Darwin: Fix i686 bootstrap when the assembler supports GOTOFF in data.

Message ID 7B38D7C6-7639-45F1-8BF1-AB1676EBAD9E@sandoe.co.uk
State New
Headers show
Series
  • Darwin: Fix i686 bootstrap when the assembler supports GOTOFF in data.
Related show

Commit Message

Iain Sandoe March 21, 2020, 2:21 p.m.
Hi,

When we use an assembler that supports " .long XX@GOTOFF", the current
combination of configuration parameters and conditional compilation
(when building an i686-darwin compiler with mdynamic-no-pic) assume that
it's OK to put jump tables in the .const section.

However, when we encounter a weak function with a jump table, this
produces relocations that directly access the weak symbol section from
the .const section - which is deemed illegal by the linker (since that
would mean that the weak symbol could not be replaced).

Arguably, this is a limitation (maybe even a bug) in the linker - but
it seems that we'd have to change the ABI to fix it - since it would
require some annotation (maybe just using a special section for the
jump tables) to tell the linker that this specific circumstance is OK
because the direct access to the weak symbol can only occur from that
symbol itself.

The fix is to force jump tables into the text section for all X86 Darwin
versions (PIC code already had this change).

Comments

Jonathan Wakely via Gcc-patches March 21, 2020, 2:27 p.m. | #1
On Sat, Mar 21, 2020 at 3:21 PM Iain Sandoe <iain@sandoe.co.uk> wrote:
>

> Hi,

>

> When we use an assembler that supports " .long XX@GOTOFF", the current

> combination of configuration parameters and conditional compilation

> (when building an i686-darwin compiler with mdynamic-no-pic) assume that

> it's OK to put jump tables in the .const section.

>

> However, when we encounter a weak function with a jump table, this

> produces relocations that directly access the weak symbol section from

> the .const section - which is deemed illegal by the linker (since that

> would mean that the weak symbol could not be replaced).

>

> Arguably, this is a limitation (maybe even a bug) in the linker - but

> it seems that we'd have to change the ABI to fix it - since it would

> require some annotation (maybe just using a special section for the

> jump tables) to tell the linker that this specific circumstance is OK

> because the direct access to the weak symbol can only occur from that

> symbol itself.

>

> The fix is to force jump tables into the text section for all X86 Darwin

> versions (PIC code already had this change).

>

> ===

>

> The change to i386.h is just a  tidy up, and I can apply the darwin.h portion

> independently of that to fix the problem if you’d prefer no changes at this

> stage.

>

> OK for master (for the  i386.h part of the change)?

> backports to open branches (for the  i386.h part of the change)?


OK everywhere, but please rewrite the condition to equivalent:

(flag_pic && !(TARGET_64BIT || HAVE_AS_GOTOFF_IN_DATA))

Thanks,
Uros.

>

> thanks

> Iain.

>

> gcc/ChangeLog:

>

> 2020-03-21  Iain Sandoe  <iain@sandoe.co.uk>

>

>         * config/i386/darwin.h (JUMP_TABLES_IN_TEXT_SECTION): Remove

>         references to Darwin.

>         * config/i386/i386.h (JUMP_TABLES_IN_TEXT_SECTION): Define this

>         unconditionally and comment on why.

>

> diff --git a/gcc/config/i386/darwin.h b/gcc/config/i386/darwin.h

> index 1b94cb88c98..5faa1f40089 100644

> --- a/gcc/config/i386/darwin.h

> +++ b/gcc/config/i386/darwin.h

> @@ -239,6 +239,16 @@ along with GCC; see the file COPYING3.  If not see

>  #undef TARGET_ASM_OUTPUT_IDENT

>  #define TARGET_ASM_OUTPUT_IDENT default_asm_output_ident_directive

>

> +/* We always want jump tables in the text section:

> +   * for PIC code, we need the subtracted symbol to be defined at

> +     assembly-time.

> +   * for mdynamic-no-pic, we cannot support jump tables in the .const

> +     section for weak functions, this looks to ld64 like direct access

> +     to the weak symbol from an anonymous atom.  */

> +

> +#undef JUMP_TABLES_IN_TEXT_SECTION

> +#define JUMP_TABLES_IN_TEXT_SECTION 1

> +

>  /* Darwin profiling -- call mcount.

>     If we need a stub, then we unconditionally mark it as used.  */

>  #undef FUNCTION_PROFILER

> diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h

> index 98235f75a9e..d4255f0d678 100644

> --- a/gcc/config/i386/i386.h

> +++ b/gcc/config/i386/i386.h

> @@ -2285,11 +2285,10 @@ extern int const svr4_dbx_register_map[FIRST_PSEUDO_REGISTER];

>

>  /* Under some conditions we need jump tables in the text section,

>     because the assembler cannot handle label differences between

> -   sections.  This is the case for x86_64 on Mach-O for example.  */

> +   sections.  */

>

>  #define JUMP_TABLES_IN_TEXT_SECTION \

> -  (flag_pic && ((TARGET_MACHO && TARGET_64BIT) \

> -   || (!TARGET_64BIT && !HAVE_AS_GOTOFF_IN_DATA)))

> +  (flag_pic && (!TARGET_64BIT && !HAVE_AS_GOTOFF_IN_DATA))

>

>  /* Switch to init or fini section via SECTION_OP, emit a call to FUNC,

>     and switch back.  For x86 we do this only to save a few bytes that

>

Patch

===

The change to i386.h is just a  tidy up, and I can apply the darwin.h portion
independently of that to fix the problem if you’d prefer no changes at this
stage.

OK for master (for the  i386.h part of the change)?
backports to open branches (for the  i386.h part of the change)?

thanks
Iain.

gcc/ChangeLog:

2020-03-21  Iain Sandoe  <iain@sandoe.co.uk>

	* config/i386/darwin.h (JUMP_TABLES_IN_TEXT_SECTION): Remove
	references to Darwin.
	* config/i386/i386.h (JUMP_TABLES_IN_TEXT_SECTION): Define this
	unconditionally and comment on why.

diff --git a/gcc/config/i386/darwin.h b/gcc/config/i386/darwin.h
index 1b94cb88c98..5faa1f40089 100644
--- a/gcc/config/i386/darwin.h
+++ b/gcc/config/i386/darwin.h
@@ -239,6 +239,16 @@  along with GCC; see the file COPYING3.  If not see
 #undef TARGET_ASM_OUTPUT_IDENT
 #define TARGET_ASM_OUTPUT_IDENT default_asm_output_ident_directive
 
+/* We always want jump tables in the text section:
+   * for PIC code, we need the subtracted symbol to be defined at
+     assembly-time.
+   * for mdynamic-no-pic, we cannot support jump tables in the .const
+     section for weak functions, this looks to ld64 like direct access
+     to the weak symbol from an anonymous atom.  */
+
+#undef JUMP_TABLES_IN_TEXT_SECTION
+#define JUMP_TABLES_IN_TEXT_SECTION 1
+
 /* Darwin profiling -- call mcount.
    If we need a stub, then we unconditionally mark it as used.  */
 #undef FUNCTION_PROFILER
diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h
index 98235f75a9e..d4255f0d678 100644
--- a/gcc/config/i386/i386.h
+++ b/gcc/config/i386/i386.h
@@ -2285,11 +2285,10 @@  extern int const svr4_dbx_register_map[FIRST_PSEUDO_REGISTER];
 
 /* Under some conditions we need jump tables in the text section,
    because the assembler cannot handle label differences between
-   sections.  This is the case for x86_64 on Mach-O for example.  */
+   sections.  */
 
 #define JUMP_TABLES_IN_TEXT_SECTION \
-  (flag_pic && ((TARGET_MACHO && TARGET_64BIT) \
-   || (!TARGET_64BIT && !HAVE_AS_GOTOFF_IN_DATA)))
+  (flag_pic && (!TARGET_64BIT && !HAVE_AS_GOTOFF_IN_DATA))
 
 /* Switch to init or fini section via SECTION_OP, emit a call to FUNC,
    and switch back.  For x86 we do this only to save a few bytes that