r200/r300: add aperture space checks

5 files changed, 213 insertions, 39 deletions

diff --git a/src/mesa/drivers/dri/radeon/common_misc.c b/src/mesa/drivers/dri/radeon/common_misc.c
index 19294487f6..1adcefb19a 100644
--- a/src/mesa/drivers/dri/radeon/common_misc.c
+++ b/src/mesa/drivers/dri/radeon/common_misc.c
@@ -685,6 +685,14 @@ void rcommonInitCmdBuf(radeonContextPtr rmesa, int max_state_size)
 	assert(rmesa->cmdbuf.cs != NULL);
 	rmesa->cmdbuf.size = size;
 
+	if (!rmesa->radeonScreen->kernel_mm) {
+		radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, rmesa->radeonScreen->texSize[0]);
+		radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, rmesa->radeonScreen->gartTextures.size);
+	} else {
+		radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, rmesa->radeonScreen->texSize[0]);
+		radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, rmesa->radeonScreen->gartTextures.size);
+	}
+
 }
 /**
  * Destroy the command buffer
@@ -907,7 +915,7 @@ void radeonCleanupContext(radeonContextPtr radeon)
 	}
 }
 
-void
+static void
 radeon_make_kernel_renderbuffer_current(radeonContextPtr radeon,
 					GLframebuffer *draw)
 {
@@ -1314,7 +1322,6 @@ void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
 {
 	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 	uint32_t *out;
-	uint32_t bo_size;
 
 	if (stride == 0) {
 		radeonAllocDmaRegion(rmesa, &aos->bo, &aos->offset, size * 4, 32);
@@ -1328,7 +1335,6 @@ void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
 	aos->components = size;
 	aos->count = count;
 
-//	radeon_bo_map(aos->bo, 1);
 	out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
 	switch (size) {
 	case 1: radeonEmitVec4(out, data, stride, count); break;
@@ -1339,7 +1345,6 @@ void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
 		assert(0);
 		break;
 	}
-//	radeon_bo_unmap(aos->bo);
 }
 
 
@@ -2320,6 +2325,9 @@ void radeonSpanRenderFinish(GLcontext * ctx)
 
 void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size)
 {
+	struct radeon_cs_space_check bos[1];
+	int flushed, ret;
+
 	size = MAX2(size, MAX_DMA_BUF_SZ * 16);
 
 	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
@@ -2330,8 +2338,6 @@ void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size)
 		rmesa->dma.flush(rmesa->glCtx);
 	}
 
-
-
 	if (rmesa->dma.nr_released_bufs > 4) {
 		rcommonFlushCmdBuf(rmesa, __FUNCTION__);
 		rmesa->dma.nr_released_bufs = 0;
@@ -2341,13 +2347,42 @@ void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size)
 		radeon_bo_unref(rmesa->dma.current);
 		rmesa->dma.current = 0;
 	}
-	
+
+again_alloc:	
 	rmesa->dma.current = radeon_bo_open(rmesa->radeonScreen->bom,
 					    0, size, 4, RADEON_GEM_DOMAIN_GTT,
 					    0);
 
+	if (!rmesa->dma.current) {
+		rcommonFlushCmdBuf(rmesa, __FUNCTION__);
+		rmesa->dma.nr_released_bufs = 0;
+		goto again_alloc;
+	}
+
 	rmesa->dma.current_used = 0;
 	rmesa->dma.current_vertexptr = 0;
+	
+	bos[0].bo = rmesa->dma.current;
+	bos[0].read_domains = RADEON_GEM_DOMAIN_GTT;
+	bos[0].write_domain =0 ;
+	bos[0].new_accounted = 0;
+
+again:
+	ret = radeon_cs_space_check(rmesa->cmdbuf.cs, bos, 1);
+	if (ret == RADEON_CS_SPACE_OP_TO_BIG) {
+		fprintf(stderr,"Got OPEARTION TO BIG ILLEGAL - this cannot happen");
+		assert(0);
+	} else if (ret == RADEON_CS_SPACE_FLUSH) {
+		rcommonFlushCmdBuf(rmesa, __FUNCTION__);
+		if (flushed) {
+			fprintf(stderr,"flushed but still no space\n");
+			assert(0);
+		}
+		flushed = 1;
+		goto again;
+	}
+		
+	
 	radeon_bo_map(rmesa->dma.current, 1);
 }
 
diff --git a/src/mesa/drivers/dri/radeon/common_misc.h b/src/mesa/drivers/dri/radeon/common_misc.h
index d17d1607db..44e464eb13 100644
--- a/src/mesa/drivers/dri/radeon/common_misc.h
+++ b/src/mesa/drivers/dri/radeon/common_misc.h
@@ -2,6 +2,7 @@
 #define COMMON_MISC_H
 
 #include "common_context.h"
+#include "radeon_buffer.h"
 void radeonRecalcScissorRects(radeonContextPtr radeon);
 void radeonSetCliprects(radeonContextPtr radeon);
 void radeonUpdateScissor( GLcontext *ctx );
@@ -122,4 +123,36 @@ void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size);
 void radeonAllocDmaRegion(radeonContextPtr rmesa,
 			  struct radeon_bo **pbo, int *poffset,
 			  int bytes, int alignment);
+void radeonReleaseDmaRegion(radeonContextPtr rmesa);
+
+void rcommon_flush_last_swtcl_prim(GLcontext *ctx);
+
+void *rcommonAllocDmaLowVerts(radeonContextPtr rmesa, int nverts, int vsize);
+
+
+static inline struct radeon_renderbuffer *radeon_get_depthbuffer(radeonContextPtr rmesa)
+{
+	struct radeon_renderbuffer *rrb;
+	rrb = rmesa->state.depth.rrb;
+	if (!rrb)
+		return NULL;
+
+	return rrb;
+}
+
+static inline struct radeon_renderbuffer *radeon_get_colorbuffer(radeonContextPtr rmesa)
+{
+	struct radeon_renderbuffer *rrb;
+	GLframebuffer *fb = rmesa->dri.drawable->driverPrivate;
+
+	rrb = rmesa->state.color.rrb;
+	if (rmesa->radeonScreen->driScreen->dri2.enabled) {
+		rrb = (struct radeon_renderbuffer *)fb->Attachment[BUFFER_BACK_LEFT].Renderbuffer;
+	}
+	if (!rrb)
+		return NULL;
+	return rrb;
+}
+
+
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c
index bd126c026c..f782d96900 100644
--- a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c
+++ b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c
@@ -413,10 +413,8 @@ static struct radeon_bo *bo_open(struct radeon_bo_manager *bom,
         r = bo_dma_alloc(&(bo_legacy->base));
         if (r) {
 	  if (legacy_wait_any_pending(boml) == -1) {
-	    fprintf(stderr, "Ran out of GART memory (for %d)!\n", size);
-            fprintf(stderr, "Please consider adjusting GARTSize option.\n");
             bo_free(bo_legacy);
-            exit(-1);
+	    return NULL;
 	  }
 	  goto retry;
 	  return NULL;
@@ -639,6 +637,24 @@ void radeon_bo_manager_legacy_dtor(struct radeon_bo_manager *bom)
     free(boml);
 }
 
+static struct bo_legacy *radeon_legacy_bo_alloc_static(struct bo_manager_legacy *bom,
+						       int size, uint32_t offset)
+{
+    struct bo_legacy *bo;
+
+    bo = bo_allocate(bom, size, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+    if (bo == NULL)
+	return NULL;
+    bo->static_bo = 1;
+    bo->offset = offset + bom->fb_location;
+    bo->base.handle = bo->offset;
+    bo->ptr = bom->screen->driScreen->pFB + offset;
+    if (bo->base.handle > bom->nhandle) {
+        bom->nhandle = bo->base.handle + 1;
+    }
+    return bo;
+}
+
 struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *scrn)
 {
     struct bo_manager_legacy *bom;
@@ -682,41 +698,30 @@ struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *sc
 
     /* biggest framebuffer size */
     size = 4096*4096*4; 
+
     /* allocate front */
-    bo = bo_allocate(bom, size, 0, RADEON_GEM_DOMAIN_VRAM, 0);
-    if (bo == NULL) {
+    bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->frontOffset);
+    if (!bo) {
         radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom);
         return NULL;
     }
     if (scrn->sarea->tiling_enabled) {
         bo->base.flags = RADEON_BO_FLAGS_MACRO_TILE;
     }
-    bo->static_bo = 1;
-    bo->offset = bom->screen->frontOffset + bom->fb_location;
-    bo->base.handle = bo->offset;
-    bo->ptr = scrn->driScreen->pFB + bom->screen->frontOffset;
-    if (bo->base.handle > bom->nhandle) {
-        bom->nhandle = bo->base.handle + 1;
-    }
+
     /* allocate back */
-    bo = bo_allocate(bom, size, 0, RADEON_GEM_DOMAIN_VRAM, 0);
-    if (bo == NULL) {
+    bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->backOffset);
+    if (!bo) {
         radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom);
         return NULL;
     }
     if (scrn->sarea->tiling_enabled) {
         bo->base.flags = RADEON_BO_FLAGS_MACRO_TILE;
     }
-    bo->static_bo = 1;
-    bo->offset = bom->screen->backOffset + bom->fb_location;
-    bo->base.handle = bo->offset;
-    bo->ptr = scrn->driScreen->pFB + bom->screen->backOffset;
-    if (bo->base.handle > bom->nhandle) {
-        bom->nhandle = bo->base.handle + 1;
-    }
+
     /* allocate depth */
-    bo = bo_allocate(bom, size, 0, RADEON_GEM_DOMAIN_VRAM, 0);
-    if (bo == NULL) {
+    bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->depthOffset);
+    if (!bo) {
         radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom);
         return NULL;
     }
@@ -725,13 +730,6 @@ struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *sc
         bo->base.flags |= RADEON_BO_FLAGS_MACRO_TILE;
         bo->base.flags |= RADEON_BO_FLAGS_MICRO_TILE;
     }
-    bo->static_bo = 1;
-    bo->offset = bom->screen->depthOffset + bom->fb_location;
-    bo->base.handle = bo->offset;
-    bo->ptr = scrn->driScreen->pFB + bom->screen->depthOffset;
-    if (bo->base.handle > bom->nhandle) {
-        bom->nhandle = bo->base.handle + 1;
-    }
     return (struct radeon_bo_manager*)bom;
 }
 
@@ -750,3 +748,10 @@ unsigned radeon_bo_legacy_relocs_size(struct radeon_bo *bo)
     }
     return bo->size;
 }
+
+int radeon_legacy_bo_is_static(struct radeon_bo *bo)
+{
+    struct bo_legacy *bo_legacy = (struct bo_legacy*)bo;
+    return bo_legacy->static_bo;
+}
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h
index 208171e372..575979cbec 100644
--- a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h
+++ b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h
@@ -44,4 +44,5 @@ void radeon_bo_manager_legacy_dtor(struct radeon_bo_manager *bom);
 void radeon_bo_legacy_texture_age(struct radeon_bo_manager *bom);
 unsigned radeon_bo_legacy_relocs_size(struct radeon_bo *bo);
 
+int radeon_legacy_bo_is_static(struct radeon_bo *bo);
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c
index 8997187d62..171818db37 100644
--- a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c
+++ b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.c
@@ -315,7 +315,7 @@ static int cs_emit(struct radeon_cs *cs)
         cmd.boxes = (drm_clip_rect_t *) csm->ctx->pClipRects;
     }
 
-    dump_cmdbuf(cs);
+    //    dump_cmdbuf(cs);
 
     r = drmCommandWrite(cs->csm->fd, DRM_RADEON_CMDBUF, &cmd, sizeof(cmd));
     if (r) {
@@ -330,6 +330,10 @@ static int cs_emit(struct radeon_cs *cs)
 	}
     }
     cs_set_age(cs);
+
+    cs->csm->read_used = 0;
+    cs->csm->vram_write_used = 0;
+    cs->csm->gart_write_used = 0;
     return 0;
 }
 
@@ -374,6 +378,101 @@ static void cs_print(struct radeon_cs *cs, FILE *file)
 {
 }
 
+static int cs_check_space(struct radeon_cs *cs, struct radeon_cs_space_check *bos, int num_bo)
+{
+    struct radeon_cs_manager *csm = cs->csm;
+    int this_op_read = 0, this_op_gart_write = 0, this_op_vram_write = 0;
+    uint32_t read_domains, write_domain;
+    int i;
+    struct radeon_bo *bo;
+
+    /* check the totals for this operation */
+
+    if (num_bo == 0)
+        return 0;
+
+    /* prepare */
+    for (i = 0; i < num_bo; i++) {
+      bo = bos[i].bo;
+
+      bos[i].new_accounted = 0;
+      read_domains = bos[i].read_domains;
+      write_domain = bos[i].write_domain;
+		
+      /* pinned bos don't count */
+      if (radeon_legacy_bo_is_static(bo))
+	  continue;
+ 
+      /* already accounted this bo */
+      if (write_domain && (write_domain == bo->space_accounted))
+	  continue;
+
+      if (read_domains && ((read_domains << 16) == bo->space_accounted))
+	  continue;
+      
+      if (bo->space_accounted == 0) {
+	  if (write_domain == RADEON_GEM_DOMAIN_VRAM)
+	      this_op_vram_write += bo->size;
+	  else if (write_domain == RADEON_GEM_DOMAIN_GTT)
+	      this_op_gart_write += bo->size;
+	  else
+	      this_op_read += bo->size;
+	  bos[i].new_accounted = (read_domains << 16) | write_domain;
+      } else {
+	  uint16_t old_read, old_write;
+	  
+	  old_read = bo->space_accounted >> 16;
+	  old_write = bo->space_accounted & 0xffff;
+
+	  if (write_domain && (old_read & write_domain)) {
+	      bos[i].new_accounted = write_domain;
+	      /* moving from read to a write domain */
+	      if (write_domain == RADEON_GEM_DOMAIN_VRAM) {
+		  this_op_read -= bo->size;
+		  this_op_vram_write += bo->size;
+	      } else if (write_domain == RADEON_GEM_DOMAIN_VRAM) {
+		  this_op_read -= bo->size;
+		  this_op_gart_write += bo->size;
+	      }
+	  } else if (read_domains & old_write) {
+	      bos[i].new_accounted = bo->space_accounted & 0xffff;
+	  } else {
+	      /* rewrite the domains */
+	      if (write_domain != old_write)
+		  fprintf(stderr,"WRITE DOMAIN RELOC FAILURE 0x%x %d %d\n", bo->handle, write_domain, old_write);
+	      if (read_domains != old_read)
+		  fprintf(stderr,"READ DOMAIN RELOC FAILURE 0x%x %d %d\n", bo->handle, read_domains, old_read);
+	      return RADEON_CS_SPACE_FLUSH;
+	  }
+      }
+	}
+	
+	if (this_op_read < 0)
+		this_op_read = 0;
+
+	/* check sizes - operation first */
+	if ((this_op_read + this_op_gart_write > csm->gart_limit) ||
+	    (this_op_vram_write > csm->vram_limit)) {
+	    return RADEON_CS_SPACE_OP_TO_BIG;
+	}
+
+	if (((csm->vram_write_used + this_op_vram_write) > csm->vram_limit) ||
+	    ((csm->read_used + csm->gart_write_used + this_op_gart_write + this_op_read) > csm->gart_limit)) {
+		return RADEON_CS_SPACE_FLUSH;
+	}
+
+	csm->gart_write_used += this_op_gart_write;
+	csm->vram_write_used += this_op_vram_write;
+	csm->read_used += this_op_read;
+	/* commit */
+	for (i = 0; i < num_bo; i++) {
+		bo = bos[i].bo;
+		bo->space_accounted = bos[i].new_accounted;
+	}
+
+	return RADEON_CS_SPACE_OK;
+}
+
 static struct radeon_cs_funcs  radeon_cs_legacy_funcs = {
     cs_create,
     cs_write_dword,
@@ -384,7 +483,8 @@ static struct radeon_cs_funcs  radeon_cs_legacy_funcs = {
     cs_destroy,
     cs_erase,
     cs_need_flush,
-    cs_print
+    cs_print,
+    cs_check_space
 };
 
 struct radeon_cs_manager *radeon_cs_manager_legacy_ctor(struct radeon_context *ctx)