	WREG32(R_006548_D1MODE_PRIORITY_A_CNT, d1mode_priority_a_cnt);
-	WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, d1mode_priority_a_cnt);
+	WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, d1mode_priority_b_cnt);
 	WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, d2mode_priority_a_cnt);
-	WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, d2mode_priority_a_cnt);
+	WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, d2mode_priority_b_cnt);
 }
 
 uint32_t rs690_mc_rreg(struct radeon_device *rdev, uint32_t reg)