[patch/rfa] Signal trampoline unwinder for hppa-hpux

[patch/rfa] Signal trampoline unwinder for hppa-hpux

[Randolph Chung]

This patch implements a signal trampoline unwinder for hpux. It also
cleans up some old sigtramp cruft that was in hppa-hpux-tdep.c.

I tested this against hppa2.0w-hp-hpux11.11 and hppa2.0w-hp-hpux11.00,
but I don't have a hpux machine running a 32-bit kernel to test. If
anyone is interested in that target, please give it a try. It *should*
work :)

Unfortunately this still fails a lot of the signal related testcases.
It is not because of a problem with the signal unwinder though:

On HPUX-SOM, there is an export stub inserted into the call sequence in
some cases; the export stub is hit on the return path from indirect
function calls, and apparently on return from a signal handler. This 
creates an extra frame in the backtrace, viz:

(gdb) bt
#0  0x000029e4 in handle_USR1 (sig=0)
    at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:18
#1  0x000029c8 in handle_USR1 ()
    at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:17
#2  <signal handler called>
#3  main () at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:40

Frame 1 comes from the export stub. Because of this, the pattern
matching in the backtrace test fails.  I am not sure if this is a
problem with the test script or with the handling of export stubs -- am
I supposed to somehow supress the export stub from showing up in the
backtrace? (if so, how?)

Comments? ok to check in?

randolph

2004-05-19  Randolph Chung  <tausq@debian.org>

    * hppa-hpux-tdep.c (offsetof): Define.
    (hppa_hpux_pc_in_sigtramp, hppa32_hpux_frame_saved_pc_in_sigtramp)
    (hppa32_hpux_frame_base_before_sigtramp)
    (hppa32_hpux_frame_find_saved_regs_in_sigtramp)
    (hppa64_hpux_frame_saved_pc_in_sigtramp)
    (hppa64_hpux_frame_base_before_sigtramp)
    (hppa64_hpux_frame_find_saved_regs_in_sigtramp): Remove unused functions.
    (hppa_hpux_sigtramp_find_sigcontext, hppa_hpux_sigtramp_unwind_cache)
    (hppa_hpux_sigtramp_frame_unwind_cache, hppa_hpux_sigtramp_frame_this_id)
    (hppa_hpux_sigtramp_frame_prev_register, hppa_hpux_sigtramp_frame_unwind)
    (hppa_hpux_sigtramp_unwind_sniffer): New signal trampoline unwinder.
    (hppa_hpux_init_abi): Register sigtramp unwinder.

Index: hppa-hpux-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/hppa-hpux-tdep.c,v
retrieving revision 1.18
diff -u -p -r1.18 hppa-hpux-tdep.c
--- hppa-hpux-tdep.c	7 May 2004 05:48:48 -0000	1.18
+++ hppa-hpux-tdep.c	20 May 2004 03:45:07 -0000
@@ -24,15 +24,21 @@ Foundation, Inc., 59 Temple Place - Suit
 #include "osabi.h"
 #include "gdb_string.h"
 #include "frame.h"
+#include "frame-unwind.h"
+#include "trad-frame.h"
 #include "symtab.h"
 #include "objfiles.h"
 #include "inferior.h"
 #include "infcall.h"
 #include "hppa-tdep.h"
 
 #include <dl.h>
 #include <machine/save_state.h>
 
+#ifndef offsetof
+#define offsetof(TYPE, MEMBER) ((unsigned long) &((TYPE *)0)->MEMBER)
+#endif
+
 /* Forward declarations.  */
 extern void _initialize_hppa_hpux_tdep (void);
 extern initialize_file_ftype _initialize_hppa_hpux_tdep;
@@ -45,125 +52,6 @@ typedef struct
   }
 args_for_find_stub;
 
-/* FIXME: brobecker 2002-12-25.  The following functions will eventually
-   become static, after the multiarching conversion is done.  */
-int hppa_hpux_pc_in_sigtramp (CORE_ADDR pc, char *name);
-void hppa32_hpux_frame_saved_pc_in_sigtramp (struct frame_info *fi,
-                                             CORE_ADDR *tmp);
-void hppa32_hpux_frame_base_before_sigtramp (struct frame_info *fi,
-                                             CORE_ADDR *tmp);
-void hppa32_hpux_frame_find_saved_regs_in_sigtramp (struct frame_info *fi,
-                                                    CORE_ADDR *fsr);
-void hppa64_hpux_frame_saved_pc_in_sigtramp (struct frame_info *fi,
-                                             CORE_ADDR *tmp);
-void hppa64_hpux_frame_base_before_sigtramp (struct frame_info *fi,
-                                             CORE_ADDR *tmp);
-void hppa64_hpux_frame_find_saved_regs_in_sigtramp (struct frame_info *fi,
-                                                    CORE_ADDR *fsr);
-
-int
-hppa_hpux_pc_in_sigtramp (CORE_ADDR pc, char *name)
-{
-  /* Actually, for a PA running HPUX the kernel calls the signal handler
-     without an intermediate trampoline.  Luckily the kernel always sets
-     the return pointer for the signal handler to point to _sigreturn.  */
-  return (name && (strcmp ("_sigreturn", name) == 0));
-}
-
-/* For hppa32_hpux_frame_saved_pc_in_sigtramp, 
-   hppa32_hpux_frame_base_before_sigtramp and
-   hppa32_hpux_frame_find_saved_regs_in_sigtramp:
-
-   The signal context structure pointer is always saved at the base
-   of the frame which "calls" the signal handler.  We only want to find
-   the hardware save state structure, which lives 10 32bit words into
-   sigcontext structure.
-
-   Within the hardware save state structure, registers are found in the
-   same order as the register numbers in GDB.
-
-   At one time we peeked at %r31 rather than the PC queues to determine
-   what instruction took the fault.  This was done on purpose, but I don't
-   remember why.  Looking at the PC queues is really the right way, and
-   I don't remember why that didn't work when this code was originally
-   written.  */
-
-void
-hppa32_hpux_frame_saved_pc_in_sigtramp (struct frame_info *fi, CORE_ADDR *tmp)
-{
-  *tmp = read_memory_integer (get_frame_base (fi) + (43 * 4), 4);
-}
-
-void
-hppa32_hpux_frame_base_before_sigtramp (struct frame_info *fi,
-                                        CORE_ADDR *tmp)
-{
-  *tmp = read_memory_integer (get_frame_base (fi) + (40 * 4), 4);
-}
-
-void
-hppa32_hpux_frame_find_saved_regs_in_sigtramp (struct frame_info *fi,
-					       CORE_ADDR *fsr)
-{
-  int i;
-  const CORE_ADDR tmp = get_frame_base (fi) + (10 * 4);
-
-  for (i = 0; i < NUM_REGS; i++)
-    {
-      if (i == HPPA_SP_REGNUM)
-	fsr[HPPA_SP_REGNUM] = read_memory_integer (tmp + HPPA_SP_REGNUM * 4, 4);
-      else
-	fsr[i] = tmp + i * 4;
-    }
-}
-
-/* For hppa64_hpux_frame_saved_pc_in_sigtramp, 
-   hppa64_hpux_frame_base_before_sigtramp and
-   hppa64_hpux_frame_find_saved_regs_in_sigtramp:
-
-   These functions are the PA64 ABI equivalents of the 32bits counterparts
-   above. See the comments there.
-
-   For PA64, the save_state structure is at an offset of 24 32-bit words
-   from the sigcontext structure. The 64 bit general registers are at an
-   offset of 640 bytes from the beginning of the save_state structure,
-   and the floating pointer register are at an offset of 256 bytes from
-   the beginning of the save_state structure.  */
-
-void
-hppa64_hpux_frame_saved_pc_in_sigtramp (struct frame_info *fi, CORE_ADDR *tmp)
-{
-  *tmp = read_memory_integer
-           (get_frame_base (fi) + (24 * 4) + 640 + (33 * 8), 8);
-}
-
-void
-hppa64_hpux_frame_base_before_sigtramp (struct frame_info *fi,
-                                        CORE_ADDR *tmp)
-{
-  *tmp = read_memory_integer
-           (get_frame_base (fi) + (24 * 4) + 640 + (30 * 8), 8);
-}
-
-void
-hppa64_hpux_frame_find_saved_regs_in_sigtramp (struct frame_info *fi,
-					       CORE_ADDR *fsr)
-{
-  int i;
-  const CORE_ADDR tmp1 = get_frame_base (fi) + (24 * 4) + 640;
-  const CORE_ADDR tmp2 = get_frame_base (fi) + (24 * 4) + 256;
-
-  for (i = 0; i < NUM_REGS; i++)
-    {
-      if (i == HPPA_SP_REGNUM)
-        fsr[HPPA_SP_REGNUM] = read_memory_integer (tmp1 + HPPA_SP_REGNUM * 8, 8);
-      else if (i >= HPPA_FP0_REGNUM)
-        fsr[i] = tmp2 + (i - HPPA_FP0_REGNUM) * 8;
-      else
-        fsr[i] = tmp1 + i * 8;
-    }
-}
-
 /* Return one if PC is in the call path of a trampoline, else return zero.
 
    Note we return one for *any* call trampoline (long-call, arg-reloc), not
@@ -1222,6 +1060,202 @@ child_get_current_exception_event (void)
   return &current_ex_event;
 }
 
+/* Signal frames.  */
+
+static CORE_ADDR
+hppa_hpux_sigtramp_find_sigcontext (CORE_ADDR sp, CORE_ADDR pc)
+{
+  unsigned int insns[5];
+
+  read_memory_nobpt (pc, (char *)insns, sizeof(insns));
+
+  /* This comes from _sigreturn:
+     379a0000 ldo 0(%r28), %r26
+     6bd33fc9 stw %r19, -28(%sp)
+     20200801 ldil -40000000, %r1
+     e420e008 be,l 4(%sr7, %rr1), %sr0, %r31
+     34160116 ldi 139, %r22  */
+
+  if (extract_unsigned_integer (&insns[0], 4) != 0x379a0000
+      || extract_unsigned_integer (&insns[1], 4) != 0x6bd33fc9
+      || extract_unsigned_integer (&insns[2], 4) != 0x20200801
+      || extract_unsigned_integer (&insns[3], 4) != 0xe420e008
+      || extract_unsigned_integer (&insns[4], 4) != 0x34160116)
+    return 0;
+
+  /* We can find a ucontext_t 1352 bytes above the stack. The first
+     element of the ucontext_t is a mcontext_t (save_state_t), which
+     contains the registers we want.  */
+  return sp - 1352;
+}
+
+struct hppa_hpux_sigtramp_unwind_cache
+{
+  CORE_ADDR base;
+  struct trad_frame_saved_reg *saved_regs;
+};
+
+static struct hppa_hpux_sigtramp_unwind_cache *
+hppa_hpux_sigtramp_frame_unwind_cache (struct frame_info *next_frame,
+					void **this_cache)
+{
+  struct gdbarch *gdbarch = get_frame_arch (next_frame);
+  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
+  struct hppa_hpux_sigtramp_unwind_cache *info;
+  unsigned int flag;
+  CORE_ADDR sp, pc, scptr;
+  int i, incr, off, szoff;
+
+  if (*this_cache)
+    return *this_cache;
+
+  info = FRAME_OBSTACK_ZALLOC (struct hppa_hpux_sigtramp_unwind_cache);
+  *this_cache = info;
+  info->saved_regs = trad_frame_alloc_saved_regs (next_frame);
+
+  sp = frame_unwind_register_unsigned (next_frame, HPPA_SP_REGNUM);
+  pc = frame_pc_unwind (next_frame);
+  scptr = hppa_hpux_sigtramp_find_sigcontext (sp, pc);
+
+  /* See /usr/include/machine/save_state.h for the structure of the save_state_t
+     structure. */
+  
+  off = scptr;
+  flag = read_memory_unsigned_integer(scptr, 4);
+    
+  if (!(flag & 0x40))
+    {
+      /* Narrow registers. */
+      off = scptr + offsetof (save_state_t, ss_narrow);
+      incr = 4;
+      szoff = 0;
+    }
+  else
+    {
+      /* Wide registers. */
+      off = scptr + offsetof (save_state_t, ss_wide) + 8;
+      incr = 8;
+      szoff = (tdep->bytes_per_address == 4 ? 4 : 0);
+    }
+
+  for (i = 1; i < 32; i++)
+    {
+      info->saved_regs[HPPA_R0_REGNUM + i].addr = off + szoff;
+      off += incr;
+    }
+
+  info->saved_regs[HPPA_SAR_REGNUM].addr = off + szoff;
+  off += incr;
+
+  info->saved_regs[HPPA_PCOQ_HEAD_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_PCSQ_HEAD_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_PCOQ_TAIL_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_PCSQ_TAIL_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_EIEM_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_IIR_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_ISR_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_IOR_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_IPSW_REGNUM].addr = off + szoff;
+  off += incr;
+
+  /* ignore ss_cpustate */
+  off += incr;
+
+  info->saved_regs[HPPA_SR4_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_SR4_REGNUM+1].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_SR4_REGNUM+2].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_SR4_REGNUM+3].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_SR4_REGNUM+4].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_SR4_REGNUM+5].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_SR4_REGNUM+6].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_SR4_REGNUM+7].addr = off + szoff;
+  off += incr;
+
+  info->saved_regs[HPPA_RCR_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_PID0_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_PID1_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_CCR_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_PID2_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_PID3_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_TR0_REGNUM].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_TR0_REGNUM + 1].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_TR0_REGNUM + 2].addr = off + szoff;
+  off += incr;
+  info->saved_regs[HPPA_CR27_REGNUM].addr = off + szoff;
+  off += incr;
+
+  /* TODO: fp regs */
+
+  info->base = frame_unwind_register_unsigned (next_frame, HPPA_SP_REGNUM);
+
+  return info;
+}
+
+static void
+hppa_hpux_sigtramp_frame_this_id (struct frame_info *next_frame,
+				   void **this_prologue_cache,
+				   struct frame_id *this_id)
+{
+  struct hppa_hpux_sigtramp_unwind_cache *info
+    = hppa_hpux_sigtramp_frame_unwind_cache (next_frame, this_prologue_cache);
+  *this_id = frame_id_build (info->base, frame_pc_unwind (next_frame));
+}
+
+static void
+hppa_hpux_sigtramp_frame_prev_register (struct frame_info *next_frame,
+					 void **this_prologue_cache,
+					 int regnum, int *optimizedp,
+					 enum lval_type *lvalp, 
+					 CORE_ADDR *addrp,
+					 int *realnump, void *valuep)
+{
+  struct hppa_hpux_sigtramp_unwind_cache *info
+    = hppa_hpux_sigtramp_frame_unwind_cache (next_frame, this_prologue_cache);
+  hppa_frame_prev_register_helper (next_frame, info->saved_regs, regnum,
+		                   optimizedp, lvalp, addrp, realnump, valuep);
+}
+
+static const struct frame_unwind hppa_hpux_sigtramp_frame_unwind = {
+  SIGTRAMP_FRAME,
+  hppa_hpux_sigtramp_frame_this_id,
+  hppa_hpux_sigtramp_frame_prev_register
+};
+
+static const struct frame_unwind *
+hppa_hpux_sigtramp_unwind_sniffer (struct frame_info *next_frame)
+{
+  CORE_ADDR sp = frame_unwind_register_unsigned (next_frame, HPPA_SP_REGNUM);
+  CORE_ADDR pc = frame_pc_unwind (next_frame);
+
+  if (hppa_hpux_sigtramp_find_sigcontext (sp, pc))
+    return &hppa_hpux_sigtramp_frame_unwind;
+
+  return NULL;
+}
+
 static void
 hppa_hpux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
 {
@@ -1237,6 +1277,8 @@ hppa_hpux_init_abi (struct gdbarch_info 
   set_gdbarch_in_solib_return_trampoline (gdbarch,
 					  hppa_hpux_in_solib_return_trampoline);
   set_gdbarch_skip_trampoline_code (gdbarch, hppa_hpux_skip_trampoline_code);
+
+  frame_unwind_append_sniffer (gdbarch, hppa_hpux_sigtramp_unwind_sniffer);
 }
 
 static void
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Joel Brobecker]

Hello Randolph,

> I tested this against hppa2.0w-hp-hpux11.11 and hppa2.0w-hp-hpux11.00,
> but I don't have a hpux machine running a 32-bit kernel to test. If
> anyone is interested in that target, please give it a try. It *should*
> work :)

How can you tell whether a kernel is 32bits or not?

I tried your patch on our machines, it we get noticeably better results,
especially in sigstep.exp, which no longer times out to death....

> 2004-05-19  Randolph Chung  <tausq@debian.org>
> 
>     * hppa-hpux-tdep.c (offsetof): Define.
>     (hppa_hpux_pc_in_sigtramp, hppa32_hpux_frame_saved_pc_in_sigtramp)
>     (hppa32_hpux_frame_base_before_sigtramp)
>     (hppa32_hpux_frame_find_saved_regs_in_sigtramp)
>     (hppa64_hpux_frame_saved_pc_in_sigtramp)
>     (hppa64_hpux_frame_base_before_sigtramp)
>     (hppa64_hpux_frame_find_saved_regs_in_sigtramp): Remove unused functions.
>     (hppa_hpux_sigtramp_find_sigcontext, hppa_hpux_sigtramp_unwind_cache)
>     (hppa_hpux_sigtramp_frame_unwind_cache, hppa_hpux_sigtramp_frame_this_id)
>     (hppa_hpux_sigtramp_frame_prev_register, hppa_hpux_sigtramp_frame_unwind)
>     (hppa_hpux_sigtramp_unwind_sniffer): New signal trampoline unwinder.
>     (hppa_hpux_init_abi): Register sigtramp unwinder.

-- 
Joel

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Randolph Chung]

> How can you tell whether a kernel is 32bits or not?

in the code? In the save_state_t structure returned by the kernel,
there's a flag, which is checked here:

+  if (!(flag & 0x40))
+    {
+      /* Narrow registers. */
+      off = scptr + offsetof (save_state_t, ss_narrow);
+      incr = 4;
+      szoff = 0;
+    }
+  else
+    {
+      /* Wide registers. */
+      off = scptr + offsetof (save_state_t, ss_wide) + 8;
+      incr = 8;
+      szoff = (tdep->bytes_per_address == 4 ? 4 : 0);
+    }

There are 3 cases, if you have a 32-bit kernel, then it should always
hit the first if case. If you have a 64-bit kernel and 32-bit userspace,
you should hit the second case with szoff == 4. For 64-bit kernel and
64-bit userspace you will hit the else case with szoff == 0.

I've only tested the 2nd scenario on hpux11.00 and hpux11.11. Too many
targets and combinations :-(

> I tried your patch on our machines, it we get noticeably better results,
> especially in sigstep.exp, which no longer times out to death....

yeah, it used to get into an infinite loop.. not very nice :)

randolph
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Joel Brobecker]

On Thu, May 20, 2004 at 01:46:25PM -0700, Joel Brobecker wrote:
> > I tested this against hppa2.0w-hp-hpux11.11 and hppa2.0w-hp-hpux11.00,
> > but I don't have a hpux machine running a 32-bit kernel to test. If
> > anyone is interested in that target, please give it a try. It *should*
> > work :)
> 
> How can you tell whether a kernel is 32bits or not?
> 
> I tried your patch on our machines, it we get noticeably better results,
> especially in sigstep.exp, which no longer times out to death....

Hmmm, using a small trace in the code you pointed out in another
message, I found out that we are also running a 64bit kernel...
So my testing didn't bring anything new, I'm afraid.

> > 2004-05-19  Randolph Chung  <tausq@debian.org>
> > 
> >     * hppa-hpux-tdep.c (offsetof): Define.
> >     (hppa_hpux_pc_in_sigtramp, hppa32_hpux_frame_saved_pc_in_sigtramp)
> >     (hppa32_hpux_frame_base_before_sigtramp)
> >     (hppa32_hpux_frame_find_saved_regs_in_sigtramp)
> >     (hppa64_hpux_frame_saved_pc_in_sigtramp)
> >     (hppa64_hpux_frame_base_before_sigtramp)
> >     (hppa64_hpux_frame_find_saved_regs_in_sigtramp): Remove unused functions.
> >     (hppa_hpux_sigtramp_find_sigcontext, hppa_hpux_sigtramp_unwind_cache)
> >     (hppa_hpux_sigtramp_frame_unwind_cache, hppa_hpux_sigtramp_frame_this_id)
> >     (hppa_hpux_sigtramp_frame_prev_register, hppa_hpux_sigtramp_frame_unwind)
> >     (hppa_hpux_sigtramp_unwind_sniffer): New signal trampoline unwinder.
> >     (hppa_hpux_init_abi): Register sigtramp unwinder.

-- 
Joel

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Joel Brobecker]

> How can you tell whether a kernel is 32bits or not?

Just FYI, I just found the answer to my question:

        % getconf KERNEL_BITS
        Will tell you which kernel you are running (32/64).

Check the getconf and sysconf man pages... We run a 64bit kernel.

-- 
Joel

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Andrew Cagney]

+/* Signal frames.  */
+
+static CORE_ADDR
+hppa_hpux_sigtramp_find_sigcontext (CORE_ADDR sp, CORE_ADDR pc)
+{
+  unsigned int insns[5];
+
+  read_memory_nobpt (pc, (char *)insns, sizeof(insns));
+
+  /* This comes from _sigreturn:
+     379a0000 ldo 0(%r28), %r26
+     6bd33fc9 stw %r19, -28(%sp)
+     20200801 ldil -40000000, %r1
+     e420e008 be,l 4(%sr7, %rr1), %sr0, %r31
+     34160116 ldi 139, %r22  */
+
+  if (extract_unsigned_integer (&insns[0], 4) != 0x379a0000
+      || extract_unsigned_integer (&insns[1], 4) != 0x6bd33fc9
+      || extract_unsigned_integer (&insns[2], 4) != 0x20200801
+      || extract_unsigned_integer (&insns[3], 4) != 0xe420e008
+      || extract_unsigned_integer (&insns[4], 4) != 0x34160116)
+    return 0;
+
+  /* We can find a ucontext_t 1352 bytes above the stack. The first
+     element of the ucontext_t is a mcontext_t (save_state_t), which
+     contains the registers we want.  */
+  return sp - 1352;
+}
Can this make use of tramp-frame for the pattern match?  For single 
stepping through trampolines to work, this code needs to be able to 
match the tramp when part way through it (what sigstep is testing).

Andrew

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Randolph Chung]

> Can this make use of tramp-frame for the pattern match?  For single 
> stepping through trampolines to work, this code needs to be able to 
> match the tramp when part way through it (what sigstep is testing).

hmm.. i think so, let me try to rewrite it.

randolph
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Randolph Chung]

> Can this make use of tramp-frame for the pattern match?  For single 
> stepping through trampolines to work, this code needs to be able to 
> match the tramp when part way through it (what sigstep is testing).

As it turns out, no, not easily....

The signal trampoline on hpux actually has a name (_sigreturn), so it
fails the check in tramp_frame_sniffer ()

:-(

randolph
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Randolph Chung]

> The signal trampoline on hpux actually has a name (_sigreturn), so it
> fails the check in tramp_frame_sniffer ()

is this more efficient/correct? It doesn't do any code reading just
checks the pc against "_sigreturn".

randolph

2004-05-22  Randolph Chung  <tausq@debian.org>

	* hppa-hpux-tdep.c (offsetof): Define.
	(hppa_hpux_pc_in_sigtramp, hppa32_hpux_frame_saved_pc_in_sigtramp)
	(hppa32_hpux_frame_base_before_sigtramp)
	(hppa32_hpux_frame_find_saved_regs_in_sigtramp)
	(hppa64_hpux_frame_saved_pc_in_sigtramp)
	(hppa64_hpux_frame_base_before_sigtramp)
	(hppa64_hpux_frame_find_saved_regs_in_sigtramp): Remove unused 
	functions.
	(struct hppa_hpux_sigtramp_unwind_cache, hppa_hpux_tramp_reg)
	(hppa_hpux_sigtramp_frame_unwind_cache)
	(hppa_hpux_sigtramp_frame_this_id)
	(hppa_hpux_sigtramp_frame_prev_register)
	(hppa_hpux_sigtramp_frame_unwind)
	(hppa_hpux_sigtramp_unwind_sniffer): New signal trampoline unwinder.
	(hppa_hpux_init_abi): Register sigtramp unwinder.
	* Makefile.in (hppa-hpux-tdep.o): Update dependencies.

Index: hppa-hpux-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/hppa-hpux-tdep.c,v
retrieving revision 1.18
diff -u -p -r1.18 hppa-hpux-tdep.c
--- hppa-hpux-tdep.c	7 May 2004 05:48:48 -0000	1.18
+++ hppa-hpux-tdep.c	23 May 2004 00:21:43 -0000
@@ -24,15 +24,21 @@ Foundation, Inc., 59 Temple Place - Suit
 #include "osabi.h"
 #include "gdb_string.h"
 #include "frame.h"
+#include "frame-unwind.h"
+#include "trad-frame.h"
 #include "symtab.h"
 #include "objfiles.h"
 #include "inferior.h"
 #include "infcall.h"
 #include "hppa-tdep.h"
 
 #include <dl.h>
 #include <machine/save_state.h>
 
+#ifndef offsetof
+#define offsetof(TYPE, MEMBER) ((unsigned long) &((TYPE *)0)->MEMBER)
+#endif
+
 /* Forward declarations.  */
 extern void _initialize_hppa_hpux_tdep (void);
 extern initialize_file_ftype _initialize_hppa_hpux_tdep;
@@ -45,125 +52,6 @@ typedef struct
   }
 args_for_find_stub;
 
-/* FIXME: brobecker 2002-12-25.  The following functions will eventually
-   become static, after the multiarching conversion is done.  */
-int hppa_hpux_pc_in_sigtramp (CORE_ADDR pc, char *name);
-void hppa32_hpux_frame_saved_pc_in_sigtramp (struct frame_info *fi,
-                                             CORE_ADDR *tmp);
-void hppa32_hpux_frame_base_before_sigtramp (struct frame_info *fi,
-                                             CORE_ADDR *tmp);
-void hppa32_hpux_frame_find_saved_regs_in_sigtramp (struct frame_info *fi,
-                                                    CORE_ADDR *fsr);
-void hppa64_hpux_frame_saved_pc_in_sigtramp (struct frame_info *fi,
-                                             CORE_ADDR *tmp);
-void hppa64_hpux_frame_base_before_sigtramp (struct frame_info *fi,
-                                             CORE_ADDR *tmp);
-void hppa64_hpux_frame_find_saved_regs_in_sigtramp (struct frame_info *fi,
-                                                    CORE_ADDR *fsr);
-
-int
-hppa_hpux_pc_in_sigtramp (CORE_ADDR pc, char *name)
-{
-  /* Actually, for a PA running HPUX the kernel calls the signal handler
-     without an intermediate trampoline.  Luckily the kernel always sets
-     the return pointer for the signal handler to point to _sigreturn.  */
-  return (name && (strcmp ("_sigreturn", name) == 0));
-}
-
-/* For hppa32_hpux_frame_saved_pc_in_sigtramp, 
-   hppa32_hpux_frame_base_before_sigtramp and
-   hppa32_hpux_frame_find_saved_regs_in_sigtramp:
-
-   The signal context structure pointer is always saved at the base
-   of the frame which "calls" the signal handler.  We only want to find
-   the hardware save state structure, which lives 10 32bit words into
-   sigcontext structure.
-
-   Within the hardware save state structure, registers are found in the
-   same order as the register numbers in GDB.
-
-   At one time we peeked at %r31 rather than the PC queues to determine
-   what instruction took the fault.  This was done on purpose, but I don't
-   remember why.  Looking at the PC queues is really the right way, and
-   I don't remember why that didn't work when this code was originally
-   written.  */
-
-void
-hppa32_hpux_frame_saved_pc_in_sigtramp (struct frame_info *fi, CORE_ADDR *tmp)
-{
-  *tmp = read_memory_integer (get_frame_base (fi) + (43 * 4), 4);
-}
-
-void
-hppa32_hpux_frame_base_before_sigtramp (struct frame_info *fi,
-                                        CORE_ADDR *tmp)
-{
-  *tmp = read_memory_integer (get_frame_base (fi) + (40 * 4), 4);
-}
-
-void
-hppa32_hpux_frame_find_saved_regs_in_sigtramp (struct frame_info *fi,
-					       CORE_ADDR *fsr)
-{
-  int i;
-  const CORE_ADDR tmp = get_frame_base (fi) + (10 * 4);
-
-  for (i = 0; i < NUM_REGS; i++)
-    {
-      if (i == HPPA_SP_REGNUM)
-	fsr[HPPA_SP_REGNUM] = read_memory_integer (tmp + HPPA_SP_REGNUM * 4, 4);
-      else
-	fsr[i] = tmp + i * 4;
-    }
-}
-
-/* For hppa64_hpux_frame_saved_pc_in_sigtramp, 
-   hppa64_hpux_frame_base_before_sigtramp and
-   hppa64_hpux_frame_find_saved_regs_in_sigtramp:
-
-   These functions are the PA64 ABI equivalents of the 32bits counterparts
-   above. See the comments there.
-
-   For PA64, the save_state structure is at an offset of 24 32-bit words
-   from the sigcontext structure. The 64 bit general registers are at an
-   offset of 640 bytes from the beginning of the save_state structure,
-   and the floating pointer register are at an offset of 256 bytes from
-   the beginning of the save_state structure.  */
-
-void
-hppa64_hpux_frame_saved_pc_in_sigtramp (struct frame_info *fi, CORE_ADDR *tmp)
-{
-  *tmp = read_memory_integer
-           (get_frame_base (fi) + (24 * 4) + 640 + (33 * 8), 8);
-}
-
-void
-hppa64_hpux_frame_base_before_sigtramp (struct frame_info *fi,
-                                        CORE_ADDR *tmp)
-{
-  *tmp = read_memory_integer
-           (get_frame_base (fi) + (24 * 4) + 640 + (30 * 8), 8);
-}
-
-void
-hppa64_hpux_frame_find_saved_regs_in_sigtramp (struct frame_info *fi,
-					       CORE_ADDR *fsr)
-{
-  int i;
-  const CORE_ADDR tmp1 = get_frame_base (fi) + (24 * 4) + 640;
-  const CORE_ADDR tmp2 = get_frame_base (fi) + (24 * 4) + 256;
-
-  for (i = 0; i < NUM_REGS; i++)
-    {
-      if (i == HPPA_SP_REGNUM)
-        fsr[HPPA_SP_REGNUM] = read_memory_integer (tmp1 + HPPA_SP_REGNUM * 8, 8);
-      else if (i >= HPPA_FP0_REGNUM)
-        fsr[i] = tmp2 + (i - HPPA_FP0_REGNUM) * 8;
-      else
-        fsr[i] = tmp1 + i * 8;
-    }
-}
-
 /* Return one if PC is in the call path of a trampoline, else return zero.
 
    Note we return one for *any* call trampoline (long-call, arg-reloc), not
@@ -1222,6 +1060,155 @@ child_get_current_exception_event (void)
   return &current_ex_event;
 }
 
+/* Signal frames.  */
+struct hppa_hpux_sigtramp_unwind_cache
+{
+  CORE_ADDR base;
+  struct trad_frame_saved_reg *saved_regs;
+};
+
+static int hppa_hpux_tramp_reg[] = {
+  HPPA_SAR_REGNUM,
+  HPPA_PCOQ_HEAD_REGNUM,
+  HPPA_PCSQ_HEAD_REGNUM,
+  HPPA_PCOQ_TAIL_REGNUM,
+  HPPA_PCSQ_TAIL_REGNUM,
+  HPPA_EIEM_REGNUM,
+  HPPA_IIR_REGNUM,
+  HPPA_ISR_REGNUM,
+  HPPA_IOR_REGNUM,
+  HPPA_IPSW_REGNUM,
+  -1,
+  HPPA_SR4_REGNUM,
+  HPPA_SR4_REGNUM + 1,
+  HPPA_SR4_REGNUM + 2,
+  HPPA_SR4_REGNUM + 3,
+  HPPA_SR4_REGNUM + 4,
+  HPPA_SR4_REGNUM + 5,
+  HPPA_SR4_REGNUM + 6,
+  HPPA_SR4_REGNUM + 7,
+  HPPA_RCR_REGNUM,
+  HPPA_PID0_REGNUM,
+  HPPA_PID1_REGNUM,
+  HPPA_CCR_REGNUM,
+  HPPA_PID2_REGNUM,
+  HPPA_PID3_REGNUM,
+  HPPA_TR0_REGNUM,
+  HPPA_TR0_REGNUM + 1,
+  HPPA_TR0_REGNUM + 2,
+  HPPA_CR27_REGNUM
+};
+
+static struct hppa_hpux_sigtramp_unwind_cache *
+hppa_hpux_sigtramp_frame_unwind_cache (struct frame_info *next_frame,
+				       void **this_cache)
+
+{
+  struct gdbarch *gdbarch = get_frame_arch (next_frame);
+  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
+  struct hppa_hpux_sigtramp_unwind_cache *info;
+  unsigned int flag;
+  CORE_ADDR sp, scptr;
+  int i, incr, off, szoff;
+
+  if (*this_cache)
+    return *this_cache;
+
+  info = FRAME_OBSTACK_ZALLOC (struct hppa_hpux_sigtramp_unwind_cache);
+  *this_cache = info;
+  info->saved_regs = trad_frame_alloc_saved_regs (next_frame);
+
+  sp = frame_unwind_register_unsigned (next_frame, HPPA_SP_REGNUM);
+
+  scptr = sp - 1352;
+  off = scptr;
+
+  /* See /usr/include/machine/save_state.h for the structure of the save_state_t
+     structure. */
+  
+  flag = read_memory_unsigned_integer(scptr, 4);
+    
+  if (!(flag & 0x40))
+    {
+      /* Narrow registers. */
+      off = scptr + offsetof (save_state_t, ss_narrow);
+      incr = 4;
+      szoff = 0;
+    }
+  else
+    {
+      /* Wide registers. */
+      off = scptr + offsetof (save_state_t, ss_wide) + 8;
+      incr = 8;
+      szoff = (tdep->bytes_per_address == 4 ? 4 : 0);
+    }
+
+  for (i = 1; i < 32; i++)
+    {
+      info->saved_regs[HPPA_R0_REGNUM + i].addr = off + szoff;
+      off += incr;
+    }
+
+  for (i = 0; 
+       i < sizeof(hppa_hpux_tramp_reg) / sizeof(hppa_hpux_tramp_reg[0]);
+       i++)
+    {
+      if (hppa_hpux_tramp_reg[i] > 0)
+        info->saved_regs[hppa_hpux_tramp_reg[i]].addr = off + szoff;
+      off += incr;
+    }
+
+  /* TODO: fp regs */
+
+  info->base = frame_unwind_register_unsigned (next_frame, HPPA_SP_REGNUM);
+
+  return info;
+}
+
+static void
+hppa_hpux_sigtramp_frame_this_id (struct frame_info *next_frame,
+				   void **this_prologue_cache,
+				   struct frame_id *this_id)
+{
+  struct hppa_hpux_sigtramp_unwind_cache *info
+    = hppa_hpux_sigtramp_frame_unwind_cache (next_frame, this_prologue_cache);
+  *this_id = frame_id_build (info->base, frame_pc_unwind (next_frame));
+}
+
+static void
+hppa_hpux_sigtramp_frame_prev_register (struct frame_info *next_frame,
+					 void **this_prologue_cache,
+					 int regnum, int *optimizedp,
+					 enum lval_type *lvalp, 
+					 CORE_ADDR *addrp,
+					 int *realnump, void *valuep)
+{
+  struct hppa_hpux_sigtramp_unwind_cache *info
+    = hppa_hpux_sigtramp_frame_unwind_cache (next_frame, this_prologue_cache);
+  hppa_frame_prev_register_helper (next_frame, info->saved_regs, regnum,
+		                   optimizedp, lvalp, addrp, realnump, valuep);
+}
+
+static const struct frame_unwind hppa_hpux_sigtramp_frame_unwind = {
+  SIGTRAMP_FRAME,
+  hppa_hpux_sigtramp_frame_this_id,
+  hppa_hpux_sigtramp_frame_prev_register
+};
+
+static const struct frame_unwind *
+hppa_hpux_sigtramp_unwind_sniffer (struct frame_info *next_frame)
+{
+  CORE_ADDR pc = frame_pc_unwind (next_frame);
+  char *name;
+
+  find_pc_partial_function (pc, &name, NULL, NULL);
+
+  if (name && strcmp(name, "_sigreturn") == 0)
+    return &hppa_hpux_sigtramp_frame_unwind;
+
+  return NULL;
+}
+
 static void
 hppa_hpux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
 {
@@ -1237,6 +1224,8 @@ hppa_hpux_init_abi (struct gdbarch_info 
   set_gdbarch_in_solib_return_trampoline (gdbarch,
 					  hppa_hpux_in_solib_return_trampoline);
   set_gdbarch_skip_trampoline_code (gdbarch, hppa_hpux_skip_trampoline_code);
+
+  frame_unwind_append_sniffer (gdbarch, hppa_hpux_sigtramp_unwind_sniffer);
 }
 
 static void
Index: Makefile.in
===================================================================
RCS file: /cvs/src/src/gdb/Makefile.in,v
retrieving revision 1.572
diff -u -p -r1.572 Makefile.in
--- Makefile.in	17 May 2004 15:16:39 -0000	1.572
+++ Makefile.in	23 May 2004 00:21:51 -0000
@@ -1829,7 +1829,10 @@ hppah-nat.o: hppah-nat.c $(defs_h) $(inf
 	$(gdb_wait_h) $(regcache_h) $(gdb_string_h) $(infttrace_h) \
 	$(hppa_tdep_h)
 hppa-hpux-tdep.o: hppa-hpux-tdep.c $(defs_h) $(arch_utils_h) $(gdbcore_h) \
-	$(osabi_h) $(gdb_string_h) $(frame_h) $(hppa_tdep_h)
+	$(osabi_h) $(gdb_string_h) $(frame_h) $(frame_unwind_h) \
+	$(trad_frame_h) $(symtab_h) $(objfiles_h) $(inferior_h) $(infcall_h) \
+	$(hppa_tdep_h)
 hppa-tdep.o: hppa-tdep.c $(defs_h) $(frame_h) $(bfd_h) $(inferior_h) \
 	$(regcache_h) $(completer_h) $(osabi_h) $(gdb_assert_h) \
 	$(arch_utils_h) $(symtab_h) $(dis_asm_h) $(trad_frame_h) \

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Andrew Cagney]

The signal trampoline on hpux actually has a name (_sigreturn), so it
fails the check in tramp_frame_sniffer ()


is this more efficient/correct? It doesn't do any code reading just
checks the pc against "_sigreturn".
Yes, definitly.  Having to do instruction examination should be 
considered a last resource.

Andrew

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Randolph Chung]

> Yes, definitly.  Having to do instruction examination should be 
> considered a last resource.

ok, this is checked in now too, but can someone please comment on this
bit:

================================================================
Unfortunately this still fails a lot of the signal related testcases.
It is not because of a problem with the signal unwinder though:

On HPUX-SOM, there is an export stub inserted into the call sequence in
some cases; the export stub is hit on the return path from indirect
function calls, and apparently on return from a signal handler. This 
creates an extra frame in the backtrace, viz:

(gdb) bt
#0  0x000029e4 in handle_USR1 (sig=0)
    at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:18
#1  0x000029c8 in handle_USR1 ()
    at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:17
#2  <signal handler called>
#3  main () at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:40

Frame 1 comes from the export stub. Because of this, the pattern
matching in the backtrace test fails.  I am not sure if this is a
problem with the test script or with the handling of export stubs -- am
I supposed to somehow supress the export stub from showing up in the
backtrace? (if so, how?)
================================================================

thanks
randolph
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[Daniel Jacobowitz]

On Tue, May 25, 2004 at 10:36:09PM -0700, Randolph Chung wrote:
> > Yes, definitly.  Having to do instruction examination should be 
> > considered a last resource.
> 
> ok, this is checked in now too, but can someone please comment on this
> bit:
> 
> ================================================================
> Unfortunately this still fails a lot of the signal related testcases.
> It is not because of a problem with the signal unwinder though:
> 
> On HPUX-SOM, there is an export stub inserted into the call sequence in
> some cases; the export stub is hit on the return path from indirect
> function calls, and apparently on return from a signal handler. This 
> creates an extra frame in the backtrace, viz:
> 
> (gdb) bt
> #0  0x000029e4 in handle_USR1 (sig=0)
>     at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:18
> #1  0x000029c8 in handle_USR1 ()
>     at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:17
> #2  <signal handler called>
> #3  main () at ../../../gdb/gdb/testsuite/gdb.base/annota1.c:40
> 
> Frame 1 comes from the export stub. Because of this, the pattern
> matching in the backtrace test fails.  I am not sure if this is a
> problem with the test script or with the handling of export stubs -- am
> I supposed to somehow supress the export stub from showing up in the
> backtrace? (if so, how?)

Right now, we don't have any mechanism to hide frames.  I'm not sure
whether it's a desirable feature or not.  I've wanted it for C++
occasionally.

I wonder if GDB should have a single knob to switch between
"simplified" and more thorough/accurate display modes?  Could any of
the existing toggles be added to this?

-- 
Daniel Jacobowitz

Re: [patch/rfa] Signal trampoline unwinder for hppa-hpux

[John David Anglin]

> > How can you tell whether a kernel is 32bits or not?
> 
> Just FYI, I just found the answer to my question:
> 
>         % getconf KERNEL_BITS
> 	Will tell you which kernel you are running (32/64).
> 
> 	Check the getconf and sysconf man pages... We run a 64bit kernel.

This only works under hpux 11.  See config.guess.  It sets the machine
string to the following:

	hppa2.0n	32-bit host kernel
	hppa2.0w	64-bit host kernel
	hppa2.0		HP-UX 10.20 host kernel

I believe that 10.20 only supported narrow PA 2.0 kernels.  So, I am
not sure that the distinction between hppa2.0 and hppa2.0n is useful.

If config.guess finds a 64-bit kernel, it checks CC to see if it
generates code for the 64-bit runtime and sets the machine string
to hppa64 if it does.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)