From: Thierry Reding treding@nvidia.com
When an IOMMU device is available on the platform bus, allocate an IOMMU domain and attach the display controllers to it. The display controllers can then scan out non-contiguous buffers by mapping them through the IOMMU.
Signed-off-by: Thierry Reding treding@nvidia.com --- Changes in v6: - clean up IOMMU resources on initialization failure
Changes in v5: - fix PRIME export of physically non-contiguous buffers - refactor and fix error cleanup paths - fix checking for failed allocations - fix build warnings on 64-bit
Changes in v4: - error out if IOMMU attachment fails
Changes in v3: - create write-combined mappings for mmap() - keep non-IOMMU fallback path working - no longer rely on iommu_attach()
Changes in v2: - don't pass GFP_KERNEL to drm_gem_get_pages()
drivers/gpu/drm/tegra/dc.c | 17 +++ drivers/gpu/drm/tegra/drm.c | 24 ++++ drivers/gpu/drm/tegra/drm.h | 5 + drivers/gpu/drm/tegra/fb.c | 16 ++- drivers/gpu/drm/tegra/gem.c | 277 ++++++++++++++++++++++++++++++++++++++------ drivers/gpu/drm/tegra/gem.h | 6 + 6 files changed, 308 insertions(+), 37 deletions(-)
diff --git a/drivers/gpu/drm/tegra/dc.c b/drivers/gpu/drm/tegra/dc.c index a058a43e6322..62c81df67216 100644 --- a/drivers/gpu/drm/tegra/dc.c +++ b/drivers/gpu/drm/tegra/dc.c @@ -9,6 +9,7 @@
#include <linux/clk.h> #include <linux/debugfs.h> +#include <linux/iommu.h> #include <linux/reset.h>
#include <soc/tegra/pmc.h> @@ -1289,6 +1290,17 @@ static int tegra_dc_init(struct host1x_client *client) struct tegra_drm *tegra = drm->dev_private; int err;
+ if (tegra->domain) { + err = iommu_attach_device(tegra->domain, dc->dev); + if (err < 0) { + dev_err(dc->dev, "failed to attach to domain: %d\n", + err); + return err; + } + + dc->domain = tegra->domain; + } + drm_crtc_init(drm, &dc->base, &tegra_crtc_funcs); drm_mode_crtc_set_gamma_size(&dc->base, 256); drm_crtc_helper_add(&dc->base, &tegra_crtc_helper_funcs); @@ -1346,6 +1358,11 @@ static int tegra_dc_exit(struct host1x_client *client) return err; }
+ if (dc->domain) { + iommu_detach_device(dc->domain, dc->dev); + dc->domain = NULL; + } + return 0; }
diff --git a/drivers/gpu/drm/tegra/drm.c b/drivers/gpu/drm/tegra/drm.c index e1632fb03a89..6c9df794b3be 100644 --- a/drivers/gpu/drm/tegra/drm.c +++ b/drivers/gpu/drm/tegra/drm.c @@ -8,6 +8,7 @@ */
#include <linux/host1x.h> +#include <linux/iommu.h>
#include "drm.h" #include "gem.h" @@ -33,6 +34,17 @@ static int tegra_drm_load(struct drm_device *drm, unsigned long flags) if (!tegra) return -ENOMEM;
+ if (iommu_present(&platform_bus_type)) { + tegra->domain = iommu_domain_alloc(&platform_bus_type); + if (IS_ERR(tegra->domain)) { + err = PTR_ERR(tegra->domain); + goto free; + } + + DRM_DEBUG("IOMMU context initialized\n"); + drm_mm_init(&tegra->mm, 0, SZ_2G); + } + mutex_init(&tegra->clients_lock); INIT_LIST_HEAD(&tegra->clients); drm->dev_private = tegra; @@ -76,6 +88,12 @@ fbdev: tegra_drm_fb_free(drm); config: drm_mode_config_cleanup(drm); + + if (tegra->domain) { + iommu_domain_free(tegra->domain); + drm_mm_takedown(&tegra->mm); + } +free: kfree(tegra); return err; } @@ -83,6 +101,7 @@ config: static int tegra_drm_unload(struct drm_device *drm) { struct host1x_device *device = to_host1x_device(drm->dev); + struct tegra_drm *tegra = drm->dev_private; int err;
drm_kms_helper_poll_fini(drm); @@ -94,6 +113,11 @@ static int tegra_drm_unload(struct drm_device *drm) if (err < 0) return err;
+ if (tegra->domain) { + iommu_domain_free(tegra->domain); + drm_mm_takedown(&tegra->mm); + } + return 0; }
diff --git a/drivers/gpu/drm/tegra/drm.h b/drivers/gpu/drm/tegra/drm.h index ef2faaef5936..96ff47d586a2 100644 --- a/drivers/gpu/drm/tegra/drm.h +++ b/drivers/gpu/drm/tegra/drm.h @@ -39,6 +39,9 @@ struct tegra_fbdev { struct tegra_drm { struct drm_device *drm;
+ struct iommu_domain *domain; + struct drm_mm mm; + struct mutex clients_lock; struct list_head clients;
@@ -121,6 +124,8 @@ struct tegra_dc { struct drm_pending_vblank_event *event;
const struct tegra_dc_soc_info *soc; + + struct iommu_domain *domain; };
static inline struct tegra_dc * diff --git a/drivers/gpu/drm/tegra/fb.c b/drivers/gpu/drm/tegra/fb.c index 0474241ac2a4..fab39eb2dae8 100644 --- a/drivers/gpu/drm/tegra/fb.c +++ b/drivers/gpu/drm/tegra/fb.c @@ -65,8 +65,12 @@ static void tegra_fb_destroy(struct drm_framebuffer *framebuffer) for (i = 0; i < fb->num_planes; i++) { struct tegra_bo *bo = fb->planes[i];
- if (bo) + if (bo) { + if (bo->pages && bo->vaddr) + vunmap(bo->vaddr); + drm_gem_object_unreference_unlocked(&bo->gem); + } }
drm_framebuffer_cleanup(framebuffer); @@ -254,6 +258,16 @@ static int tegra_fbdev_probe(struct drm_fb_helper *helper, offset = info->var.xoffset * bytes_per_pixel + info->var.yoffset * fb->pitches[0];
+ if (bo->pages) { + bo->vaddr = vmap(bo->pages, bo->num_pages, VM_MAP, + pgprot_writecombine(PAGE_KERNEL)); + if (!bo->vaddr) { + dev_err(drm->dev, "failed to vmap() framebuffer\n"); + err = -ENOMEM; + goto destroy; + } + } + drm->mode_config.fb_base = (resource_size_t)bo->paddr; info->screen_base = (void __iomem *)bo->vaddr + offset; info->screen_size = size; diff --git a/drivers/gpu/drm/tegra/gem.c b/drivers/gpu/drm/tegra/gem.c index b513df33f17a..8b1095d05c58 100644 --- a/drivers/gpu/drm/tegra/gem.c +++ b/drivers/gpu/drm/tegra/gem.c @@ -14,6 +14,7 @@ */
#include <linux/dma-buf.h> +#include <linux/iommu.h> #include <drm/tegra_drm.h>
#include "drm.h" @@ -91,6 +92,88 @@ static const struct host1x_bo_ops tegra_bo_ops = { .kunmap = tegra_bo_kunmap, };
+/* + * A generic iommu_map_sg() function is being reviewed and will hopefully be + * merged soon. At that point this function can be dropped in favour of the + * one provided by the IOMMU API. + */ +static ssize_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova, + struct scatterlist *sg, unsigned int nents, + int prot) +{ + struct scatterlist *s; + size_t offset = 0; + unsigned int i; + int err; + + for_each_sg(sg, s, nents, i) { + phys_addr_t phys = page_to_phys(sg_page(s)); + size_t length = s->offset + s->length; + + err = iommu_map(domain, iova + offset, phys, length, prot); + if (err < 0) { + iommu_unmap(domain, iova, offset); + return err; + } + + offset += length; + } + + return offset; +} + +static int tegra_bo_iommu_map(struct tegra_drm *tegra, struct tegra_bo *bo) +{ + int prot = IOMMU_READ | IOMMU_WRITE; + ssize_t err; + + if (bo->mm) + return -EBUSY; + + bo->mm = kzalloc(sizeof(*bo->mm), GFP_KERNEL); + if (!bo->mm) + return -ENOMEM; + + err = drm_mm_insert_node_generic(&tegra->mm, bo->mm, bo->gem.size, + PAGE_SIZE, 0, 0, 0); + if (err < 0) { + dev_err(tegra->drm->dev, "out of I/O virtual memory: %zd\n", + err); + goto free; + } + + bo->paddr = bo->mm->start; + + err = __iommu_map_sg(tegra->domain, bo->paddr, bo->sgt->sgl, + bo->sgt->nents, prot); + if (err < 0) { + dev_err(tegra->drm->dev, "failed to map buffer: %zd\n", err); + goto remove; + } + + bo->size = err; + + return 0; + +remove: + drm_mm_remove_node(bo->mm); +free: + kfree(bo->mm); + return err; +} + +static int tegra_bo_iommu_unmap(struct tegra_drm *tegra, struct tegra_bo *bo) +{ + if (!bo->mm) + return 0; + + iommu_unmap(tegra->domain, bo->paddr, bo->size); + drm_mm_remove_node(bo->mm); + kfree(bo->mm); + + return 0; +} + static struct tegra_bo *tegra_bo_alloc_object(struct drm_device *drm, size_t size) { @@ -121,9 +204,64 @@ free: return ERR_PTR(err); }
-static void tegra_bo_destroy(struct drm_device *drm, struct tegra_bo *bo) +static void tegra_bo_free(struct drm_device *drm, struct tegra_bo *bo) { - dma_free_writecombine(drm->dev, bo->gem.size, bo->vaddr, bo->paddr); + if (bo->pages) { + drm_gem_put_pages(&bo->gem, bo->pages, true, true); + sg_free_table(bo->sgt); + kfree(bo->sgt); + } else { + dma_free_writecombine(drm->dev, bo->gem.size, bo->vaddr, + bo->paddr); + } +} + +static int tegra_bo_get_pages(struct drm_device *drm, struct tegra_bo *bo, + size_t size) +{ + bo->pages = drm_gem_get_pages(&bo->gem); + if (IS_ERR(bo->pages)) + return PTR_ERR(bo->pages); + + bo->num_pages = size >> PAGE_SHIFT; + + bo->sgt = drm_prime_pages_to_sg(bo->pages, bo->num_pages); + if (IS_ERR(bo->sgt)) { + drm_gem_put_pages(&bo->gem, bo->pages, false, false); + return PTR_ERR(bo->sgt); + } + + return 0; +} + +static int tegra_bo_alloc(struct drm_device *drm, struct tegra_bo *bo, + size_t size) +{ + struct tegra_drm *tegra = drm->dev_private; + int err; + + if (tegra->domain) { + err = tegra_bo_get_pages(drm, bo, size); + if (err < 0) + return err; + + err = tegra_bo_iommu_map(tegra, bo); + if (err < 0) { + tegra_bo_free(drm, bo); + return err; + } + } else { + bo->vaddr = dma_alloc_writecombine(drm->dev, size, &bo->paddr, + GFP_KERNEL | __GFP_NOWARN); + if (!bo->vaddr) { + dev_err(drm->dev, + "failed to allocate buffer of size %zu\n", + size); + return -ENOMEM; + } + } + + return 0; }
struct tegra_bo *tegra_bo_create(struct drm_device *drm, unsigned int size, @@ -136,14 +274,9 @@ struct tegra_bo *tegra_bo_create(struct drm_device *drm, unsigned int size, if (IS_ERR(bo)) return bo;
- bo->vaddr = dma_alloc_writecombine(drm->dev, size, &bo->paddr, - GFP_KERNEL | __GFP_NOWARN); - if (!bo->vaddr) { - dev_err(drm->dev, "failed to allocate buffer with size %u\n", - size); - err = -ENOMEM; - goto err_dma; - } + err = tegra_bo_alloc(drm, bo, size); + if (err < 0) + goto release;
if (flags & DRM_TEGRA_GEM_CREATE_TILED) bo->tiling.mode = TEGRA_BO_TILING_MODE_TILED; @@ -153,13 +286,9 @@ struct tegra_bo *tegra_bo_create(struct drm_device *drm, unsigned int size,
return bo;
-err_mmap: +release: drm_gem_object_release(&bo->gem); -err_init: - tegra_bo_destroy(drm, bo); -err_dma: kfree(bo); - return ERR_PTR(err); }
@@ -190,6 +319,7 @@ struct tegra_bo *tegra_bo_create_with_handle(struct drm_file *file, static struct tegra_bo *tegra_bo_import(struct drm_device *drm, struct dma_buf *buf) { + struct tegra_drm *tegra = drm->dev_private; struct dma_buf_attachment *attach; struct tegra_bo *bo; int err; @@ -217,12 +347,19 @@ static struct tegra_bo *tegra_bo_import(struct drm_device *drm, goto detach; }
- if (bo->sgt->nents > 1) { - err = -EINVAL; - goto detach; + if (tegra->domain) { + err = tegra_bo_iommu_map(tegra, bo); + if (err < 0) + goto detach; + } else { + if (bo->sgt->nents > 1) { + err = -EINVAL; + goto detach; + } + + bo->paddr = sg_dma_address(bo->sgt->sgl); }
- bo->paddr = sg_dma_address(bo->sgt->sgl); bo->gem.import_attach = attach;
return bo; @@ -241,14 +378,18 @@ free:
void tegra_bo_free_object(struct drm_gem_object *gem) { + struct tegra_drm *tegra = gem->dev->dev_private; struct tegra_bo *bo = to_tegra_bo(gem);
+ if (tegra->domain) + tegra_bo_iommu_unmap(tegra, bo); + if (gem->import_attach) { dma_buf_unmap_attachment(gem->import_attach, bo->sgt, DMA_TO_DEVICE); drm_prime_gem_destroy(gem, NULL); } else { - tegra_bo_destroy(gem->dev, bo); + tegra_bo_free(gem->dev, bo); }
drm_gem_object_release(gem); @@ -303,14 +444,44 @@ int tegra_bo_dumb_map_offset(struct drm_file *file, struct drm_device *drm, return 0; }
+static int tegra_bo_fault(struct vm_area_struct *vma, struct vm_fault *vmf) +{ + struct drm_gem_object *gem = vma->vm_private_data; + struct tegra_bo *bo = to_tegra_bo(gem); + struct page *page; + pgoff_t offset; + int err; + + if (!bo->pages) + return VM_FAULT_SIGBUS; + + offset = ((unsigned long)vmf->virtual_address - vma->vm_start) >> PAGE_SHIFT; + page = bo->pages[offset]; + + err = vm_insert_page(vma, (unsigned long)vmf->virtual_address, page); + switch (err) { + case -EAGAIN: + case 0: + case -ERESTARTSYS: + case -EINTR: + case -EBUSY: + return VM_FAULT_NOPAGE; + + case -ENOMEM: + return VM_FAULT_OOM; + } + + return VM_FAULT_SIGBUS; +} + const struct vm_operations_struct tegra_bo_vm_ops = { + .fault = tegra_bo_fault, .open = drm_gem_vm_open, .close = drm_gem_vm_close, };
int tegra_drm_mmap(struct file *file, struct vm_area_struct *vma) { - unsigned long vm_pgoff = vma->vm_pgoff; struct drm_gem_object *gem; struct tegra_bo *bo; int ret; @@ -322,17 +493,28 @@ int tegra_drm_mmap(struct file *file, struct vm_area_struct *vma) gem = vma->vm_private_data; bo = to_tegra_bo(gem);
- vma->vm_flags &= ~VM_PFNMAP; - vma->vm_pgoff = 0; + if (!bo->pages) { + unsigned long vm_pgoff = vma->vm_pgoff;
- ret = dma_mmap_writecombine(gem->dev->dev, vma, bo->vaddr, bo->paddr, - gem->size); - if (ret) { - drm_gem_vm_close(vma); - return ret; - } + vma->vm_flags &= ~VM_PFNMAP; + vma->vm_pgoff = 0;
- vma->vm_pgoff = vm_pgoff; + ret = dma_mmap_writecombine(gem->dev->dev, vma, bo->vaddr, + bo->paddr, gem->size); + if (ret) { + drm_gem_vm_close(vma); + return ret; + } + + vma->vm_pgoff = vm_pgoff; + } else { + pgprot_t prot = vm_get_page_prot(vma->vm_flags); + + vma->vm_flags |= VM_MIXEDMAP; + vma->vm_flags &= ~VM_PFNMAP; + + vma->vm_page_prot = pgprot_writecombine(prot); + }
return 0; } @@ -349,21 +531,44 @@ tegra_gem_prime_map_dma_buf(struct dma_buf_attachment *attach, if (!sgt) return NULL;
- if (sg_alloc_table(sgt, 1, GFP_KERNEL)) { - kfree(sgt); - return NULL; - } + if (bo->pages) { + struct scatterlist *sg; + unsigned int i; + + if (sg_alloc_table(sgt, bo->num_pages, GFP_KERNEL)) + goto free;
- sg_dma_address(sgt->sgl) = bo->paddr; - sg_dma_len(sgt->sgl) = gem->size; + for_each_sg(sgt->sgl, sg, bo->num_pages, i) + sg_set_page(sg, bo->pages[i], PAGE_SIZE, 0); + + if (dma_map_sg(attach->dev, sgt->sgl, sgt->nents, dir) == 0) + goto free; + } else { + if (sg_alloc_table(sgt, 1, GFP_KERNEL)) + goto free; + + sg_dma_address(sgt->sgl) = bo->paddr; + sg_dma_len(sgt->sgl) = gem->size; + }
return sgt; + +free: + sg_free_table(sgt); + kfree(sgt); + return NULL; }
static void tegra_gem_prime_unmap_dma_buf(struct dma_buf_attachment *attach, struct sg_table *sgt, enum dma_data_direction dir) { + struct drm_gem_object *gem = attach->dmabuf->priv; + struct tegra_bo *bo = to_tegra_bo(gem); + + if (bo->pages) + dma_unmap_sg(attach->dev, sgt->sgl, sgt->nents, dir); + sg_free_table(sgt); kfree(sgt); } diff --git a/drivers/gpu/drm/tegra/gem.h b/drivers/gpu/drm/tegra/gem.h index 6538b56780c2..3dd4165f812a 100644 --- a/drivers/gpu/drm/tegra/gem.h +++ b/drivers/gpu/drm/tegra/gem.h @@ -38,6 +38,12 @@ struct tegra_bo { dma_addr_t paddr; void *vaddr;
+ struct drm_mm_node *mm; + unsigned long num_pages; + struct page **pages; + /* size of IOMMU mapping */ + size_t size; + struct tegra_bo_tiling tiling; };