Re: kernel oops!

[ierdnah <ierdnah@go.ro>]

On Thu, 2005-01-27 at 15:35 -0800, Linus Torvalds wrote:

the last patch works, but the load increases very much (normally with
200 VPN connections I have a load of maximum 10, with this patch I have
a load of 50-100 - after 30 min of uptime)

> You probably should. The patch you've tested is really ugly, and not a fix 
> at all - it's really just depending on the compiler generating a specific 
> code sequence that will hide the race.  As such, it's a patch I would only 
> accept in the standard kernel as an absolute last resort.
> 
> In contrast, the second patch I tested may actually _fix_ the race. 
> 
> The fact that the first patch makes the oops go away is a good thing, 
> though: it shows that your oops really was due to that small race window, 
> and as such it helps validate that it wasn't anything else.

--- 1.32/drivers/char/pty.c     2005-01-10 17:29:36 -08:00
+++ edited/drivers/char/pty.c   2005-01-23 10:21:04 -08:00
@@ -149,13 +149,17 @@
 static int pty_chars_in_buffer(struct tty_struct *tty)
 {
        struct tty_struct *to = tty->link;
-       int count;
+       int count = 0;
 
-       if (!to || !to->ldisc.chars_in_buffer)
-               return 0;
-
-       /* The ldisc must report 0 if no characters available to be read
*/
-       count = to->ldisc.chars_in_buffer(to);
+       if (to) {
+               struct tty_ldisc *ld = tty_ldisc_ref(to);
+               if (ld) {
+                       if (ld->chars_in_buffer) {
+                               count = ld->chars_in_buffer(to);
+                               tty_ldisc_deref(ld);
+                       }
+               }
+       }
 
 
       if (tty->driver->subtype == PTY_TYPE_SLAVE) return count;
 
-- 
ierdnah <ierdnah@go.ro>