diff --git a/drivers/gpu/drm/panthor/panthor_drv.c b/drivers/gpu/drm/panthor/panthor_drv.c
index 929fea7977e25..e6c50c97b9c7d 100644
--- a/drivers/gpu/drm/panthor/panthor_drv.c
+++ b/drivers/gpu/drm/panthor/panthor_drv.c
@@ -85,15 +85,15 @@ panthor_get_uobj_array(const struct drm_panthor_obj_array *in, u32 min_stride,
 	int ret = 0;
 	void *out_alloc;
 
+	if (!in->count)
+		return NULL;
+
 	/* User stride must be at least the minimum object size, otherwise it might
 	 * lack useful information.
 	 */
 	if (in->stride < min_stride)
 		return ERR_PTR(-EINVAL);
 
-	if (!in->count)
-		return NULL;
-
 	out_alloc = kvmalloc_array(in->count, obj_size, GFP_KERNEL);
 	if (!out_alloc)
 		return ERR_PTR(-ENOMEM);
diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c
index 857f3f11258aa..ef232c0c20493 100644
--- a/drivers/gpu/drm/panthor/panthor_fw.c
+++ b/drivers/gpu/drm/panthor/panthor_fw.c
@@ -1089,6 +1089,12 @@ int panthor_fw_post_reset(struct panthor_device *ptdev)
 		panthor_fw_stop(ptdev);
 		ptdev->fw->fast_reset = false;
 		drm_err(&ptdev->base, "FW fast reset failed, trying a slow reset");
+
+		ret = panthor_vm_flush_all(ptdev->fw->vm);
+		if (ret) {
+			drm_err(&ptdev->base, "FW slow reset failed (couldn't flush FW's AS l2cache)");
+			return ret;
+		}
 	}
 
 	/* Reload all sections, including RO ones. We're not supposed
@@ -1099,7 +1105,7 @@ int panthor_fw_post_reset(struct panthor_device *ptdev)
 
 	ret = panthor_fw_start(ptdev);
 	if (ret) {
-		drm_err(&ptdev->base, "FW slow reset failed");
+		drm_err(&ptdev->base, "FW slow reset failed (couldn't start the FW )");
 		return ret;
 	}
 
diff --git a/drivers/gpu/drm/panthor/panthor_mmu.c b/drivers/gpu/drm/panthor/panthor_mmu.c
index 15be29ff46c39..4f0f9cf6426b6 100644
--- a/drivers/gpu/drm/panthor/panthor_mmu.c
+++ b/drivers/gpu/drm/panthor/panthor_mmu.c
@@ -873,14 +873,27 @@ static int panthor_vm_flush_range(struct panthor_vm *vm, u64 iova, u64 size)
 	if (!drm_dev_enter(&ptdev->base, &cookie))
 		return 0;
 
-	/* Flush the PTs only if we're already awake */
-	if (pm_runtime_active(ptdev->base.dev))
-		ret = mmu_hw_do_operation(vm, iova, size, AS_COMMAND_FLUSH_PT);
+	/*
+	 * If we made it this far, that means the device is awake, because
+	 * upon device suspension, all active VMs are given an AS id of -1
+	 */
+	ret = mmu_hw_do_operation(vm, iova, size, AS_COMMAND_FLUSH_PT);
 
 	drm_dev_exit(cookie);
 	return ret;
 }
 
+/**
+ * panthor_vm_flush_all() - Flush L2 caches for the entirety of a VM's AS
+ * @vm: VM whose cache to flush
+ *
+ * Return: 0 on success, a negative error code if flush failed.
+ */
+int panthor_vm_flush_all(struct panthor_vm *vm)
+{
+	return panthor_vm_flush_range(vm, vm->base.mm_start, vm->base.mm_range);
+}
+
 static int panthor_vm_unmap_pages(struct panthor_vm *vm, u64 iova, u64 size)
 {
 	struct panthor_device *ptdev = vm->ptdev;
diff --git a/drivers/gpu/drm/panthor/panthor_mmu.h b/drivers/gpu/drm/panthor/panthor_mmu.h
index e6a2198b009b7..10635cc8a94df 100644
--- a/drivers/gpu/drm/panthor/panthor_mmu.h
+++ b/drivers/gpu/drm/panthor/panthor_mmu.h
@@ -31,6 +31,7 @@ panthor_vm_get_bo_for_va(struct panthor_vm *vm, u64 va, u64 *bo_offset);
 int panthor_vm_active(struct panthor_vm *vm);
 void panthor_vm_idle(struct panthor_vm *vm);
 int panthor_vm_as(struct panthor_vm *vm);
+int panthor_vm_flush_all(struct panthor_vm *vm);
 
 struct panthor_heap_pool *
 panthor_vm_get_heap_pool(struct panthor_vm *vm, bool create);
diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
index 79ffcbc41d78e..463bcd3cf00f3 100644
--- a/drivers/gpu/drm/panthor/panthor_sched.c
+++ b/drivers/gpu/drm/panthor/panthor_sched.c
@@ -458,6 +458,16 @@ struct panthor_queue {
 		/** @seqno: Sequence number of the last initialized fence. */
 		atomic64_t seqno;
 
+		/**
+		 * @last_fence: Fence of the last submitted job.
+		 *
+		 * We return this fence when we get an empty command stream.
+		 * This way, we are guaranteed that all earlier jobs have completed
+		 * when drm_sched_job::s_fence::finished without having to feed
+		 * the CS ring buffer with a dummy job that only signals the fence.
+		 */
+		struct dma_fence *last_fence;
+
 		/**
 		 * @in_flight_jobs: List containing all in-flight jobs.
 		 *
@@ -829,6 +839,9 @@ static void group_free_queue(struct panthor_group *group, struct panthor_queue *
 	panthor_kernel_bo_destroy(queue->ringbuf);
 	panthor_kernel_bo_destroy(queue->iface.mem);
 
+	/* Release the last_fence we were holding, if any. */
+	dma_fence_put(queue->fence_ctx.last_fence);
+
 	kfree(queue);
 }
 
@@ -2784,9 +2797,6 @@ static void group_sync_upd_work(struct work_struct *work)
 
 		spin_lock(&queue->fence_ctx.lock);
 		list_for_each_entry_safe(job, job_tmp, &queue->fence_ctx.in_flight_jobs, node) {
-			if (!job->call_info.size)
-				continue;
-
 			if (syncobj->seqno < job->done_fence->seqno)
 				break;
 
@@ -2865,11 +2875,14 @@ queue_run_job(struct drm_sched_job *sched_job)
 	static_assert(sizeof(call_instrs) % 64 == 0,
 		      "call_instrs is not aligned on a cacheline");
 
-	/* Stream size is zero, nothing to do => return a NULL fence and let
-	 * drm_sched signal the parent.
+	/* Stream size is zero, nothing to do except making sure all previously
+	 * submitted jobs are done before we signal the
+	 * drm_sched_job::s_fence::finished fence.
 	 */
-	if (!job->call_info.size)
-		return NULL;
+	if (!job->call_info.size) {
+		job->done_fence = dma_fence_get(queue->fence_ctx.last_fence);
+		return dma_fence_get(job->done_fence);
+	}
 
 	ret = pm_runtime_resume_and_get(ptdev->base.dev);
 	if (drm_WARN_ON(&ptdev->base, ret))
@@ -2926,8 +2939,13 @@ queue_run_job(struct drm_sched_job *sched_job)
 			pm_runtime_get(ptdev->base.dev);
 			sched->pm.has_ref = true;
 		}
+		panthor_devfreq_record_busy(sched->ptdev);
 	}
 
+	/* Update the last fence. */
+	dma_fence_put(queue->fence_ctx.last_fence);
+	queue->fence_ctx.last_fence = dma_fence_get(job->done_fence);
+
 	done_fence = dma_fence_get(job->done_fence);
 
 out_unlock:
@@ -3378,10 +3396,15 @@ panthor_job_create(struct panthor_file *pfile,
 		goto err_put_job;
 	}
 
-	job->done_fence = kzalloc(sizeof(*job->done_fence), GFP_KERNEL);
-	if (!job->done_fence) {
-		ret = -ENOMEM;
-		goto err_put_job;
+	/* Empty command streams don't need a fence, they'll pick the one from
+	 * the previously submitted job.
+	 */
+	if (job->call_info.size) {
+		job->done_fence = kzalloc(sizeof(*job->done_fence), GFP_KERNEL);
+		if (!job->done_fence) {
+			ret = -ENOMEM;
+			goto err_put_job;
+		}
 	}
 
 	ret = drm_sched_job_init(&job->base,