From patchwork Thu Nov 4 20:46:45 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Iain Sandoe X-Patchwork-Id: 47076 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 B4A2B385803D for ; Thu, 4 Nov 2021 20:47:53 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org B4A2B385803D DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1636058873; bh=fgqfXQvd2QaEJGIBekjlO6eSS1cd8o+udA1U6BuO/8g=; h=To:Subject:Date:List-Id:List-Unsubscribe:List-Archive:List-Post: List-Help:List-Subscribe:From:Reply-To:Cc:From; b=mgJTfutyjRWVXQLzZpi+ceMRh4ltpsSRSQcqwzmreJT8P+AFTTO/V8vlXy870bMRf 9CyiPZlA/10RG3Y598P1qMEDHAi2fuBNoOyPUEq+ny2GqUXqmdJaFIjXhxXWDzY0s8 fP0kdoGoeJ6FOFZCpaeTTqYaJp56sBmWpyYLnc6Y= X-Original-To: gcc-patches@gcc.gnu.org Delivered-To: gcc-patches@gcc.gnu.org Received: from mail-wr1-x42f.google.com (mail-wr1-x42f.google.com [IPv6:2a00:1450:4864:20::42f]) by sourceware.org (Postfix) with ESMTPS id A16BD3857C40 for ; Thu, 4 Nov 2021 20:46:52 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org A16BD3857C40 Received: by mail-wr1-x42f.google.com with SMTP id d24so10593361wra.0 for ; Thu, 04 Nov 2021 13:46:52 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:cc:subject:date:message-id:reply-to :mime-version:content-transfer-encoding; bh=fgqfXQvd2QaEJGIBekjlO6eSS1cd8o+udA1U6BuO/8g=; b=Wj2QKRiUHe4kyjVe94exT9EwgloAdfRwmjIkIoMtcZzLzfH0N4lIfo60HA6wl4DOQy Yqld9UZvSVkYJ7TjvPWe1mPImeDP0HGgkw8BNNrJamZbwGms99JhZsYXXfmJQ+vIEF9q nEvXFublwa3R3YPlATrH7GEbm6qMqlIzlLVPXJ4x0NxXHf5wxeFLXjNKc4J40ShAdPDA xzES6N+FvaihZJjzHcnFSEAVXXbG7DFHmVQVZ5mwg7x/MFUSVEqQyaWpdYKnWgXeImpI m6pAlPbIqzbRU4oU/BqFL9Uz9NJYtKU742+Qnu4WgmoP51zySuT9hYPku5M92TZgDGBT ouTQ== X-Gm-Message-State: AOAM530y2oopU4JrwQ7ayvkk4qUcQD/phj9j1sWQkjC+t8WzHFjnBZPZ o59+so1UR+6HW79XYAjtm26iNjPDua4= X-Google-Smtp-Source: ABdhPJz9LUN1zZ1ARQ6kY4o2GFAfGWQ3m4PXLjiv4qrqu1+SFh8yWaMdOWFP70OmYfKhTCWshUeGKw== X-Received: by 2002:adf:d1c2:: with SMTP id b2mr4978328wrd.114.1636058811469; Thu, 04 Nov 2021 13:46:51 -0700 (PDT) Received: from localhost.localdomain (host81-138-1-83.in-addr.btopenworld.com. [81.138.1.83]) by smtp.gmail.com with ESMTPSA id f133sm6330407wmf.31.2021.11.04.13.46.50 (version=TLS1_2 cipher=ECDHE-ECDSA-AES128-GCM-SHA256 bits=128/128); Thu, 04 Nov 2021 13:46:51 -0700 (PDT) X-Google-Original-From: Iain Sandoe To: gcc-patches@gcc.gnu.org Subject: [PATCH] IPA: Provide a mechanism to register static DTORs via cxa_atexit. Date: Thu, 4 Nov 2021 20:46:45 +0000 Message-Id: <20211104204645.25097-1-iain@sandoe.co.uk> X-Mailer: git-send-email 2.24.3 (Apple Git-128) MIME-Version: 1.0 X-Spam-Status: No, score=-8.5 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, GIT_PATCH_0, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Iain Sandoe via Gcc-patches From: Iain Sandoe Reply-To: iain@sandoe.co.uk Cc: Iain Sandoe , hubicka@kam.mff.cuni.cz Errors-To: gcc-patches-bounces+patchwork=sourceware.org@gcc.gnu.org Sender: "Gcc-patches" For at least one target (Darwin) the platform convention is to register static destructors (i.e. __attribute__((destructor))) with __cxa_atexit rather than placing them into a list that is run by some other mechanism. This patch provides a target hook that allows a target to opt into this and handling for the process in ipa_cdtor_merge (). When the mode is enabled (dtors_from_cxa_atexit is set) we: * Generate new CTORs to register static destructors with __cxa_atexit and add them to the existing list of CTORs; we then process the revised CTORs list. * We sort the DTORs into priority and then TU order, this means that they are registered in that order with __cxa_atexit () and therefore will be run in the reverse order. * Likewise, CTORs are sorted into priority and then TU order, which means that they will run in that order. This matches the behavior of using init/fini (or mod_init_func/mod_term_func) sections. I have intentionally made this an additive patch, since ideally it should be safe for backport, since we currently produce non- conforming code for Darwin from open branches. However, it would seem reasonably easy to re-use the sort compare from this to rectify the mis-ordering of CTORS when LTO is on. (as a separate patch, of course). Tested on x86_64, i686 and powerpc Darwin, x86_64-linux-gnu. comments? OK for master? Eventual backports? thanks Iain Signed-off-by: Iain Sandoe gcc/ChangeLog: * config/darwin.h (TARGET_DTORS_FROM_CXA_ATEXIT): New. * doc/tm.texi: Regenerated. * doc/tm.texi.in: Add TARGET_DTORS_FROM_CXA_ATEXIT hook. * ipa.c (ipa_discover_variable_flags): (cgraph_build_static_cdtor_1): Return the built function decl. (build_cxa_atexit_decl): New. (build_dso_handle_decl): New. (build_cxa_dtor_registrations): New. (compare_cdtor_tu_order): New. (build_cxa_atexit_fns): New. (ipa_cdtor_merge): If dtors_from_cxa_atexit is set, process the DTORs/CTORs accordingly. (pass_ipa_cdtor_merge::gate): Also run if dtors_from_cxa_atexit is set. * target.def (dtors_from_cxa_atexit): New hook. --- gcc/config/darwin.h | 5 ++ gcc/doc/tm.texi | 8 ++ gcc/doc/tm.texi.in | 2 + gcc/ipa.c | 201 +++++++++++++++++++++++++++++++++++++++++++- gcc/target.def | 10 +++ 5 files changed, 222 insertions(+), 4 deletions(-) diff --git a/gcc/config/darwin.h b/gcc/config/darwin.h index 27cb3e4bb30..5202903f5b2 100644 --- a/gcc/config/darwin.h +++ b/gcc/config/darwin.h @@ -54,6 +54,11 @@ see the files COPYING3 and COPYING.RUNTIME respectively. If not, see #define DO_GLOBAL_DTORS_BODY +/* Register static destructors to run from __cxa_atexit instead of putting + them into a .mod_term_funcs section. */ + +#define TARGET_DTORS_FROM_CXA_ATEXIT true + /* The string value for __SIZE_TYPE__. */ #ifndef SIZE_TYPE diff --git a/gcc/doc/tm.texi b/gcc/doc/tm.texi index 78a1af1ad4d..6ec1d50b3e4 100644 --- a/gcc/doc/tm.texi +++ b/gcc/doc/tm.texi @@ -9210,6 +9210,14 @@ collecting constructors and destructors to be run at startup and exit. It is false if we must use @command{collect2}. @end deftypevr +@deftypevr {Target Hook} bool TARGET_DTORS_FROM_CXA_ATEXIT +This value is true if the target wants destructors to be queued to be +run from __cxa_atexit. If this is the case then, for each priority level, +a new constructor will be entered that registers the destructors for that +level with __cxa_atexit (and there will be no destructors emitted). +It is false the method implied by @code{have_ctors_dtors} is used. +@end deftypevr + @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority}) If defined, a function that outputs assembler code to arrange to call the function referenced by @var{symbol} at initialization time. diff --git a/gcc/doc/tm.texi.in b/gcc/doc/tm.texi.in index 4401550989e..2b9960b73d7 100644 --- a/gcc/doc/tm.texi.in +++ b/gcc/doc/tm.texi.in @@ -6015,6 +6015,8 @@ encountering an @code{init_priority} attribute. @hook TARGET_HAVE_CTORS_DTORS +@hook TARGET_DTORS_FROM_CXA_ATEXIT + @hook TARGET_ASM_CONSTRUCTOR @hook TARGET_ASM_DESTRUCTOR diff --git a/gcc/ipa.c b/gcc/ipa.c index 4f62ac183ee..d234a69b9fe 100644 --- a/gcc/ipa.c +++ b/gcc/ipa.c @@ -837,7 +837,7 @@ ipa_discover_variable_flags (void) FINAL specify whether the externally visible name for collect2 should be produced. */ -static void +static tree cgraph_build_static_cdtor_1 (char which, tree body, int priority, bool final, tree optimization, tree target) @@ -916,6 +916,7 @@ cgraph_build_static_cdtor_1 (char which, tree body, int priority, bool final, set_cfun (NULL); current_function_decl = NULL; + return decl; } /* Generate and emit a static constructor or destructor. WHICH must @@ -1022,6 +1023,128 @@ build_cdtor (bool ctor_p, const vec &cdtors) } } +/* Helper functions for build_cxa_dtor_registrations (). + Build a decl for __cxa_atexit (). */ + +static tree +build_cxa_atexit_decl () +{ + /* The parameter to "__cxa_atexit" is "void (*)(void *)". */ + tree fn_type = build_function_type_list (void_type_node, + ptr_type_node, NULL_TREE); + tree fn_ptr_type = build_pointer_type (fn_type); + /* The declaration for `__cxa_atexit' is: + int __cxa_atexit (void (*)(void *), void *, void *). */ + const char *name = "__cxa_atexit"; + tree cxa_name = get_identifier (name); + fn_type = build_function_type_list (integer_type_node, fn_ptr_type, + ptr_type_node, ptr_type_node, NULL_TREE); + tree atexit_fndecl = build_decl (BUILTINS_LOCATION, FUNCTION_DECL, + cxa_name, fn_type); + SET_DECL_ASSEMBLER_NAME (atexit_fndecl, cxa_name); + DECL_VISIBILITY (atexit_fndecl) = VISIBILITY_DEFAULT; + DECL_VISIBILITY_SPECIFIED (atexit_fndecl) = true; + set_call_expr_flags (atexit_fndecl, ECF_LEAF | ECF_NOTHROW); + TREE_PUBLIC (atexit_fndecl) = true; + DECL_EXTERNAL (atexit_fndecl) = true; + DECL_ARTIFICIAL (atexit_fndecl) = true; + return atexit_fndecl; +} + +/* Build a decl for __dso_handle. */ + +static tree +build_dso_handle_decl () +{ + /* Declare the __dso_handle variable. */ + tree dso_handle_decl = build_decl (UNKNOWN_LOCATION, VAR_DECL, + get_identifier ("__dso_handle"), + ptr_type_node); + TREE_PUBLIC (dso_handle_decl) = true; + DECL_EXTERNAL (dso_handle_decl) = true; + DECL_ARTIFICIAL (dso_handle_decl) = true; +#ifdef HAVE_GAS_HIDDEN + if (dso_handle_decl != error_mark_node) + { + DECL_VISIBILITY (dso_handle_decl) = VISIBILITY_HIDDEN; + DECL_VISIBILITY_SPECIFIED (dso_handle_decl) = true; + } +#endif + return dso_handle_decl; +} + +/* This builds one or more constructor functions that register DTORs with + __cxa_atexit (). Within a priority level, DTORs are registered in TU + order - which means that they will run in reverse TU order from cxa_atexit. + This is the same behavior as using a .fini / .mod_term_funcs section. + As the functions are built, they are appended to the CTORs vector. */ + +static void +build_cxa_dtor_registrations (const vec &dtors, vec *ctors) +{ + size_t i,j; + size_t len = dtors.length (); + + location_t sav_loc = input_location; + input_location = UNKNOWN_LOCATION; + + tree atexit_fndecl = build_cxa_atexit_decl (); + tree dso_handle_decl = build_dso_handle_decl (); + + /* We want &__dso_handle. */ + tree dso_ptr = build1_loc (UNKNOWN_LOCATION, ADDR_EXPR, + ptr_type_node, dso_handle_decl); + + i = 0; + while (i < len) + { + priority_type priority = 0; + tree body = NULL_TREE; + j = i; + do + { + priority_type p; + tree fn = dtors[j]; + p = DECL_FINI_PRIORITY (fn); + if (j == i) + priority = p; + else if (p != priority) + break; + j++; + } + while (j < len); + + /* Find the next batch of constructors/destructors with the same + initialization priority. */ + for (;i < j; i++) + { + tree fn = dtors[i]; + DECL_STATIC_DESTRUCTOR (fn) = 0; + tree dtor_ptr = build1_loc (UNKNOWN_LOCATION, ADDR_EXPR, + ptr_type_node, fn); + tree call_cxa_atexit + = build_call_expr_loc (UNKNOWN_LOCATION, atexit_fndecl, 3, + dtor_ptr, null_pointer_node, dso_ptr); + + /* We do not want to optimize away pure/const calls here. + When optimizing, these should be already removed, when not + optimizing, we want user to be able to breakpoint in them. */ + TREE_SIDE_EFFECTS (call_cxa_atexit) = 1; + append_to_statement_list (call_cxa_atexit, &body); + } + + gcc_assert (body != NULL_TREE); + /* Generate a function to register the DTORs at this priority. */ + tree new_ctor + = cgraph_build_static_cdtor_1 ('I', body, priority, true, + DECL_FUNCTION_SPECIFIC_OPTIMIZATION (dtors[0]), + DECL_FUNCTION_SPECIFIC_TARGET (dtors[0])); + /* Add this to the list of ctors. */ + ctors->safe_push (new_ctor); + } + input_location = sav_loc; +} + /* Comparison function for qsort. P1 and P2 are actually of type "tree *" and point to static constructors. DECL_INIT_PRIORITY is used to determine the sort order. */ @@ -1071,7 +1194,43 @@ compare_dtor (const void *p1, const void *p2) else if (priority1 > priority2) return 1; else - /* Ensure a stable sort. */ + /* Ensure a stable sort - into TU order. */ + return DECL_UID (f1) - DECL_UID (f2); +} + +/* Comparison function for qsort. P1 and P2 are of type "tree *" and point to + a pair of static constructors or destructors. We first sort on the basis of + priority and then into TU order (on the strict assumption that DECL_UIDs are + ordered in the same way as the original functions). ???: this seems quite + fragile. */ + +static int +compare_cdtor_tu_order (const void *p1, const void *p2) +{ + tree f1; + tree f2; + int priority1; + int priority2; + + f1 = *(const tree *)p1; + f2 = *(const tree *)p2; + if (DECL_STATIC_CONSTRUCTOR (f1)) + { + priority1 = DECL_INIT_PRIORITY (f1); + priority2 = DECL_INIT_PRIORITY (f2); + } + else + { + priority1 = DECL_FINI_PRIORITY (f1); + priority2 = DECL_FINI_PRIORITY (f2); + } + + if (priority1 < priority2) + return -1; + else if (priority1 > priority2) + return 1; + else + /* For equal priority, sort into the order of definition in the TU. */ return DECL_UID (f1) - DECL_UID (f2); } @@ -1097,6 +1256,37 @@ build_cdtor_fns (vec *ctors, vec *dtors) } } +/* Generate new CTORs to register static destructors with __cxa_atexit and add + them to the existing list of CTORs; we then process the revised CTORs list. + + We sort the DTORs into priority and then TU order, this means that they are + registered in that order with __cxa_atexit () and therefore will be run in + the reverse order. + + Likewise, CTORs are sorted into priority and then TU order, which means that + they will run in that order. + + This matches the behavior of using init/fini or mod_init_func/mod_term_func + sections. */ + +static void +build_cxa_atexit_fns (vec *ctors, vec *dtors) +{ + if (!dtors->is_empty ()) + { + gcc_assert (targetm.dtors_from_cxa_atexit); + dtors->qsort (compare_cdtor_tu_order); + build_cxa_dtor_registrations (*dtors, ctors); + } + + if (!ctors->is_empty ()) + { + gcc_assert (targetm.dtors_from_cxa_atexit); + ctors->qsort (compare_cdtor_tu_order); + build_cdtor (/*ctor_p=*/true, *ctors); + } +} + /* Look for constructors and destructors and produce function calling them. This is needed for targets not supporting ctors or dtors, but we perform the transformation also at linktime to merge possibly numerous @@ -1115,7 +1305,10 @@ ipa_cdtor_merge (void) if (DECL_STATIC_CONSTRUCTOR (node->decl) || DECL_STATIC_DESTRUCTOR (node->decl)) record_cdtor_fn (node, &ctors, &dtors); - build_cdtor_fns (&ctors, &dtors); + if (targetm.dtors_from_cxa_atexit) + build_cxa_atexit_fns (&ctors, &dtors); + else + build_cdtor_fns (&ctors, &dtors); return 0; } @@ -1162,7 +1355,7 @@ pass_ipa_cdtor_merge::gate (function *) /* Perform the pass when we have no ctors/dtors support or at LTO time to merge multiple constructors into single function. */ - return !targetm.have_ctors_dtors || in_lto_p; + return !targetm.have_ctors_dtors || in_lto_p || targetm.dtors_from_cxa_atexit; } } // anon namespace diff --git a/gcc/target.def b/gcc/target.def index 51ea167172b..b803c582f01 100644 --- a/gcc/target.def +++ b/gcc/target.def @@ -6778,6 +6778,16 @@ collecting constructors and destructors to be run at startup and exit.\n\ It is false if we must use @command{collect2}.", bool, false) +/* True if the target wants DTORs to be run from cxa_atexit. */ +DEFHOOKPOD +(dtors_from_cxa_atexit, + "This value is true if the target wants destructors to be queued to be\n\ +run from __cxa_atexit. If this is the case then, for each priority level,\n\ +a new constructor will be entered that registers the destructors for that\n\ +level with __cxa_atexit (and there will be no destructors emitted).\n\ +It is false the method implied by @code{have_ctors_dtors} is used.", + bool, false) + /* True if thread-local storage is supported. */ DEFHOOKPOD (have_tls,