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]
Message-Id: <1508921765-15396-5-git-send-email-byungchul.park@lge.com>
Date:   Wed, 25 Oct 2017 17:56:00 +0900
From:   Byungchul Park <byungchul.park@....com>
To:     peterz@...radead.org, mingo@...nel.org, axboe@...nel.dk
Cc:     johan@...nel.org, tglx@...utronix.de, linux-kernel@...r.kernel.org,
        linux-mm@...ck.org, tj@...nel.org, johannes.berg@...el.com,
        oleg@...hat.com, amir73il@...il.com, david@...morbit.com,
        darrick.wong@...cle.com, linux-xfs@...r.kernel.org,
        linux-fsdevel@...r.kernel.org, linux-block@...r.kernel.org,
        hch@...radead.org, idryomov@...il.com, kernel-team@....com
Subject: [PATCH v5 4/9] locking/lockdep: Add a boot parameter allowing unwind in cross-release and disable it by default

Johan Hovold reported a heavy performance regression caused by lockdep
cross-release:

 > Boot time (from "Linux version" to login prompt) had in fact doubled
 > since 4.13 where it took 17 seconds (with my current config) compared to
 > the 35 seconds I now see with 4.14-rc4.
 >
 > I quick bisect pointed to lockdep and specifically the following commit:
 >
 >	28a903f63ec0 ("locking/lockdep: Handle non(or multi)-acquisition
 >	               of a crosslock")
 >
 > which I've verified is the commit which doubled the boot time (compared
 > to 28a903f63ec0^) (added by lockdep crossrelease series [1]).

Currently cross-release performs unwind on every acquisition, but that
is very expensive.

This patch makes unwind optional and disables it by default and only
records acquire_ip.

Full stack traces are sometimes required for full analysis, in which
case a boot paramter, crossrelease_fullstack, can be specified.

On my qemu Ubuntu machine (x86_64, 4 cores, 512M), the regression was
fixed. We measure boot times with 'perf stat --null --repeat 10 $QEMU',
where $QEMU launches a kernel with init=/bin/true:

1. No lockdep enabled:

 Performance counter stats for 'qemu_booting_time.sh bzImage' (10 runs):

       2.756558155 seconds time elapsed                    ( +-  0.09% )

2. Lockdep enabled:

 Performance counter stats for 'qemu_booting_time.sh bzImage' (10 runs):

       2.968710420 seconds time elapsed                    ( +-  0.12% )

3. Lockdep enabled + cross-release enabled:

 Performance counter stats for 'qemu_booting_time.sh bzImage' (10 runs):

       3.153839636 seconds time elapsed                    ( +-  0.31% )

4. Lockdep enabled + cross-release enabled + this patch applied:

 Performance counter stats for 'qemu_booting_time.sh bzImage' (10 runs):

       2.963669551 seconds time elapsed                    ( +-  0.11% )

I.e. lockdep cross-release performance is now indistinguishable from
vanilla lockdep.

Reported-by: Johan Hovold <johan@...nel.org>
Bisected-by: Johan Hovold <johan@...nel.org>
Analyzed-by: Thomas Gleixner <tglx@...utronix.de>
Suggested-by: Thomas Gleixner <tglx@...utronix.de>
Signed-off-by: Byungchul Park <byungchul.park@....com>
---
 Documentation/admin-guide/kernel-parameters.txt |  3 +++
 kernel/locking/lockdep.c                        | 19 +++++++++++++++++--
 2 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index ead7f40..4107b01 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -709,6 +709,9 @@
 			It will be ignored when crashkernel=X,high is not used
 			or memory reserved is below 4G.
 
+	crossrelease_fullstack
+			[KNL] Allow to record full stack trace in cross-release
+
 	cryptomgr.notests
                         [KNL] Disable crypto self-tests
 
diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
index e36e652..160b5d6 100644
--- a/kernel/locking/lockdep.c
+++ b/kernel/locking/lockdep.c
@@ -76,6 +76,15 @@
 #define lock_stat 0
 #endif
 
+static int crossrelease_fullstack;
+static int __init allow_crossrelease_fullstack(char *str)
+{
+	crossrelease_fullstack = 1;
+	return 0;
+}
+
+early_param("crossrelease_fullstack", allow_crossrelease_fullstack);
+
 /*
  * lockdep_lock: protects the lockdep graph, the hashes and the
  *               class/list/hash allocators.
@@ -4863,8 +4872,14 @@ static void add_xhlock(struct held_lock *hlock)
 	xhlock->trace.nr_entries = 0;
 	xhlock->trace.max_entries = MAX_XHLOCK_TRACE_ENTRIES;
 	xhlock->trace.entries = xhlock->trace_entries;
-	xhlock->trace.skip = 3;
-	save_stack_trace(&xhlock->trace);
+
+	if (crossrelease_fullstack) {
+		xhlock->trace.skip = 3;
+		save_stack_trace(&xhlock->trace);
+	} else {
+		xhlock->trace.nr_entries = 1;
+		xhlock->trace.entries[0] = hlock->acquire_ip;
+	}
 }
 
 static inline int same_context_xhlock(struct hist_lock *xhlock)
-- 
1.9.1

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ