Monday Patch Queue Review update (2021-12-13)

Monday Patch Queue Review update (2021-12-13)

[Carlos O'Donell]

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

Meeting: 2021-12-13 @ 0900h EST (UTC-5)

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

IRC: #glibc on OFTC.

Review new patches and restart review at the top.

 * Carlos asks what are we looking at putting into the release.
 * Adhemerval:
  * Would like to get hugepage support into 2.35.
   * Siddhesh: Reviewed the last patch set.
   * Huge page support aligns with what we are doing with the text segment
    * Makes it easier to support THP with .text.
  * Carlos: What is the status of the LD_AUDIT series?
   * Adhemerval: Florian has reviewed half of the patches.
   * Florian: Ran out of time last week. I should be able to review this week.
 * Florian:
  * _dl_find_object needs review.
   * Adhemerval: Will review.
  * Minor fix to rseq. No ABI implications.
   * Removes not-needed relocations.
  * Add membarrier support (syscall wrapper)
   * Need to do some digging for programming model.
   * May be an additional step required for membarrier setup.
 * HJ:
  * libmvec patches are very large (42 patches)
   * Submit one function. Work with community to address concerns.
     If the community agrees with the approach then the rest of the
     patches can be automated and ready.
   * Intent is to include the functions in 2.35.
  * DT_RUNPATH v3 posted.
   * Carlos to review v3.
 * Sunil:
  * libmvec needs review.
   * CFI escape? Compiler generated. Encodes the DFA expression.
    * Can we remove it? What are they?
 * Siddhesh:
  * No additional reviews required.
 * Szabolcs:
  * No additional patches for review.
 * Carlos:
  * pthread cond var fix and review. Bug fix.
  * Licked the cookie. MUST BE FIXED.
 * Series 5728: Fix 64-bit time_t ancillary socket conversion
  * Posted v3. Needs review from Florian.
 * Series 5725: [v4] elf: Add _dl_find_object function
  * Adhemerval to review.
 * Series 5714: fix p_align on PT_LOAD segment in DSO isn't honored
  * Checked in.
 * Series 2348: [v2] Use LFS and 64 bit time for installed programs (BZ #15333)
  * Needed.
 * Discussed alignment issues with some architectures from the PT_LOAD changes.
  * Still need this if p_align is large.
  * We can solve this in the loader.
  * Once we fix hte linker the normal binary will have 4KiB p_align.
  * Will only get the large alignment if we have a variable that needs it.
  * For huge page we still want the large p_align even if there is no variables.
  * TODO: Must fix unmapping for 2.35.
 * Series 5703: intl/plural.y: define macros to avoid multiple conflicting declarations of yyerror and yylex
  * Ask Arjun to review this change.
 * Series 5688: [v3] Support target specific ALIGN for variable alignment test [BZ #28676]
  * Needs review.
 * Series 5695: [v7] elf: Don't execute shared object directly [BZ #28453]
  * Needs review.
 * What is the state of the OpenRISC or1k port?
  * Joseph has reviewed some of the state.
  * Is this ready for glibc 2.35?
  * State of the port is looking good.
  * Need to check build-many-glibcs and with Jospeh for the state of the port.
  * Doesn't look like there are any large blockers for 2.35.
 * New ld sort algorithm is failing on OpenRISC.
  * Needs review.
 * New ld sort default should be flipped to new algorithm.
  * Need to flip the new sort.

-- 
Cheers,
Carlos.

Re: Monday Patch Queue Review update (2021-12-13)

[Florian Weimer]

* Carlos O'Donell via Libc-alpha:

>  * Discussed alignment issues with some architectures from the PT_LOAD changes.
>   * Still need this if p_align is large.
>   * We can solve this in the loader.
>   * Once we fix hte linker the normal binary will have 4KiB p_align.
>   * Will only get the large alignment if we have a variable that needs it.
>   * For huge page we still want the large p_align even if there is no variables.
>   * TODO: Must fix unmapping for 2.35.

I filed a binutils bug:

  p_align in ELF program headers should not exceed section alignment
  <https://sourceware.org/bugzilla/show_bug.cgi?id=28689>

But it's not that easy because glibc is unnecessarily strict about
p_align:

  PT_LOAD p_align check is too strict
  <https://sourceware.org/bugzilla/show_bug.cgi?id=28688>

Thanks,
Florian