[PATCH v2] perf: fix kernel panic when parsing user space CS saved in pt_regs

From: Liu ShuoX
Date: Thu Jun 05 2014 - 03:40:52 EST

From: Zhang Yanmin <yanmin.zhang@xxxxxxxxx>

ChangeLog V2: Before sysexit, perf NMI might arrive. There is
still a race. Here we change rsp to keep it pointing
to pt_regs->orig_ax.
In addition, after sti, before sysexit, an irq might
arrives. That causes more chances for perf NMI to jump
in.

We hit a kernel panic when running perf to collect some performance data.
kenel is x86_64 and user space apps are 32bit.

[71.965351, 1] [Binder_2] BUG: unable to handle kernel NULL pointer dereference at 0000000000000004
[71.965360, 1] [Binder_2] IP: [<ffffffff82012091>] get_segment_base+0x71/0xc0
[71.965367, 1] [Binder_2] PGD 6c65f067 PUD 0
[71.965375, 1] [Binder_2] Oops: 0000 [#1] PREEMPT SMP
[71.965413, 1] [Binder_2] Modules linked in: <...>
[71.965420, 1] [Binder_2] CPU: 1 PID: 304 Comm: Binder_2 Tainted: G W O 3.10.20-263902-g184bfbc-dirty #14
[71.965426, 1] [Binder_2] task: ffff8800764dc300 ti: ffff88006c6e8000 task.ti: ffff88006c6e8000
[71.965439, 1] [Binder_2] RIP: 0010:[<ffffffff82012091>] [<ffffffÃf82012091>] get_segment_base+0x71/0xc0
[71.965<44, 1] [Binder_2] RSP: 0018:ffff^X8007ea87b98 EFLAGS: 00010092
[71.965447, 1] [Binder_2] RAX: 0000000000000024 RBX: 0000000000000000 RCX: 0000000000000000
[71.965450, 1] [Binder_2] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000009
[71.965454, 1] [Binder_2] RBP: ffff88007ea87ba8 R08: ffffffff83143b3c R09: ffffffff831848a8
[71.965458, 1] [Binder_2] R10: 0000000000000000 R11: 00000000001bf2d8 R12: 0000000000000000
[71.965462, 1] [Binder_2] R13: ffff88006c6e9fd8 R14: ffff88006c6e9f58 R15: ffff8800764dc300
[71.965468, 1_ [Binder_2] FS: 0000000000000000(0000) GS:ffff88007ea80000(006b) knlGS:00000000f704add0
[71.965472, 1] [Binder_2] CS: 0010 DS: 002b ES: 002b CR0: 0000000080050033
[71.965476, 1] [Binder_2] CR2: 0000000000000004 CR3: 0000000076588000 CR4: 00000^P00001007e0
[71.965480, 1] [Binder_2] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[71.965485, 1] [Binder_2] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[71.966141, 1] [Binder_2] Stack:
[71.966152, 1] [Binder_2] ffff88005f266c00 0000000000000000 ffff88007ea87c18 ffffffff82013cac
[71.966161, 1] [Binder_2] ffff88007ea87d58 00000016fe4704a0 00000000000001a7 ffff88007ea87ef8
[71.966171, 1] [Binder_6] ffff88005f266c00 ffff88007ea87ef8 ffff8800!e07b400 ffff88005f266c00
[71.966173, 1] [Binder_2] Call Trace:
[71.966179, 1] [Binder_2] <NMI>
[71.966190, 1] [Binder_2] [<ffffffff82013cac>] perf_callchain_user+0x15c/0x240
[71.966202, 1] [Binder_2] [<ffffffff82160754>] perf_callchain+0x134/0x180
[71.966210, 1] [Binder_2] [<fffff&ff820e0787>] ? local_clock+0x47/0x60
[71.966221, 1] [Binder_2] [<ffffffff8215d49b>] perf_prepare_sample+0x1bb/0x240
[71.966231, 1] [Binder_2] [<ffffffff8215d667>] __perf_event_overflow+0x147/0x230
[71.966241, 1] [Binder_2] [<ffffffff82012f68>] ? x86_perf_event_set_period+0xd8/0x150
[71.966252, 1] [Binder_2] [<ffffffff8215df24>] perf_event_overflow+0x14/0x20
[71.966260, 1] [Binder_2] [<ffffffff820194d2>] intel_pmu_handle_irq+0x1c2/0x270
[71.966270, 1] [Binder_2] [<ffffffff828b5d60>] ? call_softirq+0x30/0x30
[71.966284, 1] [Binder_2]$ [<ffffffff828aff01>] perf_event_nmi_handler+0x21/0x30
[71.966293, 1] [Binder_2] [<ffffffff828af5b9>] nmi_handle.isr!.1+0x59/0x=0
[71.966303, 1] [Binder_2] [<ffffffff828af6d8>] default_do_nmi+0x58/0x240
[71.966312, 1] [Binder_2] [<ffffffff828af978>] do_nmi+0xb8/0xf0
[71.966321, 1] [Binder_2] [|ffffffgf828aebe7>] end_repeat_nmi+0x1e/0x2e
[71.966332, 1] [Binder_2] [<ffffffff828b5d60>] ? call_softirq+0x30/0x30
[71.966341, 1] [Binder_2] [<ffffffff828b5d60>] ? call_softirq+0x30/0x30
[71.966350, 1] [Binder_2] [<fFffffff828b5d60>] ? call_softirq+0x30/0x30

Basically, ia32 uses sysenter to start system calls.

sysexit_from_sys_call=>trace_hardirqs_on_thunk. Before calling,
sysexit_from_sys_call already pops up pt_regs, then trace_hardirqs_on_thunk
would reuse pt_regs space. If perf NMI happens here, perf might use a bad pt_regs.

The patch fixes it by keeping rsp pointing under pt_regs->ip.

Signed-off-by: Zhang Yanmin <yanmin.zhang@xxxxxxxxx>
Signed-off-by: Liu ShuoX <shuox.liu@xxxxxxxxx>
---
arch/x86/ia32/ia32entry.S | 11 ++++++-----
1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S
index 4299eb0..a695d8e 100644
--- a/arch/x86/ia32/ia32entry.S
+++ b/arch/x86/ia32/ia32entry.S
@@ -172,15 +172,16 @@ sysexit_from_sys_call:
andl $~0x200,EFLAGS-R11(%rsp) movl RIP-R11(%rsp),%edx /* User %eip */
CFI_REGISTER rip,rdx
- RESTORE_ARGS 0,24,0,0,0,0
- xorq %r8,%r8
+ RESTORE_ARGS 0,0,0,0,0,0
+ movq 24(%rsp),%r8 /* rflags */
+ movq 32(%rsp),%rcx /* User %esp */
xorq %r9,%r9
xorq %r10,%r10
xorq %r11,%r11
- popfq_cfi
+ pushq_cfi %r8
/*CFI_RESTORE rflags*/
- popq_cfi %rcx /* User %esp */
- CFI_REGISTER rsp,rcx
+ popfq_cfi
+ xorq %r8,%r8
TRACE_IRQS_ON
ENABLE_INTERRUPTS_SYSEXIT32
--
1.8.3.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/