linux-kernel - Re: [PATCH] arm64: add dump_stack to show

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

Date:   Thu, 23 Mar 2017 15:03:57 +0000
From:   Mark Rutland <mark.rutland@....com>
To:     Kefeng Wang <wangkefeng.wang@...wei.com>
Cc:     Ding Tianhong <dingtianhong@...wei.com>,
        Catalin Marinas <catalin.marinas@....com>,
        Will Deacon <will.deacon@....com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        linux-arm-kernel@...ts.infradead.org
Subject: Re: [PATCH] arm64: add dump_stack to show_regs

On Mon, Mar 20, 2017 at 09:05:04PM +0800, Kefeng Wang wrote:
> 
> 
> On 2017/3/20 19:02, Mark Rutland wrote:
> > On Sun, Mar 19, 2017 at 03:15:25PM +0800, Ding Tianhong wrote:
> >> Recently I found that when the system trigger a soft lockup in interrupt,
> >> there is only showing the regs, but no stack trace, it is very difficult
> >> to locate the problem:
> >>
> >> ===========================================
> >>
> >> [10072.999437] NMI watchdog: BUG: soft lockup - CPU#16 stuck for 23s! [ksoftirqd/16:88]
> >> .....
> >> [10073.041254] CPU: 16 PID: 88 Comm: ksoftirqd/16 Tainted: G         4.x.x #1
> >> [10073.041258] Hardware name: xxxxx, BIOS 1.17 01/04/2017
> >> [10073.041261] task: ffff803f6cb06200 ti: ffff803f6cb50000 task.ti: ffff803f6cb50000
> >> [10073.041274] PC is at _raw_spin_unlock_irqrestore+0x24/0x30
> >> [10073.041280] LR is at blk_run_queue+0x3c/0x48
> >> [10073.041282] pc : [<ffff800000a3df14>] lr : [<ffff8000004f3a7c>] pstate: 60000145
> >> [10073.041285] sp : ffff803f6cb53b20
> >> [10073.041286] x29: ffff803f6cb53b20 x28: 0000000000001000
> >> [10073.041290] x27: 0000000000000000 x26: ffff800001226000
> >> [10073.041294] x25: 0000000000000000 x24: 0000000000000140
> >> [10073.041297] x23: ffff803f62e108c8 x22: ffff800001037000
> >> [10073.041302] x21: ffff843f66800040 x20: 0000000000000140
> >> [10073.041305] x19: ffff803f62e108c8 x18: 0000000000000007
> >> [10073.041309] x17: 000000000000000e x16: 0000000000000001
> >> [10073.041312] x15: 0000000000000019 x14: 0000000000000033
> >> [10073.041317] x13: 000000000000004c x12: 0000000000000000
> >> [10073.041320] x11: 0000000000001000 x10: 0000000000000010
> >> [10073.041323] x9 : ffff8000004f3a7c x8 : ffff803f69b59120
> >> [10073.041327] x7 : 0000000000000000 x6 : 0000000000000002
> >> [10073.041331] x5 : 0000000000000244 x4 : 00000000000244d9
> >> [10073.041334] x3 : ffff843f653ab918 x2 : 0000000000004074
> >> [10073.041337] x1 : 0000000000000140 x0 : ffff803f62e10e58
> >>
> >> ===============================================
> >>
> >> So add the general dump_stack to show_regs to support showing the stack.
> >>
> >> Signed-off-by: Ding Tianhong <dingtianhong@...wei.com>
> >> ---
> >>  arch/arm64/kernel/process.c | 1 +
> >>  1 file changed, 1 insertion(+)
> >>
> >> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> >> index 043d373..60c5c26 100644
> >> --- a/arch/arm64/kernel/process.c
> >> +++ b/arch/arm64/kernel/process.c
> >> @@ -212,6 +212,7 @@ void show_regs(struct pt_regs * regs)
> >>  {
> >>  	printk("\n");
> >>  	__show_regs(regs);
> >> +	dump_stack();
> >>  }
> > 
> > I don't think this is quite right.
> 
> I found the same logic exists in arm32.
> 
> > I see that x86's show_regs() will dump a kernel stack, but it starts
> > from the stack described by the regs, not the stack used to call
> > dump_stack().
> > 
> > Also, for longjmp_break_handler() I think we only want the current
> > registers, and not the stack.
> 
> Is there a better way to show the kernel stack? it is not early to address issue
> if only show regs. Making a new show_regs() to call dump_mem()/dump_backtrace()/dump_instr()?

First, I think we can make longjmp_break_handler() use __show_regs().

Second, I think we can make show_regs() call dump_backtrace(), passing
the regs down. I believe that should trigger the existing frame
skipping, though we might need some fixups to cater for this particular
case.

Thanks,
Mark.