nouveau: push event block/allowing out of the fence context

There is a deadlock between the irq and fctx locks,
the irq handling takes irq then fctx lock
the fence signalling takes fctx then irq lock

This splits the fence signalling path so the code that hits
the irq lock is done in a separate work queue.

This seems to fix crashes/hangs when using nouveau gsp with
i915 primary GPU.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-11-airlied@gmail.com

drivers/gpu/drm/nouveau/nouveau_fence.c

@@ -62,7 +62,7 @@ nouveau_fence_signal(struct nouveau_fence *fence)
 	if (test_bit(DMA_FENCE_FLAG_USER_BITS, &fence->base.flags)) {
 		struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
 
-		if (!--fctx->notify_ref)
+		if (atomic_dec_and_test(&fctx->notify_ref))
 			drop = 1;
 	}
 
@@ -103,6 +103,7 @@ nouveau_fence_context_kill(struct nouveau_fence_chan *fctx, int error)
 void
 nouveau_fence_context_del(struct nouveau_fence_chan *fctx)
 {
+	cancel_work_sync(&fctx->allow_block_work);
 	nouveau_fence_context_kill(fctx, 0);
 	nvif_event_dtor(&fctx->event);
 	fctx->dead = 1;
@@ -167,6 +168,18 @@ nouveau_fence_wait_uevent_handler(struct nvif_event *event, void *repv, u32 repc
 	return ret;
 }
 
+static void
+nouveau_fence_work_allow_block(struct work_struct *work)
+{
+	struct nouveau_fence_chan *fctx = container_of(work, struct nouveau_fence_chan,
+						       allow_block_work);
+
+	if (atomic_read(&fctx->notify_ref) == 0)
+		nvif_event_block(&fctx->event);
+	else
+		nvif_event_allow(&fctx->event);
+}
+
 void
 nouveau_fence_context_new(struct nouveau_channel *chan, struct nouveau_fence_chan *fctx)
 {
@@ -178,6 +191,7 @@ nouveau_fence_context_new(struct nouveau_channel *chan, struct nouveau_fence_cha
 	} args;
 	int ret;
 
+	INIT_WORK(&fctx->allow_block_work, nouveau_fence_work_allow_block);
 	INIT_LIST_HEAD(&fctx->flip);
 	INIT_LIST_HEAD(&fctx->pending);
 	spin_lock_init(&fctx->lock);
@@ -521,15 +535,19 @@ static bool nouveau_fence_enable_signaling(struct dma_fence *f)
 	struct nouveau_fence *fence = from_fence(f);
 	struct nouveau_fence_chan *fctx = nouveau_fctx(fence);
 	bool ret;
+	bool do_work;
 
-	if (!fctx->notify_ref++)
-		nvif_event_allow(&fctx->event);
+	if (atomic_inc_return(&fctx->notify_ref) == 0)
+		do_work = true;
 
 	ret = nouveau_fence_no_signaling(f);
 	if (ret)
 		set_bit(DMA_FENCE_FLAG_USER_BITS, &fence->base.flags);
-	else if (!--fctx->notify_ref)
-		nvif_event_block(&fctx->event);
+	else if (atomic_dec_and_test(&fctx->notify_ref))
+		do_work = true;
+
+	if (do_work)
+		schedule_work(&fctx->allow_block_work);
 
 	return ret;
 }

drivers/gpu/drm/nouveau/nouveau_fence.h

@@ -3,6 +3,7 @@
 #define __NOUVEAU_FENCE_H__
 
 #include <linux/dma-fence.h>
+#include <linux/workqueue.h>
 #include <nvif/event.h>
 
 struct nouveau_drm;
@@ -45,7 +46,9 @@ struct nouveau_fence_chan {
 	char name[32];
 
 	struct nvif_event event;
-	int notify_ref, dead, killed;
+	struct work_struct allow_block_work;
+	atomic_t notify_ref;
+	int dead, killed;
 };
 
 struct nouveau_fence_priv {