On Mon, 5 May 1997, David S. Miller wrote:
> Hubert, can you try this patch instead, trigger the high stress
> situations you describe, and let me know what the kernel messages say?
Ok, currently compiling kernel.
BTW: The error not only shows up under heavy load: On Saturday I had the
error on a slightly loaded machine (another one!): When compiling kernel
(not "make -j", just a "normal" compile) my xosview died. In
/var/log/messages I found:
May 3 17:42:08 Mandelbrot kernel: SIGBUS: ayiee, couldnt allocate page
table.
This kernel does not have Ingo's scheduling patch, but his debugging patch
included.
BTW2: Your patch is missing the arguments for printk.
> Thanks.
>
> --- mm/vmscan.c.~1~ Thu May 1 00:31:31 1997
> +++ mm/vmscan.c Mon May 5 04:02:40 1997
> @@ -340,6 +340,7 @@
> stop = 3;
> if (wait)
> stop = 0;
> +we_slept:
> switch (state) {
> do {
> case 0:
> @@ -357,7 +358,13 @@
> i--;
> } while ((i - stop) >= 0);
> }
> - return 0;
> + /* This should never return failure, instead it should sleep
> + * and try again.
> + */
> + printk("try_to_free_page: state(%d) stop(%d) i(%d) sleep instead of fail\n");
> + schedule();
> + stop = 9;
> + goto we_slept;
> }
Hubert mantel@suse.de