On 10/8/20 6:58 AM, Oleg Nesterov wrote:
> On 10/05, Jens Axboe wrote:
>>
>> static inline int signal_pending_state(long state, struct task_struct *p)
>> {
>> if (!(state & (TASK_INTERRUPTIBLE | TASK_WAKEKILL)))
>> return 0;
>> - if (!signal_pending(p))
>> + if (!task_sigpending(p))
>> return 0;
>
> This looks obviously wrong. Say, schedule() in TASK_INTERRUPTIBLE should
> not block if TIF_NOTIFY_SIGNAL is set.
>
> With this change set_notify_signal() will not force the task to return
> from wait_event_interruptible, mutex_lock_interruptible, etc.
True, not sure why I made the distinction there. I'll fix that one up.
>
>> return (state & TASK_INTERRUPTIBLE) || __fatal_signal_pending(p);
>> @@ -389,7 +394,7 @@ static inline bool fault_signal_pending(vm_fault_t
>> fault_flags,
>> {
>> return unlikely((fault_flags & VM_FAULT_RETRY) &&
>> (fatal_signal_pending(current) ||
>> - (user_mode(regs) && signal_pending(current))));
>> + (user_mode(regs) && task_sigpending(current))));
>
> This looks unnecessary,
Dropped
>> @@ -773,7 +773,7 @@ static int ptrace_peek_siginfo(struct task_struct *child,
>> data += sizeof(siginfo_t);
>> i++;
>>
>> - if (signal_pending(current))
>> + if (task_sigpending(current))
>
> This too.
>
> IMO, this patch should do s/signal_pending/task_sigpending/ only if it is
> strictly needed for correctness.
Agree, I'll kill the ones you identified.
--
Jens Axboe
