Re: [PATCH][3/4] poll()/select() timeout behavior

["Richard B. Johnson" <root@chaos.analogic.com>]

On Fri, 20 Feb 2004, Bill Rugolsky Jr. wrote:

> This patch changes select() and poll() to not wait forever when a valid,
> but large timeout value is supplied.  The SUSv3 manual page for select(2)
> states:
>
>   "If the timeout argument specifies a timeout interval greater than the
>   implementation-defined maximum value, the maximum value shall be used as
>   the actual timeout value."
>
> Both select() and poll() have a well-defined mechanism to wait forever,
> so there is no need for the existing behavior.
>
> Please apply.
>
>     Bill Rugolsky

The "well-defined mechanism" to wait forever for poll() is to
use ANY negative value as the timeout value. If you chose -1,
that, currently is MAX_SCHEDULE_TIMEOUT on Linux 2.4.24.
Don't your changes muck this up or is "<0" checked somewhere
else?

>
> --- linux/fs/select.c	2004-02-20 14:27:24.784616879 -0500
> +++ linux/fs/select.c	2004-02-20 14:27:28.264660774 -0500
> @@ -316,6 +316,8 @@
>  		if ((unsigned long) sec < (MAX_SCHEDULE_TIMEOUT-1) / HZ - 1) {
>  			timeout = ROUND_UP(usec, 1000000/HZ);
>  			timeout += sec * (unsigned long) HZ;
> +		} else {
> +			timeout = MAX_SCHEDULE_TIMEOUT-1;
>  		}
>  	}
>
> @@ -476,7 +478,7 @@
>  			if (seconds <= (MAX_SCHEDULE_TIMEOUT-2) / HZ - 1)
>  				timeout += seconds*HZ;
>  			else
> -				timeout = MAX_SCHEDULE_TIMEOUT;
> +				timeout = MAX_SCHEDULE_TIMEOUT-1;
>  		}
>  	}
>

Cheers,
Dick Johnson
Penguin : Linux version 2.4.24 on an i686 machine (797.90 BogoMips).
            Note 96.31% of all statistics are fiction.