From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 26363 invoked by alias); 7 May 2002 15:18:38 -0000 Mailing-List: contact gdb-patches-help@sources.redhat.com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sources.redhat.com Received: (qmail 26325 invoked from network); 7 May 2002 15:18:33 -0000 Received: from unknown (HELO cygnus.com) (205.180.83.203) by sources.redhat.com with SMTP; 7 May 2002 15:18:33 -0000 Received: from localhost.redhat.com (romulus.sfbay.redhat.com [172.16.27.251]) by runyon.cygnus.com (8.8.7-cygnus/8.8.7) with ESMTP id IAA12318 for ; Tue, 7 May 2002 08:18:31 -0700 (PDT) Received: by localhost.redhat.com (Postfix, from userid 469) id 136CD10AAB; Tue, 7 May 2002 11:18:01 -0400 (EDT) From: Elena Zannoni MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <15575.61481.423043.405803@localhost.redhat.com> Date: Tue, 07 May 2002 08:18:00 -0000 To: thorpej@wasabisystems.com Cc: gdb-patches@sources.redhat.com Subject: Re: [PATCH/RFA] Add OS/ABI infrastructure to SuperH target In-Reply-To: <20020507014529.I503@dr-evil.shagadelic.org> References: <20020507014529.I503@dr-evil.shagadelic.org> X-SW-Source: 2002-05/txt/msg00154.txt.bz2 Jason R Thorpe writes: > The following adds OS/ABI handling a'la the ARM and Alpha targets > to the SuperH target. > > Once this is in, I'll be able to commit my NetBSD/sh variant. > > OK to commit? > > * sh-tdep.c (sh_abi_names): Declare. > (process_note_abi_tag_sections): New function. > (get_elfosabi): Ditto. > (sh_gdbarch_register_os_abi): Ditto. > (sh_dump_tdep): Ditto. > (_initialize_sh_tdep): Use gdbarch_register to register > sh_gdbarch_init and sh_dump_tdep. > * config/sh/tm-sh.h (sh_abi): Declare. > (gdbarch_tdep): Add sh_abi and abi_name members. > Ah, great! But hmmm...it would be ok, except that I am afraid there may be some conflict with the sh5 work. In there I defined: enum sh_abi { SH_ABI_UNKNOWN, SH_ABI_32, SH_ABI_64 }; The ABI is unknown for all the architectures except for sh5 where it is set to either 32 or 64 depending on the bfd. I am not saying this is completely correct, note. Is there any chance you could take a look at the sh5 patch I posted a while back, and see if/how your patch integrates (in principle, at least)? The patch is here: http://sources.redhat.com/ml/gdb-patches/2002-02/msg00218.html About your patch, I would prefer if the switch on the os_number used names instead of 0,1,2, is this possible? Same for this line: + if (elfosabi == 0) thanks Elena > -- > -- Jason R. Thorpe > Index: sh-tdep.c > =================================================================== > RCS file: /cvs/src/src/gdb/sh-tdep.c,v > retrieving revision 1.52 > diff -u -r1.52 sh-tdep.c > --- sh-tdep.c 12 Apr 2002 18:18:57 -0000 1.52 > +++ sh-tdep.c 7 May 2002 08:21:59 -0000 > @@ -41,6 +41,8 @@ > #include "regcache.h" > #include "doublest.h" > > +#include "elf-bfd.h" > + > #include "solib-svr4.h" > > void (*sh_show_regs) (void); > @@ -1895,6 +1897,182 @@ > } > #endif /* SVR4_SHARED_LIBS */ > > + > +/* This table matches the indices assigned to enum sh_abi. Keep > + them in sync. */ > +static const char * const sh_abi_names[] = > +{ > + "", > + "GNU/Linux", > + "NetBSD ELF", > + NULL > +}; > + > +static void > +process_note_abi_tag_sections (bfd *abfd, asection *sect, void *obj) > +{ > + enum sh_abi *os_ident_ptr = obj; > + const char *name; > + unsigned int sectsize; > + > + name = bfd_get_section_name (abfd, sect); > + sectsize = bfd_section_size (abfd, sect); > + > + if (strcmp (name, ".note.ABI-tag") == 0 && sectsize > 0) > + { > + unsigned int name_length, data_length, note_type; > + char *note; > + > + /* If the section is larger than this, it's probably not what we are > + looking for. */ > + if (sectsize > 128) > + sectsize = 128; > + > + note = alloca (sectsize); > + > + bfd_get_section_contents (abfd, sect, note, > + (file_ptr) 0, (bfd_size_type) sectsize); > + > + name_length = bfd_h_get_32 (abfd, note); > + data_length = bfd_h_get_32 (abfd, note + 4); > + note_type = bfd_h_get_32 (abfd, note + 8); > + > + if (name_length == 4 && data_length == 16 && note_type == 1 > + && strcmp (note + 12, "GNU") == 0) > + { > + int os_number = bfd_h_get_32 (abfd, note + 16); > + > + /* The case numbers are from abi-tags in glibc. */ > + switch (os_number) > + { > + case 0: > + *os_ident_ptr = SH_ABI_LINUX; > + break; > + > + case 1: > + internal_error > + (__FILE__, __LINE__, > + "process_note_abi_sections: Hurd objects not supported"); > + break; > + > + case 2: > + internal_error > + (__FILE__, __LINE__, > + "process_note_abi_sections: Solaris objects not supported"); > + break; > + > + default: > + internal_error > + (__FILE__, __LINE__, > + "process_note_abi_sections: unknown OS number %d", > + os_number); > + } > + } > + } > + /* NetBSD uses a similar trick. */ > + else if (strcmp (name, ".note.netbsd.ident") == 0 && sectsize > 0) > + { > + unsigned int name_length, desc_length, note_type; > + char *note; > + > + /* If the section is larger than this, it's probably not what we are > + looking for. */ > + if (sectsize > 128) > + sectsize = 128; > + > + note = alloca (sectsize); > + > + bfd_get_section_contents (abfd, sect, note, > + (file_ptr) 0, (bfd_size_type) sectsize); > + > + name_length = bfd_h_get_32 (abfd, note); > + desc_length = bfd_h_get_32 (abfd, note + 4); > + note_type = bfd_h_get_32 (abfd, note + 8); > + > + if (name_length == 7 && desc_length == 4 && note_type == 1 > + && strcmp (note + 12, "NetBSD") == 0) > + /* XXX Should we check the version here? > + Probably not necessary yet. */ > + *os_ident_ptr = SH_ABI_NETBSD_ELF; > + } > +} > + > +static int > +get_elfosabi (bfd *abfd) > +{ > + int elfosabi; > + enum sh_abi sh_abi = SH_ABI_UNKNOWN; > + > + elfosabi = elf_elfheader (abfd)->e_ident[EI_OSABI]; > + > + /* When elfosabi is 0 (ELFOSABI_NONE), this is supposed to indicate > + that we're on a SYSV system. However, GNU/Linux uses a note section > + to record OS/ABI info, but leaves e_ident[EI_OSABI] zero. So we > + have to check the note sections too. */ > + if (elfosabi == 0) > + { > + bfd_map_over_sections (abfd, > + process_note_abi_tag_sections, > + &sh_abi); > + } > + > + if (sh_abi != SH_ABI_UNKNOWN) > + return sh_abi; > + > + switch (elfosabi) > + { > + case ELFOSABI_NONE: > + /* Leave it as unknown. */ > + break; > + > + case ELFOSABI_NETBSD: > + return SH_ABI_NETBSD_ELF; > + > + case ELFOSABI_LINUX: > + return SH_ABI_LINUX; > + } > + > + return SH_ABI_UNKNOWN; > +} > + > +struct sh_abi_handler > +{ > + struct sh_abi_handler *next; > + enum sh_abi abi; > + void (*init_abi)(struct gdbarch_info, struct gdbarch *); > +}; > + > +struct sh_abi_handler *sh_abi_handler_list = NULL; > + > +void > +sh_gdbarch_register_os_abi (enum sh_abi abi, > + void (*init_abi)(struct gdbarch_info, > + struct gdbarch *)) > +{ > + struct sh_abi_handler **handler_p; > + > + for (handler_p = &sh_abi_handler_list; *handler_p != NULL; > + handler_p = &(*handler_p)->next) > + { > + if ((*handler_p)->abi == abi) > + { > + internal_error > + (__FILE__, __LINE__, > + "sh_gdbarch_register_os_abi: A handler for this ABI variant " > + "(%d) has already been registered", (int) abi); > + /* If user wants to continue, override previous definition. */ > + (*handler_p)->init_abi = init_abi; > + return; > + } > + } > + > + (*handler_p) > + = (struct sh_abi_handler *) xmalloc (sizeof (struct sh_abi_handler)); > + (*handler_p)->next = NULL; > + (*handler_p)->abi = abi; > + (*handler_p)->init_abi = init_abi; > +} > + > static gdbarch_init_ftype sh_gdbarch_init; > > static struct gdbarch * > @@ -1906,17 +2084,51 @@ > gdbarch_register_name_ftype *sh_register_name; > gdbarch_store_return_value_ftype *sh_store_return_value; > gdbarch_register_virtual_type_ftype *sh_register_virtual_type; > + enum sh_abi sh_abi = SH_ABI_UNKNOWN; > + struct sh_abi_handler *abi_handler; > + > + /* Try to determine the ABI of the object we are loading. */ > + > + if (info.abfd != NULL) > + { > + switch (bfd_get_flavour (info.abfd)) > + { > + case bfd_target_elf_flavour: > + sh_abi = get_elfosabi (info.abfd); > + break; > + > + default: > + /* Just leave it as "unkown". */ > + break; > + } > + } > > /* Find a candidate among the list of pre-declared architectures. */ > - arches = gdbarch_list_lookup_by_info (arches, &info); > - if (arches != NULL) > - return arches->gdbarch; > + for (arches = gdbarch_list_lookup_by_info (arches, &info); > + arches != NULL; > + arches = gdbarch_list_lookup_by_info (arches->next, &info)) > + { > + /* Make sure the ABI selection matches. */ > + tdep = gdbarch_tdep (arches->gdbarch); > + if (tdep && tdep->sh_abi == sh_abi) > + return arches->gdbarch; > + } > > /* None found, create a new architecture from the information > provided. */ > tdep = XMALLOC (struct gdbarch_tdep); > gdbarch = gdbarch_alloc (&info, tdep); > > + tdep->sh_abi = sh_abi; > + if (sh_abi < SH_ABI_INVALID) > + tdep->abi_name = sh_abi_names[sh_abi]; > + else > + { > + internal_error (__FILE__, __LINE__, "Invalid setting of sh_abi %d", > + (int) sh_abi); > + tdep->abi_name = ""; > + } > + > /* Initialize the register numbers that are not common to all the > variants to -1, if necessary thse will be overwritten in the case > statement below. */ > @@ -2167,15 +2379,58 @@ > set_gdbarch_frame_num_args (gdbarch, frame_num_args_unknown); > set_gdbarch_believe_pcc_promotion (gdbarch, 1); > > + /* Hook in ABI-specific overrides, if they have been registered. If > + the ABI is unknown, this is probably an embedded target, so we > + should not warn about this situation. */ > + if (sh_abi != SH_ABI_UNKNOWN) > + { > + for (abi_handler = sh_abi_handler_list; abi_handler != NULL; > + abi_handler = abi_handler->next) > + if (abi_handler->abi == sh_abi) > + break; > + > + if (abi_handler) > + abi_handler->init_abi (info, gdbarch); > + else > + { > + /* We assume that if GDB_MULTI_ARCH is less than > + GDB_MULTI_ARCH_TM that an ABI variant can be supported by > + overriding definitions in this file. */ > + if (GDB_MULTI_ARCH > GDB_MULTI_ARCH_PARTIAL) > + fprintf_filtered > + (gdb_stderr, > + "A handler for the ABI variant \"%s\" is not built into this " > + "configuration of GDB. " > + "Attempting to continue with the default SuperH settings", > + sh_abi_names[sh_abi]); > + } > + } > + > return gdbarch; > } > > +static void > +sh_dump_tdep (struct gdbarch *current_gdbarch, struct ui_file *file) > +{ > + struct gdbarch_tdep *tdep = gdbarch_tdep (current_gdbarch); > + > + if (tdep == NULL) > + return; > + > + if (tdep->abi_name != NULL) > + fprintf_unfiltered (file, "sh_dump_tdep: ABI = %s\n", tdep->abi_name); > + else > + internal_error (__FILE__, __LINE__, > + "sh_dump_tdep: illegal setting of tdep->sh_abi (%d)", > + (int) tdep->sh_abi); > +} > + > void > _initialize_sh_tdep (void) > { > struct cmd_list_element *c; > > - register_gdbarch_init (bfd_arch_sh, sh_gdbarch_init); > + gdbarch_register (bfd_arch_sh, sh_gdbarch_init, sh_dump_tdep); > > add_com ("regs", class_vars, sh_show_regs_command, "Print all registers"); > } > Index: config/sh/tm-sh.h > =================================================================== > RCS file: /cvs/src/src/gdb/config/sh/tm-sh.h,v > retrieving revision 1.16 > diff -u -r1.16 tm-sh.h > --- config/sh/tm-sh.h 12 Jul 2001 02:29:33 -0000 1.16 > +++ config/sh/tm-sh.h 7 May 2002 08:22:00 -0000 > @@ -23,6 +23,17 @@ > > #define GDB_MULTI_ARCH 1 > > +/* ABI variants that we know about. If you add to this enum, please > + update the table of names in sh-tdep.c. */ > +enum sh_abi > +{ > + SH_ABI_UNKNOWN = 0, > + SH_ABI_LINUX, > + SH_ABI_NETBSD_ELF, > + > + SH_ABI_INVALID /* Keep this last. */ > +}; > + > /* Information that is dependent on the processor variant. */ > > struct gdbarch_tdep > @@ -56,6 +67,9 @@ > int ARGLAST_REGNUM; > int FLOAT_ARGLAST_REGNUM; > int RETURN_REGNUM; > + > + enum sh_abi sh_abi; /* OS/ABI of the inferior */ > + const char *abi_name; /* Name of the above */ > }; > > /* Registers common to all the SH variants. */ > @@ -81,4 +95,3 @@ > #define LITTLE_REMOTE_BREAKPOINT { 0x20, 0xc3 } /* Used in remote.c */ > > /*#define NOP {0x20, 0x0b}*/ /* Who uses this???*/ > -