From: mengqinggang <mengqinggang@loongson.cn>
To: Xi Ruoyao <xry111@xry111.site>, binutils@sourceware.org
Cc: Nick Clifton <nickc@redhat.com>
Subject: Re: [PATCH] LoongArch: Disallow TLS transition when a section contains TLS_IE64 or TLS_DESC64 reloc
Date: Fri, 26 Jan 2024 09:43:32 +0800 [thread overview]
Message-ID: <8ffdf1ab-7727-10d9-7299-698f364479b9@loongson.cn> (raw)
In-Reply-To: <20240125134238.174841-1-xry111@xry111.site>
Hi,
For TLS transition, we want to change to that the instructions that can be
converted must have a R_LARCH_RELAX relocation.
After 32ee2b4b71c78d3dc0c0b1f87f25fe5df8786b71 commit, the gas not
emit R_LARCH_RELAX for two register macros.
在 2024/1/25 下午9:36, Xi Ruoyao 写道:
> We cannot do TLS transition for them without a psABI revision marking
> some "key" instructions with reloc. Disallow the transition when we see
> such a reloc in a section to prevent an invalid "partial" transition.
>
> Signed-off-by: Xi Ruoyao <xry111@xry111.site>
> ---
>
> To Nick:
>
> Sorry for another late update, but we need this for 2.42 or anything
> compiled with -mcmodel=extreme and using errno will just blow up :(.
>
> bfd/elfnn-loongarch.c | 53 +++++++++++++++++--
> .../ld-loongarch-elf/ld-loongarch-elf.exp | 14 +++++
> .../ld-loongarch-elf/tls-ie64-notrans.d | 3 ++
> .../ld-loongarch-elf/tls-ie64-notrans.s | 7 +++
> 4 files changed, 73 insertions(+), 4 deletions(-)
> create mode 100644 ld/testsuite/ld-loongarch-elf/tls-ie64-notrans.d
> create mode 100644 ld/testsuite/ld-loongarch-elf/tls-ie64-notrans.s
>
> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
> index b62bb424644..b9709401f56 100644
> --- a/bfd/elfnn-loongarch.c
> +++ b/bfd/elfnn-loongarch.c
> @@ -717,6 +717,40 @@ loongarch_tls_transition (struct bfd_link_info *info, unsigned int r_type,
> return loongarch_tls_transition_without_check (info, r_type, h);
> }
>
> +/* For TLS IE in extreme code model:
> + pcalau12i t0, %ie_pc_hi20(x)
> + li.d t1, %ie_pc_lo12(x)
> + lu32i.d t1, %ie64_pc_lo20(x)
> + lu52i.d t1, t1, %ie64_pc_hi12(x)
> + ldx.d t0, t0, t1
> + We've no idea how to remove the ldx.d instruction or turn it into a
> + nop because there is no reloc on it. But we have to stop turning this
> + into LE or we'll end up
> + lu12i.w t0, %le_hi20(x)
> + ori t1, t1, %le_lo12(x)
> + lu32i.d t1, %ie64_pc_lo20(x)
> + lu52i.d t1, t1, %ie64_pc_hi12(x)
> + ldx.d t0, t0, t1
> + which is completely wrong. For TLS DESC it's similar: we cannot remove
> + the add.d instruction but we have to stop the transition. */
> +static bool
> +loongarch_elf_allow_tls_transition_p (const Elf_Internal_Rela *r_begin,
> + const Elf_Internal_Rela *r_end)
> +{
> + for (; r_begin != r_end; r_begin++)
> + switch (ELFNN_R_TYPE (r_begin->r_info))
> + {
> + case R_LARCH_TLS_IE64_PC_HI12:
> + case R_LARCH_TLS_IE64_PC_LO20:
> + case R_LARCH_TLS_DESC64_PC_HI12:
> + case R_LARCH_TLS_DESC64_PC_LO20:
> + return false;
> + }
> +
> + return true;
> +}
> +
> +
> /* Look through the relocs for a section during the first phase, and
> allocate space in the global offset table or procedure linkage
> table. */
> @@ -730,6 +764,7 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> struct elf_link_hash_entry **sym_hashes;
> const Elf_Internal_Rela *rel;
> asection *sreloc = NULL;
> + bool allow_tls_transition;
>
> if (bfd_link_relocatable (info))
> return true;
> @@ -741,6 +776,9 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> if (htab->elf.dynobj == NULL)
> htab->elf.dynobj = abfd;
>
> + allow_tls_transition = loongarch_elf_allow_tls_transition_p (
> + relocs, relocs + sec->reloc_count);
> +
> for (rel = relocs; rel < relocs + sec->reloc_count; rel++)
> {
> unsigned int r_type;
> @@ -818,7 +856,8 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> int need_dynreloc = 0;
> int only_need_pcrel = 0;
>
> - r_type = loongarch_tls_transition (info, r_type, h, abfd, r_symndx);
> + if (allow_tls_transition)
> + r_type = loongarch_tls_transition (info, r_type, h, abfd, r_symndx);
> switch (r_type)
> {
> case R_LARCH_GOT_PC_HI20:
> @@ -2632,7 +2671,6 @@ loongarch_elf_relocate_section (bfd *output_bfd, struct bfd_link_info *info,
> asection **local_sections)
> {
> Elf_Internal_Rela *rel;
> - Elf_Internal_Rela *relend;
> bool fatal = false;
> asection *sreloc = elf_section_data (input_section)->sreloc;
> struct loongarch_elf_link_hash_table *htab = loongarch_elf_hash_table (info);
> @@ -2643,8 +2681,10 @@ loongarch_elf_relocate_section (bfd *output_bfd, struct bfd_link_info *info,
> bool is_dyn = elf_hash_table (info)->dynamic_sections_created;
> asection *plt = htab->elf.splt ? htab->elf.splt : htab->elf.iplt;
> asection *got = htab->elf.sgot;
> + Elf_Internal_Rela *relend = relocs + input_section->reloc_count;
> + bool allow_tls_transition =
> + loongarch_elf_allow_tls_transition_p (relocs, relend);
>
> - relend = relocs + input_section->reloc_count;
> for (rel = relocs; rel < relend; rel++)
> {
> unsigned int r_type = ELFNN_R_TYPE (rel->r_info);
> @@ -2789,7 +2829,12 @@ loongarch_elf_relocate_section (bfd *output_bfd, struct bfd_link_info *info,
>
> BFD_ASSERT (!resolved_local || defined_local);
>
> - relaxed_r_type = loongarch_tls_transition (info, r_type, h, input_bfd, r_symndx);
> + relaxed_r_type =
> + (allow_tls_transition ? loongarch_tls_transition (info, r_type,
> + h, input_bfd,
> + r_symndx)
> + : r_type);
> +
> if (relaxed_r_type != r_type)
> {
> howto = loongarch_elf_rtype_to_howto (input_bfd, relaxed_r_type);
> diff --git a/ld/testsuite/ld-loongarch-elf/ld-loongarch-elf.exp b/ld/testsuite/ld-loongarch-elf/ld-loongarch-elf.exp
> index 2ff06d62236..9e35940f5d6 100644
> --- a/ld/testsuite/ld-loongarch-elf/ld-loongarch-elf.exp
> +++ b/ld/testsuite/ld-loongarch-elf/ld-loongarch-elf.exp
> @@ -88,6 +88,20 @@ if [istarget "loongarch64-*-*"] {
> "medium-call" \
> ] \
> ]
> +
> + run_ld_link_tests \
> + [list \
> + [list \
> + "disable TLS IE transition with TLS_IE64_PC reloc" \
> + "-e 0x0" "" \
> + "" \
> + {tls-ie64-notrans.s} \
> + [list \
> + [list objdump -D tls-ie64-notrans.d] \
> + ] \
> + "tls-ie64-notrans" \
> + ] \
> + ]
> }
>
> if [istarget "loongarch64-*-*"] {
> diff --git a/ld/testsuite/ld-loongarch-elf/tls-ie64-notrans.d b/ld/testsuite/ld-loongarch-elf/tls-ie64-notrans.d
> new file mode 100644
> index 00000000000..47d507a0538
> --- /dev/null
> +++ b/ld/testsuite/ld-loongarch-elf/tls-ie64-notrans.d
> @@ -0,0 +1,3 @@
> +#...
> +.*pcalau12i.*
> +#pass
> diff --git a/ld/testsuite/ld-loongarch-elf/tls-ie64-notrans.s b/ld/testsuite/ld-loongarch-elf/tls-ie64-notrans.s
> new file mode 100644
> index 00000000000..cd8c65ff9d3
> --- /dev/null
> +++ b/ld/testsuite/ld-loongarch-elf/tls-ie64-notrans.s
> @@ -0,0 +1,7 @@
> +# it had segfaulted the linker due to invalid IE->LE transition
> +.globl _start
> +_start:
> + la.tls.ie $a0, $a1, foo
> +
> +.section .tdata
> +foo: .word 114514
next prev parent reply other threads:[~2024-01-26 1:43 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-01-25 13:36 Xi Ruoyao
2024-01-25 13:49 ` Nick Clifton
2024-01-25 13:52 ` Xi Ruoyao
2024-01-26 1:43 ` mengqinggang [this message]
2024-01-26 8:12 ` Xi Ruoyao
2024-01-26 8:19 ` Xi Ruoyao
2024-01-26 8:40 ` mengqinggang
2024-01-26 9:45 ` Xi Ruoyao
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=8ffdf1ab-7727-10d9-7299-698f364479b9@loongson.cn \
--to=mengqinggang@loongson.cn \
--cc=binutils@sourceware.org \
--cc=nickc@redhat.com \
--cc=xry111@xry111.site \
/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).