public inbox for gcc@gcc.gnu.org
 help / color / mirror / Atom feed
* [Exception Handling] Why does frame unwind label have static alignment 4?
@ 2022-01-17 17:24 Joseph Faulls
  2022-01-17 18:55 ` H.J. Lu
  0 siblings, 1 reply; 2+ messages in thread
From: Joseph Faulls @ 2022-01-17 17:24 UTC (permalink / raw)
  To: gcc

Hello all,

Small disclaimer of being new to EH, but I have come across a problem that seems quite fundamental to EH frame registry. I am targeting riscv64, but I believe the problem to be entirely platform agnostic.

When using EH frame registry, a label is put at the beginning of the frame unwind info to register and deregister frames:

libgcc/crtstuff.c:
> STATIC EH_FRAME_SECTION_CONST char __EH_FRAME_BEGIN__[]
>      __attribute__((section(__LIBGCC_EH_FRAME_SECTION_NAME__), aligned(4)))

This is then populated with struct dwarf_fde:

libgcc/unwind-dw2-fde.h:
> struct dwarf_fde
> {
>   uword length;
>   sword CIE_delta;
>   unsigned char pc_begin[];
> } __attribute__ ((packed, aligned (__alignof__ (void *))));

The trouble is that the __EH_FRAME_BEGIN__ has alignment 4, whereas the struct dwarf_fde has alignment of void*. When targeting 64-bit, these values are not the same.
It is this that is causing 4 bytes of padding. This is the disassembly of a simple program targeting riscv64 that throws an exception:
> 22954 000000000000e324 <__EH_FRAME_BEGIN__>:
> e324:       0000                    unimp
> e326:       0000                    unimp
> e328:       0010                    0x10
> e32a:       0000                    unimp
> e32c:       0000                    unimp
> e32e:       0000                    unimp
> e330:       00527a03                0x527a03

It is this that causes __register_frame_info_bases to fall flat on its face:

Libgcc/crtstuff.c in function frame_dummy:
> __register_frame_info (__EH_FRAME_BEGIN__, &object);

libgcc/unwind-dw2-fde.c
>   void
>   __register_frame_info_bases (const void *begin, struct object *ob,
>                                void *tbase, void *dbase)
>   {
>     /* If .eh_frame is empty, don't register at all.  */
>     if ((const uword *) begin == 0 || *(const uword *) begin == 0)
>       return;
> ...
> void
> __register_frame_info (const void *begin, struct object *ob)
> {
>   __register_frame_info_bases (begin, ob, 0, 0);
> }

This means no EH frames are ever registered, and so exceptions cannot be handled!

So, why does __EH_FRAME_BEGIN__ have alignment 4 and not of __alignof__ (void *) ? Is this a bug?

Thanks,
Joe

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [Exception Handling] Why does frame unwind label have static alignment 4?
  2022-01-17 17:24 [Exception Handling] Why does frame unwind label have static alignment 4? Joseph Faulls
@ 2022-01-17 18:55 ` H.J. Lu
  0 siblings, 0 replies; 2+ messages in thread
From: H.J. Lu @ 2022-01-17 18:55 UTC (permalink / raw)
  To: Joseph Faulls; +Cc: gcc

On Mon, Jan 17, 2022 at 9:24 AM Joseph Faulls <Joseph.Faulls@imgtec.com> wrote:
>
> Hello all,
>
> Small disclaimer of being new to EH, but I have come across a problem that seems quite fundamental to EH frame registry. I am targeting riscv64, but I believe the problem to be entirely platform agnostic.
>
> When using EH frame registry, a label is put at the beginning of the frame unwind info to register and deregister frames:
>
> libgcc/crtstuff.c:
> > STATIC EH_FRAME_SECTION_CONST char __EH_FRAME_BEGIN__[]
> >      __attribute__((section(__LIBGCC_EH_FRAME_SECTION_NAME__), aligned(4)))
>
> This is then populated with struct dwarf_fde:
>
> libgcc/unwind-dw2-fde.h:
> > struct dwarf_fde
> > {
> >   uword length;
> >   sword CIE_delta;
> >   unsigned char pc_begin[];
> > } __attribute__ ((packed, aligned (__alignof__ (void *))));
>
> The trouble is that the __EH_FRAME_BEGIN__ has alignment 4, whereas the struct dwarf_fde has alignment of void*. When targeting 64-bit, these values are not the same.
> It is this that is causing 4 bytes of padding. This is the disassembly of a simple program targeting riscv64 that throws an exception:
> > 22954 000000000000e324 <__EH_FRAME_BEGIN__>:
> > e324:       0000                    unimp
> > e326:       0000                    unimp
> > e328:       0010                    0x10
> > e32a:       0000                    unimp
> > e32c:       0000                    unimp
> > e32e:       0000                    unimp
> > e330:       00527a03                0x527a03
>
> It is this that causes __register_frame_info_bases to fall flat on its face:
>
> Libgcc/crtstuff.c in function frame_dummy:
> > __register_frame_info (__EH_FRAME_BEGIN__, &object);
>
> libgcc/unwind-dw2-fde.c
> >   void
> >   __register_frame_info_bases (const void *begin, struct object *ob,
> >                                void *tbase, void *dbase)
> >   {
> >     /* If .eh_frame is empty, don't register at all.  */
> >     if ((const uword *) begin == 0 || *(const uword *) begin == 0)
> >       return;
> > ...
> > void
> > __register_frame_info (const void *begin, struct object *ob)
> > {
> >   __register_frame_info_bases (begin, ob, 0, 0);
> > }
>
> This means no EH frames are ever registered, and so exceptions cannot be handled!
>
> So, why does __EH_FRAME_BEGIN__ have alignment 4 and not of __alignof__ (void *) ? Is this a bug?
>
> Thanks,
> Joe

This is:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=27576

-- 
H.J.

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2022-01-17 18:55 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-01-17 17:24 [Exception Handling] Why does frame unwind label have static alignment 4? Joseph Faulls
2022-01-17 18:55 ` H.J. Lu

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).