Re: BUG: unable to handle kernel paging request from pty_write [was: Linux 4.4.2]

From: Peter Hurley <peter@hurleysoftware.com>
To: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Greg KH <gregkh@linuxfoundation.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	stable <stable@vger.kernel.org>,
	lwn@lwn.net, Steven Rostedt <rostedt@goodmis.org>
Subject: Re: BUG: unable to handle kernel paging request from pty_write [was: Linux 4.4.2]
Date: Thu, 25 Feb 2016 16:38:39 -0800	[thread overview]
Message-ID: <56CF9E8F.3070905@hurleysoftware.com> (raw)
In-Reply-To: <56CF8124.4080003@hurleysoftware.com>

On 02/25/2016 02:33 PM, Peter Hurley wrote:
> On 02/25/2016 01:32 PM, Jiri Slaby wrote:
>> On 02/25/2016, 09:51 PM, Linus Torvalds wrote:
>>> Jiri, can you check your try_to_wake_up() disassembly for some
>>> indirect "jmp" instructions?
>>
>> Nope, there is none.
>>
>> I will reply to all your questions tomorrow.
>>
>> Just quickly, as I have to go (and don't want you to duplicate efforts)
>> the kernel which was used can be obtained here:
>> https://build.opensuse.org/package/binaries/openSUSE:Factory:Staging:I/kernel-default?repository=standard
>>
>> The issue is very weird, indeed, this is what I noted to our bugzilla:
>> The stack trace ends in call of try_to_wake_up. Then, there it has to be
>> some of the indirect calls:
>>
>> callq  *0x40(%rax)
>>   p->sched_class->select_task_rq from select_task_rq
>>
>> RAX is 0x00000000bb37e180, barely can be read with offset 0x40
>>
>> callq  *0xd85656(%rip) # ffffffff81e2aba0 <smp_ops+0x20>
>>   smp_ops.smp_send_reschedule from ttwu_queue_remote
>>
>> Which hardly can be it, given smp_ops is static.
>>
>> So it has to be some other "call *" from a nested function :(.
>>
>>
>>
>>
>> Interestingly, RBP contains address inside try_to_wake_up --
>> ffffffff810a535a (dunno why) which is:
>> ffffffff810a5355:       e8 66 a0 ff ff          callq  ffffffff8109f3c0
>> <ttwu_stat>
>> ffffffff810a535a:       e9 9d fe ff ff          jmpq   ffffffff810a51fc
>> <try_to_wake_up+0x3c>
> 
> That would imply that RSP was off by +8 when the ttwu_stat() epilog was
> executed so that RBP <= ret addr and RIP <= some local var in try_to_wake_up()
> stack frame.
> 
> Looks like R15 in the crash report could be what RBP should have been.
> 
> Now to find out why RSP is +8

Which I would investigate if I could download that kernel.
Unfortunately, OBS doesn't like me so if you could make that
kernel available some other way or send me a mixed listing
of kernel/sched/core.c

>> ttwu_stat does in the begginning:
>> mov    $0x16e80,%r14
>>
>> which is what we actually still have in r14 when it crashes. The first
>> ttwu_stat's "if" has to go through the true branch (otherwise r14 would
>> be overwritten).
>>
>>
>>
>> Another note: we die when jmp/calling to 0xffff88023fd40000.
>> RSI=RDI=0xffff88023fdd6e80. RSI-RIP is 0x96e80, which is R14 + 0x80000.
>> Coincidence?
>>
>> thanks,
>>
> 