[vserver] O(1) s_context CPU scheduling, take two:
 vserver development mailing list 
[Next/Previous Months] [Main vserver Project Homepage] [Howto Subscribe/Unsubscribe] [Paul Sladen's vserver stuff]
About this list Date view Thread view Subject view Author view Attachment view

From: Sam Vilain (sam_at_vilain.net)
Date: Mon 24 Feb 2003 - 18:04:41 GMT

For the keen,

I found the bug with my previous patch. It comes down to patch fuzz! One
ludicrous insertion of diff lines into the wrong place by patch slipped
through the cracks of my careful poring over its output. And I mean
ludicrous - not one of the context lines in kernel/exit.c where patch
chose to put the added lines were the same. This has cast great shrouds
of doubt over the reliability of the diff/patch operation to me, but
nevermind that :-)

Here is the fix:

diff -urN linux-2.4.21-pre4-ac6-ctx16-rough/kernel/exit.c
linux-2.4.21-pre4-ac6-ctx16/kernel/exit.c
--- linux-2.4.21-pre4-ac6-ctx16-rough/kernel/exit.c Tue Feb 25 06:30:15
2003
+++ linux-2.4.21-pre4-ac6-ctx16/kernel/exit.c Mon Feb 24 22:14:23 2003
@@ -42,6 +42,8 @@
        current->cmaj_flt += p->maj_flt + p->cmaj_flt;
        current->cnswap += p->nswap + p->cnswap;
        sched_exit(p);
+ sys_release_s_info(p);
+ sys_release_ip_info(p->ip_info);
        p->pid = 0;
        free_task_struct(p);
 }
@@ -66,8 +68,6 @@
                        break;
                }
                if (p->pid == pgrp)
- sys_release_s_info(p);
- sys_release_ip_info(p->ip_info);
                        fallback = p->session;
        }
        read_unlock(&tasklist_lock);

The attached patch is against the latest Alan Cox kernel (2.4.21-pre4-ac6;
I needed it to get DMA working on my KT400 motherboard) + jacque's ctx16
patch. It also applies against 2.4.20-ac2 + jacque's ctx16 patch; the
three failed hunks and one instance of fuzz seem upon first inspection to
be OK - and so the result will probably work, but I didn't test that.

I haven't had a chance to thoroughly test this on a system with many real
vservers yet, but that shocking bug in the last patch I posted is well and
truly gone. I'm running the kernel now on my workstation with several
contexts testing the new code, and all seems to be in order.

I'll post a version with some per-vserver tuning knobs once I figure out
this whole /proc business :-). Until then, hack kernel/sys.c 

-- 
Sam Vilain, sam_at_vilain.net

The reason we start a war is to fight a war, win a war, thereby
causing no more war!
 - George W. Bush during the first Presidential debate

About this list Date view Thread view Subject view Author view Attachment view
[Next/Previous Months] [Main vserver Project Homepage] [Howto Subscribe/Unsubscribe] [Paul Sladen's vserver stuff]
Generated on Mon 24 Feb 2003 - 18:34:56 GMT by hypermail 2.1.3