Add the via PCI DMA blit code.

8 files changed, 1055 insertions, 17 deletions

diff --git a/linux-core/Makefile.kernel b/linux-core/Makefile.kernel
index b483a86a..5d4dac04 100644
--- a/linux-core/Makefile.kernel
+++ b/linux-core/Makefile.kernel
@@ -23,7 +23,7 @@ sis-objs    := sis_drv.o sis_ds.o sis_mm.o
 ffb-objs    := ffb_drv.o ffb_context.o
 savage-objs := savage_drv.o savage_bci.o savage_state.o
 via-objs    := via_irq.o via_drv.o via_ds.o via_map.o via_mm.o via_dma.o via_verifier.o \
-		via_video.o
+		via_video.o via_dmablit.o
 mach64-objs := mach64_drv.o mach64_dma.o mach64_irq.o mach64_state.o 
 
 ifeq ($(CONFIG_COMPAT),y)
diff --git a/linux-core/via_dmablit.c b/linux-core/via_dmablit.c
new file mode 100644
index 00000000..c37e78c8
--- /dev/null
+++ b/linux-core/via_dmablit.c
@@ -0,0 +1,788 @@
+/* via_dmablit.c -- PCI DMA BitBlt support for the VIA Unichrome/Pro
+ * 
+ * Copyright (C) 2005 Thomas Hellstrom, All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sub license,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, 
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: 
+ *    Thomas Hellstrom.
+ *    Partially based on code obtained from Digeo Inc.
+ */
+
+
+/*
+ * Unmaps the DMA mappings. 
+ * FIXME: Is this a NoOp on x86? Also 
+ * FIXME: What happens if this one is called and a pending blit has previously done 
+ * the same DMA mappings? 
+ */
+
+#include "drmP.h"
+#include "via_drm.h"
+#include "via_drv.h"
+#include "via_dmablit.h"
+
+#define VIA_PGDN(x)             (((unsigned long)(x)) & PAGE_MASK)
+#define VIA_PGOFF(x)            (((unsigned long)(x)) & ~PAGE_MASK)
+#define VIA_PFN(x)              ((unsigned long)(x) >> PAGE_SHIFT)
+
+typedef struct _drm_via_descriptor {
+	uint32_t mem_addr;
+	uint32_t dev_addr;
+	uint32_t size;
+	uint32_t next;
+} drm_via_descriptor_t;
+
+
+/*
+ * Unmap a DMA mapping.
+ */
+
+
+
+static void
+via_unmap_blit_from_device(struct pci_dev *pdev, drm_via_sg_info_t *vsg)
+{
+	int num_desc = vsg->num_desc;
+	unsigned cur_descriptor_page = num_desc / vsg->descriptors_per_page;
+	unsigned descriptor_this_page = num_desc % vsg->descriptors_per_page;
+	drm_via_descriptor_t *desc_ptr = vsg->desc_pages[cur_descriptor_page] + 
+		descriptor_this_page;
+	dma_addr_t next = vsg->chain_start;
+
+	while(num_desc--) {
+		if (descriptor_this_page-- == 0) {
+			cur_descriptor_page--;
+			descriptor_this_page = vsg->descriptors_per_page - 1;
+			desc_ptr = vsg->desc_pages[cur_descriptor_page] + 
+				descriptor_this_page;
+		}
+		dma_unmap_single(&pdev->dev, next, sizeof(*desc_ptr), DMA_FROM_DEVICE);
+		dma_unmap_page(&pdev->dev, desc_ptr->mem_addr, desc_ptr->size, vsg->direction);
+		next = (dma_addr_t) desc_ptr->next;
+		desc_ptr--;
+	}
+}
+
+/*
+ * If mode = 0, count how many descriptors are needed.
+ * If mode = 1, Map the DMA pages for the device, put together and map also the descriptors.
+ * Descriptors are run in reverse order by the hardware because we are not allowed to update the
+ * 'next' field without syncing calls when the descriptor is already mapped.
+ */
+
+static void
+via_map_blit_for_device(struct pci_dev *pdev,
+		   const drm_via_dmablit_t *xfer,
+		   drm_via_sg_info_t *vsg, 
+		   int mode)
+{
+	unsigned cur_descriptor_page = 0;
+	unsigned num_descriptors_this_page = 0;
+	unsigned char *mem_addr = xfer->mem_addr;
+	unsigned char *cur_mem;
+	unsigned char *first_addr = (unsigned char *)VIA_PGDN(mem_addr);
+	uint32_t fb_addr = xfer->fb_addr;
+	uint32_t cur_fb;
+	unsigned long line_len;
+	unsigned remaining_len;
+	int num_desc = 0;
+	int cur_line;
+	dma_addr_t next = 0 | VIA_DMA_DPR_EC;
+	drm_via_descriptor_t *desc_ptr = 0;
+
+	if (mode == 1) 
+		desc_ptr = vsg->desc_pages[cur_descriptor_page];
+
+	for (cur_line = 0; cur_line < xfer->num_lines; ++cur_line) {
+
+		line_len = xfer->line_length;
+		cur_fb = fb_addr;
+		cur_mem = mem_addr;
+		
+		while (line_len > 0) {
+
+                        remaining_len = min(PAGE_SIZE-VIA_PGOFF(cur_mem), line_len);
+			line_len -= remaining_len;
+
+			if (mode == 1) {
+                                desc_ptr->mem_addr = 
+					dma_map_page(&pdev->dev, 
+						     vsg->pages[VIA_PFN(cur_mem) - 
+								VIA_PFN(first_addr)],
+						     VIA_PGOFF(cur_mem), remaining_len, 
+						     vsg->direction);
+                                desc_ptr->dev_addr = cur_fb;
+				
+                                desc_ptr->size = remaining_len;
+				desc_ptr->next = (uint32_t) next;
+				next = dma_map_single(&pdev->dev, desc_ptr, sizeof(*desc_ptr), 
+						      DMA_TO_DEVICE);
+				desc_ptr++;
+				if (++num_descriptors_this_page >= vsg->descriptors_per_page) {
+					num_descriptors_this_page = 0;
+					desc_ptr = vsg->desc_pages[++cur_descriptor_page];
+				}
+			}
+			
+			num_desc++;
+			cur_mem += remaining_len;
+			cur_fb += remaining_len;
+		}
+		
+		mem_addr += xfer->mem_stride;
+		fb_addr += xfer->fb_stride;
+	}
+
+	if (mode == 1) {
+		vsg->chain_start = next;
+		vsg->state = dr_via_device_mapped;
+	}
+	vsg->num_desc = num_desc;
+}
+
+/*
+ * Function that frees up all resources for a blit. It is usable even if the 
+ * blit info has only be partially built as long as the status enum is consistent
+ * with the actual status of the used resources.
+ */
+
+
+void
+via_free_sg_info(struct pci_dev *pdev, drm_via_sg_info_t *vsg) 
+{
+	struct page *page;
+	int i;
+
+	switch(vsg->state) {
+	case dr_via_device_mapped:
+		via_unmap_blit_from_device(pdev, vsg);
+	case dr_via_desc_pages_alloc:
+		for (i=0; i<vsg->num_desc_pages; ++i) {
+			if (vsg->desc_pages[i] != NULL)
+			  free_page((unsigned long)vsg->desc_pages[i]);
+		}
+		kfree(vsg->desc_pages);
+	case dr_via_pages_locked:
+		for (i=0; i<vsg->num_pages; ++i) {
+			if ( NULL != (page = vsg->pages[i])) {
+				if (! PageReserved(page) && (DMA_FROM_DEVICE == vsg->direction)) 
+					SetPageDirty(page);
+				page_cache_release(page);
+			}
+		}
+	case dr_via_pages_alloc:
+		vfree(vsg->pages);
+	default:
+		vsg->state = dr_via_sg_init;
+	}
+	if (vsg->bounce_buffer) {
+		vfree(vsg->bounce_buffer);
+		vsg->bounce_buffer = NULL;
+	}
+	vsg->free_on_sequence = 0;
+}		
+
+/*
+ * Fire a blit engine.
+ */
+
+static void
+via_fire_dmablit(drm_device_t *dev, drm_via_sg_info_t *vsg, int engine)
+{
+	drm_via_private_t *dev_priv = (drm_via_private_t *)dev->dev_private;
+
+	VIA_WRITE(VIA_PCI_DMA_MAR0 + engine*0x10, 0);
+	VIA_WRITE(VIA_PCI_DMA_DAR0 + engine*0x10, 0);
+	VIA_WRITE(VIA_PCI_DMA_CSR0 + engine*0x04, VIA_DMA_CSR_DD | VIA_DMA_CSR_TD | 
+		  VIA_DMA_CSR_DE);
+	VIA_WRITE(VIA_PCI_DMA_MR0  + engine*0x04, VIA_DMA_MR_CM | VIA_DMA_MR_TDIE);
+	VIA_WRITE(VIA_PCI_DMA_BCR0 + engine*0x10, 0);
+	VIA_WRITE(VIA_PCI_DMA_DPR0 + engine*0x10, vsg->chain_start);
+	VIA_WRITE(VIA_PCI_DMA_CSR0 + engine*0x04, VIA_DMA_CSR_DE | VIA_DMA_CSR_TS);
+}
+
+/*
+ * Obtain a page pointer array and lock all pages into system memory. A segmentation violation will
+ * occur here if the calling user does not have access to the submitted address.
+ */
+
+static int
+via_lock_all_dma_pages(drm_via_sg_info_t *vsg,  drm_via_dmablit_t *xfer)
+{
+	int ret;
+	unsigned long first_pfn = VIA_PFN(xfer->mem_addr);
+	vsg->num_pages = VIA_PFN(xfer->mem_addr + (xfer->num_lines * xfer->mem_stride -1)) - 
+		first_pfn + 1;
+	
+	if (NULL == (vsg->pages = vmalloc(sizeof(struct page *) * vsg->num_pages)))
+		return DRM_ERR(ENOMEM);
+	memset(vsg->pages, 0, sizeof(struct page *) * vsg->num_pages);
+	down_read(&current->mm->mmap_sem);
+	ret = get_user_pages(current, current->mm, (unsigned long) xfer->mem_addr,
+			     vsg->num_pages, vsg->direction, 0, vsg->pages, NULL);
+
+	up_read(&current->mm->mmap_sem);
+	if (ret != vsg->num_pages) {
+		if (ret < 0) 
+			return ret;
+		vsg->state = dr_via_pages_locked;
+		return DRM_ERR(EINVAL);
+	}
+	vsg->state = dr_via_pages_locked;
+	DRM_DEBUG("DMA pages locked\n");
+	return 0;
+}
+
+/*
+ * Allocate DMA capable memory for the blit descriptor chain, and an array that keeps track of the
+ * pages we allocate. We don't want to use kmalloc for the descriptor chain because it may be
+ * quite large for some blits, and pages don't need to be contingous.
+ */
+
+static int 
+via_alloc_desc_pages(drm_via_sg_info_t *vsg)
+{
+	int i;
+	
+	vsg->descriptors_per_page = PAGE_SIZE / sizeof( drm_via_descriptor_t);
+	vsg->num_desc_pages = (vsg->num_desc + vsg->descriptors_per_page - 1) / 
+		vsg->descriptors_per_page;
+
+	if (NULL ==  (vsg->desc_pages = kmalloc(sizeof(void *) * vsg->num_desc_pages, GFP_KERNEL))) 
+		return DRM_ERR(ENOMEM);
+	
+	memset(vsg->desc_pages, 0, sizeof(void *) * vsg->num_desc_pages);
+	vsg->state = dr_via_desc_pages_alloc;
+	for (i=0; i<vsg->num_desc_pages; ++i) {
+		if (NULL == (vsg->desc_pages[i] = 
+			     (drm_via_descriptor_t *) __get_free_page(GFP_KERNEL)))
+			return DRM_ERR(ENOMEM);
+	}
+	DRM_DEBUG("Allocated %d pages for %d descriptors.\n", vsg->num_desc_pages,
+		  vsg->num_desc);
+	return 0;
+}
+			
+static void
+via_abort_dmablit(drm_device_t *dev, int engine)
+{
+	drm_via_private_t *dev_priv = (drm_via_private_t *)dev->dev_private;
+
+	VIA_WRITE(VIA_PCI_DMA_CSR0 + engine*0x04, VIA_DMA_CSR_TA);
+}
+
+static void
+via_dmablit_engine_off(drm_device_t *dev, int engine)
+{
+	drm_via_private_t *dev_priv = (drm_via_private_t *)dev->dev_private;
+
+	VIA_WRITE(VIA_PCI_DMA_CSR0 + engine*0x04, VIA_DMA_CSR_TD | VIA_DMA_CSR_DD); 
+}
+
+
+
+/*
+ * The dmablit part of the IRQ handler. Trying to do only reasonably fast things here.
+ * The rest, like unmapping and freeing memory for done blits is done in a separate workqueue
+ * task. Basically the task of the interrupt handler is to submit a new blit to the engine, while
+ * the workqueue task takes care of processing associated with the old blit.
+ */
+		
+void
+via_dmablit_handler(drm_device_t *dev, int engine, int from_irq)
+{
+	drm_via_private_t *dev_priv = (drm_via_private_t *)dev->dev_private;
+	drm_via_blitq_t *blitq = dev_priv->blit_queues + engine;
+	int cur;
+	int done_transfer;
+	unsigned long irqsave=0;
+	uint32_t status = 0;
+
+	DRM_DEBUG("DMA blit handler called. engine = %d, from_irq = %d, blitq = 0x%lx\n",
+		  engine, from_irq, (unsigned long) blitq);
+
+	if (from_irq) {
+		spin_lock(&blitq->blit_lock);
+	} else {
+		spin_lock_irqsave(&blitq->blit_lock, irqsave);
+	}
+
+	done_transfer = blitq->is_active && 
+	  (( status = VIA_READ(VIA_PCI_DMA_CSR0 + engine*0x04)) & VIA_DMA_CSR_TD);
+	done_transfer = done_transfer || ( blitq->aborting && !(status & VIA_DMA_CSR_DE)); 
+
+	cur = blitq->cur;
+	if (done_transfer) {
+
+		blitq->blits[cur]->aborted = blitq->aborting;
+		blitq->done_blit_handle++;
+		DRM_WAKEUP(blitq->blit_queue + cur);		
+
+		cur++;
+		if (cur >= VIA_NUM_BLIT_SLOTS) 
+			cur = 0;
+		blitq->cur = cur;
+
+		/*
+		 * Clear transfer done flag.
+		 */
+
+		VIA_WRITE(VIA_PCI_DMA_CSR0 + engine*0x04,  VIA_DMA_CSR_TD);
+
+		blitq->is_active = 0;
+		blitq->aborting = 0;
+		schedule_work(&blitq->wq);	
+
+	} else if (blitq->is_active && time_after_eq(jiffies, blitq->end)) {
+
+		/*
+		 * Abort transfer after one second.
+		 */
+
+		via_abort_dmablit(dev, engine);
+		blitq->aborting = 1;
+		blitq->end = jiffies + DRM_HZ;
+	}
+	  		
+	if (!blitq->is_active) {
+		if (blitq->num_outstanding) {
+			via_fire_dmablit(dev, blitq->blits[cur], engine);
+			blitq->is_active = 1;
+			blitq->cur = cur;
+			blitq->num_outstanding--;
+			blitq->end = jiffies + DRM_HZ;
+			if (!timer_pending(&blitq->poll_timer)) {
+				blitq->poll_timer.expires = jiffies+1;
+				add_timer(&blitq->poll_timer);
+			}
+		} else {
+			if (timer_pending(&blitq->poll_timer)) {
+				del_timer(&blitq->poll_timer);
+			}
+			via_dmablit_engine_off(dev, engine);
+		}
+	}		
+
+	if (from_irq) {
+		spin_unlock(&blitq->blit_lock);
+	} else {
+		spin_unlock_irqrestore(&blitq->blit_lock, irqsave);
+	}
+} 
+
+
+
+/*
+ * Check whether this blit is still active, performing necessary locking.
+ */
+
+static int
+via_dmablit_active(drm_via_blitq_t *blitq, int engine, uint32_t handle, wait_queue_head_t **queue)
+{
+	unsigned long irqsave;
+	uint32_t slot;
+	int active;
+
+	spin_lock_irqsave(&blitq->blit_lock, irqsave);
+
+	/*
+	 * Allow for handle wraparounds.
+	 */
+
+	active = ((blitq->done_blit_handle - handle) > (1 << 23)) &&
+		((blitq->cur_blit_handle - handle) <= (1 << 23));
+
+	if (queue && active) {
+		slot = handle - blitq->done_blit_handle + blitq->cur -1;
+		if (slot >= VIA_NUM_BLIT_SLOTS) {
+			slot -= VIA_NUM_BLIT_SLOTS;
+		}
+		*queue = blitq->blit_queue + slot;
+	}
+
+	spin_unlock_irqrestore(&blitq->blit_lock, irqsave);
+
+	return active;
+}
+	
+/*
+ * Sync. Wait for at least three seconds for the blit to be performed.
+ */
+
+static int
+via_dmablit_sync(drm_device_t *dev, uint32_t handle, int engine) 
+{
+
+	drm_via_private_t *dev_priv = (drm_via_private_t *)dev->dev_private;
+	drm_via_blitq_t *blitq = dev_priv->blit_queues + engine;
+	wait_queue_head_t *queue;
+	int ret = 0;
+
+	if (via_dmablit_active(blitq, engine, handle, &queue)) {
+		DRM_WAIT_ON(ret, *queue, 3 * DRM_HZ, 
+			    !via_dmablit_active(blitq, engine, handle, NULL));
+	}
+	DRM_DEBUG("DMA blit sync handle 0x%x engine %d returned %d\n",
+		  handle, engine, ret);
+	
+	return ret;
+}
+
+
+/*
+ * A timer that regularly polls the blit engine in cases where we don't have interrupts:
+ * a) Broken hardware (typically those that don't have any video capture facility).
+ * b) Blit abort. The hardware doesn't send an interrupt when a blit is aborted.
+ * The timer and hardware IRQ's can and do work in parallel. If the hardware has
+ * irqs, it will shorten the latency somewhat.
+ */
+
+
+
+static void
+via_dmablit_timer(unsigned long data)
+{
+	drm_via_blitq_t *blitq = (drm_via_blitq_t *) data;
+	drm_device_t *dev = blitq->dev;
+	int engine = (int)
+		(blitq - ((drm_via_private_t *)dev->dev_private)->blit_queues);
+		
+	DRM_DEBUG("Polling timer called for engine %d, jiffies %lu\n", engine, 
+		  (unsigned long) jiffies);
+
+	via_dmablit_handler(dev, engine, 0);
+	
+	if (!timer_pending(&blitq->poll_timer)) {
+		blitq->poll_timer.expires = jiffies+1;
+		add_timer(&blitq->poll_timer);
+	}
+	via_dmablit_handler(dev, engine, 0);
+
+}
+
+
+
+
+/*
+ * Workqueue task that frees data and mappings associated with a blit.
+ * Also wakes up waiting processes. Each of these tasks handles one
+ * blit engine only and may not be called on each interrupt.
+ */
+
+
+static void 
+via_dmablit_workqueue(void *data)
+{
+	drm_via_blitq_t *blitq = (drm_via_blitq_t *) data;
+	drm_device_t *dev = blitq->dev;
+	unsigned long irqsave;
+	drm_via_sg_info_t *cur_sg;
+	int cur_released;
+	
+	
+	DRM_DEBUG("Workqueue task called for blit engine %ld\n",(unsigned long) 
+		  (blitq - ((drm_via_private_t *)dev->dev_private)->blit_queues));
+
+	spin_lock_irqsave(&blitq->blit_lock, irqsave);
+	
+	while(blitq->serviced != blitq->cur) {
+
+		cur_released = blitq->serviced++;
+
+		DRM_DEBUG("Releasing blit slot %d\n", cur_released);
+
+		if (blitq->serviced >= VIA_NUM_BLIT_SLOTS) 
+			blitq->serviced = 0;
+		
+		cur_sg = blitq->blits[cur_released];
+		blitq->num_free++;
+				
+		spin_unlock_irqrestore(&blitq->blit_lock, irqsave);
+		
+		DRM_WAKEUP(&blitq->busy_queue);
+		
+		via_free_sg_info(dev->pdev, cur_sg);
+		kfree(cur_sg);
+		
+		spin_lock_irqsave(&blitq->blit_lock, irqsave);
+	}
+
+	spin_unlock_irqrestore(&blitq->blit_lock, irqsave);
+}
+	
+
+/*
+ * Init all blit engines. Currently we use two, but some hardware have 4.
+ */
+
+
+void
+via_init_dmablit(drm_device_t *dev)
+{
+	int i,j;
+	drm_via_private_t *dev_priv = (drm_via_private_t *)dev->dev_private;
+	drm_via_blitq_t *blitq;
+
+	pci_set_master(dev->pdev);	
+	
+	for (i=0; i< VIA_NUM_BLIT_ENGINES; ++i) {
+		blitq = dev_priv->blit_queues + i;
+		blitq->dev = dev;
+		blitq->cur_blit_handle = 0;
+		blitq->done_blit_handle = 0;
+		blitq->head = 0;
+		blitq->cur = 0;
+		blitq->serviced = 0;
+		blitq->num_free = VIA_NUM_BLIT_SLOTS;
+		blitq->num_outstanding = 0;
+		blitq->is_active = 0;
+		blitq->aborting = 0;
+		blitq->blit_lock = SPIN_LOCK_UNLOCKED;
+		for (j=0; j<VIA_NUM_BLIT_SLOTS; ++j) {
+			DRM_INIT_WAITQUEUE(blitq->blit_queue + j);
+		}
+		DRM_INIT_WAITQUEUE(&blitq->busy_queue);
+		INIT_WORK(&blitq->wq, via_dmablit_workqueue, blitq);
+		init_timer(&blitq->poll_timer);
+		blitq->poll_timer.function = &via_dmablit_timer;
+		blitq->poll_timer.data = (unsigned long) blitq;
+	}	
+}
+
+/*
+ * Build all info and do all mappings required for a blit.
+ */
+		
+
+static int
+via_build_sg_info(drm_device_t *dev, drm_via_sg_info_t *vsg, drm_via_dmablit_t *xfer)
+{
+	int draw = xfer->to_fb;
+	int ret = 0;
+	
+	vsg->direction = (draw) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
+	vsg->bounce_buffer = 0;
+
+	vsg->state = dr_via_sg_init;
+
+	if (xfer->num_lines <= 0 || xfer->line_length <= 0) {
+		DRM_ERROR("Zero size bitblt.\n");
+		return DRM_ERR(EINVAL);
+	}
+
+	/*
+	 * Below check is a driver limitation, not a hardware one. We
+	 * don't want to lock unused pages, and don't want to incoporate the
+	 * extra logic of avoiding them. Make sure there are no. 
+	 * (Not a big limitation anyway.)
+	 */
+
+	if ((xfer->mem_stride - xfer->line_length) >= PAGE_SIZE) {
+		DRM_ERROR("Too large system memory stride.\n");
+		return DRM_ERR(EINVAL);
+	}
+
+	/* 
+	 * we allow a negative fb stride to allow flipping of images in
+	 * transfer. 
+	 */
+
+	if (xfer->mem_stride < xfer->line_length ||
+		abs(xfer->fb_stride) < xfer->line_length) {
+		DRM_ERROR("Invalid frame-buffer / memory stride.\n");
+		return DRM_ERR(EINVAL);
+	}
+
+	/*
+	 * A hardware bug seems to be worked around if system memory addresses start on
+	 * 16 byte boundaries. This seems a bit restrictive however. VIA is contacted
+	 * about this. Meanwhile, impose the following restrictions:
+	 */
+
+	if ((((unsigned long)xfer->mem_addr & 3) != ((unsigned long)xfer->fb_addr & 3)) ||
+	    ((xfer->mem_stride & 3) != (xfer->fb_stride & 3))) {
+		DRM_ERROR("Invalid DRM bitblt alignment.\n");
+	        return DRM_ERR(EINVAL);
+	}
+
+	if (0 != (ret = via_lock_all_dma_pages(vsg, xfer))) {
+		DRM_ERROR("Could not lock DMA pages.\n");
+		via_free_sg_info(dev->pdev, vsg);
+		return ret;
+	}
+
+	via_map_blit_for_device(dev->pdev, xfer, vsg, 0);
+	if (0 != (ret = via_alloc_desc_pages(vsg))) {
+		DRM_ERROR("Could not allocate DMA descriptor pages.\n");
+		via_free_sg_info(dev->pdev, vsg);
+		return ret;
+	}
+	via_map_blit_for_device(dev->pdev, xfer, vsg, 1);
+	
+	return 0;
+}
+	
+
+/*
+ * Reserve one free slot in the blit queue. Will wait for one second for one
+ * to become available. Otherwise -EBUSY is returned.
+ */
+
+static int 
+via_dmablit_grab_slot(drm_via_blitq_t *blitq, int engine)
+{
+	int ret=0;
+	unsigned long irqsave;
+
+	DRM_DEBUG("Num free is %d\n", blitq->num_free);
+	spin_lock_irqsave(&blitq->blit_lock, irqsave);
+	while(blitq->num_free == 0) {
+		spin_unlock_irqrestore(&blitq->blit_lock, irqsave);
+
+		DRM_WAIT_ON(ret, blitq->busy_queue, DRM_HZ, blitq->num_free > 0);
+		if (ret) {
+			return (DRM_ERR(EINTR) == ret) ? DRM_ERR(EAGAIN) : ret;
+		}
+		
+		spin_lock_irqsave(&blitq->blit_lock, irqsave);
+	}
+	
+	blitq->num_free--;
+	spin_unlock_irqrestore(&blitq->blit_lock, irqsave);
+
+	return 0;
+}
+
+/*
+ * Hand back a free slot if we changed our mind.
+ */
+
+static void 
+via_dmablit_release_slot(drm_via_blitq_t *blitq)
+{
+	unsigned long irqsave;
+
+	spin_lock_irqsave(&blitq->blit_lock, irqsave);
+	blitq->num_free++;
+	spin_unlock_irqrestore(&blitq->blit_lock, irqsave);
+	DRM_WAKEUP( &blitq->busy_queue );
+}
+
+/*
+ * Grab a free slot. Build blit info and queue a blit.
+ */
+
+
+static int 
+via_dmablit(drm_device_t *dev, drm_via_dmablit_t *xfer)	 
+{
+	drm_via_private_t *dev_priv = (drm_via_private_t *)dev->dev_private;
+	drm_via_sg_info_t *vsg;
+	drm_via_blitq_t *blitq;
+        int ret;
+	int engine;
+	unsigned long irqsave;
+
+	if (dev_priv == NULL) {
+		DRM_ERROR("Called without initialization.\n");
+		return DRM_ERR(EINVAL);
+	}
+
+	engine = (xfer->to_fb) ? 0 : 1;
+	blitq = dev_priv->blit_queues + engine;
+	if (0 != (ret = via_dmablit_grab_slot(blitq, engine))) {
+		return ret;
+	}
+	if (NULL == (vsg = kmalloc(sizeof(*vsg), GFP_KERNEL))) {
+		via_dmablit_release_slot(blitq);
+		return DRM_ERR(ENOMEM);
+	}
+	if (0 != (ret = via_build_sg_info(dev, vsg, xfer))) {
+		via_dmablit_release_slot(blitq);
+		kfree(vsg);
+		return ret;
+	}
+	spin_lock_irqsave(&blitq->blit_lock, irqsave);
+
+	blitq->blits[blitq->head++] = vsg;
+	if (blitq->head >= VIA_NUM_BLIT_SLOTS) 
+		blitq->head = 0;
+	blitq->num_outstanding++;
+	xfer->sync.sync_handle = ++blitq->cur_blit_handle; 
+
+	spin_unlock_irqrestore(&blitq->blit_lock, irqsave);
+	xfer->sync.engine = engine;
+
+       	via_dmablit_handler(dev, engine, 0);
+
+	return 0;
+}
+
+/*
+ * Sync on a previously submitted blit. Note that the X server use signals extensively, and
+ * that there is a very big proability that this IOCTL will be interrupted by a signal. In that
+ * case it returns with -EAGAIN for the signal to be delivered. 
+ * The caller should then reissue the IOCTL. This is similar to what is being done for drmGetLock().
+ */
+
+int
+via_dma_blit_sync( DRM_IOCTL_ARGS )
+{
+	drm_via_blitsync_t sync;
+	int err;
+	DRM_DEVICE;
+
+	DRM_COPY_FROM_USER_IOCTL(sync, (drm_via_blitsync_t *)data, sizeof(sync));
+	
+	if (sync.engine >= VIA_NUM_BLIT_ENGINES) 
+		return DRM_ERR(EINVAL);
+
+	err = via_dmablit_sync(dev, sync.sync_handle, sync.engine);
+
+	if (DRM_ERR(EINTR) == err)
+		err = DRM_ERR(EAGAIN);
+
+	return err;
+}
+	
+
+/*
+ * Queue a blit and hand back a handle to be used for sync. This IOCTL may be interrupted by a signal
+ * while waiting for a free slot in the blit queue. In that case it returns with -EAGAIN and should 
+ * be reissued. See the above IOCTL code.
+ */
+
+int 
+via_dma_blit( DRM_IOCTL_ARGS )
+{
+	drm_via_dmablit_t xfer;
+	int err;
+	DRM_DEVICE;
+
+	DRM_COPY_FROM_USER_IOCTL(xfer, (drm_via_dmablit_t __user *)data, sizeof(xfer));
+
+	err = via_dmablit(dev, &xfer);
+	DRM_ERROR("Error is %d\n", err);
+
+	DRM_COPY_TO_USER_IOCTL((void __user *)data, xfer, sizeof(xfer));
+
+	return err;
+}
diff --git a/linux-core/via_dmablit.h b/linux-core/via_dmablit.h
new file mode 100644
index 00000000..64863917
--- /dev/null
+++ b/linux-core/via_dmablit.h
@@ -0,0 +1,138 @@
+/* via_dmablit.h -- PCI DMA BitBlt support for the VIA Unichrome/Pro
+ * 
+ * Copyright 2005 Thomas Hellstrom.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sub license,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, 
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: 
+ *    Thomas Hellstrom.
+ *    Register info from Digeo Inc.
+ */
+
+#ifndef _VIA_DMABLIT_H
+#define _VIA_DMABLIT_H
+
+#define VIA_NUM_BLIT_ENGINES 2
+#define VIA_NUM_BLIT_SLOTS 8
+
+struct _drm_via_descriptor;
+
+typedef struct _drm_via_sg_info {
+	struct page **pages;
+	unsigned long num_pages;
+	struct _drm_via_descriptor **desc_pages;
+	int num_desc_pages;
+	int num_desc;
+	enum dma_data_direction direction;
+	unsigned char *bounce_buffer;
+        dma_addr_t chain_start;
+	uint32_t free_on_sequence;
+        unsigned int descriptors_per_page;
+	int aborted;
+	enum {
+	        dr_via_device_mapped,
+		dr_via_desc_pages_alloc,
+		dr_via_pages_locked,
+		dr_via_pages_alloc,
+		dr_via_sg_init
+	} state;
+} drm_via_sg_info_t;
+
+typedef struct _drm_via_blitq {
+	drm_device_t *dev;
+	uint32_t cur_blit_handle;
+	uint32_t done_blit_handle;
+	unsigned serviced;
+	unsigned head;
+	unsigned cur;
+	unsigned num_free;
+	unsigned num_outstanding;
+	unsigned long end;  
+        int aborting;
+	int is_active;
+	drm_via_sg_info_t *blits[VIA_NUM_BLIT_SLOTS];
+	spinlock_t blit_lock;
+	wait_queue_head_t blit_queue[VIA_NUM_BLIT_SLOTS];
+	wait_queue_head_t busy_queue;
+	struct work_struct wq;
+	struct timer_list poll_timer;
+} drm_via_blitq_t;
+	
+
+/* 
+ *  PCI DMA Registers
+ *  Channels 2 & 3 don't seem to be implemented in hardware.
+ */
+ 
+#define VIA_PCI_DMA_MAR0            0xE40   /* Memory Address Register of Channel 0 */ 
+#define VIA_PCI_DMA_DAR0            0xE44   /* Device Address Register of Channel 0 */ 
+#define VIA_PCI_DMA_BCR0            0xE48   /* Byte Count Register of Channel 0 */ 
+#define VIA_PCI_DMA_DPR0            0xE4C   /* Descriptor Pointer Register of Channel 0 */ 
+
+#define VIA_PCI_DMA_MAR1            0xE50   /* Memory Address Register of Channel 1 */ 
+#define VIA_PCI_DMA_DAR1            0xE54   /* Device Address Register of Channel 1 */ 
+#define VIA_PCI_DMA_BCR1            0xE58   /* Byte Count Register of Channel 1 */ 
+#define VIA_PCI_DMA_DPR1            0xE5C   /* Descriptor Pointer Register of Channel 1 */ 
+
+#define VIA_PCI_DMA_MAR2            0xE60   /* Memory Address Register of Channel 2 */ 
+#define VIA_PCI_DMA_DAR2            0xE64   /* Device Address Register of Channel 2 */ 
+#define VIA_PCI_DMA_BCR2            0xE68   /* Byte Count Register of Channel 2 */ 
+#define VIA_PCI_DMA_DPR2            0xE6C   /* Descriptor Pointer Register of Channel 2 */ 
+
+#define VIA_PCI_DMA_MAR3            0xE70   /* Memory Address Register of Channel 3 */ 
+#define VIA_PCI_DMA_DAR3            0xE74   /* Device Address Register of Channel 3 */ 
+#define VIA_PCI_DMA_BCR3            0xE78   /* Byte Count Register of Channel 3 */ 
+#define VIA_PCI_DMA_DPR3            0xE7C   /* Descriptor Pointer Register of Channel 3 */ 
+
+#define VIA_PCI_DMA_MR0             0xE80   /* Mode Register of Channel 0 */ 
+#define VIA_PCI_DMA_MR1             0xE84   /* Mode Register of Channel 1 */ 
+#define VIA_PCI_DMA_MR2             0xE88   /* Mode Register of Channel 2 */ 
+#define VIA_PCI_DMA_MR3             0xE8C   /* Mode Register of Channel 3 */ 
+
+#define VIA_PCI_DMA_CSR0            0xE90   /* Command/Status Register of Channel 0 */ 
+#define VIA_PCI_DMA_CSR1            0xE94   /* Command/Status Register of Channel 1 */ 
+#define VIA_PCI_DMA_CSR2            0xE98   /* Command/Status Register of Channel 2 */ 
+#define VIA_PCI_DMA_CSR3            0xE9C   /* Command/Status Register of Channel 3 */ 
+
+#define VIA_PCI_DMA_PTR             0xEA0   /* Priority Type Register */ 
+
+/* Define for DMA engine */ 
+/* DPR */
+#define VIA_DMA_DPR_EC		(1<<1)	/* end of chain */
+#define VIA_DMA_DPR_DDIE	(1<<2)	/* descriptor done interrupt enable */
+#define VIA_DMA_DPR_DT		(1<<3)	/* direction of transfer (RO) */
+
+/* MR */
+#define VIA_DMA_MR_CM		(1<<0)	/* chaining mode */
+#define VIA_DMA_MR_TDIE		(1<<1)	/* transfer done interrupt enable */
+#define VIA_DMA_MR_HENDMACMD		(1<<7) /* ? */
+
+/* CSR */
+#define VIA_DMA_CSR_DE		(1<<0)	/* DMA enable */
+#define VIA_DMA_CSR_TS		(1<<1)	/* transfer start */
+#define VIA_DMA_CSR_TA		(1<<2)	/* transfer abort */
+#define VIA_DMA_CSR_TD		(1<<3)	/* transfer done */
+#define VIA_DMA_CSR_DD		(1<<4)	/* descriptor done */
+#define VIA_DMA_DPR_EC          (1<<1)  /* end of chain */
+
+
+
+#endif
diff --git a/shared-core/via_dma.c b/shared-core/via_dma.c
index ff9fca4b..327d48f7 100644
--- a/shared-core/via_dma.c
+++ b/shared-core/via_dma.c
@@ -740,6 +740,19 @@ via_cmdbuf_size(DRM_IOCTL_ARGS)
 	return ret;
 }
 
+#ifndef VIA_HAVE_DMABLIT
+int 
+via_dma_blit_sync( DRM_IOCTL_ARGS ) {
+	DRM_ERROR("PCI DMA BitBlt is not implemented for this system.\n");
+	return DRM_ERR(EINVAL);
+}
+int 
+via_dma_blit( DRM_IOCTL_ARGS ) {
+	DRM_ERROR("PCI DMA BitBlt is not implemented for this system.\n");
+	return DRM_ERR(EINVAL);
+}
+#endif
+
 drm_ioctl_desc_t via_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_VIA_ALLOCMEM)] = {via_mem_alloc, DRM_AUTH},
 	[DRM_IOCTL_NR(DRM_VIA_FREEMEM)] = {via_mem_free, DRM_AUTH},
@@ -752,7 +765,9 @@ drm_ioctl_desc_t via_ioctls[] = {
 	[DRM_IOCTL_NR(DRM_VIA_FLUSH)] = {via_flush_ioctl, DRM_AUTH},
 	[DRM_IOCTL_NR(DRM_VIA_PCICMD)] = {via_pci_cmdbuffer, DRM_AUTH},
 	[DRM_IOCTL_NR(DRM_VIA_CMDBUF_SIZE)] = {via_cmdbuf_size, DRM_AUTH},
-	[DRM_IOCTL_NR(DRM_VIA_WAIT_IRQ)] = {via_wait_irq, DRM_AUTH}
+	[DRM_IOCTL_NR(DRM_VIA_WAIT_IRQ)] = {via_wait_irq, DRM_AUTH},
+	[DRM_IOCTL_NR(DRM_VIA_DMA_BLIT)] = {via_dma_blit, DRM_AUTH},
+	[DRM_IOCTL_NR(DRM_VIA_BLIT_SYNC)] = {via_dma_blit_sync, DRM_AUTH}
 };
 
 int via_max_ioctl = DRM_ARRAY_SIZE(via_ioctls);
diff --git a/shared-core/via_drm.h b/shared-core/via_drm.h
index cc129a36..292106ab 100644
--- a/shared-core/via_drm.h
+++ b/shared-core/via_drm.h
@@ -75,6 +75,8 @@
 #define DRM_VIA_CMDBUF_SIZE	0x0b
 #define NOT_USED
 #define DRM_VIA_WAIT_IRQ        0x0d
+#define DRM_VIA_DMA_BLIT        0x0e
+#define DRM_VIA_BLIT_SYNC       0x0f
 
 #define DRM_IOCTL_VIA_ALLOCMEM	  DRM_IOWR(DRM_COMMAND_BASE + DRM_VIA_ALLOCMEM, drm_via_mem_t)
 #define DRM_IOCTL_VIA_FREEMEM	  DRM_IOW( DRM_COMMAND_BASE + DRM_VIA_FREEMEM, drm_via_mem_t)
@@ -89,6 +91,8 @@
 #define DRM_IOCTL_VIA_CMDBUF_SIZE DRM_IOWR( DRM_COMMAND_BASE + DRM_VIA_CMDBUF_SIZE, \
 					    drm_via_cmdbuf_size_t)
 #define DRM_IOCTL_VIA_WAIT_IRQ    DRM_IOWR( DRM_COMMAND_BASE + DRM_VIA_WAIT_IRQ, drm_via_irqwait_t)
+#define DRM_IOCTL_VIA_DMA_BLIT    DRM_IOW(DRM_COMMAND_BASE + DRM_VIA_DMA_BLIT, drm_via_dmablit_t)
+#define DRM_IOCTL_VIA_BLIT_SYNC   DRM_IOW(DRM_COMMAND_BASE + DRM_VIA_BLIT_SYNC, drm_via_blitsync_t)
 
 /* Indices into buf.Setup where various bits of state are mirrored per
  * context and per buffer.  These can be fired at the card as a unit,
@@ -212,6 +216,14 @@ typedef enum {
 
 #define VIA_IRQ_FLAGS_MASK 0xF0000000
 
+enum drm_via_irqs{drm_via_irq_hqv0 = 0,
+		  drm_via_irq_hqv1,
+		  drm_via_irq_dma0_dd,
+		  drm_via_irq_dma0_td,
+		  drm_via_irq_dma1_dd,
+		  drm_via_irq_dma1_td,
+                  drm_via_irq_num};
+
 struct drm_via_wait_irq_request{
 	unsigned irq;
 	via_irq_seq_type_t type;
@@ -224,4 +236,26 @@ typedef union drm_via_irqwait {
 	struct drm_wait_vblank_reply reply;
 } drm_via_irqwait_t;
 
+typedef struct drm_via_blitsync { 
+	uint32_t sync_handle;
+	unsigned engine;
+} drm_via_blitsync_t;
+
+typedef struct drm_via_dmablit {
+	uint32_t num_lines;          
+	uint32_t line_length;        
+
+        uint32_t fb_addr;                
+	uint32_t fb_stride;              
+
+        unsigned char *mem_addr;        
+	uint32_t  mem_stride;        
+       
+	int bounce_buffer;
+        int to_fb;
+
+	drm_via_blitsync_t sync;   
+} drm_via_dmablit_t;
+
+
 #endif				/* _VIA_DRM_H_ */
diff --git a/shared-core/via_drv.h b/shared-core/via_drv.h
index acbf17e1..9e4f303a 100644
--- a/shared-core/via_drv.h
+++ b/shared-core/via_drv.h
@@ -24,23 +24,31 @@
 #ifndef _VIA_DRV_H_
 #define _VIA_DRV_H_
 
-#define DRIVER_AUTHOR	"VIA"
+#define DRIVER_AUTHOR	"Various"
 
 #define DRIVER_NAME		"via"
 #define DRIVER_DESC		"VIA Unichrome / Pro"
-#define DRIVER_DATE		"20050814"
+#define DRIVER_DATE		"20050925"
 
 #define DRIVER_MAJOR		2
-#define DRIVER_MINOR		6
-#define DRIVER_PATCHLEVEL	7
+#define DRIVER_MINOR		7
+#define DRIVER_PATCHLEVEL	0
 
 #include "via_verifier.h"
 
-#define VIA_PCI_BUF_SIZE 60000
-#define VIA_FIRE_BUF_SIZE  1024
-#define VIA_NUM_IRQS 2
+#if defined(__linux__)
+#include "via_dmablit.h"
 
+/*
+ * This define and all its references can be removed when
+ * the DMA blit code has been implemented for FreeBSD.
+ */
+#define VIA_HAVE_DMABLIT 1
+#endif
 
+#define VIA_PCI_BUF_SIZE 60000
+#define VIA_FIRE_BUF_SIZE  1024
+#define VIA_NUM_IRQS 4
 
 typedef struct drm_via_ring_buffer {
 	drm_local_map_t map;
@@ -84,6 +92,10 @@ typedef struct drm_via_private {
 	maskarray_t *irq_masks;
 	uint32_t irq_enable_mask; 
 	uint32_t irq_pending_mask;	
+	int *irq_map;
+#ifdef VIA_HAVE_DMABLIT
+	drm_via_blitq_t blit_queues[VIA_NUM_BLIT_ENGINES];
+#endif
 } drm_via_private_t;
 
 enum via_family {
@@ -111,6 +123,8 @@ extern int via_flush_ioctl(DRM_IOCTL_ARGS);
 extern int via_pci_cmdbuffer(DRM_IOCTL_ARGS);
 extern int via_cmdbuf_size(DRM_IOCTL_ARGS);
 extern int via_wait_irq(DRM_IOCTL_ARGS);
+extern int via_dma_blit_sync( DRM_IOCTL_ARGS );
+extern int via_dma_blit( DRM_IOCTL_ARGS );
 
 extern int via_driver_load(drm_device_t *dev, unsigned long chipset);
 extern int via_driver_unload(drm_device_t *dev);
@@ -131,5 +145,11 @@ extern int via_driver_dma_quiescent(drm_device_t * dev);
 extern void via_init_futex(drm_via_private_t *dev_priv);
 extern void via_cleanup_futex(drm_via_private_t *dev_priv);
 extern void via_release_futex(drm_via_private_t *dev_priv, int context);
+extern int via_driver_irq_wait(drm_device_t * dev, unsigned int irq, int force_sequence,
+			       unsigned int *sequence);
+#ifdef VIA_HAVE_DMABLIT
+extern void via_dmablit_handler(drm_device_t *dev, int engine, int from_irq);
+extern void via_init_dmablit(drm_device_t *dev);
+#endif
 
 #endif
diff --git a/shared-core/via_irq.c b/shared-core/via_irq.c
index f87b2f2c..96be8e3d 100644
--- a/shared-core/via_irq.c
+++ b/shared-core/via_irq.c
@@ -50,6 +50,15 @@
 #define VIA_IRQ_HQV1_ENABLE     (1 << 25)
 #define VIA_IRQ_HQV0_PENDING    (1 << 9)
 #define VIA_IRQ_HQV1_PENDING    (1 << 10)
+#define VIA_IRQ_DMA0_DD_ENABLE  (1 << 20)
+#define VIA_IRQ_DMA0_TD_ENABLE  (1 << 21)
+#define VIA_IRQ_DMA1_DD_ENABLE  (1 << 22)
+#define VIA_IRQ_DMA1_TD_ENABLE  (1 << 23)
+#define VIA_IRQ_DMA0_DD_PENDING (1 << 4)
+#define VIA_IRQ_DMA0_TD_PENDING (1 << 5)
+#define VIA_IRQ_DMA1_DD_PENDING (1 << 6)
+#define VIA_IRQ_DMA1_TD_PENDING (1 << 7)
+
 
 /*
  * Device-specific IRQs go here. This type might need to be extended with
@@ -57,13 +66,26 @@
  * Currently we activate the HQV interrupts of  Unichrome Pro group A. 
  */
 
+
 static maskarray_t via_pro_group_a_irqs[] = {
 	{VIA_IRQ_HQV0_ENABLE, VIA_IRQ_HQV0_PENDING, 0x000003D0, 0x00008010, 0x00000000 },
-	{VIA_IRQ_HQV1_ENABLE, VIA_IRQ_HQV1_PENDING, 0x000013D0, 0x00008010, 0x00000000 }};
+	{VIA_IRQ_HQV1_ENABLE, VIA_IRQ_HQV1_PENDING, 0x000013D0, 0x00008010, 0x00000000 },
+	{VIA_IRQ_DMA0_TD_ENABLE, VIA_IRQ_DMA0_TD_PENDING, VIA_PCI_DMA_CSR0, 
+	 VIA_DMA_CSR_TA | VIA_DMA_CSR_TD, 0x00000008},
+	{VIA_IRQ_DMA1_TD_ENABLE, VIA_IRQ_DMA1_TD_PENDING, VIA_PCI_DMA_CSR1,
+	VIA_DMA_CSR_TA | VIA_DMA_CSR_TD, 0x00000008}
+};
 static int via_num_pro_group_a = sizeof(via_pro_group_a_irqs)/sizeof(maskarray_t);
-
-static maskarray_t via_unichrome_irqs[] = {};
+static int via_irqmap_pro_group_a[] = {0, 1, -1, 2, -1, 3};
+
+static maskarray_t via_unichrome_irqs[] = {
+	{VIA_IRQ_DMA0_TD_ENABLE, VIA_IRQ_DMA0_TD_PENDING, VIA_PCI_DMA_CSR0, 
+	 VIA_DMA_CSR_TA | VIA_DMA_CSR_TD, 0x00000008},
+	{VIA_IRQ_DMA1_TD_ENABLE, VIA_IRQ_DMA1_TD_PENDING, VIA_PCI_DMA_CSR1,
+	 VIA_DMA_CSR_TA | VIA_DMA_CSR_TD, 0x00000008}
+};
 static int via_num_unichrome = sizeof(via_unichrome_irqs)/sizeof(maskarray_t);
+static int via_irqmap_unichrome[] = {-1, -1, -1, 0, -1, 1};
 
 
 static unsigned time_diff(struct timeval *now,struct timeval *then) 
@@ -111,9 +133,17 @@ irqreturn_t via_driver_irq_handler(DRM_IRQ_ARGS)
 
 	for (i=0; i<dev_priv->num_irqs; ++i) {
 		if (status & cur_irq->pending_mask) {
+			DRM_DEBUG("Received IRQ %d\n", i);
 			atomic_inc( &cur_irq->irq_received );
 			DRM_WAKEUP( &cur_irq->irq_queue );
 			handled = 1;
+#ifdef VIA_HAVE_DMABLIT
+			if (dev_priv->irq_map[drm_via_irq_dma0_td] == i) {
+				via_dmablit_handler(dev, 0, 1);
+			} else if (dev_priv->irq_map[drm_via_irq_dma1_td] == i) {
+				via_dmablit_handler(dev, 1, 1);
+			}
+#endif
 		}
 		cur_irq++;
 	}
@@ -167,7 +197,7 @@ int via_driver_vblank_wait(drm_device_t * dev, unsigned int *sequence)
 	return ret;
 }
 
-static int 
+int 
 via_driver_irq_wait(drm_device_t * dev, unsigned int irq, int force_sequence,
 		    unsigned int *sequence)
 {
@@ -176,6 +206,7 @@ via_driver_irq_wait(drm_device_t * dev, unsigned int irq, int force_sequence,
 	drm_via_irq_t *cur_irq = dev_priv->via_irqs;
 	int ret = 0;
 	maskarray_t *masks = dev_priv->irq_masks;
+	int real_irq;
 
 	DRM_DEBUG("%s\n", __FUNCTION__);
 
@@ -184,14 +215,22 @@ via_driver_irq_wait(drm_device_t * dev, unsigned int irq, int force_sequence,
 		return DRM_ERR(EINVAL);
 	}
 
-	if (irq >= dev_priv->num_irqs ) {
+	if (irq >= drm_via_irq_num ) {
 		DRM_ERROR("%s Trying to wait on unknown irq %d\n", __FUNCTION__, irq);
 		return DRM_ERR(EINVAL);
 	}
 		
-	cur_irq += irq;
+	real_irq = dev_priv->irq_map[irq];
+
+	if (real_irq < 0) {
+		DRM_ERROR("%s Video IRQ %d is not available on this hardware.\n", __FUNCTION__, irq);
+		return DRM_ERR(EINVAL);
+	}
+
+
+	cur_irq += real_irq;
 
-	if (masks[irq][2] && !force_sequence) {
+	if (masks[real_irq][2] && !force_sequence) {
 		DRM_WAIT_ON(ret, cur_irq->irq_queue, 3 * DRM_HZ,
 			    ((VIA_READ(masks[irq][2]) & masks[irq][3]) == masks[irq][4]));
 		cur_irq_sequence = atomic_read(&cur_irq->irq_received);
@@ -226,6 +265,8 @@ void via_driver_irq_preinstall(drm_device_t * dev)
 			via_pro_group_a_irqs : via_unichrome_irqs;
 		dev_priv->num_irqs = (dev_priv->pro_group_a) ?
 			via_num_pro_group_a : via_num_unichrome;
+		dev_priv->irq_map = (dev_priv->pro_group_a) ?
+			via_irqmap_pro_group_a : via_irqmap_unichrome;
 		
 		for(i=0; i < dev_priv->num_irqs; ++i) {
 			atomic_set(&cur_irq->irq_received, 0);
diff --git a/shared-core/via_map.c b/shared-core/via_map.c
index 8e405418..2b653d75 100644
--- a/shared-core/via_map.c
+++ b/shared-core/via_map.c
@@ -61,7 +61,9 @@ static int via_do_init_map(drm_device_t * dev, drm_via_init_t * init)
 	dev_priv->agpAddr = init->agpAddr;
 
 	via_init_futex( dev_priv );
-
+#ifdef VIA_HAVE_DMABLIT
+	via_init_dmablit( dev );
+#endif
 	dev->dev_private = (void *)dev_priv;
 	return 0;
 }