From: Ben Woodard <woodard@redhat.com>
To: Szabolcs Nagy <szabolcs.nagy@arm.com>
Cc: libc-alpha@sourceware.org
Subject: Re: [PATCH 11/15] x86_64: Avoid lazy relocation of tlsdesc [BZ #27137]
Date: Fri, 9 Apr 2021 07:55:13 -0700 [thread overview]
Message-ID: <C4949EA7-DCE1-42F8-82AF-A19DF99B9C71@redhat.com> (raw)
In-Reply-To: <20210409133809.GR23289@arm.com>
> On Apr 9, 2021, at 6:38 AM, Szabolcs Nagy <szabolcs.nagy@arm.com> wrote:
>
> The 04/08/2021 17:14, Ben Woodard wrote:
>> Don’t you also need to modify elf_machine_runtime_setup It also has a reference to _dl_tlsdesc_resolve_rela that becomes undefined when you try to compile with your patchset including patch 13 where you remove the code.
>>
>> To make a test build I just commented it out but I think that this patch should remove that if statement as well.
>
> thanks,
> indeed this was wrong, i tested the wrong branch on x86_64.
>
> i will fix this and post a v2 set with the other feedback.
On the positive side, I’ve been tracking down a problem where a library compiled with the gnu2 variant of TLS in a way that I haven’t been able to reproduce yet is crashing the dynamic loader when used with a performance tool that uses LD_AUDIT.
This patch alone (with my tiny modification below) addresses the problem. I say “addresses” because it doesn’t exactly fix the problem; it makes it so that the code with the bug in it isn’t run. Patch 13 in your patch set removes the code with the bug in it.
I see that patches 1 and 2 of your patch set have already been committed. I would encourage you to consider committing V2 of patch 11 and 13 (or maybe 11-14) even before the rest of the patch set since it addresses a bug that we are seeing in the wild.
-ben
>
>>
>> diff --git a/sysdeps/x86_64/dl-machine.h b/sysdeps/x86_64/dl-machine.h
>> index 9a876a371e..2b1b36a739 100644
>> --- a/sysdeps/x86_64/dl-machine.h
>> +++ b/sysdeps/x86_64/dl-machine.h
>> @@ -127,9 +127,11 @@ elf_machine_runtime_setup (struct link_map *l, int lazy, int profile)
>> }
>> }
>>
>> - if (l->l_info[ADDRIDX (DT_TLSDESC_GOT)] && lazy)
>> - *(ElfW(Addr)*)(D_PTR (l, l_info[ADDRIDX (DT_TLSDESC_GOT)]) + l->l_addr)
>> - = (ElfW(Addr)) &_dl_tlsdesc_resolve_rela;
>> + /* Lazy binding of TLSDESC relocations is no longer done so this logic
>> + won't apply */
>> + /* if (l->l_info[ADDRIDX (DT_TLSDESC_GOT)] && lazy) */
>> + /* *(ElfW(Addr)*)(D_PTR (l, l_info[ADDRIDX (DT_TLSDESC_GOT)]) + l->l_addr) */
>> + /* = (ElfW(Addr)) &_dl_tlsdesc_resolve_rela; */
>>
>> return lazy;
>> }
>>
>>
>>> On Feb 15, 2021, at 4:02 AM, Szabolcs Nagy via Libc-alpha <libc-alpha@sourceware.org> wrote:
>>>
>>> Lazy tlsdesc relocation is racy because the static tls optimization and
>>> tlsdesc management operations are done without holding the dlopen lock.
>>>
>>> This similar to the commit b7cf203b5c17dd6d9878537d41e0c7cc3d270a67
>>> for aarch64, but it fixes a different race: bug 27137.
>>> ---
>>> sysdeps/x86_64/dl-machine.h | 19 ++++++++++++++-----
>>> 1 file changed, 14 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/sysdeps/x86_64/dl-machine.h b/sysdeps/x86_64/dl-machine.h
>>> index 103eee6c3f..9a876a371e 100644
>>> --- a/sysdeps/x86_64/dl-machine.h
>>> +++ b/sysdeps/x86_64/dl-machine.h
>>> @@ -570,12 +570,21 @@ elf_machine_lazy_rel (struct link_map *map,
>>> }
>>> else if (__glibc_likely (r_type == R_X86_64_TLSDESC))
>>> {
>>> - struct tlsdesc volatile * __attribute__((__unused__)) td =
>>> - (struct tlsdesc volatile *)reloc_addr;
>>> + const Elf_Symndx symndx = ELFW (R_SYM) (reloc->r_info);
>>> + const ElfW (Sym) *symtab = (const void *)D_PTR (map, l_info[DT_SYMTAB]);
>>> + const ElfW (Sym) *sym = &symtab[symndx];
>>> + const struct r_found_version *version = NULL;
>>>
>>> - td->arg = (void*)reloc;
>>> - td->entry = (void*)(D_PTR (map, l_info[ADDRIDX (DT_TLSDESC_PLT)])
>>> - + map->l_addr);
>>> + if (map->l_info[VERSYMIDX (DT_VERSYM)] != NULL)
>>> + {
>>> + const ElfW (Half) *vernum =
>>> + (const void *)D_PTR (map, l_info[VERSYMIDX (DT_VERSYM)]);
>>> + version = &map->l_versions[vernum[symndx] & 0x7fff];
>>> + }
>>> +
>>> + /* Always initialize TLS descriptors completely at load time, in
>>> + case static TLS is allocated for it that requires locking. */
>>> + elf_machine_rela (map, reloc, sym, version, reloc_addr, skip_ifunc);
>>> }
>>> else if (__glibc_unlikely (r_type == R_X86_64_IRELATIVE))
>>> {
>>> --
>>> 2.17.1
>>>
>>
>
> --
next prev parent reply other threads:[~2021-04-09 14:55 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-02-15 11:56 [PATCH 00/15] Dynamic TLS related data race fixes Szabolcs Nagy
2021-02-15 11:56 ` [PATCH 01/15] aarch64: free tlsdesc data on dlclose [BZ #27403] Szabolcs Nagy
2021-04-01 12:57 ` Adhemerval Zanella
2021-04-06 13:43 ` Szabolcs Nagy
2021-04-06 16:52 ` Adhemerval Zanella
2021-02-15 11:56 ` [PATCH 02/15] elf: Fix data race in _dl_name_match_p [BZ #21349] Szabolcs Nagy
2021-04-01 14:01 ` Adhemerval Zanella
2021-04-06 16:41 ` Szabolcs Nagy
2021-02-15 11:57 ` [PATCH 03/15] Add test case for [BZ #19329] Szabolcs Nagy
2021-04-02 19:10 ` Adhemerval Zanella
2021-02-15 11:59 ` [PATCH 04/15] Add a DTV setup test [BZ #27136] Szabolcs Nagy
2021-04-02 19:35 ` Adhemerval Zanella
2021-02-15 11:59 ` [PATCH 05/15] elf: Fix a DTV setup issue " Szabolcs Nagy
2021-04-02 19:46 ` Adhemerval Zanella
2021-02-15 11:59 ` [PATCH 06/15] elf: Fix comments and logic in _dl_add_to_slotinfo Szabolcs Nagy
2021-04-02 20:50 ` Adhemerval Zanella
2021-04-06 15:48 ` Szabolcs Nagy
2021-04-06 17:47 ` Adhemerval Zanella
2021-04-07 7:57 ` Szabolcs Nagy
2021-04-07 14:20 ` Adhemerval Zanella
2021-02-15 12:00 ` [PATCH 07/15] elf: Refactor _dl_update_slotinfo to avoid use after free Szabolcs Nagy
2021-04-06 19:40 ` Adhemerval Zanella
2021-04-07 8:01 ` Szabolcs Nagy
2021-04-07 14:28 ` Adhemerval Zanella
2021-04-07 14:36 ` Adhemerval Zanella
2021-04-07 17:05 ` Adhemerval Zanella
2021-02-15 12:01 ` [PATCH 08/15] elf: Fix data races in pthread_create and TLS access [BZ #19329] Szabolcs Nagy
2021-02-15 12:01 ` [PATCH 09/15] elf: Use relaxed atomics for racy accesses " Szabolcs Nagy
2021-02-15 12:01 ` [PATCH 10/15] elf: Fix DTV gap reuse logic [BZ #27135] Szabolcs Nagy
2021-02-15 12:02 ` [PATCH 11/15] x86_64: Avoid lazy relocation of tlsdesc [BZ #27137] Szabolcs Nagy
2021-04-09 0:14 ` Ben Woodard
2021-04-09 13:38 ` Szabolcs Nagy
2021-04-09 14:55 ` Ben Woodard [this message]
2021-02-15 12:02 ` [PATCH 12/15] i386: " Szabolcs Nagy
2021-02-15 12:02 ` [PATCH 13/15] x86_64: Remove lazy tlsdesc relocation related code Szabolcs Nagy
2021-02-15 12:03 ` [PATCH 14/15] i386: " Szabolcs Nagy
2021-02-15 12:03 ` [PATCH 15/15] elf: " Szabolcs Nagy
2021-02-15 12:08 ` [PATCH 03/15] Add test case for [BZ #19329] Szabolcs Nagy
2021-02-15 12:08 ` [PATCH 06/15] elf: Fix comments and logic in _dl_add_to_slotinfo Szabolcs Nagy
[not found] ` <CGME20210215115731epcas5p45614957debad2f679230d0bd1efbd57f@epcms5p7>
2021-02-15 12:11 ` [PATCH 02/15] elf: Fix data race in _dl_name_match_p [BZ #21349] Maninder Singh
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=C4949EA7-DCE1-42F8-82AF-A19DF99B9C71@redhat.com \
--to=woodard@redhat.com \
--cc=libc-alpha@sourceware.org \
--cc=szabolcs.nagy@arm.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).