Merge branch 'drm-radeon-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6

* 'drm-radeon-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm/radeon/kms/atom: fix PHY init
  drm/radeon/kms: add missing Evergreen texture formats to the CS parser
  drm/radeon/kms: viewport height has to be even
  drm/radeon/kms: remove duplicate reg from r600 safe regs
  drm/radeon/kms: add support for Llano Fusion APUs
  drm/radeon/kms: add llano pci ids
  drm/radeon/kms: fill in asic struct for llano
  drm/radeon/kms: add family ids for llano APUs
  drm/radeon: fix oops in ttm reserve when pageflipping (v2)
  drm/radeon/kms: clean up the radeon kms Kconfig
  drm/radeon/kms: fix thermal sensor reading on juniper
  drm/radeon/kms: add missing case for cayman thermal sensor
  drm/radeon/kms: add blit support for cayman (v2)
  drm/radeon/kms/blit: workaround some hw issues on evergreen+

18 files changed, 819 insertions, 304 deletions

diff --git a/drivers/gpu/drm/radeon/Kconfig b/drivers/gpu/drm/radeon/Kconfig
index 9746fee59f5..ea92bbe3ed3 100644
--- a/drivers/gpu/drm/radeon/Kconfig
+++ b/drivers/gpu/drm/radeon/Kconfig
@@ -28,11 +28,4 @@ config DRM_RADEON_KMS
 	  The kernel will also perform security check on command stream
 	  provided by the user, we want to catch and forbid any illegal use
 	  of the GPU such as DMA into random system memory or into memory
-	  not owned by the process supplying the command stream. This part
-	  of the code is still incomplete and this why we propose that patch
-	  as a staging driver addition, future security might forbid current
-	  experimental userspace to run.
-
-	  This code support the following hardware : R1XX,R2XX,R3XX,R4XX,R5XX
-	  (radeon up to X1950). Works is underway to provide support for R6XX,
-	  R7XX and newer hardware (radeon from HD2XXX to HD4XXX).
+	  not owned by the process supplying the command stream.
diff --git a/drivers/gpu/drm/radeon/atombios_crtc.c b/drivers/gpu/drm/radeon/atombios_crtc.c
index ec848787d7d..84a69e7fa11 100644
--- a/drivers/gpu/drm/radeon/atombios_crtc.c
+++ b/drivers/gpu/drm/radeon/atombios_crtc.c
@@ -1045,7 +1045,7 @@ static int dce4_crtc_do_set_base(struct drm_crtc *crtc,
 	uint64_t fb_location;
 	uint32_t fb_format, fb_pitch_pixels, tiling_flags;
 	u32 fb_swap = EVERGREEN_GRPH_ENDIAN_SWAP(EVERGREEN_GRPH_ENDIAN_NONE);
-	u32 tmp;
+	u32 tmp, viewport_w, viewport_h;
 	int r;
 
 	/* no fb bound */
@@ -1171,8 +1171,10 @@ static int dce4_crtc_do_set_base(struct drm_crtc *crtc,
 	y &= ~1;
 	WREG32(EVERGREEN_VIEWPORT_START + radeon_crtc->crtc_offset,
 	       (x << 16) | y);
+	viewport_w = crtc->mode.hdisplay;
+	viewport_h = (crtc->mode.vdisplay + 1) & ~1;
 	WREG32(EVERGREEN_VIEWPORT_SIZE + radeon_crtc->crtc_offset,
-	       (crtc->mode.hdisplay << 16) | crtc->mode.vdisplay);
+	       (viewport_w << 16) | viewport_h);
 
 	/* pageflip setup */
 	/* make sure flip is at vb rather than hb */
@@ -1213,7 +1215,7 @@ static int avivo_crtc_do_set_base(struct drm_crtc *crtc,
 	uint64_t fb_location;
 	uint32_t fb_format, fb_pitch_pixels, tiling_flags;
 	u32 fb_swap = R600_D1GRPH_SWAP_ENDIAN_NONE;
-	u32 tmp;
+	u32 tmp, viewport_w, viewport_h;
 	int r;
 
 	/* no fb bound */
@@ -1338,8 +1340,10 @@ static int avivo_crtc_do_set_base(struct drm_crtc *crtc,
 	y &= ~1;
 	WREG32(AVIVO_D1MODE_VIEWPORT_START + radeon_crtc->crtc_offset,
 	       (x << 16) | y);
+	viewport_w = crtc->mode.hdisplay;
+	viewport_h = (crtc->mode.vdisplay + 1) & ~1;
 	WREG32(AVIVO_D1MODE_VIEWPORT_SIZE + radeon_crtc->crtc_offset,
-	       (crtc->mode.hdisplay << 16) | crtc->mode.vdisplay);
+	       (viewport_w << 16) | viewport_h);
 
 	/* pageflip setup */
 	/* make sure flip is at vb rather than hb */
diff --git a/drivers/gpu/drm/radeon/cayman_blit_shaders.c b/drivers/gpu/drm/radeon/cayman_blit_shaders.c
index e148ab04b80..7b4eeb7b4a8 100644
--- a/drivers/gpu/drm/radeon/cayman_blit_shaders.c
+++ b/drivers/gpu/drm/radeon/cayman_blit_shaders.c
@@ -39,17 +39,335 @@
 
 const u32 cayman_default_state[] =
 {
-	/* XXX fill in additional blit state */
+	0xc0066900,
+	0x00000000,
+	0x00000060, /* DB_RENDER_CONTROL */
+	0x00000000, /* DB_COUNT_CONTROL */
+	0x00000000, /* DB_DEPTH_VIEW */
+	0x0000002a, /* DB_RENDER_OVERRIDE */
+	0x00000000, /* DB_RENDER_OVERRIDE2 */
+	0x00000000, /* DB_HTILE_DATA_BASE */
 
 	0xc0026900,
-	0x00000316,
-	0x0000000e, /* VGT_VERTEX_REUSE_BLOCK_CNTL */
-	0x00000010, /*  */
+	0x0000000a,
+	0x00000000, /* DB_STENCIL_CLEAR */
+	0x00000000, /* DB_DEPTH_CLEAR */
+
+	0xc0036900,
+	0x0000000f,
+	0x00000000, /* DB_DEPTH_INFO */
+	0x00000000, /* DB_Z_INFO */
+	0x00000000, /* DB_STENCIL_INFO */
+
+	0xc0016900,
+	0x00000080,
+	0x00000000, /* PA_SC_WINDOW_OFFSET */
+
+	0xc00d6900,
+	0x00000083,
+	0x0000ffff, /* PA_SC_CLIPRECT_RULE */
+	0x00000000, /* PA_SC_CLIPRECT_0_TL */
+	0x20002000, /* PA_SC_CLIPRECT_0_BR */
+	0x00000000,
+	0x20002000,
+	0x00000000,
+	0x20002000,
+	0x00000000,
+	0x20002000,
+	0xaaaaaaaa, /* PA_SC_EDGERULE */
+	0x00000000, /* PA_SU_HARDWARE_SCREEN_OFFSET */
+	0x0000000f, /* CB_TARGET_MASK */
+	0x0000000f, /* CB_SHADER_MASK */
+
+	0xc0226900,
+	0x00000094,
+	0x80000000, /* PA_SC_VPORT_SCISSOR_0_TL */
+	0x20002000, /* PA_SC_VPORT_SCISSOR_0_BR */
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x80000000,
+	0x20002000,
+	0x00000000, /* PA_SC_VPORT_ZMIN_0 */
+	0x3f800000, /* PA_SC_VPORT_ZMAX_0 */
+
+	0xc0016900,
+	0x000000d4,
+	0x00000000, /* SX_MISC */
 
 	0xc0026900,
 	0x000000d9,
 	0x00000000, /* CP_RINGID */
 	0x00000000, /* CP_VMID */
+
+	0xc0096900,
+	0x00000100,
+	0x00ffffff, /* VGT_MAX_VTX_INDX */
+	0x00000000, /* VGT_MIN_VTX_INDX */
+	0x00000000, /* VGT_INDX_OFFSET */
+	0x00000000, /* VGT_MULTI_PRIM_IB_RESET_INDX */
+	0x00000000, /* SX_ALPHA_TEST_CONTROL */
+	0x00000000, /* CB_BLEND_RED */
+	0x00000000, /* CB_BLEND_GREEN */
+	0x00000000, /* CB_BLEND_BLUE */
+	0x00000000, /* CB_BLEND_ALPHA */
+
+	0xc0016900,
+	0x00000187,
+	0x00000100, /* SPI_VS_OUT_ID_0 */
+
+	0xc0026900,
+	0x00000191,
+	0x00000100, /* SPI_PS_INPUT_CNTL_0 */
+	0x00000101, /* SPI_PS_INPUT_CNTL_1 */
+
+	0xc0016900,
+	0x000001b1,
+	0x00000000, /* SPI_VS_OUT_CONFIG */
+
+	0xc0106900,
+	0x000001b3,
+	0x20000001, /* SPI_PS_IN_CONTROL_0 */
+	0x00000000, /* SPI_PS_IN_CONTROL_1 */
+	0x00000000, /* SPI_INTERP_CONTROL_0 */
+	0x00000000, /* SPI_INPUT_Z */
+	0x00000000, /* SPI_FOG_CNTL */
+	0x00100000, /* SPI_BARYC_CNTL */
+	0x00000000, /* SPI_PS_IN_CONTROL_2 */
+	0x00000000, /* SPI_COMPUTE_INPUT_CNTL */
+	0x00000000, /* SPI_COMPUTE_NUM_THREAD_X */
+	0x00000000, /* SPI_COMPUTE_NUM_THREAD_Y */
+	0x00000000, /* SPI_COMPUTE_NUM_THREAD_Z */
+	0x00000000, /* SPI_GPR_MGMT */
+	0x00000000, /* SPI_LDS_MGMT */
+	0x00000000, /* SPI_STACK_MGMT */
+	0x00000000, /* SPI_WAVE_MGMT_1 */
+	0x00000000, /* SPI_WAVE_MGMT_2 */
+
+	0xc0016900,
+	0x000001e0,
+	0x00000000, /* CB_BLEND0_CONTROL */
+
+	0xc00e6900,
+	0x00000200,
+	0x00000000, /* DB_DEPTH_CONTROL */
+	0x00000000, /* DB_EQAA */
+	0x00cc0010, /* CB_COLOR_CONTROL */
+	0x00000210, /* DB_SHADER_CONTROL */
+	0x00010000, /* PA_CL_CLIP_CNTL */
+	0x00000004, /* PA_SU_SC_MODE_CNTL */
+	0x00000100, /* PA_CL_VTE_CNTL */
+	0x00000000, /* PA_CL_VS_OUT_CNTL */
+	0x00000000, /* PA_CL_NANINF_CNTL */
+	0x00000000, /* PA_SU_LINE_STIPPLE_CNTL */
+	0x00000000, /* PA_SU_LINE_STIPPLE_SCALE */
+	0x00000000, /* PA_SU_PRIM_FILTER_CNTL */
+	0x00000000, /*  */
+	0x00000000, /*  */
+
+	0xc0026900,
+	0x00000229,
+	0x00000000, /* SQ_PGM_START_FS */
+	0x00000000,
+
+	0xc0016900,
+	0x0000023b,
+	0x00000000, /* SQ_LDS_ALLOC_PS */
+
+	0xc0066900,
+	0x00000240,
+	0x00000000, /* SQ_ESGS_RING_ITEMSIZE */
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+
+	0xc0046900,
+	0x00000247,
+	0x00000000, /* SQ_GS_VERT_ITEMSIZE */
+	0x00000000,
+	0x00000000,
+	0x00000000,
+
+	0xc0116900,
+	0x00000280,
+	0x00000000, /* PA_SU_POINT_SIZE */
+	0x00000000, /* PA_SU_POINT_MINMAX */
+	0x00000008, /* PA_SU_LINE_CNTL */
+	0x00000000, /* PA_SC_LINE_STIPPLE */
+	0x00000000, /* VGT_OUTPUT_PATH_CNTL */
+	0x00000000, /* VGT_HOS_CNTL */
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000, /* VGT_GS_MODE */
+
+	0xc0026900,
+	0x00000292,
+	0x00000000, /* PA_SC_MODE_CNTL_0 */
+	0x00000000, /* PA_SC_MODE_CNTL_1 */
+
+	0xc0016900,
+	0x000002a1,
+	0x00000000, /* VGT_PRIMITIVEID_EN */
+
+	0xc0016900,
+	0x000002a5,
+	0x00000000, /* VGT_MULTI_PRIM_IB_RESET_EN */
+
+	0xc0026900,
+	0x000002a8,
+	0x00000000, /* VGT_INSTANCE_STEP_RATE_0 */
+	0x00000000,
+
+	0xc0026900,
+	0x000002ad,
+	0x00000000, /* VGT_REUSE_OFF */
+	0x00000000,
+
+	0xc0016900,
+	0x000002d5,
+	0x00000000, /* VGT_SHADER_STAGES_EN */
+
+	0xc0016900,
+	0x000002dc,
+	0x0000aa00, /* DB_ALPHA_TO_MASK */
+
+	0xc0066900,
+	0x000002de,
+	0x00000000, /* PA_SU_POLY_OFFSET_DB_FMT_CNTL */
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+
+	0xc0026900,
+	0x000002e5,
+	0x00000000, /* VGT_STRMOUT_CONFIG */
+	0x00000000,
+
+	0xc01b6900,
+	0x000002f5,
+	0x76543210, /* PA_SC_CENTROID_PRIORITY_0 */
+	0xfedcba98, /* PA_SC_CENTROID_PRIORITY_1 */
+	0x00000000, /* PA_SC_LINE_CNTL */
+	0x00000000, /* PA_SC_AA_CONFIG */
+	0x00000005, /* PA_SU_VTX_CNTL */
+	0x3f800000, /* PA_CL_GB_VERT_CLIP_ADJ */
+	0x3f800000, /* PA_CL_GB_VERT_DISC_ADJ */
+	0x3f800000, /* PA_CL_GB_HORZ_CLIP_ADJ */
+	0x3f800000, /* PA_CL_GB_HORZ_DISC_ADJ */
+	0x00000000, /* PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0 */
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0x00000000,
+	0xffffffff, /* PA_SC_AA_MASK_X0Y0_X1Y0 */
+	0xffffffff,
+
+	0xc0026900,
+	0x00000316,
+	0x0000000e, /* VGT_VERTEX_REUSE_BLOCK_CNTL */
+	0x00000010, /*  */
+};
+
+const u32 cayman_vs[] =
+{
+	0x00000004,
+	0x80400400,
+	0x0000a03c,
+	0x95000688,
+	0x00004000,
+	0x15000688,
+	0x00000000,
+	0x88000000,
+	0x04000000,
+	0x67961001,
+#ifdef __BIG_ENDIAN
+	0x00020000,
+#else
+	0x00000000,
+#endif
+	0x00000000,
+	0x04000000,
+	0x67961000,
+#ifdef __BIG_ENDIAN
+	0x00020008,
+#else
+	0x00000008,
+#endif
+	0x00000000,
+};
+
+const u32 cayman_ps[] =
+{
+	0x00000004,
+	0xa00c0000,
+	0x00000008,
+	0x80400000,
+	0x00000000,
+	0x95000688,
+	0x00000000,
+	0x88000000,
+	0x00380400,
+	0x00146b10,
+	0x00380000,
+	0x20146b10,
+	0x00380400,
+	0x40146b00,
+	0x80380000,
+	0x60146b00,
+	0x00000010,
+	0x000d1000,
+	0xb0800000,
+	0x00000000,
 };
 
+const u32 cayman_ps_size = ARRAY_SIZE(cayman_ps);
+const u32 cayman_vs_size = ARRAY_SIZE(cayman_vs);
 const u32 cayman_default_size = ARRAY_SIZE(cayman_default_state);
diff --git a/drivers/gpu/drm/radeon/cayman_blit_shaders.h b/drivers/gpu/drm/radeon/cayman_blit_shaders.h
index 33b75e5d0fa..f5d0e9a6026 100644
--- a/drivers/gpu/drm/radeon/cayman_blit_shaders.h
+++ b/drivers/gpu/drm/radeon/cayman_blit_shaders.h
@@ -25,8 +25,11 @@
 #ifndef CAYMAN_BLIT_SHADERS_H
 #define CAYMAN_BLIT_SHADERS_H
 
+extern const u32 cayman_ps[];
+extern const u32 cayman_vs[];
 extern const u32 cayman_default_state[];
 
+extern const u32 cayman_ps_size, cayman_vs_size;
 extern const u32 cayman_default_size;
 
 #endif
diff --git a/drivers/gpu/drm/radeon/evergreen.c b/drivers/gpu/drm/radeon/evergreen.c
index 7c37638095f..98ea597bc76 100644
--- a/drivers/gpu/drm/radeon/evergreen.c
+++ b/drivers/gpu/drm/radeon/evergreen.c
@@ -88,21 +88,39 @@ u32 evergreen_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
 /* get temperature in millidegrees */
 int evergreen_get_temp(struct radeon_device *rdev)
 {
-	u32 temp = (RREG32(CG_MULT_THERMAL_STATUS) & ASIC_T_MASK) >>
-		ASIC_T_SHIFT;
-	u32 actual_temp = 0;
-
-	if (temp & 0x400)
-		actual_temp = -256;
-	else if (temp & 0x200)
-		actual_temp = 255;
-	else if (temp & 0x100) {
-		actual_temp = temp & 0x1ff;
-		actual_temp |= ~0x1ff;
-	} else
-		actual_temp = temp & 0xff;
+	u32 temp, toffset, actual_temp = 0;
+
+	if (rdev->family == CHIP_JUNIPER) {
+		toffset = (RREG32(CG_THERMAL_CTRL) & TOFFSET_MASK) >>
+			TOFFSET_SHIFT;
+		temp = (RREG32(CG_TS0_STATUS) & TS0_ADC_DOUT_MASK) >>
+			TS0_ADC_DOUT_SHIFT;
+
+		if (toffset & 0x100)
+			actual_temp = temp / 2 - (0x200 - toffset);
+		else
+			actual_temp = temp / 2 + toffset;
+
+		actual_temp = actual_temp * 1000;
+
+	} else {
+		temp = (RREG32(CG_MULT_THERMAL_STATUS) & ASIC_T_MASK) >>
+			ASIC_T_SHIFT;
 
-	return (actual_temp * 1000) / 2;
+		if (temp & 0x400)
+			actual_temp = -256;
+		else if (temp & 0x200)
+			actual_temp = 255;
+		else if (temp & 0x100) {
+			actual_temp = temp & 0x1ff;
+			actual_temp |= ~0x1ff;
+		} else
+			actual_temp = temp & 0xff;
+
+		actual_temp = (actual_temp * 1000) / 2;
+	}
+
+	return actual_temp;
 }
 
 int sumo_get_temp(struct radeon_device *rdev)
@@ -1415,6 +1433,8 @@ static u32 evergreen_get_tile_pipe_to_backend_map(struct radeon_device *rdev,
 	case CHIP_CEDAR:
 	case CHIP_REDWOOD:
 	case CHIP_PALM:
+	case CHIP_SUMO:
+	case CHIP_SUMO2:
 	case CHIP_TURKS:
 	case CHIP_CAICOS:
 		force_no_swizzle = false;
@@ -1544,6 +1564,8 @@ static void evergreen_program_channel_remap(struct radeon_device *rdev)
 	case CHIP_REDWOOD:
 	case CHIP_CEDAR:
 	case CHIP_PALM:
+	case CHIP_SUMO:
+	case CHIP_SUMO2:
 	case CHIP_TURKS:
 	case CHIP_CAICOS:
 	default:
@@ -1689,6 +1711,54 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
 		break;
+	case CHIP_SUMO:
+		rdev->config.evergreen.num_ses = 1;
+		rdev->config.evergreen.max_pipes = 4;
+		rdev->config.evergreen.max_tile_pipes = 2;
+		if (rdev->pdev->device == 0x9648)
+			rdev->config.evergreen.max_simds = 3;
+		else if ((rdev->pdev->device == 0x9647) ||
+			 (rdev->pdev->device == 0x964a))
+			rdev->config.evergreen.max_simds = 4;
+		else
+			rdev->config.evergreen.max_simds = 5;
+		rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
+		rdev->config.evergreen.max_gprs = 256;
+		rdev->config.evergreen.max_threads = 248;
+		rdev->config.evergreen.max_gs_threads = 32;
+		rdev->config.evergreen.max_stack_entries = 256;
+		rdev->config.evergreen.sx_num_of_sets = 4;
+		rdev->config.evergreen.sx_max_export_size = 256;
+		rdev->config.evergreen.sx_max_export_pos_size = 64;
+		rdev->config.evergreen.sx_max_export_smx_size = 192;
+		rdev->config.evergreen.max_hw_contexts = 8;
+		rdev->config.evergreen.sq_num_cf_insts = 2;
+
+		rdev->config.evergreen.sc_prim_fifo_size = 0x40;
+		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
+		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
+		break;
+	case CHIP_SUMO2:
+		rdev->config.evergreen.num_ses = 1;
+		rdev->config.evergreen.max_pipes = 4;
+		rdev->config.evergreen.max_tile_pipes = 4;
+		rdev->config.evergreen.max_simds = 2;
+		rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
+		rdev->config.evergreen.max_gprs = 256;
+		rdev->config.evergreen.max_threads = 248;
+		rdev->config.evergreen.max_gs_threads = 32;
+		rdev->config.evergreen.max_stack_entries = 512;
+		rdev->config.evergreen.sx_num_of_sets = 4;
+		rdev->config.evergreen.sx_max_export_size = 256;
+		rdev->config.evergreen.sx_max_export_pos_size = 64;
+		rdev->config.evergreen.sx_max_export_smx_size = 192;
+		rdev->config.evergreen.max_hw_contexts = 8;
+		rdev->config.evergreen.sq_num_cf_insts = 2;
+
+		rdev->config.evergreen.sc_prim_fifo_size = 0x40;
+		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
+		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
+		break;
 	case CHIP_BARTS:
 		rdev->config.evergreen.num_ses = 2;
 		rdev->config.evergreen.max_pipes = 4;
@@ -2039,6 +2109,8 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 	switch (rdev->family) {
 	case CHIP_CEDAR:
 	case CHIP_PALM:
+	case CHIP_SUMO:
+	case CHIP_SUMO2:
 	case CHIP_CAICOS:
 		/* no vertex cache */
 		sq_config &= ~VC_ENABLE;
@@ -2060,6 +2132,8 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 	switch (rdev->family) {
 	case CHIP_CEDAR:
 	case CHIP_PALM:
+	case CHIP_SUMO:
+	case CHIP_SUMO2:
 		ps_thread_count = 96;
 		break;
 	default:
@@ -2099,6 +2173,8 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 	switch (rdev->family) {
 	case CHIP_CEDAR:
 	case CHIP_PALM:
+	case CHIP_SUMO:
+	case CHIP_SUMO2:
 	case CHIP_CAICOS:
 		vgt_cache_invalidation = CACHE_INVALIDATION(TC_ONLY);
 		break;
diff --git a/drivers/gpu/drm/radeon/evergreen_blit_kms.c b/drivers/gpu/drm/radeon/evergreen_blit_kms.c
index ba06a69c6de..57f3bc17b87 100644
--- a/drivers/gpu/drm/radeon/evergreen_blit_kms.c
+++ b/drivers/gpu/drm/radeon/evergreen_blit_kms.c
@@ -31,6 +31,7 @@
 
 #include "evergreend.h"
 #include "evergreen_blit_shaders.h"
+#include "cayman_blit_shaders.h"
 
 #define DI_PT_RECTLIST        0x11
 #define DI_INDEX_SIZE_16_BIT  0x0
@@ -152,6 +153,8 @@ set_vtx_resource(struct radeon_device *rdev, u64 gpu_addr)
 
 	if ((rdev->family == CHIP_CEDAR) ||
 	    (rdev->family == CHIP_PALM) ||
+	    (rdev->family == CHIP_SUMO) ||
+	    (rdev->family == CHIP_SUMO2) ||
 	    (rdev->family == CHIP_CAICOS))
 		cp_set_surface_sync(rdev,
 				    PACKET3_TC_ACTION_ENA, 48, gpu_addr);
@@ -199,6 +202,16 @@ static void
 set_scissors(struct radeon_device *rdev, int x1, int y1,
 	     int x2, int y2)
 {
+	/* workaround some hw bugs */
+	if (x2 == 0)
+		x1 = 1;
+	if (y2 == 0)
+		y1 = 1;
+	if (rdev->family == CHIP_CAYMAN) {
+		if ((x2 == 1) && (y2 == 1))
+			x2 = 2;
+	}
+
 	radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
 	radeon_ring_write(rdev, (PA_SC_SCREEN_SCISSOR_TL - PACKET3_SET_CONTEXT_REG_START) >> 2);
 	radeon_ring_write(rdev, (x1 << 0) | (y1 << 16));
@@ -255,238 +268,284 @@ set_default_state(struct radeon_device *rdev)
 	u64 gpu_addr;
 	int dwords;
 
-	switch (rdev->family) {
-	case CHIP_CEDAR:
-	default:
-		num_ps_gprs = 93;
-		num_vs_gprs = 46;
-		num_temp_gprs = 4;
-		num_gs_gprs = 31;
-		num_es_gprs = 31;
-		num_hs_gprs = 23;
-		num_ls_gprs = 23;
-		num_ps_threads = 96;
-		num_vs_threads = 16;
-		num_gs_threads = 16;
-		num_es_threads = 16;
-		num_hs_threads = 16;
-		num_ls_threads = 16;
-		num_ps_stack_entries = 42;
-		num_vs_stack_entries = 42;
-		num_gs_stack_entries = 42;
-		num_es_stack_entries = 42;
-		num_hs_stack_entries = 42;
-		num_ls_stack_entries = 42;
-		break;
-	case CHIP_REDWOOD:
-		num_ps_gprs = 93;
-		num_vs_gprs = 46;
-		num_temp_gprs = 4;
-		num_gs_gprs = 31;
-		num_es_gprs = 31;
-		num_hs_gprs = 23;
-		num_ls_gprs = 23;
-		num_ps_threads = 128;
-		num_vs_threads = 20;
-		num_gs_threads = 20;
-		num_es_threads = 20;
-		num_hs_threads = 20;
-		num_ls_threads = 20;
-		num_ps_stack_entries = 42;
-		num_vs_stack_entries = 42;
-		num_gs_stack_entries = 42;
-		num_es_stack_entries = 42;
-		num_hs_stack_entries = 42;
-		num_ls_stack_entries = 42;
-		break;
-	case CHIP_JUNIPER:
-		num_ps_gprs = 93;
-		num_vs_gprs = 46;
-		num_temp_gprs = 4;
-		num_gs_gprs = 31;
-		num_es_gprs = 31;
-		num_hs_gprs = 23;
-		num_ls_gprs = 23;
-		num_ps_threads = 128;
-		num_vs_threads = 20;
-		num_gs_threads = 20;
-		num_es_threads = 20;
-		num_hs_threads = 20;
-		num_ls_threads = 20;
-		num_ps_stack_entries = 85;
-		num_vs_stack_entries = 85;
-		num_gs_stack_entries = 85;
-		num_es_stack_entries = 85;
-		num_hs_stack_entries = 85;
-		num_ls_stack_entries = 85;
-		break;
-	case CHIP_CYPRESS:
-	case CHIP_HEMLOCK:
-		num_ps_gprs = 93;
-		num_vs_gprs = 46;
-		num_temp_gprs = 4;
-		num_gs_gprs = 31;
-		num_es_gprs = 31;
-		num_hs_gprs = 23;
-		num_ls_gprs = 23;
-		num_ps_threads = 128;
-		num_vs_threads = 20;
-		num_gs_threads = 20;
-		num_es_threads = 20;
-		num_hs_threads = 20;
-		num_ls_threads = 20;
-		num_ps_stack_entries = 85;
-		num_vs_stack_entries = 85;
-		num_gs_stack_entries = 85;
-		num_es_stack_entries = 85;
-		num_hs_stack_entries = 85;
-		num_ls_stack_entries = 85;
-		break;
-	case CHIP_PALM:
-		num_ps_gprs = 93;
-		num_vs_gprs = 46;
-		num_temp_gprs = 4;
-		num_gs_gprs = 31;
-		num_es_gprs = 31;
-		num_hs_gprs = 23;
-		num_ls_gprs = 23;
-		num_ps_threads = 96;
-		num_vs_threads = 16;
-		num_gs_threads = 16;
-		num_es_threads = 16;
-		num_hs_threads = 16;
-		num_ls_threads = 16;
-		num_ps_stack_entries = 42;
-		num_vs_stack_entries = 42;
-		num_gs_stack_entries = 42;
-		num_es_stack_entries = 42;
-		num_hs_stack_entries = 42;
-		num_ls_stack_entries = 42;
-		break;
-	case CHIP_BARTS:
-		num_ps_gprs = 93;
-		num_vs_gprs = 46;
-		num_temp_gprs = 4;
-		num_gs_gprs = 31;
-		num_es_gprs = 31;
-		num_hs_gprs = 23;
-		num_ls_gprs = 23;
-		num_ps_threads = 128;
-		num_vs_threads = 20;
-		num_gs_threads = 20;
-		num_es_threads = 20;
-		num_hs_threads = 20;
-		num_ls_threads = 20;
-		num_ps_stack_entries = 85;
-		num_vs_stack_entries = 85;
-		num_gs_stack_entries = 85;
-		num_es_stack_entries = 85;
-		num_hs_stack_entries = 85;
-		num_ls_stack_entries = 85;
-		break;
-	case CHIP_TURKS:
-		num_ps_gprs = 93;
-		num_vs_gprs = 46;
-		num_temp_gprs = 4;
-		num_gs_gprs = 31;
-		num_es_gprs = 31;
-		num_hs_gprs = 23;
-		num_ls_gprs = 23;
-		num_ps_threads = 128;
-		num_vs_threads = 20;
-		num_gs_threads = 20;
-		num_es_threads = 20;
-		num_hs_threads = 20;
-		num_ls_threads = 20;
-		num_ps_stack_entries = 42;
-		num_vs_stack_entries = 42;
-		num_gs_stack_entries = 42;
-		num_es_stack_entries = 42;
-		num_hs_stack_entries = 42;
-		num_ls_stack_entries = 42;
-		break;
-	case CHIP_CAICOS:
-		num_ps_gprs = 93;
-		num_vs_gprs = 46;
-		num_temp_gprs = 4;
-		num_gs_gprs = 31;
-		num_es_gprs = 31;
-		num_hs_gprs = 23;
-		num_ls_gprs = 23;
-		num_ps_threads = 128;
-		num_vs_threads = 10;
-		num_gs_threads = 10;
-		num_es_threads = 10;
-		num_hs_threads = 10;
-		num_ls_threads = 10;
-		num_ps_stack_entries = 42;
-		num_vs_stack_entries = 42;
-		num_gs_stack_entries = 42;
-		num_es_stack_entries = 42;
-		num_hs_stack_entries = 42;
-		num_ls_stack_entries = 42;
-		break;
-	}
-
-	if ((rdev->family == CHIP_CEDAR) ||
-	    (rdev->family == CHIP_PALM) ||
-	    (rdev->family == CHIP_CAICOS))
-		sq_config = 0;
-	else
-		sq_config = VC_ENABLE;
-
-	sq_config |= (EXPORT_SRC_C |
-		      CS_PRIO(0) |
-		      LS_PRIO(0) |
-		      HS_PRIO(0) |
-		      PS_PRIO(0) |
-		      VS_PRIO(1) |
-		      GS_PRIO(2) |
-		      ES_PRIO(3));
-
-	sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(num_ps_gprs) |
-				  NUM_VS_GPRS(num_vs_gprs) |
-				  NUM_CLAUSE_TEMP_GPRS(num_temp_gprs));
-	sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(num_gs_gprs) |
-				  NUM_ES_GPRS(num_es_gprs));
-	sq_gpr_resource_mgmt_3 = (NUM_HS_GPRS(num_hs_gprs) |
-				  NUM_LS_GPRS(num_ls_gprs));
-	sq_thread_resource_mgmt = (NUM_PS_THREADS(num_ps_threads) |
-				   NUM_VS_THREADS(num_vs_threads) |
-				   NUM_GS_THREADS(num_gs_threads) |
-				   NUM_ES_THREADS(num_es_threads));
-	sq_thread_resource_mgmt_2 = (NUM_HS_THREADS(num_hs_threads) |
-				     NUM_LS_THREADS(num_ls_threads));
-	sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(num_ps_stack_entries) |
-				    NUM_VS_STACK_ENTRIES(num_vs_stack_entries));
-	sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(num_gs_stack_entries) |
-				    NUM_ES_STACK_ENTRIES(num_es_stack_entries));
-	sq_stack_resource_mgmt_3 = (NUM_HS_STACK_ENTRIES(num_hs_stack_entries) |
-				    NUM_LS_STACK_ENTRIES(num_ls_stack_entries));
-
 	/* set clear context state */
 	radeon_ring_write(rdev, PACKET3(PACKET3_CLEAR_STATE, 0));
 	radeon_ring_write(rdev, 0);
 
-	/* disable dyn gprs */
-	radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
-	radeon_ring_write(rdev, (SQ_DYN_GPR_CNTL_PS_FLUSH_REQ - PACKET3_SET_CONFIG_REG_START) >> 2);
-	radeon_ring_write(rdev, 0);
+	if (rdev->family < CHIP_CAYMAN) {
+		switch (rdev->family) {
+		case CHIP_CEDAR:
+		default:
+			num_ps_gprs = 93;
+			num_vs_gprs = 46;
+			num_temp_gprs = 4;
+			num_gs_gprs = 31;
+			num_es_gprs = 31;
+			num_hs_gprs = 23;
+			num_ls_gprs = 23;
+			num_ps_threads = 96;
+			num_vs_threads = 16;
+			num_gs_threads = 16;
+			num_es_threads = 16;
+			num_hs_threads = 16;
+			num_ls_threads = 16;
+			num_ps_stack_entries = 42;
+			num_vs_stack_entries = 42;
+			num_gs_stack_entries = 42;
+			num_es_stack_entries = 42;
+			num_hs_stack_entries = 42;
+			num_ls_stack_entries = 42;
+			break;
+		case CHIP_REDWOOD:
+			num_ps_gprs = 93;
+			num_vs_gprs = 46;
+			num_temp_gprs = 4;
+			num_gs_gprs = 31;
+			num_es_gprs = 31;
+			num_hs_gprs = 23;
+			num_ls_gprs = 23;
+			num_ps_threads = 128;
+			num_vs_threads = 20;
+			num_gs_threads = 20;
+			num_es_threads = 20;
+			num_hs_threads = 20;
+			num_ls_threads = 20;
+			num_ps_stack_entries = 42;
+			num_vs_stack_entries = 42;
+			num_gs_stack_entries = 42;
+			num_es_stack_entries = 42;
+			num_hs_stack_entries = 42;
+			num_ls_stack_entries = 42;
+			break;
+		case CHIP_JUNIPER:
+			num_ps_gprs = 93;
+			num_vs_gprs = 46;
+			num_temp_gprs = 4;
+			num_gs_gprs = 31;
+			num_es_gprs = 31;
+			num_hs_gprs = 23;
+			num_ls_gprs = 23;
+			num_ps_threads = 128;
+			num_vs_threads = 20;
+			num_gs_threads = 20;
+			num_es_threads = 20;
+			num_hs_threads = 20;
+			num_ls_threads = 20;
+			num_ps_stack_entries = 85;
+			num_vs_stack_entries = 85;
+			num_gs_stack_entries = 85;
+			num_es_stack_entries = 85;
+			num_hs_stack_entries = 85;
+			num_ls_stack_entries = 85;
+			break;
+		case CHIP_CYPRESS:
+		case CHIP_HEMLOCK:
+			num_ps_gprs = 93;
+			num_vs_gprs = 46;
+			num_temp_gprs = 4;
+			num_gs_gprs = 31;
+			num_es_gprs = 31;
+			num_hs_gprs = 23;
+			num_ls_gprs = 23;
+			num_ps_threads = 128;
+			num_vs_threads = 20;
+			num_gs_threads = 20;
+			num_es_threads = 20;
+			num_hs_threads = 20;
+			num_ls_threads = 20;
+			num_ps_stack_entries = 85;
+			num_vs_stack_entries = 85;
+			num_gs_stack_entries = 85;
+			num_es_stack_entries = 85;
+			num_hs_stack_entries = 85;
+			num_ls_stack_entries = 85;
+			break;
+		case CHIP_PALM:
+			num_ps_gprs = 93;
+			num_vs_gprs = 46;
+			num_temp_gprs = 4;
+			num_gs_gprs = 31;
+			num_es_gprs = 31;
+			num_hs_gprs = 23;
+			num_ls_gprs = 23;
+			num_ps_threads = 96;
+			num_vs_threads = 16;
+			num_gs_threads = 16;
+			num_es_threads = 16;
+			num_hs_threads = 16;
+			num_ls_threads = 16;
+			num_ps_stack_entries = 42;
+			num_vs_stack_entries = 42;
+			num_gs_stack_entries = 42;
+			num_es_stack_entries = 42;
+			num_hs_stack_entries = 42;
+			num_ls_stack_entries = 42;
+			break;
+		case CHIP_SUMO:
+			num_ps_gprs = 93;
+			num_vs_gprs = 46;
+			num_temp_gprs = 4;
+			num_gs_gprs = 31;
+			num_es_gprs = 31;
+			num_hs_gprs = 23;
+			num_ls_gprs = 23;
+			num_ps_threads = 96;
+			num_vs_threads = 25;
+			num_gs_threads = 25;
+			num_es_threads = 25;
+			num_hs_threads = 25;
+			num_ls_threads = 25;
+			num_ps_stack_entries = 42;
+			num_vs_stack_entries = 42;
+			num_gs_stack_entries = 42;
+			num_es_stack_entries = 42;
+			num_hs_stack_entries = 42;
+			num_ls_stack_entries = 42;
+			break;
+		case CHIP_SUMO2:
+			num_ps_gprs = 93;
+			num_vs_gprs = 46;
+			num_temp_gprs = 4;
+			num_gs_gprs = 31;
+			num_es_gprs = 31;
+			num_hs_gprs = 23;
+			num_ls_gprs = 23;
+			num_ps_threads = 96;
+			num_vs_threads = 25;
+			num_gs_threads = 25;
+			num_es_threads = 25;
+			num_hs_threads = 25;
+			num_ls_threads = 25;
+			num_ps_stack_entries = 85;
+			num_vs_stack_entries = 85;
+			num_gs_stack_entries = 85;
+			num_es_stack_entries = 85;
+			num_hs_stack_entries = 85;
+			num_ls_stack_entries = 85;
+			break;
+		case CHIP_BARTS:
+			num_ps_gprs = 93;
+			num_vs_gprs = 46;
+			num_temp_gprs = 4;
+			num_gs_gprs = 31;
+			num_es_gprs = 31;
+			num_hs_gprs = 23;
+			num_ls_gprs = 23;
+			num_ps_threads = 128;
+			num_vs_thr