linux-kernel - [PATCH] ARM:Fix incomplete stacktrace when unwind ftrace_regs

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>] [day] [month] [year] [list]

Message-ID: <20220223022428.58385-1-haibo.li@mediatek.com>
Date:   Wed, 23 Feb 2022 10:24:28 +0800
From:   Haibo Li <haibo.li@...iatek.com>
To:     Russell King <linux@...linux.org.uk>,
        Matthias Brugger <matthias.bgg@...il.com>
CC:     "GitAuthor : Haibo Li" <haibo.li@...iatek.com>,
        <xiaoming.yu@...iatek.com>, <linux-arm-kernel@...ts.infradead.org>,
        <linux-kernel@...r.kernel.org>,
        <linux-mediatek@...ts.infradead.org>
Subject: [PATCH] ARM:Fix incomplete stacktrace when unwind ftrace_regs_caller

We can use register_ftrace_function(&ops) to
register a function callback.
if we set FTRACE_OPS_FL_SAVE_REGS in ftrace flags(flags in ops),
then do stack unwind
like dump_stack or WARN in the callback,we get incomplete stacktrace.
The stack unwind ends in ftrace_regs_caller.

Below is the output from test code.
The test code do register_ftrace_function
with FTRACE_OPS_FL_SAVE_REGS set,
and set sysrq_handle_loglevel as filter,then do dump_stack
in callback.It can not continue to
unwind the stack from ftrace_regs_call

echo 7 >/proc/sysrq-trigger
sysrq: Changing Loglevel
CPU: 1 PID: 145 Comm: sh Not tainted
5.17.0-rc4-00002-gb0dc07624ccd-dirty #2
[    9.023226][  T145] Hardware name: Generic DT based system
[    9.023670][  T145]  unwind_backtrace from show_stack+0x28/0x2c
[    9.024070][  T145]  show_stack from dump_stack_lvl+0x48/0x54
[    9.024298][  T145]  dump_stack_lvl from ftrace_regs_call+0x4/0x18
[    9.024650][  T145] sysrq: Loglevel set to 7

Fix it by adding unwind directives in ftrace_regs_caller.

Signed-off-by: Haibo Li <haibo.li@...iatek.com>
---
 arch/arm/kernel/entry-ftrace.S | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/arch/arm/kernel/entry-ftrace.S b/arch/arm/kernel/entry-ftrace.S
index a74289ebc803..eafa0c9b8992 100644
--- a/arch/arm/kernel/entry-ftrace.S
+++ b/arch/arm/kernel/entry-ftrace.S
@@ -75,15 +75,18 @@
 
 	add 	ip, sp, #12	@ move in IP the value of SP as it was
 				@ before the push {lr} of the mcount mechanism
-
+ UNWIND(.movsp ip)
 	str     lr, [sp, #0]    @ store LR instead of PC
+ UNWIND(.save{pc})  @it is the caller of ftrace_regs_caller
 
 	ldr     lr, [sp, #8]    @ get previous LR
 
 	str	r0, [sp, #8]	@ write r0 as OLD_R0 over previous LR
 
 	stmdb   sp!, {ip, lr}
+ UNWIND(.save{ip,lr})
 	stmdb   sp!, {r0-r11, lr}
+ UNWIND(.save{r0-r11,lr})
 
 	@ stack content at this point:
 	@ 0  4          48   52       56            60   64    68       72
-- 
2.25.1