[llvm] r209478 - [RuntimeDyld] Teach RuntimeDyldMachO how to handle scattered VANILLA relocs on
David Blaikie
dblaikie at gmail.com
Thu May 22 16:30:41 PDT 2014
On Thu, May 22, 2014 at 4:21 PM, Lang Hames <lhames at gmail.com> wrote:
> I'm just testing with a local copy.
Right - but I assume you're seeing a bunch of failures? I just mean
that you could XFAIL those current failures, then remove the XFAILs as
you cleanup. Not expecting you to setup a bot today - would be nice at
some point - but in the interim I'd be willing to take your word for
it that the XFAILs are failing and when you remove them it's because
you've fixed them and they now pass.
> You can build an i386 LLVM on an x86-64
> MacOSX box with:
>
> cmake -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CROSSCOMPILING=True
> -DLLVM_TARGET_ARCH=i386-apple-macosx10.9 -DLLVM_TARGETS_TO_BUILD=X86
> -DLLVM_ENABLE_LIBCXX=True -DCMAKE_C_FLAGS="-m32 -target
> i386-apple-macosx10.8" -DCMAKE_CXX_FLAGS="-m32 -target
> i386-apple-macosx10.9"
>
> I would guess the Linux equivalent would be:
>
> cmake -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CROSSCOMPILING=True
> -DLLVM_TARGET_ARCH=i386-pc-linux-gnu -DLLVM_TARGETS_TO_BUILD=X86
> -DLLVM_ENABLE_LIBCXX -DCMAKE_C_FLAGS="-m32 -target i386-pc-linux-gnu"
> -DCMAKE_CXX_FLAGS="-m32 -target i386-pc-linux-gnu"
>
> I'm still a bit fuzzy on how to XFAIL things for specific Arch/OS
> combinations, which we need for the JIT (e.g. Some test works on
> i386-apple-macosx, but not on i386-pc-linux-gnu, because the object format
> support differs). Pointers and/or patches welcome.
Judging by the XFAILs for i386 (all of which are in MCJIT) that exist
already, I assume all you need to do is "XFAIL: i386" ("grep -r
XFAIL.*i386 test" shows a bunch of existing xfails for a variety of
things).
>
> - Lang.
>
>
>
> On Thu, May 22, 2014 at 4:07 PM, David Blaikie <dblaikie at gmail.com> wrote:
>>
>> On Thu, May 22, 2014 at 4:03 PM, Lang Hames <lhames at gmail.com> wrote:
>> > We don't have an i386 bot, I suspect nobody does (or they've been
>> > ignoring
>> > the fact that these have been red).
>> >
>> > Something like that sounds very useful. :)
>>
>> I'm assuming you've got some kind of i386 machine to test this on - is
>> it worth baselining (marking as XFAIL) that so at least /someone/
>> (presumably at least you, maybe other people at some point) can run
>> make check, get a green result, get an XPASS failure when you fix
>> things, etc?
>>
>> >
>> > - Lang.
>> >
>> >
>> > On Thu, May 22, 2014 at 3:48 PM, David Blaikie <dblaikie at gmail.com>
>> > wrote:
>> >>
>> >> On Thu, May 22, 2014 at 3:30 PM, Lang Hames <lhames at gmail.com> wrote:
>> >> > Author: lhames
>> >> > Date: Thu May 22 17:30:13 2014
>> >> > New Revision: 209478
>> >> >
>> >> > URL: http://llvm.org/viewvc/llvm-project?rev=209478&view=rev
>> >> > Log:
>> >> > [RuntimeDyld] Teach RuntimeDyldMachO how to handle scattered VANILLA
>> >> > relocs on
>> >> > i386.
>> >> >
>> >> > This fixes two more MCJIT regression tests on i386:
>> >>
>> >> Do we have a public bot running these tests? (why weren't they XFAIL'd
>> >> until now?) Do you have one internally at least that's green (how is
>> >> it green if these tests aren't XFAIL'd on i386?) so you'll know (&
>> >> presumably tell the community) if these regress?
>> >>
>> >> >
>> >> > ExecutionEngine/MCJIT/2003-05-06-LivenessClobber.ll
>> >> > ExecutionEngine/MCJIT/2013-04-04-RelocAddend.ll
>> >> >
>> >> > The implementation of processScatteredVANILLA is tasteless
>> >> > (*ba-dum-ching*),
>> >> > but I'm working on a substantial tidy-up of RuntimeDyldMachO that
>> >> > should
>> >> > improve things.
>> >> >
>> >> > This patch also fixes a type-o in
>> >> > RuntimeDyldMachO::processSECTDIFFRelocation,
>> >> > and teaches that method to skip over the PAIR reloc following the
>> >> > SECTDIFF.
>> >> >
>> >> > <rdar://problem/16961886>
>> >> >
>> >> >
>> >> > Modified:
>> >> > llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp
>> >> > llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h
>> >> >
>> >> > Modified:
>> >> > llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp
>> >> > URL:
>> >> >
>> >> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp?rev=209478&r1=209477&r2=209478&view=diff
>> >> >
>> >> >
>> >> > ==============================================================================
>> >> > --- llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp
>> >> > (original)
>> >> > +++ llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp
>> >> > Thu
>> >> > May 22 17:30:13 2014
>> >> > @@ -460,7 +460,7 @@ relocation_iterator RuntimeDyldMachO::pr
>> >> >
>> >> > uint32_t AddrB = MachO->getScatteredRelocationValue(RE2);
>> >> > section_iterator SBI = getSectionByAddress(*MachO, AddrB);
>> >> > - assert(SBI != MachO->section_end() && "Can't find seciton for
>> >> > address
>> >> > B");
>> >> > + assert(SBI != MachO->section_end() && "Can't find section for
>> >> > address
>> >> > B");
>> >> > uint64_t SectionBBase;
>> >> > SBI->getAddress(SectionBBase);
>> >> > uint64_t SectionBOffset = AddrB - SectionBBase;
>> >> > @@ -483,7 +483,48 @@ relocation_iterator RuntimeDyldMachO::pr
>> >> > addRelocationForSection(R, SectionAID);
>> >> > addRelocationForSection(R, SectionBID);
>> >> >
>> >> > - return RelI;
>> >> > + return ++RelI;
>> >> > +}
>> >> > +
>> >> > +relocation_iterator RuntimeDyldMachO::processI386ScatteredVANILLA(
>> >> > + unsigned SectionID,
>> >> > + relocation_iterator
>> >> > RelI,
>> >> > + ObjectImage &Obj,
>> >> > + ObjSectionToIDMap
>> >> > &ObjSectionToID) {
>> >> > + const MachOObjectFile *MachO =
>> >> > + static_cast<const MachOObjectFile*>(Obj.getObjectFile());
>> >> > + MachO::any_relocation_info RE =
>> >> > + MachO->getRelocation(RelI->getRawDataRefImpl());
>> >> > +
>> >> > + SectionEntry &Section = Sections[SectionID];
>> >> > + uint32_t RelocType = MachO->getAnyRelocationType(RE);
>> >> > + bool IsPCRel = MachO->getAnyRelocationPCRel(RE);
>> >> > + unsigned Size = MachO->getAnyRelocationLength(RE);
>> >> > + uint64_t Offset;
>> >> > + RelI->getOffset(Offset);
>> >> > + uint8_t *LocalAddress = Section.Address + Offset;
>> >> > + unsigned NumBytes = 1 << Size;
>> >> > + int64_t Addend = 0;
>> >> > + memcpy(&Addend, LocalAddress, NumBytes);
>> >> > +
>> >> > + unsigned SymbolBaseAddr = MachO->getScatteredRelocationValue(RE);
>> >> > + section_iterator TargetSI = getSectionByAddress(*MachO,
>> >> > SymbolBaseAddr);
>> >> > + assert(TargetSI != MachO->section_end() && "Can't find section for
>> >> > symbol");
>> >> > + uint64_t SectionBaseAddr;
>> >> > + TargetSI->getAddress(SectionBaseAddr);
>> >> > + SectionRef TargetSection = *TargetSI;
>> >> > + bool IsCode;
>> >> > + TargetSection.isText(IsCode);
>> >> > + uint32_t TargetSectionID = findOrEmitSection(Obj, TargetSection,
>> >> > IsCode,
>> >> > + ObjSectionToID);
>> >> > +
>> >> > + Addend -= SectionBaseAddr;
>> >> > + RelocationEntry R(SectionID, Offset, RelocType, Addend,
>> >> > + IsPCRel, Size);
>> >> > +
>> >> > + addRelocationForSection(R, TargetSectionID);
>> >> > +
>> >> > + return ++RelI;
>> >> > }
>> >> >
>> >> > relocation_iterator RuntimeDyldMachO::processRelocationRef(
>> >> > @@ -498,17 +539,22 @@ relocation_iterator RuntimeDyldMachO::pr
>> >> > uint32_t RelType = MachO->getAnyRelocationType(RE);
>> >> >
>> >> > // FIXME: Properly handle scattered relocations.
>> >> > - // For now, optimistically skip these: they can often be
>> >> > ignored, as
>> >> > - // the static linker will already have applied the
>> >> > relocation,
>> >> > and it
>> >> > - // only needs to be reapplied if symbols move relative to
>> >> > one
>> >> > another.
>> >> > - // Note: This will fail horribly where the relocations *do*
>> >> > need to be
>> >> > - // applied, but that was already the case.
>> >> > + // Special case the couple of scattered relocations that we
>> >> > know how
>> >> > + // to handle: SECTDIFF relocations, and scattered VANILLA
>> >> > relocations
>> >> > + // on I386.
>> >> > + // For all other scattered relocations, just bail out and
>> >> > hope
>> >> > for the
>> >> > + // best, since the offsets computed by scattered
>> >> > relocations
>> >> > have often
>> >> > + // been optimisticaly filled in by the compiler. This will
>> >> > fail
>> >> > + // horribly where the relocations *do* need to be applied,
>> >> > but
>> >> > that was
>> >> > + // already the case.
>> >> > if (MachO->isRelocationScattered(RE)) {
>> >> > if (RelType == MachO::GENERIC_RELOC_SECTDIFF ||
>> >> > RelType == MachO::GENERIC_RELOC_LOCAL_SECTDIFF)
>> >> > return processSECTDIFFRelocation(SectionID, RelI, Obj,
>> >> > ObjSectionToID);
>> >> > -
>> >> > - return ++RelI;
>> >> > + else if (Arch == Triple::x86 && RelType ==
>> >> > MachO::GENERIC_RELOC_VANILLA)
>> >> > + return processI386ScatteredVANILLA(SectionID, RelI, Obj,
>> >> > ObjSectionToID);
>> >> > + else
>> >> > + return ++RelI;
>> >> > }
>> >> >
>> >> > RelocationValueRef Value;
>> >> >
>> >> > Modified:
>> >> > llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h
>> >> > URL:
>> >> >
>> >> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h?rev=209478&r1=209477&r2=209478&view=diff
>> >> >
>> >> >
>> >> > ==============================================================================
>> >> > --- llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h
>> >> > (original)
>> >> > +++ llvm/trunk/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h Thu
>> >> > May 22 17:30:13 2014
>> >> > @@ -71,6 +71,12 @@ private:
>> >> > ObjectImage &ObjImg,
>> >> > ObjSectionToIDMap
>> >> > &ObjSectionToID);
>> >> >
>> >> > + relocation_iterator processI386ScatteredVANILLA(
>> >> > + unsigned SectionID,
>> >> > + relocation_iterator
>> >> > RelI,
>> >> > + ObjectImage &ObjImg,
>> >> > + ObjSectionToIDMap
>> >> > &ObjSectionToID);
>> >> > +
>> >> > struct EHFrameRelatedSections {
>> >> > EHFrameRelatedSections()
>> >> > : EHFrameSID(RTDYLD_INVALID_SECTION_ID),
>> >> >
>> >> >
>> >> > _______________________________________________
>> >> > llvm-commits mailing list
>> >> > llvm-commits at cs.uiuc.edu
>> >> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> >
>> >
>
>
More information about the llvm-commits
mailing list