From patchwork Mon Dec 19 16:45:57 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Tom Tromey X-Patchwork-Id: 62157 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id ED5E53858035 for ; Mon, 19 Dec 2022 16:46:40 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org ED5E53858035 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1671468401; bh=c5jKEWROi4qOwSS90kwTcT0ptoKidUTzdeOI8Osa3Sc=; h=To:Cc:Subject:Date:In-Reply-To:References:List-Id: List-Unsubscribe:List-Archive:List-Post:List-Help:List-Subscribe: From:Reply-To:From; b=kZxPo1K8C33P5pk6eOKVPRE3s56v1+/bEGXZNnABbt3EGve5yK1BsPOZP480YoFYG /17H6UsGyp18Rq9uGHOvxoGGQ66g1S9LIWZTlkQUWsnSPg+3O5oABG2uylxA7jyKPk R/eFXx4qjgnC4K1s2ZMv+pHrwwJfi0hXxH8llJTA= X-Original-To: gdb-patches@sourceware.org Delivered-To: gdb-patches@sourceware.org Received: from mail-il1-x12e.google.com (mail-il1-x12e.google.com [IPv6:2607:f8b0:4864:20::12e]) by sourceware.org (Postfix) with ESMTPS id B6C1B3858D32 for ; Mon, 19 Dec 2022 16:46:09 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org B6C1B3858D32 Received: by mail-il1-x12e.google.com with SMTP id i25so5005141ila.8 for ; Mon, 19 Dec 2022 08:46:09 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=c5jKEWROi4qOwSS90kwTcT0ptoKidUTzdeOI8Osa3Sc=; b=ZWY7nrWYtC+k8YmzhkDqs21eucNCqzvCfLTlfe4WPKLdAniiafXTaJBREfvVhnSG4v tTjGM1khMaZmeSyW7AtrfUcgp+ZKXNxfH8JY2WmBGoI28s7dKPb2Ll+NiAYwnoFBkD73 WzYiexKjr0IO7VIQQv2NIjdoa90OipPi18tPZb2WUL2K3R9FkLX/umIy8HDyLuu+mN6V 1ieoLKevlkX0gj8QZUhOkilTYajNtCtUuO0M3cnBvxOqVlcjdH26agfnOCL3+KAY+GPb xUiN/SgOpl5kf4oYKYZlWvmTTw53AQQyahRf6dbv5CDaeRIKBn51QZmP3mUVw/Ya7bYL iJXQ== X-Gm-Message-State: ANoB5pnNtTnv5PGBsQOACH0Kor5TGXNLmgKoCFvqPCM0/iCpg7y2mUfL oNgWoQXqESWnZmGFAUYLQqM5+tuGjqep/OG1 X-Google-Smtp-Source: AA0mqf6zD35uG8ejEkNXU2QwoO3QE6Sc/rBNjdNXAkopYY+gHW1wSSm7v3BGOTCmNszeQK6hTv4BsA== X-Received: by 2002:a92:2904:0:b0:302:f5cf:ec8a with SMTP id l4-20020a922904000000b00302f5cfec8amr24099917ilg.18.1671468367626; Mon, 19 Dec 2022 08:46:07 -0800 (PST) Received: from localhost.localdomain (97-122-76-186.hlrn.qwest.net. [97.122.76.186]) by smtp.gmail.com with ESMTPSA id x22-20020a056638027600b00363fe31cf55sm3645787jaq.40.2022.12.19.08.46.07 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 19 Dec 2022 08:46:07 -0800 (PST) To: gdb-patches@sourceware.org Cc: Tom Tromey Subject: [PATCH v3 1/2] Only use the per-BFD object to write a DWARF index Date: Mon, 19 Dec 2022 09:45:57 -0700 Message-Id: <20221219164558.378363-2-tromey@adacore.com> X-Mailer: git-send-email 2.38.1 In-Reply-To: <20221219164558.378363-1-tromey@adacore.com> References: <20221219164558.378363-1-tromey@adacore.com> MIME-Version: 1.0 X-Spam-Status: No, score=-10.9 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, OBFU_UNSUB_UL, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org X-BeenThere: gdb-patches@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gdb-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Tom Tromey via Gdb-patches From: Tom Tromey Reply-To: Tom Tromey Errors-To: gdb-patches-bounces+patchwork=sourceware.org@sourceware.org Sender: "Gdb-patches" The DWARF index does not need access to the objfile or per-objfile objects when writing -- it's entirely based on the objfile-independent per-BFD data. This patch implements this idea by changing the entire API to only be passed the per-BFD object. This simplifies some lifetime reasoning for the next patch. This patch removes some code that ensures that the BFD came from a file. It seems to me that checking for the existence of a build-id is good enough for the index cache. --- gdb/dwarf2/index-cache.c | 20 ++++------ gdb/dwarf2/index-cache.h | 4 +- gdb/dwarf2/index-write.c | 81 ++++++++++++++++++---------------------- gdb/dwarf2/index-write.h | 2 +- gdb/dwarf2/read.c | 2 +- 5 files changed, 48 insertions(+), 61 deletions(-) diff --git a/gdb/dwarf2/index-cache.c b/gdb/dwarf2/index-cache.c index 6de58592050..b2902de5dcd 100644 --- a/gdb/dwarf2/index-cache.c +++ b/gdb/dwarf2/index-cache.c @@ -89,23 +89,17 @@ index_cache::disable () /* See dwarf-index-cache.h. */ void -index_cache::store (dwarf2_per_objfile *per_objfile) +index_cache::store (dwarf2_per_bfd *per_bfd) { - objfile *obj = per_objfile->objfile; - if (!enabled ()) return; - /* If the objfile does not correspond to an actual file, skip it. */ - if ((obj->flags & OBJF_NOT_FILENAME) != 0) - return; - /* Get build id of objfile. */ - const bfd_build_id *build_id = build_id_bfd_get (obj->obfd.get ()); + const bfd_build_id *build_id = build_id_bfd_get (per_bfd->obfd); if (build_id == nullptr) { index_cache_debug ("objfile %s has no build id", - objfile_name (obj)); + bfd_get_filename (per_bfd->obfd)); return; } @@ -113,7 +107,7 @@ index_cache::store (dwarf2_per_objfile *per_objfile) /* Get build id of dwz file, if present. */ gdb::optional dwz_build_id_str; - const dwz_file *dwz = dwarf2_get_dwz_file (per_objfile->per_bfd); + const dwz_file *dwz = dwarf2_get_dwz_file (per_bfd); const char *dwz_build_id_ptr = NULL; if (dwz != nullptr) @@ -148,18 +142,18 @@ index_cache::store (dwarf2_per_objfile *per_objfile) } index_cache_debug ("writing index cache for objfile %s", - objfile_name (obj)); + bfd_get_filename (per_bfd->obfd)); /* Write the index itself to the directory, using the build id as the filename. */ - write_dwarf_index (per_objfile, m_dir.c_str (), + write_dwarf_index (per_bfd, m_dir.c_str (), build_id_str.c_str (), dwz_build_id_ptr, dw_index_kind::GDB_INDEX); } catch (const gdb_exception_error &except) { index_cache_debug ("couldn't store index cache for objfile %s: %s", - objfile_name (obj), except.what ()); + bfd_get_filename (per_bfd->obfd), except.what ()); } } diff --git a/gdb/dwarf2/index-cache.h b/gdb/dwarf2/index-cache.h index 6366a9a9360..a6ac0112e41 100644 --- a/gdb/dwarf2/index-cache.h +++ b/gdb/dwarf2/index-cache.h @@ -24,6 +24,8 @@ #include "gdbsupport/array-view.h" #include "symfile.h" +class dwarf2_per_bfd; + /* Base of the classes used to hold the resources of the indices loaded from the cache (e.g. mmapped files). */ @@ -53,7 +55,7 @@ class index_cache void disable (); /* Store an index for the specified object file in the cache. */ - void store (dwarf2_per_objfile *per_objfile); + void store (dwarf2_per_bfd *per_bfd); /* Look for an index file matching BUILD_ID. If found, return the contents as an array_view and store the underlying resources (allocated memory, diff --git a/gdb/dwarf2/index-write.c b/gdb/dwarf2/index-write.c index 3d215a6307b..a63c461da0b 100644 --- a/gdb/dwarf2/index-write.c +++ b/gdb/dwarf2/index-write.c @@ -522,7 +522,7 @@ write_address_map (struct addrmap *addrmap, data_buf &addr_vec, class debug_names { public: - debug_names (dwarf2_per_objfile *per_objfile, bool is_dwarf64, + debug_names (dwarf2_per_bfd *per_bfd, bool is_dwarf64, bfd_endian dwarf5_byte_order) : m_dwarf5_byte_order (dwarf5_byte_order), m_dwarf32 (dwarf5_byte_order), @@ -532,7 +532,7 @@ class debug_names : static_cast (m_dwarf32)), m_name_table_string_offs (m_dwarf.name_table_string_offs), m_name_table_entry_offs (m_dwarf.name_table_entry_offs), - m_debugstrlookup (per_objfile) + m_debugstrlookup (per_bfd) {} int dwarf5_offset_size () const @@ -785,23 +785,23 @@ class debug_names { public: - /* Object constructor to be called for current DWARF2_PER_OBJFILE. + /* Object constructor to be called for current DWARF2_PER_BFD. All .debug_str section strings are automatically stored. */ - debug_str_lookup (dwarf2_per_objfile *per_objfile) - : m_abfd (per_objfile->objfile->obfd.get ()), - m_per_objfile (per_objfile) + debug_str_lookup (dwarf2_per_bfd *per_bfd) + : m_abfd (per_bfd->obfd), + m_per_bfd (per_bfd) { - per_objfile->per_bfd->str.read (per_objfile->objfile); - if (per_objfile->per_bfd->str.buffer == NULL) + gdb_assert (per_bfd->str.readin); + if (per_bfd->str.buffer == NULL) return; - for (const gdb_byte *data = per_objfile->per_bfd->str.buffer; - data < (per_objfile->per_bfd->str.buffer - + per_objfile->per_bfd->str.size);) + for (const gdb_byte *data = per_bfd->str.buffer; + data < (per_bfd->str.buffer + + per_bfd->str.size);) { const char *const s = reinterpret_cast (data); const auto insertpair = m_str_table.emplace (c_str_view (s), - data - per_objfile->per_bfd->str.buffer); + data - per_bfd->str.buffer); if (!insertpair.second) complaint (_("Duplicate string \"%s\" in " ".debug_str section [in module %s]"), @@ -818,7 +818,7 @@ class debug_names const auto it = m_str_table.find (c_str_view (s)); if (it != m_str_table.end ()) return it->second; - const size_t offset = (m_per_objfile->per_bfd->str.size + const size_t offset = (m_per_bfd->str.size + m_str_add_buf.size ()); m_str_table.emplace (c_str_view (s), offset); m_str_add_buf.append_cstr0 (s); @@ -834,7 +834,7 @@ class debug_names private: std::unordered_map m_str_table; bfd *const m_abfd; - dwarf2_per_objfile *m_per_objfile; + dwarf2_per_bfd *m_per_bfd; /* Data to add at the end of .debug_str for new needed symbol names. */ data_buf m_str_add_buf; @@ -1046,9 +1046,9 @@ class debug_names .debug_names section. */ static bool -check_dwarf64_offsets (dwarf2_per_objfile *per_objfile) +check_dwarf64_offsets (dwarf2_per_bfd *per_bfd) { - for (const auto &per_cu : per_objfile->per_bfd->all_units) + for (const auto &per_cu : per_bfd->all_units) { if (to_underlying (per_cu->sect_off) >= (static_cast (1) << 32)) @@ -1197,7 +1197,7 @@ write_cooked_index (cooked_index_vector *table, associated dwz file, DWZ_OUT_FILE must be NULL. */ static void -write_gdbindex (dwarf2_per_objfile *per_objfile, +write_gdbindex (dwarf2_per_bfd *per_bfd, cooked_index_vector *table, FILE *out_file, FILE *dwz_out_file) { @@ -1210,7 +1210,7 @@ write_gdbindex (dwarf2_per_objfile *per_objfile, in the index file). This will later be needed to write the address table. */ cu_index_map cu_index_htab; - cu_index_htab.reserve (per_objfile->per_bfd->all_units.size ()); + cu_index_htab.reserve (per_bfd->all_units.size ()); /* Store out the .debug_type CUs, if any. */ data_buf types_cu_list; @@ -1221,10 +1221,9 @@ write_gdbindex (dwarf2_per_objfile *per_objfile, int counter = 0; int types_counter = 0; - for (int i = 0; i < per_objfile->per_bfd->all_units.size (); ++i) + for (int i = 0; i < per_bfd->all_units.size (); ++i) { - dwarf2_per_cu_data *per_cu - = per_objfile->per_bfd->all_units[i].get (); + dwarf2_per_cu_data *per_cu = per_bfd->all_units[i].get (); int &this_counter = per_cu->is_debug_types ? types_counter : counter; @@ -1287,27 +1286,25 @@ static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 }; many bytes were expected to be written into OUT_FILE. */ static void -write_debug_names (dwarf2_per_objfile *per_objfile, +write_debug_names (dwarf2_per_bfd *per_bfd, cooked_index_vector *table, FILE *out_file, FILE *out_file_str) { - const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (per_objfile); - struct objfile *objfile = per_objfile->objfile; + const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (per_bfd); const enum bfd_endian dwarf5_byte_order - = gdbarch_byte_order (objfile->arch ()); + = bfd_big_endian (per_bfd->obfd) ? BFD_ENDIAN_BIG : BFD_ENDIAN_LITTLE; /* The CU list is already sorted, so we don't need to do additional work here. Also, the debug_types entries do not appear in all_units, but only in their own hash table. */ data_buf cu_list; data_buf types_cu_list; - debug_names nametable (per_objfile, dwarf5_is_dwarf64, dwarf5_byte_order); + debug_names nametable (per_bfd, dwarf5_is_dwarf64, dwarf5_byte_order); int counter = 0; int types_counter = 0; - for (int i = 0; i < per_objfile->per_bfd->all_units.size (); ++i) + for (int i = 0; i < per_bfd->all_units.size (); ++i) { - dwarf2_per_cu_data *per_cu - = per_objfile->per_bfd->all_units[i].get (); + dwarf2_per_cu_data *per_cu = per_bfd->all_units[i].get (); int &this_counter = per_cu->is_debug_types ? types_counter : counter; data_buf &this_list = per_cu->is_debug_types ? types_cu_list : cu_list; @@ -1320,8 +1317,8 @@ write_debug_names (dwarf2_per_objfile *per_objfile, } /* Verify that all units are represented. */ - gdb_assert (counter == per_objfile->per_bfd->all_comp_units.size ()); - gdb_assert (types_counter == per_objfile->per_bfd->all_type_units.size ()); + gdb_assert (counter == per_bfd->all_units.size ()); + gdb_assert (types_counter == per_bfd->all_type_units.size ()); for (const cooked_index_entry *entry : table->all_entries ()) nametable.insert (entry); @@ -1454,23 +1451,17 @@ struct index_wip_file /* See dwarf-index-write.h. */ void -write_dwarf_index (dwarf2_per_objfile *per_objfile, const char *dir, +write_dwarf_index (dwarf2_per_bfd *per_bfd, const char *dir, const char *basename, const char *dwz_basename, dw_index_kind index_kind) { - struct objfile *objfile = per_objfile->objfile; - - if (per_objfile->per_bfd->index_table == nullptr) + if (per_bfd->index_table == nullptr) error (_("No debugging symbols")); - cooked_index_vector *table - = per_objfile->per_bfd->index_table->index_for_writing (); + cooked_index_vector *table = per_bfd->index_table->index_for_writing (); - if (per_objfile->per_bfd->types.size () > 1) + if (per_bfd->types.size () > 1) error (_("Cannot make an index when the file has multiple .debug_types sections")); - - gdb_assert ((objfile->flags & OBJF_NOT_FILENAME) == 0); - const char *index_suffix = (index_kind == dw_index_kind::DEBUG_NAMES ? INDEX5_SUFFIX : INDEX4_SUFFIX); @@ -1484,13 +1475,13 @@ write_dwarf_index (dwarf2_per_objfile *per_objfile, const char *dir, { index_wip_file str_wip_file (dir, basename, DEBUG_STR_SUFFIX); - write_debug_names (per_objfile, table, objfile_index_wip.out_file.get (), + write_debug_names (per_bfd, table, objfile_index_wip.out_file.get (), str_wip_file.out_file.get ()); str_wip_file.finalize (); } else - write_gdbindex (per_objfile, table, objfile_index_wip.out_file.get (), + write_gdbindex (per_bfd, table, objfile_index_wip.out_file.get (), (dwz_index_wip.has_value () ? dwz_index_wip->out_file.get () : NULL)); @@ -1545,8 +1536,8 @@ save_gdb_index_command (const char *arg, int from_tty) if (dwz != NULL) dwz_basename = lbasename (dwz->filename ()); - write_dwarf_index (per_objfile, arg, basename, dwz_basename, - index_kind); + write_dwarf_index (per_objfile->per_bfd, arg, basename, + dwz_basename, index_kind); } catch (const gdb_exception_error &except) { diff --git a/gdb/dwarf2/index-write.h b/gdb/dwarf2/index-write.h index b8855dcae7c..3a9c1432c7f 100644 --- a/gdb/dwarf2/index-write.h +++ b/gdb/dwarf2/index-write.h @@ -33,7 +33,7 @@ same, but for the dwz file's index. */ extern void write_dwarf_index - (dwarf2_per_objfile *per_objfile, const char *dir, const char *basename, + (dwarf2_per_bfd *per_bfd, const char *dir, const char *basename, const char *dwz_basename, dw_index_kind index_kind); #endif /* DWARF_INDEX_WRITE_H */ diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c index ddea38cf596..fc8239b4261 100644 --- a/gdb/dwarf2/read.c +++ b/gdb/dwarf2/read.c @@ -5434,7 +5434,7 @@ dwarf2_build_psymtabs (struct objfile *objfile) dwarf2_build_psymtabs_hard (per_objfile); /* (maybe) store an index in the cache. */ - global_index_cache.store (per_objfile); + global_index_cache.store (per_objfile->per_bfd); } catch (const gdb_exception_error &except) { From patchwork Mon Dec 19 16:45:58 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Tom Tromey X-Patchwork-Id: 62156 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 0B1153858035 for ; Mon, 19 Dec 2022 16:46:37 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 0B1153858035 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1671468397; bh=QaddITAsFaiaXhBT8uj16QX4FyHxOZAmSo4+wYoaGwQ=; h=To:Cc:Subject:Date:In-Reply-To:References:List-Id: List-Unsubscribe:List-Archive:List-Post:List-Help:List-Subscribe: From:Reply-To:From; b=wP7JsYcsXED85YoLwh8UpUZxFarR2Cjv5mZZdT3hsT1qX0QdaQzVjur/G/4joh87w 66oOnkR6YmsZeJTX7MaPfB4SEy+va8ETdbehunmrSqplTZlLXCbVyt8LpVtUJgZgLu 4alf2H/bJ8R+UsHboBbiBHzm0nKIgYlkrQR3bIOA= X-Original-To: gdb-patches@sourceware.org Delivered-To: gdb-patches@sourceware.org Received: from mail-il1-x134.google.com (mail-il1-x134.google.com [IPv6:2607:f8b0:4864:20::134]) by sourceware.org (Postfix) with ESMTPS id 56CD23858D28 for ; Mon, 19 Dec 2022 16:46:09 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 56CD23858D28 Received: by mail-il1-x134.google.com with SMTP id y2so5010521ily.5 for ; Mon, 19 Dec 2022 08:46:09 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=QaddITAsFaiaXhBT8uj16QX4FyHxOZAmSo4+wYoaGwQ=; b=2q8Ym7q7EXDOPnBrVW9aNBpOBKTnJ5avRZAlA2v3PCDMpMJwtrow/SMnUCljrkcSiI pvhxgZrKvT3/iH9wo9JyZz+wt468ucA5fcXu5ZGx7jAeKZF8yURjC0CaMho1Nh+Jn/OC aHCn+Uo25uo2zX987SFnXZi2reb1navpJ79Pea6rduFhsdven/Sy31WLTy0gN1LfdDhr NkAbf/3BQAf+QCx9uUtzZuLa43fTbsdmm3cwsFg7FaThOsoNZRnua7KebNbGowS4lH8y DdOo5ZiQIQS7nJ40DIkZK/kYhI2yU5Xb9sJhkzFLvECLroVBXn9RQ5lvNe+krrwEKYRX foow== X-Gm-Message-State: ANoB5plzfPuz1kADBLE/a2jw03S1pEI3CxQukNEGniUcaIwEIxJkV9fY eLVfaXmveAMbLOWncdF77C7SznWy0KQhVM8g X-Google-Smtp-Source: AA0mqf55XPZOaIRPp8a15aKNO9kkpYp/wm+hh5oTIhSb+WIYAsE+s6KOUy4Vc8f4rKuI2KQGN4qP9A== X-Received: by 2002:a92:c911:0:b0:303:9265:a73c with SMTP id t17-20020a92c911000000b003039265a73cmr26197510ilp.22.1671468368352; Mon, 19 Dec 2022 08:46:08 -0800 (PST) Received: from localhost.localdomain (97-122-76-186.hlrn.qwest.net. [97.122.76.186]) by smtp.gmail.com with ESMTPSA id x22-20020a056638027600b00363fe31cf55sm3645787jaq.40.2022.12.19.08.46.07 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 19 Dec 2022 08:46:07 -0800 (PST) To: gdb-patches@sourceware.org Cc: Tom Tromey Subject: [PATCH v3 2/2] Write the DWARF index in the background Date: Mon, 19 Dec 2022 09:45:58 -0700 Message-Id: <20221219164558.378363-3-tromey@adacore.com> X-Mailer: git-send-email 2.38.1 In-Reply-To: <20221219164558.378363-1-tromey@adacore.com> References: <20221219164558.378363-1-tromey@adacore.com> MIME-Version: 1.0 X-Spam-Status: No, score=-11.4 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org X-BeenThere: gdb-patches@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gdb-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Tom Tromey via Gdb-patches From: Tom Tromey Reply-To: Tom Tromey Errors-To: gdb-patches-bounces+patchwork=sourceware.org@sourceware.org Sender: "Gdb-patches" The new DWARF cooked indexer interacts poorly with the DWARF index cache. In particular, the cache will require gdb to wait for the cooked index to be finalized. As this happens in the foreground, it means that users with this setting enabled will see a slowdown. This patch changes gdb to write the cache entry a worker thread. (As usual, in the absence of threads, this work is simply done immediately in the main thread.) Some care is taken to ensure that this can't crash, and that gdb will not exit before the task is complete. To avoid use-after-free problems, the DWARF per-BFD object explicitly waits for the index cache task to complete. To avoid gdb exiting early, an exit observer is used to wait for all such pending tasks. In normal use, neither of these waits will be very visible. For users using "-batch" to pre-generate the index, though, it would be. However I don't think there is much to be done about this, as it was the status quo ante. --- gdb/dwarf2/cooked-index.c | 73 ++++++++++++++++++++++++++++++++++++++- gdb/dwarf2/cooked-index.h | 31 ++++++++++------- gdb/dwarf2/mapped-index.h | 9 +++++ gdb/dwarf2/read.c | 11 +++--- 4 files changed, 107 insertions(+), 17 deletions(-) diff --git a/gdb/dwarf2/cooked-index.c b/gdb/dwarf2/cooked-index.c index 0aa026c7779..172a4dd96d2 100644 --- a/gdb/dwarf2/cooked-index.c +++ b/gdb/dwarf2/cooked-index.c @@ -20,11 +20,20 @@ #include "defs.h" #include "dwarf2/cooked-index.h" #include "dwarf2/read.h" +#include "dwarf2/index-cache.h" #include "cp-support.h" #include "c-lang.h" #include "ada-lang.h" #include "split-name.h" +#include "observable.h" +#include "run-on-main-thread.h" #include +#include + +/* We don't want gdb to exit while it is in the process of writing to + the index cache. So, all live cooked index vectors are stored + here, and then these are all waited for before exit proceeds. */ +static std::unordered_set active_vectors; /* Hash function for cooked_index_entry. */ @@ -282,11 +291,46 @@ cooked_index::find (gdb::string_view name, bool completing) return range (lower, upper); } -cooked_index_vector::cooked_index_vector (vec_type &&vec) +cooked_index_vector::cooked_index_vector (vec_type &&vec, + dwarf2_per_bfd *per_bfd) : m_vector (std::move (vec)) { for (auto &idx : m_vector) idx->finalize (); + + /* This must be set after all the finalization tasks have been + started, because it may call 'wait'. */ + m_write_future + = gdb::thread_pool::g_thread_pool->post_task ([this, per_bfd] () + { + maybe_write_index (per_bfd); + }); + + /* ACTIVE_VECTORS is not locked, and this assert ensures that this + will be caught if ever moved to the background. */ + gdb_assert (is_main_thread ()); + active_vectors.insert (this); +} + +cooked_index_vector::~cooked_index_vector () +{ + /* The 'finalize' method may be run in a different thread. If + this object is destroyed before this completes, then the method + will end up writing to freed memory. Waiting for this to + complete avoids this problem; and the cost seems ignorable + because creating and immediately destroying the debug info is a + relatively rare thing to do. */ + wait (); + + /* Likewise for the index-creating future, though this one must also + waited for by the per-BFD object to ensure the required data + remains live. */ + wait_completely (); + + /* Remove our entry from the global list. See the assert in the + constructor to understand this. */ + gdb_assert (is_main_thread ()); + active_vectors.erase (this); } /* See cooked-index.h. */ @@ -345,3 +389,30 @@ cooked_index_vector::get_main () const return result; } + +void +cooked_index_vector::maybe_write_index (dwarf2_per_bfd *per_bfd) +{ + /* Wait for finalization. */ + wait (); + + /* (maybe) store an index in the cache. */ + global_index_cache.store (per_bfd); +} + +/* Wait for all the index cache entries to be written before gdb + exits. */ +static void +wait_for_index_cache (int) +{ + gdb_assert (is_main_thread ()); + for (cooked_index_vector *item : active_vectors) + item->wait_completely (); +} + +void _initialize_cooked_index (); +void +_initialize_cooked_index () +{ + gdb::observers::gdb_exiting.attach (wait_for_index_cache, "cooked-index"); +} diff --git a/gdb/dwarf2/cooked-index.h b/gdb/dwarf2/cooked-index.h index 2ea32781be5..fb901c6f87f 100644 --- a/gdb/dwarf2/cooked-index.h +++ b/gdb/dwarf2/cooked-index.h @@ -36,6 +36,7 @@ #include "gdbsupport/range-chain.h" struct dwarf2_per_cu_data; +struct dwarf2_per_bfd; /* Flags that describe an entry in the index. */ enum cooked_index_flag_enum : unsigned char @@ -307,7 +308,8 @@ class cooked_index_vector : public dwarf_scanner_base object. */ typedef std::vector> vec_type; - explicit cooked_index_vector (vec_type &&vec); + cooked_index_vector (vec_type &&vec, dwarf2_per_bfd *per_bfd); + ~cooked_index_vector () override; DISABLE_COPY_AND_ASSIGN (cooked_index_vector); /* Wait until the finalization of the entire cooked_index_vector is @@ -318,17 +320,6 @@ class cooked_index_vector : public dwarf_scanner_base item->wait (); } - ~cooked_index_vector () - { - /* The 'finalize' methods may be run in a different thread. If - this object is destroyed before these complete, then one will - end up writing to freed memory. Waiting for finalization to - complete avoids this problem; and the cost seems ignorable - because creating and immediately destroying the debug info is a - relatively rare thing to do. */ - wait (); - } - /* A range over a vector of subranges. */ typedef range_chain range; @@ -367,11 +358,27 @@ class cooked_index_vector : public dwarf_scanner_base quick_symbol_functions_up make_quick_functions () const override; + /* Wait for the index to be completely finished. For ordinary uses, + the index code ensures this itself -- e.g., 'all_entries' will + wait on the 'finalize' future. However, on destruction, if an + index is being written, it's also necessary to wait for that to + complete. */ + void wait_completely () override + { + m_write_future.wait (); + } + private: + /* Maybe write the index to the index cache. */ + void maybe_write_index (dwarf2_per_bfd *per_bfd); + /* The vector of cooked_index objects. This is stored because the entries are stored on the obstacks in those objects. */ vec_type m_vector; + + /* A future that tracks when the 'index_write' method is done. */ + std::future m_write_future; }; #endif /* GDB_DWARF2_COOKED_INDEX_H */ diff --git a/gdb/dwarf2/mapped-index.h b/gdb/dwarf2/mapped-index.h index 7d71347f9f4..e0c3a447991 100644 --- a/gdb/dwarf2/mapped-index.h +++ b/gdb/dwarf2/mapped-index.h @@ -73,6 +73,15 @@ struct dwarf_scanner_base will return 'this' as a cooked index. For other forms, it will throw an exception with an appropriate error message. */ virtual cooked_index_vector *index_for_writing () = 0; + + /* Wait for reading of the debuginfo to be completely finished. + This normally has a trivial implementation, but if a subclass + does any background reading, it's needed to ensure that the + reading is completed before destroying the containing per-BFD + object. */ + virtual void wait_completely () + { + } }; /* Base class containing bits shared by both .gdb_index and diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c index fc8239b4261..6d27fbf1d75 100644 --- a/gdb/dwarf2/read.c +++ b/gdb/dwarf2/read.c @@ -1467,6 +1467,11 @@ dwarf2_per_bfd::dwarf2_per_bfd (bfd *obfd, const dwarf2_debug_sections *names, dwarf2_per_bfd::~dwarf2_per_bfd () { + /* Data from the per-BFD may be needed when finalizing the cooked + index table, so wait here while this happens. */ + if (index_table != nullptr) + index_table->wait_completely (); + for (auto &per_cu : all_units) { per_cu->imported_symtabs_free (); @@ -5432,9 +5437,6 @@ dwarf2_build_psymtabs (struct objfile *objfile) try { dwarf2_build_psymtabs_hard (per_objfile); - - /* (maybe) store an index in the cache. */ - global_index_cache.store (per_objfile->per_bfd); } catch (const gdb_exception_error &except) { @@ -7182,7 +7184,8 @@ dwarf2_build_psymtabs_hard (dwarf2_per_objfile *per_objfile) indexes.end ()); indexes.shrink_to_fit (); - cooked_index_vector *vec = new cooked_index_vector (std::move (indexes)); + cooked_index_vector *vec = new cooked_index_vector (std::move (indexes), + per_bfd); per_bfd->index_table.reset (vec); const cooked_index_entry *main_entry = vec->get_main ();