return IRQ_HANDLED;
}
+/*
+ * hl_ts_free_objects - handler of the free objects workqueue.
+ * This function should put refcount to objects that the registration node
+ * took refcount to them.
+ * @work: workqueue object pointer
+ */
+static void hl_ts_free_objects(struct work_struct *work)
+{
+ struct timestamp_reg_work_obj *job =
+ container_of(work, struct timestamp_reg_work_obj, free_obj);
+ struct timestamp_reg_free_node *free_obj, *temp_free_obj;
+ struct list_head *free_list_head = job->free_obj_head;
+ struct hl_device *hdev = job->hdev;
+
+ list_for_each_entry_safe(free_obj, temp_free_obj, free_list_head, free_objects_node) {
+ dev_dbg(hdev->dev, "About to put refcount to ts_buff (%p) cq_cb(%p)\n",
+ free_obj->ts_buff,
+ free_obj->cq_cb);
+
+ hl_ts_put(free_obj->ts_buff);
+ hl_cb_put(free_obj->cq_cb);
+ kfree(free_obj);
+ }
+
+ kfree(free_list_head);
+ kfree(job);
+}
+
+/*
+ * This function called with spin_lock of wait_list_lock taken
+ * This function will set timestamp and delete the registration node from the
+ * wait_list_lock.
+ * and since we're protected with spin_lock here, so we cannot just put the refcount
+ * for the objects here, since the release function may be called and it's also a long
+ * logic (which might sleep also) that cannot be handled in irq context.
+ * so here we'll be filling a list with nodes of "put" jobs and then will send this
+ * list to a dedicated workqueue to do the actual put.
+ */
+static int handle_registration_node(struct hl_device *hdev, struct hl_user_pending_interrupt *pend,
+ struct list_head **free_list)
+{
+ struct timestamp_reg_free_node *free_node;
+ u64 timestamp;
+
+ if (!(*free_list)) {
+ /* Alloc/Init the timestamp registration free objects list */
+ *free_list = kmalloc(sizeof(struct list_head), GFP_ATOMIC);
+ if (!(*free_list))
+ return -ENOMEM;
+
+ INIT_LIST_HEAD(*free_list);
+ }
+
+ free_node = kmalloc(sizeof(*free_node), GFP_ATOMIC);
+ if (!free_node)
+ return -ENOMEM;
+
+ timestamp = ktime_get_ns();
+
+ *pend->ts_reg_info.timestamp_kernel_addr = timestamp;
+
+ dev_dbg(hdev->dev, "Timestamp is set to ts cb address (%p), ts: 0x%llx\n",
+ pend->ts_reg_info.timestamp_kernel_addr,
+ *(u64 *)pend->ts_reg_info.timestamp_kernel_addr);
+
+ list_del(&pend->wait_list_node);
+
+ /* Mark kernel CB node as free */
+ pend->ts_reg_info.in_use = 0;
+
+ /* Putting the refcount for ts_buff and cq_cb objects will be handled
+ * in workqueue context, just add job to free_list.
+ */
+ free_node->ts_buff = pend->ts_reg_info.ts_buff;
+ free_node->cq_cb = pend->ts_reg_info.cq_cb;
+ list_add(&free_node->free_objects_node, *free_list);
+
+ return 0;
+}
+
static void handle_user_cq(struct hl_device *hdev,
struct hl_user_interrupt *user_cq)
{
- struct hl_user_pending_interrupt *pend;
+ struct hl_user_pending_interrupt *pend, *temp_pend;
+ struct list_head *ts_reg_free_list_head = NULL;
+ struct timestamp_reg_work_obj *job;
+ bool reg_node_handle_fail = false;
ktime_t now = ktime_get();
+ int rc;
+
+ /* For registration nodes:
+ * As part of handling the registration nodes, we should put refcount to
+ * some objects. the problem is that we cannot do that under spinlock
+ * or in irq handler context at all (since release functions are long and
+ * might sleep), so we will need to handle that part in workqueue context.
+ * To avoid handling kmalloc failure which compels us rolling back actions
+ * and move nodes hanged on the free list back to the interrupt wait list
+ * we always alloc the job of the WQ at the beginning.
+ */
+ job = kmalloc(sizeof(*job), GFP_ATOMIC);
+ if (!job)
+ return;
spin_lock(&user_cq->wait_list_lock);
- list_for_each_entry(pend, &user_cq->wait_list_head, wait_list_node) {
- if ((pend->cq_kernel_addr &&
- *(pend->cq_kernel_addr) >= pend->cq_target_value) ||
+ list_for_each_entry_safe(pend, temp_pend, &user_cq->wait_list_head, wait_list_node) {
+ if ((pend->cq_kernel_addr && *(pend->cq_kernel_addr) >= pend->cq_target_value) ||
!pend->cq_kernel_addr) {
- pend->fence.timestamp = now;
- complete_all(&pend->fence.completion);
+ if (pend->ts_reg_info.ts_buff) {
+ if (!reg_node_handle_fail) {
+ rc = handle_registration_node(hdev, pend,
+ &ts_reg_free_list_head);
+ if (rc)
+ reg_node_handle_fail = true;
+ }
+ } else {
+ /* Handle wait target value node */
+ pend->fence.timestamp = now;
+ complete_all(&pend->fence.completion);
+ }
}
}
spin_unlock(&user_cq->wait_list_lock);
+
+ if (ts_reg_free_list_head) {
+ INIT_WORK(&job->free_obj, hl_ts_free_objects);
+ job->free_obj_head = ts_reg_free_list_head;
+ job->hdev = hdev;
+ queue_work(hdev->ts_free_obj_wq, &job->free_obj);
+ } else {
+ kfree(job);
+ }
}
/**
projects / linux-2.6-microblaze.git / blobdiff