drm/i915: Document locking guidelines

To ensure cross-driver locking compatibility, document the expected
guidelines for implementing the GEM locking in i915. Note that this
is a description of how things should end up after being reworked,
and does not reflect the current state of things.

v2: Use rst note:: tag (Rodrigo)

Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Dave Airlie <airlied@redhat.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
Cc: CQ Tang <cq.tang@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Acked-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190830105053.17491-1-joonas.lahtinen@linux.intel.com

1 files changed, 46 insertions, 0 deletions

diff --git a/Documentation/gpu/i915.rst b/Documentation/gpu/i915.rst
index 429b08aac797..33cc6ddf8f64 100644
--- a/Documentation/gpu/i915.rst
+++ b/Documentation/gpu/i915.rst
@@ -329,6 +329,52 @@ for execution also include a list of all locations within buffers that
 refer to GPU-addresses so that the kernel can edit the buffer correctly.
 This process is dubbed relocation.
 
+Locking Guidelines
+------------------
+
+.. note::
+   This is a description of how the locking should be after
+   refactoring is done. Does not necessarily reflect what the locking
+   looks like while WIP.
+
+#. All locking rules and interface contracts with cross-driver interfaces
+   (dma-buf, dma_fence) need to be followed.
+
+#. No struct_mutex anywhere in the code
+
+#. dma_resv will be the outermost lock (when needed) and ww_acquire_ctx
+   is to be hoisted at highest level and passed down within i915_gem_ctx
+   in the call chain
+
+#. While holding lru/memory manager (buddy, drm_mm, whatever) locks
+   system memory allocations are not allowed
+
+	* Enforce this by priming lockdep (with fs_reclaim). If we
+	  allocate memory while holding these looks we get a rehash
+	  of the shrinker vs. struct_mutex saga, and that would be
+	  real bad.
+
+#. Do not nest different lru/memory manager locks within each other.
+   Take them in turn to update memory allocations, relying on the object’s
+   dma_resv ww_mutex to serialize against other operations.
+
+#. The suggestion for lru/memory managers locks is that they are small
+   enough to be spinlocks.
+
+#. All features need to come with exhaustive kernel selftests and/or
+   IGT tests when appropriate
+
+#. All LMEM uAPI paths need to be fully restartable (_interruptible()
+   for all locks/waits/sleeps)
+
+	* Error handling validation through signal injection.
+	  Still the best strategy we have for validating GEM uAPI
+          corner cases.
+	  Must be excessively used in the IGT, and we need to check
+	  that we really have full path coverage of all error cases.
+
+	* -EDEADLK handling with ww_mutex
+
 GEM BO Management Implementation Details
 ----------------------------------------