linux-kernel - Re: [BUG] completely bonkers use of set_need_resched + VM_FAULT

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 for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5232C2BB.9070303@vmware.com>
Date:	Fri, 13 Sep 2013 09:46:03 +0200
From:	Thomas Hellstrom <thellstrom@...are.com>
To:	Maarten Lankhorst <maarten.lankhorst@...onical.com>
CC:	Daniel Vetter <daniel.vetter@...ll.ch>,
	Peter Zijlstra <peterz@...radead.org>,
	Dave Airlie <airlied@...ux.ie>,
	intel-gfx <intel-gfx@...ts.freedesktop.org>,
	dri-devel <dri-devel@...ts.freedesktop.org>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Ingo Molnar <mingo@...nel.org>,
	Thomas Gleixner <tglx@...utronix.de>
Subject: Re: [BUG] completely bonkers use of set_need_resched + VM_FAULT_NOPAGE

On 09/13/2013 09:16 AM, Maarten Lankhorst wrote:
> Op 13-09-13 08:44, Thomas Hellstrom schreef:
>> On 09/12/2013 11:50 PM, Maarten Lankhorst wrote:
>>> Op 12-09-13 18:44, Thomas Hellstrom schreef:
>>>> On 09/12/2013 05:45 PM, Maarten Lankhorst wrote:
>>>>> Op 12-09-13 17:36, Daniel Vetter schreef:
>>>>>> On Thu, Sep 12, 2013 at 5:06 PM, Peter Zijlstra <peterz@...radead.org> wrote:
>>>>>>> So I'm poking around the preemption code and stumbled upon:
>>>>>>>
>>>>>>> drivers/gpu/drm/i915/i915_gem.c:                set_need_resched();
>>>>>>> drivers/gpu/drm/ttm/ttm_bo_vm.c:                        set_need_resched();
>>>>>>> drivers/gpu/drm/ttm/ttm_bo_vm.c:                        set_need_resched();
>>>>>>> drivers/gpu/drm/udl/udl_gem.c:          set_need_resched();
>>>>>>>
>>>>>>> All these sites basically do:
>>>>>>>
>>>>>>>      while (!trylock())
>>>>>>>            yield();
>>>>>>>
>>>>>>> which is a horrible and broken locking pattern.
>>>>>>>
>>>>>>> Firstly its deadlock prone, suppose the faulting process is a FIFOn+1
>>>>>>> task that preempted the lock holder at FIFOn.
>>>>>>>
>>>>>>> Secondly the implementation is worse than usual by abusing
>>>>>>> VM_FAULT_NOPAGE, which is supposed to install a PTE so that the fault
>>>>>>> doesn't retry, but you're using it as a get out of fault path. And
>>>>>>> you're using set_need_resched() which is not something a driver should
>>>>>>> _ever_ touch.
>>>>>>>
>>>>>>> Now I'm going to take away set_need_resched() -- and while you can
>>>>>>> 'reimplement' it using set_thread_flag() you're not going to do that
>>>>>>> because it will be broken due to changes to the preempt code.
>>>>>>>
>>>>>>> So please as to fix ASAP and don't allow anybody to trick you into
>>>>>>> merging silly things like that again ;-)
>>>>>> The set_need_resched in i915_gem.c:i915_gem_fault can actually be
>>>>>> removed. It was there to give the error handler a chance to sneak in
>>>>>> and reset the hw/sw tracking when the gpu is dead. That hack goes back
>>>>>> to the days when the locking around our error handler was somewhere
>>>>>> between nonexistent and totally broken, nowadays we keep things from
>>>>>> live-locking by a bit of magic in i915_mutex_lock_interruptible. I'll
>>>>>> whip up a patch to rip this out. I'll also check that our testsuite
>>>>>> properly exercises this path (needs a bit of work on a quick look for
>>>>>> better coverage).
>>>>>>
>>>>>> The one in ttm is just bonghits to shut up lockdep: ttm can recurse
>>>>>> into it's own pagefault handler and then deadlock, the trylock just
>>>>>> keeps lockdep quiet. We've had that bug arise in drm/i915 due to some
>>>>>> fun userspace did and now have testcases for them. The right solution
>>>>>> to fix this is to use copy_to|from_user_atomic in ttm everywhere it
>>>>>> holds locks and have slowpaths which drops locks, copies stuff into a
>>>>>> temp allocation and then continues. At least that's how we've fixed
>>>>>> all those inversions in i915-gem. I'm not volunteering to fix this ;-)
>>>>> Ah the case where a mmap'd address is passed to the execbuf ioctl? :P
>>>>>
>>>>> Fine I'll look into it a bit, hopefully before tuesday. Else it might take a bit longer since I'll be on my way to plumbers..
>>>> I think a possible fix would be if fault() were allowed to return an error and drop the mmap_sem() before returning.
>>>>
>>>> Otherwise we need to track down all copy_to_user / copy_from_user which happen with bo::reserve held.
>> Actually, from looking at the mm code, it seems OK to do the following:
>>
>> if (!bo_tryreserve()) {
>>      up_read mmap_sem(); // Release the mmap_sem to avoid deadlocks.
>>      bo_reserve();               // Wait for the BO to become available (interruptible)
>>      bo_unreserve();           // Where is bo_wait_unreserved() when we need it, Maarten :P
>>      return VM_FAULT_RETRY; // Go ahead and retry the VMA walk, after regrabbing
>> }
> Is this meant as a jab at me? You're doing locking wrong here! Again!

It's not meant as a jab at you.  I'm sorry if it came out that way. It 
was meant as a joke. I wasn't aware the topic was sensitive.

Anyway, could you describe what is wrong, with the above solution, 
because it seems perfectly legal to me.
There is no substantial overhead, and there is no risc of deadlocks. Or 
do you mean it's bad because it confuses lockdep?

/Thomas
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/