Mirror of the gdb-patches mailing list
 help / color / mirror / Atom feed
From: Simon Marchi <simon.marchi@efficios.com>
To: gdb-patches@sourceware.org
Cc: Tom Tromey <tom@tromey.com>
Subject: [PATCH v2 42/42] Share DWARF partial symtabs
Date: Tue, 12 May 2020 17:18:24 -0400	[thread overview]
Message-ID: <20200512211824.6834-43-simon.marchi@efficios.com> (raw)
In-Reply-To: <20200512210913.5593-1-simon.marchi@efficios.com>

From: Tom Tromey <tom@tromey.com>

This changes the DWARF reader to share partial symtabs (or indices if
they are available) across objfiles.  This has a few parts.

* If multiple objfiles backed by the same BFD can share partial symtabs
  (see below), a single dwarf2_per_bfd is created.  It is stored in the
  per-bfd `dwarf2_per_bfd_bfd_data_key` registry.  Multiple
  dwarf2_per_objfile objects will point to the same instance.  The
  lifetime of these dwarf2_per_bfd objects is naturally handled.  When
  all the objfiles using the BFD are destroyed, the BFD's refount drops
  to 0, which triggers the removal of the corresponding dwarf2_per_bfd
  object from the registry and its destruction.

* If multiple objfiles backed by the same BFD can't share partial
  symtabs (see below), one dwarf2_per_bfd object is created for each
  objfile.  Each dwarf2_per_objfile will point to their own instance of
  dwarf2_per_bfd.  These instances of dwarf2_per_bfd are kept in a
  per-objfile registry, meaning that when the objfile gets destroyed,
  the dwarf2_per_bfd instance gets destroyed as well.

* objfile::partial_symtabs is changed to be a shared_ptr again.  This
  lets us stash a second reference in dwarf2_per_bfd; if the DWARF
  data is being shared, we can simply copy this value to the new
  objfile.

* Two dwarf2_per_objfile objects backed by the same BFD may share a
  dwarf2_per_bfd instance if:

  * No other symbol reader has found symbols, and
  * No BFD section rqeuires relocation

YYYY-MM-DD  Tom Tromey  <tom@tromey.com>
YYYY-MM-DD  Simon Marchi  <simon.marchi@efficios.com>

	* objfiles.h (struct objfile) <partial_symtabs>: Now a
	shared_ptr.
	* dwarf2/read.h (struct dwarf2_per_objfile) <partial_symtabs>: New
	member.
	* dwarf2/read.c (dwarf2_per_bfd_bfd_data_key,
	dwarf2_per_bfd_objfile_data_key>: New globals.
	(dwarf2_has_info): Use shared dwarf2_per_bfd if possible.
	(dwarf2_get_section_info): Use get_dwarf2_per_objfile.
	(dwarf2_initialize_objfile): Consider cases where per_bfd can be
	shared.
	(dwarf2_build_psymtabs): Set objfile::partial_symtabs and
	short-circuit when sharing.
	(dwarf2_build_psymtabs): Set dwarf2_per_objfile::partial_symtabs.
	(dwarf2_psymtab::expand_psymtab): Use free_cached_comp_units.
---
 gdb/dwarf2/read.c | 105 +++++++++++++++++++++++++++++++++++++++-------
 gdb/dwarf2/read.h |   5 +++
 gdb/objfiles.h    |   2 +-
 3 files changed, 97 insertions(+), 15 deletions(-)

diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c
index 3f93a51381f..2de4ead004f 100644
--- a/gdb/dwarf2/read.c
+++ b/gdb/dwarf2/read.c
@@ -105,7 +105,19 @@ static bool check_physname = false;
 /* When true, do not reject deprecated .gdb_index sections.  */
 static bool use_deprecated_index_sections = false;
 
-static const struct objfile_key<dwarf2_per_objfile> dwarf2_objfile_data_key;
+/* This is used to store the data that is always per objfile.  */
+static const objfile_key<dwarf2_per_objfile> dwarf2_objfile_data_key;
+
+/* These are used to store the dwarf2_per_bfd objects.
+
+   objfiles having the same BFD, which doesn't require relocations, are going to
+   share a dwarf2_per_bfd object, which is held in the _bfd_data_key version.
+
+   Other objfiles are not going to share a dwarf2_per_bfd with any other
+   objfiles, so they'll have their own version kept in the _objfile_data_key
+   version.  */
+static const struct bfd_key<dwarf2_per_bfd> dwarf2_per_bfd_bfd_data_key;
+static const struct objfile_key<dwarf2_per_bfd> dwarf2_per_bfd_objfile_data_key;
 
 /* The "aclass" indices for various kinds of computed DWARF symbols.  */
 
@@ -1853,9 +1865,30 @@ dwarf2_has_info (struct objfile *objfile,
 
   if (dwarf2_per_objfile == NULL)
     {
-      /* For now, each dwarf2_per_objfile owns its own dwarf2_per_bfd (no
-         sharing yet).  */
-      dwarf2_per_bfd *per_bfd = new dwarf2_per_bfd (objfile->obfd, names, can_copy);
+      dwarf2_per_bfd *per_bfd;
+
+      /* We can share a "dwarf2_per_bfd" with other objfiles if the BFD
+         doesn't require relocations and if there aren't partial symbols
+	 from some other reader.  */
+      if (!objfile_has_partial_symbols (objfile)
+	  && !gdb_bfd_requires_relocations (objfile->obfd))
+	{
+	  /* See if one has been created for this BFD yet.  */
+	  per_bfd = dwarf2_per_bfd_bfd_data_key.get (objfile->obfd);
+
+	  if (per_bfd == nullptr)
+	    {
+	      /* No, create it now.  */
+	      per_bfd = new dwarf2_per_bfd (objfile->obfd, names, can_copy);
+	      dwarf2_per_bfd_bfd_data_key.set (objfile->obfd, per_bfd);
+	    }
+	}
+      else
+	{
+	  /* No sharing possible, create one specifically for this objfile.  */
+	  per_bfd = new dwarf2_per_bfd (objfile->obfd, names, can_copy);
+	  dwarf2_per_bfd_objfile_data_key.set (objfile, per_bfd);
+	}
 
       dwarf2_per_objfile = dwarf2_objfile_data_key.emplace (objfile, objfile, per_bfd);
     }
@@ -2017,7 +2050,7 @@ dwarf2_get_section_info (struct objfile *objfile,
                          asection **sectp, const gdb_byte **bufp,
                          bfd_size_type *sizep)
 {
-  struct dwarf2_per_objfile *data = dwarf2_objfile_data_key.get (objfile);
+  struct dwarf2_per_objfile *data = get_dwarf2_per_objfile (objfile);
   struct dwarf2_section_info *info;
 
   /* We may see an objfile without any DWARF, in which case we just
@@ -5874,6 +5907,7 @@ dwarf2_initialize_objfile (struct objfile *objfile, dw_index_kind *index_kind)
 {
   struct dwarf2_per_objfile *dwarf2_per_objfile
     = get_dwarf2_per_objfile (objfile);
+  dwarf2_per_bfd *per_bfd = dwarf2_per_objfile->per_bfd;
 
   /* If we're about to read full symbols, don't bother with the
      indices.  In this case we also don't care if some other debug
@@ -5881,20 +5915,28 @@ dwarf2_initialize_objfile (struct objfile *objfile, dw_index_kind *index_kind)
      expanded anyway.  */
   if ((objfile->flags & OBJF_READNOW))
     {
-      dwarf2_per_objfile->per_bfd->using_index = 1;
+      /* When using READNOW, the using_index flag (set below) indicates that
+	 PER_BFD was already initialized, when we loaded some other objfile.  */
+      if (per_bfd->using_index)
+	{
+	  *index_kind = dw_index_kind::GDB_INDEX;
+	  dwarf2_per_objfile->resize_symtabs ();
+	  return true;
+	}
+
+      per_bfd->using_index = 1;
       create_all_comp_units (dwarf2_per_objfile);
       create_all_type_units (dwarf2_per_objfile);
-      dwarf2_per_objfile->per_bfd->quick_file_names_table
-	= create_quick_file_names_table
-	    (dwarf2_per_objfile->per_bfd->all_comp_units.size ());
+      per_bfd->quick_file_names_table
+	= create_quick_file_names_table (per_bfd->all_comp_units.size ());
       dwarf2_per_objfile->resize_symtabs ();
 
-      for (int i = 0; i < (dwarf2_per_objfile->per_bfd->all_comp_units.size ()
-			   + dwarf2_per_objfile->per_bfd->all_type_units.size ()); ++i)
+      for (int i = 0; i < (per_bfd->all_comp_units.size ()
+			   + per_bfd->all_type_units.size ()); ++i)
 	{
-	  dwarf2_per_cu_data *per_cu = dwarf2_per_objfile->per_bfd->get_cutu (i);
+	  dwarf2_per_cu_data *per_cu = per_bfd->get_cutu (i);
 
-	  per_cu->v.quick = OBSTACK_ZALLOC (&dwarf2_per_objfile->per_bfd->obstack,
+	  per_cu->v.quick = OBSTACK_ZALLOC (&per_bfd->obstack,
 					    struct dwarf2_per_cu_quick_data);
 	}
 
@@ -5905,6 +5947,24 @@ dwarf2_initialize_objfile (struct objfile *objfile, dw_index_kind *index_kind)
       return true;
     }
 
+  /* Was a debug names index already read when we processed an objfile sharing
+     PER_BFD?  */
+  if (per_bfd->debug_names_table != nullptr)
+    {
+      *index_kind = dw_index_kind::DEBUG_NAMES;
+      dwarf2_per_objfile->resize_symtabs ();
+      return true;
+    }
+
+  /* Was a GDB index already read when we processed an objfile sharing
+     PER_BFD?  */
+  if (per_bfd->index_table != nullptr)
+    {
+      *index_kind = dw_index_kind::GDB_INDEX;
+      dwarf2_per_objfile->resize_symtabs ();
+      return true;
+    }
+
   if (dwarf2_read_debug_names (dwarf2_per_objfile))
     {
       *index_kind = dw_index_kind::DEBUG_NAMES;
@@ -5945,6 +6005,16 @@ dwarf2_build_psymtabs (struct objfile *objfile)
 {
   struct dwarf2_per_objfile *dwarf2_per_objfile
     = get_dwarf2_per_objfile (objfile);
+  dwarf2_per_bfd *per_bfd = dwarf2_per_objfile->per_bfd;
+
+  if (per_bfd->partial_symtabs != nullptr)
+    {
+      /* Partial symbols were already read, so now we can simply
+	 attach them.  */
+      objfile->partial_symtabs = per_bfd->partial_symtabs;
+      dwarf2_per_objfile->resize_symtabs ();
+      return;
+    }
 
   init_psymbol_list (objfile, 1024);
 
@@ -5966,6 +6036,12 @@ dwarf2_build_psymtabs (struct objfile *objfile)
     {
       exception_print (gdb_stderr, except);
     }
+
+  /* Finish by setting the local reference to partial symtabs, so that
+     we don't try to read them again if reading another objfile with the same
+     BFD.  If we can't in fact share, this won't make a difference anyway as
+     the dwarf2_per_bfd object won't be shared.  */
+  per_bfd->partial_symtabs = objfile->partial_symtabs;
 }
 
 /* Find the base address of the compilation unit for range lists and
@@ -9014,9 +9090,10 @@ dwarf2_psymtab::expand_psymtab (struct objfile *objfile)
 {
   gdb_assert (!readin_p (objfile));
 
+  dwarf2_per_objfile *per_objfile = get_dwarf2_per_objfile (objfile);
+  free_cached_comp_units freer (per_objfile);
   expand_dependencies (objfile);
 
-  dwarf2_per_objfile *per_objfile = get_dwarf2_per_objfile (objfile);
   dw2_do_instantiate_symtab (per_cu_data, per_objfile, false);
   gdb_assert (get_compunit_symtab (objfile) != nullptr);
 }
diff --git a/gdb/dwarf2/read.h b/gdb/dwarf2/read.h
index 996cf55af22..53f0c7361a3 100644
--- a/gdb/dwarf2/read.h
+++ b/gdb/dwarf2/read.h
@@ -254,6 +254,11 @@ struct dwarf2_per_bfd
   /* CUs that are queued to be read.  */
   std::queue<dwarf2_queue_item> queue;
 
+  /* We keep a separate reference to the partial symtabs, in case we
+     are sharing them between objfiles.  This is only set after
+     partial symbols have been read the first time.  */
+  std::shared_ptr<psymtab_storage> partial_symtabs;
+
 private:
 
   /* The total number of per_cu and signatured_type objects that have
diff --git a/gdb/objfiles.h b/gdb/objfiles.h
index 0b47bd0c1e2..56ff52119dc 100644
--- a/gdb/objfiles.h
+++ b/gdb/objfiles.h
@@ -576,7 +576,7 @@ struct objfile
 
   /* The partial symbol tables.  */
 
-  std::unique_ptr<psymtab_storage> partial_symtabs;
+  std::shared_ptr<psymtab_storage> partial_symtabs;
 
   /* The object file's BFD.  Can be null if the objfile contains only
      minimal symbols, e.g. the run time common symbols for SunOS4.  */
-- 
2.26.2



  parent reply	other threads:[~2020-05-12 21:19 UTC|newest]

Thread overview: 92+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-05-12 21:08 [PATCH v2 00/42] Share DWARF partial symtabs between objfiles Simon Marchi
2020-05-12 21:08 ` [PATCH v2 01/42] Introduce dwarf2_per_objfile::obstack Simon Marchi
2020-05-12 21:08 ` [PATCH v2 02/42] Add "objfile" parameter to two partial_symtab methods Simon Marchi
2020-05-12 21:08 ` [PATCH v2 03/42] Add dwarf2_per_cu_data::index Simon Marchi
2020-05-12 21:08 ` [PATCH v2 04/42] Add dwarf2_per_objfile member to DWARF batons Simon Marchi
2020-05-12 21:08 ` [PATCH v2 05/42] Split dwarf2_per_objfile into dwarf2_per_objfile and dwarf2_per_bfd Simon Marchi
2020-05-12 21:08 ` [PATCH v2 06/42] Remove symtab links from dwarf2_psymtab and dwarf2_per_cu_quick_data Simon Marchi
2020-05-12 21:08 ` [PATCH v2 07/42] Move die_type_hash to dwarf2_per_objfile Simon Marchi
2020-05-12 21:08 ` [PATCH v2 08/42] Add dwarf2_per_objfile field to dwarf2_cu Simon Marchi
2020-05-12 21:08 ` [PATCH v2 09/42] Remove reference to dwarf2_per_cu_data::dwarf2_per_objfile in dw2_do_instantiate_symtab Simon Marchi
2020-05-12 21:11 ` [PATCH v2 10/42] Remove dwarf2_cu->per_cu->dwarf2_per_objfile references Simon Marchi
2020-05-12 21:11 ` [PATCH v2 11/42] Add dwarf2_per_bfd field to dwarf2_per_cu_data Simon Marchi
2020-05-12 21:11 ` [PATCH v2 12/42] Make dwarf2_get_dwz_file take a dwarf2_per_bfd Simon Marchi
2020-05-12 21:11 ` [PATCH v2 13/42] Use bfd_get_filename instead of objfile_name in lookup_dwo_unit Simon Marchi
2020-05-12 21:11 ` [PATCH v2 14/42] Add dwarf2_per_objfile parameter to cutu_reader's constructors Simon Marchi
2020-05-12 21:11 ` [PATCH v2 15/42] Make queue_and_load_dwo_tu receive a dwarf2_cu Simon Marchi
2020-05-22 20:45   ` Tom Tromey
2020-05-25 19:10     ` Simon Marchi
2020-05-12 21:11 ` [PATCH v2 16/42] Remove dwarf2_per_cu_data::dwarf2_per_objfile reference in cutu_reader::keep Simon Marchi
2020-05-12 21:11 ` [PATCH v2 17/42] Add dwarf2_per_objfile parameter to create_partial_symtab Simon Marchi
2020-05-12 21:11 ` [PATCH v2 18/42] Add dwarf2_per_objfile parameter to recursively_compute_inclusions Simon Marchi
2020-05-12 21:11 ` [PATCH v2 19/42] Add dwarf2_per_objfile parameter to process_full_{comp, type}_unit Simon Marchi
2020-05-12 21:12 ` [PATCH v2 20/42] Pass dwarf2_cu objects to dwo-related functions, instead of dwarf2_per_cu_data Simon Marchi
2020-05-12 21:12 ` [PATCH v2 21/42] Remove reference to dwarf2_per_cu_data::dwarf2_per_objfile in queue_and_load_all_dwo_tus Simon Marchi
2020-05-12 21:12 ` [PATCH v2 22/42] Move int type methods out of dwarf2_per_cu_data Simon Marchi
2020-05-12 21:12 ` [PATCH v2 23/42] Add dwarf2_per_objfile to dwarf_expr_context and dwarf2_frame_cache Simon Marchi
2020-05-22 21:04   ` Tom Tromey
2020-05-25 19:50     ` Simon Marchi
2020-05-28  1:44       ` Simon Marchi
2020-05-12 21:12 ` [PATCH v2 24/42] Remove dwarf2_per_cu_data::text_offset Simon Marchi
2020-05-12 21:12 ` [PATCH v2 25/42] Add dwarf2_per_objfile parameter to dwarf2_read_addr_index Simon Marchi
2020-05-12 21:12 ` [PATCH v2 26/42] Add dwarf2_per_objfile parameter to allocate_piece_closure Simon Marchi
2020-05-12 21:12 ` [PATCH v2 27/42] Add dwarf2_per_objfile parameters to dwarf2_fetch_* functions Simon Marchi
2020-05-12 21:12 ` [PATCH v2 28/42] Remove dwarf2_per_cu_data::objfile () Simon Marchi
2020-05-27 16:27   ` Tom de Vries
2020-05-27 16:52     ` Tom de Vries
2020-05-27 20:03     ` Simon Marchi
2020-05-27 21:08       ` Tom de Vries
2020-05-27 21:55         ` Simon Marchi
2020-05-27 22:16           ` Simon Marchi
2020-05-28  2:00             ` Simon Marchi
2020-05-28 13:05               ` Tom de Vries
2020-05-28 15:33                 ` Simon Marchi
2020-05-28 16:24                   ` Christian Biesinger
2020-05-28 16:52                     ` Simon Marchi
2020-05-28 19:49                       ` Simon Marchi
2020-05-12 21:12 ` [PATCH v2 29/42] Add dwarf2_per_objfile parameter to free_one_cached_comp_unit Simon Marchi
2020-05-12 21:17 ` [PATCH v2 30/42] Add dwarf2_per_objfile parameter to get_die_type_at_offset Simon Marchi
2020-05-12 21:17 ` [PATCH v2 31/42] Remove leftover references to dwarf2_per_cu_data::dwarf2_per_objfile Simon Marchi
2020-05-12 21:17 ` [PATCH v2 32/42] Remove dwarf2_per_cu_data::dwarf2_per_objfile Simon Marchi
2020-05-12 21:17 ` [PATCH v2 33/42] Split type_unit_group Simon Marchi
2020-05-13  9:54   ` Tom de Vries
2020-05-13 15:06     ` Simon Marchi
2020-05-12 21:17 ` [PATCH v2 34/42] Move signatured_type::type to unshareable object Simon Marchi
2020-05-12 21:17 ` [PATCH v2 35/42] Pass dwarf2_per_bfd instead of dwarf2_per_objfile to some index-related functions Simon Marchi
2020-05-12 21:17 ` [PATCH v2 36/42] Pass dwarf2_cu to process_full_{comp,type}_unit Simon Marchi
2020-05-12 21:17 ` [PATCH v2 37/42] Make load_cu return the loaded dwarf2_cu Simon Marchi
2020-05-12 21:17 ` [PATCH v2 38/42] Add comp_unit_head to dwarf2_per_cu_data Simon Marchi
2020-05-12 21:17 ` [PATCH v2 39/42] Pass existing_cu object to cutu_reader Simon Marchi
2020-05-22 20:57   ` Tom Tromey
2020-05-25 19:10     ` Simon Marchi
2020-05-12 21:18 ` [PATCH v2 40/42] Replace dwarf2_per_cu_data::cu backlink with per-objfile map Simon Marchi
2020-05-12 21:18 ` [PATCH v2 41/42] Make mapped_debug_names independent of objfile Simon Marchi
2020-05-22 21:01   ` Tom Tromey
2020-05-25 19:53     ` Simon Marchi
2020-05-12 21:18 ` Simon Marchi [this message]
2020-05-13 10:17 ` [PATCH v2 00/42] Share DWARF partial symtabs between objfiles Tom de Vries
2020-05-13 15:46   ` Simon Marchi
2020-05-22 21:02     ` Tom Tromey
2020-05-25 19:53       ` Simon Marchi
2020-05-13 14:52 ` Simon Marchi
2020-05-22 21:07 ` Tom Tromey
2020-05-23 12:24   ` Pedro Alves
2020-05-25 19:56     ` Simon Marchi
2020-05-26 11:17       ` Pedro Alves
2020-05-26 15:35         ` Simon Marchi
2020-05-26 21:34           ` Simon Marchi
2020-05-27  5:08             ` Simon Marchi
2020-05-27 14:53               ` Simon Marchi
2020-05-27 15:51                 ` Simon Marchi
2020-05-29 10:23                   ` Tom de Vries
2020-05-29 12:55                     ` Tom de Vries
2020-05-31  4:16                     ` Simon Marchi
2020-05-31 14:22                       ` Tom de Vries
2020-06-02 21:27                         ` Simon Marchi
2020-06-04 17:55                           ` Tom de Vries
2020-06-04 18:04                             ` Simon Marchi
2020-06-10  3:43                               ` Simon Marchi
2020-06-12  3:25                                 ` Tom Tromey
2020-05-27 14:50 ` [PATCH v2 41.5/42] Move line_header_hash to dwarf2_per_objfile Simon Marchi
2020-05-27 15:07   ` Tom Tromey
2020-05-27 15:10     ` Simon Marchi

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200512211824.6834-43-simon.marchi@efficios.com \
    --to=simon.marchi@efficios.com \
    --cc=gdb-patches@sourceware.org \
    --cc=tom@tromey.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox