target/linux/bcm27xx/patches-5.4/950-0729-udmabuf-fix-dma-buf-cpu-access.patch

   1 From 153f1e1def226f87cc4c307d2806b000a820f64b Mon Sep 17 00:00:00 2001
   2 From: Gurchetan Singh <gurchetansingh@chromium.org>
   3 Date: Tue, 17 Dec 2019 15:02:28 -0800
   4 Subject: [PATCH] udmabuf: fix dma-buf cpu access
   5
   6 Commit 1ffe09590121fbb3786d6c860acdd200f7ab095c upstream.
   7
   8 I'm just going to put Chia's review comment here since it sums
   9 the issue rather nicely:
  10
  11 "(1) Semantically, a dma-buf is in DMA domain.  CPU access from the
  12 importer must be surrounded by {begin,end}_cpu_access.  This gives the
  13 exporter a chance to move the buffer to the CPU domain temporarily.
  14
  15 (2) When the exporter itself has other means to do CPU access, it is
  16 only reasonable for the exporter to move the buffer to the CPU domain
  17 before access, and to the DMA domain after access.  The exporter can
  18 potentially reuse {begin,end}_cpu_access for that purpose.
  19
  20 Because of (1), udmabuf does need to implement the
  21 {begin,end}_cpu_access hooks.  But "begin" should mean
  22 dma_sync_sg_for_cpu and "end" should mean dma_sync_sg_for_device.
  23
  24 Because of (2), if userspace wants to continuing accessing through the
  25 memfd mapping, it should call udmabuf's {begin,end}_cpu_access to
  26 avoid cache issues."
  27
  28 Reported-by: Chia-I Wu <olvaffe@gmail.com>
  29 Suggested-by: Chia-I Wu <olvaffe@gmail.com>
  30 Fixes: 284562e1f348 ("udmabuf: implement begin_cpu_access/end_cpu_access hooks")
  31 Signed-off-by: Gurchetan Singh <gurchetansingh@chromium.org>
  32 Link: http://patchwork.freedesktop.org/patch/msgid/20191217230228.453-1-gurchetansingh@chromium.org
  33 Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
  34 ---
  35  drivers/dma-buf/udmabuf.c | 7 +++----
  36  1 file changed, 3 insertions(+), 4 deletions(-)
  37
  38 --- a/drivers/dma-buf/udmabuf.c
  39 +++ b/drivers/dma-buf/udmabuf.c
  40 @@ -122,9 +122,8 @@ static int begin_cpu_udmabuf(struct dma_
  41                 if (IS_ERR(ubuf->sg))
  42                         return PTR_ERR(ubuf->sg);
  43         } else {
  44 -               dma_sync_sg_for_device(dev, ubuf->sg->sgl,
  45 -                                      ubuf->sg->nents,
  46 -                                      direction);
  47 +               dma_sync_sg_for_cpu(dev, ubuf->sg->sgl, ubuf->sg->nents,
  48 +                                   direction);
  49         }
  50
  51         return 0;
  52 @@ -139,7 +138,7 @@ static int end_cpu_udmabuf(struct dma_bu
  53         if (!ubuf->sg)
  54                 return -EINVAL;
  55
  56 -       dma_sync_sg_for_cpu(dev, ubuf->sg->sgl, ubuf->sg->nents, direction);
  57 +       dma_sync_sg_for_device(dev, ubuf->sg->sgl, ubuf->sg->nents, direction);
  58         return 0;
  59  }
  60