drm/i915: Wean off drm_pci_alloc/drm_pci_free
authorChris Wilson <chris@chris-wilson.co.uk>
Sun, 2 Feb 2020 15:39:34 +0000 (15:39 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 28 Feb 2020 16:22:21 +0000 (17:22 +0100)
commit aa3146193ae25d0fe4b96d815169a135db2e8f01 upstream.

drm_pci_alloc and drm_pci_free are just very thin wrappers around
dma_alloc_coherent, with a note that we should be removing them.
Furthermore since

commit de09d31dd38a50fdce106c15abd68432eebbd014
Author: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Date:   Fri Jan 15 16:51:42 2016 -0800

    page-flags: define PG_reserved behavior on compound pages

    As far as I can see there's no users of PG_reserved on compound pages.
    Let's use PF_NO_COMPOUND here.

drm_pci_alloc has been declared broken since it mixes GFP_COMP and
SetPageReserved. Avoid this conflict by weaning ourselves off using the
abstraction and using the dma functions directly.

Reported-by: Taketo Kabe
Closes: https://gitlab.freedesktop.org/drm/intel/issues/1027
Fixes: de09d31dd38a ("page-flags: define PG_reserved behavior on compound pages")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: <stable@vger.kernel.org> # v4.5+
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20200202153934.3899472-1-chris@chris-wilson.co.uk
(cherry picked from commit c6790dc22312f592c1434577258b31c48c72d52a)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/gpu/drm/i915/display/intel_display.c
drivers/gpu/drm/i915/gem/i915_gem_object_types.h
drivers/gpu/drm/i915/gem/i915_gem_phys.c
drivers/gpu/drm/i915/i915_gem.c

index af50f05f4e9dfcf66290f1947770918531f8282c..272503615378873d2ec9663a1c4ab84aa3419d36 100644 (file)
@@ -10510,7 +10510,7 @@ static u32 intel_cursor_base(const struct intel_plane_state *plane_state)
        u32 base;
 
        if (INTEL_INFO(dev_priv)->display.cursor_needs_physical)
-               base = obj->phys_handle->busaddr;
+               base = sg_dma_address(obj->mm.pages->sgl);
        else
                base = intel_plane_ggtt_offset(plane_state);
 
index 646859fea2246fad71e79f4ce14f8102759f573d..08b35587bc6dc30f92d971ad7cd1bdaa6bd6e690 100644 (file)
@@ -240,9 +240,6 @@ struct drm_i915_gem_object {
 
                void *gvt_info;
        };
-
-       /** for phys allocated objects */
-       struct drm_dma_handle *phys_handle;
 };
 
 static inline struct drm_i915_gem_object *
index 76835690816081c39aa6b2fcffc7ef3fa3ea62a3..0cfe9bd763775b5afcc2a46f7bfdf5610f415f5f 100644 (file)
 static int i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
 {
        struct address_space *mapping = obj->base.filp->f_mapping;
-       struct drm_dma_handle *phys;
-       struct sg_table *st;
        struct scatterlist *sg;
-       char *vaddr;
+       struct sg_table *st;
+       dma_addr_t dma;
+       void *vaddr;
+       void *dst;
        int i;
-       int err;
 
        if (WARN_ON(i915_gem_object_needs_bit17_swizzle(obj)))
                return -EINVAL;
 
-       /* Always aligning to the object size, allows a single allocation
+       /*
+        * Always aligning to the object size, allows a single allocation
         * to handle all possible callers, and given typical object sizes,
         * the alignment of the buddy allocation will naturally match.
         */
-       phys = drm_pci_alloc(obj->base.dev,
-                            roundup_pow_of_two(obj->base.size),
-                            roundup_pow_of_two(obj->base.size));
-       if (!phys)
+       vaddr = dma_alloc_coherent(&obj->base.dev->pdev->dev,
+                                  roundup_pow_of_two(obj->base.size),
+                                  &dma, GFP_KERNEL);
+       if (!vaddr)
                return -ENOMEM;
 
-       vaddr = phys->vaddr;
+       st = kmalloc(sizeof(*st), GFP_KERNEL);
+       if (!st)
+               goto err_pci;
+
+       if (sg_alloc_table(st, 1, GFP_KERNEL))
+               goto err_st;
+
+       sg = st->sgl;
+       sg->offset = 0;
+       sg->length = obj->base.size;
+
+       sg_assign_page(sg, (struct page *)vaddr);
+       sg_dma_address(sg) = dma;
+       sg_dma_len(sg) = obj->base.size;
+
+       dst = vaddr;
        for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
                struct page *page;
-               char *src;
+               void *src;
 
                page = shmem_read_mapping_page(mapping, i);
-               if (IS_ERR(page)) {
-                       err = PTR_ERR(page);
-                       goto err_phys;
-               }
+               if (IS_ERR(page))
+                       goto err_st;
 
                src = kmap_atomic(page);
-               memcpy(vaddr, src, PAGE_SIZE);
-               drm_clflush_virt_range(vaddr, PAGE_SIZE);
+               memcpy(dst, src, PAGE_SIZE);
+               drm_clflush_virt_range(dst, PAGE_SIZE);
                kunmap_atomic(src);
 
                put_page(page);
-               vaddr += PAGE_SIZE;
+               dst += PAGE_SIZE;
        }
 
        intel_gt_chipset_flush(&to_i915(obj->base.dev)->gt);
 
-       st = kmalloc(sizeof(*st), GFP_KERNEL);
-       if (!st) {
-               err = -ENOMEM;
-               goto err_phys;
-       }
-
-       if (sg_alloc_table(st, 1, GFP_KERNEL)) {
-               kfree(st);
-               err = -ENOMEM;
-               goto err_phys;
-       }
-
-       sg = st->sgl;
-       sg->offset = 0;
-       sg->length = obj->base.size;
-
-       sg_dma_address(sg) = phys->busaddr;
-       sg_dma_len(sg) = obj->base.size;
-
-       obj->phys_handle = phys;
-
        __i915_gem_object_set_pages(obj, st, sg->length);
 
        return 0;
 
-err_phys:
-       drm_pci_free(obj->base.dev, phys);
-
-       return err;
+err_st:
+       kfree(st);
+err_pci:
+       dma_free_coherent(&obj->base.dev->pdev->dev,
+                         roundup_pow_of_two(obj->base.size),
+                         vaddr, dma);
+       return -ENOMEM;
 }
 
 static void
 i915_gem_object_put_pages_phys(struct drm_i915_gem_object *obj,
                               struct sg_table *pages)
 {
+       dma_addr_t dma = sg_dma_address(pages->sgl);
+       void *vaddr = sg_page(pages->sgl);
+
        __i915_gem_object_release_shmem(obj, pages, false);
 
        if (obj->mm.dirty) {
                struct address_space *mapping = obj->base.filp->f_mapping;
-               char *vaddr = obj->phys_handle->vaddr;
+               void *src = vaddr;
                int i;
 
                for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
@@ -114,15 +113,16 @@ i915_gem_object_put_pages_phys(struct drm_i915_gem_object *obj,
                                continue;
 
                        dst = kmap_atomic(page);
-                       drm_clflush_virt_range(vaddr, PAGE_SIZE);
-                       memcpy(dst, vaddr, PAGE_SIZE);
+                       drm_clflush_virt_range(src, PAGE_SIZE);
+                       memcpy(dst, src, PAGE_SIZE);
                        kunmap_atomic(dst);
 
                        set_page_dirty(page);
                        if (obj->mm.madv == I915_MADV_WILLNEED)
                                mark_page_accessed(page);
                        put_page(page);
-                       vaddr += PAGE_SIZE;
+
+                       src += PAGE_SIZE;
                }
                obj->mm.dirty = false;
        }
@@ -130,7 +130,9 @@ i915_gem_object_put_pages_phys(struct drm_i915_gem_object *obj,
        sg_free_table(pages);
        kfree(pages);
 
-       drm_pci_free(obj->base.dev, obj->phys_handle);
+       dma_free_coherent(&obj->base.dev->pdev->dev,
+                         roundup_pow_of_two(obj->base.size),
+                         vaddr, dma);
 }
 
 static void phys_release(struct drm_i915_gem_object *obj)
index 98305d987ac14b366ccad231752e7ed4e1be4577..4d561da3dceaded4d75c21179e52d76af2c3fa83 100644 (file)
@@ -136,7 +136,7 @@ i915_gem_phys_pwrite(struct drm_i915_gem_object *obj,
                     struct drm_i915_gem_pwrite *args,
                     struct drm_file *file)
 {
-       void *vaddr = obj->phys_handle->vaddr + args->offset;
+       void *vaddr = sg_page(obj->mm.pages->sgl) + args->offset;
        char __user *user_data = u64_to_user_ptr(args->data_ptr);
 
        /*
@@ -802,10 +802,10 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
                ret = i915_gem_gtt_pwrite_fast(obj, args);
 
        if (ret == -EFAULT || ret == -ENOSPC) {
-               if (obj->phys_handle)
-                       ret = i915_gem_phys_pwrite(obj, args, file);
-               else
+               if (i915_gem_object_has_struct_page(obj))
                        ret = i915_gem_shmem_pwrite(obj, args);
+               else
+                       ret = i915_gem_phys_pwrite(obj, args, file);
        }
 
        i915_gem_object_unpin_pages(obj);