1 files changed, 524 insertions, 40 deletions

diff --git a/drivers/gpu/drm/radeon/evergreen.c b/drivers/gpu/drm/radeon/evergreen.c
index 2f93d46ae69..f12a5b3ec05 100644
--- a/drivers/gpu/drm/radeon/evergreen.c
+++ b/drivers/gpu/drm/radeon/evergreen.c
@@ -32,6 +32,7 @@
 #include "atom.h"
 #include "avivod.h"
 #include "evergreen_reg.h"
+#include "evergreen_blit_shaders.h"
 
 #define EVERGREEN_PFP_UCODE_SIZE 1120
 #define EVERGREEN_PM4_UCODE_SIZE 1376
@@ -284,9 +285,444 @@ void evergreen_hpd_fini(struct radeon_device *rdev)
 	}
 }
 
+/* watermark setup */
+
+static u32 evergreen_line_buffer_adjust(struct radeon_device *rdev,
+					struct radeon_crtc *radeon_crtc,
+					struct drm_display_mode *mode,
+					struct drm_display_mode *other_mode)
+{
+	u32 tmp = 0;
+	/*
+	 * Line Buffer Setup
+	 * There are 3 line buffers, each one shared by 2 display controllers.
+	 * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
+	 * the display controllers.  The paritioning is done via one of four
+	 * preset allocations specified in bits 2:0:
+	 * first display controller
+	 *  0 - first half of lb (3840 * 2)
+	 *  1 - first 3/4 of lb (5760 * 2)
+	 *  2 - whole lb (7680 * 2)
+	 *  3 - first 1/4 of lb (1920 * 2)
+	 * second display controller
+	 *  4 - second half of lb (3840 * 2)
+	 *  5 - second 3/4 of lb (5760 * 2)
+	 *  6 - whole lb (7680 * 2)
+	 *  7 - last 1/4 of lb (1920 * 2)
+	 */
+	if (mode && other_mode) {
+		if (mode->hdisplay > other_mode->hdisplay) {
+			if (mode->hdisplay > 2560)
+				tmp = 1; /* 3/4 */
+			else
+				tmp = 0; /* 1/2 */
+		} else if (other_mode->hdisplay > mode->hdisplay) {
+			if (other_mode->hdisplay > 2560)
+				tmp = 3; /* 1/4 */
+			else
+				tmp = 0; /* 1/2 */
+		} else
+			tmp = 0; /* 1/2 */
+	} else if (mode)
+		tmp = 2; /* whole */
+	else if (other_mode)
+		tmp = 3; /* 1/4 */
+
+	/* second controller of the pair uses second half of the lb */
+	if (radeon_crtc->crtc_id % 2)
+		tmp += 4;
+	WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset, tmp);
+
+	switch (tmp) {
+	case 0:
+	case 4:
+	default:
+		return 3840 * 2;
+	case 1:
+	case 5:
+		return 5760 * 2;
+	case 2:
+	case 6:
+		return 7680 * 2;
+	case 3:
+	case 7:
+		return 1920 * 2;
+	}
+}
+
+static u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev)
+{
+	u32 tmp = RREG32(MC_SHARED_CHMAP);
+
+	switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
+	case 0:
+	default:
+		return 1;
+	case 1:
+		return 2;
+	case 2:
+		return 4;
+	case 3:
+		return 8;
+	}
+}
+
+struct evergreen_wm_params {
+	u32 dram_channels; /* number of dram channels */
+	u32 yclk;          /* bandwidth per dram data pin in kHz */
+	u32 sclk;          /* engine clock in kHz */
+	u32 disp_clk;      /* display clock in kHz */
+	u32 src_width;     /* viewport width */
+	u32 active_time;   /* active display time in ns */
+	u32 blank_time;    /* blank time in ns */
+	bool interlaced;    /* mode is interlaced */
+	fixed20_12 vsc;    /* vertical scale ratio */
+	u32 num_heads;     /* number of active crtcs */
+	u32 bytes_per_pixel; /* bytes per pixel display + overlay */
+	u32 lb_size;       /* line buffer allocated to pipe */
+	u32 vtaps;         /* vertical scaler taps */
+};
+
+static u32 evergreen_dram_bandwidth(struct evergreen_wm_params *wm)
+{
+	/* Calculate DRAM Bandwidth and the part allocated to display. */
+	fixed20_12 dram_efficiency; /* 0.7 */
+	fixed20_12 yclk, dram_channels, bandwidth;
+	fixed20_12 a;
+
+	a.full = dfixed_const(1000);
+	yclk.full = dfixed_const(wm->yclk);
+	yclk.full = dfixed_div(yclk, a);
+	dram_channels.full = dfixed_const(wm->dram_channels * 4);
+	a.full = dfixed_const(10);
+	dram_efficiency.full = dfixed_const(7);
+	dram_efficiency.full = dfixed_div(dram_efficiency, a);
+	bandwidth.full = dfixed_mul(dram_channels, yclk);
+	bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
+
+	return dfixed_trunc(bandwidth);
+}
+
+static u32 evergreen_dram_bandwidth_for_display(struct evergreen_wm_params *wm)
+{
+	/* Calculate DRAM Bandwidth and the part allocated to display. */
+	fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
+	fixed20_12 yclk, dram_channels, bandwidth;
+	fixed20_12 a;
+
+	a.full = dfixed_const(1000);
+	yclk.full = dfixed_const(wm->yclk);
+	yclk.full = dfixed_div(yclk, a);
+	dram_channels.full = dfixed_const(wm->dram_channels * 4);
+	a.full = dfixed_const(10);
+	disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
+	disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
+	bandwidth.full = dfixed_mul(dram_channels, yclk);
+	bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
+
+	return dfixed_trunc(bandwidth);
+}
+
+static u32 evergreen_data_return_bandwidth(struct evergreen_wm_params *wm)
+{
+	/* Calculate the display Data return Bandwidth */
+	fixed20_12 return_efficiency; /* 0.8 */
+	fixed20_12 sclk, bandwidth;
+	fixed20_12 a;
+
+	a.full = dfixed_const(1000);
+	sclk.full = dfixed_const(wm->sclk);
+	sclk.full = dfixed_div(sclk, a);
+	a.full = dfixed_const(10);
+	return_efficiency.full = dfixed_const(8);
+	return_efficiency.full = dfixed_div(return_efficiency, a);
+	a.full = dfixed_const(32);
+	bandwidth.full = dfixed_mul(a, sclk);
+	bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
+
+	return dfixed_trunc(bandwidth);
+}
+
+static u32 evergreen_dmif_request_bandwidth(struct evergreen_wm_params *wm)
+{
+	/* Calculate the DMIF Request Bandwidth */
+	fixed20_12 disp_clk_request_efficiency; /* 0.8 */
+	fixed20_12 disp_clk, bandwidth;
+	fixed20_12 a;
+
+	a.full = dfixed_const(1000);
+	disp_clk.full = dfixed_const(wm->disp_clk);
+	disp_clk.full = dfixed_div(disp_clk, a);
+	a.full = dfixed_const(10);
+	disp_clk_request_efficiency.full = dfixed_const(8);
+	disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
+	a.full = dfixed_const(32);
+	bandwidth.full = dfixed_mul(a, disp_clk);
+	bandwidth.full = dfixed_mul(bandwidth, disp_clk_request_efficiency);
+
+	return dfixed_trunc(bandwidth);
+}
+
+static u32 evergreen_available_bandwidth(struct evergreen_wm_params *wm)
+{
+	/* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
+	u32 dram_bandwidth = evergreen_dram_bandwidth(wm);
+	u32 data_return_bandwidth = evergreen_data_return_bandwidth(wm);
+	u32 dmif_req_bandwidth = evergreen_dmif_request_bandwidth(wm);
+
+	return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
+}
+
+static u32 evergreen_average_bandwidth(struct evergreen_wm_params *wm)
+{
+	/* Calculate the display mode Average Bandwidth
+	 * DisplayMode should contain the source and destination dimensions,
+	 * timing, etc.
+	 */
+	fixed20_12 bpp;
+	fixed20_12 line_time;
+	fixed20_12 src_width;
+	fixed20_12 bandwidth;
+	fixed20_12 a;
+
+	a.full = dfixed_const(1000);
+	line_time.full = dfixed_const(wm->active_time + wm->blank_time);
+	line_time.full = dfixed_div(line_time, a);
+	bpp.full = dfixed_const(wm->bytes_per_pixel);
+	src_width.full = dfixed_const(wm->src_width);
+	bandwidth.full = dfixed_mul(src_width, bpp);
+	bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
+	bandwidth.full = dfixed_div(bandwidth, line_time);
+
+	return dfixed_trunc(bandwidth);
+}
+
+static u32 evergreen_latency_watermark(struct evergreen_wm_params *wm)
+{
+	/* First calcualte the latency in ns */
+	u32 mc_latency = 2000; /* 2000 ns. */
+	u32 available_bandwidth = evergreen_available_bandwidth(wm);
+	u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
+	u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
+	u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
+	u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
+		(wm->num_heads * cursor_line_pair_return_time);
+	u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
+	u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
+	fixed20_12 a, b, c;
+
+	if (wm->num_heads == 0)
+		return 0;
+
+	a.full = dfixed_const(2);
+	b.full = dfixed_const(1);
+	if ((wm->vsc.full > a.full) ||
+	    ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
+	    (wm->vtaps >= 5) ||
+	    ((wm->vsc.full >= a.full) && wm->interlaced))
+		max_src_lines_per_dst_line = 4;
+	else
+		max_src_lines_per_dst_line = 2;
+
+	a.full = dfixed_const(available_bandwidth);
+	b.full = dfixed_const(wm->num_heads);
+	a.full = dfixed_div(a, b);
+
+	b.full = dfixed_const(1000);
+	c.full = dfixed_const(wm->disp_clk);
+	b.full = dfixed_div(c, b);
+	c.full = dfixed_const(wm->bytes_per_pixel);
+	b.full = dfixed_mul(b, c);
+
+	lb_fill_bw = min(dfixed_trunc(a), dfixed_trunc(b));
+
+	a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
+	b.full = dfixed_const(1000);
+	c.full = dfixed_const(lb_fill_bw);
+	b.full = dfixed_div(c, b);
+	a.full = dfixed_div(a, b);
+	line_fill_time = dfixed_trunc(a);
+
+	if (line_fill_time < wm->active_time)
+		return latency;
+	else
+		return latency + (line_fill_time - wm->active_time);
+
+}
+
+static bool evergreen_average_bandwidth_vs_dram_bandwidth_for_display(struct evergreen_wm_params *wm)
+{
+	if (evergreen_average_bandwidth(wm) <=
+	    (evergreen_dram_bandwidth_for_display(wm) / wm->num_heads))
+		return true;
+	else
+		return false;
+};
+
+static bool evergreen_average_bandwidth_vs_available_bandwidth(struct evergreen_wm_params *wm)
+{
+	if (evergreen_average_bandwidth(wm) <=
+	    (evergreen_available_bandwidth(wm) / wm->num_heads))
+		return true;
+	else
+		return false;
+};
+
+static bool evergreen_check_latency_hiding(struct evergreen_wm_params *wm)
+{
+	u32 lb_partitions = wm->lb_size / wm->src_width;
+	u32 line_time = wm->active_time + wm->blank_time;
+	u32 latency_tolerant_lines;
+	u32 latency_hiding;
+	fixed20_12 a;
+
+	a.full = dfixed_const(1);
+	if (wm->vsc.full > a.full)
+		latency_tolerant_lines = 1;
+	else {
+		if (lb_partitions <= (wm->vtaps + 1))
+			latency_tolerant_lines = 1;
+		else
+			latency_tolerant_lines = 2;
+	}
+
+	latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
+
+	if (evergreen_latency_watermark(wm) <= latency_hiding)
+		return true;
+	else
+		return false;
+}
+
+static void evergreen_program_watermarks(struct radeon_device *rdev,
+					 struct radeon_crtc *radeon_crtc,
+					 u32 lb_size, u32 num_heads)
+{
+	struct drm_display_mode *mode = &radeon_crtc->base.mode;
+	struct evergreen_wm_params wm;
+	u32 pixel_period;
+	u32 line_time = 0;
+	u32 latency_watermark_a = 0, latency_watermark_b = 0;
+	u32 priority_a_mark = 0, priority_b_mark = 0;
+	u32 priority_a_cnt = PRIORITY_OFF;
+	u32 priority_b_cnt = PRIORITY_OFF;
+	u32 pipe_offset = radeon_crtc->crtc_id * 16;
+	u32 tmp, arb_control3;
+	fixed20_12 a, b, c;
+
+	if (radeon_crtc->base.enabled && num_heads && mode) {
+		pixel_period = 1000000 / (u32)mode->clock;
+		line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
+		priority_a_cnt = 0;
+		priority_b_cnt = 0;
+
+		wm.yclk = rdev->pm.current_mclk * 10;
+		wm.sclk = rdev->pm.current_sclk * 10;
+		wm.disp_clk = mode->clock;
+		wm.src_width = mode->crtc_hdisplay;
+		wm.active_time = mode->crtc_hdisplay * pixel_period;
+		wm.blank_time = line_time - wm.active_time;
+		wm.interlaced = false;
+		if (mode->flags & DRM_MODE_FLAG_INTERLACE)
+			wm.interlaced = true;
+		wm.vsc = radeon_crtc->vsc;
+		wm.vtaps = 1;
+		if (radeon_crtc->rmx_type != RMX_OFF)
+			wm.vtaps = 2;
+		wm.bytes_per_pixel = 4; /* XXX: get this from fb config */
+		wm.lb_size = lb_size;
+		wm.dram_channels = evergreen_get_number_of_dram_channels(rdev);
+		wm.num_heads = num_heads;
+
+		/* set for high clocks */
+		latency_watermark_a = min(evergreen_latency_watermark(&wm), (u32)65535);
+		/* set for low clocks */
+		/* wm.yclk = low clk; wm.sclk = low clk */
+		latency_watermark_b = min(evergreen_latency_watermark(&wm), (u32)65535);
+
+		/* possibly force display priority to high */
+		/* should really do this at mode validation time... */
+		if (!evergreen_average_bandwidth_vs_dram_bandwidth_for_display(&wm) ||
+		    !evergreen_average_bandwidth_vs_available_bandwidth(&wm) ||
+		    !evergreen_check_latency_hiding(&wm) ||
+		    (rdev->disp_priority == 2)) {
+			DRM_INFO("force priority to high\n");
+			priority_a_cnt |= PRIORITY_ALWAYS_ON;
+			priority_b_cnt |= PRIORITY_ALWAYS_ON;
+		}
+
+		a.full = dfixed_const(1000);
+		b.full = dfixed_const(mode->clock);
+		b.full = dfixed_div(b, a);
+		c.full = dfixed_const(latency_watermark_a);
+		c.full = dfixed_mul(c, b);
+		c.full = dfixed_mul(c, radeon_crtc->hsc);
+		c.full = dfixed_div(c, a);
+		a.full = dfixed_const(16);
+		c.full = dfixed_div(c, a);
+		priority_a_mark = dfixed_trunc(c);
+		priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK;
+
+		a.full = dfixed_const(1000);
+		b.full = dfixed_const(mode->clock);
+		b.full = dfixed_div(b, a);
+		c.full = dfixed_const(latency_watermark_b);
+		c.full = dfixed_mul(c, b);
+		c.full = dfixed_mul(c, radeon_crtc->hsc);
+		c.full = dfixed_div(c, a);
+		a.full = dfixed_const(16);
+		c.full = dfixed_div(c, a);
+		priority_b_mark = dfixed_trunc(c);
+		priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK;
+	}
+
+	/* select wm A */
+	arb_control3 = RREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset);
+	tmp = arb_control3;
+	tmp &= ~LATENCY_WATERMARK_MASK(3);
+	tmp |= LATENCY_WATERMARK_MASK(1);
+	WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, tmp);
+	WREG32(PIPE0_LATENCY_CONTROL + pipe_offset,
+	       (LATENCY_LOW_WATERMARK(latency_watermark_a) |
+		LATENCY_HIGH_WATERMARK(line_time)));
+	/* select wm B */
+	tmp = RREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset);
+	tmp &= ~LATENCY_WATERMARK_MASK(3);
+	tmp |= LATENCY_WATERMARK_MASK(2);
+	WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, tmp);
+	WREG32(PIPE0_LATENCY_CONTROL + pipe_offset,
+	       (LATENCY_LOW_WATERMARK(latency_watermark_b) |
+		LATENCY_HIGH_WATERMARK(line_time)));
+	/* restore original selection */
+	WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, arb_control3);
+
+	/* write the priority marks */
+	WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt);
+	WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt);
+
+}
+
 void evergreen_bandwidth_update(struct radeon_device *rdev)
 {
-	/* XXX */
+	struct drm_display_mode *mode0 = NULL;
+	struct drm_display_mode *mode1 = NULL;
+	u32 num_heads = 0, lb_size;
+	int i;
+
+	radeon_update_display_priority(rdev);
+
+	for (i = 0; i < rdev->num_crtc; i++) {
+		if (rdev->mode_info.crtcs[i]->base.enabled)
+			num_heads++;
+	}
+	for (i = 0; i < rdev->num_crtc; i += 2) {
+		mode0 = &rdev->mode_info.crtcs[i]->base.mode;
+		mode1 = &rdev->mode_info.crtcs[i+1]->base.mode;
+		lb_size = evergreen_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1);
+		evergreen_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
+		lb_size = evergreen_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0);
+		evergreen_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads);
+	}
 }
 
 static int evergreen_mc_wait_for_idle(struct radeon_device *rdev)
@@ -677,7 +1113,7 @@ static int evergreen_cp_load_microcode(struct radeon_device *rdev)
 
 static int evergreen_cp_start(struct radeon_device *rdev)
 {
-	int r;
+	int r, i;
 	uint32_t cp_me;
 
 	r = radeon_ring_lock(rdev, 7);
@@ -697,16 +1133,39 @@ static int evergreen_cp_start(struct radeon_device *rdev)
 	cp_me = 0xff;
 	WREG32(CP_ME_CNTL, cp_me);
 
-	r = radeon_ring_lock(rdev, 4);
+	r = radeon_ring_lock(rdev, evergreen_default_size + 15);
 	if (r) {
 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
 		return r;
 	}
-	/* init some VGT regs */
-	radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
-	radeon_ring_write(rdev, (VGT_VERTEX_REUSE_BLOCK_CNTL - PACKET3_SET_CONTEXT_REG_START) >> 2);
-	radeon_ring_write(rdev, 0xe);
-	radeon_ring_write(rdev, 0x10);
+
+	/* setup clear context state */
+	radeon_ring_write(rdev, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
+	radeon_ring_write(rdev, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
+
+	for (i = 0; i < evergreen_default_size; i++)
+		radeon_ring_write(rdev, evergreen_default_state[i]);
+
+	radeon_ring_write(rdev, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
+	radeon_ring_write(rdev, PACKET3_PREAMBLE_END_CLEAR_STATE);
+
+	/* set clear context state */
+	radeon_ring_write(rdev, PACKET3(PACKET3_CLEAR_STATE, 0));
+	radeon_ring_write(rdev, 0);
+
+	/* SQ_VTX_BASE_VTX_LOC */
+	radeon_ring_write(rdev, 0xc0026f00);
+	radeon_ring_write(rdev, 0x00000000);
+	radeon_ring_write(rdev, 0x00000000);
+	radeon_ring_write(rdev, 0x00000000);
+
+	/* Clear consts */
+	radeon_ring_write(rdev, 0xc0036f00);
+	radeon_ring_write(rdev, 0x00000bc4);
+	radeon_ring_write(rdev, 0xffffffff);
+	radeon_ring_write(rdev, 0xffffffff);
+	radeon_ring_write(rdev, 0xffffffff);
+
 	radeon_ring_unlock_commit(rdev);
 
 	return 0;
@@ -731,7 +1190,7 @@ int evergreen_cp_resume(struct radeon_device *rdev)
 
 	/* Set ring buffer size */
 	rb_bufsz = drm_order(rdev->cp.ring_size / 8);
-	tmp = RB_NO_UPDATE | (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
+	tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
 #ifdef __BIG_ENDIAN
 	tmp |= BUF_SWAP_32BIT;
 #endif
@@ -745,8 +1204,19 @@ int evergreen_cp_resume(struct radeon_device *rdev)
 	WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
 	WREG32(CP_RB_RPTR_WR, 0);
 	WREG32(CP_RB_WPTR, 0);
-	WREG32(CP_RB_RPTR_ADDR, rdev->cp.gpu_addr & 0xFFFFFFFF);
-	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->cp.gpu_addr));
+
+	/* set the wb address wether it's enabled or not */
+	WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
+	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
+	WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
+
+	if (rdev->wb.enabled)
+		WREG32(SCRATCH_UMSK, 0xff);
+	else {
+		tmp |= RB_NO_UPDATE;
+		WREG32(SCRATCH_UMSK, 0);
+	}
+
 	mdelay(1);
 	WREG32(CP_RB_CNTL, tmp);
 
@@ -1584,6 +2054,7 @@ int evergreen_irq_set(struct radeon_device *rdev)
 	if (rdev->irq.sw_int) {
 		DRM_DEBUG("evergreen_irq_set: sw int\n");
 		cp_int_cntl |= RB_INT_ENABLE;
+		cp_int_cntl |= TIME_STAMP_INT_ENABLE;
 	}
 	if (rdev->irq.crtc_vblank_int[0]) {
 		DRM_DEBUG("evergreen_irq_set: vblank 0\n");
@@ -1760,8 +2231,10 @@ static inline u32 evergreen_get_ih_wptr(struct radeon_device *rdev)
 {
 	u32 wptr, tmp;
 
-	/* XXX use writeback */
-	wptr = RREG32(IH_RB_WPTR);
+	if (rdev->wb.enabled)
+		wptr = rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4];
+	else
+		wptr = RREG32(IH_RB_WPTR);
 
 	if (wptr & RB_OVERFLOW) {
 		/* When a ring buffer overflow happen start parsing interrupt
@@ -2000,6 +2473,7 @@ restart_ih:
 			break;
 		case 181: /* CP EOP event */
 			DRM_DEBUG("IH: CP EOP\n");
+			radeon_fence_process(rdev);
 			break;
 		case 233: /* GUI IDLE */
 			DRM_DEBUG("IH: CP EOP\n");
@@ -2048,26 +2522,18 @@ static int evergreen_startup(struct radeon_device *rdev)
 			return r;
 	}
 	evergreen_gpu_init(rdev);
-#if 0
-	if (!rdev->r600_blit.shader_obj) {
-		r = r600_blit_init(rdev);
-		if (r) {
-			DRM_ERROR("radeon: failed blitter (%d).\n", r);
-			return r;
-		}
-	}
 
-	r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
-	if (unlikely(r != 0))
-		return r;
-	r = radeon_bo_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
-			&rdev->r600_blit.shader_gpu_addr);
-	radeon_bo_unreserve(rdev->r600_blit.shader_obj);
+	r = evergreen_blit_init(rdev);
 	if (r) {
-		DRM_ERROR("failed to pin blit object %d\n", r);
-		return r;
+		evergreen_blit_fini(rdev);
+		rdev->asic->copy = NULL;
+		dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
 	}
-#endif
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
 
 	/* Enable IRQ */
 	r = r600_irq_init(rdev);
@@ -2087,8 +2553,6 @@ static int evergreen_startup(struct radeon_device *rdev)
 	r = evergreen_cp_resume(rdev);
 	if (r)
 		return r;
-	/* write back buffer are not vital so don't worry about failure */
-	r600_wb_enable(rdev);
 
 	return 0;
 }
@@ -2122,23 +2586,43 @@ int evergreen_resume(struct radeon_device *rdev)
 
 int evergreen_suspend(struct radeon_device *rdev)
 {
-#if 0
 	int r;
-#endif
+
 	/* FIXME: we should wait for ring to be empty */
 	r700_cp_stop(rdev);
 	rdev->cp.ready = false;
 	evergreen_irq_suspend(rdev);
-	r600_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	evergreen_pcie_gart_disable(rdev);
-#if 0
+
 	/* unpin shaders bo */
 	r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
 	if (likely(r == 0)) {
 		radeon_bo_unpin(rdev->r600_blit.shader_obj);
 		radeon_bo_unreserve(rdev->r600_blit.shader_obj);
 	}
-#endif
+
+	return 0;
+}
+
+int evergreen_copy_blit(struct radeon_device *rdev,
+			uint64_t src_offset, uint64_t dst_offset,
+			unsigned num_pages, struct radeon_fence *fence)
+{
+	int r;
+
+	mutex_lock(&rdev->r600_blit.mutex);
+	rdev->r600_blit.vb_ib = NULL;
+	r = evergreen_blit_prepare_copy(rdev, num_pages * RADEON_GPU_PAGE_SIZE);
+	if (r) {
+		if (rdev->r600_blit.vb_ib)
+			radeon_ib_free(rdev, &rdev->r600_blit.vb_ib);
+		mutex_unlock(&rdev->r600_blit.mutex);
+		return r;
+	}
+	evergreen_kms_blit_copy(rdev, src_offset, dst_offset, num_pages * RADEON_GPU_PAGE_SIZE);
+	evergreen_blit_done_copy(rdev, fence);
+	mutex_unlock(&rdev->r600_blit.mutex);
 	return 0;
 }
 
@@ -2246,8 +2730,8 @@ int evergreen_init(struct radeon_device *rdev)
 	if (r) {
 		dev_err(rdev->dev, "disabling GPU acceleration\n");
 		r700_cp_fini(rdev);
-		r600_wb_fini(rdev);
 		r600_irq_fini(rdev);
+		radeon_wb_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		evergreen_pcie_gart_fini(rdev);
 		rdev->accel_working = false;
@@ -2269,10 +2753,10 @@ int evergreen_init(struct radeon_device *rdev)
 
 void evergreen_fini(struct radeon_device *rdev)
 {
-	/*r600_blit_fini(rdev);*/
+	evergreen_blit_fini(rdev);
 	r700_cp_fini(rdev);
-	r600_wb_fini(rdev);
 	r600_irq_fini(rdev);
+	radeon_wb_fini(rdev);
 	radeon_irq_kms_fini(rdev);
 	evergreen_pcie_gart_fini(rdev);
 	radeon_gem_fini(rdev);