After the previous patch the "task == tt->this_task" checks in back_trace() and is_task_active() lead to falsely "active" state if ACTIVE() && !LOCAL_ACTIVE().
This is easy to fix, but it seems to me that we can just simplify this logic and remove the tt->this_task check from back_trace(), although I'm afraid I missed something. It look really simple: if LOCAL_ACTIVE() than "this_task" is /usr/bin/crash and it is obviously running, otherwise we can never assume this if !task_has_cpu(). Not to mention that "deprecation of usage of has_cpu on non-SMP systems" looks confusing, task_has_cpu() can work without VALID_MEMBER(task_struct_has_cpu). BTW, I think that task_init() should use "on_cpu" if MEMBER_OFFSET_INIT(has_cpu) fails, but this is off-topic. Signed-off-by: Oleg Nesterov <[email protected]> --- kernel.c | 3 +-- task.c | 8 +++----- 2 files changed, 4 insertions(+), 7 deletions(-) diff --git a/kernel.c b/kernel.c index c82b854..5a392fc 100644 --- a/kernel.c +++ b/kernel.c @@ -2800,8 +2800,7 @@ back_trace(struct bt_info *bt) return; } - if (LIVE() && !(bt->flags & BT_EFRAME_SEARCH) && - ((bt->task == tt->this_task) || is_task_active(bt->task))) { + if (LIVE() && !(bt->flags & BT_EFRAME_SEARCH) && is_task_active(bt->task)) { if (BT_REFERENCE_CHECK(bt) || bt->flags & (BT_TEXT_SYMBOLS_PRINT|BT_TEXT_SYMBOLS_NOPRINT)) diff --git a/task.c b/task.c index fd4b700..a6bc9e7 100644 --- a/task.c +++ b/task.c @@ -5359,18 +5359,16 @@ is_task_active(ulong task) { int has_cpu; + if (LOCAL_ACTIVE() && (task == tt->this_task)) + return TRUE; if (DUMPFILE() && is_panic_thread(task)) return TRUE; fill_task_struct(task); - has_cpu = tt->last_task_read ? + has_cpu = tt->last_task_read ? task_has_cpu(task, tt->task_struct) : 0; - if (!(kt->flags & SMP) && !has_cpu && ACTIVE() && - (task == tt->this_task)) - has_cpu = TRUE; - return(has_cpu); } -- 2.5.0 -- Crash-utility mailing list [email protected] https://www.redhat.com/mailman/listinfo/crash-utility
