Monday Patch Queue Review update (2022-05-16)

Monday Patch Queue Review update (2022-05-16)

[Adhemerval Zanella]

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

Meeting: 2022-05-23 @ 0900h EST5EDT

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

IRC: #glibc on OFTC.

Review new patches and attendee requested reviews.

 * Status NEW delgate NOBODY 224 patches.
 * Start at patch 54271.
 * dlfcn: Move RTLD_DEFAULT/RTLD_NEXT outside __USE_GNU
  - Looks ok, Florian will review.
 * linux: use statx for fstat if neither newfstatat nor fstatat64 is present
  - Looks ok, Adhemerval will review.
 * x86-64: Ignore r_addend for R_X86_64_GLOB_DAT/R_X86_64_JUMP_SLOT
  - Fangrui state is is already done on other environments, Florian will review.
 * dlsym: Make RTLD_NEXT prefer default version definition [#BZ #14932]
  - Current status is too broken to keep a versioned symbol, change looks reasonable.
 * [v11,1/6] elf: Refactor dl_new_hash so it can be tested / benchmarked
  - Accepted by Siddhesh.
 * [1/5] locale: Turn ADDC and ADDS into functions in linereader.c
  - Not urgent, skipped for now.
 * Avoid RMW of flags2 outside lock (BZ #27842)
  - Still need some discussion that lock can be skipped if flockfile is called.
 * [v2] x86_64: Implement evex512 version of strlen, strnlen, wcslen and wcsnlen
  - Noah is reviewing it.
 * [v3] Fix deadlock when pthread_atfork handler calls pthread_atfork or dlclose
  - Accepted by Adhemerval.
 * Add arc4random support
  - Still need reviewers.
 * [v3] Remove special flags of libc.5.so and libc.4.so
  - Patch looks ok, some changers required.
 * socket: Check lengths before advancing pointer in CMSG_NXTHDR
  - Siddhesh will review it.
 * [v4,3/3] linux: Add fallback for clone failure on posix_spawn (BZ #29115)
  - Skipped, most likely will result in WONTFIX and programs will need to use
    any new kernel time namespace fix kernel will provide.
 * GLIBC LoongArch PATCHES
  - Patchset looks ok, still requires some upstream fixes.  Skipped due it will
    require at least a v5.
 * [v3] elf: Rewrite long RESOLVE_MAP macro to a debug friendly function
  - Siddhesh is ok with the __always_inline__ change.
 * [v3] math: Add math-use-builtins-fabs (BZ#29027)
  - The patch is ok, it does not change any code.
 * stdio-common: Add the fgetln function
  - Still need review.
 * elf: Simplify version test when searching a versioned symbol
  - Florian will review.
 * Simplify ELF_RTYPE_CLASS_EXTERN_PROTECTED_DATA and revert aarch64/arm's extern protected data handle
  - Florian will review.
 * Optimize struct pthread size
  - Skipped, Arjun will review.
 * libio: Remove _IO_adjust_column, _IO_adjust_wcolumn, column tracking
  - Adhemerval will review.
 * powerpc64: small-model @got@tprel accesses
 * [v3] powerpc64: Enable static-pie
   - skipped, waiting review from powerpc maintainers.

 * Stopped at 52777.

 * Siddhesh
  - Nothing to review.
 * Szabolcs
  - Nothing
 * Sunil
  - Need to check https://sourceware.org/bugzilla/show_bug.cgi?id=29127 
 * Noah
  - Can we change pthread cond to use default mutex and increment the counter
    outside? Is this a GNU extension or is related to some different mutex
    type?
 * Patrick
  - Nothing
 * Florian
  - Nothing

Monday Patch Queue Review update (2022-05-16)

[Carlos O'Donell]

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

Meeting: 2022-05-16 @ 0900h EST5EDT

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

IRC: #glibc on OFTC.

Review new patches and attendee requested reviews.

 * Status NEW delgate NOBODY 252 patches.
 * Start at patch 53982
 * [v2] Remove special flags of libc.5.so and libc.4.so
  * Adhemerval will review.
  * Florian: Cleanup is valuable.
 * 53980 Enable DT_RELR automatically
  * Carlos: Make first commit message line more descriptive
 * 53963 Check for ISO C compilers should also allow C++
  * Fangrui and Florian reviewed.
 * 53886 fortify: Ensure that __glibc_fortify condition is a constant [BZ #29141]
  * Florian noted Siddhesh is the subject matter expert.
 * [v8,1/6] elf: Refactor dl_new_hash so it can be tested / benchmarked
  * Adhemerval: Is this something we can get the compiler to do?
  * Carlos: Yes, it would be better if the compiler did it, but today it doesn't.
  * Ahemerval: Not a blocking objection, but don't like the asm.
  * Siddhesh: Elevate the objection about the asm. Will comment on the list.
 * 53756 Linux: Fix posix_spawn when user with time namespaces
  * Florian: Ask kernel for new way of entering time namespace
  * The kernel can't change the current behaviour.
  * Many C and C++ libraries have interfaces implemented from scratch.
  * All the interfaces would have to be changed to support this.
  * Florian: Reached out to Christian Brauner about this on the musl thread.
  * Do we need to speak with Andrei Vagin (CRIU)?
 * Stopped at 53612.
 * Adhemerval: Looking for arc4random review.
  * arc4random at v5 review.
  * Carlos will review kernel syscall patches.
  * Florian has not reviewed this as author of the previous attempt.
  * Carlos: We are talking about commiting to the ABI? Any objections to the ABI?
  * Florian: We should add the ABI.
 * Arjun: Have two patches to reivew. New patch to post v3 for pthread_atfork.
 * Florian:
  * brk change: 
   * Not sure where I'm heading with the brk change.
   * Adhemerval: OK with the brk changes.
  * vfprintf rework still needs review of the buffer management.
  * UTF-8 locale source work:
   * Siddhesh can change one.
   * Carlos can change Spanish locales (22 of them)
 * HJ: Nothing to review.
  * Long term goal to get rid of copy relocation in x86.
  * There are a couple of platform that support copy relocations.
  * Many don't support copy relocations.
  * From a performance and security POV we want to get rid of them.
  * Especially with protected symbols they are supposed to improve performance.
  * Semantics of copy relocations break this.
  * gcc 12 has option to avoid copy relocations on x86.
  * Explore removing copy relocations?
  * Florian: From Fedora and downstream perspective copy relocations don't matter.
   * We use LTO and the builds localize the symbols.
   * The change to copy reocations is a NOP.
   * Likewise for SUSE.
   * HJ: Note in Windows with __declspec(dllimport/dllexport) we can improve code gen.
    * HJ: Anything we can do here that is similar?
  * Florian: COPY relocations may still be relevant on some architectures.
   * For example cost to access global data through the GOT
 * Szabolcs: Outstanding patch was reviewed. Need to change and push.

-- 
Cheers,
Carlos.