drm/asahi: harden work-queue lifetime across teardown (does not yet fix the M1 Pro firmware crash)

drivers/gpu/drm/asahi/fw/workqueue.rs

@@ -152,7 +152,7 @@ pub(crate) mod raw {
         pub(crate) unk_94: u32,
         pub(crate) pending: AtomicU32,
         pub(crate) unk_9c: u32,
-        pub(crate) gpu_context: GpuPointer<'a, super::GpuContextData>,
+        pub(crate) gpu_context: GpuWeakPointer<super::GpuContextData>,
         pub(crate) unk_a8: U64,
         #[ver(V >= V13_2 && G < G14X)]
         pub(crate) unk_b0: u32,
@@ -171,14 +171,17 @@ impl GpuStruct for GpuContextData {
     type Raw<'a> = raw::GpuContextData;
 }
 
+/// The firmware structures of a work queue.
+///
+/// The raw structure points to the queue's GPU context weakly. The `GpuContext` keeps these
+/// structures until it has been invalidated, and only then frees its own data.
 #[versions(AGX)]
 #[derive(Debug)]
 pub(crate) struct QueueInfo {
     pub(crate) state: GpuObject<RingState>,
     pub(crate) ring: GpuArray<u64>,
     pub(crate) gpu_buf: GpuArray<u8>,
     pub(crate) notifier_list: Arc<GpuObject<event::NotifierList>>,
-    pub(crate) gpu_context: Arc<crate::workqueue::GpuContext>,
 }
 
 #[versions(AGX)]

drivers/gpu/drm/asahi/gpu.rs

@@ -260,9 +260,8 @@ pub(crate) struct GpuManager {
     stats: Arc<stats::StatsSnapshot>,
     buffer_mgr: buffer::BufferManager::ver,
     ids: SequenceIDs,
-    #[allow(clippy::vec_box)]
     #[pin]
-    garbage_contexts: Mutex<KVec<KBox<fw::types::GpuObject<fw::workqueue::GpuContextData>>>>,
+    garbage_contexts: Mutex<KVec<workqueue::RetiredContext>>,
 }
 
 /// Operations the DRM driver performs on a GPU, independent of its firmware interface.
@@ -401,8 +400,8 @@ pub(crate) trait GpuManager: Send + Sync {
     fn get_cfg(&self) -> &'static hw::HwConfig;
     /// Get the dynamic GPU configuration for this SoC.
     fn get_dyncfg(&self) -> &hw::DynConfig;
-    /// Register an unused context as garbage
-    fn free_context(&self, data: KBox<fw::types::GpuObject<fw::workqueue::GpuContextData>>);
+    /// Register an unused context, with the objects it retains, for invalidation.
+    fn free_context(&self, ctx: workqueue::RetiredContext);
 }
 
 /// Private generic trait for functions that don't need to escape this module.
@@ -1456,7 +1455,7 @@ impl GpuManager for GpuManager::ver {
         core::mem::swap(&mut *self.garbage_contexts.lock(), &mut garbage_ctx);
 
         for ctx in garbage_ctx {
-            if self.invalidate_context(&ctx).is_err() {
+            if self.invalidate_context(&ctx.data).is_err() {
                 dev_err!(
                     self.dev.as_ref(),
                     "GpuContext: Failed to invalidate GPU context!\n"
@@ -1465,6 +1464,9 @@ impl GpuManager for GpuManager::ver {
                     panic!("GPU firmware timed out");
                 }
             }
+            // The firmware no longer knows the context's queues, so their structures may go now,
+            // ahead of the context data they point to.
+            core::mem::drop(ctx.retained);
         }
 
         let mut guard = self.alloc.lock();
@@ -1733,7 +1735,7 @@ impl GpuManager for GpuManager::ver {
         &self.dyncfg
     }
 
-    fn free_context(&self, ctx: KBox<fw::types::GpuObject<fw::workqueue::GpuContextData>>) {
+    fn free_context(&self, ctx: workqueue::RetiredContext) {
         let mut garbage = self.garbage_contexts.lock();
 
         if garbage.push(ctx, GFP_KERNEL).is_err() {

drivers/gpu/drm/asahi/queue/mod.rs

@@ -525,6 +525,15 @@ impl Queue::ver {
 
         let buffer = buffer::Buffer::ver::new(gpu, alloc, ualloc.clone(), ualloc_priv, mgr)?;
 
+        let gpu_context = Arc::new(
+            workqueue::GpuContext::new(dev.clone(), alloc, buffer.any_ref())?,
+            GFP_KERNEL,
+        )?;
+        // Every command of this queue points the firmware at the notifier, and the notifier links
+        // into the notifier list of the work queues. Keep it as long as they are kept, until the
+        // context is invalidated, instead of freeing it with the last command.
+        gpu_context.retain(KBox::new(notifier.clone(), GFP_KERNEL)?)?;
+
         let mut ret = Queue::ver {
             dev: dev.clone(),
             _sched: sched,
@@ -538,11 +547,7 @@ impl Queue::ver {
             inner: QueueInner::ver {
                 dev: dev.clone(),
                 ualloc,
-                gpu_context: Arc::new(
-                    workqueue::GpuContext::new(dev.clone(), alloc, buffer.any_ref())?,
-                    GFP_KERNEL,
-                )?,
-
+                gpu_context,
                 buffer,
                 notifier_list: Arc::new(notifier_list, GFP_KERNEL)?,
                 notifier,