[cfe-dev] libclang crash when parsing MS-style inline assembly

Alp Toker alp at nuanti.com
Thu Nov 7 22:57:12 PST 2013 

Richard,

I've replaced this with a better test in r194249.

Please keep an eye on this, let's see if it works everywhere.

Alp.

On 08/11/2013 04:16, Alp Toker wrote:
> On 07/11/2013 18:10, Richard Barton wrote:
>> Hi all
>>
>> [Adding Alp - didn't realise he was missed off my original mail.]
>>
>> Looking at the code in lib/Parse/ParseStmt.cpp it looks as though this test
>> can only be expected to pass when targeting x86 or x86_64. Any other
>> architecture emits the "Unsupported architecture 'aarch64' for MS-style
>> inline assembly" message.
>>
>> I'd like to XFAIL the test for all the other backends. I've committed the
>> patch (r194211) Please holler if you disagree.
> Hello Richard, thanks for CCing me in and for the XFAIL quick fix.
>
> Your suggestion to combine the two messages could be a good idea.
>
> Alternatively maybe we could move the err_msasm_unable_to_create_target
> check before err_msasm_unsupported_arch, or maybe even remove
> err_msasm_unsupported_arch completely and get the target ASM parser to
> diagnose the condition?
>
> I haven't been able to try these out because c-index-test doesn't accept
> a target triple as far as I can tell, so it's been difficult to work with.
>
> Open to suggestions here, and I've already got it on my long-term TODO
> list to revisit this.
>
> Also the original reporter should file a bug regarding how to enable the
> feature in libclang or it will be forgotten. I think that one is a
> legitimate feature request.
>
> Alp.
>
>
>
>> Thanks,
>> Rich
>>
>>
>>> -----Original Message-----
>>> From: cfe-dev-bounces at cs.uiuc.edu [mailto:cfe-dev-bounces at cs.uiuc.edu] On
>>> Behalf Of Richard Barton
>>> Sent: 04 November 2013 15:25
>>> To: mcrosier at codeaurora.org; William Ledoux
>>> Cc: cfe-dev at cs.uiuc.edu
>>> Subject: Re: [cfe-dev] libclang crash when parsing MS-style inline
>> assembly
>>> Hi all
>>>
>>> Apologies in advance that I don't fully understand the technicalities of
>> this area,
>>> but could that regression test be made more robust?
>>>
>>> My clang build has LLVM_DEFAULT_TARGET=aarch64-none-eabi (Host is
>>> x86_64-linux-gnueabi) and the error message that I get is slightly
>> different
>>> causing the test to fail:
>>>
>>> .../ms-asm-no-target.cpp:8:3: error: Unsupported architecture 'aarch64'
>> for MS-
>>> style inline assembly
>>>
>>> Again, I don't fully understand the area, but the test seems only to
>> require that
>>> an x86 backend is available. It doesn't account for cross-compiling or
>> when the
>>> default triple is not x86. If the test needs to be run on x86, shouldn't
>> the
>>> REQUIRES line be more strict? Alternatively, if my failure mode is equally
>>> acceptable behaviour, could the two error messages be combined?
>>>
>>> Thanks,
>>> Rich
>>>
>>>
>>>> -----Original Message-----
>>>> From: cfe-dev-bounces at cs.uiuc.edu [mailto:cfe-dev-bounces at cs.uiuc.edu]
>>>> On Behalf Of mcrosier at codeaurora.org
>>>> Sent: 30 October 2013 15:58
>>>> To: William Ledoux
>>>> Cc: cfe-dev at cs.uiuc.edu
>>>> Subject: Re: [cfe-dev] libclang crash when parsing MS-style inline
>>> assembly
>>>>> Hi Chad,
>>>>>
>>>>> So now I understand that there is no way to parse MS-style inline
>>>>> assembly without a target asm parser.
>>>>> But I still need to figure out if I can and how to tell libclang to
>>>>> initialize a specific target.
>>>> Unfortunately, I don't have an answer here.  Perhaps Alp or another
>>>> member
>>> of
>>>> the list has a solution.
>>>>
>>>>> On Wed, Oct 30, 2013 at 4:27 PM, <mcrosier at codeaurora.org> wrote:
>>>>>
>>>>>> Hi William,
>>>>>> I can answer your second question.  GCC style inline assembly does
>>>>>> not require a target AsmParser.  MS-style inline assembly requires
>>>>>> a parser because this is how we discover the constraints.  Hope
>>>>>> this
>>> helps.
>>>>>>  Chad
>>>>>>
>>>>>>> Hi Alp,
>>>>>>>
>>>>>>> Thanks for helping !
>>>>>>>
>>>>>>> 1] Is there a way I could tell libclang which target to
>>>>>>> initialize
>>>>>> without
>>>>>>> modifying it ?
>>>>>>> 2] How does it work for gcc style assembly, is there a similar
>>>>>>> error message or does it work without the need of any targets ?
>>>>>>> 3] Since you decided to diagnose this, you might also want to
>>>>>>> check
>>>>>> the
>>>>>>> case were the target have been initialized, but not the used
>>>>>>> function pointers of this target
>>>>>>>
>>>>>>> William.
>>>>>>>
>>>>>>>
>>>>>>> On Wed, Oct 30, 2013 at 3:53 PM, Alp Toker <alp at nuanti.com> wrote:
>>>>>>>
>>>>>>>>  Hello William,
>>>>>>>>
>>>>>>>> I've landed a fix in r193685 to diagnose this instead of crashing:
>>>>>>>>
>>>>>>>> clang/test/Index/ms-asm-no-target.cpp:8:3: error: MS-style
>>>>>>>> inline assembly is not available: Unable to find target for this
>>>>>>>> triple (no targets
>>>>>> are
>>>>>>>> registered)
>>>>>>>>
>>>>>>>> It's up to the embedder to decide what targets are linked in, so
>>>>>>>> how
>>>>>> the
>>>>>>>> API exposes that is still an open question.
>>>>>>>>
>>>>>>>> Alp.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> On 30/10/2013 11:50, William Ledoux wrote:
>>>>>>>>
>>>>>>>> Hello,
>>>>>>>>
>>>>>>>> The following minimal code that contain MS-style inline assembly
>>>>>>>> will compile fine with clang, but libclang fails to parse it
>>>>>>>> (clang_parseTranslationUnit will return NULL).
>>>>>>>>
>>>>>>>>    void Break(){ __asm { int 3 } }
>>>>>>>>
>>>>>>>> In yesterday's llvm and clang sources, the problem occurs in
>>>>>>>> ParseMicrosoftAsmStatement.
>>>>>>>> In the code below, because no target have been registered, the
>>>>>>>> first line will set TheTarget to NULL, and the second line will
>>>>>>>> dereference TheTarget, thus causing the problem.
>>>>>>>>
>>>>>>>>    const llvm::Target *TheTarget =
>>>>>>>> llvm::TargetRegistry::lookupTarget(TT,
>>>>>>>> Error);
>>>>>>>>    OwningPtr<llvm::MCRegisterInfo>
>>>>>> MRI(TheTarget->createMCRegInfo(TT));
>>>>>>>> For what I understood, clang, in cc1_main, will initialize
>>>>>>>> targets
>>>>>> and
>>>>>>>> targets' functions with the following 4 lines, whereas libclang
>>>>>> won't.
>>>>>>>>   llvm::InitializeAllTargets();
>>>>>>>>   llvm::InitializeAllTargetMCs();
>>>>>>>>   llvm::InitializeAllAsmPrinters();
>>>>>>>>   llvm::InitializeAllAsmParsers();
>>>>>>>>
>>>>>>>> Just for testing purpose, adding those 4 lines somewhere in
>>>>>>>> clang_createIndex fixes the problem. I know this is probably
>>>>>>>> wrong,
>>>>>> but
>>>>>>>> did
>>>>>>>> it just to see if more problems were hiding behind.
>>>>>>>>
>>>>>>>> So my questions are:
>>>>>>>>   1] Is it wanted that libclang doesn't initialize any target ?
>>>>>>>>   2] if yes, how shloud it behave with MS inline assembly ?
>>>>>>>>   3] if no, what is the proper way to make it initialize targets ?
>>>>>>>>
>>>>>>>> Many thanks for you work !
>>>>>>>> William
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> View this message in context:
>>>>>>>>
>>>>>> http://clang-developers.42468.n3.nabble.com/libclang-crash-when-par
>>>>>> si ng-MS-style-inline-assembly-tp4035432.html
>>>>>>>> Sent from the Clang Developers mailing list archive at Nabble.com.
>>>>>>>> _______________________________________________
>>>>>>>> cfe-dev mailing
>>>>>>>> listcfe-dev at cs.uiuc.eduhttp://
>>>>>> lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>>>>>>> -- http://www.nuanti.com
>>>>>>>> the browser experts
>>>>>>>>
>>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> cfe-dev mailing list
>>>>>>> cfe-dev at cs.uiuc.edu
>>>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>>>>>>
>>>>>>
>>>> _______________________________________________
>>>> cfe-dev mailing list
>>>> cfe-dev at cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>>
>>>
>>> _______________________________________________
>>> cfe-dev mailing list
>>> cfe-dev at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>

-- 
http://www.nuanti.com
the browser experts

More information about the cfe-commits mailing list