Merge tag 'xf86-video-ati-6.9.0' into debian-unstable

10 files changed, 173 insertions, 22 deletions

diff --git a/configure.ac b/configure.ac
index beafb7e..b8c18a6 100644
--- a/configure.ac
+++ b/configure.ac
@@ -22,7 +22,7 @@
 
 AC_PREREQ(2.57)
 AC_INIT([xf86-video-ati],
-        6.8.192,
+        6.9.0,
         [https://bugs.freedesktop.org/enter_bug.cgi?product=xorg],
         xf86-video-ati)
 
@@ -244,6 +244,66 @@ if test "x$XSERVER_LIBPCIACCESS" = xyes; then
 fi
 AM_CONDITIONAL(XSERVER_LIBPCIACCESS, test "x$XSERVER_LIBPCIACCESS" = xyes)
 
+# Checks for headers/macros for byte swapping
+# Known variants:
+#	<byteswap.h> bswap_16, bswap_32, bswap_64  (glibc)
+#	<sys/endian.h> __swap16, __swap32, __swap64 (OpenBSD)
+#	<sys/endian.h> bswap16, bswap32, bswap64 (other BSD's)
+#	and a fallback to local macros if none of the above are found
+
+# if <byteswap.h> is found, assume it's the correct version
+AC_CHECK_HEADERS([byteswap.h])
+
+# if <sys/endian.h> is found, have to check which version
+AC_CHECK_HEADER([sys/endian.h], [HAVE_SYS_ENDIAN_H="yes"], [HAVE_SYS_ENDIAN_H="no"])
+
+if test "x$HAVE_SYS_ENDIAN_H" = "xyes" ; then
+	AC_MSG_CHECKING([for __swap16 variant of <sys/endian.h> byteswapping macros])
+	AC_LINK_IFELSE([AC_LANG_PROGRAM([
+#include <sys/types.h>
+#include <sys/endian.h>
+ ], [
+int a = 1, b;
+b = __swap16(a);
+ ])
+], [SYS_ENDIAN__SWAP='yes'], [SYS_ENDIAN__SWAP='no'])
+	AC_MSG_RESULT([$SYS_ENDIAN__SWAP])
+
+	AC_MSG_CHECKING([for bswap16 variant of <sys/endian.h> byteswapping macros])
+	AC_LINK_IFELSE([AC_LANG_PROGRAM([
+#include <sys/types.h>
+#include <sys/endian.h>
+ ], [
+int a = 1, b;
+b = bswap16(a);
+ ])
+], [SYS_ENDIAN_BSWAP='yes'], [SYS_ENDIAN_BSWAP='no'])
+	AC_MSG_RESULT([$SYS_ENDIAN_BSWAP])
+
+    	if test "$SYS_ENDIAN_BSWAP" = "yes" ; then
+		USE_SYS_ENDIAN_H=yes
+		BSWAP=bswap
+	else	
+	    	if test "$SYS_ENDIAN__SWAP" = "yes" ; then
+			USE_SYS_ENDIAN_H=yes
+			BSWAP=__swap
+		else
+			USE_SYS_ENDIAN_H=no
+		fi
+	fi
+
+	if test "$USE_SYS_ENDIAN_H" = "yes" ; then
+	    AC_DEFINE([USE_SYS_ENDIAN_H], 1, 
+		[Define to use byteswap macros from <sys/endian.h>])
+	    AC_DEFINE_UNQUOTED([bswap_16], ${BSWAP}16, 
+			[Define to 16-bit byteswap macro])
+	    AC_DEFINE_UNQUOTED([bswap_32], ${BSWAP}32, 
+			[Define to 32-bit byteswap macro])
+	    AC_DEFINE_UNQUOTED([bswap_64], ${BSWAP}64, 
+			[Define to 64-bit byteswap macro])
+	fi
+fi
+
 case $host_os in
   *linux*)
   AC_DEFINE(FGL_LINUX, 1, [Use linux pragma pack]) ;;
diff --git a/src/radeon_accel.c b/src/radeon_accel.c
index d45e932..65ad33d 100644
--- a/src/radeon_accel.c
+++ b/src/radeon_accel.c
@@ -371,7 +371,7 @@ void RADEONEngineInit(ScrnInfoPtr pScrn)
 		   info->CurrentLayout.bitsPerPixel);
 
 #ifdef XF86DRI
-    if (info->directRenderingEnabled && (IS_R300_3D | IS_R500_3D)) {
+    if (info->directRenderingEnabled && (IS_R300_3D || IS_R500_3D)) {
 	drmRadeonGetParam np;
 	int num_pipes;
 
@@ -419,11 +419,11 @@ void RADEONEngineInit(ScrnInfoPtr pScrn)
 	}
     }
 
-    if (IS_R300_3D | IS_R500_3D)
+    if (IS_R300_3D || IS_R500_3D)
 	xf86DrvMsg(pScrn->scrnIndex, X_INFO,
 		   "num pipes is %d\n", info->num_gb_pipes);
 
-    if (IS_R300_3D | IS_R500_3D) {
+    if (IS_R300_3D || IS_R500_3D) {
 	uint32_t gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16 | R300_SUBPIXEL_1_16);
 
 	switch(info->num_gb_pipes) {
diff --git a/src/radeon_accelfuncs.c b/src/radeon_accelfuncs.c
index 3c0b8a0..56793cd 100644
--- a/src/radeon_accelfuncs.c
+++ b/src/radeon_accelfuncs.c
@@ -151,6 +151,11 @@ FUNC_NAME(RADEONSetupForSolidFill)(ScrnInfoPtr pScrn,
 					      | RADEON_DST_Y_TOP_TO_BOTTOM));
 
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 }
 
 /* Subsequent XAA SolidFillRect
@@ -205,6 +210,11 @@ FUNC_NAME(RADEONSetupForSolidLine)(ScrnInfoPtr pScrn,
     OUT_ACCEL_REG(RADEON_DP_WRITE_MASK,      planemask);
 
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 }
 
 /* Subsequent XAA solid horizontal and vertical lines */
@@ -324,6 +334,11 @@ FUNC_NAME(RADEONSetupForDashedLine)(ScrnInfoPtr pScrn,
     OUT_ACCEL_REG(RADEON_BRUSH_DATA0,        pat);
 
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 }
 
 /* Helper function to draw last point for dashed lines */
@@ -358,6 +373,11 @@ FUNC_NAME(RADEONDashedLastPel)(ScrnInfoPtr pScrn,
     OUT_ACCEL_REG(RADEON_DP_BRUSH_FRGD_CLR,  info->dash_fg);
 
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 }
 
 /* Subsequent XAA dashed line */
@@ -461,6 +481,11 @@ FUNC_NAME(RADEONSetupForScreenToScreenCopy)(ScrnInfoPtr pScrn,
 		   (ydir >= 0 ? RADEON_DST_Y_TOP_TO_BOTTOM : 0)));
 
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 
     info->trans_color = trans_color;
     FUNC_NAME(RADEONSetTransparency)(pScrn, trans_color);
@@ -553,6 +578,11 @@ FUNC_NAME(RADEONSetupForMono8x8PatternFill)(ScrnInfoPtr pScrn,
 #endif
 
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 }
 
 /* Subsequent XAA 8x8 pattern color expansion.  Because they are used in
@@ -1086,6 +1116,11 @@ FUNC_NAME(RADEONSetClippingRectangle)(ScrnInfoPtr pScrn,
     OUT_ACCEL_REG(RADEON_SC_BOTTOM_RIGHT,    tmp2);
 
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 
     FUNC_NAME(RADEONSetTransparency)(pScrn, info->trans_color);
 }
@@ -1105,6 +1140,11 @@ FUNC_NAME(RADEONDisableClipping)(ScrnInfoPtr pScrn)
 					      RADEON_DEFAULT_SC_BOTTOM_MAX));
 
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 
     FUNC_NAME(RADEONSetTransparency)(pScrn, info->trans_color);
 }
diff --git a/src/radeon_atombios.c b/src/radeon_atombios.c
index 900e928..20aa722 100644
--- a/src/radeon_atombios.c
+++ b/src/radeon_atombios.c
@@ -1787,6 +1787,13 @@ RADEONGetATOMConnectorInfoFromBIOSConnectorTable (ScrnInfoPtr pScrn)
 	    continue;
 	}
 #endif
+#if 1
+	if (i == ATOM_DEVICE_TV1_INDEX) {
+	    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "Skipping TV-Out\n");
+	    info->BiosConnector[i].valid = FALSE;
+	    continue;
+	}
+#endif
 
 	info->BiosConnector[i].valid = TRUE;
 	info->BiosConnector[i].output_id = ci.sucI2cId.sbfAccess.bfI2C_LineMux;
diff --git a/src/radeon_exa.c b/src/radeon_exa.c
index 0193a28..f461f3c 100644
--- a/src/radeon_exa.c
+++ b/src/radeon_exa.c
@@ -323,7 +323,7 @@ do {									\
 	case EXA_ENGINEMODE_UNKNOWN:					\
 	    wait_until |= RADEON_WAIT_HOST_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN;	\
 	case EXA_ENGINEMODE_2D:						\
-	    wait_until |= RADEON_WAIT_2D_IDLECLEAN;			\
+	    wait_until |= RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE;		\
 	case EXA_ENGINEMODE_3D:						\
 	    break;							\
 	}								\
diff --git a/src/radeon_exa_funcs.c b/src/radeon_exa_funcs.c
index 13a7de5..56de23e 100644
--- a/src/radeon_exa_funcs.c
+++ b/src/radeon_exa_funcs.c
@@ -124,7 +124,6 @@ FUNC_NAME(RADEONPrepareSolid)(PixmapPtr pPix, int alu, Pixel pm, Pixel fg)
 static void
 FUNC_NAME(RADEONSolid)(PixmapPtr pPix, int x1, int y1, int x2, int y2)
 {
-
     RINFO_FROM_SCREEN(pPix->drawable.pScreen);
     ACCEL_PREAMBLE();
 
@@ -139,7 +138,16 @@ FUNC_NAME(RADEONSolid)(PixmapPtr pPix, int x1, int y1, int x2, int y2)
 static void
 FUNC_NAME(RADEONDoneSolid)(PixmapPtr pPix)
 {
+    RINFO_FROM_SCREEN(pPix->drawable.pScreen);
+    ACCEL_PREAMBLE();
+
     TRACE;
+
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 }
 
 void
@@ -206,7 +214,6 @@ FUNC_NAME(RADEONCopy)(PixmapPtr pDst,
 		      int dstX, int dstY,
 		      int w, int h)
 {
-
     RINFO_FROM_SCREEN(pDst->drawable.pScreen);
     ACCEL_PREAMBLE();
 
@@ -233,7 +240,16 @@ FUNC_NAME(RADEONCopy)(PixmapPtr pDst,
 static void
 FUNC_NAME(RADEONDoneCopy)(PixmapPtr pDst)
 {
+    RINFO_FROM_SCREEN(pDst->drawable.pScreen);
+    ACCEL_PREAMBLE();
+
     TRACE;
+
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 }
 
 static Bool
@@ -343,6 +359,11 @@ RADEONBlitChunk(ScrnInfoPtr pScrn, uint32_t datatype, uint32_t src_pitch_offset,
     OUT_ACCEL_REG(RADEON_DST_Y_X, (dstY << 16) | dstX);
     OUT_ACCEL_REG(RADEON_DST_HEIGHT_WIDTH, (h << 16) | w);
     FINISH_ACCEL();
+    BEGIN_ACCEL(2);
+    OUT_ACCEL_REG(RADEON_DSTCACHE_CTLSTAT, RADEON_RB2D_DC_FLUSH_ALL);
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL,
+                  RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_DMA_GUI_IDLE);
+    FINISH_ACCEL();
 }
 #endif
 
diff --git a/src/radeon_exa_render.c b/src/radeon_exa_render.c
index 2319e3b..4736e4f 100644
--- a/src/radeon_exa_render.c
+++ b/src/radeon_exa_render.c
@@ -1978,7 +1978,7 @@ static void FUNC_NAME(RadeonCompositeTile)(PixmapPtr pDst,
 		 RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE |
 		 (4 << RADEON_CP_VC_CNTL_NUM_SHIFT));
     } else {
-	if (IS_R300_3D | IS_R500_3D)
+	if (IS_R300_3D || IS_R500_3D)
 	    BEGIN_RING(4 * vtx_count + 4);
 	else
 	    BEGIN_RING(4 * vtx_count + 2);
@@ -1991,7 +1991,7 @@ static void FUNC_NAME(RadeonCompositeTile)(PixmapPtr pDst,
     }
 
 #else /* ACCEL_CP */
-    if (IS_R300_3D | IS_R500_3D)
+    if (IS_R300_3D || IS_R500_3D)
 	BEGIN_ACCEL(2 + vtx_count * 4);
     else
 	BEGIN_ACCEL(1 + vtx_count * 4);
@@ -2032,7 +2032,7 @@ static void FUNC_NAME(RadeonCompositeTile)(PixmapPtr pDst,
 		xFixedToFloat(srcTopRight.x) / info->texW[0],     xFixedToFloat(srcTopRight.y) / info->texH[0]);
     }
 
-    if (IS_R300_3D | IS_R500_3D)
+    if (IS_R300_3D || IS_R500_3D)
 	/* flushing is pipelined, free/finish is not */
 	OUT_ACCEL_REG(R300_RB3D_DSTCACHE_CTLSTAT, R300_DC_FLUSH_3D);
 
@@ -2114,18 +2114,13 @@ static void FUNC_NAME(RadeonDoneComposite)(PixmapPtr pDst)
 
     ENTER_DRAW(0);
 
-    if (IS_R500_3D || ((info->ChipFamily == CHIP_FAMILY_RS400) ||
-		       (info->ChipFamily == CHIP_FAMILY_RS480) ||
-		       (info->ChipFamily == CHIP_FAMILY_RS600) ||
-		       (info->ChipFamily == CHIP_FAMILY_RS690) ||
-		       (info->ChipFamily == CHIP_FAMILY_RS740))) {
-	/* r500 shows corruption on small things like glyphs without a 3D idle 
-	 * IGP shows more substantial corruption
-	 */
+    if (IS_R300_3D || IS_R500_3D) {
+	BEGIN_ACCEL(2);
+	OUT_ACCEL_REG(R300_RB3D_DSTCACHE_CTLSTAT, R300_RB3D_DC_FLUSH_ALL);
+    } else
 	BEGIN_ACCEL(1);
-	OUT_ACCEL_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
-	FINISH_ACCEL();
-    }
+    OUT_ACCEL_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
+    FINISH_ACCEL();
 
     LEAVE_DRAW(0);
 }
diff --git a/src/radeon_macros.h b/src/radeon_macros.h
index 3675dc5..afe442e 100644
--- a/src/radeon_macros.h
+++ b/src/radeon_macros.h
@@ -51,7 +51,23 @@
 
 #include "compiler.h"
 
+#if HAVE_BYTESWAP_H
 #include <byteswap.h>
+#elif defined(USE_SYS_ENDIAN_H)
+#include <sys/endian.h>
+#else
+#define	bswap_16(value)  \
+ 	((((value) & 0xff) << 8) | ((value) >> 8))
+
+#define	bswap_32(value)	\
+ 	(((uint32_t)bswap_16((uint16_t)((value) & 0xffff)) << 16) | \
+ 	(uint32_t)bswap_16((uint16_t)((value) >> 16)))
+ 
+#define	bswap_64(value)	\
+ 	(((uint64_t)bswap_32((uint32_t)((value) & 0xffffffff)) \
+ 	    << 32) | \
+ 	(uint64_t)bswap_32((uint32_t)((value) >> 32)))
+#endif
 
 #if X_BYTE_ORDER == X_BIG_ENDIAN
 #define le32_to_cpu(x) bswap_32(x)
diff --git a/src/radeon_reg.h b/src/radeon_reg.h
index 52795b1..59e2f12 100644
--- a/src/radeon_reg.h
+++ b/src/radeon_reg.h
@@ -1458,6 +1458,7 @@
 #       define RADEON_RB2D_DC_FLUSH_ALL     0xf
 #       define RADEON_RB2D_DC_BUSY          (1 << 31)
 #define RADEON_RB2D_DSTCACHE_MODE           0x3428
+#define RADEON_DSTCACHE_CTLSTAT             0x1714
 
 #define RADEON_RB3D_ZCACHE_MODE             0x3250
 #define RADEON_RB3D_ZCACHE_CTLSTAT          0x3254
diff --git a/src/radeon_textured_videofuncs.c b/src/radeon_textured_videofuncs.c
index d5d1b1c..277d9b2 100644
--- a/src/radeon_textured_videofuncs.c
+++ b/src/radeon_textured_videofuncs.c
@@ -117,7 +117,10 @@ FUNC_NAME(RADEONDisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv
 	OUT_VIDEO_REG(RADEON_RB3D_DSTCACHE_CTLSTAT, RADEON_RB3D_DC_FLUSH);
     /* We must wait for 3d to idle, in case source was just written as a dest. */
     OUT_VIDEO_REG(RADEON_WAIT_UNTIL,
-		RADEON_WAIT_HOST_IDLECLEAN | RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
+		  RADEON_WAIT_HOST_IDLECLEAN |
+		  RADEON_WAIT_2D_IDLECLEAN |
+		  RADEON_WAIT_3D_IDLECLEAN |
+		  RADEON_WAIT_DMA_GUI_IDLE);
     FINISH_VIDEO();
 
     if (IS_R300_3D || IS_R500_3D) {
@@ -640,6 +643,14 @@ FUNC_NAME(RADEONDisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv
 	pBox++;
     }
 
+    if (IS_R300_3D || IS_R500_3D) {
+	BEGIN_VIDEO(2);
+	OUT_VIDEO_REG(R300_RB3D_DSTCACHE_CTLSTAT, R300_RB3D_DC_FLUSH_ALL);
+    } else
+	BEGIN_VIDEO(1);
+    OUT_VIDEO_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
+    FINISH_VIDEO();
+
     DamageDamageRegion(pPriv->pDraw, &pPriv->clip);
 }