Re: [patch/rfa] hppa-linux target, 2nd try

[Andrew Cagney <cagney@gnu.org>]

>>>+static struct insn_pattern hppa_sigtramp[] = {
>>>> >+  /* ldi 0, %r25 or ldi 1, %r25 */
>>>> >+  { 0x34190000, 0xfffffffd },
>>>> >+  /* ldi __NR_rt_sigreturn, %r20 */
>>>> >+  { 0x3414015a, 0xffffffff },
>>>> >+  /* be,l 0x100(%sr2, %r0), %sr0, %r31 */
>>>> >+  { 0xe4008200, 0xffffffff },
>>>> >+  /* nop */
>>>> >+  { 0x08000240, 0xffffffff },
>>>> >+  { 0, 0 }
>>>> >+};
>>
>>> 
>>> hmm, perhaps the mask can be integrated into tramp-frame.[hc], anyone?
> 
> 
> well, hppa still can't use it because of the variable offset in the
> stack frame to locate the trampoline :(

As with the following comment?

> +/* (This is derived from MD_FALLBACK_FRAME_STATE_FOR in gcc.)
> + 
> +   Unfortunately, because of various bugs and changes to the kernel,
> +   we have several cases to deal with.
> +
> +   In 2.4, the signal trampoline is 4 bytes, and pc should point directly at 
> +   the beginning of the trampoline and struct rt_sigframe.
> +
> +   In <= 2.6.5-rc2-pa3, the signal trampoline is 9 bytes, and pc points at
> +   the 4th word in the trampoline structure.  This is wrong, it should point 
> +   at the 5th word.  This is fixed in 2.6.5-rc2-pa4.
> +
> +   To detect these cases, we first take pc, align it to 64-bytes
> +   to get the beginning of the signal frame, and then check offsets 0, 4
> +   and 5 to see if we found the beginning of the trampoline.  This will
> +   tell us how to locate the sigcontext structure.
> +
> +   Note that with a 2.4 64-bit kernel, the signal context is not properly
> +   passed back to userspace so the unwind will not work correctly.  */

The method is passed the start address of the trampoline vis:

   /* Initialize a trad-frame cache corresponding to the tramp-frame.
      FUNC is the address of the instruction TRAMP[0] in memory.  */
   void (*init) (const struct tramp_frame *self,
                 struct frame_info *next_frame,
                 struct trad_frame_cache *this_cache,
                 CORE_ADDR func);

and from that it should be possible to compute the offset.

Andrew