From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-oi1-x22e.google.com (mail-oi1-x22e.google.com [IPv6:2607:f8b0:4864:20::22e]) by sourceware.org (Postfix) with ESMTPS id 8FF803857830 for ; Tue, 13 Apr 2021 14:04:39 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.3.2 sourceware.org 8FF803857830 Received: by mail-oi1-x22e.google.com with SMTP id b3so2313635oie.5 for ; Tue, 13 Apr 2021 07:04:39 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc; bh=WpBi7J60Aa3VnETDt/8froqBnqy4+F9bXbEUKpuVA5U=; b=Kt3KQKZadRQXvO5BSB1J9LZ3L8U6G4FX1KCXfUzkp9vgLh925fiMuxlU2OSGBn+DtW BVzmQffTZ+sAcK9ALaNj1wjOJeWqc0v95bSxiCRsff97YgtJ70QFhr3GUPajPAixZPTV BFd4ybSFNMNMmhzqXcwGvGAJQSYRSOW8gSnWhbjGW5TKIuDP9oDcU/l9aH0VOaL07MQH D8VdA3lM2pzlrge6yO70BXlPjyAgH8nUAqMB+xQs+6CSQc198y8H4ktRlpyxGPT3J7Vx KJt8r1mSIF9hZrzDSm0ZxCGyJ7uiB0qat7ETGeKwRNaxKWqFcbJUmJEBeQNYAo2A9xuP y8/w== X-Gm-Message-State: AOAM531DgOEBPfD4o0g8/C7v4ptNjBEreR7x3RqgSt7Oo5euDWSVKmQK fAjp0sJnPiCr2ZDxBiVXRu5dP0FZKu6y4T84ne8= X-Google-Smtp-Source: ABdhPJxRLbZ+ApMH5Lua2A9w8jrogB5jjg/2wWKWcNCqltPJD3tFtl9X6WGinDnjD+sZm1XfyD8nKXhQnQSY+PIoiNI= X-Received: by 2002:aca:b208:: with SMTP id b8mr122523oif.79.1618322678621; Tue, 13 Apr 2021 07:04:38 -0700 (PDT) MIME-Version: 1.0 References: <13859e5a8b0dab26c922c4d1df63d854957c33b3.1618301209.git.szabolcs.nagy@arm.com> In-Reply-To: <13859e5a8b0dab26c922c4d1df63d854957c33b3.1618301209.git.szabolcs.nagy@arm.com> From: "H.J. Lu" Date: Tue, 13 Apr 2021 07:04:02 -0700 Message-ID: Subject: Re: [PATCH v2 12/14] i386: Remove lazy tlsdesc relocation related code To: Szabolcs Nagy Cc: GNU C Library Content-Type: text/plain; charset="UTF-8" X-Spam-Status: No, score=-3034.9 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, GIT_PATCH_0, KAM_SHORT, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.2 X-Spam-Checker-Version: SpamAssassin 3.4.2 (2018-09-13) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 13 Apr 2021 14:04:43 -0000 On Tue, Apr 13, 2021 at 2:34 AM Szabolcs Nagy via Libc-alpha wrote: > > Like in commit e75711ebfa976d5468ec292282566a18b07e4d67 for x86_64, > remove unused lazy tlsdesc relocation processing code: > > _dl_tlsdesc_resolve_abs_plus_addend > _dl_tlsdesc_resolve_rel > _dl_tlsdesc_resolve_rela > _dl_tlsdesc_resolve_hold > --- > sysdeps/i386/dl-tlsdesc.S | 156 ------------------------- > sysdeps/i386/dl-tlsdesc.h | 6 +- > sysdeps/i386/tlsdesc.c | 231 +------------------------------------- > 3 files changed, 2 insertions(+), 391 deletions(-) > > diff --git a/sysdeps/i386/dl-tlsdesc.S b/sysdeps/i386/dl-tlsdesc.S > index e781d973b7..255fe88651 100644 > --- a/sysdeps/i386/dl-tlsdesc.S > +++ b/sysdeps/i386/dl-tlsdesc.S > @@ -134,159 +134,3 @@ _dl_tlsdesc_dynamic: > cfi_endproc > .size _dl_tlsdesc_dynamic, .-_dl_tlsdesc_dynamic > #endif /* SHARED */ > - > - /* This function is a wrapper for a lazy resolver for TLS_DESC > - REL relocations that reference the *ABS* segment in their own > - link maps. %ebx points to the caller's GOT. %eax points to a > - TLS descriptor, such that 0(%eax) holds the address of the > - resolver wrapper itself (unless some other thread beat us to > - it) and 4(%eax) holds the addend in the relocation. > - > - When the actual resolver returns, it will have adjusted the > - TLS descriptor such that we can tail-call it for it to return > - the TP offset of the symbol. */ > - > - .hidden _dl_tlsdesc_resolve_abs_plus_addend > - .global _dl_tlsdesc_resolve_abs_plus_addend > - .type _dl_tlsdesc_resolve_abs_plus_addend,@function > - cfi_startproc > - .align 16 > -_dl_tlsdesc_resolve_abs_plus_addend: > -0: > - _CET_ENDBR > - pushl %eax > - cfi_adjust_cfa_offset (4) > - pushl %ecx > - cfi_adjust_cfa_offset (4) > - pushl %edx > - cfi_adjust_cfa_offset (4) > - movl $1f - 0b, %ecx > - movl 4(%ebx), %edx > - call _dl_tlsdesc_resolve_abs_plus_addend_fixup > -1: > - popl %edx > - cfi_adjust_cfa_offset (-4) > - popl %ecx > - cfi_adjust_cfa_offset (-4) > - popl %eax > - cfi_adjust_cfa_offset (-4) > - jmp *(%eax) > - cfi_endproc > - .size _dl_tlsdesc_resolve_abs_plus_addend, .-_dl_tlsdesc_resolve_abs_plus_addend > - > - /* This function is a wrapper for a lazy resolver for TLS_DESC > - REL relocations that had zero addends. %ebx points to the > - caller's GOT. %eax points to a TLS descriptor, such that > - 0(%eax) holds the address of the resolver wrapper itself > - (unless some other thread beat us to it) and 4(%eax) holds a > - pointer to the relocation. > - > - When the actual resolver returns, it will have adjusted the > - TLS descriptor such that we can tail-call it for it to return > - the TP offset of the symbol. */ > - > - .hidden _dl_tlsdesc_resolve_rel > - .global _dl_tlsdesc_resolve_rel > - .type _dl_tlsdesc_resolve_rel,@function > - cfi_startproc > - .align 16 > -_dl_tlsdesc_resolve_rel: > -0: > - _CET_ENDBR > - pushl %eax > - cfi_adjust_cfa_offset (4) > - pushl %ecx > - cfi_adjust_cfa_offset (4) > - pushl %edx > - cfi_adjust_cfa_offset (4) > - movl $1f - 0b, %ecx > - movl 4(%ebx), %edx > - call _dl_tlsdesc_resolve_rel_fixup > -1: > - popl %edx > - cfi_adjust_cfa_offset (-4) > - popl %ecx > - cfi_adjust_cfa_offset (-4) > - popl %eax > - cfi_adjust_cfa_offset (-4) > - jmp *(%eax) > - cfi_endproc > - .size _dl_tlsdesc_resolve_rel, .-_dl_tlsdesc_resolve_rel > - > - /* This function is a wrapper for a lazy resolver for TLS_DESC > - RELA relocations. %ebx points to the caller's GOT. %eax > - points to a TLS descriptor, such that 0(%eax) holds the > - address of the resolver wrapper itself (unless some other > - thread beat us to it) and 4(%eax) holds a pointer to the > - relocation. > - > - When the actual resolver returns, it will have adjusted the > - TLS descriptor such that we can tail-call it for it to return > - the TP offset of the symbol. */ > - > - .hidden _dl_tlsdesc_resolve_rela > - .global _dl_tlsdesc_resolve_rela > - .type _dl_tlsdesc_resolve_rela,@function > - cfi_startproc > - .align 16 > -_dl_tlsdesc_resolve_rela: > -0: > - _CET_ENDBR > - pushl %eax > - cfi_adjust_cfa_offset (4) > - pushl %ecx > - cfi_adjust_cfa_offset (4) > - pushl %edx > - cfi_adjust_cfa_offset (4) > - movl $1f - 0b, %ecx > - movl 4(%ebx), %edx > - call _dl_tlsdesc_resolve_rela_fixup > -1: > - popl %edx > - cfi_adjust_cfa_offset (-4) > - popl %ecx > - cfi_adjust_cfa_offset (-4) > - popl %eax > - cfi_adjust_cfa_offset (-4) > - jmp *(%eax) > - cfi_endproc > - .size _dl_tlsdesc_resolve_rela, .-_dl_tlsdesc_resolve_rela > - > - /* This function is a placeholder for lazy resolving of TLS > - relocations. Once some thread starts resolving a TLS > - relocation, it sets up the TLS descriptor to use this > - resolver, such that other threads that would attempt to > - resolve it concurrently may skip the call to the original lazy > - resolver and go straight to a condition wait. > - > - When the actual resolver returns, it will have adjusted the > - TLS descriptor such that we can tail-call it for it to return > - the TP offset of the symbol. */ > - > - .hidden _dl_tlsdesc_resolve_hold > - .global _dl_tlsdesc_resolve_hold > - .type _dl_tlsdesc_resolve_hold,@function > - cfi_startproc > - .align 16 > -_dl_tlsdesc_resolve_hold: > -0: > - _CET_ENDBR > - pushl %eax > - cfi_adjust_cfa_offset (4) > - pushl %ecx > - cfi_adjust_cfa_offset (4) > - pushl %edx > - cfi_adjust_cfa_offset (4) > - movl $1f - 0b, %ecx > - movl 4(%ebx), %edx > - call _dl_tlsdesc_resolve_hold_fixup > -1: > - popl %edx > - cfi_adjust_cfa_offset (-4) > - popl %ecx > - cfi_adjust_cfa_offset (-4) > - popl %eax > - cfi_adjust_cfa_offset (-4) > - jmp *(%eax) > - cfi_endproc > - .size _dl_tlsdesc_resolve_hold, .-_dl_tlsdesc_resolve_hold > diff --git a/sysdeps/i386/dl-tlsdesc.h b/sysdeps/i386/dl-tlsdesc.h > index 753c03e79c..12e90da3a8 100644 > --- a/sysdeps/i386/dl-tlsdesc.h > +++ b/sysdeps/i386/dl-tlsdesc.h > @@ -43,11 +43,7 @@ struct tlsdesc_dynamic_arg > > extern ptrdiff_t attribute_hidden __attribute__ ((regparm (1))) > _dl_tlsdesc_return (struct tlsdesc *), > - _dl_tlsdesc_undefweak (struct tlsdesc *), > - _dl_tlsdesc_resolve_abs_plus_addend (struct tlsdesc *), > - _dl_tlsdesc_resolve_rel (struct tlsdesc *), > - _dl_tlsdesc_resolve_rela (struct tlsdesc *), > - _dl_tlsdesc_resolve_hold (struct tlsdesc *); > + _dl_tlsdesc_undefweak (struct tlsdesc *); > > # ifdef SHARED > extern void *_dl_make_tlsdesc_dynamic (struct link_map *map, > diff --git a/sysdeps/i386/tlsdesc.c b/sysdeps/i386/tlsdesc.c > index 0bc646541f..436a21f66b 100644 > --- a/sysdeps/i386/tlsdesc.c > +++ b/sysdeps/i386/tlsdesc.c > @@ -16,242 +16,13 @@ > License along with the GNU C Library; if not, see > . */ > > -#include > #include > -#include > #include > #include > #include > +#define _dl_tlsdesc_resolve_hold 0 > #include > > -/* The following 4 functions take an entry_check_offset argument. > - It's computed by the caller as an offset between its entry point > - and the call site, such that by adding the built-in return address > - that is implicitly passed to the function with this offset, we can > - easily obtain the caller's entry point to compare with the entry > - point given in the TLS descriptor. If it's changed, we want to > - return immediately. */ > - > -/* This function is used to lazily resolve TLS_DESC REL relocations > - that reference the *ABS* segment in their own link maps. The > - argument is the addend originally stored there. */ > - > -void > -__attribute__ ((regparm (3))) attribute_hidden > -_dl_tlsdesc_resolve_abs_plus_addend_fixup (struct tlsdesc volatile *td, > - struct link_map *l, > - ptrdiff_t entry_check_offset) > -{ > - ptrdiff_t addend = (ptrdiff_t) td->arg; > - > - if (_dl_tlsdesc_resolve_early_return_p (td, __builtin_return_address (0) > - - entry_check_offset)) > - return; > - > -#ifndef SHARED > - CHECK_STATIC_TLS (l, l); > -#else > - if (!TRY_STATIC_TLS (l, l)) > - { > - td->arg = _dl_make_tlsdesc_dynamic (l, addend); > - td->entry = _dl_tlsdesc_dynamic; > - } > - else > -#endif > - { > - td->arg = (void*) (addend - l->l_tls_offset); > - td->entry = _dl_tlsdesc_return; > - } > - > - _dl_tlsdesc_wake_up_held_fixups (); > -} > - > -/* This function is used to lazily resolve TLS_DESC REL relocations > - that originally had zero addends. The argument location, that > - originally held the addend, is used to hold a pointer to the > - relocation, but it has to be restored before we call the function > - that applies relocations. */ > - > -void > -__attribute__ ((regparm (3))) attribute_hidden > -_dl_tlsdesc_resolve_rel_fixup (struct tlsdesc volatile *td, > - struct link_map *l, > - ptrdiff_t entry_check_offset) > -{ > - const ElfW(Rel) *reloc = td->arg; > - > - if (_dl_tlsdesc_resolve_early_return_p (td, __builtin_return_address (0) > - - entry_check_offset)) > - return; > - > - /* The code below was borrowed from _dl_fixup(), > - except for checking for STB_LOCAL. */ > - const ElfW(Sym) *const symtab > - = (const void *) D_PTR (l, l_info[DT_SYMTAB]); > - const char *strtab = (const void *) D_PTR (l, l_info[DT_STRTAB]); > - const ElfW(Sym) *sym = &symtab[ELFW(R_SYM) (reloc->r_info)]; > - lookup_t result; > - > - /* Look up the target symbol. If the normal lookup rules are not > - used don't look in the global scope. */ > - if (ELFW(ST_BIND) (sym->st_info) != STB_LOCAL > - && __builtin_expect (ELFW(ST_VISIBILITY) (sym->st_other), 0) == 0) > - { > - const struct r_found_version *version = NULL; > - > - if (l->l_info[VERSYMIDX (DT_VERSYM)] != NULL) > - { > - const ElfW(Half) *vernum = > - (const void *) D_PTR (l, l_info[VERSYMIDX (DT_VERSYM)]); > - ElfW(Half) ndx = vernum[ELFW(R_SYM) (reloc->r_info)] & 0x7fff; > - version = &l->l_versions[ndx]; > - if (version->hash == 0) > - version = NULL; > - } > - > - result = _dl_lookup_symbol_x (strtab + sym->st_name, l, &sym, > - l->l_scope, version, ELF_RTYPE_CLASS_PLT, > - DL_LOOKUP_ADD_DEPENDENCY, NULL); > - } > - else > - { > - /* We already found the symbol. The module (and therefore its load > - address) is also known. */ > - result = l; > - } > - > - if (!sym) > - { > - td->arg = 0; > - td->entry = _dl_tlsdesc_undefweak; > - } > - else > - { > -# ifndef SHARED > - CHECK_STATIC_TLS (l, result); > -# else > - if (!TRY_STATIC_TLS (l, result)) > - { > - td->arg = _dl_make_tlsdesc_dynamic (result, sym->st_value); > - td->entry = _dl_tlsdesc_dynamic; > - } > - else > -# endif > - { > - td->arg = (void*)(sym->st_value - result->l_tls_offset); > - td->entry = _dl_tlsdesc_return; > - } > - } > - > - _dl_tlsdesc_wake_up_held_fixups (); > -} > - > -/* This function is used to lazily resolve TLS_DESC RELA relocations. > - The argument location is used to hold a pointer to the relocation. */ > - > -void > -__attribute__ ((regparm (3))) attribute_hidden > -_dl_tlsdesc_resolve_rela_fixup (struct tlsdesc volatile *td, > - struct link_map *l, > - ptrdiff_t entry_check_offset) > -{ > - const ElfW(Rela) *reloc = td->arg; > - > - if (_dl_tlsdesc_resolve_early_return_p (td, __builtin_return_address (0) > - - entry_check_offset)) > - return; > - > - /* The code below was borrowed from _dl_fixup(), > - except for checking for STB_LOCAL. */ > - const ElfW(Sym) *const symtab > - = (const void *) D_PTR (l, l_info[DT_SYMTAB]); > - const char *strtab = (const void *) D_PTR (l, l_info[DT_STRTAB]); > - const ElfW(Sym) *sym = &symtab[ELFW(R_SYM) (reloc->r_info)]; > - lookup_t result; > - > - /* Look up the target symbol. If the normal lookup rules are not > - used don't look in the global scope. */ > - if (ELFW(ST_BIND) (sym->st_info) != STB_LOCAL > - && __builtin_expect (ELFW(ST_VISIBILITY) (sym->st_other), 0) == 0) > - { > - const struct r_found_version *version = NULL; > - > - if (l->l_info[VERSYMIDX (DT_VERSYM)] != NULL) > - { > - const ElfW(Half) *vernum = > - (const void *) D_PTR (l, l_info[VERSYMIDX (DT_VERSYM)]); > - ElfW(Half) ndx = vernum[ELFW(R_SYM) (reloc->r_info)] & 0x7fff; > - version = &l->l_versions[ndx]; > - if (version->hash == 0) > - version = NULL; > - } > - > - result = _dl_lookup_symbol_x (strtab + sym->st_name, l, &sym, > - l->l_scope, version, ELF_RTYPE_CLASS_PLT, > - DL_LOOKUP_ADD_DEPENDENCY, NULL); > - } > - else > - { > - /* We already found the symbol. The module (and therefore its load > - address) is also known. */ > - result = l; > - } > - > - if (!sym) > - { > - td->arg = (void*) reloc->r_addend; > - td->entry = _dl_tlsdesc_undefweak; > - } > - else > - { > -# ifndef SHARED > - CHECK_STATIC_TLS (l, result); > -# else > - if (!TRY_STATIC_TLS (l, result)) > - { > - td->arg = _dl_make_tlsdesc_dynamic (result, sym->st_value > - + reloc->r_addend); > - td->entry = _dl_tlsdesc_dynamic; > - } > - else > -# endif > - { > - td->arg = (void*) (sym->st_value - result->l_tls_offset > - + reloc->r_addend); > - td->entry = _dl_tlsdesc_return; > - } > - } > - > - _dl_tlsdesc_wake_up_held_fixups (); > -} > - > -/* This function is used to avoid busy waiting for other threads to > - complete the lazy relocation. Once another thread wins the race to > - relocate a TLS descriptor, it sets the descriptor up such that this > - function is called to wait until the resolver releases the > - lock. */ > - > -void > -__attribute__ ((regparm (3))) attribute_hidden > -_dl_tlsdesc_resolve_hold_fixup (struct tlsdesc volatile *td, > - struct link_map *l __attribute__((__unused__)), > - ptrdiff_t entry_check_offset) > -{ > - /* Maybe we're lucky and can return early. */ > - if (__builtin_return_address (0) - entry_check_offset != td->entry) > - return; > - > - /* Locking here will stop execution until the running resolver runs > - _dl_tlsdesc_wake_up_held_fixups(), releasing the lock. > - > - FIXME: We'd be better off waiting on a condition variable, such > - that we didn't have to hold the lock throughout the relocation > - processing. */ > - __rtld_lock_lock_recursive (GL(dl_load_lock)); > - __rtld_lock_unlock_recursive (GL(dl_load_lock)); > -} > - > - > /* Unmap the dynamic object, but also release its TLS descriptor table > if there is one. */ > > -- > 2.17.1 > LGTM. Thanks. -- H.J.