Re: Nasty regression from .27.7 to .27.8: idle samba goes crazy

From: Manfred Spraul
Date: Wed Dec 10 2008 - 12:43:20 EST


Chuck Ebbert wrote:
On Mon, 8 Dec 2008 23:22:46 +0100
Jan Rekorajski <baggins@xxxxxxxxxxxxxxxxx> wrote:

On Mon, 08 Dec 2008, Stefan Richter wrote:

On Monday, 8 of December 2008, Holger Hoffstaette wrote:
The system is a generic old single-core P4 box with a single SATA drive,
Gentoo userland and Samba is 3.0.33 (in async mode). The kernel has no
patches or binary drivers.
Holger, it may be unrelated to the issue, but to be sure: Which network
card driver do you use?
I think you can safely rule out NIC, I'm also seeing this behaviour on a
brand new server with imap hanging in some busy-loop.
Network card in my case:
Broadcom Corporation NetXtreme II BCM5708 Gigabit Ethernet (rev 12)

What I observer was one CPU doing 100% system work, and the number of
timer interrupts went from 1k per second to 4k (for the whole system).

Could you try the attached patch?
It should fix the bug.


--
Manfred
rom ae060e0b7bc071bd73dd5319b93c3344d9e10212 Mon Sep 17 00:00:00 2001
From: Manfred Spraul <manfred@xxxxxxxxxxxxxxxx>
To: torvalds@xxxxxxxxxxxxxxxxxxxx
Cc: linux-kernel@xxxxxxxxxxxxxxx
Cc: cebbert@xxxxxxxxxx
Cc: airlied@xxxxxxxxx
Cc: akpm@xxxxxxxxxxxxxxxxxxxx
Bcc: manfred@xxxxxxxxxxxxxxxx
Date: Wed, 10 Dec 2008 18:17:06 +0100
Subject: [PATCH] lib/idr.c: Fix bug introduced by RCU fix

The last patch to lib/idr.c caused a bug if idr_get_new_above() was
called on an empty idr:
Usually, nodes stay on the same layer. New layers are added to the top
of the tree.
The exception is idr_get_new_above() on an empty tree: In this case,
the new root node is first added on layer 0, then moved upwards.
p->layer was not updated.

As usual: You shall never rely on the source code comments, they
will only mislead you.

Signed-off-by: Manfred Spraul <manfred@xxxxxxxxxxxxxxxx>
---
lib/idr.c | 8 +++++++-
1 files changed, 7 insertions(+), 1 deletions(-)

diff --git a/lib/idr.c b/lib/idr.c
index 7a785a0..1c4f928 100644
--- a/lib/idr.c
+++ b/lib/idr.c
@@ -220,8 +220,14 @@ build_up:
*/
while ((layers < (MAX_LEVEL - 1)) && (id >= (1 << (layers*IDR_BITS)))) {
layers++;
- if (!p->count)
+ if (!p->count) {
+ /* special case: if the tree is currently empty,
+ * then we grow the tree by moving the top node
+ * upwards.
+ */
+ p->layer++;
continue;
+ }
if (!(new = get_from_free_list(idp))) {
/*
* The allocation failed. If we built part of
--
1.5.6.5