Mirror of the gdb-patches mailing list
 help / color / mirror / Atom feed
* Re: [rfa/dwarf/doc] Inter-compilation-unit reference support for partial DIEs
@ 2004-04-22 22:58 Michael Elizabeth Chastain
  0 siblings, 0 replies; 26+ messages in thread
From: Michael Elizabeth Chastain @ 2004-04-22 22:58 UTC (permalink / raw)
  To: drow, jimb; +Cc: eliz, ezannoni, gdb-patches

I did more Perl hacking this week.  Feast on this set of tables:

  http://www.shout.net/~mec/sunday/current/Compare-by-gdb-HEAD-drow.html

It's a comparison of gdb HEAD versus gdb drow.*, using the same test
suite (from gdb HEAD) so that the results compare nicely.

Right now the only drow* is drow_intercu-20040221-branch.

This covers a multitude of gcc's and both dwarf-2 and stabs+.

There are no regressions on this branch.

Pretty nifty, huh?

Michael C


^ permalink raw reply	[flat|nested] 26+ messages in thread
* [rfa/dwarf/doc] Inter-compilation-unit reference support for partial DIEs
@ 2004-04-20 17:09 Daniel Jacobowitz
  2004-04-20 19:43 ` Eli Zaretskii
                   ` (3 more replies)
  0 siblings, 4 replies; 26+ messages in thread
From: Daniel Jacobowitz @ 2004-04-20 17:09 UTC (permalink / raw)
  To: gdb-patches; +Cc: jimb, ezannoni, eliz

Here we go.  This patch is enough to load binaries which use
inter-compilation-unit references, as long as you don't do anything which
would cause full symbol tables to be generated - of course, that's not
especially useful!

Here's how it works.  First, when we read in the abbrevs for each
compilation unit, we check whether there are any attributes which might
indicate inter-cu references.  If they are, we set a flag, and we build a
tree of dwarf2_per_cu_data structures, which describe a minimal amount of
information about every compilation unit in the objfile.  If we see a
reference, we look up the containing compilation unit in the tree.
This lets us find the start offset of the needed compilation unit.  We load
it, and store a pointer to it in the tree.

When we're finished processing a compilation unit, we age everything in the
cache, and free things which have exceeded the user-settable cache age
limit - it's a limit on how many compilation units can be processed in a row
that do not mention a particular cached item.  The default (5) is a
reasonable tradeoff between speed and memory use; it's a factor of six
faster than disabling caching, and a factor of twenty less memory usage than
never releasing cached items, in some crude testing.


The machinery used to age the cache probably looks a little overcomplicated;
in particular, there's no need for dwarf2_mark in this patch.  That's
because a partial symbol table won't have dependencies on other partial
symbol tables, given the way we currently support references in partial
DIEs - more specifically, because of the way we build partial symbol tables.
We only fetch name and flag information from references, so we can use a
referenced compilation unit before we've built a symbol table for it.

The machinery is necessary for full DIEs, because we can't use references to
a comp unit before we've built its symbol table.  I maintain the invariant
that we load DIEs for all CUs which we are going to need in the processing
of a particular CU before we begin building any symbol tables, so we have to
track dependencies, and make sure things don't fall out of the cache
unexpectedly.  I think there's another valid approach to this problem,
but it doesn't save much over this one, and it adds considerable complexity
to demand-(re-)load the extra CUs.


Dwarf changes OK?  Doc changes OK?  Comments/suggestions?

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

2004-04-20  Daniel Jacobowitz  <drow@mvista.com>

	* Makefile.in (dwarf2read.o): Update dependencies.
	* dwarf2read.c: Include "splay-tree.h", "command.h", and
	"gdbcmd.h".
	(struct dwarf2_per_objfile): Add cu_tree and read_in_chain.
	(struct dwarf2_cu): Add read_in_chain, per_cu, last_used,
	mark, and has_form_ref_addr.
	(struct dwarf2_per_cu_data): New.
	(dwarf2_max_cache_age): New.
	(dwarf2_build_psymtabs_hard): Free cached compilation units
	after loading.  Create and manage the cu_tree.
	(psymtab_to_symtab_1): Initialize cu.
	(dwarf2_read_abbrevs): Set has_form_ref_addr.
	(find_partial_die): Use dwarf2_find_containing_comp_unit
	and load_comp_unit.
	(free_stack_comp_unit): Update comments.  Clear the per-cu
	pointer.  Handle aging.
	(dwarf2_find_containing_comp_unit, free_cached_comp_units)
	(age_cached_comp_units, free_one_cached_comp_unit)
	(dwarf2_mark, dwarf2_clear_marks, splay_tree_obstack_allocate)
	(create_comp_unit_tree, load_comp_unit, set_dwarf2_cmdlist)
	(show_dwarf2_cmdlist, set_dwarf2_cmd, show_dwarf2_cmd): New.
	(_initialize_dwarf2_read): Provide "maint set dwarf2 max-cache-age"
	and "maint show dwarf2 max-cache-age".
	* gdbcmd.h (maintenance_set_cmdlist, maintenance_show_cmdlist): New
	externs.
	* maint.c (maintenance_set_cmdlist, maintenance_show_cmdlist): Make
	global.

2004-04-20  Daniel Jacobowitz  <drow@mvista.com>

	* gdb.texinfo (Maintenance Commands): Document "maint set dwarf2
	max-cache-age" and "maint show dwarf2 max-cache-age".

Index: Makefile.in
===================================================================
RCS file: /cvs/src/src/gdb/Makefile.in,v
retrieving revision 1.546
diff -u -p -r1.546 Makefile.in
--- Makefile.in	18 Apr 2004 16:50:29 -0000	1.546
+++ Makefile.in	20 Apr 2004 16:29:31 -0000
@@ -1730,7 +1730,8 @@ dwarf2read.o: dwarf2read.c $(defs_h) $(b
 	$(objfiles_h) $(elf_dwarf2_h) $(buildsym_h) $(demangle_h) \
 	$(expression_h) $(filenames_h) $(macrotab_h) $(language_h) \
 	$(complaints_h) $(bcache_h) $(dwarf2expr_h) $(dwarf2loc_h) \
-	$(cp_support_h) $(gdb_string_h) $(gdb_assert_h) $(hashtab_h)
+	$(cp_support_h) $(gdb_string_h) $(gdb_assert_h) $(hashtab_h) \
+	$(splay_tree_h) $(command_h) $(gdbcmd_h)
 dwarfread.o: dwarfread.c $(defs_h) $(symtab_h) $(gdbtypes_h) $(objfiles_h) \
 	$(elf_dwarf_h) $(buildsym_h) $(demangle_h) $(expression_h) \
 	$(language_h) $(complaints_h) $(gdb_string_h)
Index: dwarf2read.c
===================================================================
RCS file: /cvs/src/src/gdb/dwarf2read.c,v
retrieving revision 1.148
diff -u -p -r1.148 dwarf2read.c
--- dwarf2read.c	19 Apr 2004 18:20:50 -0000	1.148
+++ dwarf2read.c	20 Apr 2004 16:29:32 -0000
@@ -44,7 +44,10 @@
 #include "dwarf2expr.h"
 #include "dwarf2loc.h"
 #include "cp-support.h"
+#include "splay-tree.h"
 #include "hashtab.h"
+#include "command.h"
+#include "gdbcmd.h"
 
 #include <fcntl.h>
 #include "gdb_string.h"
@@ -166,6 +169,15 @@ struct dwarf2_per_objfile
   char *macinfo_buffer;
   char *ranges_buffer;
   char *loc_buffer;
+
+  /* A tree of all the compilation units.  Each member is a pointer
+     to a struct dwarf2_cu.  This will be set if and only if we have
+     encountered a compilation unit with inter-CU references.  */
+  splay_tree cu_tree;
+
+  /* A chain of compilation units that are currently read in, so that
+     they can be freed later.  */
+  struct dwarf2_per_cu_data *read_in_chain;
 };
 
 static struct dwarf2_per_objfile *dwarf2_per_objfile;
@@ -297,6 +309,25 @@ struct dwarf2_cu
      unit, including partial DIEs.  */
   struct obstack comp_unit_obstack;
 
+  /* When multiple dwarf2_cu structures are living in memory, this field
+     chains them all together, so that they can be released efficiently.
+     We will probably also want a generation counter so that most-recently-used
+     compilation units are cached...  */
+  struct dwarf2_per_cu_data *read_in_chain;
+
+  /* Backchain to our per_cu entry if the tree has been built.  */
+  struct dwarf2_per_cu_data *per_cu;
+
+  /* How many compilation units ago was this CU last referenced?  */
+  int last_used;
+
+  /* Mark used when releasing cached dies.  */
+  unsigned int mark : 1;
+
+  /* This flag will be set if this compilation unit might include
+     inter-compilation-unit references.  */
+  unsigned int has_form_ref_addr : 1;
+
   /* This flag will be set if this compilation unit includes any
      DW_TAG_namespace DIEs.  If we know that there are explicit
      DIEs for namespaces, we don't need to try to infer them
@@ -304,6 +335,22 @@ struct dwarf2_cu
   unsigned int has_namespace_info : 1;
 };
 
+struct dwarf2_per_cu_data
+{
+  /* The start offset and length of this compilation unit.  2**31-1
+     bytes should suffice to store the length of any compilation unit
+     - if it doesn't, GDB will fall over anyway.  */
+  unsigned long offset;
+  unsigned long length : 31;
+
+  /* Flag indicating this compilation unit will be read in before
+     any of the current compilation units are processed.  */
+  unsigned long queued : 1;
+
+  /* Set iff currently read in.  */
+  struct dwarf2_cu *cu;
+};
+
 /* The line number information for a compilation unit (found in the
    .debug_line section) begins with a "statement program header",
    which contains the following information.  */
@@ -577,6 +624,13 @@ struct field_info
     int nfnfields;
   };
 
+/* Loaded secondary compilation units are kept in memory until they
+   have not been referenced for the processing of this many
+   compilation units.  Set this to zero to disable caching.  Cache
+   sizes of up to at least twenty will improve startup time for
+   typical inter-CU-reference binaries, at an obvious memory cost.  */
+static int dwarf2_max_cache_age = 5;
+
 /* Various complaints about symbol reading that don't abort the process */
 
 static void
@@ -933,12 +987,27 @@ dwarf2_symbol_mark_computed (struct attr
 static char *skip_one_die (char *info_ptr, struct abbrev_info *abbrev,
 			   struct dwarf2_cu *cu);
 
+static struct dwarf2_per_cu_data *dwarf2_find_containing_comp_unit
+  (unsigned long offset, struct dwarf2_cu *cu);
+
 static void free_stack_comp_unit (void *);
 
+static void free_cached_comp_units (void *);
+
+static void age_cached_comp_units (void);
+
+static void free_one_cached_comp_unit (void *);
+
+static void dwarf2_mark (struct dwarf2_cu *);
+
+static void dwarf2_clear_marks (struct dwarf2_per_cu_data *);
+
 static void *hashtab_obstack_allocate (void *data, size_t size, size_t count);
 
 static void dummy_obstack_deallocate (void *object, void *data);
 
+static void *splay_tree_obstack_allocate (int size, void *data);
+
 static hashval_t partial_die_hash (const void *item);
 
 static int partial_die_eq (const void *item_lhs, const void *item_rhs);
@@ -1194,6 +1263,52 @@ partial_read_comp_unit_head (struct comp
   return info_ptr;
 }
 
+/* Create a tree of all compilation units in OBJFILE.  We do this only
+   if an inter-comp-unit reference is found; presumably if there is one,
+   there will be many, and one will occur early in the .debug_info section.
+   So there's no point in building this tree incrementally.  */
+
+static void
+create_comp_unit_tree (struct objfile *objfile)
+{
+  char *info_ptr = dwarf2_per_objfile->info_buffer;
+
+  /* Initialize the compilation unit tree.  */
+  dwarf2_per_objfile->cu_tree
+    = splay_tree_new_with_allocator (splay_tree_compare_ints,
+				     NULL, NULL,
+				     splay_tree_obstack_allocate,
+				     dummy_obstack_deallocate,
+				     &objfile->objfile_obstack);
+
+  while (info_ptr < dwarf2_per_objfile->info_buffer + dwarf2_per_objfile->info_size)
+    {
+      struct comp_unit_head cu_header;
+      char *beg_of_comp_unit;
+      struct dwarf2_per_cu_data *this_cu;
+      unsigned long offset;
+      int bytes_read;
+
+      offset = info_ptr - dwarf2_per_objfile->info_buffer;
+
+      /* Read just enough information to find out where the next
+	 compilation unit is.  */
+      cu_header.length = read_initial_length (objfile->obfd, info_ptr,
+					      &cu_header, &bytes_read);
+
+      /* Save the compilation unit for later lookup.  */
+      this_cu = obstack_alloc (&objfile->objfile_obstack,
+			       sizeof (struct dwarf2_per_cu_data));
+      memset (this_cu, 0, sizeof (*this_cu));
+      this_cu->offset = offset;
+      this_cu->length = cu_header.length + cu_header.initial_length_size;
+      splay_tree_insert (dwarf2_per_objfile->cu_tree, this_cu->offset,
+			 (splay_tree_value) this_cu);
+
+      info_ptr = info_ptr + this_cu->length;
+    }
+}
+
 /* Build the partial symbol table by doing a quick pass through the
    .debug_info and .debug_abbrev sections.  */
 
@@ -1207,13 +1322,19 @@ dwarf2_build_psymtabs_hard (struct objfi
   char *beg_of_comp_unit;
   struct partial_die_info comp_unit_die;
   struct partial_symtab *pst;
+  struct cleanup *back_to;
   CORE_ADDR lowpc, highpc, baseaddr;
+  splay_tree cu_tree = NULL;
 
   info_ptr = dwarf2_per_objfile->info_buffer;
 
+  /* Any cached compilation units will be linked by the per-objfile
+     read_in_chain.  Make sure to free them when we're done.  */
+  back_to = make_cleanup (free_cached_comp_units, NULL);
+
   /* Since the objects we're extracting from .debug_info vary in
      length, only the individual functions to extract them (like
-     read_comp_unit_head and load_partial_die) can really know whether
+     read_comp_unit_head and read_partial_die) can really know whether
      the buffer is large enough to hold another complete object.
 
      At the moment, they don't actually check that.  If .debug_info
@@ -1251,12 +1372,13 @@ dwarf2_build_psymtabs_hard (struct objfi
 
       cu.list_in_scope = &file_symbols;
 
-      cu.partial_dies = NULL;
-
       /* Read the abbrevs for this compilation unit into a table */
       dwarf2_read_abbrevs (abfd, &cu);
       make_cleanup (dwarf2_free_abbrev_table, &cu);
 
+      if (cu.has_form_ref_addr && dwarf2_per_objfile->cu_tree == NULL)
+	create_comp_unit_tree (objfile);
+
       /* Read the compilation unit die */
       abbrev = peek_die_abbrev (info_ptr, &bytes_read, &cu);
       info_ptr = read_partial_die (&comp_unit_die, abbrev, bytes_read,
@@ -1280,6 +1402,36 @@ dwarf2_build_psymtabs_hard (struct objfi
       /* Store the function that reads in the rest of the symbol table */
       pst->read_symtab = dwarf2_psymtab_to_symtab;
 
+      if (dwarf2_per_objfile->cu_tree != NULL)
+	{
+	  splay_tree_node node;
+	  struct dwarf2_per_cu_data *per_cu;
+
+	  node = splay_tree_lookup (dwarf2_per_objfile->cu_tree,
+				    cu.header.offset);
+	  gdb_assert (node != NULL);
+	  per_cu = (struct dwarf2_per_cu_data *) node->value;
+
+	  /* If this compilation unit was already read in, free the
+	     cached copy in order to read it in again.  This is
+	     necessary because we skipped some symbols when we first
+	     read in the compilation unit (see load_partial_dies).
+	     This problem could be avoided, but the benefit is
+	     unclear.  */
+	  if (per_cu->cu != NULL)
+	    free_one_cached_comp_unit (per_cu->cu);
+
+	  cu.per_cu = per_cu;
+
+	  /* Note that this is a pointer to our stack frame, being
+	     added to a global data structure.  It will be cleaned up
+	     in free_stack_comp_unit when we finish with this
+	     compilation unit.  */
+	  per_cu->cu = &cu;
+	}
+      else
+	cu.per_cu = NULL;
+
       /* Check if comp unit has_children.
          If so, read the rest of the partial symbols from this comp unit.
          If not, there's no more debug_info for this comp unit. */
@@ -1326,6 +1478,64 @@ dwarf2_build_psymtabs_hard (struct objfi
 
       do_cleanups (back_to_inner);
     }
+  do_cleanups (back_to);
+}
+
+/* Load the DIEs for a secondary CU into memory.  */
+
+static void
+load_comp_unit (struct dwarf2_per_cu_data *this_cu, struct objfile *objfile)
+{
+  bfd *abfd = objfile->obfd;
+  char *info_ptr, *beg_of_comp_unit;
+  struct partial_die_info comp_unit_die;
+  struct dwarf2_cu *cu;
+  struct abbrev_info *abbrev;
+  unsigned int bytes_read;
+  struct cleanup *back_to;
+
+  info_ptr = dwarf2_per_objfile->info_buffer + this_cu->offset;
+  beg_of_comp_unit = info_ptr;
+
+  cu = xmalloc (sizeof (struct dwarf2_cu));
+  memset (cu, 0, sizeof (struct dwarf2_cu));
+
+  obstack_init (&cu->comp_unit_obstack);
+
+  cu->objfile = objfile;
+  info_ptr = partial_read_comp_unit_head (&cu->header, info_ptr, abfd);
+
+  /* Complete the CU header.  */
+  cu->header.offset = beg_of_comp_unit - dwarf2_per_objfile->info_buffer;
+  cu->header.first_die_ptr = info_ptr;
+  cu->header.cu_head_ptr = beg_of_comp_unit;
+
+  /* Read the abbrevs for this compilation unit into a table */
+  dwarf2_read_abbrevs (abfd, cu);
+  back_to = make_cleanup (dwarf2_free_abbrev_table, cu);
+
+  /* Read the compilation unit die.  */
+  abbrev = peek_die_abbrev (info_ptr, &bytes_read, cu);
+  info_ptr = read_partial_die (&comp_unit_die, abbrev, bytes_read,
+			       abfd, info_ptr, cu);
+
+  /* Set the language we're debugging.  */
+  set_cu_language (comp_unit_die.language, cu);
+
+  /* Link into the compilation unit tree.  */
+  this_cu->cu = cu;
+  cu->per_cu = this_cu;
+
+  /* Check if comp unit has_children.
+     If so, read the rest of the partial symbols from this comp unit.
+     If not, there's no more debug_info for this comp unit. */
+  if (comp_unit_die.has_children)
+    load_partial_dies (abfd, info_ptr, 0, cu);
+
+  do_cleanups (back_to);
+
+  cu->read_in_chain = dwarf2_per_objfile->read_in_chain;
+  dwarf2_per_objfile->read_in_chain = this_cu;
 }
 
 /* Process all loaded DIEs for compilation unit CU, starting at FIRST_DIE.
@@ -2018,6 +2228,7 @@ psymtab_to_symtab_1 (struct partial_symt
   /* We're in the global namespace.  */
   processing_current_prefix = "";
 
+  memset (&cu, 0, sizeof (cu));
   obstack_init (&cu.comp_unit_obstack);
   back_to = make_cleanup (free_stack_comp_unit, &cu);
 
@@ -4437,6 +4648,17 @@ dwarf2_read_abbrevs (bfd *abfd, struct d
 		= xrealloc (cur_attrs, (allocated_attrs
 					* sizeof (struct attr_abbrev)));
 	    }
+
+	  /* Record whether this compilation unit might have
+	     inter-compilation-unit references.  If we don't know what form
+	     this attribute will have, then it might potentially be a
+	     DW_FORM_ref_addr, so we conservatively expect inter-CU
+	     references.  */
+
+	  if (abbrev_form == DW_FORM_ref_addr
+	      || abbrev_form == DW_FORM_indirect)
+	    cu->has_form_ref_addr = 1;
+
 	  cur_attrs[cur_abbrev->num_attrs].name = abbrev_name;
 	  cur_attrs[cur_abbrev->num_attrs++].form = abbrev_form;
 	  abbrev_name = read_unsigned_leb128 (abfd, abbrev_ptr, &bytes_read);
@@ -4880,8 +5102,15 @@ find_partial_die (unsigned long offset, 
       return find_partial_die_in_comp_unit (offset, cu);
     }
 
-  internal_error (__FILE__, __LINE__,
-		  "unsupported inter-compilation-unit reference");
+  per_cu = dwarf2_find_containing_comp_unit (offset, cu);
+  gdb_assert (per_cu != NULL);
+
+  if (per_cu->cu == NULL)
+    load_comp_unit (per_cu, cu->objfile);
+
+  per_cu->cu->last_used = 0;
+  *target_cu = per_cu->cu;
+  return find_partial_die_in_comp_unit (offset, per_cu->cu);
 }
 
 /* Adjust PART_DIE before generating a symbol for it.  This function
@@ -8557,11 +8786,56 @@ dwarf2_symbol_mark_computed (struct attr
     }
 }
 
-/* This cleanup function is passed the address of a dwarf2_cu on the stack
-   when we're finished with it.  We can't free the pointer itself, but
-   release any associated storage.
+/* Locate the compilation unit from CU's objfile which contains the
+   DIE at OFFSET.  Returns NULL on failure.
+
+   We assume that OFFSET is not the start of the compilation unit header
+   (it can be the compilation unit DIE, though, which comes after the
+   header).  */
+
+static struct dwarf2_per_cu_data *
+dwarf2_find_containing_comp_unit (unsigned long offset,
+				  struct dwarf2_cu *cu)
+{
+  struct objfile *objfile = cu->objfile;
+  struct dwarf2_per_cu_data *this_cu;
+  splay_tree cu_tree;
+  splay_tree_node node;
+
+  cu_tree = dwarf2_per_objfile->cu_tree;
+  gdb_assert (cu_tree != NULL);
+  
+  node = splay_tree_predecessor (cu_tree, offset);
+  gdb_assert (node != NULL);
+
+  this_cu = (struct dwarf2_per_cu_data *) node->value;
+  gdb_assert (offset >= this_cu->offset);
+  gdb_assert (offset < this_cu->offset + this_cu->length);
+  return this_cu;
+}
+
+/* Release one cached compilation unit, CU.  We unlink it from the tree
+   of compilation units, but we don't remove it from the read_in_chain;
+   the caller is responsible for that.  */
+
+static void
+free_one_comp_unit (void *data)
+{
+  struct dwarf2_cu *cu = data;
 
-   Only used during partial symbol parsing.  */
+  if (cu->per_cu != NULL)
+    cu->per_cu->cu = NULL;
+  cu->per_cu = NULL;
+
+  obstack_free (&cu->comp_unit_obstack, NULL);
+
+  xfree (cu);
+}
+
+/* This cleanup function is passed the address of a dwarf2_cu on the stack
+   when we're finished with it.  We can't free the pointer itself, but be
+   sure to unlink it from the cache.  Also release any associated storage
+   and perform cache maintenance.  */
 
 static void
 free_stack_comp_unit (void *data)
@@ -8570,6 +8844,127 @@ free_stack_comp_unit (void *data)
 
   obstack_free (&cu->comp_unit_obstack, NULL);
   cu->partial_dies = NULL;
+
+  if (cu->per_cu != NULL)
+    {
+      /* This compilation unit is on the stack in our caller, so we
+	 should not xfree it.  Just unlink it.  */
+      cu->per_cu->cu = NULL;
+      cu->per_cu = NULL;
+
+      /* If we had a per-cu pointer, then we may have other compilation
+	 units loaded, so age them now.  */
+      age_cached_comp_units ();
+    }
+}
+
+/* Free all cached compilation units.  */
+
+static void
+free_cached_comp_units (void *data)
+{
+  struct dwarf2_per_cu_data *per_cu, **last_chain;
+
+  per_cu = dwarf2_per_objfile->read_in_chain;
+  last_chain = &dwarf2_per_objfile->read_in_chain;
+  while (per_cu != NULL)
+    {
+      struct dwarf2_per_cu_data *next_cu;
+
+      next_cu = per_cu->cu->read_in_chain;
+
+      free_one_comp_unit (per_cu->cu);
+      *last_chain = next_cu;
+
+      per_cu = next_cu;
+    }
+}
+
+/* Increase the age counter on each cached compilation unit, and free
+   any that are too old.  */
+
+static void
+age_cached_comp_units (void)
+{
+  struct dwarf2_per_cu_data *per_cu, **last_chain;
+
+  dwarf2_clear_marks (dwarf2_per_objfile->read_in_chain);
+  per_cu = dwarf2_per_objfile->read_in_chain;
+  while (per_cu != NULL)
+    {
+      per_cu->cu->last_used ++;
+      if (per_cu->cu->last_used <= dwarf2_max_cache_age)
+	dwarf2_mark (per_cu->cu);
+      per_cu = per_cu->cu->read_in_chain;
+    }
+
+  per_cu = dwarf2_per_objfile->read_in_chain;
+  last_chain = &dwarf2_per_objfile->read_in_chain;
+  while (per_cu != NULL)
+    {
+      struct dwarf2_per_cu_data *next_cu;
+
+      next_cu = per_cu->cu->read_in_chain;
+
+      if (!per_cu->cu->mark)
+	{
+	  free_one_comp_unit (per_cu->cu);
+	  *last_chain = next_cu;
+	}
+      else
+	last_chain = &per_cu->cu->read_in_chain;
+
+      per_cu = next_cu;
+    }
+}
+
+/* Remove a single compilation unit from the cache.  */
+
+static void
+free_one_cached_comp_unit (void *target_cu)
+{
+  struct dwarf2_per_cu_data *per_cu, **last_chain;
+
+  per_cu = dwarf2_per_objfile->read_in_chain;
+  last_chain = &dwarf2_per_objfile->read_in_chain;
+  while (per_cu != NULL)
+    {
+      struct dwarf2_per_cu_data *next_cu;
+
+      next_cu = per_cu->cu->read_in_chain;
+
+      if (per_cu->cu == target_cu)
+	{
+	  free_one_comp_unit (per_cu->cu);
+	  *last_chain = next_cu;
+	  break;
+	}
+      else
+	last_chain = &per_cu->cu->read_in_chain;
+
+      per_cu = next_cu;
+    }
+}
+
+/* Set the mark field in CU and in every other compilation unit in the
+   cache that we must keep because we are keeping CU.  */
+
+static void
+dwarf2_mark (struct dwarf2_cu *cu)
+{
+  if (cu->mark)
+    return;
+  cu->mark = 1;
+}
+
+static void
+dwarf2_clear_marks (struct dwarf2_per_cu_data *per_cu)
+{
+  while (per_cu)
+    {
+      per_cu->cu->mark = 0;
+      per_cu = per_cu->cu->read_in_chain;
+    }
 }
 
 /* Allocation function for the libiberty hash table which uses an
@@ -8594,6 +8989,14 @@ dummy_obstack_deallocate (void *object, 
   return;
 }
 
+/* Allocation function for the libiberty splay tree which uses an obstack.  */
+
+static void *
+splay_tree_obstack_allocate (int size, void *data)
+{
+  return obstack_alloc ((struct obstack *) data, size);
+}
+
 /* Trivial hash function for partial_die_info: the hash value of a DIE
    is its offset in .debug_info for this objfile.  */
 
@@ -8615,10 +9018,51 @@ partial_die_eq (const void *item_lhs, co
   return part_die_lhs->offset == part_die_rhs->offset;
 }
 
+static struct cmd_list_element *set_dwarf2_cmdlist;
+static struct cmd_list_element *show_dwarf2_cmdlist;
+
+static void
+set_dwarf2_cmd (char *args, int from_tty)
+{
+  help_list (set_dwarf2_cmdlist, "maintenance set dwarf2 ", -1, gdb_stdout);
+}
+
+static void
+show_dwarf2_cmd (char *args, int from_tty)
+{ 
+  cmd_show_list (show_dwarf2_cmdlist, from_tty, "");
+}
+
 void _initialize_dwarf2_read (void);
 
 void
 _initialize_dwarf2_read (void)
 {
   dwarf2_objfile_data_key = register_objfile_data ();
+
+  add_prefix_cmd ("dwarf2", class_maintenance, set_dwarf2_cmd,
+		  "Set DWARF 2 specific variables.\n"
+		  "Configure DWARF 2 variables such as the cache size",
+                  &set_dwarf2_cmdlist, "maintenance set dwarf2 ",
+                  0/*allow-unknown*/, &maintenance_set_cmdlist);
+
+  add_prefix_cmd ("dwarf2", class_maintenance, show_dwarf2_cmd,
+		  "Show DWARF 2 specific variables\n"
+		  "Show DWARF 2 variables such as the cache size",
+                  &show_dwarf2_cmdlist, "maintenance show dwarf2 ",
+                  0/*allow-unknown*/, &maintenance_show_cmdlist);
+
+  add_setshow_cmd ("max-cache-age", class_obscure, var_zinteger,
+		   &dwarf2_max_cache_age,
+		   "Set an upper bound on the age of cached "
+		   "compilation units.\n"
+		   "A higher limit means that cached "
+		   "compilation units will be stored\n"
+		   "in memory longer, and more total memory will "
+		   "be used.  Zero disables\n"
+		   "caching, which can slow down startup.",
+		   "Show the upper bound on the age of cached "
+		   "dwarf2 compilation units.",
+		   NULL, NULL, &set_dwarf2_cmdlist,
+		   &show_dwarf2_cmdlist);
 }
Index: gdbcmd.h
===================================================================
RCS file: /cvs/src/src/gdb/gdbcmd.h,v
retrieving revision 1.10
diff -u -p -r1.10 gdbcmd.h
--- gdbcmd.h	7 May 2003 21:42:47 -0000	1.10
+++ gdbcmd.h	20 Apr 2004 16:29:32 -0000
@@ -98,6 +98,14 @@ extern struct cmd_list_element *maintena
 
 extern struct cmd_list_element *maintenanceprintlist;
 
+/* Chain containing all defined "maintenance set" subcommands. */
+
+extern struct cmd_list_element *maintenance_set_cmdlist;
+
+/* Chain containing all defined "maintenance show" subcommands. */
+
+extern struct cmd_list_element *maintenance_show_cmdlist;
+
 extern struct cmd_list_element *setprintlist;
 
 extern struct cmd_list_element *showprintlist;
Index: maint.c
===================================================================
RCS file: /cvs/src/src/gdb/maint.c,v
retrieving revision 1.41
diff -u -p -r1.41 maint.c
--- maint.c	19 Jan 2004 01:20:11 -0000	1.41
+++ maint.c	20 Apr 2004 16:29:32 -0000
@@ -609,8 +609,8 @@ maintenance_do_deprecate (char *text, in
 
 /* Maintenance set/show framework.  */
 
-static struct cmd_list_element *maintenance_set_cmdlist;
-static struct cmd_list_element *maintenance_show_cmdlist;
+struct cmd_list_element *maintenance_set_cmdlist;
+struct cmd_list_element *maintenance_show_cmdlist;
 
 static void
 maintenance_set_cmd (char *args, int from_tty)
Index: doc/gdb.texinfo
===================================================================
RCS file: /cvs/src/src/gdb/doc/gdb.texinfo,v
retrieving revision 1.202
diff -u -p -r1.202 gdb.texinfo
--- doc/gdb.texinfo	28 Mar 2004 12:22:55 -0000	1.202
+++ doc/gdb.texinfo	20 Apr 2004 16:29:35 -0000
@@ -19499,6 +19499,19 @@ data in a @file{gmon.out} file, be sure 
 Configuring with @samp{--enable-profiling} arranges for @value{GDBN} to be
 compiled with the @samp{-pg} compiler option.
 
+@kindex maint set dwarf2 max-cache-age
+@kindex maint show dwarf2 max-cache-age
+@item maint set dwarf2 max-cache-age
+@itemx maint show dwarf2 max-cache-age
+Control the DWARF 2 compilation unit cache.
+
+In object files with inter-compilation-unit references, such as those
+produced by the GCC option @samp{-feliminate-dwarf2-dups}, the DWARF 2
+reader needs to frequently refer to previously read compilation units.
+This setting controls how long a compilation unit will remain in the cache
+if it is not referenced.  Setting it to zero disables caching, which will
+slow down @value{GDBN} startup but reduce memory consumption.
+
 @end table
 
 


^ permalink raw reply	[flat|nested] 26+ messages in thread
end of thread, other threads:[~2004-09-20 23:59 UTC | newest]

Thread overview: 26+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2004-04-22 22:58 [rfa/dwarf/doc] Inter-compilation-unit reference support for partial DIEs Michael Elizabeth Chastain
  -- strict thread matches above, loose matches on Subject: below --
2004-04-20 17:09 Daniel Jacobowitz
2004-04-20 19:43 ` Eli Zaretskii
2004-04-22 22:15 ` Jim Blandy
2004-06-17  3:42 ` Daniel Jacobowitz
2004-06-17  4:19   ` Eli Zaretskii
2004-06-17  4:21     ` Daniel Jacobowitz
2004-06-17  4:26       ` Eli Zaretskii
2004-07-15 18:45   ` Daniel Jacobowitz
2004-07-16 11:18     ` Eli Zaretskii
2004-08-04 23:07     ` Daniel Jacobowitz
2004-08-05 17:23       ` Jim Blandy
2004-08-05 18:05         ` Daniel Jacobowitz
2004-08-06 22:25           ` Jim Blandy
2004-08-07 22:01             ` Daniel Jacobowitz
2004-08-08  4:42               ` Jim Blandy
2004-08-08 18:17                 ` Daniel Jacobowitz
2004-08-08 19:30                   ` Daniel Jacobowitz
2004-08-10  0:24                     ` Jim Blandy
2004-08-06 22:28           ` Jim Blandy
2004-08-06 22:58           ` Jim Blandy
2004-08-07  0:05           ` Jim Blandy
2004-09-16 21:15 ` Daniel Jacobowitz
2004-09-20 16:41   ` Daniel Jacobowitz
2004-09-20 20:28   ` Jim Blandy
2004-09-20 23:59     ` Daniel Jacobowitz

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox