linux-kernel - Re: [RFC PATCH 0/3] restartable sequences v2: fast user-space percpu critical sections

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]

Date:	Sun, 10 Apr 2016 14:07:36 +0000 (UTC)
From:	Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
To:	Linus Torvalds <torvalds@...ux-foundation.org>
Cc:	Andy Lutomirski <luto@...capital.net>,
	Peter Zijlstra <peterz@...radead.org>,
	"Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>,
	Ingo Molnar <mingo@...hat.com>,
	Paul Turner <commonly@...il.com>,
	Andi Kleen <andi@...stfloor.org>, Chris Lameter <cl@...ux.com>,
	Dave Watson <davejwatson@...com>,
	Josh Triplett <josh@...htriplett.org>,
	linux-api <linux-api@...r.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Andrew Hunter <ahh@...gle.com>
Subject: Re: [RFC PATCH 0/3] restartable sequences v2: fast user-space
 percpu critical sections

----- On Apr 8, 2016, at 5:25 PM, Linus Torvalds torvalds@...ux-foundation.org wrote:

> On Fri, Apr 8, 2016 at 10:46 AM, Mathieu Desnoyers
> <mathieu.desnoyers@...icios.com> wrote:
>>
>> By the way, the debugger can always decide to single-step through the
>> first iteration of the rseq, and then after it loops, decide to skip
>> single-stepping until the exit points are reached.
> 
> A _human_ debugger may decide to do that yes.
> 
> But the the debugger _program_ may not be that smart. For example,
> let's say that you - as a human - set a few watchpoints. The debugger
> will use hardware breakpoints for the first few, but in more complex
> cases the debugger will actually say "oops, no more hw breakpoints,
> I'll just start single-stepping instead".
> 
> The human operator may not even be aware that the debugger has gone
> into that slower mode. Normally it's just slower. But you'd want it to
> be _only_ slower, not "oops, the program no longer makes any forward
> progress at all, because a library that the user didn't even know or
> care about - and never sees, because the single-stepping is all
> internal = happened to use a code sequence that doesn't work under
> single-stepping".

Which is why I'm proposing to extend gdb to support this automatically,
without requiring interaction or knowledge from the user.

The idea is to let gdb detect entry points into those restartable
critical sections as it single-steps through the program. It would
know about all rseq c.s. exit points too, so it can track whether
it has single-stepped over an entire rseq c.s. and thus caused a
restart. At that point, it can put the breakpoint at each exit point
associated with the entry point, thus skipping single-step of the
second iteration of the critical section.

I think this could be achieved by populating a section that contains
information about entry and exit points of those critical sections
in the rseq_{start,finish,abort} functions. Those sections would end
up in the app/lib ELF binary, may not have to be necessarily loaded
into program's memory.

Does it make sense to try it out, or am I missing something obvious ?

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com