Support R_ARM_REL32 and R_ARM_GOT_PREL in RuntimeDyldELF (SR-387)

[Lang Hames via llvm-commits]

Hi William,

Unfortunately I had to revert this patch in r279275 as it broke some unit
tests on ARM Linux. Do you see any unit test failures like these in your
build:
http://lab.llvm.org:8011/builders/clang-cmake-armv7-a15-selfhost/builds/7600
?

- Lang.

On Thu, Aug 18, 2016 at 6:53 PM, Lang Hames <lhames at gmail.com> wrote:

> No worries. :)
>
> Thanks again for the patch!
>
> - Lang.
>
> On Thu, Aug 18, 2016 at 6:29 PM, William Dillon <william at housedillon.com>
> wrote:
>
>> Hi Lang,
>>
>> That's great, thank you so much.  I'm glad you took a look because most
>> of what you just said went over my head. :)
>>
>> Cheers,
>> - Will
>>
>> On Aug 18, 2016, at 6:25 PM, Lang Hames <lhames at gmail.com> wrote:
>>
>> Hi William,
>>
>> Daniel told me you were still hoping to land this, so I took a stab at
>> writing a test for it and discovered that, unfortunately, RuntimeDyldELF's
>> GOT building mechanism (which uses a separate section for GOT entries)
>> isn't compatible with RuntimeDyldChecker.
>>
>> The proper solution to that is to fix RuntimeDyldELF's GOT handling (it's
>> not safe to use a separate section in general, as it may be allocated out
>> of range of a PC-rel GOT entry load), but that's non-trivial.
>>
>> In the mean time, since this patch doesn't change any existing behavior
>> I've committed in r279182.
>>
>> Cheers,
>> Lang.
>>
>>
>> On Tue, Mar 22, 2016 at 1:05 PM, Lang Hames via llvm-commits <
>> llvm-commits at lists.llvm.org> wrote:
>>
>>> Hi Michael, William,
>>>
>>> William - thank you very much for this!
>>>
>>> If possible, could you please add an ELF test case to
>>> llvm/test/ExecutionEngine/RuntimeDyld/ARM? There is an existing MachO
>>> test case there that may be of some help, though it does not contain any
>>> GOT relocation tests. From the x86-64 GOT test:
>>>
>>> # Test PC-rel GOT relocation.
>>> # Verify both the contents of the GOT entry for y, and that the movq
>>> instruction
>>> # references the correct GOT entry address:
>>> # rtdyld-check: *{8}(stub_addr(test_x86-64.o, __text, y)) = y
>>> # rtdyld-check: decode_operand(insn3, 4) = stub_addr(test_x86-64.o,
>>> __text, y) - next_pc(insn3)
>>> insn3:
>>>         movq    y at GOTPCREL(%rip), %rax
>>>
>>> The first test is reading the value of the stub (in this case GOT entry)
>>> for y and verifying that it contains the address of y.
>>> The second test is verifying that the PC-relative offset for the
>>> instruction is equal to the difference between the stub address and the
>>> next instruction.
>>>
>>> Cheers,
>>> Lang.
>>>
>>> On Mar 22, 2016, at 12:58 PM, Michael Gottesman <mgottesman at apple.com>
>>> wrote:
>>>
>>> I am not actually reviewing this, but just as a drive by: This is
>>> definitely going to need tests.
>>>
>>> On Mar 22, 2016, at 12:56 PM, Michael Gottesman via llvm-commits <
>>> llvm-commits at lists.llvm.org> wrote:
>>>
>>> Lang, can you take a look at this one?
>>>
>>> On Mar 22, 2016, at 10:59 AM, William Dillon via llvm-commits <
>>> llvm-commits at lists.llvm.org> wrote:
>>>
>>> <llvm.patch>This change supports the swift interpreter on armv7-based
>>> linux machines.  This is required for a number of tests and the package
>>> manager.
>>>
>>> - Will_______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at lists.llvm.org
>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at lists.llvm.org
>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at lists.llvm.org
>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>>
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160819/1ebb7dc3/attachment.html>