From: Mark Wielaard <mark@klomp.org>
To: ying.huang@oss.cipunited.com, elfutils-devel@sourceware.org
Subject: Re: [PATCH 5/5] backends: Fix run-native-test.sh and run-funcretval++11.sh run fail on mips
Date: Thu, 11 May 2023 18:38:13 +0200 [thread overview]
Message-ID: <4958c19be43be2315c26d104b1ada85dbde89142.camel@klomp.org> (raw)
In-Reply-To: <20230411081141.1762395-6-ying.huang@oss.cipunited.com>
Hi,
On Tue, 2023-04-11 at 16:12 +0800, Ying Huang wrote:
> From: Ying Huang <ying.huang@oss.cipunited.com>
>
> add register_info, return_value_location function on mips
> ---
> backends/Makefile.am | 2 +-
> backends/mips_init.c | 2 +
> backends/mips_regs.c | 109 +++++++++++++++++
> backends/mips_retval.c | 261 +++++++++++++++++++++++++++++++++++++++++
> 4 files changed, 373 insertions(+), 1 deletion(-)
> create mode 100644 backends/mips_regs.c
> create mode 100644 backends/mips_retval.c
>
> diff --git a/backends/Makefile.am b/backends/Makefile.am
> index ddc31c9d..5453f787 100644
> --- a/backends/Makefile.am
> +++ b/backends/Makefile.am
> @@ -101,7 +101,7 @@ loongarch_SRCS = loongarch_init.c loongarch_symbol.c
> arc_SRCS = arc_init.c arc_symbol.c
>
> mips_SRCS = mips_init.c mips_symbol.c mips_attrs.c mips_initreg.c \
> - mips_cfi.c mips_unwind.c
> + mips_cfi.c mips_unwind.c mips_regs.c mips_retval.c
OK.
> libebl_backends_a_SOURCES = $(i386_SRCS) $(sh_SRCS) $(x86_64_SRCS) \
> $(ia64_SRCS) $(alpha_SRCS) $(arm_SRCS) \
> diff --git a/backends/mips_init.c b/backends/mips_init.c
> index 3caa9fee..7ca93314 100644
> --- a/backends/mips_init.c
> +++ b/backends/mips_init.c
> @@ -58,6 +58,8 @@ mips_init (Elf *elf __attribute__ ((unused)),
> HOOK (eh, set_initial_registers_tid);
> HOOK (eh, abi_cfi);
> HOOK (eh, unwind);
> + HOOK (eh, register_info);
> + HOOK (eh, return_value_location);
> eh->frame_nregs = 32;
> return eh;
> }
OK
> diff --git a/backends/mips_regs.c b/backends/mips_regs.c
> new file mode 100644
> index 00000000..733caeee
> --- /dev/null
> +++ b/backends/mips_regs.c
> @@ -0,0 +1,109 @@
> +/* Register names and numbers for mips DWARF.
> + Copyright (C) 2006 Red Hat, Inc.
> + Copyright (C) 2023 CIP United Inc.
> + This file is part of elfutils.
> +
> + This file is free software; you can redistribute it and/or modify
> + it under the terms of either
> +
> + * the GNU Lesser General Public License as published by the Free
> + Software Foundation; either version 3 of the License, or (at
> + your option) any later version
> +
> + or
> +
> + * the GNU General Public License as published by the Free
> + Software Foundation; either version 2 of the License, or (at
> + your option) any later version
> +
> + or both in parallel, as here.
> +
> + elfutils is distributed in the hope that it will be useful, but
> + WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + General Public License for more details.
> +
> + You should have received copies of the GNU General Public License and
> + the GNU Lesser General Public License along with this program. If
> + not, see <http://www.gnu.org/licenses/>. */
> +
> +#ifdef HAVE_CONFIG_H
> +# include <config.h>
> +#endif
> +
> +#include <assert.h>
> +#include <dwarf.h>
> +#include <string.h>
> +
> +#define BACKEND mips_
> +#include "libebl_CPU.h"
> +
> +ssize_t
> +mips_register_info (Ebl *ebl __attribute__ ((unused)),
> + int regno, char *name, size_t namelen,
> + const char **prefix, const char **setname,
> + int *bits, int *type)
> +{
> + if (name == NULL)
> + return 66;
> +
> + if (regno < 0 || regno > 65 || namelen < 4)
> + return -1;
> +
> + *prefix = "$";
> +
> + if (regno < 32)
> + {
> + *setname = "integer";
> + *type = DW_ATE_signed;
> + *bits = 32;
> + if (regno < 32 + 10)
> + {
> + name[0] = regno + '0';
> + namelen = 1;
> + }
> + else
> + {
> + name[0] = (regno / 10) + '0';
> + name[1] = (regno % 10) + '0';
> + namelen = 2;
> + }
> + }
> + else if (regno < 64)
> + {
> + *setname = "FPU";
> + *type = DW_ATE_float;
> + *bits = 32;
> + name[0] = 'f';
> + if (regno < 32 + 10)
> + {
> + name[1] = (regno - 32) + '0';
> + namelen = 2;
> + }
> + else
> + {
> + name[1] = (regno - 32) / 10 + '0';
> + name[2] = (regno - 32) % 10 + '0';
> + namelen = 3;
> + }
> + }
> + else if (regno == 64)
> + {
> + *type = DW_ATE_signed;
> + *bits = 32;
> + name[0] = 'h';
> + name[1] = 'i';
> + namelen = 2;
> + }
> + else
> + {
> + *type = DW_ATE_signed;
> + *bits = 32;
> + name[0] = 'l';
> + name[1] = 'o';
> + namelen = 2;
> + }
> +
> + name[namelen++] = '\0';
> + return namelen;
> +}
OK, but indentation seems slightly off. space vs tabs?
> diff --git a/backends/mips_retval.c b/backends/mips_retval.c
> new file mode 100644
> index 00000000..fd9aaefa
> --- /dev/null
> +++ b/backends/mips_retval.c
> @@ -0,0 +1,261 @@
> +/* Function return value location for Linux/mips ABI.
> + Copyright (C) 2005 Red Hat, Inc.
> + Copyright (C) 2023 CIP United Inc.
> + This file is part of elfutils.
> +
> + This file is free software; you can redistribute it and/or modify
> + it under the terms of either
> +
> + * the GNU Lesser General Public License as published by the Free
> + Software Foundation; either version 3 of the License, or (at
> + your option) any later version
> +
> + or
> +
> + * the GNU General Public License as published by the Free
> + Software Foundation; either version 2 of the License, or (at
> + your option) any later version
> +
> + or both in parallel, as here.
> +
> + elfutils is distributed in the hope that it will be useful, but
> + WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + General Public License for more details.
> +
> + You should have received copies of the GNU General Public License and
> + the GNU Lesser General Public License along with this program. If
> + not, see <http://www.gnu.org/licenses/>. */
> +
> +#ifdef HAVE_CONFIG_H
> +# include <config.h>
> +#endif
> +
> +#include <assert.h>
> +#include <dwarf.h>
> +#include <string.h>
> +#include <elf.h>
> +#include <stdio.h>
> +
> +#define BACKEND mips_
> +#include "libebl_CPU.h"
> +#include "libdwP.h"
> +
> +/* All the possible MIPS ARCHs. */
> +enum mips_arch
> + {
> + MIPS_ARCH_UNKNOWN = 0,
> + MIPS_ARCH_32,
> + MIPS_ARCH_64,
> + MIPS_ARCH_LAST
> + };
> +
> +/* Find the mips ARCH of the current file */
> +enum mips_arch find_mips_arch(Elf *elf)
> +{
> + GElf_Ehdr ehdr_mem;
> + GElf_Ehdr *ehdr = gelf_getehdr (elf, &ehdr_mem);
> +
> + if (ehdr == NULL)
> + return MIPS_ARCH_LAST;
> +
> + GElf_Word elf_flags = ehdr->e_flags;
> +
> + /* Check elf_flags to see if it specifies the ARCH being used. */
> + switch ((elf_flags & EF_MIPS_ARCH))
> + {
> + case E_MIPS_ARCH_32:
> + case EF_MIPS_ARCH_32R2:
> + case E_MIPS_ARCH_32R6:
> + return MIPS_ARCH_32;
> + case E_MIPS_ARCH_64:
> + case EF_MIPS_ARCH_64R2:
> + case E_MIPS_ARCH_64R6:
> + return MIPS_ARCH_64;
> + default:
> + return MIPS_ARCH_32;
> + }
> +
> + return MIPS_ARCH_UNKNOWN;
> +}
>
> +unsigned int
> +mips_arch_regsize (enum mips_arch arch)
> +{
> + switch (arch)
> + {
> + case MIPS_ARCH_32:
> + return 4;
> + case MIPS_ARCH_64:
> + return 8;
> + case MIPS_ARCH_UNKNOWN:
> + case MIPS_ARCH_LAST:
> + default:
> + return 0;
> + }
> +}
So this is different from checking ELFCLASS32/64?
> +/* $v0 or pair $v0, $v1 */
> +static const Dwarf_Op loc_intreg_o32[] =
> + {
> + { .atom = DW_OP_reg2 }, { .atom = DW_OP_piece, .number = 4 },
> + { .atom = DW_OP_reg3 }, { .atom = DW_OP_piece, .number = 4 },
> + };
> +
> +static const Dwarf_Op loc_intreg[] =
> + {
> + { .atom = DW_OP_reg2 }, { .atom = DW_OP_piece, .number = 8 },
> + { .atom = DW_OP_reg3 }, { .atom = DW_OP_piece, .number = 8 },
> + };
> +#define nloc_intreg 1
> +#define nloc_intregpair 4
> +
> +/* $f0 (float), or pair $f0, $f1 (double).
> + * f2/f3 are used for COMPLEX (= 2 doubles) returns in Fortran */
> +static const Dwarf_Op loc_fpreg_o32[] =
> + {
> + { .atom = DW_OP_regx, .number = 32 }, { .atom = DW_OP_piece, .number = 4 },
> + { .atom = DW_OP_regx, .number = 33 }, { .atom = DW_OP_piece, .number = 4 },
> + { .atom = DW_OP_regx, .number = 34 }, { .atom = DW_OP_piece, .number = 4 },
> + { .atom = DW_OP_regx, .number = 35 }, { .atom = DW_OP_piece, .number = 4 },
> + };
> +
> +/* $f0, or pair $f0, $f2. */
> +static const Dwarf_Op loc_fpreg[] =
> + {
> + { .atom = DW_OP_regx, .number = 32 }, { .atom = DW_OP_piece, .number = 8 },
> + { .atom = DW_OP_regx, .number = 34 }, { .atom = DW_OP_piece, .number = 8 },
> + };
> +#define nloc_fpreg 1
> +#define nloc_fpregpair 4
> +#define nloc_fpregquad 8
> +
> +/* The return value is a structure and is actually stored in stack space
> + passed in a hidden argument by the caller. But, the compiler
> + helpfully returns the address of that space in $v0. */
> +static const Dwarf_Op loc_aggregate[] =
> + {
> + { .atom = DW_OP_breg2, .number = 0 }
> + };
> +#define nloc_aggregate 1
> +
> +int
> +mips_return_value_location (Dwarf_Die *functypedie, const Dwarf_Op **locp)
> +{
> + /* First find the ARCH used by the elf object */
> + enum mips_arch arch = find_mips_arch(functypedie->cu->dbg->elf);
> + /* Something went seriously wrong while trying to figure out the ARCH */
> + if (arch == MIPS_ARCH_LAST)
> + return -1;
> +
> + /* We couldn't identify the ARCH, but the file seems valid */
> + if (arch == MIPS_ARCH_UNKNOWN)
> + return -3;
> +
> + unsigned int regsize = mips_arch_regsize (arch);
> + if (!regsize)
> + return -2;
> +
> + /* Start with the function's type, and get the DW_AT_type attribute,
> + which is the type of the return value. */
> +
> + Dwarf_Attribute attr_mem;
> + Dwarf_Attribute *attr = dwarf_attr_integrate (functypedie, DW_AT_type, &attr_mem);
> + if (attr == NULL)
> + /* The function has no return value, like a `void' function in C. */
> + return 0;
> +
> + Dwarf_Die die_mem;
> + Dwarf_Die *typedie = dwarf_formref_die (attr, &die_mem);
> + int tag = dwarf_tag (typedie);
> +
> + /* Follow typedefs and qualifiers to get to the actual type. */
> + while (tag == DW_TAG_typedef
> + || tag == DW_TAG_const_type || tag == DW_TAG_volatile_type
> + || tag == DW_TAG_restrict_type)
> + {
> + attr = dwarf_attr_integrate (typedie, DW_AT_type, &attr_mem);
> + typedie = dwarf_formref_die (attr, &die_mem);
> + tag = dwarf_tag (typedie);
> + }
> +
> + switch (tag)
> + {
> + case -1:
> + return -1;
> +
> + case DW_TAG_subrange_type:
> + if (! dwarf_hasattr_integrate (typedie, DW_AT_byte_size))
> + {
> + attr = dwarf_attr_integrate (typedie, DW_AT_type, &attr_mem);
> + typedie = dwarf_formref_die (attr, &die_mem);
> + tag = dwarf_tag (typedie);
> + }
> + /* Fall through. */
> + FALLTHROUGH;
> +
> + case DW_TAG_base_type:
> + case DW_TAG_enumeration_type:
> + CASE_POINTER:
> + {
> + Dwarf_Word size;
> + if (dwarf_formudata (dwarf_attr_integrate (typedie, DW_AT_byte_size,
> + &attr_mem), &size) != 0)
> + {
> + if (dwarf_is_pointer (tag))
> + size = regsize;
> + else
> + return -1;
> + }
> + if (tag == DW_TAG_base_type)
> + {
> + Dwarf_Word encoding;
> + if (dwarf_formudata (dwarf_attr_integrate (typedie, DW_AT_encoding,
> + &attr_mem), &encoding) != 0)
> + return -1;
> +
> +#define ARCH_LOC(loc, regsize) ((regsize) == 4 ? (loc ## _o32) : (loc))
> +
> + if (encoding == DW_ATE_float)
> + {
> + *locp = ARCH_LOC(loc_fpreg, regsize);
> + if (size <= regsize)
> + return nloc_fpreg;
> +
> + if (size <= 2*regsize)
> + return nloc_fpregpair;
> +
> + if (size <= 4*regsize && arch == MIPS_ARCH_32)
> + return nloc_fpregquad;
> +
> + goto aggregate;
> + }
> + }
> + *locp = ARCH_LOC(loc_intreg, regsize);
> + if (size <= regsize)
> + return nloc_intreg;
> + if (size <= 2*regsize)
> + return nloc_intregpair;
> +
> + /* Else fall through. Shouldn't happen though (at least with gcc) */
> + }
> + FALLTHROUGH;
> +
> + case DW_TAG_structure_type:
> + case DW_TAG_class_type:
> + case DW_TAG_union_type:
> + case DW_TAG_array_type:
> + aggregate:
> + /* XXX TODO: Can't handle structure return with other ABI's yet :-/ */
> + if ((arch != MIPS_ARCH_32) && (arch != MIPS_ARCH_64))
> + return -2;
> +
> + *locp = loc_aggregate;
> + return nloc_aggregate;
> + }
> +
> + /* XXX We don't have a good way to return specific errors from ebl calls.
> + This value means we do not understand the type, but it is well-formed
> + DWARF and might be valid. */
> + return -2;
> +}
OK. This looks plausible. Would be good to have a reference to the
actual calling convention.
Thanks,
Mark
next prev parent reply other threads:[~2023-05-11 16:38 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-04-11 8:12 [PATCH 0/5] Add support for MIPS Ying Huang
2023-04-11 8:12 ` [PATCH 1/5] strip: Adapt src/strip -o -f on mips Ying Huang
2023-05-09 15:15 ` Mark Wielaard
2023-05-16 6:38 ` Ying Huang
2023-05-21 21:13 ` Mark Wielaard
2023-05-24 6:21 ` Ying Huang
2023-05-16 6:46 ` Ying Huang
2023-05-16 7:34 ` Ying Huang
2023-05-21 21:14 ` Mark Wielaard
2023-05-26 2:48 ` Ying Huang
2023-05-16 8:05 ` [EXTERNAL] " Luke Diamand
2023-04-11 8:12 ` [PATCH 2/5] readelf: Adapt src/readelf -h/-S/-r/-w/-l/-d/-a " Ying Huang
2023-05-11 14:31 ` Mark Wielaard
2023-05-16 8:01 ` Ying Huang
2023-07-24 8:35 ` Ying Huang
2023-07-25 8:15 ` Ying Huang
2023-07-27 6:08 ` Ying Huang
2023-08-01 21:43 ` Mark Wielaard
2023-08-01 13:14 ` Mark Wielaard
2023-08-01 9:25 ` Mark Wielaard
2023-04-11 8:12 ` [PATCH 3/5] elflint: Fix invalid type of relocation info and other issues " Ying Huang
2023-05-11 15:59 ` Mark Wielaard
2023-05-17 9:14 ` Ying Huang
2023-04-11 8:12 ` [PATCH 4/5] stack: Fix stack unwind failure " Ying Huang
2023-05-11 16:07 ` Mark Wielaard
2023-05-18 6:14 ` Ying Huang
2023-04-11 8:12 ` [PATCH 5/5] backends: Fix run-native-test.sh and run-funcretval++11.sh run fail " Ying Huang
2023-05-11 16:38 ` Mark Wielaard [this message]
2023-05-18 9:06 ` Ying Huang
2023-05-04 2:24 ` [PATCH 0/5] Add support for MIPS 黄莺
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=4958c19be43be2315c26d104b1ada85dbde89142.camel@klomp.org \
--to=mark@klomp.org \
--cc=elfutils-devel@sourceware.org \
--cc=ying.huang@oss.cipunited.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;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).