[v3] LoongArch: Fix eh_return epilogue for normal returns
Checks
Commit Message
On LoongArch, the regitsters $r4 - $r7 (EH_RETURN_DATA_REGNO) will be saved
and restored in the function prologue and epilogue if the given function calls
__builtin_eh_return. This causes the return value to be overwritten on normal
return paths and breaks a rare case of libgcc's _Unwind_RaiseException.
gcc/ChangeLog:
* config/loongarch/loongarch.cc: Do not restore the saved eh_return
data registers ($r4-$r7) for a normal return of a function that calls
__builtin_eh_return elsewhere.
* config/loongarch/loongarch-protos.h: Same.
* config/loongarch/loongarch.md: Same.
gcc/testsuite/ChangeLog:
* gcc.target/loongarch/eh_return-normal-return.c: New test.
---
gcc/config/loongarch/loongarch-protos.h | 2 +-
gcc/config/loongarch/loongarch.cc | 41 ++++++++++++-------
gcc/config/loongarch/loongarch.md | 18 +++++++-
.../loongarch/eh_return-normal-return.c | 32 +++++++++++++++
4 files changed, 76 insertions(+), 17 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/loongarch/eh_return-normal-return.c
Comments
On Thu, 2023-12-07 at 09:40 +0800, Yang Yujie wrote:
> static void
> loongarch_for_each_saved_reg (HOST_WIDE_INT sp_offset,
> - loongarch_save_restore_fn fn)
> + loongarch_save_restore_fn fn,
> + bool skip_eh_data_regs_p)
> {
> HOST_WIDE_INT offset;
>
> /* Save the link register and s-registers. */
> offset = cfun->machine->frame.gp_sp_offset - sp_offset;
> for (int regno = GP_REG_FIRST; regno <= GP_REG_LAST; regno++)
> - if (BITSET_P (cfun->machine->frame.mask, regno - GP_REG_FIRST))
> - {
> - if (!cfun->machine->reg_is_wrapped_separately[regno])
> - loongarch_save_restore_reg (word_mode, regno, offset, fn);
> + {
> + /* Special care needs to be taken for $r4-$r7 (EH_RETURN_DATA_REGNO)
> + when returning normally from a function that calls __builtin_eh_return.
> + In this case, these registers are saved but should not be restored,
> + or the return value may be clobbered. */
>
> - offset -= UNITS_PER_WORD;
> - }
> + if (BITSET_P (cfun->machine->frame.mask, regno - GP_REG_FIRST))
> + {
> + if (!(cfun->machine->reg_is_wrapped_separately[regno]
> + || (skip_eh_data_regs_p
> + && GP_ARG_FIRST <= regno && regno < GP_ARG_FIRST + 4)))
> + loongarch_save_restore_reg (word_mode, regno, offset, fn);
> +
> + offset -= UNITS_PER_WORD;
> + }
> + }
I don't like this pair of {} for the for statement. It's not necessary
and it changes the indent level, causing the diff hard to review.
Otherwise LGTM. I'm not sure why I didn't notice the eh_return issue
when I learnt shrink wrapping from RISC-V...
On Thu, Dec 07, 2023 at 11:02:58AM +0800, Xi Ruoyao wrote:
>
> I don't like this pair of {} for the for statement. It's not necessary
> and it changes the indent level, causing the diff hard to review.
>
> Otherwise LGTM. I'm not sure why I didn't notice the eh_return issue
> when I learnt shrink wrapping from RISC-V...
>
Thanks for the review! This problem on LoongArch was first noticed in a
failed libphobos test case, and the fix is partially borrowed from i386,
which seemed to be the only architecture without this issue.
So despite the extra braces (which I'd say I prefer to have because of the
new block of comment inserted on top of the if statement :P), I am going to
ask Lulu for pushing this.
Yujie
On Thu, 2023-12-07 at 14:18 +0800, Yang Yujie wrote:
> On Thu, Dec 07, 2023 at 11:02:58AM +0800, Xi Ruoyao wrote:
> >
> > I don't like this pair of {} for the for statement. It's not necessary
> > and it changes the indent level, causing the diff hard to review.
> >
> > Otherwise LGTM. I'm not sure why I didn't notice the eh_return issue
> > when I learnt shrink wrapping from RISC-V...
> >
>
> Thanks for the review! This problem on LoongArch was first noticed in a
> failed libphobos test case, and the fix is partially borrowed from i386,
> which seemed to be the only architecture without this issue.
>
> So despite the extra braces (which I'd say I prefer to have because of the
> new block of comment inserted on top of the if statement :P), I am going to
> ask Lulu for pushing this.
I understand and I don't think adding {} is wrong. The problem is the
indent change causes a large chunk of diff and it makes reviewing more
difficult. Thus generally we should not mix real code change and format
change in a commit.
i. e. it would be better to separate it into two patches, the first adds
{} and changes the indent, and the second changes the logic. But now I
don't think it's needed to make a V4, just pushing this should be fine.
On Thu, Dec 07, 2023 at 04:13:51PM +0800, Xi Ruoyao wrote:
>
> I understand and I don't think adding {} is wrong. The problem is the
> indent change causes a large chunk of diff and it makes reviewing more
> difficult. Thus generally we should not mix real code change and format
> change in a commit.
>
> i. e. it would be better to separate it into two patches, the first adds
> {} and changes the indent, and the second changes the logic. But now I
> don't think it's needed to make a V4, just pushing this should be fine.
>
Thanks for the suggestion. It is clearer that way.
Yujie
@@ -60,7 +60,7 @@ enum loongarch_symbol_type {
extern rtx loongarch_emit_move (rtx, rtx);
extern HOST_WIDE_INT loongarch_initial_elimination_offset (int, int);
extern void loongarch_expand_prologue (void);
-extern void loongarch_expand_epilogue (bool);
+extern void loongarch_expand_epilogue (int);
extern bool loongarch_can_use_return_insn (void);
extern bool loongarch_symbolic_constant_p (rtx, enum loongarch_symbol_type *);
@@ -1015,20 +1015,30 @@ loongarch_save_restore_reg (machine_mode mode, int regno, HOST_WIDE_INT offset,
static void
loongarch_for_each_saved_reg (HOST_WIDE_INT sp_offset,
- loongarch_save_restore_fn fn)
+ loongarch_save_restore_fn fn,
+ bool skip_eh_data_regs_p)
{
HOST_WIDE_INT offset;
/* Save the link register and s-registers. */
offset = cfun->machine->frame.gp_sp_offset - sp_offset;
for (int regno = GP_REG_FIRST; regno <= GP_REG_LAST; regno++)
- if (BITSET_P (cfun->machine->frame.mask, regno - GP_REG_FIRST))
- {
- if (!cfun->machine->reg_is_wrapped_separately[regno])
- loongarch_save_restore_reg (word_mode, regno, offset, fn);
+ {
+ /* Special care needs to be taken for $r4-$r7 (EH_RETURN_DATA_REGNO)
+ when returning normally from a function that calls __builtin_eh_return.
+ In this case, these registers are saved but should not be restored,
+ or the return value may be clobbered. */
- offset -= UNITS_PER_WORD;
- }
+ if (BITSET_P (cfun->machine->frame.mask, regno - GP_REG_FIRST))
+ {
+ if (!(cfun->machine->reg_is_wrapped_separately[regno]
+ || (skip_eh_data_regs_p
+ && GP_ARG_FIRST <= regno && regno < GP_ARG_FIRST + 4)))
+ loongarch_save_restore_reg (word_mode, regno, offset, fn);
+
+ offset -= UNITS_PER_WORD;
+ }
+ }
/* This loop must iterate over the same space as its companion in
loongarch_compute_frame_info. */
@@ -1297,7 +1307,7 @@ loongarch_expand_prologue (void)
GEN_INT (-step1));
RTX_FRAME_RELATED_P (emit_insn (insn)) = 1;
size -= step1;
- loongarch_for_each_saved_reg (size, loongarch_save_reg);
+ loongarch_for_each_saved_reg (size, loongarch_save_reg, false);
}
/* Set up the frame pointer, if we're using one. */
@@ -1382,11 +1392,11 @@ loongarch_can_use_return_insn (void)
return reload_completed && cfun->machine->frame.total_size == 0;
}
-/* Expand an "epilogue" or "sibcall_epilogue" pattern; SIBCALL_P
- says which. */
+/* Expand function epilogue for the following insn patterns:
+ "epilogue" (style == 0) / "sibcall_epilogue" (1) / "eh_return" (2). */
void
-loongarch_expand_epilogue (bool sibcall_p)
+loongarch_expand_epilogue (int style)
{
/* Split the frame into two. STEP1 is the amount of stack we should
deallocate before restoring the registers. STEP2 is the amount we
@@ -1403,7 +1413,8 @@ loongarch_expand_epilogue (bool sibcall_p)
bool need_barrier_p
= (get_frame_size () + cfun->machine->frame.arg_pointer_offset) != 0;
- if (!sibcall_p && loongarch_can_use_return_insn ())
+ /* Handle simple returns. */
+ if (style == 0 && loongarch_can_use_return_insn ())
{
emit_jump_insn (gen_return ());
return;
@@ -1479,7 +1490,8 @@ loongarch_expand_epilogue (bool sibcall_p)
/* Restore the registers. */
loongarch_for_each_saved_reg (frame->total_size - step2,
- loongarch_restore_reg);
+ loongarch_restore_reg,
+ crtl->calls_eh_return && style != 2);
if (need_barrier_p)
loongarch_emit_stack_tie ();
@@ -1504,7 +1516,8 @@ loongarch_expand_epilogue (bool sibcall_p)
emit_insn (gen_add3_insn (stack_pointer_rtx, stack_pointer_rtx,
EH_RETURN_STACKADJ_RTX));
- if (!sibcall_p)
+ /* Emit return unless doing sibcall. */
+ if (style != 1)
emit_jump_insn (gen_simple_return_internal (ra));
}
@@ -3197,7 +3197,7 @@ (define_expand "epilogue"
[(const_int 2)]
""
{
- loongarch_expand_epilogue (false);
+ loongarch_expand_epilogue (0);
DONE;
})
@@ -3205,7 +3205,7 @@ (define_expand "sibcall_epilogue"
[(const_int 2)]
""
{
- loongarch_expand_epilogue (true);
+ loongarch_expand_epilogue (1);
DONE;
})
@@ -3262,6 +3262,20 @@ (define_expand "eh_return"
emit_insn (gen_eh_set_ra_di (operands[0]));
else
emit_insn (gen_eh_set_ra_si (operands[0]));
+
+ emit_jump_insn (gen_eh_return_internal ());
+ emit_barrier ();
+ DONE;
+})
+
+(define_insn_and_split "eh_return_internal"
+ [(eh_return)]
+ ""
+ "#"
+ "epilogue_completed"
+ [(const_int 0)]
+{
+ loongarch_expand_epilogue (2);
DONE;
})
new file mode 100644
@@ -0,0 +1,32 @@
+/* { dg-do run } */
+/* { dg-options "-O2" } */
+
+#include <stdlib.h>
+
+int foo () __attribute__((noinline));
+int main ();
+
+int
+foo () {
+
+ int t;
+
+ /* prevent optimization using asm */
+ asm ("" : "=r" (t) : "0" (-1));
+ asm ("" : "=r" (t) : "0" (t ? 1 : 0));
+
+ if (t == 0)
+ /* never reached */
+ __builtin_eh_return (0, __builtin_return_address (0));
+ else
+ return 202312;
+}
+
+int
+main ()
+{
+ if (foo() == 202312)
+ return 0;
+ else
+ abort ();
+}