[RFC PATCH 0/8] [x86] Fix 16-bit addressing modes (bug 18220) and attempt to implement .code16 (bug 8684)

[David Woodhouse]

There follows a series of patches which attempts $SUBJECT.

It could do with careful review from someone who's spent more than two
days looking at LLVM code. In particular, my implementation of
reloc_signed_2byte is almost certainly not correct in some details, and
is only tested for ELF output. And was the code in EmitImmediate()
already doing the wrong thing and turning FK_Data_8 into FK_SecRel_4? I
ignored that and simply did 'Size == 2 ? FK_SecRel_2 : FK_SecRel_4',
preserving the original 'bug' if it is a bug.

I would like feedback on the introduction of OpSize16 before I actually
complete the final patch in the series, which is only a rough pass so
far and covers the instructions which were needed by my test cases.

I should probably pull out the addition of 'popaw' and 'pushaw'
instructions (and the renaming of popal and pushal) from the final patch
into a separate commit... or just draw explicit attention to it here.

This implementation of .code16 isn't perfect — we do need to explicitly
encode the word size on some instructions, such as retw because a simple
'ret' will alias to 'retl' by default. But that can possibly be
addressed in subsequent patches, and it's certainly good enough for my
purposes as it is.

-- 
David Woodhouse                            Open Source Technology Centre
David.Woodhouse at intel.com                              Intel Corporation
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 5745 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131213/3c43ff31/attachment.bin>