mach64: make buffer emission macros normal functions

2 files changed, 239 insertions, 221 deletions

diff --git a/shared-core/mach64_dma.c b/shared-core/mach64_dma.c
index dd3547f8..9aa3f768 100644
--- a/shared-core/mach64_dma.c
+++ b/shared-core/mach64_dma.c
@@ -559,6 +559,233 @@ void mach64_dump_ring_info(drm_mach64_private_t * dev_priv)
 
 
 /*******************************************************************/
+/** \name DMA descriptor ring macros */
+/*@{*/
+
+static __inline__ void mach64_set_dma_eol(volatile u32 * addr)
+{
+#if defined(__i386__)
+	int nr = 31;
+
+	/* Taken from include/asm-i386/bitops.h linux header */
+	__asm__ __volatile__("lock;" "btsl %1,%0":"=m"(*addr)
+			     :"Ir"(nr));
+#elif defined(__powerpc__)
+	u32 old;
+	u32 mask = cpu_to_le32(MACH64_DMA_EOL);
+
+	/* Taken from the include/asm-ppc/bitops.h linux header */
+	__asm__ __volatile__("\n\
+1:	lwarx	%0,0,%3 \n\
+	or	%0,%0,%2 \n\
+	stwcx.	%0,0,%3 \n\
+	bne-	1b":"=&r"(old), "=m"(*addr)
+			     :"r"(mask), "r"(addr), "m"(*addr)
+			     :"cc");
+#elif defined(__alpha__)
+	u32 temp;
+	u32 mask = MACH64_DMA_EOL;
+
+	/* Taken from the include/asm-alpha/bitops.h linux header */
+	__asm__ __volatile__("1:	ldl_l %0,%3\n"
+			     "	bis %0,%2,%0\n"
+			     "	stl_c %0,%1\n"
+			     "	beq %0,2f\n"
+			     ".subsection 2\n"
+			     "2:	br 1b\n"
+			     ".previous":"=&r"(temp), "=m"(*addr)
+			     :"Ir"(mask), "m"(*addr));
+#else
+	u32 mask = cpu_to_le32(MACH64_DMA_EOL);
+
+	*addr |= mask;
+#endif
+}
+
+static __inline__ void mach64_clear_dma_eol(volatile u32 * addr)
+{
+#if defined(__i386__)
+	int nr = 31;
+
+	/* Taken from include/asm-i386/bitops.h linux header */
+	__asm__ __volatile__("lock;" "btrl %1,%0":"=m"(*addr)
+			     :"Ir"(nr));
+#elif defined(__powerpc__)
+	u32 old;
+	u32 mask = cpu_to_le32(MACH64_DMA_EOL);
+
+	/* Taken from the include/asm-ppc/bitops.h linux header */
+	__asm__ __volatile__("\n\
+1:	lwarx	%0,0,%3 \n\
+	andc	%0,%0,%2 \n\
+	stwcx.	%0,0,%3 \n\
+	bne-	1b":"=&r"(old), "=m"(*addr)
+			     :"r"(mask), "r"(addr), "m"(*addr)
+			     :"cc");
+#elif defined(__alpha__)
+	u32 temp;
+	u32 mask = ~MACH64_DMA_EOL;
+
+	/* Taken from the include/asm-alpha/bitops.h linux header */
+	__asm__ __volatile__("1:	ldl_l %0,%3\n"
+			     "	and %0,%2,%0\n"
+			     "	stl_c %0,%1\n"
+			     "	beq %0,2f\n"
+			     ".subsection 2\n"
+			     "2:	br 1b\n"
+			     ".previous":"=&r"(temp), "=m"(*addr)
+			     :"Ir"(mask), "m"(*addr));
+#else
+	u32 mask = cpu_to_le32(~MACH64_DMA_EOL);
+
+	*addr &= mask;
+#endif
+}
+
+#define RING_LOCALS									\
+	int _ring_tail, _ring_write; unsigned int _ring_mask; volatile u32 *_ring
+
+#define RING_WRITE_OFS  _ring_write
+
+#define BEGIN_RING( n )									\
+do {											\
+	if ( MACH64_VERBOSE ) {								\
+		DRM_INFO( "BEGIN_RING( %d ) in %s\n",					\
+			   (n), __FUNCTION__ );						\
+	}										\
+	if ( dev_priv->ring.space <= (n) * sizeof(u32) ) {				\
+		int ret;								\
+		if ((ret=mach64_wait_ring( dev_priv, (n) * sizeof(u32))) < 0 ) {	\
+			DRM_ERROR( "wait_ring failed, resetting engine\n");		\
+			mach64_dump_engine_info( dev_priv );				\
+			mach64_do_engine_reset( dev_priv );				\
+			return ret;							\
+		}									\
+	}										\
+	dev_priv->ring.space -= (n) * sizeof(u32);					\
+	_ring = (u32 *) dev_priv->ring.start;						\
+	_ring_tail = _ring_write = dev_priv->ring.tail;					\
+	_ring_mask = dev_priv->ring.tail_mask;						\
+} while (0)
+
+#define OUT_RING( x )						\
+do {								\
+	if ( MACH64_VERBOSE ) {					\
+		DRM_INFO( "   OUT_RING( 0x%08x ) at 0x%x\n",	\
+			   (unsigned int)(x), _ring_write );	\
+	}							\
+	_ring[_ring_write++] = cpu_to_le32( x );		\
+	_ring_write &= _ring_mask;				\
+} while (0)
+
+#define ADVANCE_RING()							\
+do {									\
+	if ( MACH64_VERBOSE ) {						\
+		DRM_INFO( "ADVANCE_RING() wr=0x%06x tail=0x%06x\n",	\
+			  _ring_write, _ring_tail );			\
+	}								\
+	DRM_MEMORYBARRIER();						\
+	mach64_clear_dma_eol( &_ring[(_ring_tail - 2) & _ring_mask] );	\
+	DRM_MEMORYBARRIER();						\
+	dev_priv->ring.tail = _ring_write;				\
+	mach64_ring_tick( dev_priv, &(dev_priv)->ring );		\
+} while (0)
+
+
+int mach64_add_buf_to_ring(drm_mach64_private_t *dev_priv,
+                           drm_mach64_freelist_t *entry)
+{
+	int bytes, pages, remainder;
+	u32 address, page;
+	int i;
+	struct drm_buf *buf = entry->buf;
+	RING_LOCALS;
+
+	bytes = buf->used;
+	address = GETBUFADDR( buf );
+	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;
+
+	BEGIN_RING( pages * 4 );
+
+	for ( i = 0 ; i < pages-1 ; i++ ) {
+		page = address + i * MACH64_DMA_CHUNKSIZE;
+		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );
+		OUT_RING( page );
+		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );
+		OUT_RING( 0 );
+	}
+
+	/* generate the final descriptor for any remaining commands in this buffer */
+	page = address + i * MACH64_DMA_CHUNKSIZE;
+	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;
+
+	/* Save dword offset of last descriptor for this buffer.
+	 * This is needed to check for completion of the buffer in freelist_get
+	 */
+	entry->ring_ofs = RING_WRITE_OFS;
+
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );
+	OUT_RING( page );
+	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );
+	OUT_RING( 0 );
+
+	ADVANCE_RING();
+	
+	return 0;
+}
+
+int mach64_add_hostdata_buf_to_ring(drm_mach64_private_t *dev_priv,
+                                    drm_mach64_freelist_t *entry)
+{
+	int bytes, pages, remainder;
+	u32 address, page;
+	int i;
+	struct drm_buf *buf = entry->buf;
+	RING_LOCALS;
+	
+	bytes = buf->used - MACH64_HOSTDATA_BLIT_OFFSET;
+	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;
+	address = GETBUFADDR( buf );
+	
+	BEGIN_RING( 4 + pages * 4 );
+	
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );
+	OUT_RING( address );
+	OUT_RING( MACH64_HOSTDATA_BLIT_OFFSET | MACH64_DMA_HOLD_OFFSET );
+	OUT_RING( 0 );
+	address += MACH64_HOSTDATA_BLIT_OFFSET;
+	
+	for ( i = 0 ; i < pages-1 ; i++ ) {
+		page = address + i * MACH64_DMA_CHUNKSIZE;
+		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );
+		OUT_RING( page );
+		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );
+		OUT_RING( 0 );
+	}
+	
+	/* generate the final descriptor for any remaining commands in this buffer */
+	page = address + i * MACH64_DMA_CHUNKSIZE;
+	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;
+	
+	/* Save dword offset of last descriptor for this buffer.
+	 * This is needed to check for completion of the buffer in freelist_get
+	 */
+	entry->ring_ofs = RING_WRITE_OFS;
+	
+	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );
+	OUT_RING( page );
+	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );
+	OUT_RING( 0 );
+	
+	ADVANCE_RING();
+	
+	return 0;
+}
+
+/*@}*/
+
+
+/*******************************************************************/
 /** \name DMA test and initialization */
 /*@{*/
 
diff --git a/shared-core/mach64_drv.h b/shared-core/mach64_drv.h
index 347c9426..7bd40a68 100644
--- a/shared-core/mach64_drv.h
+++ b/shared-core/mach64_drv.h
@@ -140,6 +140,11 @@ extern void mach64_dump_engine_info(drm_mach64_private_t * dev_priv);
 extern void mach64_dump_ring_info(drm_mach64_private_t * dev_priv);
 extern int mach64_do_engine_reset(drm_mach64_private_t * dev_priv);
 
+extern int mach64_add_buf_to_ring(drm_mach64_private_t *dev_priv,
+                                  drm_mach64_freelist_t *_entry);
+extern int mach64_add_hostdata_buf_to_ring(drm_mach64_private_t *dev_priv,
+                                           drm_mach64_freelist_t *_entry);
+
 extern int mach64_do_dma_idle(drm_mach64_private_t * dev_priv);
 extern int mach64_do_dma_flush(drm_mach64_private_t * dev_priv);
 extern int mach64_do_cleanup_dma(struct drm_device * dev);
@@ -521,89 +526,9 @@ extern void mach64_driver_irq_uninstall(struct drm_device * dev);
 #define MACH64_APERTURE_OFFSET	        0x7ff800	/* frame-buffer offset for gui-masters */
 
 /* ================================================================
- * Misc helper macros
+ * Ring operations
  */
 
-static __inline__ void mach64_set_dma_eol(volatile u32 * addr)
-{
-#if defined(__i386__)
-	int nr = 31;
-
-	/* Taken from include/asm-i386/bitops.h linux header */
-	__asm__ __volatile__("lock;" "btsl %1,%0":"=m"(*addr)
-			     :"Ir"(nr));
-#elif defined(__powerpc__)
-	u32 old;
-	u32 mask = cpu_to_le32(MACH64_DMA_EOL);
-
-	/* Taken from the include/asm-ppc/bitops.h linux header */
-	__asm__ __volatile__("\n\
-1:	lwarx	%0,0,%3 \n\
-	or	%0,%0,%2 \n\
-	stwcx.	%0,0,%3 \n\
-	bne-	1b":"=&r"(old), "=m"(*addr)
-			     :"r"(mask), "r"(addr), "m"(*addr)
-			     :"cc");
-#elif defined(__alpha__)
-	u32 temp;
-	u32 mask = MACH64_DMA_EOL;
-
-	/* Taken from the include/asm-alpha/bitops.h linux header */
-	__asm__ __volatile__("1:	ldl_l %0,%3\n"
-			     "	bis %0,%2,%0\n"
-			     "	stl_c %0,%1\n"
-			     "	beq %0,2f\n"
-			     ".subsection 2\n"
-			     "2:	br 1b\n"
-			     ".previous":"=&r"(temp), "=m"(*addr)
-			     :"Ir"(mask), "m"(*addr));
-#else
-	u32 mask = cpu_to_le32(MACH64_DMA_EOL);
-
-	*addr |= mask;
-#endif
-}
-
-static __inline__ void mach64_clear_dma_eol(volatile u32 * addr)
-{
-#if defined(__i386__)
-	int nr = 31;
-
-	/* Taken from include/asm-i386/bitops.h linux header */
-	__asm__ __volatile__("lock;" "btrl %1,%0":"=m"(*addr)
-			     :"Ir"(nr));
-#elif defined(__powerpc__)
-	u32 old;
-	u32 mask = cpu_to_le32(MACH64_DMA_EOL);
-
-	/* Taken from the include/asm-ppc/bitops.h linux header */
-	__asm__ __volatile__("\n\
-1:	lwarx	%0,0,%3 \n\
-	andc	%0,%0,%2 \n\
-	stwcx.	%0,0,%3 \n\
-	bne-	1b":"=&r"(old), "=m"(*addr)
-			     :"r"(mask), "r"(addr), "m"(*addr)
-			     :"cc");
-#elif defined(__alpha__)
-	u32 temp;
-	u32 mask = ~MACH64_DMA_EOL;
-
-	/* Taken from the include/asm-alpha/bitops.h linux header */
-	__asm__ __volatile__("1:	ldl_l %0,%3\n"
-			     "	and %0,%2,%0\n"
-			     "	stl_c %0,%1\n"
-			     "	beq %0,2f\n"
-			     ".subsection 2\n"
-			     "2:	br 1b\n"
-			     ".previous":"=&r"(temp), "=m"(*addr)
-			     :"Ir"(mask), "m"(*addr));
-#else
-	u32 mask = cpu_to_le32(~MACH64_DMA_EOL);
-
-	*addr &= mask;
-#endif
-}
-
 static __inline__ void mach64_ring_start(drm_mach64_private_t * dev_priv)
 {
 	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
@@ -750,59 +675,6 @@ mach64_update_ring_snapshot(drm_mach64_private_t * dev_priv)
 }
 
 /* ================================================================
- * DMA descriptor ring macros
- */
-
-#define RING_LOCALS									\
-	int _ring_tail, _ring_write; unsigned int _ring_mask; volatile u32 *_ring
-
-#define RING_WRITE_OFS  _ring_write
-
-#define BEGIN_RING( n )									\
-do {											\
-	if ( MACH64_VERBOSE ) {								\
-		DRM_INFO( "BEGIN_RING( %d ) in %s\n",					\
-			   (n), __FUNCTION__ );						\
-	}										\
-	if ( dev_priv->ring.space <= (n) * sizeof(u32) ) {				\
-		int ret;								\
-		if ((ret=mach64_wait_ring( dev_priv, (n) * sizeof(u32))) < 0 ) {	\
-			DRM_ERROR( "wait_ring failed, resetting engine\n");		\
-			mach64_dump_engine_info( dev_priv );				\
-			mach64_do_engine_reset( dev_priv );				\
-			return ret;							\
-		}									\
-	}										\
-	dev_priv->ring.space -= (n) * sizeof(u32);					\
-	_ring = (u32 *) dev_priv->ring.start;						\
-	_ring_tail = _ring_write = dev_priv->ring.tail;					\
-	_ring_mask = dev_priv->ring.tail_mask;						\
-} while (0)
-
-#define OUT_RING( x )						\
-do {								\
-	if ( MACH64_VERBOSE ) {					\
-		DRM_INFO( "   OUT_RING( 0x%08x ) at 0x%x\n",	\
-			   (unsigned int)(x), _ring_write );	\
-	}							\
-	_ring[_ring_write++] = cpu_to_le32( x );		\
-	_ring_write &= _ring_mask;				\
-} while (0)
-
-#define ADVANCE_RING()							\
-do {									\
-	if ( MACH64_VERBOSE ) {						\
-		DRM_INFO( "ADVANCE_RING() wr=0x%06x tail=0x%06x\n",	\
-			  _ring_write, _ring_tail );			\
-	}								\
-	DRM_MEMORYBARRIER();						\
-	mach64_clear_dma_eol( &_ring[(_ring_tail - 2) & _ring_mask] );	\
-	DRM_MEMORYBARRIER();						\
-	dev_priv->ring.tail = _ring_write;				\
-	mach64_ring_tick( dev_priv, &(dev_priv)->ring );		\
-} while (0)
-
-/* ================================================================
  * DMA macros
  */
 
@@ -889,7 +761,7 @@ do {								\
 #define DMAADVANCE( dev_priv, _discard )						     \
 do {											     \
 	struct list_head *ptr;								     \
-	RING_LOCALS;									     \
+	int ret;									     \
 											     \
 	if ( MACH64_VERBOSE ) {								     \
 		DRM_INFO( "DMAADVANCE() in %s\n", __FUNCTION__ );			     \
@@ -902,7 +774,6 @@ do {											     \
 	}										     \
 	if (_buf->pending) {								     \
                 /* This is a resued buffer, so we need to find it in the pending list */     \
-		int ret;								     \
 		if ( (ret=mach64_find_pending_buf_entry(dev_priv, &_entry, _buf)) ) {	     \
 			DRM_ERROR( "DMAADVANCE() in %s: couldn't find pending buf %d\n",     \
 				   __FUNCTION__, _buf->idx );				     \
@@ -927,7 +798,8 @@ do {											     \
 		list_add_tail(ptr, &dev_priv->pending);					     \
 	}										     \
 	_entry->discard = (_discard);							     \
-	ADD_BUF_TO_RING( dev_priv );							     \
+	if ( (ret = mach64_add_buf_to_ring( dev_priv, _entry )) )			     \
+		return ret;								     \
 } while (0)
 
 #define DMADISCARDBUF()									\
@@ -943,48 +815,10 @@ do {											\
 	_entry->discard = 1;								\
 } while(0)
 
-#define ADD_BUF_TO_RING( dev_priv )							\
-do {											\
-	int bytes, pages, remainder;							\
-	u32 address, page;								\
-	int i;										\
-											\
-	bytes = _buf->used;								\
-	address = GETBUFADDR( _buf );							\
-											\
-	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;		\
-											\
-	BEGIN_RING( pages * 4 );							\
-											\
-	for ( i = 0 ; i < pages-1 ; i++ ) {						\
-		page = address + i * MACH64_DMA_CHUNKSIZE;				\
-		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );			\
-		OUT_RING( page );							\
-		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );		\
-		OUT_RING( 0 );								\
-	}										\
-											\
-	/* generate the final descriptor for any remaining commands in this buffer */	\
-	page = address + i * MACH64_DMA_CHUNKSIZE;					\
-	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;					\
-											\
-	/* Save dword offset of last descriptor for this buffer.			\
-	 * This is needed to check for completion of the buffer in freelist_get		\
-	 */										\
-	_entry->ring_ofs = RING_WRITE_OFS;						\
-											\
-	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );				\
-	OUT_RING( page );								\
-	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );		\
-	OUT_RING( 0 );									\
-											\
-	ADVANCE_RING();									\
-} while(0)
-
 #define DMAADVANCEHOSTDATA( dev_priv )							\
 do {											\
 	struct list_head *ptr;								\
-	RING_LOCALS;									\
+	int ret;									\
 											\
 	if ( MACH64_VERBOSE ) {								\
 		DRM_INFO( "DMAADVANCEHOSTDATA() in %s\n", __FUNCTION__ );		\
@@ -1008,51 +842,8 @@ do {											\
 	_entry->buf->pending = 1;							\
 	list_add_tail(ptr, &dev_priv->pending);						\
 	_entry->discard = 1;								\
-	ADD_HOSTDATA_BUF_TO_RING( dev_priv );						\
+	if ( (ret = mach64_add_hostdata_buf_to_ring( dev_priv, _entry )) )		\
+		return ret;								\
 } while (0)
 
-#define ADD_HOSTDATA_BUF_TO_RING( dev_priv )						 \
-do {											 \
-	int bytes, pages, remainder;							 \
-	u32 address, page;								 \
-	int i;										 \
-											 \
-	bytes = _buf->used - MACH64_HOSTDATA_BLIT_OFFSET;				 \
-	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;		 \
-	address = GETBUFADDR( _buf );							 \
-											 \
-	BEGIN_RING( 4 + pages * 4 );							 \
-											 \
-	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );				 \
-	OUT_RING( address );								 \
-	OUT_RING( MACH64_HOSTDATA_BLIT_OFFSET | MACH64_DMA_HOLD_OFFSET );		 \
-	OUT_RING( 0 );									 \
-											 \
-	address += MACH64_HOSTDATA_BLIT_OFFSET;						 \
-											 \
-	for ( i = 0 ; i < pages-1 ; i++ ) {						 \
-		page = address + i * MACH64_DMA_CHUNKSIZE;				 \
-		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );		 \
-		OUT_RING( page );							 \
-		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );		 \
-		OUT_RING( 0 );								 \
-	}										 \
-											 \
-	/* generate the final descriptor for any remaining commands in this buffer */	 \
-	page = address + i * MACH64_DMA_CHUNKSIZE;					 \
-	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;					 \
-											 \
-	/* Save dword offset of last descriptor for this buffer.			 \
-	 * This is needed to check for completion of the buffer in freelist_get		 \
-	 */										 \
-	_entry->ring_ofs = RING_WRITE_OFS;						 \
-											 \
-	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );			 \
-	OUT_RING( page );								 \
-	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );		 \
-	OUT_RING( 0 );									 \
-											 \
-	ADVANCE_RING();									 \
-} while(0)
-
 #endif				/* __MACH64_DRV_H__ */