Re: [Nasm-devel] NASM 2.12rc3 - FINAL RELEASE CANDIDATE

[H. P. A. <hp...@zy...>]

On 02/24/16 05:30, Ismail Donmez wrote:
> H. Peter Anvin <hpa <at> zytor.com> writes:
>
>>
>> I have just tagged NASM 2.12rc3.  I am hoping this is the final rc
>> before a 2.12 release.
>>
>> If you can, please test it out as I would like to make a final 2.12
>> release as soon as possible.
>
> This might be too late to report this (I originally reported at
> forum.nasm.us but with no reply) but with nams.git and 2.12rc7 all tests
> fail for me on openSUSE Tumbleweed x86-64, here is one example:
>
> ? gdb --args ../nasm -fbin -ovmread.bin vmread.asm
> Reading symbols from ../nasm...done.
> (gdb) r
> Starting program: /havana/t/n/nasm/nasm -fbin -ovmread.bin vmread.asm
>
> Program received signal SIGSEGV, Segmentation fault.
> __memcpy_sse2_unaligned () at ../sysdeps/x86_64/multiarch/memcpy-sse2-
> unaligned.S:36
> 36      ../sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S: No such file
> or directory.
> (gdb) bt
> #0  __memcpy_sse2_unaligned () at ../sysdeps/x86_64/multiarch/memcpy-sse2-
> unaligned.S:36
> #1  0x0000000000407915 in saa_wbytes (s=0x7a5db0, data=0x7ffd96f364d0,
> len=18446744073709551612) at saa.c:132
> #2  0x0000000000416b30 in bin_out (segto=2, data=0x7ffd96f364d0,
> type=OUT_RAWDATA, size=18446744073709551612, segment=-1, wrt=-1) at
> output/outbin.c:797
> #3  0x000000000040a8f6 in out (offset=32, segto=2, data=0x7ffd96f364d0,
> type=OUT_RAWDATA, size=18446744073709551612, segment=-1, wrt=-1) at
> assemble.c:372
> #4  0x000000000040e4fc in gencode (segment=2, offset=32, bits=64,
> ins=0x7ffd96f367a0, temp=0x4bd9f0 <instrux_VMREAD+80>, insn_end=36) at
> assemble.c:1945
> #5  0x000000000040b436 in assemble (segment=2, offset=28, bits=64, cp=...,
> instruction=0x7ffd96f367a0, output=0x51e660 <of_bin>, error=0x405fe6
> <nasm_error>, listgen=0x51f7c0 <nasmlist>) at assemble.c:692
> #6  0x00000000004054ad in assemble_file (fname=0x51f8c0 <inname>
> "vmread.asm", depend_ptr=0x0) at nasm.c:1761
> #7  0x00000000004026c4 in main (argc=4, argv=0x7ffd96f36b48) at nasm.c:466
>
> Tried with gcc 5.3.1, gcc-6 trunk and clang 3.8. All gives the same
> problem.
>
> Any help is appreciated.
>

I cannot reproduce this with gcc 5.3.1 or clang 3.5 on Fedora.  It would 
be highly useful if you could:

a) configure with --enable-werror
b) make sure you do "make clean" between trying compilers
c) try compiling with -O0
d) try changing uint64_t on line 311 of assemble.c to int64_t

The most logical explanation is that the compiler miscompiles 
abs((int)size) where size is uint64_t.  However, this is certainly not 
true for all the three compilers listed.

If we can narrow this down to a specific compiler it might be worthwhile 
to make a bug report.

	-hpa