Re: [DynInst_API:] PatchFunction blocks() crash

[Bill Williams <bill@xxxxxxxxxxx>]

On 03/12/2015 04:21 PM, Donghai wrote:
> Hi,
>
>  From the back trace info, the *_obj* pointer is NULL.
> PatchObject *pbObj = _obj->addrSpace()->findObject(block->obj());
It's not actually the _obj, but its address space, that's NULL. The 
example is, I think, missing a step; you need to create an address space 
from your PatchObject with AddrSpace::create().
It's also possible that the correct logic here is:

PatchObject* pbObj;
if(_obj->addrSpace()) // block may not be in this PatchObject
{
    pbObj = _obj->addrSpace()->findObject(block->obj());
}
else // Everything must be local, as there's no address space
{
    pbObj = _obj;
}

Certainly in order to use PatchAPI for anything that you can't do with 
ParseAPI, you need an address space, which is why we haven't seen this 
in real tests...
> On Wed, Mar 11, 2015 at 5:20 PM, Donghai <donghaitad@xxxxxxxxx
> <mailto:donghaitad@xxxxxxxxx>> wrote:
>
>     Hi Bill,
>
>     The backtrace info is as follows:
>
>     #0  Dyninst::PatchAPI::AddrSpace::findObject (this=0x0, co=0x8071648)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/patchAPI/src/AddrSpace.C:104
>     #1  0xb7e18deb in Dyninst::PatchAPI::PatchParseCallback::add_edge_cb (
>          this=0x8089890, block=0x8087708, edge=0x8082958,
>          type=Dyninst::ParseAPI::ParseCallback::target)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/patchAPI/src/ParseCallback.C:148
>     #2  0xb7e7cfb3 in add_edge_cb
>     (t=Dyninst::ParseAPI::ParseCallback::target,
>          e=0x8082958, b=0x8087708, this=0x8075c78)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/src/ParseCallback.C:254
>     #3  Dyninst::ParseAPI::ParseCallbackManager::addEdge (this=0x8075c78,
>          b=0x8087708, e=0x8082958,
>     t=Dyninst::ParseAPI::ParseCallback::target)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/src/ParseCallback.C:147
>     #4  0xb7e5e9fe in Dyninst::ParseAPI::Parser::link (this=0x8074f20,
>          src=0x8087708, dst=0x80875c0, et=Dyninst::ParseAPI::RET,
>     sink=false)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/src/Parser.C:1720
>     #5  0xb7e744b9 in Dyninst::ParseAPI::CodeObject::add_edge
>     (this=0x8071648,
>          src=0x8087708, trg=0x80875c0, et=Dyninst::ParseAPI::RET)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/src/CodeObject.C:184
>     #6  0xb7e6d4a6 in Dyninst::ParseAPI::Function::delayed_link_return (
>          this=0x8076088, o=0x8071648, retblk=0x8087708)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/src/Function.C:358
>     #7  0xb7e6dc80 in Dyninst::ParseAPI::Function::blocks_int
>     (this=0x8076088)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/src/Function.C:306
>     ---Type <return> to continue, or q <return> to quit---
>     #8  0xb7e65843 in Dyninst::ParseAPI::Parser::finalize (this=0x8074f20,
>          f=0x8076088)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/src/Parser.C:653
>     #9  0xb7e6c87e in Dyninst::ParseAPI::Function::finalize (this=0x8076088)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/src/Function.C:184
>     #10 0xb7e063e8 in num_blocks (this=0x8076088)
>          at /home/tad/Desktop/Dyninst/Dyninst-8.2.1/parseAPI/h/CFG.h:489
>     #11 Dyninst::PatchAPI::PatchFunction::blocks (this=0x808b270)
>          at
>     /home/tad/Desktop/Dyninst/Dyninst-8.2.1/patchAPI/src/PatchFunction.C:54
>     #12 0x0804915f in main ()
>
>     On Wed, Mar 11, 2015 at 12:00 PM, Bill Williams <bill@xxxxxxxxxxx
>     <mailto:bill@xxxxxxxxxxx>> wrote:
>
>         On 03/10/2015 08:26 PM, Donghai wrote:
>
>             Hi,
>
>             I use a very simple program to test the first example of
>             PatchAPI.
>
>             Unfortunately, the dyninst gets crash on this example. More
>
>             specifically, func->blocks() cause the system segment fault.
>
>             The code is pasted as follows, can someone give me some hint?
>
>             Thanks.
>
>         Can you send us a stack trace from gdb of the fault?
>
>         It seems likely that some of our parseAPI optimizations are not
>         playing nicely with this example, and that we're assuming data
>         structures are populated before they actually are. If adding
>         co->parse() before you create the PatchObject eliminates the
>         crash, that's a good way to check whether we're being
>         inappropriately lazy.
>
>
>             Best
>             Hai
>
>             ParseAPI::SymtabCodeSource *sts;
>             ParseAPI::CodeObject *co;
>             Address code_base = 0x8048000;
>
>             sts = new ParseAPI::SymtabCodeSource( argv[1] );
>             co = new ParseAPI::CodeObject( sts );
>
>             PatchObject *obj = PatchObject::create(co, code_base);
>
>             // Find all functions in the object
>             std::vector<PatchFunction*> all;
>             obj->funcs(back_inserter(all))__;
>
>             for (std::vector<PatchFunction*>::__iterator fi = all.begin();
>                    fi != all.end(); fi++) {
>                 // Print out each function's name
>                 PatchFunction* func = *fi;
>                 std::cout << func->name() << std::endl;
>
>                 const PatchFunction::Blockset& blks = func->blocks();
>                 for (PatchFunction::BlockSet::__iterator bi = blks.begin();
>                      bi != blks.end(); bi++) {
>                   // Print out each block's size
>                   PatchBlock* blk = *bi;
>                   std::cout << "\tBlock size:" << blk->size() << std::endl;
>                 }
>             }
>
>
>             _________________________________________________
>             Dyninst-api mailing list
>             Dyninst-api@xxxxxxxxxxx <mailto:Dyninst-api@xxxxxxxxxxx>
>             https://lists.cs.wisc.edu/__mailman/listinfo/dyninst-api
>             <https://lists.cs.wisc.edu/mailman/listinfo/dyninst-api>
>
>
>
>         --
>         --bw
>
>         Bill Williams
>         Paradyn Project
>         bill@xxxxxxxxxxx <mailto:bill@xxxxxxxxxxx>
>         _________________________________________________
>         Dyninst-api mailing list
>         Dyninst-api@xxxxxxxxxxx <mailto:Dyninst-api@xxxxxxxxxxx>
>         https://lists.cs.wisc.edu/__mailman/listinfo/dyninst-api
>         <https://lists.cs.wisc.edu/mailman/listinfo/dyninst-api>

--
--bw

Bill Williams
Paradyn Project
bill@xxxxxxxxxxx