The add/double shortcut in ecp_nistz256-x86_64.pl left one instruction
point that did not unwind, and the "slow" path in AES_cbc_encrypt was
not annotated correctly. For the latter, add
.cfi_{remember,restore}_state support to perlasm.
Next, fill in a bunch of functions that are missing no-op .cfi_startproc
and .cfi_endproc blocks. libunwind cannot unwind those stack frames
otherwise.
Finally, work around a bug in libunwind by not encoding rflags. (rflags
isn't a callee-saved register, so there's not much need to annotate it
anyway.)
These were found as part of ABI testing work in BoringSSL.
Reviewed-by: Richard Levitte <levitte@openssl.org>
GH: #8109
(cherry picked from commit c0e8e5007b)
Around 138 distinct errors found and fixed; thanks!
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/3459)
Perl, multiple versions, for some reason occasionally takes issue with
letter b[?] in ox([0-9a-f]+) regex. As result some constants, such as
0xb1 came out wrong when generating code for MASM. Fixes GH#3241.
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/3385)
CFI directives annotate instructions that are significant for stack
unwinding procedure. In addition to directives recognized by GNU
assembler this module implements three synthetic ones:
- .cfi_push annotates push instructions in prologue and translates to
.cfi_adjust_cfa_offset (if needed) and .cfi_offset;
- .cfi_pop annotates pop instructions in epilogue and translates to
.cfi_adjust_cfs_offset (if needed) and .cfi_restore;
- .cfi_cfa_expression encodes DW_CFA_def_cfa_expression and passes it
to .cfi_escape as byte vector;
CFA expression syntax is made up mix of DWARF operator suffixes [subset
of] and references to registers with optional bias. Following example
describes offloaded original stack pointer at specific offset from
current stack pointer:
.cfi_cfa_expression %rsp+40,deref,+8
Final +8 has everything to do with the fact that CFA, Canonical Frame
Address, is reference to top of caller's stack, and on x86_64 call to
subroutine pushes 8-byte return address.
Triggered by request from Adam Langley.
Reviewed-by: Rich Salz <rsalz@openssl.org>
$1<<32>>32 worked fine with either 32- or 64-bit perl for a good while,
relying on quirk that [pure] 32-bit perl performed it as $1<<0>>0. But
this apparently changed in some version past minimally required 5.10,
and operation result became 0. Yet, it went unnoticed for another while,
because most perl package providers configure their packages with
-Duse64bitint option.
Reviewed-by: Rich Salz <rsalz@openssl.org>
The prevailing style seems to not have trailing whitespace, but a few
lines do. This is mostly in the perlasm files, but a few C files got
them after the reformat. This is the result of:
find . -name '*.pl' | xargs sed -E -i '' -e 's/( |'$'\t'')*$//'
find . -name '*.c' | xargs sed -E -i '' -e 's/( |'$'\t'')*$//'
find . -name '*.h' | xargs sed -E -i '' -e 's/( |'$'\t'')*$//'
Then bn_prime.h was excluded since this is a generated file.
Note mkerr.pl has some changes in a heredoc for some help output, but
other lines there lack trailing whitespace too.
Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
use strict would have caught a number of historical bugs in the perlasm
code, some in the repository and some found during review. It even found
a fresh masm-only bug (see below).
This required some tweaks. The "single instance is enough" globals got
switched to proper blessed objects rather than relying on symbolic refs.
A few types need $opcode passed in as a result.
The $$line thing is a little bit of a nuisance. There may be a clearer
pattern to use instead.
This even a bug in the masm code.
9b634c9b37 added logic to make labels
global or function-global based on whether something starts with a $,
seemingly intended to capture the $decor setting of '$L$'. However, it
references $ret which is not defined in label::out. label::out is always
called after label::re, so $ret was always the label itself, so the line
always ran.
I've removed the regular expression so as not to change the behavior of
the script. A number of the assembly files now routinely jump across
functions, so this seems to be the desired behavior now.
GH#1165
Signed-off-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
Add copyright to most .pl files
This does NOT cover any .pl file that has other copyright in it.
Most of those are Andy's but some are public domain.
Fix typo's in some existing files.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Not all assemblers of "gas" flavour handle binary constants, e.g.
seasoned MacOS Xcode doesn't, so give them a hand.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Once upon a time, there was chop, which somply chopped off the last
character of $_ or a given variable, and it was used to take off the
EOL character (\n) of strings.
... but then, you had to check for the presence of such character.
So came chomp, the better chop which checks for \n before chopping it
off. And this worked well, as long as Perl made internally sure that
all EOLs were converted to \n.
These days, though, there seems to be a mixture of perls, so lines
from files in the "wrong" environment might have \r\n as EOL, or just
\r (Mac OS, unless I'm misinformed).
So it's time we went for the more generic variant and use s|\R$||, the
better chomp which recognises all kinds of known EOLs and chops them
off.
A few chops were left alone, as they are use as surgical tools to
remove one last slash or one last comma.
NOTE: \R came with perl 5.10.0. It means that from now on, our
scripts will fail with any older version.
Reviewed-by: Rich Salz <rsalz@openssl.org>
