tree e75c4536ab7a13bc55430d677710c0961ee12d6c
parent 03d1b7c544851d9f44df1e9ff21839742e08c819
author David Benjamin <davidben@google.com> 1712206259 -0400
committer Boringssl LUCI CQ <boringssl-scoped@luci-project-accounts.iam.gserviceaccount.com> 1715415043 +0000

Align perlasm SEH directives with gas/clang-assembler

perlasm broadly uses gas syntax. gas and clang-assembler already have
SEH directives. From what I can tell, no one ever properly documented
this, but this mail describes this. LLVM's test data also has examples.
https://sourceware.org/legacy-ml/binutils/2009-08/msg00193.html

First, we named ours based on the MASM directives and prepended ".seh_".
gas says "endprologue" instead of "endprolog", "savexmm" instead of
"savexmm128", and "stackalloc" instead of "allocstack".  Since perlasm
mostly looks like gas, I've switched to the gas spellings.

Second, we made .seh_endprologue implicit because it's always
immediately after the last directive. Both MASM and clang-assembler make
it explicit. Synthesizing an .seh_endprologue for those syntaxes would
require buffering the up the whole function, so just require it be
explicit in the source.

The last difference is that gas says ".seh_proc name_of_function". I've
not aligned on that one because MASM actually integrates it into the
PROC line. You add the FRAME keyword or not depending on whether it's a
frame function. To make the MASM output easier, I think we need to
diverge from both gas and what we've currently done. I'll resolve that
in a follow-up change.

Along the way, fix a couple of instances where the _CET_ENDBR got put on
the wrong side of the SEH directive. I don't think that macro works on
Windows anyway, so it's moot. But if it did emit anything, it should be
included in the prologue.

Bug: 571
Change-Id: I39701a952a654afe6bfc8b3b908ca8fe65d6f1a1
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/68292
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Bob Beck <bbe@google.com>