ACPICA: Fix AcpiWalkNamespace race condition with table unload

Added a reader/writer locking mechanism to allow multiple
concurrent namespace walks (readers), but a dynamic table unload
will have exclusive access to the namespace. This fixes a problem
where a table unload could delete the portion of the namespace that
is currently being examined by a walk.  Adds a new file, utlock.c
that implements the reader/writer lock mechanism. ACPICA BZ 749.

http://www.acpica.org/bugzilla/show_bug.cgi?id=749

Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>

1 files changed, 26 insertions, 7 deletions

diff --git a/drivers/acpi/acpica/nsxfeval.c b/drivers/acpi/acpica/nsxfeval.c
index 2583a66a60a7..045054037c2d 100644
--- a/drivers/acpi/acpica/nsxfeval.c
+++ b/drivers/acpi/acpica/nsxfeval.c
@@ -475,21 +475,40 @@ acpi_walk_namespace(acpi_object_type type,
         }
 
         /*
-         * Lock the namespace around the walk.
-         * The namespace will be unlocked/locked around each call
-         * to the user function - since this function
-         * must be allowed to make Acpi calls itself.
+         * Need to acquire the namespace reader lock to prevent interference
+         * with any concurrent table unloads (which causes the deletion of
+         * namespace objects). We cannot allow the deletion of a namespace node
+         * while the user function is using it. The exception to this are the
+         * nodes created and deleted during control method execution -- these
+         * nodes are marked as temporary nodes and are ignored by the namespace
+         * walk. Thus, control methods can be executed while holding the
+         * namespace deletion lock (and the user function can execute control
+         * methods.)
+         */
+        status = acpi_ut_acquire_read_lock(&acpi_gbl_namespace_rw_lock);
+        if (ACPI_FAILURE(status)) {
+                return status;
+        }
+
+        /*
+         * Lock the namespace around the walk. The namespace will be
+         * unlocked/locked around each call to the user function - since the user
+         * function must be allowed to make ACPICA calls itself (for example, it
+         * will typically execute control methods during device enumeration.)
          */
         status = acpi_ut_acquire_mutex(ACPI_MTX_NAMESPACE);
         if (ACPI_FAILURE(status)) {
-                return_ACPI_STATUS(status);
+                goto unlock_and_exit;
         }
 
         status = acpi_ns_walk_namespace(type, start_object, max_depth,
-                                        ACPI_NS_WALK_UNLOCK,
-                                        user_function, context, return_value);
+                                        ACPI_NS_WALK_UNLOCK, user_function,
+                                        context, return_value);
 
         (void)acpi_ut_release_mutex(ACPI_MTX_NAMESPACE);
+
+      unlock_and_exit:
+        (void)acpi_ut_release_read_lock(&acpi_gbl_namespace_rw_lock);
         return_ACPI_STATUS(status);
 }