[Lldb-commits] [lldb] r202956 - Add support for JIT debugging on Linux using the GDB JIT interface. Patch written with Keno Fischer.

Thu Mar 6 10:06:31 PST 2014

The make files should be changed to be included on all platforms. 

I don't know why people keep making plug-ins that will work fine remotely, but make them only build natively on the host they want to use them on. LLDB does remote debugging, so we need and all plug-ins that can be built for cross debugging to always be built.

On Mar 6, 2014, at 3:26 AM, Ahmed Charles <acharles at outlook.com> wrote:

> This patch seems to break builds on windows, since the JITLoaderGDB class is only included in cmake on FreeBSD and Linux, but it is used in lldb.cpp on all platforms.
> 
> 
> 
> ________________________________
>> From: andrew.macp at gmail.com 
>> Date: Thu, 6 Mar 2014 10:52:36 +0100 
>> To: jingham at apple.com 
>> CC: lldb-commits at cs.uiuc.edu 
>> Subject: Re: [Lldb-commits] [lldb] r202956 - Add support for JIT  
>> debugging on Linux using the GDB JIT interface. Patch written with Keno  
>> Fischer. 
>> 
>> Hi Jim, 
>> 
>> I agree that this part of the patch isn't pretty but I wasn't able to  
>> find a way to be notified when a shared library is loaded (exactly as  
>> you suggest). We need to check on launch or attach and then whenever a  
>> new library is loaded, if there's a better way to do that it would be  
>> great to clean this up. 
>> 
>> I can't reproduce the crash on Linux but it looks like we need to  
>> unregister the notification callback that's set, that's the only reason  
>> I can think of that you would end up in the ProcessStateChangedCallback  
>> with an invalid baton pointer. I've attached a patch that I'm hoping  
>> will resolve the issue here, it's definitely a bug that it wasn't being  
>> unregistered. 
>> 
>> I hadn't intended for the JITLoaderGDB to be enabled on OSX as I wasn't  
>> able to test it there, sorry about that. It's likely not useful there  
>> right now as it would need modifications to the ObjectFileMachO to do  
>> some relocations, and additionally MCJIT in LLVM would need to be  
>> modified to call the GDB hook on OSX (though other JIT hosts may work).  
>> That said it shouldn't be causing crashes if enabled. 
>> 
>> Thanks for looking into this. 
>> 
>> Cheers, 
>> Andrew 
>> 
>> 
>> 
>> On Thu, Mar 6, 2014 at 4:28 AM, Jim Ingham  
>> <jingham at apple.com<mailto:jingham at apple.com>> wrote: 
>> This part of the patch worries me.  If I am debugging a process that  
>> doesn’t have this JIT loader symbol, this bit means every time that the  
>> process stops for any reason you will search the whole world for some  
>> symbol that won’t be found.  That’s something we really avoid doing if  
>> we can, programs get pretty big and this is not the sort of thing you  
>> want to do. 
>> 
>> I don’t know how this symbol comes about, is there no event (shared  
>> library load or something) that you can hook into to find this symbol? 
>> 
>> This patch is also causing a crash on Mac OS X just running a program.  
>>  The crash looks like: 
>> 
>> (lldb) bt 
>> * thread #8: tid = 0xf68e5, name =  
>> <lldb.process.internal-state(pid=40372)>, function:  
>> lldb_private::Process::GetTarget() , stop reason = EXC_BAD_ACCESS  
>> (code=1, address=0x100) 
>>     frame #0: 0x0000000106f8072c  
>> LLDB`lldb_private::Process::GetTarget() at Process.h:2516 
>>     frame #1: 0x0000000108e7aa5a  
>> LLDB`JITLoaderGDB::GetSymbolAddress(lldb_private::ConstString const&,  
>> lldb::SymbolType) const at JITLoaderGDB.cpp:368 
>>     frame #2: 0x0000000108e7a8bf LLDB`JITLoaderGDB::SetJITBreakpoint()  
>> at JITLoaderGDB.cpp:99 
>>     frame #3: 0x0000000108e7a6d8  
>> LLDB`JITLoaderGDB::ProcessStateChangedCallback(void*,  
>> lldb_private::Process*, lldb::StateType) at JITLoaderGDB.cpp:354 
>>     frame #4: 0x0000000108b7b29b  
>> LLDB`lldb_private::Process::SynchronouslyNotifyStateChanged(lldb::StateType)  
>> at Process.cpp:1223 
>>     frame #5: 0x0000000108b89762  
>> LLDB`lldb_private::Process::ShouldBroadcastEvent(lldb_private::Event*)  
>> at Process.cpp:3846 
>>     frame #6: 0x0000000108b8454d  
>> LLDB`lldb_private::Process::HandlePrivateEvent(std::__1::shared_ptr<lldb_private::Event>&)  
>> at Process.cpp:4141 
>>     frame #7: 0x0000000108b8a755  
>> LLDB`lldb_private::Process::RunPrivateStateThread() at Process.cpp:4290 
>>     frame #8: 0x0000000108b89bfd  
>> LLDB`lldb_private::Process::PrivateStateThread(void*) at  
>> Process.cpp:4221 
>>     frame #9: 0x00000001087d811a LLDB`ThreadCreateTrampoline(void*) at  
>> Host.cpp:629 
>>     frame #10: 0x00007fff815df899 libsystem_pthread.dylib`_pthread_body 
>>     frame #11: 0x00007fff815df72a libsystem_pthread.dylib`_pthread_start 
>>     frame #12: 0x00007fff815e3fc9 libsystem_pthread.dylib`thread_start 
>> (lldb) f 2 
>> frame #2: 0x0000000108e7a8bf LLDB`JITLoaderGDB::SetJITBreakpoint() at  
>> JITLoaderGDB.cpp:99 
>>    96          log->Printf("JITLoaderGDB::%s looking for JIT register hook", 
>>    97                      __FUNCTION__); 
>>    98 
>> -> 99      addr_t jit_addr =  
>> GetSymbolAddress(ConstString("__jit_debug_register_code"), 
>>    100                                        eSymbolTypeAny); 
>>    101     if (jit_addr == LLDB_INVALID_ADDRESS) 
>>    102         return; 
>> (lldb) expr *this 
>> (JITLoaderGDB) $13 = { 
>>   lldb_private::JITLoader = { 
>>     m_process = 0x0000000000000000 Public:  
>> lldb_private::ThreadSafeValue<lldb::StateType> @  Private:  
>> lldb_private::ThreadSafeValue<lldb::StateType> @ 
>>   } 
>>   m_jit_objects = size=160215376 { 
>>     [0] = { 
>>       first = <parent is NULL> 
>>       second = <parent is NULL> 
>>     } 
>>     ... 
>>   } 
>>   m_jit_break_id = 0 
>>   m_notification_callbacks = { 
>>     baton = 0x0000000000000001 
>>     initialize = 0x00007fc54b00f3b0 
>>     process_state_changed = 0x00000001098cb150 (vtable for  
>> std::__1::__shared_ptr_pointer<lldb_private::Section*,  
>> std::__1::default_delete<lldb_private::Section>,  
>> std::__1::allocator<lldb_private::Section>> + 16) 
>>   } 
>> } 
>> 
>> Looks like the JIT instance that is getting passed in is not good for  
>> some reason. 
>> 
>> Jim 
>> 
>> 
>> On Mar 5, 2014, at 2:12 AM, Andrew MacPherson  
>> <andrew.macp at gmail.com<mailto:andrew.macp at gmail.com>> wrote: 
>> 
>> +void 
>> +JITLoaderGDB::ProcessStateChangedCallback(void *baton, 
>> +                                          lldb_private::Process *process, 
>> +                                          lldb::StateType state) 
>> +{ 
>> +    JITLoaderGDB* instance = static_cast<JITLoaderGDB*>(baton); 
>> + 
>> +    switch (state) 
>> +    { 
>> +    case eStateConnected: 
>> +    case eStateAttaching: 
>> +    case eStateLaunching: 
>> +    case eStateInvalid: 
>> +    case eStateUnloaded: 
>> +    case eStateExited: 
>> +    case eStateDetached: 
>> +        // instance->Clear(false); 
>> +        break; 
>> + 
>> +    case eStateRunning: 
>> +    case eStateStopped: 
>> +        // Keep trying to set our JIT breakpoint each time we stop until we 
>> +        // succeed 
>> +        if (!instance->DidSetJITBreakpoint() && process->IsAlive()) 
>> +            instance->SetJITBreakpoint(); 
>> +        break; 
>> + 
>> +    case eStateStepping: 
>> +    case eStateCrashed: 
>> +    case eStateSuspended: 
>> +        break; 
>> +    } 
>> +} 
>> + 
>> 
>> 
>> 
>> _______________________________________________ lldb-commits mailing  
>> list lldb-commits at cs.uiuc.edu  
>> http://lists.cs.uiuc.edu/mailman/listinfo/lldb-commits 		 	   		  
> _______________________________________________
> lldb-commits mailing list
> lldb-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/lldb-commits