linux-kernel - Re: [PATCH] arm64: Clear OS lock in enable_debug

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: <20220609115716.GA2427@willie-the-truck>
Date:   Thu, 9 Jun 2022 12:57:17 +0100
From:   Will Deacon <will@...nel.org>
To:     Mark-PK Tsai <mark-pk.tsai@...iatek.com>
Cc:     Catalin Marinas <catalin.marinas@....com>,
        Matthias Brugger <matthias.bgg@...il.com>,
        yj.chiang@...iatek.com, Marc Zyngier <maz@...nel.org>,
        Alexandru Elisei <alexandru.elisei@....com>,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        linux-mediatek@...ts.infradead.org, jean-philippe.brucker@....com
Subject: Re: [PATCH] arm64: Clear OS lock in enable_debug_monitors

On Thu, Jun 09, 2022 at 11:33:18AM +0800, Mark-PK Tsai wrote:
> Always clear OS lock before enable debug event.
> 
> The OS lock is clear in cpuhp ops in recent kernel,
> but when the debug exception happened before it
> kernel might crash because debug event enable didn't
> take effect when OS lock is hold.
> 
> Below is the use case that having this problem:
> 
> Register kprobe in console_unlock and kernel will
> panic at secondary_start_kernel on secondary core.
> 
> CPU: 1 PID: 0 Comm: swapper/1 Tainted: P
> ...
> pstate: 004001c5 (nzcv dAIF +PAN -UAO)
> pc : do_undefinstr+0x5c/0x60
> lr : do_undefinstr+0x2c/0x60
> sp : ffffffc01338bc50
> pmr_save: 000000f0
> x29: ffffffc01338bc50 x28: ffffff8115e95a00 T
> x27: ffffffc01258e000 x26: ffffff8115e95a00
> x25: 00000000ffffffff x24: 0000000000000000
> x23: 00000000604001c5 x22: ffffffc014015008
> x21: 000000002232f000 x20: 00000000000000f0 j
> x19: ffffffc01338bc70 x18: ffffffc0132ed040
> x17: ffffffc01258eb48 x16: 0000000000000403 L&
> x15: 0000000000016480 x14: ffffffc01258e000 i/
> x13: 0000000000000006 x12: 0000000000006985
> x11: 00000000d5300000 x10: 0000000000000000
> x9 : 9f6c79217a8a0400 x8 : 00000000000000c5
> x7 : 0000000000000000 x6 : ffffffc01338bc08 2T
> x5 : ffffffc01338bc08 x4 : 0000000000000002
> x3 : 0000000000000000 x2 : 0000000000000004
> x1 : 0000000000000000 x0 : 0000000000000001 *q
> Call trace:
>  do_undefinstr+0x5c/0x60
>  el1_undef+0x10/0xb4
>  0xffffffc014015008
>  vprintk_func+0x210/0x290
>  printk+0x64/0x90
>  cpuinfo_detect_icache_policy+0x80/0xe0
>  __cpuinfo_store_cpu+0x150/0x160
>  secondary_start_kernel+0x154/0x440
> 
> The root cause is that OS_LSR_EL1.OSLK is reset
> to 1 on a cold reset[1] and the firmware didn't
> unlock it by default.
> So the core didn't go to el1_dbg as expected after
> kernel_enable_single_step and eret.

Hmm, I thought we didn't use hardware single-step for kprobes after
7ee31a3aa8f4 ("arm64: kprobes: Use BRK instead of single-step when executing
instructions out-of-line"). What is triggering this exception?

Will