[RFC 5/21] DRM: Add VIA DRM driver
James Simmons
jsimmons at infradead.org
Sat Jun 8 09:47:06 PDT 2013
commit fa8599a207fbd0b4ffaa8c6961d34797f615ed7d
Author: James Simmons <jsimmons at infradead.org>
Date: Fri Jun 7 19:58:14 2013 -0400
via: ttm bo driver implementation
The openchrome project decided to use the TTM api as the backend to
support GEM. The core needed to support TTM is struct ttm_bo_driver
which is implemented in via_ttm.c. This back end supports both AGP
and PCIe.
Signed-Off-by: James Simmons <jsimmons at infradead.org>
diff --git a/drivers/gpu/drm/via/via_ttm.c b/drivers/gpu/drm/via/via_ttm.c
new file mode 100644
index 0000000..5d6aead
--- /dev/null
+++ b/drivers/gpu/drm/via/via_ttm.c
@@ -0,0 +1,495 @@
+/*
+ * Copyright 2012 James Simmons <jsimmons at infradead.org>. 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 AUTHOR(S) OR COPYRIGHT HOLDER(S) 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.
+ */
+#include <linux/dma-mapping.h>
+#ifdef CONFIG_SWIOTLB
+#include <linux/swiotlb.h>
+#endif
+
+#include "via_drv.h"
+
+static struct ttm_tt *
+via_ttm_tt_create(struct ttm_bo_device *bdev, unsigned long size,
+ uint32_t page_flags, struct page *dummy_read_page)
+{
+#if __OS_HAS_AGP
+ struct drm_via_private *dev_priv =
+ container_of(bdev, struct drm_via_private, bdev);
+
+ if (drm_pci_device_is_agp(dev_priv->dev))
+ return ttm_agp_tt_create(bdev, dev_priv->dev->agp->bridge,
+ size, page_flags, dummy_read_page);
+#endif
+ return via_sgdma_backend_init(bdev, size, page_flags, dummy_read_page);
+}
+
+static int
+via_ttm_tt_populate(struct ttm_tt *ttm)
+{
+ struct sgdma_tt *dma_tt = (struct sgdma_tt *) ttm;
+ struct ttm_dma_tt *sgdma = &dma_tt->sgdma;
+ struct ttm_bo_device *bdev = ttm->bdev;
+ struct drm_via_private *dev_priv =
+ container_of(bdev, struct drm_via_private, bdev);
+ struct drm_device *dev = dev_priv->dev;
+ unsigned int i;
+ int ret = 0;
+
+ if (ttm->state != tt_unpopulated)
+ return 0;
+
+#if __OS_HAS_AGP
+ if (drm_pci_device_is_agp(dev_priv->dev))
+ return ttm_agp_tt_populate(ttm);
+#endif
+
+#ifdef CONFIG_SWIOTLB
+ if (swiotlb_nr_tbl())
+ return ttm_dma_populate(sgdma, dev->dev);
+#endif
+
+ ret = ttm_pool_populate(ttm);
+ if (ret)
+ return ret;
+
+ for (i = 0; i < ttm->num_pages; i++) {
+ sgdma->dma_address[i] = pci_map_page(dev->pdev, ttm->pages[i],
+ 0, PAGE_SIZE,
+ PCI_DMA_BIDIRECTIONAL);
+ if (pci_dma_mapping_error(dev->pdev, sgdma->dma_address[i])) {
+ while (--i) {
+ pci_unmap_page(dev->pdev, sgdma->dma_address[i],
+ PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+ sgdma->dma_address[i] = 0;
+ }
+ ttm_pool_unpopulate(ttm);
+ return -EFAULT;
+ }
+ }
+ return ret;
+}
+
+static void
+via_ttm_tt_unpopulate(struct ttm_tt *ttm)
+{
+ struct sgdma_tt *dma_tt = (struct sgdma_tt *) ttm;
+ struct ttm_dma_tt *sgdma = &dma_tt->sgdma;
+ struct ttm_bo_device *bdev = ttm->bdev;
+ struct drm_via_private *dev_priv =
+ container_of(bdev, struct drm_via_private, bdev);
+ struct drm_device *dev = dev_priv->dev;
+ unsigned int i;
+
+#if __OS_HAS_AGP
+ if (drm_pci_device_is_agp(dev_priv->dev)) {
+ ttm_agp_tt_unpopulate(ttm);
+ return;
+ }
+#endif
+
+#ifdef CONFIG_SWIOTLB
+ if (swiotlb_nr_tbl()) {
+ ttm_dma_unpopulate(sgdma, dev->dev);
+ return;
+ }
+#endif
+
+ for (i = 0; i < ttm->num_pages; i++) {
+ if (sgdma->dma_address[i]) {
+ pci_unmap_page(dev->pdev, sgdma->dma_address[i],
+ PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+ }
+ }
+
+ ttm_pool_unpopulate(ttm);
+}
+
+static int
+via_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
+{
+ /*
+ * FIXME: Invalidate texture caches here.
+ */
+ return 0;
+}
+
+static int
+via_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
+ struct ttm_mem_type_manager *man)
+{
+#if __OS_HAS_AGP
+ struct drm_via_private *dev_priv =
+ container_of(bdev, struct drm_via_private, bdev);
+ struct drm_device *dev = dev_priv->dev;
+#endif
+
+ switch (type) {
+ case TTM_PL_SYSTEM:
+ /* System memory */
+ man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
+ man->available_caching = TTM_PL_MASK_CACHING;
+ man->default_caching = TTM_PL_FLAG_CACHED;
+ break;
+
+ case TTM_PL_TT:
+ man->func = &ttm_bo_manager_func;
+
+ /* By default we handle PCI/PCIe DMA. If AGP is avaliable
+ * then we use that instead */
+ man->flags = TTM_MEMTYPE_FLAG_MAPPABLE | TTM_MEMTYPE_FLAG_CMA;
+ man->available_caching = TTM_PL_MASK_CACHING;
+ man->default_caching = TTM_PL_FLAG_CACHED;
+
+#if __OS_HAS_AGP
+ if (drm_pci_device_is_agp(dev)) {
+ if (drm_core_has_AGP(dev) && dev->agp) {
+ man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
+ man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC;
+ man->default_caching = TTM_PL_FLAG_WC;
+ } else
+ DRM_ERROR("AGP is possible but not enabled\n");
+ }
+#endif
+ break;
+
+ case TTM_PL_VRAM:
+ /* "On-card" video ram */
+ man->func = &ttm_bo_manager_func;
+ man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_MAPPABLE;
+ man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC;
+ man->default_caching = TTM_PL_FLAG_WC;
+ /* The display base address does not always equal the start of
+ * the memory region of the VRAM. In our case it is */
+ man->gpu_offset = 0;
+ break;
+
+ case TTM_PL_PRIV0:
+ /* MMIO region */
+ man->func = &ttm_bo_manager_func;
+ man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_MAPPABLE;
+ man->available_caching = TTM_PL_FLAG_UNCACHED;
+ man->default_caching = TTM_PL_FLAG_UNCACHED;
+ break;
+
+ default:
+ DRM_ERROR("Unsupported memory type %u\n", (unsigned)type);
+ return -EINVAL;
+ }
+ return 0;
+}
+
+static void
+via_evict_flags(struct ttm_buffer_object *bo, struct ttm_placement *placement)
+{
+ switch (bo->mem.mem_type) {
+ case TTM_PL_VRAM:
+ ttm_placement_from_domain(bo, placement, TTM_PL_FLAG_TT | TTM_PL_FLAG_SYSTEM, bo->bdev);
+ break;
+
+ case TTM_PL_TT:
+ default:
+ ttm_placement_from_domain(bo, placement, TTM_PL_FLAG_SYSTEM, bo->bdev);
+ break;
+ }
+}
+
+/*
+ * 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.
+ */
+struct drm_via_sg_info *
+via_alloc_desc_pages(struct ttm_tt *ttm, struct drm_device *dev,
+ unsigned long dev_start, enum dma_data_direction direction)
+{
+ struct drm_via_sg_info *vsg = kzalloc(sizeof(*vsg), GFP_KERNEL);
+ struct drm_via_private *dev_priv = dev->dev_private;
+ int desc_size = dev_priv->desc_size, i;
+
+ vsg->ttm = ttm;
+ vsg->dev_start = dev_start;
+ vsg->direction = direction;
+ vsg->num_desc = ttm->num_pages; // + 1;
+ vsg->descriptors_per_page = PAGE_SIZE / desc_size;
+ vsg->num_desc_pages = (vsg->num_desc + vsg->descriptors_per_page - 1) /
+ vsg->descriptors_per_page;
+
+ vsg->desc_pages = kzalloc(vsg->num_desc_pages * sizeof(void *), GFP_KERNEL);
+ if (!vsg->desc_pages)
+ return ERR_PTR(-ENOMEM);
+
+ vsg->state = dr_via_desc_pages_alloc;
+
+ /* Alloc pages for descriptor chain */
+ for (i = 0; i < vsg->num_desc_pages; ++i) {
+ vsg->desc_pages[i] = (unsigned long *) __get_free_page(GFP_KERNEL);
+
+ if (!vsg->desc_pages[i])
+ return ERR_PTR(-ENOMEM);
+ }
+ return vsg;
+}
+
+/* Move between GART and VRAM */
+static int
+via_move_blit(struct ttm_buffer_object *bo, bool evict, bool no_wait_gpu,
+ struct ttm_mem_reg *new_mem, struct ttm_mem_reg *old_mem)
+{
+ struct drm_via_private *dev_priv =
+ container_of(bo->bdev, struct drm_via_private, bdev);
+ enum dma_data_direction direction = DMA_TO_DEVICE;
+ unsigned long old_start, new_start, dev_addr = 0;
+ struct drm_via_sg_info *vsg;
+ int ret = -ENXIO;
+ struct via_fence *fence;
+
+ /* Real CPU physical address */
+ old_start = (old_mem->start << PAGE_SHIFT) + old_mem->bus.base;
+ new_start = (new_mem->start << PAGE_SHIFT) + new_mem->bus.base;
+
+ if (old_mem->mem_type == TTM_PL_VRAM) {
+ direction = DMA_FROM_DEVICE;
+ dev_addr = old_start;
+ } else if (new_mem->mem_type == TTM_PL_VRAM) {
+ /* direction is DMA_TO_DEVICE */
+ dev_addr = new_start;
+ }
+
+ /* device addr must be 16 byte align */
+ if (dev_addr & 0x0F)
+ return ret;
+
+ vsg = via_alloc_desc_pages(bo->ttm, dev_priv->dev, dev_addr, direction);
+ if (unlikely(IS_ERR(vsg)))
+ return PTR_ERR(vsg);
+
+ fence = via_fence_create_and_emit(dev_priv->dma_fences, vsg, 0);
+ if (unlikely(IS_ERR(fence)))
+ return PTR_ERR(fence);
+ return ttm_bo_move_accel_cleanup(bo, (void *)fence, evict, no_wait_gpu, new_mem);
+}
+
+static int
+via_move_from_vram(struct ttm_buffer_object *bo, bool interruptible,
+ bool no_wait_gpu, struct ttm_mem_reg *new_mem)
+{
+ struct ttm_mem_reg *old_mem = &bo->mem;
+ struct ttm_mem_reg tmp_mem;
+ struct ttm_placement placement;
+ u32 placements;
+ int ret;
+
+ tmp_mem = *new_mem;
+ tmp_mem.mm_node = NULL;
+ placement.fpfn = 0;
+ placement.lpfn = 0;
+ placement.num_placement = 1;
+ placement.placement = &placements;
+ placement.num_busy_placement = 1;
+ placement.busy_placement = &placements;
+ placements = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
+ ret = ttm_bo_mem_space(bo, &placement, &tmp_mem,
+ interruptible, no_wait_gpu);
+ if (unlikely(ret))
+ return ret;
+
+ /* Allocate some DMA memory for the GART address space */
+ ret = ttm_tt_set_placement_caching(bo->ttm, tmp_mem.placement);
+ if (unlikely(ret))
+ goto out_cleanup;
+
+ ret = ttm_tt_bind(bo->ttm, &tmp_mem);
+ if (unlikely(ret))
+ goto out_cleanup;
+
+ /* Move from the VRAM to GART space */
+ ret = via_move_blit(bo, true, no_wait_gpu, &tmp_mem, old_mem);
+ if (unlikely(ret))
+ goto out_cleanup;
+
+ /* Expose the GART region to the system memory */
+ ret = ttm_bo_move_ttm(bo, true, no_wait_gpu, new_mem);
+out_cleanup:
+ ttm_bo_mem_put(bo, &tmp_mem);
+ return ret;
+}
+
+static int
+via_move_to_vram(struct ttm_buffer_object *bo, bool interruptible,
+ bool no_wait_gpu, struct ttm_mem_reg *new_mem)
+{
+ struct ttm_mem_reg *old_mem = &bo->mem;
+ struct ttm_mem_reg tmp_mem;
+ struct ttm_placement placement;
+ u32 placements;
+ int ret;
+
+ tmp_mem = *new_mem;
+ tmp_mem.mm_node = NULL;
+ placement.fpfn = 0;
+ placement.lpfn = 0;
+ placement.num_placement = 1;
+ placement.placement = &placements;
+ placement.num_busy_placement = 1;
+ placement.busy_placement = &placements;
+ placements = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
+ ret = ttm_bo_mem_space(bo, &placement, &tmp_mem,
+ interruptible, no_wait_gpu);
+ if (unlikely(ret))
+ return ret;
+
+ /* Expose the GART region to the system memory */
+ ret = ttm_bo_move_ttm(bo, true, no_wait_gpu, &tmp_mem);
+ if (unlikely(ret))
+ goto out_cleanup;
+
+ /* Move from the GART to VRAM */
+ ret = via_move_blit(bo, true, no_wait_gpu, new_mem, old_mem);
+out_cleanup:
+ ttm_bo_mem_put(bo, &tmp_mem);
+ return ret;
+}
+
+static int
+via_bo_move(struct ttm_buffer_object *bo, bool evict, bool interruptible,
+ bool no_wait_gpu, struct ttm_mem_reg *new_mem)
+{
+ struct ttm_mem_reg *old_mem = &bo->mem;
+ int ret = 0;
+
+ /* No real memory copy. Just use the new_mem
+ * directly. */
+ if ((old_mem->mem_type == TTM_PL_SYSTEM &&
+ (new_mem->mem_type == TTM_PL_TT || !bo->ttm)) ||
+ (old_mem->mem_type == TTM_PL_TT &&
+ new_mem->mem_type == TTM_PL_SYSTEM) ||
+ (new_mem->mem_type == TTM_PL_PRIV0)) {
+ BUG_ON(old_mem->mm_node != NULL);
+ *old_mem = *new_mem;
+ new_mem->mm_node = NULL;
+ return ret;
+ }
+
+ /* Accelerated copy involving the VRAM. */
+ if (old_mem->mem_type == TTM_PL_VRAM &&
+ new_mem->mem_type == TTM_PL_SYSTEM) {
+ ret = via_move_from_vram(bo, interruptible, no_wait_gpu, new_mem);
+ } else if (old_mem->mem_type == TTM_PL_SYSTEM &&
+ new_mem->mem_type == TTM_PL_VRAM) {
+ ret = via_move_to_vram(bo, interruptible, no_wait_gpu, new_mem);
+ } else {
+ ret = via_move_blit(bo, evict, no_wait_gpu, new_mem, old_mem);
+ }
+
+ if (ret) {
+ ret = ttm_bo_move_memcpy(bo, evict, no_wait_gpu, new_mem);
+ }
+ return ret;
+}
+
+static int
+via_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+ struct drm_via_private *dev_priv =
+ container_of(bdev, struct drm_via_private, bdev);
+ struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
+ struct drm_device *dev = dev_priv->dev;
+
+ mem->bus.base = 0;
+ mem->bus.addr = NULL;
+ mem->bus.offset = mem->start << PAGE_SHIFT;
+ mem->bus.size = mem->num_pages << PAGE_SHIFT;
+ mem->bus.is_iomem = true;
+ if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
+ return -EINVAL;
+
+ switch (mem->mem_type) {
+ case TTM_PL_SYSTEM:
+ /* system memory */
+ mem->bus.is_iomem = false;
+ mem->bus.offset = 0;
+ return 0;
+
+ case TTM_PL_TT:
+#if __OS_HAS_AGP
+ if (drm_pci_device_is_agp(dev)) {
+ mem->bus.is_iomem = !dev->agp->cant_use_aperture;
+ mem->bus.base = dev->agp->base;
+ }
+#endif
+ break;
+
+ case TTM_PL_PRIV0:
+ mem->bus.base = pci_resource_start(dev->pdev, 1);
+ break;
+
+ case TTM_PL_VRAM:
+ if (dev->pci_device == PCI_DEVICE_ID_VIA_VX900)
+ mem->bus.base = pci_resource_start(dev->pdev, 2);
+ else
+ mem->bus.base = pci_resource_start(dev->pdev, 0);
+ break;
+
+ default:
+ return -EINVAL;
+ }
+ return 0;
+}
+
+static void via_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+}
+
+static int via_verify_access(struct ttm_buffer_object *bo, struct file *filp)
+{
+ return 0;
+}
+
+static struct ttm_bo_driver via_bo_driver = {
+ .ttm_tt_create = via_ttm_tt_create,
+ .ttm_tt_populate = via_ttm_tt_populate,
+ .ttm_tt_unpopulate = via_ttm_tt_unpopulate,
+ .invalidate_caches = via_invalidate_caches,
+ .init_mem_type = via_init_mem_type,
+ .evict_flags = via_evict_flags,
+ .move = via_bo_move,
+ .verify_access = via_verify_access,
+ .sync_obj_signaled = via_fence_signaled,
+ .sync_obj_wait = via_fence_wait,
+ .sync_obj_flush = via_fence_flush,
+ .sync_obj_unref = via_fence_unref,
+ .sync_obj_ref = via_fence_ref,
+ .io_mem_reserve = via_ttm_io_mem_reserve,
+ .io_mem_free = via_ttm_io_mem_free,
+};
+
+int via_ttm_init(struct drm_via_private *dev_priv)
+{
+ int ret = ttm_global_init(&dev_priv->mem_global_ref,
+ &dev_priv->bo_global_ref,
+ &via_bo_driver,
+ &dev_priv->bdev, false);
+ if (!ret)
+ dev_priv->bdev.dev_mapping = dev_priv->dev->dev_mapping;
+ return ret;
+}
More information about the dri-devel
mailing list