[lldb-dev] More linux process control and IOHandler races
Shawn Best
sbest at blueshiftinc.com
Wed Aug 6 17:54:50 PDT 2014
Matt,
I think you are probably right, although there are other places where it
directly calls SetPublicState(). I was wondering about the possibility
there could be other listeners waiting for a broadcast public Stop
event. Is that a possibility?
Some others here were investigating some unit tests that were failing
intermittently (StopHook). Their description of the problem sounds
unrelated to the launch code, but this patch also magically fixes that.
Shawn.
On 8/6/2014 6:26 AM, Matthew Gardiner wrote:
> Shawn,
>
> Like I said earlier your patch worked. However I think the right fix
> is to arrange that ShouldBroadcast returns false for this first stop.
> I believe this, because firstly no stops should be reported here since
> the user is only interested in launching a program, and additionally
> because it enables us to fix lldb without removing the call to
> HandlePrivateEvent. This, I think, is important to preserve as the
> central point for process state change handling.
>
> Matt
>
>
>
> Shawn Best wrote:
>> Hi Matthew,
>>
>> I have also been tracking this bug. I believe there are other bugs
>> in the unit tests failing indirectly because of this. I also have a
>> patch that will fix it, but was sitting on it until the other one
>> landed. These bugs do not show up on OSX since the inferiors are
>> launched separately then attached to.
>>
>> The first odd thing the launching code does is push an IOHandler when
>> it sees the state transition to 'launching'. This is odd because I
>> believe the launching program will always come up in a stopped state
>> which will immediately pop the IOHandler.
>>
>> At launch, the process comes up in the stopped state. The launch
>> code manually calls HandlePrivateEvent() with the stop event, which
>> then broadcasts the Event. When HandleProcessEvent gets the public
>> stop, it dumps out the current thread state just as if an executing
>> inferior hit a breakpoint and stopped.
>>
>> One way to fix this would be:
>>
>> 1. Don't push io handler when state is 'launching'
>> 2. Instead of manually calling HandlePrivateEvent, call
>> SetPublicState().
>>
>> Alternately, we could try and debug why ShouldBroadcast() returns
>> true, but that appears to be by design since it is expecting the
>> public stop event to pop the IOHandler that had been pushed when
>> launching.
>>
>> I have attached a patch demonstrating this. In conjunction with the
>> other patch for IOHandler race condition, it will fix a bunch of this
>> kind of behaviour.
>>
>> Shawn.
>>
>> On 8/5/2014 6:59 AM, Matthew Gardiner wrote:
>>> Jim,
>>>
>>> I've been trying to debug an issue (I see it on 64-bit linux) where,
>>> I do "target create" and "process launch" and despite not requesting
>>> *stop at entry*, the first stop (which I believe is just the initial
>>> ptrace attach stop) is reported to the lldb command line. I added
>>> some fprintf to Process::HandlePrivateEvent, which counts the number
>>> of eStoppedState events seen and whether ShouldBroadcastEvent
>>> returns true for this event. Here's the output from my program with
>>> diagnostic:
>>>
>>> (lldb) target create ~/me/i64-hello.elf
>>> Current executable set to '~/me/i64-hello.elf' (x86_64).
>>> (lldb) process launch
>>> MG Process::HandlePrivateEvent launching stopped_count 0
>>> should_broadcast 1
>>> Process 31393 launching
>>> MG Process::HandlePrivateEvent stopped stopped_count 1
>>> should_broadcast 1
>>> MG Process::HandlePrivateEvent running stopped_count 1
>>> should_broadcast 1
>>> Process 31393 launched: 'i64-hello.elf' (x86_64)
>>> Process 31393 stopped
>>> * thread #1: tid = 31393, 0x0000003675a011f0, name =
>>> 'i64-hello.elf', stop reason = trace
>>> frame #0: 0x0000003675a011f0
>>> -> 0x3675a011f0: movq %rsp, %rdi
>>> 0x3675a011f3: callq 0x3675a046e0
>>> 0x3675a011f8: movq %rax, %r12
>>> 0x3675a011fb: movl 0x21eb97(%rip), %eax
>>> (lldb) MG Process::HandlePrivateEvent stopped stopped_count 2
>>> should_broadcast 0
>>> MG Process::HandlePrivateEvent running stopped_count 2
>>> should_broadcast 0
>>> MG Process::HandlePrivateEvent stopped stopped_count 3
>>> should_broadcast 0
>>> MG Process::HandlePrivateEvent running stopped_count 3
>>> should_broadcast 0
>>>
>>> In summary, lldb reports the inferior to be stopped (even though
>>> /proc/pid/status and lldb "target list" say it is running). Clearly
>>> this is wrong (hence my earlier post).
>>>
>>> Am I correct in assuming that when ShouldBroadcastEvent returns
>>> true this means that lldb should show this event to the debug user?
>>> (And thus hide other events where ShouldBroadcastEvent=false).
>>>
>>> What puzzled me was why ShouldBroadcastEvent return true for this
>>> very first stop. Is this a bug?
>>>
>>> I also spent sometime at ShouldBroadcastEvent and saw that this:
>>>
>>> case eStateStopped:
>>> case eStateCrashed:
>>> case eStateSuspended:
>>> {
>>> ....
>>> if (was_restarted || should_resume ||
>>> m_resume_requested)
>>> {
>>>
>>> evaluates as false, and hence the PrivateResume code is not
>>> called... does this seem buggy to you for this very first stop?
>>>
>>> I thought I'd try asking you, since in a previous mail from Greg, he
>>> cited you as being a thread-plan expert. (Hope that's ok!). I'd
>>> really appreciate your help in clarifying the above questions for
>>> me, and if you have time, giving me some ideas as to how to trace
>>> this one further e.g. how m_thread_list.ShouldStop and
>>> m_thread_list.ShouldReportStop should behave, etc.
>>>
>>> thanks for your help
>>> Matt
>>>
>>> Matthew Gardiner wrote:
>>>> Folks,
>>>>
>>>> In addition to the overlapping prompt race Shawn Best and myself
>>>> are looking at, I'm seeing another issue where if I launch a
>>>> process, I get a stop (presumably the in) being reported to the UI.
>>>>
>>>> (lldb) target create ~/mydir/i64-hello.elf
>>>> Current executable set to '~/mydir/i64-hello.elf' (x86_64).
>>>> (lldb) process launch
>>>> Process 27238 launching
>>>> Process 27238 launched: '64-hello.elf' (x86_64)
>>>> Process 27238 stopped
>>>> * thread #1: tid = 27238, 0x0000003675a011f0, name = 'i64-hello.elf'
>>>> frame #0:
>>>> (lldb) target list
>>>> Current targets:
>>>> * target #0: i64-hello.elf ( arch=x86_64-unknown-linux,
>>>> platform=host, pid=27238, state=running )
>>>> (lldb)
>>>>
>>>> As you can see the "target list" reflects that the process is
>>>> running. Which I confirmed by looking at /proc/27238/status.
>>>>
>>>> Anyone else seeing this?
>>>>
>>>> thanks
>>>> Matt
>>>>
>>>>
>>>>
>>>> Member of the CSR plc group of companies. CSR plc registered in
>>>> England and Wales, registered number 4187346, registered office
>>>> Churchill House, Cambridge Business Park, Cowley Road, Cambridge,
>>>> CB4 0WZ, United Kingdom
>>>> More information can be found at www.csr.com <http://www.csr.com>.
>>>> Keep up to date with CSR on our technical blog, www.csr.com/blog
>>>> <http://www.csr.com/blog>, CSR people blog, www.csr.com/people
>>>> <http://www.csr.com/people>, YouTube, www.youtube.com/user/CSRplc
>>>> <http://www.youtube.com/user/CSRplc>, Facebook,
>>>> www.facebook.com/pages/CSR/191038434253534
>>>> <http://www.facebook.com/pages/CSR/191038434253534>, or follow us
>>>> on Twitter at www.twitter.com/CSR_plc
>>>> <http://www.twitter.com/CSR_plc>.
>>>> New for 2014, you can now access the wide range of products powered
>>>> by aptX at www.aptx.com <http://www.aptx.com>.
>>>> _______________________________________________
>>>> lldb-dev mailing list
>>>> lldb-dev at cs.uiuc.edu <mailto:lldb-dev at cs.uiuc.edu>
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
>>>>
>>>>
>>>> To report this email as spam click
>>>> https://www.mailcontrol.com/sr/EjKNgqvIx0TGX2PQPOmvUj!GOBh06pKKNwnTW0ZqkNYNbZeofOurgZMo6Cl2EgPiaCw7kl6fPUTCXaTERp6oIw==
>>>> <https://www.mailcontrol.com/sr/EjKNgqvIx0TGX2PQPOmvUj%21GOBh06pKKNwnTW0ZqkNYNbZeofOurgZMo6Cl2EgPiaCw7kl6fPUTCXaTERp6oIw==>
>>>> .
>>>
>>> _______________________________________________
>>> lldb-dev mailing list
>>> lldb-dev at cs.uiuc.edu <mailto:lldb-dev at cs.uiuc.edu>
>>> http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
>>
>
More information about the lldb-dev
mailing list