drm/radeon: fix wait to actually occur after the signaling callback

A normal wait adds to the front of the tail. By doing something
similar to fence_default_wait the fence code can run without racing.

This is a complete fix for "panic on suspend from KDE with radeon",
and a partial fix for "Radeon: System pauses on TAHITI". On tahiti
si_irq_set needs to be fixed too, to completely flush the writes
before radeon_fence_activity is called in radeon_fence_enable_signaling.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=90741
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=90861
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@ubuntu.com>
Reported-by: Jon Arne Jørgensen <jonjon.arnearne@gmail.com>
Reported-and-tested-by: Gustaw Smolarczyk <wielkiegie@gmail.com>
Cc: stable@vger.kernel.org (v3.18+)
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 45 insertions, 23 deletions

diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
index d13d1b5a859f..df09ca7c4889 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -1030,37 +1030,59 @@ static inline bool radeon_test_signaled(struct radeon_fence *fence)
         return test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags);
 }
 
+struct radeon_wait_cb {
+        struct fence_cb base;
+        struct task_struct *task;
+};
+
+static void
+radeon_fence_wait_cb(struct fence *fence, struct fence_cb *cb)
+{
+        struct radeon_wait_cb *wait =
+                container_of(cb, struct radeon_wait_cb, base);
+
+        wake_up_process(wait->task);
+}
+
 static signed long radeon_fence_default_wait(struct fence *f, bool intr,
                                              signed long t)
 {
         struct radeon_fence *fence = to_radeon_fence(f);
         struct radeon_device *rdev = fence->rdev;
-        bool signaled;
+        struct radeon_wait_cb cb;
 
-        fence_enable_sw_signaling(&fence->base);
+        cb.task = current;
 
-        /*
-         * This function has to return -EDEADLK, but cannot hold
-         * exclusive_lock during the wait because some callers
-         * may already hold it. This means checking needs_reset without
-         * lock, and not fiddling with any gpu internals.
-         *
-         * The callback installed with fence_enable_sw_signaling will
-         * run before our wait_event_*timeout call, so we will see
-         * both the signaled fence and the changes to needs_reset.
-         */
+        if (fence_add_callback(f, &cb.base, radeon_fence_wait_cb))
+                return t;
+
+        while (t > 0) {
+                if (intr)
+                        set_current_state(TASK_INTERRUPTIBLE);
+                else
+                        set_current_state(TASK_UNINTERRUPTIBLE);
+
+                /*
+                 * radeon_test_signaled must be called after
+                 * set_current_state to prevent a race with wake_up_process
+                 */
+                if (radeon_test_signaled(fence))
+                        break;
+
+                if (rdev->needs_reset) {
+                        t = -EDEADLK;
+                        break;
+                }
+
+                t = schedule_timeout(t);
+
+                if (t > 0 && intr && signal_pending(current))
+                        t = -ERESTARTSYS;
+        }
+
+        __set_current_state(TASK_RUNNING);
+        fence_remove_callback(f, &cb.base);
 
-        if (intr)
-                t = wait_event_interruptible_timeout(rdev->fence_queue,
-                        ((signaled = radeon_test_signaled(fence)) ||
-                         rdev->needs_reset), t);
-        else
-                t = wait_event_timeout(rdev->fence_queue,
-                        ((signaled = radeon_test_signaled(fence)) ||
-                         rdev->needs_reset), t);
-
-        if (t > 0 && !signaled)
-                return -EDEADLK;
         return t;
 }