diff mbox series

[1/3] drm/i915: allow prefaulting discontiguous objects in remap_io_mapping()

Message ID 20191219112122.4892-1-abdiel.janulgue@linux.intel.com (mailing list archive)
State New, archived
Headers show
Series [1/3] drm/i915: allow prefaulting discontiguous objects in remap_io_mapping() | expand

Commit Message

Abdiel Janulgue Dec. 19, 2019, 11:21 a.m. UTC
Provide a way to set the PTE of the physical address of kernel memory in
addition to just incrementing the pfn for a page range.

Signed-off-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/gem/i915_gem_mman.c |  1 +
 drivers/gpu/drm/i915/i915_drv.h          |  4 +++-
 drivers/gpu/drm/i915/i915_mm.c           | 17 ++++++++++++++---
 3 files changed, 18 insertions(+), 4 deletions(-)

Comments

Chris Wilson Dec. 19, 2019, 11:24 a.m. UTC | #1
Quoting Abdiel Janulgue (2019-12-19 11:21:20)
> Provide a way to set the PTE of the physical address of kernel memory in
> addition to just incrementing the pfn for a page range.
> 
> Signed-off-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/gem/i915_gem_mman.c |  1 +
>  drivers/gpu/drm/i915/i915_drv.h          |  4 +++-
>  drivers/gpu/drm/i915/i915_mm.c           | 17 ++++++++++++++---
>  3 files changed, 18 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> index 879fff8adc48..aa5d6623d86c 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> @@ -360,6 +360,7 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
>                                area->vm_start + (vma->ggtt_view.partial.offset << PAGE_SHIFT),
>                                (ggtt->gmadr.start + vma->node.start) >> PAGE_SHIFT,
>                                min_t(u64, vma->size, area->vm_end - area->vm_start),
> +                              NULL,
>                                &ggtt->iomap);
>         if (ret)
>                 goto err_fence;
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 0781b6326b8c..5ee0fe4d492b 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -2015,9 +2015,11 @@ int i915_reg_read_ioctl(struct drm_device *dev, void *data,
>         intel_de_wait_for_register((dev_priv_), (reg_), (mask_), 0, (timeout_))
>  
>  /* i915_mm.c */
> +typedef unsigned long (*io_pfn_t)(struct drm_i915_gem_object *obj,
> +                                 unsigned long n);
>  int remap_io_mapping(struct vm_area_struct *vma,
>                      unsigned long addr, unsigned long pfn, unsigned long size,
> -                    struct io_mapping *iomap);
> +                    io_pfn_t fn, struct io_mapping *iomap);
>  
>  static inline int intel_hws_csb_write_index(struct drm_i915_private *i915)
>  {
> diff --git a/drivers/gpu/drm/i915/i915_mm.c b/drivers/gpu/drm/i915/i915_mm.c
> index 318562ce64c0..86a73444bed4 100644
> --- a/drivers/gpu/drm/i915/i915_mm.c
> +++ b/drivers/gpu/drm/i915/i915_mm.c
> @@ -32,7 +32,11 @@
>  struct remap_pfn {
>         struct mm_struct *mm;
>         unsigned long pfn;
> +       unsigned long start_pfn;
>         pgprot_t prot;
> +
> +       io_pfn_t fn;
> +       struct drm_i915_gem_object *obj;
>  };
>  
>  static int remap_pfn(pte_t *pte, unsigned long addr, void *data)
> @@ -41,7 +45,10 @@ static int remap_pfn(pte_t *pte, unsigned long addr, void *data)
>  
>         /* Special PTE are not associated with any struct page */
>         set_pte_at(r->mm, addr, pte, pte_mkspecial(pfn_pte(r->pfn, r->prot)));
> -       r->pfn++;
> +       if (r->fn)
> +               r->pfn = r->fn(r->obj, r->start_pfn++);

Just say no to a vfunc for every pte. retrampolines are not your friend.

I was anticipating a second remap_io_sg() that knew how to increment
onto the next sg at dma boundaries within this setter.
-Chris
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
index 879fff8adc48..aa5d6623d86c 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
@@ -360,6 +360,7 @@  static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
 			       area->vm_start + (vma->ggtt_view.partial.offset << PAGE_SHIFT),
 			       (ggtt->gmadr.start + vma->node.start) >> PAGE_SHIFT,
 			       min_t(u64, vma->size, area->vm_end - area->vm_start),
+			       NULL,
 			       &ggtt->iomap);
 	if (ret)
 		goto err_fence;
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 0781b6326b8c..5ee0fe4d492b 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2015,9 +2015,11 @@  int i915_reg_read_ioctl(struct drm_device *dev, void *data,
 	intel_de_wait_for_register((dev_priv_), (reg_), (mask_), 0, (timeout_))
 
 /* i915_mm.c */
+typedef unsigned long (*io_pfn_t)(struct drm_i915_gem_object *obj,
+				  unsigned long n);
 int remap_io_mapping(struct vm_area_struct *vma,
 		     unsigned long addr, unsigned long pfn, unsigned long size,
-		     struct io_mapping *iomap);
+		     io_pfn_t fn, struct io_mapping *iomap);
 
 static inline int intel_hws_csb_write_index(struct drm_i915_private *i915)
 {
diff --git a/drivers/gpu/drm/i915/i915_mm.c b/drivers/gpu/drm/i915/i915_mm.c
index 318562ce64c0..86a73444bed4 100644
--- a/drivers/gpu/drm/i915/i915_mm.c
+++ b/drivers/gpu/drm/i915/i915_mm.c
@@ -32,7 +32,11 @@ 
 struct remap_pfn {
 	struct mm_struct *mm;
 	unsigned long pfn;
+	unsigned long start_pfn;
 	pgprot_t prot;
+
+	io_pfn_t fn;
+	struct drm_i915_gem_object *obj;
 };
 
 static int remap_pfn(pte_t *pte, unsigned long addr, void *data)
@@ -41,7 +45,10 @@  static int remap_pfn(pte_t *pte, unsigned long addr, void *data)
 
 	/* Special PTE are not associated with any struct page */
 	set_pte_at(r->mm, addr, pte, pte_mkspecial(pfn_pte(r->pfn, r->prot)));
-	r->pfn++;
+	if (r->fn)
+		r->pfn = r->fn(r->obj, r->start_pfn++);
+	else
+		r->pfn++;
 
 	return 0;
 }
@@ -52,15 +59,17 @@  static int remap_pfn(pte_t *pte, unsigned long addr, void *data)
  * @addr: target user address to start at
  * @pfn: physical address of kernel memory
  * @size: size of map area
+ * @fn: if provided, the function translates each increment of @pfn to io pfn
  * @iomap: the source io_mapping
  *
  *  Note: this is only safe if the mm semaphore is held when called.
  */
 int remap_io_mapping(struct vm_area_struct *vma,
 		     unsigned long addr, unsigned long pfn, unsigned long size,
-		     struct io_mapping *iomap)
+		     io_pfn_t fn, struct io_mapping *iomap)
 {
-	struct remap_pfn r;
+	struct remap_pfn r = { 0 };
+	struct i915_mmap_offset *priv = vma->vm_private_data;
 	int err;
 
 #define EXPECTED_FLAGS (VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP)
@@ -69,6 +78,8 @@  int remap_io_mapping(struct vm_area_struct *vma,
 	/* We rely on prevalidation of the io-mapping to skip track_pfn(). */
 	r.mm = vma->vm_mm;
 	r.pfn = pfn;
+	r.fn = fn;
+	r.obj = priv->obj;
 	r.prot = __pgprot((pgprot_val(iomap->prot) & _PAGE_CACHE_MASK) |
 			  (pgprot_val(vma->vm_page_prot) & ~_PAGE_CACHE_MASK));