[PATCH 2/2] x86: make sure all PUSH/POP honor DefaultSize
Jan Beulich
jbeulich@suse.com
Mon Nov 25 07:29:00 GMT 2019
More information about the Binutils mailing list
Mon Nov 25 07:29:00 GMT 2019
- Previous message (by thread): [PATCH 1/2] x86: drop some stray/bogus DefaultSize
- Next message (by thread): [PATCH v2] x86: SYSENTER/SYSEXIT are unavailable in 64-bit mode on AMD
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
While segment registers are registers, their use doesn't allow sizing of insns without suffix / explicit operand size specifier. Prevent PUSH and POP of segment registers from entering that path, instead allowing them to observe the stackop_size setting just like other PUSH/POP and alike do. gas/ 2019-11-XX Jan Beulich <jbeulich@suse.com> * config/tc-i386.c (process_suffix): Arrange for insns with a single non-GPR register operand to not have its suffix guessed from GPR operands. Extend DefaultSize handling to cover PUSH/POP of segment registers. * testsuite/gas/i386/general.s: Add PUSH/POP sreg to .code16gcc set of insns. * testsuite/gas/i386/general.l: Adjust expectations. --- a/gas/config/tc-i386.c +++ b/gas/config/tc-i386.c @@ -6207,10 +6207,11 @@ process_suffix (void) i.suffix = LONG_MNEM_SUFFIX; else if (i.tm.opcode_modifier.size == SIZE64) i.suffix = QWORD_MNEM_SUFFIX; - else if (i.reg_operands) + else if (i.reg_operands + && (i.operands > 1 || i.types[0].bitfield.class == Reg)) { /* If there's no instruction mnemonic suffix we try to invent one - based on register operands. */ + based on GPR operands. */ if (!i.suffix) { /* We take i.suffix from the last register operand specified, @@ -6315,19 +6316,24 @@ process_suffix (void) /* exclude sysret */ && i.tm.base_opcode != 0x0f07) { - if (stackop_size == LONG_MNEM_SUFFIX - && i.tm.base_opcode == 0xcf) + i.suffix = stackop_size; + if (stackop_size == LONG_MNEM_SUFFIX) { /* stackop_size is set to LONG_MNEM_SUFFIX for the .code16gcc directive to support 16-bit mode with 32-bit address. For IRET without a suffix, generate 16-bit IRET (opcode 0xcf) to return from an interrupt handler. */ - i.suffix = WORD_MNEM_SUFFIX; - as_warn (_("generating 16-bit `iret' for .code16gcc directive")); + if (i.tm.base_opcode == 0xcf) + { + i.suffix = WORD_MNEM_SUFFIX; + as_warn (_("generating 16-bit `iret' for .code16gcc directive")); + } + /* Warn about changed behavior for segment register push/pop. */ + else if ((i.tm.base_opcode | 1) == 0x07) + as_warn (_("generating 32-bit `%s', unlike earlier gas versions"), + i.tm.name); } - else - i.suffix = stackop_size; } else if (intel_syntax && !i.suffix --- a/gas/testsuite/gas/i386/general.l +++ b/gas/testsuite/gas/i386/general.l @@ -32,6 +32,10 @@ .*:144: Warning:.* .*:178: Warning:.* .*:224: Warning:.* +.*:233: Warning:.* +.*:234: Warning:.* +.*:238: Warning:.* +.*:239: Warning:.* 1 .psize 0 2 .text 3 #test jumps and calls @@ -298,9 +302,17 @@ [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+666A00[ ]+push \$0 [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+66683412 0000[ ]+push \$0x1234 [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+66FF37[ ]+push \(%bx\) +[ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+6606[ ]+push %es +.*Warning:.*32-bit.*push.* +[ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+660FA0[ ]+push %fs +.*Warning:.*32-bit.*push.* [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+6660[ ]+pusha [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+669C[ ]+pushf [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+668F07[ ]+pop \(%bx\) +[ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+6607[ ]+pop %es +.*Warning:.*32-bit.*pop.* +[ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+660FA1[ ]+pop %fs +.*Warning:.*32-bit.*pop.* [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+6661[ ]+popa [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+669D[ ]+popf [ ]*[1-9][0-9]*[ ]+[0-9a-f]*[ ]+66C3[ ]+ret --- a/gas/testsuite/gas/i386/general.s +++ b/gas/testsuite/gas/i386/general.s @@ -230,9 +230,13 @@ push $0 push $0x1234 push (%bx) + push %es + push %fs pusha pushf pop (%bx) + pop %es + pop %fs popa popf ret
- Previous message (by thread): [PATCH 1/2] x86: drop some stray/bogus DefaultSize
- Next message (by thread): [PATCH v2] x86: SYSENTER/SYSEXIT are unavailable in 64-bit mode on AMD
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Binutils mailing list