x86: accept whitespace around prefix separator

... and prediction suffix comma. Other than documented /**/ comments
currently aren't really converted to a single space, at least not for
x86 in its most common configurations. That'll be fixed subsequently, at
which point blanks may appear where so far none were expected.
Furthermore not permitting blanks around these separators wasn't quite
logical anyway - such constructs are composite ones, and hence
components ought to have been permitted to be separated by whitespace
from the very beginning. Furthermore note how, due to the scrubber being
overly aggressive in removing whitespace, some similar construct with a
prefix were already accepted.

Note how certain other checks in parse_insn() can be simplified as a
result.

While there for the prediction suffix also make checks case-insensitive
and check for a proper trailing separator.
This commit is contained in:
Jan Beulich 2024-07-26 07:59:53 +02:00
parent 1cd36be7c9
commit c97f0d71ea
4 changed files with 99 additions and 19 deletions

View File

@ -7957,6 +7957,8 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
while (1)
{
const char *split;
mnem_p = mnemonic;
/* Pseudo-prefixes start with an opening figure brace. */
if ((*mnem_p = *l) == '{')
@ -7981,9 +7983,10 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
}
l++;
}
/* Pseudo-prefixes end with a closing figure brace. */
if (*mnemonic == '{' && is_space_char (*l))
split = l;
if (is_space_char (*l))
++l;
/* Pseudo-prefixes end with a closing figure brace. */
if (*mnemonic == '{' && *l == '}')
{
*mnem_p++ = *l++;
@ -7991,12 +7994,10 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
goto too_long;
*mnem_p = '\0';
/* Point l at the closing brace if there's no other separator. */
if (*l != END_OF_INSN && !is_space_char (*l)
&& *l != PREFIX_SEPARATOR)
--l;
if (is_space_char (*l))
++l;
}
else if (!is_space_char (*l)
else if (l == split
&& *l != END_OF_INSN
&& (intel_syntax
|| (*l != PREFIX_SEPARATOR && *l != ',')))
@ -8004,7 +8005,7 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
if (mode != parse_all)
break;
as_bad (_("invalid character %s in mnemonic"),
output_invalid (*l));
output_invalid (*split));
return NULL;
}
if (token_start == l)
@ -8020,7 +8021,6 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
op_lookup (mnemonic);
if (*l != END_OF_INSN
&& (!is_space_char (*l) || l[1] != END_OF_INSN)
&& current_templates.start
&& current_templates.start->opcode_modifier.isprefix)
{
@ -8142,7 +8142,10 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
}
}
/* Skip past PREFIX_SEPARATOR and reset token_start. */
token_start = ++l;
l += (!intel_syntax && *l == PREFIX_SEPARATOR);
if (is_space_char (*l))
++l;
token_start = l;
}
else
break;
@ -8234,8 +8237,7 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
}
/* For compatibility reasons accept MOVSD and CMPSD without
operands even in AT&T mode. */
else if (*l == END_OF_INSN
|| (is_space_char (*l) && l[1] == END_OF_INSN))
else if (*l == END_OF_INSN)
{
mnem_p[-1] = '\0';
op_lookup (mnemonic);
@ -8277,8 +8279,9 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
l += length;
}
if (current_templates.start->opcode_modifier.jump == JUMP
|| current_templates.start->opcode_modifier.jump == JUMP_BYTE)
if ((current_templates.start->opcode_modifier.jump == JUMP
|| current_templates.start->opcode_modifier.jump == JUMP_BYTE)
&& *l == ',')
{
/* Check for a branch hint. We allow ",pt" and ",pn" for
predict taken and predict not taken respectively.
@ -8286,21 +8289,29 @@ parse_insn (const char *line, char *mnemonic, enum parse_mode mode)
and jcxz insns (JumpByte) for current Pentium4 chips. They
may work in the future and it doesn't hurt to accept them
now. */
if (l[0] == ',' && l[1] == 'p')
token_start = l++;
if (is_space_char (*l))
++l;
if (TOLOWER (*l) == 'p' && ISALPHA (l[1])
&& (l[2] == END_OF_INSN || is_space_char (l[2])))
{
if (l[2] == 't')
if (TOLOWER (l[1]) == 't')
{
if (!add_prefix (DS_PREFIX_OPCODE))
return NULL;
l += 3;
l += 2;
}
else if (l[2] == 'n')
else if (TOLOWER (l[1]) == 'n')
{
if (!add_prefix (CS_PREFIX_OPCODE))
return NULL;
l += 3;
l += 2;
}
else
l = token_start;
}
else
l = token_start;
}
/* Any other comma loses. */
if (*l == ',')

View File

@ -106,6 +106,7 @@ if [gas_32_check] then {
run_list_test "equ-2" "-al"
run_list_test "equ-bad"
run_dump_test "curly"
run_dump_test "separator"
run_dump_test "divide"
run_dump_test "quoted"
run_dump_test "quoted2"

View File

@ -0,0 +1,27 @@
#objdump: -dw
#name: whitespace around special separators
.*: +file format .*
Disassembly of section \.text:
0+ <separators>:
[ ]*[a-f0-9]+: 3e 72 fd + j[cb],pt .*
[ ]*[a-f0-9]+: 3e 72 fd + j[cb],pt .*
[ ]*[a-f0-9]+: 3e 72 fd + j[cb],pt .*
[ ]*[a-f0-9]+: 3e 72 fd + j[cb],pt .*
[ ]*[a-f0-9]+: 3e 72 fd + j[cb],pt .*
[ ]*[a-f0-9]+: 3e 0f 82 f9 ff ff ff j[cb],pt .*
[ ]*[a-f0-9]+: 3e 0f 82 f9 ff ff ff j[cb],pt .*
[ ]*[a-f0-9]+: 3e 0f 82 f9 ff ff ff j[cb],pt .*
[ ]*[a-f0-9]+: 3e 0f 82 f9 ff ff ff j[cb],pt .*
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
[ ]*[a-f0-9]+: 65 f7 d8 + gs neg %eax
#pass

View File

@ -0,0 +1,41 @@
.text
separators:
jc,pt .
jc ,pt .
jc, pt .
jc , pt .
jc/**/,/**/pt .
{disp32} jc,pt .
{disp32} jc ,pt .
{disp32} jc, pt .
{disp32} jc , pt .
# Which block to use depends on whether / starts a comment.
.ifeq 1/2
gs/neg %eax
gs /neg %eax
gs/ neg %eax
gs / neg %eax
gs/**///**/neg %eax
{disp32} gs/neg %eax
{disp32} gs /neg %eax
{disp32} gs/ neg %eax
{disp32} gs / neg %eax
.else
gs\neg %eax
gs \neg %eax
gs\ neg %eax
gs \ neg %eax
gs/**/\/**/neg %eax
{disp32} gs\neg %eax
{disp32} gs \neg %eax
{disp32} gs\ neg %eax
{disp32} gs \ neg %eax
.endif