Monday Patch Queue Review update (2022-06-06)

Monday Patch Queue Review update (2022-06-06)

[Carlos O'Donell]

Most recent meeting status is always here:
https://sourceware.org/glibc/wiki/PatchworkReviewMeetings#Update

Meeting: 2022-06-06 @ 0900h EST5EDT

Video/Audio: https://bluejeans.com/9093064454

IRC: #glibc on OFTC.

Review new patches and attendee requested reviews.

 * State NEW delegate NOBODY 238
 * Start at 54820
 * aarch64: Handle fewer relocations for RTLD_BOOTSTRAP
  * Arm should review for AEABI compliance?
 * riscv: Change the relocations handled for RTLD_BOOTSTRAP
  * Need machine maintainer review.
 * [v4,1/2] nss: add assert to DB_LOOKUP_FCT (BZ #28752)
  * Carlos: Will ask DJ to review.
 * [v1] x86: ZERO_UPPER_VEC_REGISTERS_RETURN_XTEST expect no transactions
  * Noah: Will resubmit after teh VEC classes gets approved.
  * Fails CI because previous series needs application.
 * [v3,1/8] x86: Create header for VEC classes in x86 strings library
  * Noah: HJ should have a look at these.
 * x86_64: Add strstr function with 512-bit EVEX
  * Noah: Looking into this.
 * [v2] nptl_db: disable DT_RELR on libthread_db.so
  * Carlos: Will review.
 * Avoid -Wstringop-overflow= warning in iconv module. \
  * Siddhesh: Will review.
 * resolv: Implement no-aaaa stub resolver option
  * Carlos: Already reviewed. Changes request.
 * New syscalls for mount and namespace
  * Carlos: Adhemerval can you please rebase and send the new syscalls?
 * Add PT_AARCH64_MEMTAG_MTE from Linux 5.18 to elf.h
  * Already reviewed.
 * [v3] elf: Remove ELF_RTYPE_CLASS_EXTERN_PROTECTED_DATA
  * Szabolcs: Semantics have changed with this patch and it needs more
    serious review. Simple test cases previously worked and now it
    doesn't.
 * GLIBC LoongArch PATCHES (caiyinyu)
  * Needs review.
  * Carlos: What is the status of the ABI? It has been submitted to musl.
  * Adhemerval: I think musl's ABI is done, but there is kernel discussion.
 * nptl: Fix __libc_cleanup_pop_restore asynchronous restore (BZ#29214) (Adhemerval)
  * Carlos: Will review.
  * Broken in 2.35.
 * v3: Fix various NPTL synchronization issues (Adhemerval)
  * Robustness issue, and fixes BZ #19951, which is a real issue.
 * Review ended 54535.
 * Noah: Why can't we use #define to define rseq offset?
  * Noah: Florian isn't here, and I was going to ask him the question.
  * Szabolcs: It's for maintenance reasons. The kernel wanted to grow
    the size of rseq, and we didn't want to have to define a constant
    location for all arches, and we didn't want to define a different
    constant per arch, so we just made it another lookup. You can cache
    the values in  globals if the two TLS accesses are slower than you
    expect.

-- 
Cheers,
Carlos.

Re: Monday Patch Queue Review update (2022-06-06)

[Florian Weimer]

* Carlos O'Donell via Libc-alpha:

>  * Noah: Why can't we use #define to define rseq offset?
>   * Noah: Florian isn't here, and I was going to ask him the question.

This would make more of the thread descriptor/TCB layout part of the
ABI.  I guess we could do that for select ABIs if there is no concern
about future alignment requirements (e.g., if a future TCB requires 128
byte alignment, the rseq offset is a multiple of 32 from the TCB, but
not a multiple of 128, and a future rseq requires 64 byte alignment, we
would be in trouble).

Thanks,
Florian