linux-kernel - Re: ptrace performance (was: [Bug #12208] uml is very slow on 2.6.28 host)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Sat, 21 Mar 2009 16:22:17 +0100
From:	Ingo Molnar <mingo@...e.hu>
To:	Michael Riepe <michael.riepe@...glemail.com>
Cc:	"Rafael J. Wysocki" <rjw@...k.pl>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Kernel Testers List <kernel-testers@...r.kernel.org>,
	Miklos Szeredi <miklos@...redi.hu>
Subject: Re: ptrace performance (was: [Bug #12208] uml is very slow on
	2.6.28 host)


* Michael Riepe <michael.riepe@...glemail.com> wrote:

> Disclaimer: I'm not using UML, but these problems may be related.
> 
> > Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=12208
> > Subject		: uml is very slow on 2.6.28 host
> > Submitter	: Miklos Szeredi <miklos@...redi.hu>
> > Date		: 2008-12-12 9:35 (93 days old)
> > References	: http://marc.info/?l=linux-kernel&m=122907463518593&w=4
> 
> The other day I noticed a dramatic ptrace slowdown between 2.6.27 and
> 2.6.28.x (verified with 2.6.28.8). In particular, a command like
> 
> 	dd if=/dev/zero of=/dev/null bs=1024k count=1024
> 
> will normally report a throughput in the GB/s range. On 2.6.27, this is
> also true if you run
> 
> 	strace -o /dev/null <dd command as above>
> 
> which is only a little slower. But if I do the same on 2.6.28.x, I 
> get a throughput of about 100 MB/s or less, i.e. less than 10%. I 
> tried the commands on three different machines (an Athlon64 3000+, 
> a Core Duo T2400 and an Atom 330), and they all behave similar. 
> The more system calls a program uses, the worse the slowdown (try 
> the dd command with bs=16k and count=65536, for example - but 
> don't hold your breath).
> 
> Interestingly, the CPUs are mostly idle while the command is 
> executing on 2.6.28.x, but there is a high (system) load on 
> 2.6.27. Therefore, I suspect that it's a scheduling or maybe timer 
> problem that was introduced between 2.6.27 and 2.6.28. I haven't 
> had the time to check the rc kernels yet; perhaps someone else can 
> run a quick check to verify that it's gone in the latest 
> 2.6.29-rc.

that's almost certainly due to the wait_task_inactive() change. Does 
the patch below improve things?

	Ingo

diff --git a/kernel/sched.c b/kernel/sched.c
index 3e827b8..2d60f23 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -2119,7 +2119,8 @@ unsigned long wait_task_inactive(struct task_struct *p, long match_state)
 		 * yield - it could be a while.
 		 */
 		if (unlikely(on_rq)) {
-			schedule_timeout_uninterruptible(1);
+			cpu_relax();
+			cond_resched();
 			continue;
 		}
 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/