On Mon, 28 Feb 2000, Andrea Arcangeli wrote:
> This is the diff for the free_inode conditional schedule. The spin_unlock
> thing probably couldn't harm since such code is all under the big kernel
> lock too in 2.2.x.
>
> --- 2.2.14/fs/inode.c Fri Jan 7 18:19:18 2000
> +++ /tmp/inode.c Mon Feb 28 19:57:05 2000
> @@ -347,7 +347,9 @@
> {
> struct list_head *entry;
> int found = 0;
> + int resched = 0;
>
> + again:
> entry = inode_in_use.next;
> while (entry != &inode_in_use) {
> struct list_head *tmp = entry;
> @@ -361,6 +363,18 @@
> list_add(tmp, freeable);
> list_entry(tmp, struct inode, i_list)->i_state = I_FREEING;
> found++;
> +
> + if (current->need_resched && !resched) {
> + INODE(tmp)->i_count++;
> + spin_unlock(&inode_lock);
> +
> + schedule();
> + iput(INODE(tmp));
> +
> + spin_lock(&inode_lock);
> + resched = 1;
> + goto again;
> + }
> }
>
> return found;
>
I am testing this code now. This code appears to allow a long latency
path. If you have 2 SCHED_FIFO tasks which are both ready to run and
the kernel is in this (long) free_inode loop, one task will get scheduled
and if it finishes quickly the other task will have to wait.
Could you explain again what the problem is when we always allow
schedule?
Wm
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 21:00:22 EST