Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1491268 > unrolled thread

[PATCH v4 0/3] drm/nouveau: set DMA mask before mapping scratch page

Back to article view | Back to linux.kernel

Contents

  [PATCH v4 0/3] drm/nouveau: set DMA mask before mapping scratch page Ard Biesheuvel <ard.biesheuvel@linaro.org> - 2016-09-26 14:40 +0200
    [PATCH v4 2/3] drm/nouveau/fb/gf100: defer DMA mapping of scratch page to init() hook Ard Biesheuvel <ard.biesheuvel@linaro.org> - 2016-09-26 14:40 +0200
      Re: [Nouveau] [PATCH v4 2/3] drm/nouveau/fb/gf100: defer DMA mapping  of scratch page to init() hook Alexandre Courbot <gnurou@gmail.com> - 2016-10-03 07:50 +0200

#1491268 — [PATCH v4 0/3] drm/nouveau: set DMA mask before mapping scratch page

This v4 is now a 3 piece series, after Alexandre pointed out that both
GF 100 and NV50 are affected by the same issue, and that a related issue
has been solved already for Tegra in commit 9d0394c6bed5
("drm/nouveau/instmem/gk20a: set DMA mask early").

The issue that this series addresses is the fact that the Nouveau driver
invokes the DMA API before setting the DMA mask. In both cases addressed
here, these are simply static bidirectional mappings of scratch pages whose
purpose is not well understood, and in most cases, it does not matter that
these pages are always allocated below 4 GB even if the hardware can access
memory much higher up.

However, on platforms without any RAM below 4 GB, the preliminary DMA mask
of 32 is preventing the nouveau driver from loading on GF100 and NV50
hardware with an error like the following one:

   nouveau 0000:02:00.0: enabling device (0000 -> 0003)
   nouveau 0000:02:00.0: NVIDIA GT218 (0a8280b1)
   nouveau 0000:02:00.0: bios: version 70.18.a6.00.00
   nouveau 0000:02:00.0: fb ctor failed, -14
   nouveau: probe of 0000:02:00.0 failed with error -14

So fix this by setting a preliminary DMA mask based on the MMU device 'dma_bits'
property (patch #1), and postpone mapping the scratch pages to the respective
FB .init() hooks. (#2 and #3)

v4: split and move dma_set_mask to probe hook (Alexander)
v3: rework code to get rid of DMA_ERROR_CODE references, which is not
    defined on all architectures
v2: replace incorrect comparison of dma_addr_t type var against NULL

Ard Biesheuvel (3):
  drm/nouveau: set streaming DMA mask early
  drm/nouveau/fb/gf100: defer DMA mapping of scratch page to init() hook
  drm/nouveau/fb/nv50: defer DMA mapping of scratch page to init() hook

 drivers/gpu/drm/nouveau/nouveau_drm.c          | 11 +++++++
 drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c | 26 +++++++++++------
 drivers/gpu/drm/nouveau/nvkm/subdev/fb/nv50.c  | 30 +++++++++++++-------
 3 files changed, 48 insertions(+), 19 deletions(-)

-- 
2.7.4

[toc] | [next] | [standalone]

#1491270 — [PATCH v4 2/3] drm/nouveau/fb/gf100: defer DMA mapping of scratch page to init() hook

The 100c10 scratch page is mapped using dma_map_page() before the TTM
layer has had a chance to set the DMA mask. This means we are still
running with the default of 32 when this code executes, and this causes
problems for platforms with no memory below 4 GB (such as AMD Seattle)

So move the dma_map_page() to the .init hook, which executes after the
DMA mask has been set.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
---
 drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c | 26 ++++++++++++++------
 1 file changed, 18 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
index 76433cc66fff..5c8132873e60 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
@@ -93,7 +93,18 @@ gf100_fb_init(struct nvkm_fb *base)
 	struct gf100_fb *fb = gf100_fb(base);
 	struct nvkm_device *device = fb->base.subdev.device;
 
-	if (fb->r100c10_page)
+	if (!fb->r100c10) {
+		dma_addr_t addr = dma_map_page(device->dev, fb->r100c10_page, 0,
+					       PAGE_SIZE, DMA_BIDIRECTIONAL);
+		if (!dma_mapping_error(device->dev, addr)) {
+			fb->r100c10 = addr;
+		} else {
+			nvkm_warn(&fb->base.subdev,
+				 "dma_map_page() failed on 100c10 page\n");
+		}
+	}
+
+	if (fb->r100c10)
 		nvkm_wr32(device, 0x100c10, fb->r100c10 >> 8);
 }
 
@@ -103,12 +114,13 @@ gf100_fb_dtor(struct nvkm_fb *base)
 	struct gf100_fb *fb = gf100_fb(base);
 	struct nvkm_device *device = fb->base.subdev.device;
 
-	if (fb->r100c10_page) {
+	if (fb->r100c10) {
 		dma_unmap_page(device->dev, fb->r100c10, PAGE_SIZE,
 			       DMA_BIDIRECTIONAL);
-		__free_page(fb->r100c10_page);
 	}
 
+	__free_page(fb->r100c10_page);
+
 	return fb;
 }
 
@@ -124,11 +136,9 @@ gf100_fb_new_(const struct nvkm_fb_func *func, struct nvkm_device *device,
 	*pfb = &fb->base;
 
 	fb->r100c10_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
-	if (fb->r100c10_page) {
-		fb->r100c10 = dma_map_page(device->dev, fb->r100c10_page, 0,
-					   PAGE_SIZE, DMA_BIDIRECTIONAL);
-		if (dma_mapping_error(device->dev, fb->r100c10))
-			return -EFAULT;
+	if (!fb->r100c10_page) {
+		nvkm_error(&fb->base.subdev, "failed 100c10 page alloc\n");
+		return -ENOMEM;
 	}
 
 	return 0;
-- 
2.7.4

[toc] | [prev] | [next] | [standalone]

#1494704 — Re: [Nouveau] [PATCH v4 2/3] drm/nouveau/fb/gf100: defer DMA mapping of scratch page to init() hook

On Mon, Sep 26, 2016 at 9:32 PM, Ard Biesheuvel
<ard.biesheuvel@linaro.org> wrote:
> The 100c10 scratch page is mapped using dma_map_page() before the TTM
> layer has had a chance to set the DMA mask. This means we are still
> running with the default of 32 when this code executes, and this causes
> problems for platforms with no memory below 4 GB (such as AMD Seattle)
>
> So move the dma_map_page() to the .init hook, which executes after the
> DMA mask has been set.
>
> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> ---
>  drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c | 26 ++++++++++++++------
>  1 file changed, 18 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
> index 76433cc66fff..5c8132873e60 100644
> --- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
> +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
> @@ -93,7 +93,18 @@ gf100_fb_init(struct nvkm_fb *base)
>         struct gf100_fb *fb = gf100_fb(base);
>         struct nvkm_device *device = fb->base.subdev.device;
>
> -       if (fb->r100c10_page)
> +       if (!fb->r100c10) {
> +               dma_addr_t addr = dma_map_page(device->dev, fb->r100c10_page, 0,
> +                                              PAGE_SIZE, DMA_BIDIRECTIONAL);
> +               if (!dma_mapping_error(device->dev, addr)) {
> +                       fb->r100c10 = addr;
> +               } else {
> +                       nvkm_warn(&fb->base.subdev,
> +                                "dma_map_page() failed on 100c10 page\n");
> +               }
> +       }
> +
> +       if (fb->r100c10)
>                 nvkm_wr32(device, 0x100c10, fb->r100c10 >> 8);

gf100_fb_oneinit() seems to be a better place for this, since it will
be executed exactly once, which is what you want for a memory
allocation. As you can see other memory allocations are also performed
there, which hints it should have been done there (and not in ctor) in
the first place. Maybe you can also move the alloc_page() there so
everything is done in the same place.

>  }
>
> @@ -103,12 +114,13 @@ gf100_fb_dtor(struct nvkm_fb *base)
>         struct gf100_fb *fb = gf100_fb(base);
>         struct nvkm_device *device = fb->base.subdev.device;
>
> -       if (fb->r100c10_page) {
> +       if (fb->r100c10) {
>                 dma_unmap_page(device->dev, fb->r100c10, PAGE_SIZE,
>                                DMA_BIDIRECTIONAL);
> -               __free_page(fb->r100c10_page);
>         }
>
> +       __free_page(fb->r100c10_page);
> +

If you move the allocation/mapping to gf100_fb_oneinit() then I
suppose you don't need this change.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web