linux-kernel - [RFC PATCH] Fix abnormal rcu dynticks_nesting values related to async page fault

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-next>] [day] [month] [year] [list]

Date:	Tue, 27 Nov 2012 13:15:25 +0800
From:	Li Zhong <zhong@...ux.vnet.ibm.com>
To:	linux-next list <linux-next@...r.kernel.org>,
	LKML <linux-kernel@...r.kernel.org>
Cc:	paulmck@...ux.vnet.ibm.com, sasha.levin@...cle.com,
	gleb@...hat.com, avi@...hat.com, fweisbec@...il.com
Subject: [RFC PATCH] Fix abnormal rcu dynticks_nesting values related to
 async page fault

I noticed some warnings complaining about dynticks_nesting value, like 

[  267.545032] ------------[ cut here ]------------
[  267.545032] WARNING: at kernel/rcutree.c:382 rcu_eqs_enter+0xab/0xc0()
[  267.545032] Hardware name: Bochs
[  267.545032] Modules linked in:
[  267.545032] Pid: 0, comm: swapper/2 Not tainted 3.7.0-rc5-next-20121115 #8
[  267.545032] Call Trace:
[  267.545032]  [<ffffffff8104714f>] warn_slowpath_common+0x7f/0xc0
[  267.545032]  [<ffffffff810471aa>] warn_slowpath_null+0x1a/0x20
[  267.545032]  [<ffffffff810e607b>] rcu_eqs_enter+0xab/0xc0
[  267.545032]  [<ffffffff810e60bb>] rcu_idle_enter+0x2b/0x70
[  267.545032]  [<ffffffff8100d44f>] cpu_idle+0x6f/0x100
[  267.545032]  [<ffffffff814bf055>] start_secondary+0x205/0x20c
[  267.545032] ---[ end trace 924ae80da035028d ]---

After enabling rcu-dyntick tracing, I got following abnormal
dynticks_nesting values (13fffffffffffff, ff00000000000001,etc):
			...
 1	<idle>-0     [002] dN.2 18739.518567: rcu_dyntick: End 0 140000000000000		rcu_idle_exit
 2	  sshd-696   [002] d..1 18739.518675: rcu_dyntick: ++= 140000000000000 140000000000001	rcu_irq_enter   - apf (not present)

 3	<idle>-0     [002] d..2 18739.518705: rcu_dyntick: Start 140000000000001 0		rcu_idle_enter
 4	<idle>-0     [002] d..2 18739.521252: rcu_dyntick: End 0 1				rcu_irq_enter	- apf (page ready)
 5	<idle>-0     [002] dN.2 18739.521261: rcu_dyntick: Start 1 0				rcu_irq_exit	- apf (page ready)
 6	<idle>-0     [002] dN.2 18739.521263: rcu_dyntick: End 0 140000000000000		rcu_idle_exit

 7	  sshd-696   [002] d..1 18739.521299: rcu_dyntick: --= 140000000000000 13fffffffffffff	rcu_irq_exit	- apf (not present)
 8	  sshd-696   [002] d..1 18739.521302: rcu_dyntick: Start 13fffffffffffff 0		rcu_user_enter
 9	  sshd-696   [002] d..1 18739.521330: rcu_dyntick: End 0 1				rcu_irq_enter   - apf (not present)

10	<idle>-0     [002] d..2 18739.521346: rcu_dyntick: Start 1 0				rcu_idle_enter - old value 1, warning
11	<idle>-0     [002] d..2 18739.530021: rcu_dyntick: ++= ff00000000000001 ff00000000000002
12	<idle>-0     [002] dN.2 18739.530029: rcu_dyntick: --= ff00000000000002 ff00000000000001
		...

I added the functions I guess which printed the tracing after each
line. 

Line #1, the idle-0 process calls rcu_idle_exit(), and finishes one
loop, to switch to sshd-696

Line #2, sshd-696 calls rcu_irq_enter() because of async page fault(page
not present), and puts itself to wait for page ready

Line #3, idle-0 is switched in, and clears the dynticks_nesting to 0

Line #4-5, I think the rcu_irq_enter/exit() is called because the page
for sshd-696 is ready

Line #6, idle-0 calls rcu_idle_exit(), to switch to sshd-696

Line #7, sshd-696 calls rcu_irq_exit() in the apf (page not present)
code path, decreasing dynticks_nesting to 13fffffffffffff. 

Line #8-9, sshd-696 calls rcu_user_enter() to start user eqs, and gets
async page fault again. It puts itself sleep again, with
dynticks_nesting value as 1.

Line #10, idle-0 switches in, as the dynticks_nesting value is 1, so
warning is reported in rcu_idle_enter(), then the value is decreased to
ff00000000000001. (In the tracing log, the new value is 0, that's
because rcu hard-code the value to be 0. I will send another patch for
this.)

This patch below tries to replace the rcu_irq_enter/exit() with
rcu_idle_exit/enter(), if it is in rcu idle, and it is idle process;
otherwise, rcu_user_exit() is called to exit user eqs if necessary. 

Signed-off-by: Li Zhong <zhong@...ux.vnet.ibm.com>
---
 arch/x86/kernel/kvm.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
index 4180a87..f65648d 100644
--- a/arch/x86/kernel/kvm.c
+++ b/arch/x86/kernel/kvm.c
@@ -247,10 +247,17 @@ do_async_page_fault(struct pt_regs *regs, unsigned long error_code)
 		break;
 	case KVM_PV_REASON_PAGE_NOT_PRESENT:
 		/* page is swapped out by the host. */
-		rcu_irq_enter();
+		if (is_idle_task(current) && rcu_is_cpu_idle())
+			rcu_idle_exit();
+		else
+			rcu_user_exit();
+
 		exit_idle();
 		kvm_async_pf_task_wait((u32)read_cr2());
-		rcu_irq_exit();
+
+		if (is_idle_task(current) && rcu_is_cpu_idle())
+			rcu_idle_enter();
+
 		break;
 	case KVM_PV_REASON_PAGE_READY:
 		rcu_irq_enter();
-- 
1.7.11.4

--
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/