[Bug target/99530] New: [i386] 'P' inline assembly operand modifier should obey -fno-plt

[Bug target/99530] New: [i386] 'P' inline assembly operand modifier should obey -fno-plt

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

            Bug ID: 99530
           Summary: [i386] 'P' inline assembly operand modifier should
                    obey -fno-plt
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: thiago at kde dot org
  Target Milestone: ---

$ cat test.c
extern void f(void); 
void g() { asm("call %P0 ; asm" : : "X" (f)); } 
int h() { f(); return 0; }

The normal modes work:

$ gcc -fno-pic -S -o - -O2 test.c| grep call              
        call f ; asm
        call    f
$ gcc -fPIC -S -o - -O2 test.c| grep call              
        call f@PLT ; asm
        call    f@PLT

But the -fno-plt modes still generate PLT references:

$ gcc -fno-pic -fno-plt -S -o - -O2 test.c| grep call              
        call f ; asm
        call    *f@GOTPCREL(%rip)
$ gcc -fPIC -fno-plt -S -o - -O2 test.c| grep call              
        call f@PLT ; asm
        call    *f@GOTPCREL(%rip)

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #1 from Thiago Macieira <thiago at kde dot org> ---
Matching Clang/LLVM bug report:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2021-03-11
     Ever confirmed|0                           |1
           Assignee|unassigned at gcc dot gnu.org      |hjl.tools at gmail dot com
                 CC|                            |hjl.tools at gmail dot com

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> ---
It is OK to use indirect branch via GOT in 64-bit. But it isn't OK for 32-bit
since PIC register may not be available at call site.

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> ---
Created attachment 50366
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50366&action=edit
A patch

I am testing this.

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #4 from Thiago Macieira <thiago at kde dot org> ---
(In reply to H.J. Lu from comment #3)
> Created attachment 50366 [details]
> A patch
> 
> I am testing this.

The same string pattern for loading from the GOT exists elsewhere in the file.
Maybe this is an opportunity to merge them into a common function?

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #5 from Thiago Macieira <thiago at kde dot org> ---
(In reply to Thiago Macieira from comment #1)
> Matching Clang/LLVM bug report:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

Bad copy & paste. That's https://bugs.llvm.org/show_bug.cgi?id=49525

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #6 from Thiago Macieira <thiago at kde dot org> ---
(In reply to H.J. Lu from comment #3)
> Created attachment 50366 [details]
> A patch
> 
> I am testing this.

This has now changed to:

$ ~/dev/gcc/bin/gcc -fPIC -fno-plt -S -o - -O2 test.c | grep call
        call    f@GOTPCREL(%rip)
        call    *f@GOTPCREL(%rip)

The * is missing for indirect calls in AT&T syntax. Intel syntax looks correct:

$ ~/dev/gcc/bin/gcc -masm=intel -fPIC -fno-plt -S -o - -O2 test.c | grep call
        call    [QWORD PTR f@GOTPCREL[rip]]
        call    [QWORD PTR f@GOTPCREL[rip]]

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #50366|0                           |1
        is obsolete|                            |

--- Comment #7 from H.J. Lu <hjl.tools at gmail dot com> ---
Created attachment 50369
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50369&action=edit
The v2 patch

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #8 from Thiago Macieira <thiago at kde dot org> ---
(In reply to H.J. Lu from comment #7)
> Created attachment 50369 [details]
> The v2 patch

Code generation with "call %P0" is now identical to what GCC generates on its
own.

Even the alignment is now right, somehow, though it also added a newline?

$ ~/dev/gcc/bin/gcc -fPIC -fno-plt -S -o - -O2 -xc - <<<'extern void f(void);
void g() { asm("cmp %P0,$0" : : "X" (f)); }' | grep -A1 GOTPC                   
        cmp     *f@GOTPCREL(%rip)
,$0

That's probably not right.

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #9 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Thiago Macieira from comment #8)
> (In reply to H.J. Lu from comment #7)
> > Created attachment 50369 [details]
> > The v2 patch
> 
> Code generation with "call %P0" is now identical to what GCC generates on
> its own.
> 
> Even the alignment is now right, somehow, though it also added a newline?
> 
> $ ~/dev/gcc/bin/gcc -fPIC -fno-plt -S -o - -O2 -xc - <<<'extern void
> f(void); void g() { asm("cmp %P0,$0" : : "X" (f)); }' | grep -A1 GOTPC      
> 
>         cmp     *f@GOTPCREL(%rip)
> ,$0
> 
> That's probably not right.

Don't use %P with cmp.

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #10 from Thiago Macieira <thiago at kde dot org> ---
(In reply to H.J. Lu from comment #9)
> Don't use %P with cmp.

I know, but that's besides the point. I was merely trying to find something
that would have a reason to type more after the %P and show that a newline
could be wrong.

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|                            |https://gcc.gnu.org/piperma
                   |                            |il/gcc-patches/2021-March/5
                   |                            |66651.html

--- Comment #11 from H.J. Lu <hjl.tools at gmail dot com> ---
A patch is posted at

https://gcc.gnu.org/pipermail/gcc-patches/2021-March/566651.html

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #12 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Thiago Macieira from comment #10)
> (In reply to H.J. Lu from comment #9)
> > Don't use %P with cmp.
> 
> I know, but that's besides the point. I was merely trying to find something
> that would have a reason to type more after the %P and show that a newline
> could be wrong.

Since my patch uses output_asm_insn to finish the instruction, %P must be
the last operand.

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

--- Comment #13 from Thiago Macieira <thiago at kde dot org> ---
> Since my patch uses output_asm_insn to finish the instruction, %P must be
> the last operand.

Okay.

If I wanted to abuse, I could just swap it around:

$ ~/dev/gcc/bin/gcc -fPIC -fno-plt -S -o - -O2 -xc - <<<'extern void f(void);
void g() { asm("cmp%z0 $0, %P0" : : "X" (f)); }' | grep GOTPC
        cmpq $0,        *f@GOTPCREL(%rip)

I can't see the need to do that.

The reason I needed to "call %P0" was to call an assembly function that did not
respect the ABI (intel-ipsec-mb's sha512_x8_avx512). The code looks like this:

#ifdef _WIN32
#  define EXTRA_CLOBBER "rsi", "rdi",
#  define OUT0          "+c" /* rcx */
#  define OUT1          "+d" /* rdx */
#else
#  define EXTRA_CLOBBER "rcx", "rdx",
#  define OUT0          "+D" /* rdi */
#  define OUT1          "+S" /* rsi */
#endif
    extern void sha512_x8_avx512(void *data, size_t size_in_blocks);
    __asm__ volatile ("call %P[func]"
            : OUT0 (data),
              OUT1 (size_in_blocks),
              "+m" (*(char (*)[])data)
            : [func] "X" (sha512_x8_avx512)
            :
    // caller-save registers
              "rax", "r8", "r9", "r10", "r11",
              EXTRA_CLOBBER
    // the current implementation does not use RBX and RBP
              // "rbx", "rbp",
    // the current implementation does clobber these callee-save registers
              "r12", "r13", "r14", "r15"
    );

https://github.com/intel/intel-ipsec-mb/blob/master/lib/avx512/sha512_x8_avx512.asm#L417

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #50369|0                           |1
        is obsolete|                            |

--- Comment #14 from H.J. Lu <hjl.tools at gmail dot com> ---
Created attachment 50381
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50381&action=edit
The v3 patch

[Bug target/99530] [i386] 'P' inline assembly operand modifier should obey -fno-plt

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99530

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
   Target Milestone|---                         |11.0
             Status|NEW                         |RESOLVED

--- Comment #15 from H.J. Lu <hjl.tools at gmail dot com> ---
It is fixed by

The master branch has been updated by H.J. Lu <hjl@gcc.gnu.org>:

https://gcc.gnu.org/g:adf14bdbc10d4114865a08cf20020a2616039057

commit r11-7701-gadf14bdbc10d4114865a08cf20020a2616039057
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Thu Mar 11 06:48:24 2021 -0800

    x86: Update 'P' operand modifier for -fno-plt

    Update 'P' operand modifier for -fno-plt to support inline assembly
    statements.  In 64-bit, we can always load function address with
    @GOTPCREL.  In 32-bit, we load function address with @GOT only for
    non-PIC since PIC register may not be available at call site.

    gcc/

            PR target/99504
            * config/i386/i386.c (ix86_force_load_from_GOT_p): Support
            inline assembly statements.
            (ix86_print_operand): Update 'P' handling for -fno-plt.

    gcc/testsuite/

            PR target/99504
            * gcc.target/i386/pr99530-1.c: New test.
            * gcc.target/i386/pr99530-2.c: Likewise.
            * gcc.target/i386/pr99530-3.c: Likewise.
            * gcc.target/i386/pr99530-4.c: Likewise.
            * gcc.target/i386/pr99530-5.c: Likewise.
            * gcc.target/i386/pr99530-6.c: Likewise.