[PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Mark Kettenis]

Ok, here it is.  The brand new DWARF CFI frame unwinder.  I'm not yet
fully happy with it, especially the way the frame info gets "built"
and stored.  I currently abuse the `sym_private' member of `struct
objfile' as a pointer the the object's FDEs.  The FDEs are stored in a
linked list, which of course isn't optimal.  But I wanted to get the
code out such that people can take a look at it and test it, so I
committed it to the i386newframe branch.

Anyway, apart from the issues above, this only handles .debug_frame
sections; not .eh_frame sections.  And only DWARF2 CFA instructions
are fully supported.  Most DWARF3 extensions and GNU extensions still
have to be implemented.  That said, this seems to be working pretty
well for me.

Mark

Index: ChangeLog
from  Mark Kettenis  <kettenis@gnu.org>
 
	* dwarf-frame.c, dwarf-frame.h: New files.
	* i386-tdep.c: Include "dwarf-frame.h".
	(i386_gdbarch_init): Hook in the DWARF CFI frame unwinder.
	* Makefile.in (SFILES): Add dwarf-frame.c
	(dwarf_frame_h): Define.
	(COMMON_OBS): Add dwarf-frame.o
	(dwarf-frame.o): Add dependencies.

Index: Makefile.in
===================================================================
RCS file: /cvs/src/src/gdb/Makefile.in,v
retrieving revision 1.364.2.1
diff -u -p -r1.364.2.1 Makefile.in
--- Makefile.in 4 May 2003 11:37:38 -0000 1.364.2.1
+++ Makefile.in 4 May 2003 21:53:50 -0000
@@ -513,8 +513,8 @@ SFILES = ada-exp.y ada-lang.c ada-typepr
 	charset.c cli-out.c coffread.c coff-pe-read.c \
 	complaints.c completer.c corefile.c \
 	cp-abi.c cp-support.c cp-namespace.c cp-valprint.c \
-	dbxread.c demangle.c disasm.c doublest.c \
-	dummy-frame.c dwarfread.c dwarf2expr.c dwarf2loc.c dwarf2read.c \
+	dbxread.c demangle.c disasm.c doublest.c dummy-frame.c \
+	dwarf-frame.c dwarfread.c dwarf2expr.c dwarf2loc.c dwarf2read.c \
 	elfread.c environ.c eval.c event-loop.c event-top.c expprint.c \
 	f-exp.y f-lang.c f-typeprint.c f-valprint.c findvar.c frame.c \
 	frame-base.c \
@@ -635,6 +635,7 @@ disasm_h = disasm.h
 doublest_h = doublest.h $(floatformat_h)
 dst_h = dst.h
 dummy_frame_h = dummy-frame.h
+dwarf_frame_h = dwarf-frame.h
 dwarf2cfi_h = dwarf2cfi.h
 dwarf2expr_h = dwarf2expr.h
 dwarf2loc_h = dwarf2loc.h
@@ -854,6 +855,7 @@ COMMON_OBS = version.o blockframe.o brea
 	dbxread.o coffread.o coff-pe-read.o elfread.o \
 	dwarfread.o dwarf2read.o mipsread.o stabsread.o corefile.o \
 	dwarf2expr.o dwarf2loc.o \
+	dwarf-frame.o \
 	c-lang.o f-lang.o objc-lang.o \
 	ui-out.o cli-out.o \
 	varobj.o wrapper.o \
@@ -1665,6 +1667,9 @@ dummy-frame.o: dummy-frame.c $(defs_h) $
 	$(frame_h) $(inferior_h) $(gdb_assert_h) $(frame_unwind_h)
 dve3900-rom.o: dve3900-rom.c $(defs_h) $(gdbcore_h) $(target_h) $(monitor_h) \
 	$(serial_h) $(inferior_h) $(command_h) $(gdb_string_h) $(regcache_h)
+dwarf-frame.o: dwarf-frame.c $(defs_h) $(dwarf_frame_h) $(frame_h) \
+	$(gdbcore_h) $(gdbtypes_h) $(symtab_h) $(objfiles_h) $(regcache_h) \
+	$(gdb_assert_h) $(gdb_string_h) $(dwarf_frame_h)
 dwarf2cfi.o: dwarf2cfi.c $(defs_h) $(gdbcore_h) $(symtab_h) $(symfile_h) \
 	$(objfiles_h) $(target_h) $(elf_dwarf2_h) $(inferior_h) \
 	$(regcache_h) $(dwarf2cfi_h) $(gdb_assert_h)
Index: i386-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/i386-tdep.c,v
retrieving revision 1.138.2.5
diff -u -p -r1.138.2.5 i386-tdep.c
--- i386-tdep.c 4 May 2003 21:47:01 -0000 1.138.2.5
+++ i386-tdep.c 4 May 2003 21:53:51 -0000
@@ -24,6 +24,7 @@
 #include "arch-utils.h"
 #include "command.h"
 #include "dummy-frame.h"
+#include "dwarf-frame.h"
 #include "doublest.h"
 #include "floatformat.h"
 #include "frame.h"
@@ -1717,6 +1718,11 @@ i386_gdbarch_init (struct gdbarch_info i
   /* Hook in ABI-specific overrides, if they have been registered.  */
   gdbarch_init_osabi (info, gdbarch);
 
+  /* Hook in the DWARF CFI frame unwinder.  */
+  frame_unwind_append_predicate (gdbarch, dwarf_frame_p);
+  frame_base_append_predicate (gdbarch, dwarf_frame_base_p);
+  set_gdbarch_dwarf2_build_frame_info (gdbarch, dwarf2_build_frame_info);
+  
   frame_unwind_append_predicate (gdbarch, i386_frame_p);
   frame_base_set_default (gdbarch, &i386_frame_base);
 
Index: dwarf-frame.c
===================================================================
RCS file: dwarf-frame.c
diff -N dwarf-frame.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ dwarf-frame.c 4 May 2003 21:53:51 -0000
@@ -0,0 +1,967 @@
+/* Frame unwinder for frames with DWARF Call Frame Information.
+
+   Copyright 2003 Free Software Foundation, Inc.
+
+   Contributed by Mark Kettenis.
+
+   This file is part of GDB.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place - Suite 330,
+   Boston, MA 02111-1307, USA.  */
+
+#include "defs.h"
+#include "dwarf2expr.h"
+#include "elf/dwarf2.h"
+#include "frame.h"
+#include "frame-base.h"
+#include "frame-unwind.h"
+#include "gdbcore.h"
+#include "gdbtypes.h"
+#include "symtab.h"
+#include "objfiles.h"
+#include "regcache.h"
+
+#include "gdb_assert.h"
+#include <stddef.h>
+#include "gdb_string.h"
+
+#include "dwarf-frame.h"
+
+/* Call Frame Information (CFI).  */
+
+/* Common Information Entry (CIE).  */
+
+struct dwarf_cie
+{
+  /* Offset into the .debug_frame section where this CIE was found.
+     Used to identify this CIE.  */
+  ULONGEST cie_pointer;
+
+  /* Constant that is factored out of all advance location
+     instructions.  */
+  ULONGEST code_alignment_factor;
+
+  /* Constants that is factored out of all offset instructions.  */
+  LONGEST data_alignment_factor;
+
+  /* Return address column.  */
+  ULONGEST return_address_register;
+
+  /* Instruction sequence to initialize a register set.  */
+  unsigned char *initial_instructions;
+  unsigned char *end;
+
+  struct dwarf_cie *next;
+};
+
+/* Frame Description Entry (FDE).  */
+
+struct dwarf_fde
+{
+  /* CIE for this FDE.  */
+  struct dwarf_cie *cie;
+
+  /* First location associated with this FDE.  */
+  CORE_ADDR initial_location;
+
+  /* Number of bytes of program instructions described by this FDE.  */
+  CORE_ADDR address_range;
+
+  /* Instruction sequence.  */
+  unsigned char *instructions;
+  unsigned char *end;
+
+  struct dwarf_fde *next;
+};
+
+static struct dwarf_fde *dwarf_frame_find_fde (CORE_ADDR *pc);
+\f
+
+/* Structure describing a frame state.  */
+
+struct dwarf_frame_state
+{
+  /* Each register save state can be described in terms of a CFA slot,
+     another register, or a location expression.  */
+  struct dwarf_frame_state_reg_info
+  {
+    struct dwarf_frame_state_reg
+    {
+      union {
+	LONGEST offset;
+	ULONGEST reg;
+	unsigned char *exp;
+      } loc;
+      enum {
+	REG_UNSAVED,
+	REG_SAVED_OFFSET,
+	REG_SAVED_REG,
+	REG_SAVED_EXP,
+	REG_UNMODIFIED
+      } how;
+    } *reg;
+    int num_regs;
+
+    /* Used to implement DW_CFA_remember_state.  */
+    struct dwarf_frame_state_reg_info *prev;
+  } regs;
+
+  LONGEST cfa_offset;
+  ULONGEST cfa_reg;
+  unsigned char *cfa_exp;
+  enum {
+    CFA_UNSET,
+    CFA_REG_OFFSET,
+    CFA_EXP
+  } cfa_how;
+
+  /* The PC described by the current frame state.  */
+  CORE_ADDR pc;
+
+  /* Initial register set from the CIE.
+     Used to implement DW_CFA_restore.  */
+  struct dwarf_frame_state_reg_info initial;
+
+  /* The information we care about from the CIE.  */
+  LONGEST data_align;
+  ULONGEST code_align;
+  ULONGEST retaddr_column;
+};
+
+/* Store the length the expression for the CFA in the `cfa_reg' field,
+   which is unused in that case.  */
+#define cfa_len cfa_reg
+
+/* Assert that the register set RS is large enough to store NUM_REGS
+   columns.  If necessary, enlarge the register set.  */
+
+static void
+dwarf_frame_state_alloc_regs (struct dwarf_frame_state_reg_info *rs,
+			      int num_regs)
+{
+  size_t size = sizeof (struct dwarf_frame_state_reg);
+
+  if (num_regs <= rs->num_regs)
+    return;
+
+  rs->reg = (struct dwarf_frame_state_reg *)
+    xrealloc (rs->reg, num_regs * size);
+
+  /* Initialize newly allocated registers.  */
+  memset (rs->reg + rs->num_regs * size, 0, (num_regs - rs->num_regs) * size);
+  rs->num_regs = num_regs;
+}
+
+/* Copy the register columns in register set RS into newly allocated
+   memory and return a pointer to this newly created copy.  */
+
+static struct dwarf_frame_state_reg *
+dwarf_frame_state_copy_regs (struct dwarf_frame_state_reg_info *rs)
+{
+  size_t size = rs->num_regs * sizeof (struct dwarf_frame_state_reg_info);
+  struct dwarf_frame_state_reg *reg;
+
+  reg = (struct dwarf_frame_state_reg *) xmalloc (size);
+  memcpy (reg, rs->reg, size);
+
+  return reg;
+}
+
+/* Release the memory allocated to register set RS.  */
+
+static void
+dwarf_frame_state_free_regs (struct dwarf_frame_state_reg_info *rs)
+{
+  if (rs)
+    {
+      dwarf_frame_state_free_regs (rs->prev);
+
+      xfree (rs->reg);
+      xfree (rs);
+    }
+}
+
+/* Release the memory allocated to the frame state FS.  */
+
+static void
+dwarf_frame_state_free (void *p)
+{
+  struct dwarf_frame_state *fs = p;
+
+  dwarf_frame_state_free_regs (fs->initial.prev);
+  dwarf_frame_state_free_regs (fs->regs.prev);
+  xfree (fs->initial.reg);
+  xfree (fs->regs.reg);
+  xfree (fs);
+}
+\f
+
+/* Helper functions for execute_stack_op.  */
+
+static CORE_ADDR
+read_reg (void *baton, int reg)
+{
+  struct frame_info *next_frame = (struct frame_info *) baton;
+  int regnum;
+  char *buf;
+
+  regnum = DWARF2_REG_TO_REGNUM (reg);
+
+  buf = (char *) alloca (register_size (current_gdbarch, regnum));
+  frame_unwind_register (next_frame, regnum, buf);
+  return extract_typed_address (buf, builtin_type_void_data_ptr);
+}
+
+static void
+read_mem (void *baton, char *buf, CORE_ADDR addr, size_t len)
+{
+  read_memory (addr, buf, len);
+}
+
+static CORE_ADDR
+execute_stack_op (unsigned char *exp, ULONGEST len,
+		  struct frame_info *next_frame, CORE_ADDR initial)
+{
+  struct dwarf_expr_context *ctx;
+  CORE_ADDR result;
+
+  ctx = new_dwarf_expr_context ();
+  ctx->baton = next_frame;
+  ctx->read_reg = read_reg;
+  ctx->read_mem = read_mem;
+
+  dwarf_expr_push (ctx, initial);
+  dwarf_expr_eval (ctx, exp, len);
+  result = dwarf_expr_fetch (ctx, 0);
+
+  free_dwarf_expr_context (ctx);
+
+  return result;
+}
+\f
+
+static void
+execute_cfa_program (unsigned char *insn_ptr, unsigned char *insn_end,
+		     struct frame_info *next_frame,
+		     struct dwarf_frame_state *fs)
+{
+  CORE_ADDR pc = frame_pc_unwind (next_frame);
+  int bytes_read;
+
+  while (insn_ptr < insn_end && fs->pc <= pc)
+    {
+      unsigned char insn = *insn_ptr++;
+      ULONGEST utmp, reg;
+      LONGEST offset;
+
+      if ((insn & 0xc0) == DW_CFA_advance_loc)
+	fs->pc += (insn & 0x3f) * fs->code_align;
+      else if ((insn & 0xc0) == DW_CFA_offset)
+	{
+	  reg = insn & 0x3f;
+	  insn_ptr = read_uleb128 (insn_ptr, insn_end, &utmp);
+	  offset = utmp * fs->data_align;
+	  dwarf_frame_state_alloc_regs (&fs->regs, reg + 1);
+	  fs->regs.reg[reg].how = REG_SAVED_OFFSET;
+	  fs->regs.reg[reg].loc.offset = offset;
+	}
+      else if ((insn & 0xc0) == DW_CFA_restore)
+	{
+	  gdb_assert (fs->initial.reg);
+	  reg = insn & 0x3f;
+	  dwarf_frame_state_alloc_regs (&fs->regs, reg + 1);
+	  fs->regs.reg[reg] = fs->initial.reg[reg];
+	}
+      else
+	{
+	  switch (insn)
+	    {
+	    case DW_CFA_set_loc:
+	      fs->pc = dwarf2_read_address (insn_ptr, insn_end, &bytes_read);
+	      insn_ptr += bytes_read;
+	      break;
+
+	    case DW_CFA_advance_loc1:
+	      utmp = extract_unsigned_integer (insn_ptr, 1);
+	      fs->pc += utmp * fs->code_align;
+	      insn_ptr++;
+	      break;
+	    case DW_CFA_advance_loc2:
+	      utmp = extract_unsigned_integer (insn_ptr, 2);
+	      fs->pc += utmp * fs->code_align;
+	      insn_ptr += 2;
+	      break;
+	    case DW_CFA_advance_loc4:
+	      utmp = extract_unsigned_integer (insn_ptr, 4);
+	      fs->pc += utmp * fs->code_align;
+	      insn_ptr += 4;
+	      break;
+
+	    case DW_CFA_offset_extended:
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &reg);
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &utmp);
+	      offset = utmp * fs->data_align;
+	      dwarf_frame_state_alloc_regs (&fs->regs, reg + 1);
+	      fs->regs.reg[reg].how = REG_SAVED_OFFSET;
+	      fs->regs.reg[reg].loc.offset = offset;
+	      break;
+
+	    case DW_CFA_restore_extended:
+	      gdb_assert (fs->initial.reg);
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &reg);
+	      dwarf_frame_state_alloc_regs (&fs->regs, reg + 1);
+	      fs->regs.reg[reg] = fs->initial.reg[reg];
+	      break;
+
+	    case DW_CFA_undefined:
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &reg);
+	      dwarf_frame_state_alloc_regs (&fs->regs, reg + 1);
+	      fs->regs.reg[reg].how = REG_UNSAVED;
+	      break;
+
+	    case DW_CFA_same_value:
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &reg);
+	      dwarf_frame_state_alloc_regs (&fs->regs, reg + 1);
+	      fs->regs.reg[reg].how = REG_UNMODIFIED;
+	      break;
+
+	    case DW_CFA_register:
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &reg);
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &utmp);
+	      dwarf_frame_state_alloc_regs (&fs->regs, reg + 1);
+	      fs->regs.reg[reg].loc.reg = utmp;
+	      break;
+
+	    case DW_CFA_remember_state:
+	      {
+		struct dwarf_frame_state_reg_info *new_rs;
+
+		new_rs = XMALLOC (struct dwarf_frame_state_reg_info);
+		*new_rs = fs->regs;
+		fs->regs.reg = dwarf_frame_state_copy_regs (&fs->regs);
+		fs->regs.prev = new_rs;
+	      }
+	      break;
+
+	    case DW_CFA_restore_state:
+	      {
+		struct dwarf_frame_state_reg_info *old_rs = fs->regs.prev;
+
+		gdb_assert (old_rs);
+
+		xfree (fs->regs.reg);
+		fs->regs = *old_rs;
+		xfree (old_rs);
+	      }
+	      break;
+
+	    case DW_CFA_def_cfa:
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &fs->cfa_reg);
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &utmp);
+	      fs->cfa_offset = utmp;
+	      fs->cfa_how = CFA_REG_OFFSET;
+	      break;
+
+	    case DW_CFA_def_cfa_register:
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &fs->cfa_reg);
+	      fs->cfa_how = CFA_REG_OFFSET;
+	      break;
+
+	    case DW_CFA_def_cfa_offset:
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &fs->cfa_offset);
+	      /* cfa_how deliberately not set.  */
+	      break;
+
+	    case DW_CFA_def_cfa_expression:
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &fs->cfa_len);
+	      fs->cfa_exp = insn_ptr;
+	      fs->cfa_how = CFA_EXP;
+	      insn_ptr += fs->cfa_len;
+	      break;
+
+	    case DW_CFA_nop:
+	      break;
+
+	    case DW_CFA_GNU_args_size:
+	      /* Ignored.  */
+	      insn_ptr = read_uleb128 (insn_ptr, insn_end, &utmp);
+	      break;
+
+	    default:
+	      internal_error (__FILE__, __LINE__, "Unknown CFI encountered.");
+	    }
+	}
+    }
+
+  /* Don't allow remember/restore between CIE and FDE programs.  */
+  dwarf_frame_state_free_regs (fs->regs.prev);
+  fs->regs.prev = NULL;
+}
+
+struct dwarf_frame_cache
+{
+  /* DWARF Call Frame Address.  */
+  CORE_ADDR cfa;
+
+  /* Saved registers, indexed by GDB register number, not by DWARF
+     register number.  */
+  struct dwarf_frame_state_reg *reg;
+};
+
+struct dwarf_frame_cache *
+dwarf_frame_cache (struct frame_info *next_frame, void **this_cache)
+{
+  struct cleanup *old_chain;
+  int num_regs = NUM_REGS + NUM_PSEUDO_REGS;
+  struct dwarf_frame_cache *cache;
+  struct dwarf_frame_state *fs;
+  struct dwarf_fde *fde;
+  int reg;
+
+  if (*this_cache)
+    return *this_cache;
+
+  /* Allocate a new cache.  */
+  cache = FRAME_OBSTACK_ZALLOC (struct dwarf_frame_cache);
+  cache->reg = FRAME_OBSTACK_CALLOC (num_regs, struct dwarf_frame_state_reg);
+
+  /* Allocate and initialize the frame state.  */
+  fs = XMALLOC (struct dwarf_frame_state);
+  memset (fs, 0, sizeof (struct dwarf_frame_state));
+  old_chain = make_cleanup (dwarf_frame_state_free, fs);
+
+  /* Unwind the PC and find the correct FDE.  */
+  fs->pc = frame_pc_unwind (next_frame);
+  fde = dwarf_frame_find_fde (&fs->pc);
+
+  /* Extract any interesting information from the CIE.  */
+  fs->data_align = fde->cie->data_alignment_factor;
+  fs->code_align = fde->cie->code_alignment_factor;
+  fs->retaddr_column = fde->cie->return_address_register;
+
+  /* First decode all the insns in the CIE.  */
+  execute_cfa_program (fde->cie->initial_instructions,
+		       fde->cie->end, next_frame, fs);
+
+  /* Save the initialized register set.  */
+  fs->initial = fs->regs;
+  fs->initial.reg = dwarf_frame_state_copy_regs (&fs->regs);
+
+  /* Then decode the insns in the FDE up to our target PC.  */
+  execute_cfa_program (fde->instructions, fde->end, next_frame, fs);
+
+  /* Caclulate the CFA.  */
+  switch (fs->cfa_how)
+    {
+    case CFA_REG_OFFSET:
+      cache->cfa = read_reg (next_frame, fs->cfa_reg);
+      cache->cfa += fs->cfa_offset;
+      break;
+
+    case CFA_EXP:
+      cache->cfa = execute_stack_op (fs->cfa_exp, fs->cfa_len, next_frame, 0);
+      break;
+
+    default:
+      internal_error (__FILE__, __LINE__, "Unknown CFA rule.");
+    }
+
+  /* Save the register info in the cache.  */
+  for (reg = 0; reg < fs->regs.num_regs; reg++)
+    {
+      /* Use the GDB register number as index.  */
+      int regnum = DWARF2_REG_TO_REGNUM (reg);
+
+      if (regnum >= 0 && regnum < num_regs)
+	cache->reg[regnum] = fs->regs.reg[reg];
+    }
+
+  /* Make sure we have stored the return addess value.  */
+  if (cache->reg[PC_REGNUM].how == REG_UNSAVED)
+    cache->reg[PC_REGNUM] = fs->regs.reg[fs->retaddr_column];
+
+  do_cleanups (old_chain);
+
+  *this_cache = cache;
+  return cache;
+}
+
+static void
+dwarf_frame_this_id (struct frame_info *next_frame, void **this_cache,
+		     struct frame_id *this_id)
+{
+  struct dwarf_frame_cache *cache = dwarf_frame_cache (next_frame, this_cache);
+
+  (*this_id) = frame_id_build (cache->cfa, frame_func_unwind (next_frame));
+}
+
+static void
+dwarf_frame_prev_register (struct frame_info *next_frame, void **this_cache,
+			    int regnum, int *optimizedp,
+			    enum lval_type *lvalp, CORE_ADDR *addrp,
+			    int *realnump, void *valuep)
+{
+  struct dwarf_frame_cache *cache = dwarf_frame_cache (next_frame, this_cache);
+
+  switch (cache->reg[regnum].how)
+    {
+    case REG_UNSAVED:
+      *optimizedp = 1;
+      *lvalp = not_lval;
+      *addrp = 0;
+      *realnump = -1;
+      if (regnum == SP_REGNUM)
+	{
+	  /* GCC defines the CFA as the value of the stack pointer
+	     just before the call instruction is executed.  Do other
+	     compilers use the same definition?  */
+	  *optimizedp = 0;
+	  if (valuep)
+	    {
+	      /* Store the value.  */
+	      store_typed_address (valuep, builtin_type_void_data_ptr,
+				   cache->cfa);
+	    }
+	}
+      else if (valuep)
+	{
+	  /* In some cases, for example %eflags on the i386, we have
+	     to provide a sane value, even though this register wasn't
+	     saved.  Assume we can get it from NEXT_FRAME.  */
+	  frame_unwind_register (next_frame, regnum, valuep);
+	}
+      break;
+
+    case REG_SAVED_OFFSET:
+      *optimizedp = 0;
+      *lvalp = lval_memory;
+      *addrp = cache->cfa + cache->reg[regnum].loc.offset;
+      *realnump = -1;
+      if (valuep)
+	{
+	  /* Read the value in from memory.  */
+	  read_memory (*addrp, valuep,
+		       register_size (current_gdbarch, regnum));
+	}
+      break;
+
+    case REG_SAVED_REG:
+      *optimizedp = 0;
+      *lvalp = lval_register;
+      *addrp = 0;
+      *realnump = DWARF2_REG_TO_REGNUM (cache->reg[regnum].loc.reg);
+      if (valuep)
+	{
+	  /* Read the value from the register.  */
+	  frame_unwind_register (next_frame, *realnump, valuep);
+	}
+      break;
+
+    case REG_UNMODIFIED:
+      frame_register_unwind (next_frame, regnum,
+			     optimizedp, lvalp, addrp, realnump, valuep);
+      break;
+
+    default:
+      internal_error (__FILE__, __LINE__, "Unknown register rule.");
+    }
+}
+
+static const struct frame_unwind dwarf_frame_unwind =
+{
+  NORMAL_FRAME,
+  dwarf_frame_this_id,
+  dwarf_frame_prev_register
+};
+
+const struct frame_unwind *
+dwarf_frame_p (CORE_ADDR pc)
+{
+  if (dwarf_frame_find_fde (&pc))
+    return &dwarf_frame_unwind;
+
+  return NULL;
+}
+\f
+static CORE_ADDR
+dwarf_frame_base_address (struct frame_info *next_frame, void **this_cache)
+{
+  struct dwarf_frame_cache *cache = dwarf_frame_cache (next_frame, this_cache);
+
+  return cache->cfa;
+}
+
+static const struct frame_base dwarf_frame_base =
+{
+  &dwarf_frame_unwind,
+  dwarf_frame_base_address,
+  dwarf_frame_base_address,
+  dwarf_frame_base_address
+};
+
+const struct frame_base *
+dwarf_frame_base_p (CORE_ADDR pc)
+{
+  if (dwarf_frame_find_fde (&pc))
+    return &dwarf_frame_base;
+
+  return NULL;
+}
+\f
+/* A minimal decoding of DWARF2 compilation units.  We only decode
+   what's needed to get to the call frame information.  */
+
+struct comp_unit
+{
+  /* Keep the bfd convenient.  */
+  bfd *abfd;
+
+  struct objfile *objfile;
+
+  /* Linked list of CIEs for this object.  */
+  struct dwarf_cie *cie;
+
+  /* Address size for this unit - from unit header.  */
+  unsigned char addr_size;
+
+  /* Pointer to the .debug_frame section loaded into memory.  */
+  char *dwarf_frame_buffer;
+
+  /* Length of the loaded .debug_frame section.  */
+  unsigned long dwarf_frame_size;
+};
+
+static unsigned int
+read_1_byte (bfd *bfd, char *buf)
+{
+  return bfd_get_8 (abfd, (bfd_byte *) buf);
+}
+
+static unsigned int
+read_4_bytes (bfd *abfd, char *buf)
+{
+  return bfd_get_32 (abfd, (bfd_byte *) buf);
+}
+
+static ULONGEST
+read_8_bytes (bfd *abfd, char *buf)
+{
+  return bfd_get_64 (abfd, (bfd_byte *) buf);
+}
+
+static ULONGEST
+read_unsigned_leb128 (bfd *abfd, char *buf, unsigned int *bytes_read_ptr)
+{
+  ULONGEST result;
+  unsigned int num_read;
+  int shift;
+  unsigned char byte;
+
+  result = 0;
+  shift = 0;
+  num_read = 0;
+
+  do
+    {
+      byte = bfd_get_8 (abfd, (bfd_byte *) buf);
+      buf++;
+      num_read++;
+      result |= ((byte & 0x7f) << shift);
+      shift += 7;
+    }
+  while (byte & 0x80);
+
+  *bytes_read_ptr = num_read;
+
+  return result;
+}
+
+static LONGEST
+read_signed_leb128 (bfd *abfd, char *buf, unsigned int *bytes_read_ptr)
+{
+  LONGEST result;
+  int shift;
+  unsigned int num_read;
+  unsigned char byte;
+
+  result = 0;
+  shift = 0;
+  num_read = 0;
+
+  do
+    {
+      byte = bfd_get_8 (abfd, (bfd_byte *) buf);
+      buf++;
+      num_read++;
+      result |= ((byte & 0x7f) << shift);
+      shift += 7;
+    }
+  while (byte & 0x80);
+
+  if ((shift < 32) && (byte & 0x40))
+    result |= -(1 << shift);
+
+  *bytes_read_ptr = num_read;
+
+  return result;
+}
+
+static ULONGEST
+read_initial_length (bfd *abfd, char *buf, unsigned int *bytes_read_ptr)
+{
+  LONGEST result;
+
+  result = bfd_get_32 (abfd, (bfd_byte *) buf);
+  if (result == 0xffffffff)
+    {
+      result = bfd_get_64 (abfd, (bfd_byte *) buf + 4);
+      *bytes_read_ptr = 12;
+    }
+  else
+    *bytes_read_ptr = 4;
+
+  return result;
+}
+
+CORE_ADDR
+read_address (struct comp_unit *unit, char *buf)
+{
+  switch (unit->addr_size)
+    {
+    case 8:
+      return bfd_get_64 (unit->abfd, (bfd_byte *) buf);
+    case 4:
+      return bfd_get_32 (unit->abfd, (bfd_byte *) buf);
+    case 2:
+      return bfd_get_16 (unit->abfd, (bfd_byte *) buf);
+    default:
+      internal_error (__FILE__, __LINE__, "Invalid address");
+    }  
+}
+\f
+
+/* GCC uses a single CIE for all FDEs in a .debug_frame section.
+   That's why we use a simple linked list here.  */
+
+static struct dwarf_cie *
+find_cie (struct comp_unit *unit, ULONGEST cie_pointer)
+{
+  struct dwarf_cie *cie = unit->cie;
+
+  while (cie)
+    {
+      if (cie->cie_pointer == cie_pointer)
+	return cie;
+
+      cie = cie->next;
+    }
+
+  return NULL;
+}
+
+static void
+add_cie (struct comp_unit *unit, struct dwarf_cie *cie)
+{
+  cie->next = unit->cie;
+  unit->cie = cie;
+}
+
+/* Find the FDE for *PC.  Return a pointer to the FDE, and store the
+   inital location associated with it into *PC.  */
+
+static struct dwarf_fde *
+dwarf_frame_find_fde (CORE_ADDR *pc)
+{
+  struct objfile *objfile;
+
+  ALL_OBJFILES (objfile)
+    {
+      struct dwarf_fde *fde = objfile->sym_private;
+      CORE_ADDR baseaddr;
+
+      baseaddr = ANOFFSET (objfile->section_offsets, SECT_OFF_TEXT (objfile));
+
+      while (fde)
+	{
+	  if (*pc >= fde->initial_location + baseaddr
+	      && *pc < fde->initial_location + baseaddr + fde->address_range)
+	    {
+	      *pc = fde->initial_location + baseaddr;
+	      return fde;
+	    }
+
+	  fde = fde->next;
+	}
+    }
+
+  return NULL;
+}
+
+static void
+add_fde (struct comp_unit *unit, struct dwarf_fde *fde)
+{
+  fde->next = unit->objfile->sym_private;
+  unit->objfile->sym_private = fde;
+}
+
+/* Read and the CIE or FDE in BUF and decode it.  */
+
+static char *
+decode_frame_entry (struct comp_unit *unit, char *buf)
+{
+  LONGEST length;
+  unsigned int bytes_read;
+  int dwarf64_p = 0;
+  int cie_p = 0;
+  ULONGEST cie_pointer;
+  char *start = buf;
+  char *end;
+
+  length = read_initial_length (unit->abfd, buf, &bytes_read);
+  buf += bytes_read;
+  end = buf + length;
+
+  if (bytes_read == 12)
+    dwarf64_p = 1;
+
+  if (dwarf64_p)
+    {
+      cie_pointer = read_8_bytes (unit->abfd, buf);
+      buf += 8;
+#ifdef CC_HAS_LONG_LONG
+      if (cie_pointer == 0xffffffffffffffffULL)
+	cie_p = 1;
+#endif
+    }
+  else
+    {
+      cie_pointer = read_4_bytes (unit->abfd, buf);
+      buf += 4;
+      if (cie_pointer == DW_CIE_ID)
+	cie_p = 1;
+    }
+
+  if (cie_p)
+    {
+      /* This is a CIE.  */
+      struct dwarf_cie *cie;
+
+      /* Record the offset into the .debug_frame section of this CIE.  */
+      cie_pointer = start - unit->dwarf_frame_buffer;
+
+      /* Check whether we've already read it.  */
+      if (find_cie (unit, cie_pointer))
+	return end;
+
+      cie = (struct dwarf_cie *)
+	obstack_alloc (&unit->objfile->psymbol_obstack,
+		       sizeof (struct dwarf_cie));
+      cie->cie_pointer = cie_pointer;
+
+      /* Check version number.  */
+      gdb_assert (read_1_byte (unit->abfd, buf) == DW_CIE_VERSION);
+      buf += 1;
+
+      /* Skip augmentation.  */
+      gdb_assert (read_1_byte (unit->abfd, buf) == 0);
+      buf += 1;
+
+      cie->code_alignment_factor =
+	read_unsigned_leb128 (unit->abfd, buf, &bytes_read);
+      buf += bytes_read;
+
+      cie->data_alignment_factor =
+	read_signed_leb128 (unit->abfd, buf, &bytes_read);
+      buf += bytes_read;
+
+      cie->return_address_register = read_1_byte (unit->abfd, buf);
+      buf += 1;
+
+      cie->initial_instructions = buf;
+      cie->end = end;
+
+      add_cie (unit, cie);
+    }
+  else
+    {
+      /* This is a FDE.  */
+      struct dwarf_fde *fde;
+
+      fde = (struct dwarf_fde *)
+	obstack_alloc (&unit->objfile->psymbol_obstack,
+		       sizeof (struct dwarf_fde));
+      fde->cie = find_cie (unit, cie_pointer);
+      if (fde->cie == NULL)
+	{
+	  decode_frame_entry (unit, unit->dwarf_frame_buffer + cie_pointer);
+	  fde->cie = find_cie (unit, cie_pointer);
+	}
+
+      gdb_assert (fde->cie != NULL);
+
+      fde->initial_location = read_address (unit, buf);
+      buf += unit->addr_size;
+
+      fde->address_range = read_address (unit, buf);
+      buf += unit->addr_size;
+
+      fde->instructions = buf;
+      fde->end = end;
+
+      add_fde (unit, fde);
+    }
+
+  return end;
+}
+\f
+
+/* FIXME: kettenis/20030504: This still needs to be integrated with
+   dwarf2read.c in a better way.  */
+
+/* Imported from dwarf2read.c.  */
+extern file_ptr dwarf_frame_offset;
+extern unsigned int dwarf_frame_size;
+extern asection *dwarf_frame_section;
+
+/* Imported from dwarf2read.c.  */
+extern char *dwarf2_read_section (struct objfile *objfile, file_ptr offset,
+				  unsigned int size, asection *sectp);
+
+void
+dwarf2_build_frame_info (struct objfile *objfile)
+{
+  struct comp_unit unit;
+  char *frame_ptr;
+
+  if (!dwarf_frame_offset)
+    return;
+
+  /* Build a minimal decoding of the DWARF2 compilation unit.  */
+
+  unit.abfd = objfile->obfd;
+  unit.objfile = objfile;
+  unit.cie = NULL;
+  unit.addr_size = objfile->obfd->arch_info->bits_per_address / 8;
+  unit.dwarf_frame_buffer = dwarf2_read_section (objfile,
+						 dwarf_frame_offset,
+						 dwarf_frame_size,
+						 dwarf_frame_section);
+  unit.dwarf_frame_size = dwarf_frame_size;
+
+  frame_ptr = unit.dwarf_frame_buffer;
+
+  while (frame_ptr < unit.dwarf_frame_buffer + unit.dwarf_frame_size)
+    frame_ptr = decode_frame_entry (&unit, frame_ptr);
+}
Index: dwarf-frame.h
===================================================================
RCS file: dwarf-frame.h
diff -N dwarf-frame.h
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ dwarf-frame.h 4 May 2003 21:53:51 -0000
@@ -0,0 +1,43 @@
+/* Frame unwinder for frames with DWARF Call Frame Information.
+
+   Copyright 2003 Free Software Foundation, Inc.
+
+   Contributed by Mark Kettenis.
+
+   This file is part of GDB.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place - Suite 330,
+   Boston, MA 02111-1307, USA.  */
+
+#ifndef DWARF_FRAME_H
+#define DWARF_FRAME_H 1
+
+struct objfile;
+
+/* Return the frame unwind methods for the function that contains PC,
+   or NULL if it can't be handled by DWARF CFI frame unwinder.  */
+
+const struct frame_unwind *dwarf_frame_p (CORE_ADDR pc);
+
+/* Return the frame base methods for the function that contains PC, or
+   NULL if it can't be handled by the DWARF CFI frame unwinder.  */
+
+const struct frame_base *dwarf_frame_base_p (CORE_ADDR pc);
+
+/* Register the DWARF CFI for OBJFILE.  */
+
+void dwarf_frame_build_info (struct objfile *objfile);
+
+#endif /* dwarf-frame.h */

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Andrew Cagney]

Mark, fyi,

> +    case REG_SAVED_REG:
> +      *optimizedp = 0;
> +      *lvalp = lval_register;
> +      *addrp = 0;
> +      *realnump = DWARF2_REG_TO_REGNUM (cache->reg[regnum].loc.reg);
> +      if (valuep)
> +	{
> +	  /* Read the value from the register.  */
> +	  frame_unwind_register (next_frame, *realnump, valuep);
> +	}
> +      break;
> +

Set *addrp to the register offset hack (Otherwize something mysterious 
fails.  What? I don't remember).

Using the frame obstack for memory looks right.

> +static CORE_ADDR
> +dwarf_frame_base_address (struct frame_info *next_frame, void **this_cache)
> +{
> +  struct dwarf_frame_cache *cache = dwarf_frame_cache (next_frame, this_cache);
> +
> +  return cache->cfa;
> +}

This isn't right.  It should return DW_AT_frame_base.  However, since 
dwarf2expr.c doesn't yet use these methods it doesn't [?] really matter. 
  Only affects ``info frame''.

I suspect things like execute_stack_op might need more cleanups.  A dead 
target, or bad memory read, can throw an error.

Looks like the interface works (and oh so much cleaner than the old 
code) :-)

Andrew

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Daniel Jacobowitz]

On Sun, May 04, 2003 at 11:35:27PM -0400, Andrew Cagney wrote:
> Mark, fyi,
> 
> >+    case REG_SAVED_REG:
> >+      *optimizedp = 0;
> >+      *lvalp = lval_register;
> >+      *addrp = 0;
> >+      *realnump = DWARF2_REG_TO_REGNUM (cache->reg[regnum].loc.reg);
> >+      if (valuep)
> >+	{
> >+	  /* Read the value from the register.  */
> >+	  frame_unwind_register (next_frame, *realnump, valuep);
> >+	}
> >+      break;
> >+
> 
> Set *addrp to the register offset hack (Otherwize something mysterious 
> fails.  What? I don't remember).
> 
> Using the frame obstack for memory looks right.
> 
> >+static CORE_ADDR
> >+dwarf_frame_base_address (struct frame_info *next_frame, void 
> >**this_cache)
> >+{
> >+  struct dwarf_frame_cache *cache = dwarf_frame_cache (next_frame, 
> >this_cache);
> >+
> >+  return cache->cfa;
> >+}
> 
> This isn't right.  It should return DW_AT_frame_base.  However, since 
> dwarf2expr.c doesn't yet use these methods it doesn't [?] really matter. 
>  Only affects ``info frame''.

I don't think it should.

The frame's CFA is the basis for identifying the frame and locating
saved registers in the CFI.  It is always present when you have CFI.

DW_AT_frame_base is the basis for locating saved variables and locals. 
It is generally present when you have DWARF-2 debug info.

The two are not necessarily related.  I don't remember how we settled
on providing DW_AT_frame_base.  Possibly a debug info auxiliary to the
function symbol or to the block.


By the way, I don't remember something else I believe we've
discussed...  Does each target that wants to use the CFI unwinder have
to add it in its gdbarch initialization?

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Mark Kettenis]

   Date: Sun, 04 May 2003 23:35:27 -0400
   From: Andrew Cagney <ac131313@redhat.com>

   Mark, fyi,

   > +    case REG_SAVED_REG:
   > +      *optimizedp = 0;
   > +      *lvalp = lval_register;
   > +      *addrp = 0;
   > +      *realnump = DWARF2_REG_TO_REGNUM (cache->reg[regnum].loc.reg);
   > +      if (valuep)
   > +	{
   > +	  /* Read the value from the register.  */
   > +	  frame_unwind_register (next_frame, *realnump, valuep);
   > +	}
   > +      break;
   > +

   Set *addrp to the register offset hack (Otherwize something mysterious 
   fails.  What? I don't remember).

Yes, findvar.c:value_of_register() and findvar.c:value_from_register()
use this.  Worse, our whole value subsystem seems to rely on this.
Ughh, that's really gross.  We should do something about that!

Anyway.  Thinking about it a bit more, I suspect the whole handling of
REG_SAVED_REG is wrong.  Instead, we should just change the register
number according to the DWARF CFI rule and let the unwinder for
NEXT_FRAME handle the request:

    case REG_SAVED_REG:
      regnum = DWARF2_REG_TO_REGNUM (cache->reg[regnum].loc.reg);
      /* FALLTHROUGH */

    case REG_UNMODIFIED:
      frame_register_unwind (next_frame, regnum,
			     optimizedp, lvalp, addrp, realnump, valuep);
      break;

Eventually this means that sentinel_frame_prev_register will provide
the register offset hack.

Do you agree with my analysis?

Mark

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Mark Kettenis]

   Date: Sun, 04 May 2003 23:35:27 -0400
   From: Andrew Cagney <ac131313@redhat.com>

   Using the frame obstack for memory looks right.

Well, the frame obstack only gets used for the frame cache.  When
building the frame cache temporary memory for the frame state is
allocated using xmalloc().  I considered using the obstack, but
growing objects after they've been finished seems to be impossible
without ugly hacks.

Memory for the CIE's and FDE's when reading the debug info is
allocated from the object's psymtab obstack.

Mark

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Andrew Cagney]

>> 
>> This isn't right.  It should return DW_AT_frame_base.  However, since 
>> dwarf2expr.c doesn't yet use these methods it doesn't [?] really matter. 
>>  Only affects ``info frame''.


> I don't think it should.

Er ...

> The frame's CFA is the basis for identifying the frame and locating
> saved registers in the CFI.  It is always present when you have CFI.
> 
> DW_AT_frame_base is the basis for locating saved variables and locals. 
> It is generally present when you have DWARF-2 debug info.

You and I went through all this not too long ago.  frame-base is for 
this high level thingie, frame-unwind is for the low level register 
information.

> The two are not necessarily related.  I don't remember how we settled
> on providing DW_AT_frame_base.  Possibly a debug info auxiliary to the
> function symbol or to the block.
> 
> 
> By the way, I don't remember something else I believe we've
> discussed...  Does each target that wants to use the CFI unwinder have
> to add it in its gdbarch initialization?

At present yes.

Given the amount of upheval required before a target will work with this 
code, it doesn't really matter.  As I,and now Mark, discovered, it is 
something of an all or nothing afair.

Andrew

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Mark Kettenis]

   Date: Sun, 4 May 2003 23:42:42 -0400
   From: Daniel Jacobowitz <drow@mvista.com>

   > >+static CORE_ADDR
   > >+dwarf_frame_base_address (struct frame_info *next_frame, void 
   > >**this_cache)
   > >+{
   > >+  struct dwarf_frame_cache *cache = dwarf_frame_cache (next_frame, 
   > >this_cache);
   > >+
   > >+  return cache->cfa;
   > >+}
   > 
   > This isn't right.

Probably yes.  I think I added this stuff when I was confused about
what exectly was the value printed by "info frame".

   > It should return DW_AT_frame_base.

I don't think that it is the job of the DWARF CFI module to provide
that.  I probably should simply remove the frame base code from the
DWARF CFI code, and trust the default frame base methods to do
something sane.

   > However, since 
   > dwarf2expr.c doesn't yet use these methods it doesn't [?] really matter. 
   >  Only affects ``info frame''.

   I don't think it should.

You mean return DW_AT_frame_base.

   The frame's CFA is the basis for identifying the frame and locating
   saved registers in the CFI.  It is always present when you have CFI.

   DW_AT_frame_base is the basis for locating saved variables and locals. 
   It is generally present when you have DWARF-2 debug info.

   The two are not necessarily related.

Indeed, there is no way of referencing the CFA from outside the DWARF
CFI info.

Mark

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Andrew Cagney]

>    Date: Sun, 04 May 2003 23:35:27 -0400
>    From: Andrew Cagney <ac131313@redhat.com>
> 
>    Mark, fyi,
> 
>    > +    case REG_SAVED_REG:
>    > +      *optimizedp = 0;
>    > +      *lvalp = lval_register;
>    > +      *addrp = 0;
>    > +      *realnump = DWARF2_REG_TO_REGNUM (cache->reg[regnum].loc.reg);
>    > +      if (valuep)
>    > +	{
>    > +	  /* Read the value from the register.  */
>    > +	  frame_unwind_register (next_frame, *realnump, valuep);
>    > +	}
>    > +      break;
>    > +
> 
>    Set *addrp to the register offset hack (Otherwize something mysterious 
>    fails.  What? I don't remember).
> 
> Yes, findvar.c:value_of_register() and findvar.c:value_from_register()
> use this.  Worse, our whole value subsystem seems to rely on this.
> Ughh, that's really gross.  We should do something about that!

[tell me about it :-)]

> Anyway.  Thinking about it a bit more, I suspect the whole handling of
> REG_SAVED_REG is wrong.  Instead, we should just change the register
> number according to the DWARF CFI rule and let the unwinder for
> NEXT_FRAME handle the request:
> 
>     case REG_SAVED_REG:
>       regnum = DWARF2_REG_TO_REGNUM (cache->reg[regnum].loc.reg);
>       /* FALLTHROUGH */

Please, anything but a fallthrough :-)

>     case REG_UNMODIFIED:
>       frame_register_unwind (next_frame, regnum,
> 			     optimizedp, lvalp, addrp, realnump, valuep);
>       break;
> 
> Eventually this means that sentinel_frame_prev_register will provide
> the register offset hack.
> 
> Do you agree with my analysis?

Yes.  Stores could otherwize go to the wrong register (a really horrible 
bug to track down).

I should note that the part of ``info registers'' where it prints out 
where a register was saved needs a re-think.  Now that things are 
recursive it has the potential for reporting a register that was saved N 
levels further in on the stack (it does this already with the d10v and 
PC vs LR).

BTW, suggest adding a few name prefixes to register number variables - 
it is really hard to know which space (DWARF 2 or GDB) the register's 
value falls in.

Andrew

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Daniel Jacobowitz]

On Mon, May 05, 2003 at 10:08:22AM -0400, Andrew Cagney wrote:
> 
> >>
> >>This isn't right.  It should return DW_AT_frame_base.  However, since 
> >>dwarf2expr.c doesn't yet use these methods it doesn't [?] really matter. 
> >> Only affects ``info frame''.
> 
> 
> >I don't think it should.
> 
> Er ...
> 
> >The frame's CFA is the basis for identifying the frame and locating
> >saved registers in the CFI.  It is always present when you have CFI.
> >
> >DW_AT_frame_base is the basis for locating saved variables and locals. 
> >It is generally present when you have DWARF-2 debug info.
> 
> You and I went through all this not too long ago.  frame-base is for 
> this high level thingie, frame-unwind is for the low level register 
> information.

Then, as Mark said, it shouldn't be providing a frame base at all.  The
CFA information is not the right frame base, and the use of
DW_AT_frame_base is exactly orthogonal to the use of CFI.

> >The two are not necessarily related.  I don't remember how we settled
> >on providing DW_AT_frame_base.  Possibly a debug info auxiliary to the
> >function symbol or to the block.
> >
> >
> >By the way, I don't remember something else I believe we've
> >discussed...  Does each target that wants to use the CFI unwinder have
> >to add it in its gdbarch initialization?
> 
> At present yes.
> 
> Given the amount of upheval required before a target will work with this 
> code, it doesn't really matter.  As I,and now Mark, discovered, it is 
> something of an all or nothing afair.

Sure, makes sense to me.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Andrew Cagney]

>    Date: Sun, 04 May 2003 23:35:27 -0400
>    From: Andrew Cagney <ac131313@redhat.com>
> 
>    Using the frame obstack for memory looks right.
> 
> Well, the frame obstack only gets used for the frame cache.  When
> building the frame cache temporary memory for the frame state is
> allocated using xmalloc().  I considered using the obstack, but
> growing objects after they've been finished seems to be impossible
> without ugly hacks.

Yes (sorry I should have been more specific).

> Memory for the CIE's and FDE's when reading the debug info is
> allocated from the object's psymtab obstack.
> 
> Mark
> 

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Andrew Cagney]

>> >The frame's CFA is the basis for identifying the frame and locating
>> >saved registers in the CFI.  It is always present when you have CFI.
>> >
>> >DW_AT_frame_base is the basis for locating saved variables and locals. 
>> >It is generally present when you have DWARF-2 debug info.
> 
>> 
>> You and I went through all this not too long ago.  frame-base is for 
>> this high level thingie, frame-unwind is for the low level register 
>> information.
> 
> 
> Then, as Mark said, it shouldn't be providing a frame base at all.  The
> CFA information is not the right frame base, and the use of
> DW_AT_frame_base is exactly orthogonal to the use of CFI.

Daniel, you and I went through all this not too long ago.

Not providing a dwarf2 specific frame-base (returning DW_AT_frame_base) 
leads to ``info frame'' printing inconsistent information.  It will 
report ``id.stack_addr'' as the frame's base and that is definitly not 
correct - it won't match the high level ``base'' that the user expects 
to match a disassembler.

Again, this is why the high level frame-base is separate to the low 
level frame-unwind.  It is possible to mix 'n' match.

However, it doesn't need to be implemented right now - at present 
dwarf2loc short circutes frame-base, implementing DW_AT_frame_base locally.

Andrew

Re: [PATCH/i386newframe/RFC] DWARF CFI frame unwinder

[Daniel Jacobowitz]

On Mon, May 05, 2003 at 10:52:54AM -0400, Andrew Cagney wrote:
> 
> >>>The frame's CFA is the basis for identifying the frame and locating
> >>>saved registers in the CFI.  It is always present when you have CFI.
> >>>
> >>>DW_AT_frame_base is the basis for locating saved variables and locals. 
> >>>It is generally present when you have DWARF-2 debug info.
> >
> >>
> >>You and I went through all this not too long ago.  frame-base is for 
> >>this high level thingie, frame-unwind is for the low level register 
> >>information.
> >
> >
> >Then, as Mark said, it shouldn't be providing a frame base at all.  The
> >CFA information is not the right frame base, and the use of
> >DW_AT_frame_base is exactly orthogonal to the use of CFI.
> 
> Daniel, you and I went through all this not too long ago.
> 
> Not providing a dwarf2 specific frame-base (returning DW_AT_frame_base) 
> leads to ``info frame'' printing inconsistent information.  It will 
> report ``id.stack_addr'' as the frame's base and that is definitly not 
> correct - it won't match the high level ``base'' that the user expects 
> to match a disassembler.
> 
> Again, this is why the high level frame-base is separate to the low 
> level frame-unwind.  It is possible to mix 'n' match.
> 
> However, it doesn't need to be implemented right now - at present 
> dwarf2loc short circutes frame-base, implementing DW_AT_frame_base locally.

I think we've managed to end up in violent agreement.  Since right now
the frame base is associated with the unwinder, shall we leave it as
the CFA with a comment saying something like:

  /* When we have DWARF-2 debugging information, this should be
     DW_AT_frame_base.  It should probably be provided by a method
     specific to the function's debug information instead of its
     unwind type.  */

Not sure I got the wording right.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer