diff mbox series

[ARM/FDPIC,v5,09/21,ARM] FDPIC: Add support for taking address of nested function

Message ID 20190515124006.25840-10-christophe.lyon@st.com
State New
Headers show
Series FDPIC ABI for ARM | expand

Commit Message

Christophe Lyon May 15, 2019, 12:39 p.m. UTC
In FDPIC mode, the trampoline generated to support pointers to nested
functions looks like:

	   .word	trampoline address
	   .word	trampoline GOT address
	   ldr 		r12, [pc, #8]
	   ldr 		r9, [pc, #8]
	   ldr		pc, [pc, #8]
	   .word	static chain value
	   .word	GOT address
	   .word	function's address

because in FDPIC function pointers are actually pointers to function
descriptors, we have to actually generate a function descriptor for
the trampoline.

2019-XX-XX  Christophe Lyon  <christophe.lyon@st.com>
	Mickaël Guêné <mickael.guene@st.com>

	gcc/
	* config/arm/arm.c (arm_asm_trampoline_template): Add FDPIC
	support.
	(arm_trampoline_init): Likewise.
	(arm_trampoline_init): Likewise.
	* config/arm/arm.h (TRAMPOLINE_SIZE): Likewise.

Change-Id: Idc4d5f629ae4f8d79bdf9623517481d524a0c144

Comments

Kyrill Tkachov July 16, 2019, 11:18 a.m. UTC | #1
Hi Christophe

On 5/15/19 1:39 PM, Christophe Lyon wrote:
> In FDPIC mode, the trampoline generated to support pointers to nested
> functions looks like:
>
>            .word trampoline address
>            .word trampoline GOT address
>            ldr            r12, [pc, #8]
>            ldr            r9, [pc, #8]
>            ldr           pc, [pc, #8]
>            .word static chain value
>            .word GOT address
>            .word function's address
>
> because in FDPIC function pointers are actually pointers to function
> descriptors, we have to actually generate a function descriptor for
> the trampoline.
>
> 2019-XX-XX  Christophe Lyon  <christophe.lyon@st.com>
>         Mickaël Guêné <mickael.guene@st.com>
>
>         gcc/
>         * config/arm/arm.c (arm_asm_trampoline_template): Add FDPIC
>         support.
>         (arm_trampoline_init): Likewise.
>         (arm_trampoline_init): Likewise.
>         * config/arm/arm.h (TRAMPOLINE_SIZE): Likewise.
>
> Change-Id: Idc4d5f629ae4f8d79bdf9623517481d524a0c144
>
> diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
> index 40e3f3b..99d13bf 100644
> --- a/gcc/config/arm/arm.c
> +++ b/gcc/config/arm/arm.c
> @@ -3976,13 +3976,50 @@ arm_warn_func_return (tree decl)
>             .word static chain value
>             .word function's address
>     XXX FIXME: When the trampoline returns, r8 will be clobbered.  */
> +/* In FDPIC mode, the trampoline looks like:
> +          .word trampoline address
> +          .word trampoline GOT address
> +          ldr            r12, [pc, #8] ; #4 for Thumb2
> +          ldr            r9,  [pc, #8] ; #4 for Thumb2
> +          ldr           pc,  [pc, #8] ; #4 for Thumb2
> +          .word static chain value
> +          .word GOT address
> +          .word function's address
> +*/


I think this comment is not right for Thumb2.

These load instructionshave 32-bit encodings, even in Thumb2 (they use 
high registers).

Also, please merge this comment with the one above (no separate /**/)

>
>  static void
>  arm_asm_trampoline_template (FILE *f)
>  {
>    fprintf (f, "\t.syntax unified\n");
>
> -  if (TARGET_ARM)
> +  if (TARGET_FDPIC)
> +    {
> +      /* The first two words are a function descriptor pointing to the
> +        trampoline code just below.  */
> +      if (TARGET_ARM)
> +       fprintf (f, "\t.arm\n");
> +      else if (TARGET_THUMB2)
> +       fprintf (f, "\t.thumb\n");
> +      else
> +       /* Only ARM and Thumb-2 are supported.  */
> +       gcc_unreachable ();
> +
> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> +      /* Trampoline code which sets the static chain register but also
> +        PIC register before jumping into real code.  */
> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> +                  STATIC_CHAIN_REGNUM, PC_REGNUM,
> +                  TARGET_THUMB2 ? 8 : 4);
> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> +                  PIC_OFFSET_TABLE_REGNUM, PC_REGNUM,
> +                  TARGET_THUMB2 ? 8 : 4);
> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> +                  PC_REGNUM, PC_REGNUM,
> +                  TARGET_THUMB2 ? 8 : 4);


As above, I think the offset should be 8 for both Arm and Thumb2.

Thanks,

Kyrill


> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> +    }
> +  else if (TARGET_ARM)
>      {
>        fprintf (f, "\t.arm\n");
>        asm_fprintf (f, "\tldr\t%r, [%r, #0]\n", STATIC_CHAIN_REGNUM, 
> PC_REGNUM);
> @@ -4023,12 +4060,40 @@ arm_trampoline_init (rtx m_tramp, tree fndecl, 
> rtx chain_value)
>    emit_block_move (m_tramp, assemble_trampoline_template (),
>                     GEN_INT (TRAMPOLINE_SIZE), BLOCK_OP_NORMAL);
>
> -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
> -  emit_move_insn (mem, chain_value);
> +  if (TARGET_FDPIC)
> +    {
> +      rtx funcdesc = XEXP (DECL_RTL (fndecl), 0);
> +      rtx fnaddr = gen_rtx_MEM (Pmode, funcdesc);
> +      rtx gotaddr = gen_rtx_MEM (Pmode, plus_constant (Pmode, 
> funcdesc, 4));
> +      /* The function start address is at offset 8, but in Thumb mode
> +        we want bit 0 set to 1 to indicate Thumb-ness, hence 9
> +        below.  */
> +      rtx trampoline_code_start
> +       = plus_constant (Pmode, XEXP (m_tramp, 0), TARGET_THUMB2 ? 9 : 8);
> +
> +      /* Write initial funcdesc which points to the trampoline.  */
> +      mem = adjust_address (m_tramp, SImode, 0);
> +      emit_move_insn (mem, trampoline_code_start);
> +      mem = adjust_address (m_tramp, SImode, 4);
> +      emit_move_insn (mem, gen_rtx_REG (Pmode, PIC_OFFSET_TABLE_REGNUM));
> +      /* Setup static chain.  */
> +      mem = adjust_address (m_tramp, SImode, 20);
> +      emit_move_insn (mem, chain_value);
> +      /* GOT + real function entry point.  */
> +      mem = adjust_address (m_tramp, SImode, 24);
> +      emit_move_insn (mem, gotaddr);
> +      mem = adjust_address (m_tramp, SImode, 28);
> +      emit_move_insn (mem, fnaddr);
> +    }
> +  else
> +    {
> +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
> +      emit_move_insn (mem, chain_value);
>
> -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
> -  fnaddr = XEXP (DECL_RTL (fndecl), 0);
> -  emit_move_insn (mem, fnaddr);
> +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
> +      fnaddr = XEXP (DECL_RTL (fndecl), 0);
> +      emit_move_insn (mem, fnaddr);
> +    }
>
>    a_tramp = XEXP (m_tramp, 0);
>    emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__clear_cache"),
> @@ -4042,7 +4107,9 @@ arm_trampoline_init (rtx m_tramp, tree fndecl, 
> rtx chain_value)
>  static rtx
>  arm_trampoline_adjust_address (rtx addr)
>  {
> -  if (TARGET_THUMB)
> +  /* For FDPIC don't fix trampoline address since it's a function
> +     descriptor and not a function address.  */
> +  if (TARGET_THUMB && !TARGET_FDPIC)
>      addr = expand_simple_binop (Pmode, IOR, addr, const1_rtx,
>                                  NULL, 0, OPTAB_LIB_WIDEN);
>    return addr;
> diff --git a/gcc/config/arm/arm.h b/gcc/config/arm/arm.h
> index 45c0e2b..f80df63 100644
> --- a/gcc/config/arm/arm.h
> +++ b/gcc/config/arm/arm.h
> @@ -1602,7 +1602,7 @@ typedef struct
>  #define INIT_EXPANDERS  arm_init_expanders ()
>
>  /* Length in units of the trampoline for entering a nested function.  */
> -#define TRAMPOLINE_SIZE  (TARGET_32BIT ? 16 : 20)
> +#define TRAMPOLINE_SIZE  (TARGET_FDPIC ? 32 : (TARGET_32BIT ? 16 : 20))
>
>  /* Alignment required for a trampoline in bits.  */
>  #define TRAMPOLINE_ALIGNMENT  32
> -- 
> 2.6.3
>
Kyrill Tkachov July 16, 2019, 12:42 p.m. UTC | #2
On 7/16/19 12:18 PM, Kyrill Tkachov wrote:
> Hi Christophe
>
> On 5/15/19 1:39 PM, Christophe Lyon wrote:
> > In FDPIC mode, the trampoline generated to support pointers to nested
> > functions looks like:
> >
> >            .word trampoline address
> >            .word trampoline GOT address
> >            ldr            r12, [pc, #8]
> >            ldr            r9, [pc, #8]
> >            ldr           pc, [pc, #8]
> >            .word static chain value
> >            .word GOT address
> >            .word function's address
> >
> > because in FDPIC function pointers are actually pointers to function
> > descriptors, we have to actually generate a function descriptor for
> > the trampoline.
> >
> > 2019-XX-XX  Christophe Lyon <christophe.lyon@st.com>
> >         Mickaël Guêné <mickael.guene@st.com>
> >
> >         gcc/
> >         * config/arm/arm.c (arm_asm_trampoline_template): Add FDPIC
> >         support.
> >         (arm_trampoline_init): Likewise.
> >         (arm_trampoline_init): Likewise.
> >         * config/arm/arm.h (TRAMPOLINE_SIZE): Likewise.
> >
> > Change-Id: Idc4d5f629ae4f8d79bdf9623517481d524a0c144
> >
> > diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
> > index 40e3f3b..99d13bf 100644
> > --- a/gcc/config/arm/arm.c
> > +++ b/gcc/config/arm/arm.c
> > @@ -3976,13 +3976,50 @@ arm_warn_func_return (tree decl)
> >             .word static chain value
> >             .word function's address
> >     XXX FIXME: When the trampoline returns, r8 will be clobbered.  */
> > +/* In FDPIC mode, the trampoline looks like:
> > +          .word trampoline address
> > +          .word trampoline GOT address
> > +          ldr            r12, [pc, #8] ; #4 for Thumb2
> > +          ldr            r9,  [pc, #8] ; #4 for Thumb2
> > +          ldr           pc,  [pc, #8] ; #4 for Thumb2
> > +          .word static chain value
> > +          .word GOT address
> > +          .word function's address
> > +*/
>
>
> I think this comment is not right for Thumb2.
>
> These load instructionshave 32-bit encodings, even in Thumb2 (they use
> high registers).

Andre and Wilco pointed out to me offline that the offset should be #4 
for Arm mode.

The Arm ARM at E1.2.3 says:

PC, the program counter

* When executing an A32 instruction, PC reads as the address of the 
current instruction plus 8.

* When executing a T32 instruction, PC reads as the address of the 
current instruction plus 4.

Thanks,

Kyrill


>
> Also, please merge this comment with the one above (no separate /**/)
>
> >
> >  static void
> >  arm_asm_trampoline_template (FILE *f)
> >  {
> >    fprintf (f, "\t.syntax unified\n");
> >
> > -  if (TARGET_ARM)
> > +  if (TARGET_FDPIC)
> > +    {
> > +      /* The first two words are a function descriptor pointing to the
> > +        trampoline code just below.  */
> > +      if (TARGET_ARM)
> > +       fprintf (f, "\t.arm\n");
> > +      else if (TARGET_THUMB2)
> > +       fprintf (f, "\t.thumb\n");
> > +      else
> > +       /* Only ARM and Thumb-2 are supported.  */
> > +       gcc_unreachable ();
> > +
> > +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> > +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> > +      /* Trampoline code which sets the static chain register but also
> > +        PIC register before jumping into real code. */
> > +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> > +                  STATIC_CHAIN_REGNUM, PC_REGNUM,
> > +                  TARGET_THUMB2 ? 8 : 4);
> > +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> > +                  PIC_OFFSET_TABLE_REGNUM, PC_REGNUM,
> > +                  TARGET_THUMB2 ? 8 : 4);
> > +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> > +                  PC_REGNUM, PC_REGNUM,
> > +                  TARGET_THUMB2 ? 8 : 4);
>
>
> As above, I think the offset should be 8 for both Arm and Thumb2.
>
> Thanks,
>
> Kyrill
>
>
> > +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> > +    }
> > +  else if (TARGET_ARM)
> >      {
> >        fprintf (f, "\t.arm\n");
> >        asm_fprintf (f, "\tldr\t%r, [%r, #0]\n", STATIC_CHAIN_REGNUM,
> > PC_REGNUM);
> > @@ -4023,12 +4060,40 @@ arm_trampoline_init (rtx m_tramp, tree fndecl,
> > rtx chain_value)
> >    emit_block_move (m_tramp, assemble_trampoline_template (),
> >                     GEN_INT (TRAMPOLINE_SIZE), BLOCK_OP_NORMAL);
> >
> > -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
> > -  emit_move_insn (mem, chain_value);
> > +  if (TARGET_FDPIC)
> > +    {
> > +      rtx funcdesc = XEXP (DECL_RTL (fndecl), 0);
> > +      rtx fnaddr = gen_rtx_MEM (Pmode, funcdesc);
> > +      rtx gotaddr = gen_rtx_MEM (Pmode, plus_constant (Pmode,
> > funcdesc, 4));
> > +      /* The function start address is at offset 8, but in Thumb mode
> > +        we want bit 0 set to 1 to indicate Thumb-ness, hence 9
> > +        below.  */
> > +      rtx trampoline_code_start
> > +       = plus_constant (Pmode, XEXP (m_tramp, 0), TARGET_THUMB2 ? 9 
> : 8);
> > +
> > +      /* Write initial funcdesc which points to the trampoline.  */
> > +      mem = adjust_address (m_tramp, SImode, 0);
> > +      emit_move_insn (mem, trampoline_code_start);
> > +      mem = adjust_address (m_tramp, SImode, 4);
> > +      emit_move_insn (mem, gen_rtx_REG (Pmode, 
> PIC_OFFSET_TABLE_REGNUM));
> > +      /* Setup static chain.  */
> > +      mem = adjust_address (m_tramp, SImode, 20);
> > +      emit_move_insn (mem, chain_value);
> > +      /* GOT + real function entry point.  */
> > +      mem = adjust_address (m_tramp, SImode, 24);
> > +      emit_move_insn (mem, gotaddr);
> > +      mem = adjust_address (m_tramp, SImode, 28);
> > +      emit_move_insn (mem, fnaddr);
> > +    }
> > +  else
> > +    {
> > +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
> > +      emit_move_insn (mem, chain_value);
> >
> > -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
> > -  fnaddr = XEXP (DECL_RTL (fndecl), 0);
> > -  emit_move_insn (mem, fnaddr);
> > +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
> > +      fnaddr = XEXP (DECL_RTL (fndecl), 0);
> > +      emit_move_insn (mem, fnaddr);
> > +    }
> >
> >    a_tramp = XEXP (m_tramp, 0);
> >    emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__clear_cache"),
> > @@ -4042,7 +4107,9 @@ arm_trampoline_init (rtx m_tramp, tree fndecl,
> > rtx chain_value)
> >  static rtx
> >  arm_trampoline_adjust_address (rtx addr)
> >  {
> > -  if (TARGET_THUMB)
> > +  /* For FDPIC don't fix trampoline address since it's a function
> > +     descriptor and not a function address.  */
> > +  if (TARGET_THUMB && !TARGET_FDPIC)
> >      addr = expand_simple_binop (Pmode, IOR, addr, const1_rtx,
> >                                  NULL, 0, OPTAB_LIB_WIDEN);
> >    return addr;
> > diff --git a/gcc/config/arm/arm.h b/gcc/config/arm/arm.h
> > index 45c0e2b..f80df63 100644
> > --- a/gcc/config/arm/arm.h
> > +++ b/gcc/config/arm/arm.h
> > @@ -1602,7 +1602,7 @@ typedef struct
> >  #define INIT_EXPANDERS  arm_init_expanders ()
> >
> >  /* Length in units of the trampoline for entering a nested 
> function.  */
> > -#define TRAMPOLINE_SIZE  (TARGET_32BIT ? 16 : 20)
> > +#define TRAMPOLINE_SIZE  (TARGET_FDPIC ? 32 : (TARGET_32BIT ? 16 : 20))
> >
> >  /* Alignment required for a trampoline in bits.  */
> >  #define TRAMPOLINE_ALIGNMENT  32
> > --
> > 2.6.3
> >
Christophe Lyon July 31, 2019, 2:44 p.m. UTC | #3
On Tue, 16 Jul 2019 at 14:42, Kyrill Tkachov
<kyrylo.tkachov@foss.arm.com> wrote:
>
>
> On 7/16/19 12:18 PM, Kyrill Tkachov wrote:
> > Hi Christophe
> >
> > On 5/15/19 1:39 PM, Christophe Lyon wrote:
> > > In FDPIC mode, the trampoline generated to support pointers to nested
> > > functions looks like:
> > >
> > >            .word trampoline address
> > >            .word trampoline GOT address
> > >            ldr            r12, [pc, #8]
> > >            ldr            r9, [pc, #8]
> > >            ldr           pc, [pc, #8]
> > >            .word static chain value
> > >            .word GOT address
> > >            .word function's address
> > >
> > > because in FDPIC function pointers are actually pointers to function
> > > descriptors, we have to actually generate a function descriptor for
> > > the trampoline.
> > >
> > > 2019-XX-XX  Christophe Lyon <christophe.lyon@st.com>
> > >         Mickaël Guêné <mickael.guene@st.com>
> > >
> > >         gcc/
> > >         * config/arm/arm.c (arm_asm_trampoline_template): Add FDPIC
> > >         support.
> > >         (arm_trampoline_init): Likewise.
> > >         (arm_trampoline_init): Likewise.
> > >         * config/arm/arm.h (TRAMPOLINE_SIZE): Likewise.
> > >
> > > Change-Id: Idc4d5f629ae4f8d79bdf9623517481d524a0c144
> > >
> > > diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
> > > index 40e3f3b..99d13bf 100644
> > > --- a/gcc/config/arm/arm.c
> > > +++ b/gcc/config/arm/arm.c
> > > @@ -3976,13 +3976,50 @@ arm_warn_func_return (tree decl)
> > >             .word static chain value
> > >             .word function's address
> > >     XXX FIXME: When the trampoline returns, r8 will be clobbered.  */
> > > +/* In FDPIC mode, the trampoline looks like:
> > > +          .word trampoline address
> > > +          .word trampoline GOT address
> > > +          ldr            r12, [pc, #8] ; #4 for Thumb2
> > > +          ldr            r9,  [pc, #8] ; #4 for Thumb2
> > > +          ldr           pc,  [pc, #8] ; #4 for Thumb2
> > > +          .word static chain value
> > > +          .word GOT address
> > > +          .word function's address
> > > +*/
> >
> >
> > I think this comment is not right for Thumb2.
> >
> > These load instructionshave 32-bit encodings, even in Thumb2 (they use
> > high registers).
>
> Andre and Wilco pointed out to me offline that the offset should be #4
> for Arm mode.
>
> The Arm ARM at E1.2.3 says:
>
> PC, the program counter
>
> * When executing an A32 instruction, PC reads as the address of the
> current instruction plus 8.
>
> * When executing a T32 instruction, PC reads as the address of the
> current instruction plus 4.
>

Yes, it looks like the code is right, and the comment is wrong:
- offset 8 for thumb2 mode
- offset 4 for arm mode

Thanks,

Christophe

> Thanks,
>
> Kyrill
>
>
> >
> > Also, please merge this comment with the one above (no separate /**/)
> >
> > >
> > >  static void
> > >  arm_asm_trampoline_template (FILE *f)
> > >  {
> > >    fprintf (f, "\t.syntax unified\n");
> > >
> > > -  if (TARGET_ARM)
> > > +  if (TARGET_FDPIC)
> > > +    {
> > > +      /* The first two words are a function descriptor pointing to the
> > > +        trampoline code just below.  */
> > > +      if (TARGET_ARM)
> > > +       fprintf (f, "\t.arm\n");
> > > +      else if (TARGET_THUMB2)
> > > +       fprintf (f, "\t.thumb\n");
> > > +      else
> > > +       /* Only ARM and Thumb-2 are supported.  */
> > > +       gcc_unreachable ();
> > > +
> > > +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> > > +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> > > +      /* Trampoline code which sets the static chain register but also
> > > +        PIC register before jumping into real code. */
> > > +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> > > +                  STATIC_CHAIN_REGNUM, PC_REGNUM,
> > > +                  TARGET_THUMB2 ? 8 : 4);
> > > +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> > > +                  PIC_OFFSET_TABLE_REGNUM, PC_REGNUM,
> > > +                  TARGET_THUMB2 ? 8 : 4);
> > > +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
> > > +                  PC_REGNUM, PC_REGNUM,
> > > +                  TARGET_THUMB2 ? 8 : 4);
> >
> >
> > As above, I think the offset should be 8 for both Arm and Thumb2.
> >
> > Thanks,
> >
> > Kyrill
> >
> >
> > > +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
> > > +    }
> > > +  else if (TARGET_ARM)
> > >      {
> > >        fprintf (f, "\t.arm\n");
> > >        asm_fprintf (f, "\tldr\t%r, [%r, #0]\n", STATIC_CHAIN_REGNUM,
> > > PC_REGNUM);
> > > @@ -4023,12 +4060,40 @@ arm_trampoline_init (rtx m_tramp, tree fndecl,
> > > rtx chain_value)
> > >    emit_block_move (m_tramp, assemble_trampoline_template (),
> > >                     GEN_INT (TRAMPOLINE_SIZE), BLOCK_OP_NORMAL);
> > >
> > > -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
> > > -  emit_move_insn (mem, chain_value);
> > > +  if (TARGET_FDPIC)
> > > +    {
> > > +      rtx funcdesc = XEXP (DECL_RTL (fndecl), 0);
> > > +      rtx fnaddr = gen_rtx_MEM (Pmode, funcdesc);
> > > +      rtx gotaddr = gen_rtx_MEM (Pmode, plus_constant (Pmode,
> > > funcdesc, 4));
> > > +      /* The function start address is at offset 8, but in Thumb mode
> > > +        we want bit 0 set to 1 to indicate Thumb-ness, hence 9
> > > +        below.  */
> > > +      rtx trampoline_code_start
> > > +       = plus_constant (Pmode, XEXP (m_tramp, 0), TARGET_THUMB2 ? 9
> > : 8);
> > > +
> > > +      /* Write initial funcdesc which points to the trampoline.  */
> > > +      mem = adjust_address (m_tramp, SImode, 0);
> > > +      emit_move_insn (mem, trampoline_code_start);
> > > +      mem = adjust_address (m_tramp, SImode, 4);
> > > +      emit_move_insn (mem, gen_rtx_REG (Pmode,
> > PIC_OFFSET_TABLE_REGNUM));
> > > +      /* Setup static chain.  */
> > > +      mem = adjust_address (m_tramp, SImode, 20);
> > > +      emit_move_insn (mem, chain_value);
> > > +      /* GOT + real function entry point.  */
> > > +      mem = adjust_address (m_tramp, SImode, 24);
> > > +      emit_move_insn (mem, gotaddr);
> > > +      mem = adjust_address (m_tramp, SImode, 28);
> > > +      emit_move_insn (mem, fnaddr);
> > > +    }
> > > +  else
> > > +    {
> > > +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
> > > +      emit_move_insn (mem, chain_value);
> > >
> > > -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
> > > -  fnaddr = XEXP (DECL_RTL (fndecl), 0);
> > > -  emit_move_insn (mem, fnaddr);
> > > +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
> > > +      fnaddr = XEXP (DECL_RTL (fndecl), 0);
> > > +      emit_move_insn (mem, fnaddr);
> > > +    }
> > >
> > >    a_tramp = XEXP (m_tramp, 0);
> > >    emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__clear_cache"),
> > > @@ -4042,7 +4107,9 @@ arm_trampoline_init (rtx m_tramp, tree fndecl,
> > > rtx chain_value)
> > >  static rtx
> > >  arm_trampoline_adjust_address (rtx addr)
> > >  {
> > > -  if (TARGET_THUMB)
> > > +  /* For FDPIC don't fix trampoline address since it's a function
> > > +     descriptor and not a function address.  */
> > > +  if (TARGET_THUMB && !TARGET_FDPIC)
> > >      addr = expand_simple_binop (Pmode, IOR, addr, const1_rtx,
> > >                                  NULL, 0, OPTAB_LIB_WIDEN);
> > >    return addr;
> > > diff --git a/gcc/config/arm/arm.h b/gcc/config/arm/arm.h
> > > index 45c0e2b..f80df63 100644
> > > --- a/gcc/config/arm/arm.h
> > > +++ b/gcc/config/arm/arm.h
> > > @@ -1602,7 +1602,7 @@ typedef struct
> > >  #define INIT_EXPANDERS  arm_init_expanders ()
> > >
> > >  /* Length in units of the trampoline for entering a nested
> > function.  */
> > > -#define TRAMPOLINE_SIZE  (TARGET_32BIT ? 16 : 20)
> > > +#define TRAMPOLINE_SIZE  (TARGET_FDPIC ? 32 : (TARGET_32BIT ? 16 : 20))
> > >
> > >  /* Alignment required for a trampoline in bits.  */
> > >  #define TRAMPOLINE_ALIGNMENT  32
> > > --
> > > 2.6.3
> > >
Christophe Lyon Aug. 29, 2019, 3:36 p.m. UTC | #4
On 31/07/2019 16:44, Christophe Lyon wrote:
> On Tue, 16 Jul 2019 at 14:42, Kyrill Tkachov
> <kyrylo.tkachov@foss.arm.com> wrote:
>>
>>
>> On 7/16/19 12:18 PM, Kyrill Tkachov wrote:
>>> Hi Christophe
>>>
>>> On 5/15/19 1:39 PM, Christophe Lyon wrote:
>>>> In FDPIC mode, the trampoline generated to support pointers to nested
>>>> functions looks like:
>>>>
>>>>             .word trampoline address
>>>>             .word trampoline GOT address
>>>>             ldr            r12, [pc, #8]
>>>>             ldr            r9, [pc, #8]
>>>>             ldr           pc, [pc, #8]
>>>>             .word static chain value
>>>>             .word GOT address
>>>>             .word function's address
>>>>
>>>> because in FDPIC function pointers are actually pointers to function
>>>> descriptors, we have to actually generate a function descriptor for
>>>> the trampoline.
>>>>
>>>> 2019-XX-XX  Christophe Lyon <christophe.lyon@st.com>
>>>>          Mickaël Guêné <mickael.guene@st.com>
>>>>
>>>>          gcc/
>>>>          * config/arm/arm.c (arm_asm_trampoline_template): Add FDPIC
>>>>          support.
>>>>          (arm_trampoline_init): Likewise.
>>>>          (arm_trampoline_init): Likewise.
>>>>          * config/arm/arm.h (TRAMPOLINE_SIZE): Likewise.
>>>>
>>>> Change-Id: Idc4d5f629ae4f8d79bdf9623517481d524a0c144
>>>>
>>>> diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
>>>> index 40e3f3b..99d13bf 100644
>>>> --- a/gcc/config/arm/arm.c
>>>> +++ b/gcc/config/arm/arm.c
>>>> @@ -3976,13 +3976,50 @@ arm_warn_func_return (tree decl)
>>>>              .word static chain value
>>>>              .word function's address
>>>>      XXX FIXME: When the trampoline returns, r8 will be clobbered.  */
>>>> +/* In FDPIC mode, the trampoline looks like:
>>>> +          .word trampoline address
>>>> +          .word trampoline GOT address
>>>> +          ldr            r12, [pc, #8] ; #4 for Thumb2
>>>> +          ldr            r9,  [pc, #8] ; #4 for Thumb2
>>>> +          ldr           pc,  [pc, #8] ; #4 for Thumb2
>>>> +          .word static chain value
>>>> +          .word GOT address
>>>> +          .word function's address
>>>> +*/
>>>
>>>
>>> I think this comment is not right for Thumb2.
>>>
>>> These load instructionshave 32-bit encodings, even in Thumb2 (they use
>>> high registers).
>>
>> Andre and Wilco pointed out to me offline that the offset should be #4
>> for Arm mode.
>>
>> The Arm ARM at E1.2.3 says:
>>
>> PC, the program counter
>>
>> * When executing an A32 instruction, PC reads as the address of the
>> current instruction plus 8.
>>
>> * When executing a T32 instruction, PC reads as the address of the
>> current instruction plus 4.
>>
> 
> Yes, it looks like the code is right, and the comment is wrong:
> - offset 8 for thumb2 mode
> - offset 4 for arm mode
> 
Here is the updated version

> Thanks,
> 
> Christophe
> 
>> Thanks,
>>
>> Kyrill
>>
>>
>>>
>>> Also, please merge this comment with the one above (no separate /**/)
>>>
>>>>
>>>>   static void
>>>>   arm_asm_trampoline_template (FILE *f)
>>>>   {
>>>>     fprintf (f, "\t.syntax unified\n");
>>>>
>>>> -  if (TARGET_ARM)
>>>> +  if (TARGET_FDPIC)
>>>> +    {
>>>> +      /* The first two words are a function descriptor pointing to the
>>>> +        trampoline code just below.  */
>>>> +      if (TARGET_ARM)
>>>> +       fprintf (f, "\t.arm\n");
>>>> +      else if (TARGET_THUMB2)
>>>> +       fprintf (f, "\t.thumb\n");
>>>> +      else
>>>> +       /* Only ARM and Thumb-2 are supported.  */
>>>> +       gcc_unreachable ();
>>>> +
>>>> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
>>>> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
>>>> +      /* Trampoline code which sets the static chain register but also
>>>> +        PIC register before jumping into real code. */
>>>> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
>>>> +                  STATIC_CHAIN_REGNUM, PC_REGNUM,
>>>> +                  TARGET_THUMB2 ? 8 : 4);
>>>> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
>>>> +                  PIC_OFFSET_TABLE_REGNUM, PC_REGNUM,
>>>> +                  TARGET_THUMB2 ? 8 : 4);
>>>> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
>>>> +                  PC_REGNUM, PC_REGNUM,
>>>> +                  TARGET_THUMB2 ? 8 : 4);
>>>
>>>
>>> As above, I think the offset should be 8 for both Arm and Thumb2.
>>>
>>> Thanks,
>>>
>>> Kyrill
>>>
>>>
>>>> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
>>>> +    }
>>>> +  else if (TARGET_ARM)
>>>>       {
>>>>         fprintf (f, "\t.arm\n");
>>>>         asm_fprintf (f, "\tldr\t%r, [%r, #0]\n", STATIC_CHAIN_REGNUM,
>>>> PC_REGNUM);
>>>> @@ -4023,12 +4060,40 @@ arm_trampoline_init (rtx m_tramp, tree fndecl,
>>>> rtx chain_value)
>>>>     emit_block_move (m_tramp, assemble_trampoline_template (),
>>>>                      GEN_INT (TRAMPOLINE_SIZE), BLOCK_OP_NORMAL);
>>>>
>>>> -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
>>>> -  emit_move_insn (mem, chain_value);
>>>> +  if (TARGET_FDPIC)
>>>> +    {
>>>> +      rtx funcdesc = XEXP (DECL_RTL (fndecl), 0);
>>>> +      rtx fnaddr = gen_rtx_MEM (Pmode, funcdesc);
>>>> +      rtx gotaddr = gen_rtx_MEM (Pmode, plus_constant (Pmode,
>>>> funcdesc, 4));
>>>> +      /* The function start address is at offset 8, but in Thumb mode
>>>> +        we want bit 0 set to 1 to indicate Thumb-ness, hence 9
>>>> +        below.  */
>>>> +      rtx trampoline_code_start
>>>> +       = plus_constant (Pmode, XEXP (m_tramp, 0), TARGET_THUMB2 ? 9
>>> : 8);
>>>> +
>>>> +      /* Write initial funcdesc which points to the trampoline.  */
>>>> +      mem = adjust_address (m_tramp, SImode, 0);
>>>> +      emit_move_insn (mem, trampoline_code_start);
>>>> +      mem = adjust_address (m_tramp, SImode, 4);
>>>> +      emit_move_insn (mem, gen_rtx_REG (Pmode,
>>> PIC_OFFSET_TABLE_REGNUM));
>>>> +      /* Setup static chain.  */
>>>> +      mem = adjust_address (m_tramp, SImode, 20);
>>>> +      emit_move_insn (mem, chain_value);
>>>> +      /* GOT + real function entry point.  */
>>>> +      mem = adjust_address (m_tramp, SImode, 24);
>>>> +      emit_move_insn (mem, gotaddr);
>>>> +      mem = adjust_address (m_tramp, SImode, 28);
>>>> +      emit_move_insn (mem, fnaddr);
>>>> +    }
>>>> +  else
>>>> +    {
>>>> +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
>>>> +      emit_move_insn (mem, chain_value);
>>>>
>>>> -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
>>>> -  fnaddr = XEXP (DECL_RTL (fndecl), 0);
>>>> -  emit_move_insn (mem, fnaddr);
>>>> +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
>>>> +      fnaddr = XEXP (DECL_RTL (fndecl), 0);
>>>> +      emit_move_insn (mem, fnaddr);
>>>> +    }
>>>>
>>>>     a_tramp = XEXP (m_tramp, 0);
>>>>     emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__clear_cache"),
>>>> @@ -4042,7 +4107,9 @@ arm_trampoline_init (rtx m_tramp, tree fndecl,
>>>> rtx chain_value)
>>>>   static rtx
>>>>   arm_trampoline_adjust_address (rtx addr)
>>>>   {
>>>> -  if (TARGET_THUMB)
>>>> +  /* For FDPIC don't fix trampoline address since it's a function
>>>> +     descriptor and not a function address.  */
>>>> +  if (TARGET_THUMB && !TARGET_FDPIC)
>>>>       addr = expand_simple_binop (Pmode, IOR, addr, const1_rtx,
>>>>                                   NULL, 0, OPTAB_LIB_WIDEN);
>>>>     return addr;
>>>> diff --git a/gcc/config/arm/arm.h b/gcc/config/arm/arm.h
>>>> index 45c0e2b..f80df63 100644
>>>> --- a/gcc/config/arm/arm.h
>>>> +++ b/gcc/config/arm/arm.h
>>>> @@ -1602,7 +1602,7 @@ typedef struct
>>>>   #define INIT_EXPANDERS  arm_init_expanders ()
>>>>
>>>>   /* Length in units of the trampoline for entering a nested
>>> function.  */
>>>> -#define TRAMPOLINE_SIZE  (TARGET_32BIT ? 16 : 20)
>>>> +#define TRAMPOLINE_SIZE  (TARGET_FDPIC ? 32 : (TARGET_32BIT ? 16 : 20))
>>>>
>>>>   /* Alignment required for a trampoline in bits.  */
>>>>   #define TRAMPOLINE_ALIGNMENT  32
>>>> --
>>>> 2.6.3
>>>>
> .
>
Kyrill Tkachov Aug. 30, 2019, 8:32 a.m. UTC | #5
On 8/29/19 4:36 PM, Christophe Lyon wrote:
> On 31/07/2019 16:44, Christophe Lyon wrote:
>> On Tue, 16 Jul 2019 at 14:42, Kyrill Tkachov
>> <kyrylo.tkachov@foss.arm.com> wrote:
>>>
>>>
>>> On 7/16/19 12:18 PM, Kyrill Tkachov wrote:
>>>> Hi Christophe
>>>>
>>>> On 5/15/19 1:39 PM, Christophe Lyon wrote:
>>>>> In FDPIC mode, the trampoline generated to support pointers to nested
>>>>> functions looks like:
>>>>>
>>>>>             .word trampoline address
>>>>>             .word trampoline GOT address
>>>>>             ldr            r12, [pc, #8]
>>>>>             ldr            r9, [pc, #8]
>>>>>             ldr           pc, [pc, #8]
>>>>>             .word static chain value
>>>>>             .word GOT address
>>>>>             .word function's address
>>>>>
>>>>> because in FDPIC function pointers are actually pointers to function
>>>>> descriptors, we have to actually generate a function descriptor for
>>>>> the trampoline.
>>>>>
>>>>> 2019-XX-XX  Christophe Lyon <christophe.lyon@st.com>
>>>>>          Mickaël Guêné <mickael.guene@st.com>
>>>>>
>>>>>          gcc/
>>>>>          * config/arm/arm.c (arm_asm_trampoline_template): Add FDPIC
>>>>>          support.
>>>>>          (arm_trampoline_init): Likewise.
>>>>>          (arm_trampoline_init): Likewise.
>>>>>          * config/arm/arm.h (TRAMPOLINE_SIZE): Likewise.
>>>>>
>>>>> Change-Id: Idc4d5f629ae4f8d79bdf9623517481d524a0c144
>>>>>
>>>>> diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
>>>>> index 40e3f3b..99d13bf 100644
>>>>> --- a/gcc/config/arm/arm.c
>>>>> +++ b/gcc/config/arm/arm.c
>>>>> @@ -3976,13 +3976,50 @@ arm_warn_func_return (tree decl)
>>>>>              .word static chain value
>>>>>              .word function's address
>>>>>      XXX FIXME: When the trampoline returns, r8 will be 
>>>>> clobbered.  */
>>>>> +/* In FDPIC mode, the trampoline looks like:
>>>>> +          .word trampoline address
>>>>> +          .word trampoline GOT address
>>>>> +          ldr            r12, [pc, #8] ; #4 for Thumb2
>>>>> +          ldr            r9,  [pc, #8] ; #4 for Thumb2
>>>>> +          ldr           pc,  [pc, #8] ; #4 for Thumb2
>>>>> +          .word static chain value
>>>>> +          .word GOT address
>>>>> +          .word function's address
>>>>> +*/
>>>>
>>>>
>>>> I think this comment is not right for Thumb2.
>>>>
>>>> These load instructionshave 32-bit encodings, even in Thumb2 (they use
>>>> high registers).
>>>
>>> Andre and Wilco pointed out to me offline that the offset should be #4
>>> for Arm mode.
>>>
>>> The Arm ARM at E1.2.3 says:
>>>
>>> PC, the program counter
>>>
>>> * When executing an A32 instruction, PC reads as the address of the
>>> current instruction plus 8.
>>>
>>> * When executing a T32 instruction, PC reads as the address of the
>>> current instruction plus 4.
>>>
>>
>> Yes, it looks like the code is right, and the comment is wrong:
>> - offset 8 for thumb2 mode
>> - offset 4 for arm mode
>>
> Here is the updated version
>
Ok with a fixed ChangeLog (it currently mentions arm_trampoline_init 
twice but doesn't mention arm_trampoline_adjust_address)

Thanks,

Kyrill


>> Thanks,
>>
>> Christophe
>>
>>> Thanks,
>>>
>>> Kyrill
>>>
>>>
>>>>
>>>> Also, please merge this comment with the one above (no separate /**/)
>>>>
>>>>>
>>>>>   static void
>>>>>   arm_asm_trampoline_template (FILE *f)
>>>>>   {
>>>>>     fprintf (f, "\t.syntax unified\n");
>>>>>
>>>>> -  if (TARGET_ARM)
>>>>> +  if (TARGET_FDPIC)
>>>>> +    {
>>>>> +      /* The first two words are a function descriptor pointing 
>>>>> to the
>>>>> +        trampoline code just below.  */
>>>>> +      if (TARGET_ARM)
>>>>> +       fprintf (f, "\t.arm\n");
>>>>> +      else if (TARGET_THUMB2)
>>>>> +       fprintf (f, "\t.thumb\n");
>>>>> +      else
>>>>> +       /* Only ARM and Thumb-2 are supported.  */
>>>>> +       gcc_unreachable ();
>>>>> +
>>>>> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
>>>>> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
>>>>> +      /* Trampoline code which sets the static chain register but 
>>>>> also
>>>>> +        PIC register before jumping into real code. */
>>>>> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
>>>>> +                  STATIC_CHAIN_REGNUM, PC_REGNUM,
>>>>> +                  TARGET_THUMB2 ? 8 : 4);
>>>>> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
>>>>> +                  PIC_OFFSET_TABLE_REGNUM, PC_REGNUM,
>>>>> +                  TARGET_THUMB2 ? 8 : 4);
>>>>> +      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
>>>>> +                  PC_REGNUM, PC_REGNUM,
>>>>> +                  TARGET_THUMB2 ? 8 : 4);
>>>>
>>>>
>>>> As above, I think the offset should be 8 for both Arm and Thumb2.
>>>>
>>>> Thanks,
>>>>
>>>> Kyrill
>>>>
>>>>
>>>>> +      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
>>>>> +    }
>>>>> +  else if (TARGET_ARM)
>>>>>       {
>>>>>         fprintf (f, "\t.arm\n");
>>>>>         asm_fprintf (f, "\tldr\t%r, [%r, #0]\n", STATIC_CHAIN_REGNUM,
>>>>> PC_REGNUM);
>>>>> @@ -4023,12 +4060,40 @@ arm_trampoline_init (rtx m_tramp, tree 
>>>>> fndecl,
>>>>> rtx chain_value)
>>>>>     emit_block_move (m_tramp, assemble_trampoline_template (),
>>>>>                      GEN_INT (TRAMPOLINE_SIZE), BLOCK_OP_NORMAL);
>>>>>
>>>>> -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
>>>>> -  emit_move_insn (mem, chain_value);
>>>>> +  if (TARGET_FDPIC)
>>>>> +    {
>>>>> +      rtx funcdesc = XEXP (DECL_RTL (fndecl), 0);
>>>>> +      rtx fnaddr = gen_rtx_MEM (Pmode, funcdesc);
>>>>> +      rtx gotaddr = gen_rtx_MEM (Pmode, plus_constant (Pmode,
>>>>> funcdesc, 4));
>>>>> +      /* The function start address is at offset 8, but in Thumb 
>>>>> mode
>>>>> +        we want bit 0 set to 1 to indicate Thumb-ness, hence 9
>>>>> +        below.  */
>>>>> +      rtx trampoline_code_start
>>>>> +       = plus_constant (Pmode, XEXP (m_tramp, 0), TARGET_THUMB2 ? 9
>>>> : 8);
>>>>> +
>>>>> +      /* Write initial funcdesc which points to the trampoline.  */
>>>>> +      mem = adjust_address (m_tramp, SImode, 0);
>>>>> +      emit_move_insn (mem, trampoline_code_start);
>>>>> +      mem = adjust_address (m_tramp, SImode, 4);
>>>>> +      emit_move_insn (mem, gen_rtx_REG (Pmode,
>>>> PIC_OFFSET_TABLE_REGNUM));
>>>>> +      /* Setup static chain.  */
>>>>> +      mem = adjust_address (m_tramp, SImode, 20);
>>>>> +      emit_move_insn (mem, chain_value);
>>>>> +      /* GOT + real function entry point.  */
>>>>> +      mem = adjust_address (m_tramp, SImode, 24);
>>>>> +      emit_move_insn (mem, gotaddr);
>>>>> +      mem = adjust_address (m_tramp, SImode, 28);
>>>>> +      emit_move_insn (mem, fnaddr);
>>>>> +    }
>>>>> +  else
>>>>> +    {
>>>>> +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
>>>>> +      emit_move_insn (mem, chain_value);
>>>>>
>>>>> -  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
>>>>> -  fnaddr = XEXP (DECL_RTL (fndecl), 0);
>>>>> -  emit_move_insn (mem, fnaddr);
>>>>> +      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 
>>>>> 16);
>>>>> +      fnaddr = XEXP (DECL_RTL (fndecl), 0);
>>>>> +      emit_move_insn (mem, fnaddr);
>>>>> +    }
>>>>>
>>>>>     a_tramp = XEXP (m_tramp, 0);
>>>>>     emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__clear_cache"),
>>>>> @@ -4042,7 +4107,9 @@ arm_trampoline_init (rtx m_tramp, tree fndecl,
>>>>> rtx chain_value)
>>>>>   static rtx
>>>>>   arm_trampoline_adjust_address (rtx addr)
>>>>>   {
>>>>> -  if (TARGET_THUMB)
>>>>> +  /* For FDPIC don't fix trampoline address since it's a function
>>>>> +     descriptor and not a function address.  */
>>>>> +  if (TARGET_THUMB && !TARGET_FDPIC)
>>>>>       addr = expand_simple_binop (Pmode, IOR, addr, const1_rtx,
>>>>>                                   NULL, 0, OPTAB_LIB_WIDEN);
>>>>>     return addr;
>>>>> diff --git a/gcc/config/arm/arm.h b/gcc/config/arm/arm.h
>>>>> index 45c0e2b..f80df63 100644
>>>>> --- a/gcc/config/arm/arm.h
>>>>> +++ b/gcc/config/arm/arm.h
>>>>> @@ -1602,7 +1602,7 @@ typedef struct
>>>>>   #define INIT_EXPANDERS  arm_init_expanders ()
>>>>>
>>>>>   /* Length in units of the trampoline for entering a nested
>>>> function.  */
>>>>> -#define TRAMPOLINE_SIZE (TARGET_32BIT ? 16 : 20)
>>>>> +#define TRAMPOLINE_SIZE  (TARGET_FDPIC ? 32 : (TARGET_32BIT ? 16 
>>>>> : 20))
>>>>>
>>>>>   /* Alignment required for a trampoline in bits.  */
>>>>>   #define TRAMPOLINE_ALIGNMENT  32
>>>>> -- 
>>>>> 2.6.3
>>>>>
>> .
>>
>
diff mbox series

Patch

diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
index 40e3f3b..99d13bf 100644
--- a/gcc/config/arm/arm.c
+++ b/gcc/config/arm/arm.c
@@ -3976,13 +3976,50 @@  arm_warn_func_return (tree decl)
 	   .word	static chain value
 	   .word	function's address
    XXX FIXME: When the trampoline returns, r8 will be clobbered.  */
+/* In FDPIC mode, the trampoline looks like:
+	   .word	trampoline address
+	   .word	trampoline GOT address
+	   ldr 		r12, [pc, #8] ; #4 for Thumb2
+	   ldr 		r9,  [pc, #8] ; #4 for Thumb2
+	   ldr		pc,  [pc, #8] ; #4 for Thumb2
+	   .word	static chain value
+	   .word	GOT address
+	   .word	function's address
+*/
 
 static void
 arm_asm_trampoline_template (FILE *f)
 {
   fprintf (f, "\t.syntax unified\n");
 
-  if (TARGET_ARM)
+  if (TARGET_FDPIC)
+    {
+      /* The first two words are a function descriptor pointing to the
+	 trampoline code just below.  */
+      if (TARGET_ARM)
+	fprintf (f, "\t.arm\n");
+      else if (TARGET_THUMB2)
+	fprintf (f, "\t.thumb\n");
+      else
+	/* Only ARM and Thumb-2 are supported.  */
+	gcc_unreachable ();
+
+      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
+      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
+      /* Trampoline code which sets the static chain register but also
+	 PIC register before jumping into real code.  */
+      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
+		   STATIC_CHAIN_REGNUM, PC_REGNUM,
+		   TARGET_THUMB2 ? 8 : 4);
+      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
+		   PIC_OFFSET_TABLE_REGNUM, PC_REGNUM,
+		   TARGET_THUMB2 ? 8 : 4);
+      asm_fprintf (f, "\tldr\t%r, [%r, #%d]\n",
+		   PC_REGNUM, PC_REGNUM,
+		   TARGET_THUMB2 ? 8 : 4);
+      assemble_aligned_integer (UNITS_PER_WORD, const0_rtx);
+    }
+  else if (TARGET_ARM)
     {
       fprintf (f, "\t.arm\n");
       asm_fprintf (f, "\tldr\t%r, [%r, #0]\n", STATIC_CHAIN_REGNUM, PC_REGNUM);
@@ -4023,12 +4060,40 @@  arm_trampoline_init (rtx m_tramp, tree fndecl, rtx chain_value)
   emit_block_move (m_tramp, assemble_trampoline_template (),
 		   GEN_INT (TRAMPOLINE_SIZE), BLOCK_OP_NORMAL);
 
-  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
-  emit_move_insn (mem, chain_value);
+  if (TARGET_FDPIC)
+    {
+      rtx funcdesc = XEXP (DECL_RTL (fndecl), 0);
+      rtx fnaddr = gen_rtx_MEM (Pmode, funcdesc);
+      rtx gotaddr = gen_rtx_MEM (Pmode, plus_constant (Pmode, funcdesc, 4));
+      /* The function start address is at offset 8, but in Thumb mode
+	 we want bit 0 set to 1 to indicate Thumb-ness, hence 9
+	 below.  */
+      rtx trampoline_code_start
+	= plus_constant (Pmode, XEXP (m_tramp, 0), TARGET_THUMB2 ? 9 : 8);
+
+      /* Write initial funcdesc which points to the trampoline.  */
+      mem = adjust_address (m_tramp, SImode, 0);
+      emit_move_insn (mem, trampoline_code_start);
+      mem = adjust_address (m_tramp, SImode, 4);
+      emit_move_insn (mem, gen_rtx_REG (Pmode, PIC_OFFSET_TABLE_REGNUM));
+      /* Setup static chain.  */
+      mem = adjust_address (m_tramp, SImode, 20);
+      emit_move_insn (mem, chain_value);
+      /* GOT + real function entry point.  */
+      mem = adjust_address (m_tramp, SImode, 24);
+      emit_move_insn (mem, gotaddr);
+      mem = adjust_address (m_tramp, SImode, 28);
+      emit_move_insn (mem, fnaddr);
+    }
+  else
+    {
+      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 8 : 12);
+      emit_move_insn (mem, chain_value);
 
-  mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
-  fnaddr = XEXP (DECL_RTL (fndecl), 0);
-  emit_move_insn (mem, fnaddr);
+      mem = adjust_address (m_tramp, SImode, TARGET_32BIT ? 12 : 16);
+      fnaddr = XEXP (DECL_RTL (fndecl), 0);
+      emit_move_insn (mem, fnaddr);
+    }
 
   a_tramp = XEXP (m_tramp, 0);
   emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__clear_cache"),
@@ -4042,7 +4107,9 @@  arm_trampoline_init (rtx m_tramp, tree fndecl, rtx chain_value)
 static rtx
 arm_trampoline_adjust_address (rtx addr)
 {
-  if (TARGET_THUMB)
+  /* For FDPIC don't fix trampoline address since it's a function
+     descriptor and not a function address.  */
+  if (TARGET_THUMB && !TARGET_FDPIC)
     addr = expand_simple_binop (Pmode, IOR, addr, const1_rtx,
 				NULL, 0, OPTAB_LIB_WIDEN);
   return addr;
diff --git a/gcc/config/arm/arm.h b/gcc/config/arm/arm.h
index 45c0e2b..f80df63 100644
--- a/gcc/config/arm/arm.h
+++ b/gcc/config/arm/arm.h
@@ -1602,7 +1602,7 @@  typedef struct
 #define INIT_EXPANDERS  arm_init_expanders ()
 
 /* Length in units of the trampoline for entering a nested function.  */
-#define TRAMPOLINE_SIZE  (TARGET_32BIT ? 16 : 20)
+#define TRAMPOLINE_SIZE  (TARGET_FDPIC ? 32 : (TARGET_32BIT ? 16 : 20))
 
 /* Alignment required for a trampoline in bits.  */
 #define TRAMPOLINE_ALIGNMENT  32