dma-buf: mmap support

Compared to Rob Clark's RFC I've ditched the prepare/finish hooks
and corresponding ioctls on the dma_buf file. The major reason for
that is that many people seem to be under the impression that this is
also for synchronization with outstanding asynchronous processsing.
I'm pretty massively opposed to this because:

- It boils down reinventing a new rather general-purpose userspace
  synchronization interface. If we look at things like futexes, this
  is hard to get right.
- Furthermore a lot of kernel code has to interact with this
  synchronization primitive. This smells a look like the dri1 hw_lock,
  a horror show I prefer not to reinvent.
- Even more fun is that multiple different subsystems would interact
  here, so we have plenty of opportunities to create funny deadlock
  scenarios.

I think synchronization is a wholesale different problem from data
sharing and should be tackled as an orthogonal problem.

Now we could demand that prepare/finish may only ensure cache
coherency (as Rob intended), but that runs up into the next problem:
We not only need mmap support to facilitate sw-only processing nodes
in a pipeline (without jumping through hoops by importing the dma_buf
into some sw-access only importer), which allows for a nicer
ION->dma-buf upgrade path for existing Android userspace. We also need
mmap support for existing importing subsystems to support existing
userspace libraries. And a loot of these subsystems are expected to
export coherent userspace mappings.

So prepare/finish can only ever be optional and the exporter /needs/
to support coherent mappings. Given that mmap access is always
somewhat fallback-y in nature I've decided to drop this optimization,
instead of just making it optional. If we demonstrate a clear need for
this, supported by benchmark results, we can always add it in again
later as an optional extension.

Other differences compared to Rob's RFC is the above mentioned support
for mapping a dma-buf through facilities provided by the importer.
Which results in mmap support no longer being optional.

Note that this dma-buf mmap patch does _not_ support every possible
insanity an existing subsystem could pull of with mmap: Because it
does not allow to intercept pagefaults and shoot down ptes importing
subsystems can't add some magic of their own at these points (e.g. to
automatically synchronize with outstanding rendering or set up some
special resources). I've done a cursory read through a few mmap
implementions of various subsytems and I'm hopeful that we can avoid
this (and the complexity it'd bring with it).

Additonally I've extended the documentation a bit to explain the hows
and whys of this mmap extension.

In case we ever want to add support for explicitly cache maneged
userspace mmap with a prepare/finish ioctl pair, we could specify that
userspace needs to mmap a different part of the dma_buf, e.g. the
range starting at dma_buf->size up to dma_buf->size*2. This works
because the size of a dma_buf is invariant over it's lifetime. The
exporter would obviously need to fall back to coherent mappings for
both ranges if a legacy clients maps the coherent range and the
architecture cannot suppor conflicting caching policies. Also, this
would obviously be optional and userspace needs to be able to fall
back to coherent mappings.

v2:
- Spelling fixes from Rob Clark.
- Compile fix for !DMA_BUF from Rob Clark.
- Extend commit message to explain how explicitly cache managed mmap
  support could be added later.
- Extend the documentation with implementations notes for exporters
  that need to manually fake coherency.

Change-Id: Ia8f2ae5d8a1b1c87ed12ca1c89d7bf2067239ee4
Cc: Rob Clark <rob.clark@linaro.org>
Cc: Rebecca Schultz Zavin <rebecca@android.com>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

1 files changed, 91 insertions, 7 deletions

diff --git a/Documentation/dma-buf-sharing.txt b/Documentation/dma-buf-sharing.txt
index 3bbd5c51605a..5ff4d2b84f72 100644
--- a/Documentation/dma-buf-sharing.txt
+++ b/Documentation/dma-buf-sharing.txt
@@ -29,13 +29,6 @@ The buffer-user
    in memory, mapped into its own address space, so it can access the same area
    of memory.
 
-*IMPORTANT*: [see https://lkml.org/lkml/2011/12/20/211 for more details]
-For this first version, A buffer shared using the dma_buf sharing API:
-- *may* be exported to user space using "mmap" *ONLY* by exporter, outside of
-  this framework.
-- with this new iteration of the dma-buf api cpu access from the kernel has been
-  enable, see below for the details.
-
 dma-buf operations for device dma only
 --------------------------------------
 
@@ -313,6 +306,83 @@ Access to a dma_buf from the kernel context involves three steps:
 				  enum dma_data_direction dir);
 
 
+Direct Userspace Access/mmap Support
+------------------------------------
+
+Being able to mmap an export dma-buf buffer object has 2 main use-cases:
+- CPU fallback processing in a pipeline and
+- supporting existing mmap interfaces in importers.
+
+1. CPU fallback processing in a pipeline
+
+   In many processing pipelines it is sometimes required that the cpu can access
+   the data in a dma-buf (e.g. for thumbnail creation, snapshots, ...). To avoid
+   the need to handle this specially in userspace frameworks for buffer sharing
+   it's ideal if the dma_buf fd itself can be used to access the backing storage
+   from userspace using mmap.
+
+   Furthermore Android's ION framework already supports this (and is otherwise
+   rather similar to dma-buf from a userspace consumer side with using fds as
+   handles, too). So it's beneficial to support this in a similar fashion on
+   dma-buf to have a good transition path for existing Android userspace.
+
+   No special interfaces, userspace simply calls mmap on the dma-buf fd.
+
+2. Supporting existing mmap interfaces in exporters
+
+   Similar to the motivation for kernel cpu access it is again important that
+   the userspace code of a given importing subsystem can use the same interfaces
+   with a imported dma-buf buffer object as with a native buffer object. This is
+   especially important for drm where the userspace part of contemporary OpenGL,
+   X, and other drivers is huge, and reworking them to use a different way to
+   mmap a buffer rather invasive.
+
+   The assumption in the current dma-buf interfaces is that redirecting the
+   initial mmap is all that's needed. A survey of some of the existing
+   subsystems shows that no driver seems to do any nefarious thing like syncing
+   up with outstanding asynchronous processing on the device or allocating
+   special resources at fault time. So hopefully this is good enough, since
+   adding interfaces to intercept pagefaults and allow pte shootdowns would
+   increase the complexity quite a bit.
+
+   Interface:
+      int dma_buf_mmap(struct dma_buf *, struct vm_area_struct *,
+		       unsigned long);
+
+   If the importing subsystem simply provides a special-purpose mmap call to set
+   up a mapping in userspace, calling do_mmap with dma_buf->file will equally
+   achieve that for a dma-buf object.
+
+3. Implementation notes for exporters
+
+   Because dma-buf buffers have invariant size over their lifetime, the dma-buf
+   core checks whether a vma is too large and rejects such mappings. The
+   exporter hence does not need to duplicate this check.
+
+   Because existing importing subsystems might presume coherent mappings for
+   userspace, the exporter needs to set up a coherent mapping. If that's not
+   possible, it needs to fake coherency by manually shooting down ptes when
+   leaving the cpu domain and flushing caches at fault time. Note that all the
+   dma_buf files share the same anon inode, hence the exporter needs to replace
+   the dma_buf file stored in vma->vm_file with it's own if pte shootdown is
+   requred. This is because the kernel uses the underlying inode's address_space
+   for vma tracking (and hence pte tracking at shootdown time with
+   unmap_mapping_range).
+
+   If the above shootdown dance turns out to be too expensive in certain
+   scenarios, we can extend dma-buf with a more explicit cache tracking scheme
+   for userspace mappings. But the current assumption is that using mmap is
+   always a slower path, so some inefficiencies should be acceptable.
+
+   Exporters that shoot down mappings (for any reasons) shall not do any
+   synchronization at fault time with outstanding device operations.
+   Synchronization is an orthogonal issue to sharing the backing storage of a
+   buffer and hence should not be handled by dma-buf itself. This is explictly
+   mentioned here because many people seem to want something like this, but if
+   different exporters handle this differently, buffer sharing can fail in
+   interesting ways depending upong the exporter (if userspace starts depending
+   upon this implicit synchronization).
+
 Miscellaneous notes
 -------------------
 
@@ -336,6 +406,20 @@ Miscellaneous notes
   the exporting driver to create a dmabuf fd must provide a way to let
   userspace control setting of O_CLOEXEC flag passed in to dma_buf_fd().
 
+- If an exporter needs to manually flush caches and hence needs to fake
+  coherency for mmap support, it needs to be able to zap all the ptes pointing
+  at the backing storage. Now linux mm needs a struct address_space associated
+  with the struct file stored in vma->vm_file to do that with the function
+  unmap_mapping_range. But the dma_buf framework only backs every dma_buf fd
+  with the anon_file struct file, i.e. all dma_bufs share the same file.
+
+  Hence exporters need to setup their own file (and address_space) association
+  by setting vma->vm_file and adjusting vma->vm_pgoff in the dma_buf mmap
+  callback. In the specific case of a gem driver the exporter could use the
+  shmem file already provided by gem (and set vm_pgoff = 0). Exporters can then
+  zap ptes by unmapping the corresponding range of the struct address_space
+  associated with their own file.
+
 References:
 [1] struct dma_buf_ops in include/linux/dma-buf.h
 [2] All interfaces mentioned above defined in include/linux/dma-buf.h