Re: [Qemu-devel] [RISU PATCH 0/5] Fix RISU build for i386

[Jan Bobek]

Sorry for the delayed reply, the U.S. tax deadline has caught up with
me, so I spent the last two evenings doing my taxes. (Yuck!)

Anyway...

On 4/8/19 6:18 PM, Richard Henderson wrote:
> On 4/8/19 8:27 AM, Jan Bobek wrote:
>> 2. Note the '-std=c99' switch in the command-line above; without it,
>>    GCC defines the symbol 'i386' to 1 and the preprocessor magic for
>>    including arch-specific headers in risu.h breaks. Does anyone have
>>    an idea how to fix this in a more robust way?
> 
> Adding -U$(ARCH) to the command line is probably as good a fix as any.

I didn't know about -U, nice!

>> 3. gas (the GNU assembler) chokes on the syntax of test_i386.s; that's
>>    why I'm using nasm as the assembler above. Is that intentional? I
>>    haven't found the nasm dependency mentioned anywhere.
> 
> I think rewriting to not require nasm is better.

Agreed.

>>    Also, nasm will happily emit the UD1 opcode (0F B9) with no
>>    operands (see test_i386.s). That's a bit surprising to me, since
>>    Intel's Software Developer's Manual says UD1 has two operands; I'd
>>    expect at least a follow-up ModR/M byte. gas refuses to assemble
>>    UD1 with no operands, and gdb's disassembler gets confused when I
>>    load up the nasm's binary into risu. Is there something obvious
>>    that I'm missing?
> 
> You are not missing anything -- ud1 should require a modrm byte.
> 
> My suggestion is to use only UD1 as the "break" insn, with the different OP_*
> codes encoded into the modrm byte.

I had to laugh when I read this; this is *exactly* what I had in mind,
but then I found out there was no ModR/M byte.

>> P.S. This is my first time using git send-email, so please bear with
>>      me if something goes wrong and/or let me know how I can improve
>>      my future submissions. Thank you!
> 
> You've done well with git send-email.  ;-)

Thanks a lot! :)

-Jan

signature.asc
Description: OpenPGP digital signature