drm/nv50: allocate page for unknown PFB object in nv50_fb.c
authorBen Skeggs <bskeggs@redhat.com>
Mon, 15 Nov 2010 02:50:50 +0000 (12:50 +1000)
committerBen Skeggs <bskeggs@redhat.com>
Fri, 3 Dec 2010 05:11:50 +0000 (15:11 +1000)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drivers/gpu/drm/nouveau/nouveau_drv.h
drivers/gpu/drm/nouveau/nv50_fb.c

index 822cd40..e1619c6 100644 (file)
@@ -311,6 +311,7 @@ struct nouveau_timer_engine {
 struct nouveau_fb_engine {
        int num_tiles;
        struct drm_mm tag_heap;
+       void *priv;
 
        int  (*init)(struct drm_device *dev);
        void (*takedown)(struct drm_device *dev);
index d745c95..aa6d783 100644 (file)
@@ -3,16 +3,58 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 
+struct nv50_fb_priv {
+       struct page *r100c08_page;
+       dma_addr_t r100c08;
+};
+
+static int
+nv50_fb_create(struct drm_device *dev)
+{
+       struct drm_nouveau_private *dev_priv = dev->dev_private;
+       struct nv50_fb_priv *priv;
+
+       priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+       if (!priv)
+               return -ENOMEM;
+
+       priv->r100c08_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+       if (!priv->r100c08_page) {
+               kfree(priv);
+               return -ENOMEM;
+       }
+
+       priv->r100c08 = pci_map_page(dev->pdev, priv->r100c08_page, 0,
+                                    PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+       if (pci_dma_mapping_error(dev->pdev, priv->r100c08)) {
+               __free_page(priv->r100c08_page);
+               kfree(priv);
+               return -EFAULT;
+       }
+
+       dev_priv->engine.fb.priv = priv;
+       return 0;
+}
+
 int
 nv50_fb_init(struct drm_device *dev)
 {
        struct drm_nouveau_private *dev_priv = dev->dev_private;
+       struct nv50_fb_priv *priv;
+       int ret;
+
+       if (!dev_priv->engine.fb.priv) {
+               ret = nv50_fb_create(dev);
+               if (ret)
+                       return ret;
+       }
+       priv = dev_priv->engine.fb.priv;
 
        /* Not a clue what this is exactly.  Without pointing it at a
         * scratch page, VRAM->GART blits with M2MF (as in DDX DFS)
         * cause IOMMU "read from address 0" errors (rh#561267)
         */
-       nv_wr32(dev, 0x100c08, dev_priv->gart_info.sg_dummy_bus >> 8);
+       nv_wr32(dev, 0x100c08, priv->r100c08 >> 8);
 
        /* This is needed to get meaningful information from 100c90
         * on traps. No idea what these values mean exactly. */
@@ -36,6 +78,18 @@ nv50_fb_init(struct drm_device *dev)
 void
 nv50_fb_takedown(struct drm_device *dev)
 {
+       struct drm_nouveau_private *dev_priv = dev->dev_private;
+       struct nv50_fb_priv *priv;
+
+       priv = dev_priv->engine.fb.priv;
+       if (!priv)
+               return;
+       dev_priv->engine.fb.priv = NULL;
+
+       pci_unmap_page(dev->pdev, priv->r100c08, PAGE_SIZE,
+                      PCI_DMA_BIDIRECTIONAL);
+       __free_page(priv->r100c08_page);
+       kfree(priv);
 }
 
 void