[PATCH] drm.rst: Some notes

[Daniel Vetter]

Some early notes from my first few pull requests. Lots of things that
need to be discussed/clarified/agreed upon, but should be good for a
starting point at least.

Not entirely sure how we should split this all up, so all in one file
for now.

Cc: Dave Airlie <airlied at gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter at intel.com>
---
 drm.rst   | 66 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 index.rst |  1 +
 2 files changed, 67 insertions(+)
 create mode 100644 drm.rst

diff --git a/drm.rst b/drm.rst
new file mode 100644
index 000000000000..473e9b102010
--- /dev/null
+++ b/drm.rst
@@ -0,0 +1,66 @@
+drm
+===
+
+Introduction
+------------
+
+drm is the overall graphics integration tree, and as such works slightly
+differently than feature trees managed with dim_:
+
+- Normally only takes pull requests.
+
+- Freezes for features from -rc6 to the end of the merge window as kernel
+  subsystem trees usually do. There`s no constantly open feature branch. After
+  -rc6 `drm-next` only accepts bugfixes and smaller cleanups aimed for the merge
+  window.
+
+- Doesn`t have committers, just maintainers, since the pull request load is
+  fairly minimal (for now). To keep it that way small trees are encouraged to
+  collaborate together in drm-misc or other groups of drivers.
+
+.. _dim: dim.html
+
+Backmerges
+----------
+
+All backmerges for `-next` trees should first land in `drm-next`, with an
+explicit merge commit.  This includes pull requests from driver trees based on
+newer upstream. In that case first apply the backmerge, then take the pull
+request.
+
+Only except is right after -rc1 when `drm-next` reopens for features, where a
+fast-forward is all that's needed.
+
+Only backmerge tagged releases.
+
+Pull Request Review
+-------------------
+
+Special care should be taken to review commits which:
+
+- Touch files outside of what the maintainer maintains (drm core code, other
+  drivers, or even other subsystems).
+
+- Not reviewed patches - occasionally the lack of review is a process fumble and
+  patches never even made it to any list.
+
+- Changing uapi. Look both for include/uapi and anything adding new KMS
+  properties.
+
+- Check for last-minute rebases (all the patches have roughly the same commit
+  timestamp). Especially rebasing onto latest upstream right before sending out
+  is discouraged by Linus (since it invalidates the testing that happened).
+
+FIXME: Script as much as possible of the above.
+
+Opens
+-----
+
+- Lots of the above needs to be discussed.
+
+- Hard rule against being both drm and sub-tree maintainer, to prevent glaring
+  conflicts of interest? Commit rights for Dave in drm-misc?
+
+- To rebase or not to rebase (probably no, except the tree is on fire)
+
+- Recipient list for pulls to Linus (Daniel botched this, should be scripted?)
diff --git a/index.rst b/index.rst
index 48982cbde5a2..14086bb4ae68 100644
--- a/index.rst
+++ b/index.rst
@@ -24,6 +24,7 @@ Contents:
 
    repositories
    drm-tip
+   drm
    drm-misc
    drm-intel
    committer-guidelines
-- 
2.20.0.rc1