mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Ernie Petrides <petrides@redhat.com>
To: Marcelo Tosatti <marcelo.tosatti@cyclades.com>
Cc: Dave Kleikamp <shaggy@austin.ibm.com>,
	Marcelo Tosatti <marcelo.tosatti@cyclades.com.br>,
	linux-kernel <linux-kernel@vger.kernel.org>
Subject: Re: 2.4.x performance tests Re: [PATCH] BUG() in exec_mmap()
Date: Mon, 13 Oct 2003 02:52:44 -0400	[thread overview]
Message-ID: <200310130652.h9D6qiib005952@pasta.boston.redhat.com> (raw)
In-Reply-To: Your message of "Thu, 09 Oct 2003 17:25:52 -0300."

On Thursday, 9-Oct-2003 at 17:25 -0300, Marcelo Tosatti wrote:

> BTW, further performance testing of the removal of this optimization is
> VERY welcome.
>
> I've done some tests and no big performance harm has showed up, but thats
> just me.


In a variation of a (loosely) 2.4.21-based kernel, I get a 0% degradation
in the speed of a program exec'ing itself 1,000,000 times with the addition
of the missing locking in exec_mmap():

--- linux-2.4.21/fs/exec.c.orig
+++ linux-2.4.21/fs/exec.c
@@ -452,9 +452,11 @@ static int exec_mmap(void)
 
 	old_mm = current->mm;
 	if (old_mm && atomic_read(&old_mm->mm_users) == 1) {
+		down_write(&old_mm->mmap_sem);
 		mm_release();
 		exit_aio(old_mm);
 		exit_mmap(old_mm);
+		up_write(&old_mm->mmap_sem);
 		return 0;
 	}
 

Applying your change to the same kernel, I get a 2.5% degradation in the
same test case:

--- linux-2.4.21/fs/exec.c.orig
+++ linux-2.4.21/fs/exec.c
@@ -451,12 +451,6 @@ static int exec_mmap(void)
 	struct mm_struct * mm, * old_mm;
 
 	old_mm = current->mm;
-	if (old_mm && atomic_read(&old_mm->mm_users) == 1) {
-		mm_release();
-		exit_aio(old_mm);
-		exit_mmap(old_mm);
-		return 0;
-	}
 
 	mm = mm_alloc();
 	if (mm) {


Average times over 3 runs (of 1,000,000 execs each) were:

	base kernel:	529 elapsed seconds
	w/1st change:	529 elapsed seconds
	w/2nd change:	542 elapsed seconds

I wouldn't bother optimizing for an exec() syscall, but the 1st change
also eliminates the possibility of two ENOMEM error paths in the typical
case of not sharing an mm_struct.


Also, a reproducer that could expose the race condition (in typically 5-20
seconds) on a dual-Xeon Dell box ran for 10's of minutes without problems
with either fix.

Cheers.  -ernie

             reply	other threads:[~2003-10-13  6:47 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2003-10-13  6:52 Ernie Petrides [this message]
2003-10-13 12:09 ` Andrea Arcangeli
  -- strict thread matches above, loose matches on Subject: below --
2003-10-09 20:19 Marcelo Tosatti
2003-10-09 20:25 ` 2.4.x performance tests " Marcelo Tosatti

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=200310130652.h9D6qiib005952@pasta.boston.redhat.com \
    --to=petrides@redhat.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=marcelo.tosatti@cyclades.com \
    --cc=marcelo.tosatti@cyclades.com.br \
    --cc=shaggy@austin.ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®