first attempt to fix PCI GART and PPC (client side still broken for PPC)

libdrm/xf86drm.c

@@ -997,6 +997,28 @@ unsigned int drmAgpDeviceId(int fd)
     return i.id_device;
 }
 
+int drmScatterGatherAlloc(int fd, unsigned long size,
+		unsigned long *handle)
+{
+    drm_agp_buffer_t b;
+    *handle = 0;
+    b.size   = size;
+    b.handle = 0;
+    if (ioctl(fd, DRM_IOCTL_SG_ALLOC, &b)) return -errno;
+    *handle = b.handle;
+    return 0;
+}
+
+int drmScatterGatherFree(int fd, unsigned long handle)
+{
+    drm_agp_buffer_t b;
+
+    b.size   = 0;
+    b.handle = handle;
+    if (ioctl(fd, DRM_IOCTL_SG_FREE, &b)) return -errno;
+    return 0;
+}
+
 int drmError(int err, const char *label)
 {
     switch (err) {

linux-core/drmP.h

@@ -144,6 +144,7 @@
 #define DRM_MEM_BOUNDAGP  17
 #define DRM_MEM_CTXBITMAP 18
 #define DRM_MEM_STUB      19
+#define DRM_MEM_SGLISTS   20
 
 #define DRM_MAX_CTXBITMAP (PAGE_SIZE * 8)
 
@@ -540,7 +541,8 @@ typedef struct drm_device_dma {
 	unsigned long	  *pagelist;
 	unsigned long	  byte_count;
 	enum {
-		_DRM_DMA_USE_AGP = 0x01
+		_DRM_DMA_USE_AGP = 0x01,
+		_DRM_DMA_USE_SG  = 0x02
 	} flags;
 
 				/* DMA support */

linux-core/drm_bufs.c

@@ -136,7 +136,7 @@ int DRM(addmap)( struct inode *inode, struct file *filp,
 #endif
 	case _DRM_SCATTER_GATHER:
 		if (!dev->sg) {
-			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+			DRM(free)(map, sizeof(*map), DRM_MEM_MAPS);
 			return -EINVAL;
 		}
 		map->offset = map->offset + dev->sg->handle;

linux-core/drm_memory.h

@@ -63,6 +63,7 @@ static drm_mem_stats_t	  DRM(mem_stats)[]   = {
 	[DRM_MEM_MAPPINGS]  = { "mappings" },
 	[DRM_MEM_BUFLISTS]  = { "buflists" },
 	[DRM_MEM_AGPLISTS]  = { "agplist"  },
+	[DRM_MEM_SGLISTS]   = { "sglist"   },
 	[DRM_MEM_TOTALAGP]  = { "totalagp" },
 	[DRM_MEM_BOUNDAGP]  = { "boundagp" },
 	[DRM_MEM_CTXBITMAP] = { "ctxbitmap"},

linux-core/r128_drv.c

@@ -33,6 +33,7 @@
 #include "r128.h"
 #include "drmP.h"
 #include "r128_drv.h"
+#include "ati_pcigart.h"
 
 #define DRIVER_AUTHOR		"Gareth Hughes, VA Linux Systems Inc."
 
@@ -86,5 +87,6 @@
 #include "drm_lock.h"
 #include "drm_memory.h"
 #include "drm_proc.h"
+#include "drm_scatter.h"
 #include "drm_vm.h"
 #include "drm_stub.h"

linux/Makefile.linux

@@ -47,7 +47,7 @@
 
 # **** End of SMP/MODVERSIONS detection
 
-MODS =		gamma.o tdfx.o
+MODS =		gamma.o tdfx.o r128.o
 LIBS =
 
 DRMTEMPLATES =	drm_auth.h drm_bufs.h drm_context.h drm_dma.h drm_drawable.h \
@@ -58,6 +58,9 @@ DRMHEADERS =	drm.h drmP.h compat-pre24.h
 GAMMAOBJS =	gamma_drv.o gamma_dma.o
 GAMMAHEADERS =	gamma_drv.h $(DRMHEADERS) $(DRMTEMPLATES)
 
+R128OBJS =	r128_drv.o r128_cce.o r128_state.o
+R128HEADERS =	r128.h r128_drv.h r128_drm.h $(DRMHEADERS) $(DRMTEMPLATES)
+
 TDFXOBJS =	tdfx_drv.o
 TDFXHEADERS =	tdfx.h $(DRMHEADERS)
 
@@ -132,7 +135,7 @@ ifeq ($(AGP),1)
 MODCFLAGS += -DCONFIG_AGP -DCONFIG_AGP_MODULE
 DRMTEMPLATES += drm_agpsupport.h
 DRMHEADERS += agpsupport-pre24.h
-MODS += mga.o r128.o radeon.o
+MODS += mga.o radeon.o
 ifeq ($(MACHINE),i386)
 MODS += i810.o
 endif
@@ -147,9 +150,6 @@ MGAHEADERS =	mga.h mga_drv.h mga_drm.h $(DRMHEADERS) $(DRMTEMPLATES)
 I810OBJS =	i810_drv.o i810_dma.o
 I810HEADERS =	i810.h i810_drv.h i810_drm.h $(DRMHEADERS) $(DRMTEMPLATES)
 
-R128OBJS =	r128_drv.o r128_cce.o r128_state.o
-R128HEADERS =	r128.h r128_drv.h r128_drm.h $(DRMHEADERS) $(DRMTEMPLATES)
-
 RADEONOBJS =	radeon_drv.o radeon_cp.o radeon_state.o
 RADEONHEADERS =	radeon.h radeon_drv.h radeon_drm.h $(DRMHEADERS) \
 		$(DRMTEMPLATES)
@@ -210,6 +210,11 @@ gamma_drv.o: gamma_drv.c
 gamma.o: $(GAMMAOBJS)
 	$(LD) -r $^ -o $@
 
+r128_drv.o: r128_drv.c
+	$(CC) $(MODCFLAGS) -DEXPORT_SYMTAB -I$(TREE) -c $< -o $@
+r128.o: $(R128OBJS) $(LIBS)
+	$(LD) -r $^ -o $@
+
 tdfx_drv.o: tdfx_drv.c
 	$(CC) $(MODCFLAGS) -DEXPORT_SYMTAB -I$(TREE) -c $< -o $@
 tdfx.o: $(TDFXOBJS) $(LIBS)
@@ -229,11 +234,6 @@ i810_drv.o: i810_drv.c
 i810.o: $(I810OBJS) $(LIBS)
 	$(LD) -r $^ -o $@
 
-r128_drv.o: r128_drv.c
-	$(CC) $(MODCFLAGS) -DEXPORT_SYMTAB -I$(TREE) -c $< -o $@
-r128.o: $(R128OBJS) $(LIBS)
-	$(LD) -r $^ -o $@
-
 radeon_drv.o: radeon_drv.c
 	$(CC) $(MODCFLAGS) -DEXPORT_SYMTAB -I$(TREE) -c $< -o $@
 radeon.o: $(RADEONOBJS) $(LIBS)
@@ -254,11 +254,11 @@ ChangeLog:
 	$(CC) $(MODCFLAGS) -I$(TREE) -c $< -o $@
 
 $(GAMMAOBJS): $(GAMMAHEADERS)
+$(R128OBJS): $(R128HEADERS)
 $(TDFXOBJS): $(TDFXHEADERS)
 ifeq ($(AGP),1)
 $(MGAOBJS): $(MGAHEADERS)
 $(I810OBJS): $(I810HEADERS)
-$(R128OBJS): $(R128HEADERS)
 $(RADEONOBJS): $(RADEONHEADERS)
 endif
 

linux/drmP.h

@@ -144,6 +144,7 @@
 #define DRM_MEM_BOUNDAGP  17
 #define DRM_MEM_CTXBITMAP 18
 #define DRM_MEM_STUB      19
+#define DRM_MEM_SGLISTS   20
 
 #define DRM_MAX_CTXBITMAP (PAGE_SIZE * 8)
 
@@ -540,7 +541,8 @@ typedef struct drm_device_dma {
 	unsigned long	  *pagelist;
 	unsigned long	  byte_count;
 	enum {
-		_DRM_DMA_USE_AGP = 0x01
+		_DRM_DMA_USE_AGP = 0x01,
+		_DRM_DMA_USE_SG  = 0x02
 	} flags;
 
 				/* DMA support */

linux/drm_bufs.h

@@ -136,7 +136,7 @@ int DRM(addmap)( struct inode *inode, struct file *filp,
 #endif
 	case _DRM_SCATTER_GATHER:
 		if (!dev->sg) {
-			drm_free(map, sizeof(*map), DRM_MEM_MAPS);
+			DRM(free)(map, sizeof(*map), DRM_MEM_MAPS);
 			return -EINVAL;
 		}
 		map->offset = map->offset + dev->sg->handle;

linux/drm_memory.h

@@ -63,6 +63,7 @@ static drm_mem_stats_t	  DRM(mem_stats)[]   = {
 	[DRM_MEM_MAPPINGS]  = { "mappings" },
 	[DRM_MEM_BUFLISTS]  = { "buflists" },
 	[DRM_MEM_AGPLISTS]  = { "agplist"  },
+	[DRM_MEM_SGLISTS]   = { "sglist"   },
 	[DRM_MEM_TOTALAGP]  = { "totalagp" },
 	[DRM_MEM_BOUNDAGP]  = { "boundagp" },
 	[DRM_MEM_CTXBITMAP] = { "ctxbitmap"},

linux/r128.h

@@ -37,10 +37,11 @@
 /* General customization:
  */
 #define __HAVE_AGP		1
-#define __MUST_HAVE_AGP		1
+#define __MUST_HAVE_AGP		0
 #define __HAVE_MTRR		1
 #define __HAVE_CTX_BITMAP	1
 #define __HAVE_SG		1
+#define __HAVE_PCI_DMA		1
 
 /* Driver customization:
  */

linux/r128_cce.c

@@ -206,9 +206,10 @@ static void r128_do_cce_flush( drm_r128_private_t *dev_priv )
 int r128_do_cce_idle( drm_r128_private_t *dev_priv )
 {
 	int i;
+	drm_r128_ring_buffer_t *ring = &dev_priv->ring;
 	
 	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
-		if ( *dev_priv->ring.head == dev_priv->ring.tail ) {
+		if ( GET_RING_HEAD(ring) == dev_priv->ring.tail ) {
 			int pm4stat = R128_READ( R128_PM4_STAT );
 			if ( ( (pm4stat & R128_PM4_FIFOCNT_MASK) >=
 			       dev_priv->cce_fifo_size ) &&
@@ -249,7 +250,7 @@ static void r128_do_cce_reset( drm_r128_private_t *dev_priv )
 {
 	R128_WRITE( R128_PM4_BUFFER_DL_WPTR, 0 );
 	R128_WRITE( R128_PM4_BUFFER_DL_RPTR, 0 );
-	*dev_priv->ring.head = 0;
+	SET_RING_HEAD( &dev_priv->ring, 0 );
 	dev_priv->ring.tail = 0;
 }
 
@@ -328,7 +329,7 @@ static void r128_cce_init_ring_buffer( drm_device_t *dev )
 	R128_WRITE( R128_PM4_BUFFER_DL_RPTR, 0 );
 
 	/* DL_RPTR_ADDR is a physical address in AGP space. */
-	*dev_priv->ring.head = 0;
+	SET_RING_HEAD( &dev_priv->ring, 0 );
 
 	if ( !dev_priv->is_pci ) {
 		R128_WRITE( R128_PM4_BUFFER_DL_RPTR_ADDR,
@@ -380,7 +381,7 @@ static int r128_do_init_cce( drm_device_t *dev, drm_r128_init_t *init )
 
 	if ( dev_priv->is_pci && !dev->sg ) {
 		DRM_ERROR( "PCI GART memory not allocated!\n" );
-		drm_free( dev_priv, sizeof(*dev_priv), DRM_MEM_DRIVER );
+		DRM(free)( dev_priv, sizeof(*dev_priv), DRM_MEM_DRIVER );
 		dev->dev_private = NULL;
 		return -EINVAL;
        }
@@ -887,7 +888,7 @@ int r128_wait_ring( drm_r128_private_t *dev_priv, int n )
 	int i;
 
 	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
-		ring->space = *ring->head - ring->tail;
+		ring->space = GET_RING_HEAD( ring ) - ring->tail;
 		if ( ring->space <= 0 )
 			ring->space += ring->size;
 
@@ -906,7 +907,7 @@ void r128_update_ring_snapshot( drm_r128_private_t *dev_priv )
 {
 	drm_r128_ring_buffer_t *ring = &dev_priv->ring;
 
-	ring->space = *ring->head - ring->tail;
+	ring->space = GET_RING_HEAD( ring ) - ring->tail;
 #if R128_PERFORMANCE_BOXES
 	if ( ring->space == 0 )
 		atomic_inc( &dev_priv->idle_count );

linux/r128_drv.c

@@ -33,6 +33,7 @@
 #include "r128.h"
 #include "drmP.h"
 #include "r128_drv.h"
+#include "ati_pcigart.h"
 
 #define DRIVER_AUTHOR		"Gareth Hughes, VA Linux Systems Inc."
 
@@ -86,5 +87,6 @@
 #include "drm_lock.h"
 #include "drm_memory.h"
 #include "drm_proc.h"
+#include "drm_scatter.h"
 #include "drm_vm.h"
 #include "drm_stub.h"

linux/r128_drv.h

@@ -67,6 +67,8 @@ typedef struct drm_r128_private {
 
 	int usec_timeout;
 	int is_pci;
+	unsigned long phys_pci_gart;
+	unsigned long cce_buffers_offset;
 
 	atomic_t idle_count;
 
@@ -241,6 +243,7 @@ extern int r128_cce_indirect( struct inode *inode, struct file *filp,
 #	define R128_PC_FLUSH_ALL		0x00ff
 #	define R128_PC_BUSY			(1 << 31)
 
+#define R128_PCI_GART_PAGE              0x017c
 #define R128_PRIM_TEX_CNTL_C		0x1cb0
 
 #define R128_SCALE_3D_CNTL		0x1a00
@@ -370,8 +373,8 @@ extern int r128_cce_indirect( struct inode *inode, struct file *filp,
 #define R128_ADDR(reg)		(R128_BASE( reg ) + reg)
 
 #define R128_DEREF(reg)		*(volatile u32 *)R128_ADDR( reg )
-#define R128_READ(reg)		R128_DEREF( reg )
+#define R128_READ(reg)		le32_to_cpu( R128_DEREF( reg ) )
-#define R128_WRITE(reg,val)	do { R128_DEREF( reg ) = val; } while (0)
+#define R128_WRITE(reg,val)	do { R128_DEREF( reg ) = cpu_to_le32( val ); } while (0)
 
 #define R128_DEREF8(reg)	*(volatile u8 *)R128_ADDR( reg )
 #define R128_READ8(reg)		R128_DEREF8( reg )
@@ -400,6 +403,10 @@ extern int R128_READ_PLL(drm_device_t *dev, int addr);
  * Misc helper macros
  */
 
+#define GET_RING_HEAD(ring)	le32_to_cpu( *(ring)->head )
+#define SET_RING_HEAD(ring,x)	*(ring)->head = cpu_to_le32( x )
+
+
 #define LOCK_TEST_WITH_RETURN( dev )					\
 do {									\
 	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||		\
@@ -415,7 +422,7 @@ do {									\
 	drm_r128_ring_buffer_t *ring = &dev_priv->ring; int i;		\
 	if ( ring->space < ring->high_mark ) {				\
 		for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {	\
-			ring->space = *ring->head - ring->tail;		\
+			ring->space = GET_RING_HEAD(ring) - ring->tail;	\
 			if ( ring->space <= 0 )				\
 				ring->space += ring->size;		\
 			if ( ring->space >= ring->high_mark )		\
@@ -498,7 +505,7 @@ do {									\
 		DRM_INFO( "   OUT_RING( 0x%08x ) at 0x%x\n",		\
 			   (unsigned int)(x), write );			\
 	}								\
-	ring[write++] = (x);						\
+	ring[write++] = cpu_to_le32( x );				\
 	write &= tail_mask;						\
 } while (0)
 

linux/r128_state.c

@@ -670,7 +670,7 @@ static void r128_cce_dispatch_indirect( drm_device_t *dev,
 			u32 *data = (u32 *)
 				((char *)dev_priv->buffers->handle
 				 + buf->offset + start);
-			data[dwords++] = R128_CCE_PACKET2;
+			data[dwords++] = cpu_to_le32( R128_CCE_PACKET2 );
 		}
 
 		buf_priv->dispatched = 1;
@@ -739,16 +739,16 @@ static void r128_cce_dispatch_indices( drm_device_t *dev,
 		data = (u32 *)((char *)dev_priv->buffers->handle
 			       + buf->offset + start);
 
-		data[0] = CCE_PACKET3( R128_3D_RNDR_GEN_INDX_PRIM, dwords-2 );
+		data[0] = cpu_to_le32( CCE_PACKET3( R128_3D_RNDR_GEN_INDX_PRIM, dwords-2 ) );
 
-		data[1] = offset;
+		data[1] = cpu_to_le32( offset );
-		data[2] = R128_MAX_VB_VERTS;
+		data[2] = cpu_to_le32( R128_MAX_VB_VERTS );
-		data[3] = format;
+		data[3] = cpu_to_le32( format );
-		data[4] = (prim | R128_CCE_VC_CNTL_PRIM_WALK_IND |
+		data[4] = cpu_to_le32( (prim | R128_CCE_VC_CNTL_PRIM_WALK_IND |
-			   (count << 16));
+			   (count << 16)) );
 
 		if ( count & 0x1 ) {
-			data[dwords-1] &= 0x0000ffff;
+			data[dwords-1] &= cpu_to_le32( 0x0000ffff );
 		}
 
 		do {
@@ -856,8 +856,8 @@ static int r128_cce_dispatch_blit( drm_device_t *dev,
 
 	data = (u32 *)((char *)dev_priv->buffers->handle + buf->offset);
 
-	data[0] = CCE_PACKET3( R128_CNTL_HOSTDATA_BLT, dwords + 6 );
+	data[0] = cpu_to_le32( CCE_PACKET3( R128_CNTL_HOSTDATA_BLT, dwords + 6 ) );
-	data[1] = (R128_GMC_DST_PITCH_OFFSET_CNTL |
+	data[1] = cpu_to_le32( (R128_GMC_DST_PITCH_OFFSET_CNTL |
 		   R128_GMC_BRUSH_NONE |
 		   (blit->format << 8) |
 		   R128_GMC_SRC_DATATYPE_COLOR |
@@ -865,14 +865,14 @@ static int r128_cce_dispatch_blit( drm_device_t *dev,
 		   R128_DP_SRC_SOURCE_HOST_DATA |
 		   R128_GMC_CLR_CMP_CNTL_DIS |
 		   R128_GMC_AUX_CLIP_DIS |
-		   R128_GMC_WR_MSK_DIS);
+		   R128_GMC_WR_MSK_DIS) );
 
-	data[2] = (blit->pitch << 21) | (blit->offset >> 5);
+	data[2] = cpu_to_le32( (blit->pitch << 21) | (blit->offset >> 5) );
 	data[3] = 0xffffffff;
 	data[4] = 0xffffffff;
-	data[5] = (blit->y << 16) | blit->x;
+	data[5] = cpu_to_le32( (blit->y << 16) | blit->x );
-	data[6] = (blit->height << 16) | blit->width;
+	data[6] = cpu_to_le32( (blit->height << 16) | blit->width );
-	data[7] = dwords;
+	data[7] = cpu_to_le32( dwords );
 
 	buf->used = (dwords + 8) * sizeof(u32);