drm/radeon/kms: fix bandwidth computation on avivo hardware

Fix bandwidth computation and crtc priority in memory controller
so that crtc memory request are fullfill in time to avoid display
artifact.

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

19 files changed, 2377 insertions, 1354 deletions

diff --git a/drivers/gpu/drm/radeon/atombios_crtc.c b/drivers/gpu/drm/radeon/atombios_crtc.c
index eac26cdb5da..74d034f77c6 100644
--- a/drivers/gpu/drm/radeon/atombios_crtc.c
+++ b/drivers/gpu/drm/radeon/atombios_crtc.c
@@ -31,6 +31,132 @@
 #include "atom.h"
 #include "atom-bits.h"
 
+static void atombios_overscan_setup(struct drm_crtc *crtc,
+				    struct drm_display_mode *mode,
+				    struct drm_display_mode *adjusted_mode)
+{
+	struct drm_device *dev = crtc->dev;
+	struct radeon_device *rdev = dev->dev_private;
+	struct radeon_crtc *radeon_crtc = to_radeon_crtc(crtc);
+	SET_CRTC_OVERSCAN_PS_ALLOCATION args;
+	int index = GetIndexIntoMasterTable(COMMAND, SetCRTC_OverScan);
+	int a1, a2;
+
+	memset(&args, 0, sizeof(args));
+
+	args.usOverscanRight = 0;
+	args.usOverscanLeft = 0;
+	args.usOverscanBottom = 0;
+	args.usOverscanTop = 0;
+	args.ucCRTC = radeon_crtc->crtc_id;
+
+	switch (radeon_crtc->rmx_type) {
+	case RMX_CENTER:
+		args.usOverscanTop = (adjusted_mode->crtc_vdisplay - mode->crtc_vdisplay) / 2;
+		args.usOverscanBottom = (adjusted_mode->crtc_vdisplay - mode->crtc_vdisplay) / 2;
+		args.usOverscanLeft = (adjusted_mode->crtc_hdisplay - mode->crtc_hdisplay) / 2;
+		args.usOverscanRight = (adjusted_mode->crtc_hdisplay - mode->crtc_hdisplay) / 2;
+		atom_execute_table(rdev->mode_info.atom_context, index, (uint32_t *)&args);
+		break;
+	case RMX_ASPECT:
+		a1 = mode->crtc_vdisplay * adjusted_mode->crtc_hdisplay;
+		a2 = adjusted_mode->crtc_vdisplay * mode->crtc_hdisplay;
+
+		if (a1 > a2) {
+			args.usOverscanLeft = (adjusted_mode->crtc_hdisplay - (a2 / mode->crtc_vdisplay)) / 2;
+			args.usOverscanRight = (adjusted_mode->crtc_hdisplay - (a2 / mode->crtc_vdisplay)) / 2;
+		} else if (a2 > a1) {
+			args.usOverscanLeft = (adjusted_mode->crtc_vdisplay - (a1 / mode->crtc_hdisplay)) / 2;
+			args.usOverscanRight = (adjusted_mode->crtc_vdisplay - (a1 / mode->crtc_hdisplay)) / 2;
+		}
+		atom_execute_table(rdev->mode_info.atom_context, index, (uint32_t *)&args);
+		break;
+	case RMX_FULL:
+	default:
+		args.usOverscanRight = 0;
+		args.usOverscanLeft = 0;
+		args.usOverscanBottom = 0;
+		args.usOverscanTop = 0;
+		atom_execute_table(rdev->mode_info.atom_context, index, (uint32_t *)&args);
+		break;
+	}
+}
+
+static void atombios_scaler_setup(struct drm_crtc *crtc)
+{
+	struct drm_device *dev = crtc->dev;
+	struct radeon_device *rdev = dev->dev_private;
+	struct radeon_crtc *radeon_crtc = to_radeon_crtc(crtc);
+	ENABLE_SCALER_PS_ALLOCATION args;
+	int index = GetIndexIntoMasterTable(COMMAND, EnableScaler);
+	/* fixme - fill in enc_priv for atom dac */
+	enum radeon_tv_std tv_std = TV_STD_NTSC;
+
+	if (!ASIC_IS_AVIVO(rdev) && radeon_crtc->crtc_id)
+		return;
+
+	memset(&args, 0, sizeof(args));
+
+	args.ucScaler = radeon_crtc->crtc_id;
+
+	if (radeon_crtc->devices & (ATOM_DEVICE_TV_SUPPORT)) {
+		switch (tv_std) {
+		case TV_STD_NTSC:
+		default:
+			args.ucTVStandard = ATOM_TV_NTSC;
+			break;
+		case TV_STD_PAL:
+			args.ucTVStandard = ATOM_TV_PAL;
+			break;
+		case TV_STD_PAL_M:
+			args.ucTVStandard = ATOM_TV_PALM;
+			break;
+		case TV_STD_PAL_60:
+			args.ucTVStandard = ATOM_TV_PAL60;
+			break;
+		case TV_STD_NTSC_J:
+			args.ucTVStandard = ATOM_TV_NTSCJ;
+			break;
+		case TV_STD_SCART_PAL:
+			args.ucTVStandard = ATOM_TV_PAL; /* ??? */
+			break;
+		case TV_STD_SECAM:
+			args.ucTVStandard = ATOM_TV_SECAM;
+			break;
+		case TV_STD_PAL_CN:
+			args.ucTVStandard = ATOM_TV_PALCN;
+			break;
+		}
+		args.ucEnable = SCALER_ENABLE_MULTITAP_MODE;
+	} else if (radeon_crtc->devices & (ATOM_DEVICE_CV_SUPPORT)) {
+		args.ucTVStandard = ATOM_TV_CV;
+		args.ucEnable = SCALER_ENABLE_MULTITAP_MODE;
+	} else {
+		switch (radeon_crtc->rmx_type) {
+		case RMX_FULL:
+			args.ucEnable = ATOM_SCALER_EXPANSION;
+			break;
+		case RMX_CENTER:
+			args.ucEnable = ATOM_SCALER_CENTER;
+			break;
+		case RMX_ASPECT:
+			args.ucEnable = ATOM_SCALER_EXPANSION;
+			break;
+		default:
+			if (ASIC_IS_AVIVO(rdev))
+				args.ucEnable = ATOM_SCALER_DISABLE;
+			else
+				args.ucEnable = ATOM_SCALER_CENTER;
+			break;
+		}
+	}
+	atom_execute_table(rdev->mode_info.atom_context, index, (uint32_t *)&args);
+	if (radeon_crtc->devices & (ATOM_DEVICE_CV_SUPPORT | ATOM_DEVICE_TV_SUPPORT)
+	    && rdev->family >= CHIP_RV515 && rdev->family <= CHIP_RV570) {
+		atom_rv515_force_tv_scaler(rdev);
+	}
+}
+
 static void atombios_lock_crtc(struct drm_crtc *crtc, int lock)
 {
 	struct radeon_crtc *radeon_crtc = to_radeon_crtc(crtc);
@@ -522,6 +648,9 @@ int atombios_crtc_mode_set(struct drm_crtc *crtc,
 		radeon_crtc_set_base(crtc, x, y, old_fb);
 		radeon_legacy_atom_set_surface(crtc);
 	}
+	atombios_overscan_setup(crtc, mode, adjusted_mode);
+	atombios_scaler_setup(crtc);
+	radeon_bandwidth_update(rdev);
 	return 0;
 }
 
@@ -529,6 +658,8 @@ static bool atombios_crtc_mode_fixup(struct drm_crtc *crtc,
 				     struct drm_display_mode *mode,
 				     struct drm_display_mode *adjusted_mode)
 {
+	if (!radeon_crtc_scaling_mode_fixup(crtc, mode, adjusted_mode))
+		return false;
 	return true;
 }
 
@@ -561,148 +692,3 @@ void radeon_atombios_init_crtc(struct drm_device *dev,
 		    AVIVO_D2CRTC_H_TOTAL - AVIVO_D1CRTC_H_TOTAL;
 	drm_crtc_helper_add(&radeon_crtc->base, &atombios_helper_funcs);
 }
-
-void radeon_init_disp_bw_avivo(struct drm_device *dev,
-			       struct drm_display_mode *mode1,
-			       uint32_t pixel_bytes1,
-			       struct drm_display_mode *mode2,
-			       uint32_t pixel_bytes2)
-{
-	struct radeon_device *rdev = dev->dev_private;
-	fixed20_12 min_mem_eff;
-	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff;
-	fixed20_12 sclk_ff, mclk_ff;
-	uint32_t dc_lb_memory_split, temp;
-
-	min_mem_eff.full = rfixed_const_8(0);
-	if (rdev->disp_priority == 2) {
-		uint32_t mc_init_misc_lat_timer = 0;
-		if (rdev->family == CHIP_RV515)
-			mc_init_misc_lat_timer =
-			    RREG32_MC(RV515_MC_INIT_MISC_LAT_TIMER);
-		else if (rdev->family == CHIP_RS690)
-			mc_init_misc_lat_timer =
-			    RREG32_MC(RS690_MC_INIT_MISC_LAT_TIMER);
-
-		mc_init_misc_lat_timer &=
-		    ~(R300_MC_DISP1R_INIT_LAT_MASK <<
-		      R300_MC_DISP1R_INIT_LAT_SHIFT);
-		mc_init_misc_lat_timer &=
-		    ~(R300_MC_DISP0R_INIT_LAT_MASK <<
-		      R300_MC_DISP0R_INIT_LAT_SHIFT);
-
-		if (mode2)
-			mc_init_misc_lat_timer |=
-			    (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
-		if (mode1)
-			mc_init_misc_lat_timer |=
-			    (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
-
-		if (rdev->family == CHIP_RV515)
-			WREG32_MC(RV515_MC_INIT_MISC_LAT_TIMER,
-				  mc_init_misc_lat_timer);
-		else if (rdev->family == CHIP_RS690)
-			WREG32_MC(RS690_MC_INIT_MISC_LAT_TIMER,
-				  mc_init_misc_lat_timer);
-	}
-
-	/*
-	 * determine is there is enough bw for current mode
-	 */
-	temp_ff.full = rfixed_const(100);
-	mclk_ff.full = rfixed_const(rdev->clock.default_mclk);
-	mclk_ff.full = rfixed_div(mclk_ff, temp_ff);
-	sclk_ff.full = rfixed_const(rdev->clock.default_sclk);
-	sclk_ff.full = rfixed_div(sclk_ff, temp_ff);
-
-	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
-	temp_ff.full = rfixed_const(temp);
-	mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
-	mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
-
-	pix_clk.full = 0;
-	pix_clk2.full = 0;
-	peak_disp_bw.full = 0;
-	if (mode1) {
-		temp_ff.full = rfixed_const(1000);
-		pix_clk.full = rfixed_const(mode1->clock);	/* convert to fixed point */
-		pix_clk.full = rfixed_div(pix_clk, temp_ff);
-		temp_ff.full = rfixed_const(pixel_bytes1);
-		peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
-	}
-	if (mode2) {
-		temp_ff.full = rfixed_const(1000);
-		pix_clk2.full = rfixed_const(mode2->clock);	/* convert to fixed point */
-		pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
-		temp_ff.full = rfixed_const(pixel_bytes2);
-		peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
-	}
-
-	if (peak_disp_bw.full >= mem_bw.full) {
-		DRM_ERROR
-		    ("You may not have enough display bandwidth for current mode\n"
-		     "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
-		printk("peak disp bw %d, mem_bw %d\n",
-		       rfixed_trunc(peak_disp_bw), rfixed_trunc(mem_bw));
-	}
-
-	/*
-	 * Line Buffer Setup
-	 * There is a single line buffer shared by both display controllers.
-	 * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between the display
-	 * controllers.  The paritioning can either be done manually or via one of four
-	 * preset allocations specified in bits 1:0:
-	 * 0 - line buffer is divided in half and shared between each display controller
-	 * 1 - D1 gets 3/4 of the line buffer, D2 gets 1/4
-	 * 2 - D1 gets the whole buffer
-	 * 3 - D1 gets 1/4 of the line buffer, D2 gets 3/4
-	 * Setting bit 2 of DC_LB_MEMORY_SPLIT controls switches to manual allocation mode.
-	 * In manual allocation mode, D1 always starts at 0, D1 end/2 is specified in bits
-	 * 14:4; D2 allocation follows D1.
-	 */
-
-	/* is auto or manual better ? */
-	dc_lb_memory_split =
-	    RREG32(AVIVO_DC_LB_MEMORY_SPLIT) & ~AVIVO_DC_LB_MEMORY_SPLIT_MASK;
-	dc_lb_memory_split &= ~AVIVO_DC_LB_MEMORY_SPLIT_SHIFT_MODE;
-#if 1
-	/* auto */
-	if (mode1 && mode2) {
-		if (mode1->hdisplay > mode2->hdisplay) {
-			if (mode1->hdisplay > 2560)
-				dc_lb_memory_split |=
-				    AVIVO_DC_LB_MEMORY_SPLIT_D1_3Q_D2_1Q;
-			else
-				dc_lb_memory_split |=
-				    AVIVO_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
-		} else if (mode2->hdisplay > mode1->hdisplay) {
-			if (mode2->hdisplay > 2560)
-				dc_lb_memory_split |=
-				    AVIVO_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
-			else
-				dc_lb_memory_split |=
-				    AVIVO_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
-		} else
-			dc_lb_memory_split |=
-			    AVIVO_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
-	} else if (mode1) {
-		dc_lb_memory_split |= AVIVO_DC_LB_MEMORY_SPLIT_D1_ONLY;
-	} else if (mode2) {
-		dc_lb_memory_split |= AVIVO_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
-	}
-#else
-	/* manual */
-	dc_lb_memory_split |= AVIVO_DC_LB_MEMORY_SPLIT_SHIFT_MODE;
-	dc_lb_memory_split &=
-	    ~(AVIVO_DC_LB_DISP1_END_ADR_MASK <<
-	      AVIVO_DC_LB_DISP1_END_ADR_SHIFT);
-	if (mode1) {
-		dc_lb_memory_split |=
-		    ((((mode1->hdisplay / 2) + 64) & AVIVO_DC_LB_DISP1_END_ADR_MASK)
-		     << AVIVO_DC_LB_DISP1_END_ADR_SHIFT);
-	} else if (mode2) {
-		dc_lb_memory_split |= (0 << AVIVO_DC_LB_DISP1_END_ADR_SHIFT);
-	}
-#endif
-	WREG32(AVIVO_DC_LB_MEMORY_SPLIT, dc_lb_memory_split);
-}
diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c
index 69bd7cb5997..0e00fef0b84 100644
--- a/drivers/gpu/drm/radeon/r100.c
+++ b/drivers/gpu/drm/radeon/r100.c
@@ -1784,3 +1784,486 @@ void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
 	int surf_index = reg * 16;
 	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
 }
+
+void r100_bandwidth_update(struct radeon_device *rdev)
+{
+	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
+	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
+	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
+	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
+	fixed20_12 memtcas_ff[8] = {
+		fixed_init(1),
+		fixed_init(2),
+		fixed_init(3),
+		fixed_init(0),
+		fixed_init_half(1),
+		fixed_init_half(2),
+		fixed_init(0),
+	};
+	fixed20_12 memtcas_rs480_ff[8] = {
+		fixed_init(0),
+		fixed_init(1),
+		fixed_init(2),
+		fixed_init(3),
+		fixed_init(0),
+		fixed_init_half(1),
+		fixed_init_half(2),
+		fixed_init_half(3),
+	};
+	fixed20_12 memtcas2_ff[8] = {
+		fixed_init(0),
+		fixed_init(1),
+		fixed_init(2),
+		fixed_init(3),
+		fixed_init(4),
+		fixed_init(5),
+		fixed_init(6),
+		fixed_init(7),
+	};
+	fixed20_12 memtrbs[8] = {
+		fixed_init(1),
+		fixed_init_half(1),
+		fixed_init(2),
+		fixed_init_half(2),
+		fixed_init(3),
+		fixed_init_half(3),
+		fixed_init(4),
+		fixed_init_half(4)
+	};
+	fixed20_12 memtrbs_r4xx[8] = {
+		fixed_init(4),
+		fixed_init(5),
+		fixed_init(6),
+		fixed_init(7),
+		fixed_init(8),
+		fixed_init(9),
+		fixed_init(10),
+		fixed_init(11)
+	};
+	fixed20_12 min_mem_eff;
+	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
+	fixed20_12 cur_latency_mclk, cur_latency_sclk;
+	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
+		disp_drain_rate2, read_return_rate;
+	fixed20_12 time_disp1_drop_priority;
+	int c;
+	int cur_size = 16;       /* in octawords */
+	int critical_point = 0, critical_point2;
+/* 	uint32_t read_return_rate, time_disp1_drop_priority; */
+	int stop_req, max_stop_req;
+	struct drm_display_mode *mode1 = NULL;
+	struct drm_display_mode *mode2 = NULL;
+	uint32_t pixel_bytes1 = 0;
+	uint32_t pixel_bytes2 = 0;
+
+	if (rdev->mode_info.crtcs[0]->base.enabled) {
+		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
+		pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
+	}
+	if (rdev->mode_info.crtcs[1]->base.enabled) {
+		mode2 = &rdev->mode_info.crtcs[1]->base.mode;
+		pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
+	}
+
+	min_mem_eff.full = rfixed_const_8(0);
+	/* get modes */
+	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
+		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
+		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
+		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
+		/* check crtc enables */
+		if (mode2)
+			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
+		if (mode1)
+			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
+		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
+	}
+
+	/*
+	 * determine is there is enough bw for current mode
+	 */
+	mclk_ff.full = rfixed_const(rdev->clock.default_mclk);
+	temp_ff.full = rfixed_const(100);
+	mclk_ff.full = rfixed_div(mclk_ff, temp_ff);
+	sclk_ff.full = rfixed_const(rdev->clock.default_sclk);
+	sclk_ff.full = rfixed_div(sclk_ff, temp_ff);
+
+	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
+	temp_ff.full = rfixed_const(temp);
+	mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
+
+	pix_clk.full = 0;
+	pix_clk2.full = 0;
+	peak_disp_bw.full = 0;
+	if (mode1) {
+		temp_ff.full = rfixed_const(1000);
+		pix_clk.full = rfixed_const(mode1->clock); /* convert to fixed point */
+		pix_clk.full = rfixed_div(pix_clk, temp_ff);
+		temp_ff.full = rfixed_const(pixel_bytes1);
+		peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
+	}
+	if (mode2) {
+		temp_ff.full = rfixed_const(1000);
+		pix_clk2.full = rfixed_const(mode2->clock); /* convert to fixed point */
+		pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
+		temp_ff.full = rfixed_const(pixel_bytes2);
+		peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
+	}
+
+	mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
+	if (peak_disp_bw.full >= mem_bw.full) {
+		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
+			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
+	}
+
+	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
+	temp = RREG32(RADEON_MEM_TIMING_CNTL);
+	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
+		mem_trcd = ((temp >> 2) & 0x3) + 1;
+		mem_trp  = ((temp & 0x3)) + 1;
+		mem_tras = ((temp & 0x70) >> 4) + 1;
+	} else if (rdev->family == CHIP_R300 ||
+		   rdev->family == CHIP_R350) { /* r300, r350 */
+		mem_trcd = (temp & 0x7) + 1;
+		mem_trp = ((temp >> 8) & 0x7) + 1;
+		mem_tras = ((temp >> 11) & 0xf) + 4;
+	} else if (rdev->family == CHIP_RV350 ||
+		   rdev->family <= CHIP_RV380) {
+		/* rv3x0 */
+		mem_trcd = (temp & 0x7) + 3;
+		mem_trp = ((temp >> 8) & 0x7) + 3;
+		mem_tras = ((temp >> 11) & 0xf) + 6;
+	} else if (rdev->family == CHIP_R420 ||
+		   rdev->family == CHIP_R423 ||
+		   rdev->family == CHIP_RV410) {
+		/* r4xx */
+		mem_trcd = (temp & 0xf) + 3;
+		if (mem_trcd > 15)
+			mem_trcd = 15;
+		mem_trp = ((temp >> 8) & 0xf) + 3;
+		if (mem_trp > 15)
+			mem_trp = 15;
+		mem_tras = ((temp >> 12) & 0x1f) + 6;
+		if (mem_tras > 31)
+			mem_tras = 31;
+	} else { /* RV200, R200 */
+		mem_trcd = (temp & 0x7) + 1;
+		mem_trp = ((temp >> 8) & 0x7) + 1;
+		mem_tras = ((temp >> 12) & 0xf) + 4;
+	}
+	/* convert to FF */
+	trcd_ff.full = rfixed_const(mem_trcd);
+	trp_ff.full = rfixed_const(mem_trp);
+	tras_ff.full = rfixed_const(mem_tras);
+
+	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
+	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
+	data = (temp & (7 << 20)) >> 20;
+	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
+		if (rdev->family == CHIP_RS480) /* don't think rs400 */
+			tcas_ff = memtcas_rs480_ff[data];
+		else
+			tcas_ff = memtcas_ff[data];
+	} else
+		tcas_ff = memtcas2_ff[data];
+
+	if (rdev->family == CHIP_RS400 ||
+	    rdev->family == CHIP_RS480) {
+		/* extra cas latency stored in bits 23-25 0-4 clocks */
+		data = (temp >> 23) & 0x7;
+		if (data < 5)
+			tcas_ff.full += rfixed_const(data);
+	}
+
+	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
+		/* on the R300, Tcas is included in Trbs.
+		 */
+		temp = RREG32(RADEON_MEM_CNTL);
+		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
+		if (data == 1) {
+			if (R300_MEM_USE_CD_CH_ONLY & temp) {
+				temp = RREG32(R300_MC_IND_INDEX);
+				temp &= ~R300_MC_IND_ADDR_MASK;
+				temp |= R300_MC_READ_CNTL_CD_mcind;
+				WREG32(R300_MC_IND_INDEX, temp);
+				temp = RREG32(R300_MC_IND_DATA);
+				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
+			} else {
+				temp = RREG32(R300_MC_READ_CNTL_AB);
+				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
+			}
+		} else {
+			temp = RREG32(R300_MC_READ_CNTL_AB);
+			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
+		}
+		if (rdev->family == CHIP_RV410 ||
+		    rdev->family == CHIP_R420 ||
+		    rdev->family == CHIP_R423)
+			trbs_ff = memtrbs_r4xx[data];
+		else
+			trbs_ff = memtrbs[data];
+		tcas_ff.full += trbs_ff.full;
+	}
+
+	sclk_eff_ff.full = sclk_ff.full;
+
+	if (rdev->flags & RADEON_IS_AGP) {
+		fixed20_12 agpmode_ff;
+		agpmode_ff.full = rfixed_const(radeon_agpmode);
+		temp_ff.full = rfixed_const_666(16);
+		sclk_eff_ff.full -= rfixed_mul(agpmode_ff, temp_ff);
+	}
+	/* TODO PCIE lanes may affect this - agpmode == 16?? */
+
+	if (ASIC_IS_R300(rdev)) {
+		sclk_delay_ff.full = rfixed_const(250);
+	} else {
+		if ((rdev->family == CHIP_RV100) ||
+		    rdev->flags & RADEON_IS_IGP) {
+			if (rdev->mc.vram_is_ddr)
+				sclk_delay_ff.full = rfixed_const(41);
+			else
+				sclk_delay_ff.full = rfixed_const(33);
+		} else {
+			if (rdev->mc.vram_width == 128)
+				sclk_delay_ff.full = rfixed_const(57);
+			else
+				sclk_delay_ff.full = rfixed_const(41);
+		}
+	}
+
+	mc_latency_sclk.full = rfixed_div(sclk_delay_ff, sclk_eff_ff);
+
+	if (rdev->mc.vram_is_ddr) {
+		if (rdev->mc.vram_width == 32) {
+			k1.full = rfixed_const(40);
+			c  = 3;
+		} else {
+			k1.full = rfixed_const(20);
+			c  = 1;
+		}
+	} else {
+		k1.full = rfixed_const(40);
+		c  = 3;
+	}
+
+	temp_ff.full = rfixed_const(2);
+	mc_latency_mclk.full = rfixed_mul(trcd_ff, temp_ff);
+	temp_ff.full = rfixed_const(c);
+	mc_latency_mclk.full += rfixed_mul(tcas_ff, temp_ff);
+	temp_ff.full = rfixed_const(4);
+	mc_latency_mclk.full += rfixed_mul(tras_ff, temp_ff);
+	mc_latency_mclk.full += rfixed_mul(trp_ff, temp_ff);
+	mc_latency_mclk.full += k1.full;
+
+	mc_latency_mclk.full = rfixed_div(mc_latency_mclk, mclk_ff);
+	mc_latency_mclk.full += rfixed_div(temp_ff, sclk_eff_ff);
+
+	/*
+	  HW cursor time assuming worst case of full size colour cursor.
+	*/
+	temp_ff.full = rfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
+	temp_ff.full += trcd_ff.full;
+	if (temp_ff.full < tras_ff.full)
+		temp_ff.full = tras_ff.full;
+	cur_latency_mclk.full = rfixed_div(temp_ff, mclk_ff);
+
+	temp_ff.full = rfixed_const(cur_size);
+	cur_latency_sclk.full = rfixed_div(temp_ff, sclk_eff_ff);
+	/*
+	  Find the total latency for the display data.
+	*/
+	disp_latency_overhead.full = rfixed_const(80);
+	disp_latency_overhead.full = rfixed_div(disp_latency_overhead, sclk_ff);
+	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
+	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
+
+	if (mc_latency_mclk.full > mc_latency_sclk.full)
+		disp_latency.full = mc_latency_mclk.full;
+	else
+		disp_latency.full = mc_latency_sclk.full;
+
+	/* setup Max GRPH_STOP_REQ default value */
+	if (ASIC_IS_RV100(rdev))
+		max_stop_req = 0x5c;
+	else
+		max_stop_req = 0x7c;
+
+	if (mode1) {
+		/*  CRTC1
+		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
+		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
+		*/
+		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
+
+		if (stop_req > max_stop_req)
+			stop_req = max_stop_req;
+
+		/*
+		  Find the drain rate of the display buffer.
+		*/
+		temp_ff.full = rfixed_const((16/pixel_bytes1));
+		disp_drain_rate.full = rfixed_div(pix_clk, temp_ff);
+
+		/*
+		  Find the critical point of the display buffer.
+		*/
+		crit_point_ff.full = rfixed_mul(disp_drain_rate, disp_latency);
+		crit_point_ff.full += rfixed_const_half(0);
+
+		critical_point = rfixed_trunc(crit_point_ff);
+
+		if (rdev->disp_priority == 2) {
+			critical_point = 0;
+		}
+
+		/*
+		  The critical point should never be above max_stop_req-4.  Setting
+		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
+		*/
+		if (max_stop_req - critical_point < 4)
+			critical_point = 0;
+
+		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
+			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
+			critical_point = 0x10;
+		}
+
+		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
+		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
+		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
+		temp &= ~(RADEON_GRPH_START_REQ_MASK);
+		if ((rdev->family == CHIP_R350) &&
+		    (stop_req > 0x15)) {
+			stop_req -= 0x10;
+		}
+		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
+		temp |= RADEON_GRPH_BUFFER_SIZE;
+		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
+			  RADEON_GRPH_CRITICAL_AT_SOF |
+			  RADEON_GRPH_STOP_CNTL);
+		/*
+		  Write the result into the register.
+		*/
+		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
+						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
+
+#if 0
+		if ((rdev->family == CHIP_RS400) ||
+		    (rdev->family == CHIP_RS480)) {
+			/* attempt to program RS400 disp regs correctly ??? */
+			temp = RREG32(RS400_DISP1_REG_CNTL);
+			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
+				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
+			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
+						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
+						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
+			temp = RREG32(RS400_DMIF_MEM_CNTL1);
+			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
+				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
+			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
+						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
+						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
+		}
+#endif
+
+		DRM_DEBUG("GRPH_BUFFER_CNTL from to %x\n",
+			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
+			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
+	}
+
+	if (mode2) {
+		u32 grph2_cntl;
+		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
+
+		if (stop_req > max_stop_req)
+			stop_req = max_stop_req;
+
+		/*
+		  Find the drain rate of the display buffer.
+		*/
+		temp_ff.full = rfixed_const((16/pixel_bytes2));
+		disp_drain_rate2.full = rfixed_div(pix_clk2, temp_ff);
+
+		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
+		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
+		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
+		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
+		if ((rdev->family == CHIP_R350) &&
+		    (stop_req > 0x15)) {
+			stop_req -= 0x10;
+		}
+		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
+		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
+		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
+			  RADEON_GRPH_CRITICAL_AT_SOF |
+			  RADEON_GRPH_STOP_CNTL);
+
+		if ((rdev->family == CHIP_RS100) ||
+		    (rdev->family == CHIP_RS200))
+			critical_point2 = 0;
+		else {
+			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
+			temp_ff.full = rfixed_const(temp);
+			temp_ff.full = rfixed_mul(mclk_ff, temp_ff);
+			if (sclk_ff.full < temp_ff.full)
+				temp_ff.full = sclk_ff.full;
+
+			read_return_rate.full = temp_ff.full;
+
+			if (mode1) {
+				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
+				time_disp1_drop_priority.full = rfixed_div(crit_point_ff, temp_ff);
+			} else {
+				time_disp1_drop_priority.full = 0;
+			}
+			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
+			crit_point_ff.full = rfixed_mul(crit_point_ff, disp_drain_rate2);
+			crit_point_ff.full += rfixed_const_half(0);
+
+			critical_point2 = rfixed_trunc(crit_point_ff);
+
+			if (rdev->disp_priority == 2) {
+				critical_point2 = 0;
+			}
+
+			if (max_stop_req - critical_point2 < 4)
+				critical_point2 = 0;
+
+		}
+
+		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
+			/* some R300 cards have problem with this set to 0 */
+			critical_point2 = 0x10;
+		}
+
+		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
+						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
+
+		if ((rdev->family == CHIP_RS400) ||
+		    (rdev->family == CHIP_RS480)) {
+#if 0
+			/* attempt to program RS400 disp2 regs correctly ??? */
+			temp = RREG32(RS400_DISP2_REQ_CNTL1);
+			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
+				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
+			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
+						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
+						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
+			temp = RREG32(RS400_DISP2_REQ_CNTL2);
+			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
+				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
+			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
+						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
+						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
+#endif
+			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
+			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
+			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
+			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
+		}
+
+		DRM_DEBUG("GRPH2_BUFFER_CNTL from to %x\n",
+			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
+	}
+}
diff --git a/drivers/gpu/drm/radeon/r300.c b/drivers/gpu/drm/radeon/r300.c
index 28e5777658b..9c8d41534a5 100644
--- a/drivers/gpu/drm/radeon/r300.c
+++ b/drivers/gpu/drm/radeon/r300.c
@@ -31,6 +31,7 @@
 #include "radeon_reg.h"
 #include "radeon.h"
 #include "radeon_drm.h"
+#include "radeon_share.h"
 
 /* r300,r350,rv350,rv370,rv380 depends on : */
 void r100_hdp_reset(struct radeon_device *rdev);
diff --git a/drivers/gpu/drm/radeon/r520.c b/drivers/gpu/drm/radeon/r520.c
index b6bd3758db6..0a981e2ee2f 100644
--- a/drivers/gpu/drm/radeon/r520.c
+++ b/drivers/gpu/drm/radeon/r520.c
@@ -28,6 +28,7 @@
 #include "drmP.h"
 #include "radeon_reg.h"
 #include "radeon.h"
+#include "radeon_share.h"
 
 /* r520,rv530,rv560,rv570,r580 depends on : */
 void r100_hdp_reset(struct radeon_device *rdev);
@@ -226,7 +227,20 @@ static void r520_vram_get_type(struct radeon_device *rdev)
 
 void r520_vram_info(struct radeon_device *rdev)
 {
+	fixed20_12 a;
+
 	r520_vram_get_type(rdev);
 
 	r100_vram_init_sizes(rdev);
+	/* FIXME: we should enforce default clock in case GPU is not in
+	 * default setup
+	 */
+	a.full = rfixed_const(100);
+	rdev->pm.sclk.full = rfixed_const(rdev->clock.default_sclk);
+	rdev->pm.sclk.full = rfixed_div(rdev->pm.sclk, a);
+}
+
+void r520_bandwidth_update(struct radeon_device *rdev)
+{
+	rv515_bandwidth_avivo_update(rdev);
 }
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index af12a2fe322..63a3fe32e58 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -113,6 +113,7 @@ enum radeon_family {
 	CHIP_RV770,
 	CHIP_RV730,
 	CHIP_RV710,
+	CHIP_RS880,
 	CHIP_LAST,
 };
 
@@ -490,6 +491,39 @@ struct radeon_wb {
 	uint64_t		gpu_addr;
 };
 
+/**
+ * struct radeon_pm - power management datas
+ * @max_bandwidth:      maximum bandwidth the gpu has (MByte/s)
+ * @igp_sideport_mclk:  sideport memory clock Mhz (rs690,rs740,rs780,rs880)
+ * @igp_system_mclk:    system clock Mhz (rs690,rs740,rs780,rs880)
+ * @igp_ht_link_clk:    ht link clock Mhz (rs690,rs740,rs780,rs880)
+ * @igp_ht_link_width:  ht link width in bits (rs690,rs740,rs780,rs880)
+ * @k8_bandwidth:       k8 bandwidth the gpu has (MByte/s) (IGP)
+ * @sideport_bandwidth: sideport bandwidth the gpu has (MByte/s) (IGP)
+ * @ht_bandwidth:       ht bandwidth the gpu has (MByte/s) (IGP)
+ * @core_bandwidth:     core GPU bandwidth the gpu has (MByte/s) (IGP)
+ * @sclk:          	GPU clock Mhz (core bandwith depends of this clock)
+ * @needed_bandwidth:   current bandwidth needs
+ *
+ * It keeps track of various data needed to take powermanagement decision.
+ * Bandwith need is used to determine minimun clock of the GPU and memory.
+ * Equation between gpu/memory clock and available bandwidth is hw dependent
+ * (type of memory, bus size, efficiency, ...)
+ */
+struct radeon_pm {
+	fixed20_12		max_bandwidth;
+	fixed20_12		igp_sideport_mclk;
+	fixed20_12		igp_system_mclk;
+	fixed20_12		igp_ht_link_clk;
+	fixed20_12		igp_ht_link_width;
+	fixed20_12		k8_bandwidth;
+	fixed20_12		sideport_bandwidth;
+	fixed20_12		ht_bandwidth;
+	fixed20_12		core_bandwidth;
+	fixed20_12		sclk;
+	fixed20_12		neede