[PATCH] Improve performance of IO locks

[Mark Wielaard]

Hi,

On Mon, Aug 01, 2022 at 11:06:07AM +0000, Wilco Dijkstra via Libc-alpha wrote:
> Improve performance of recursive IO locks by adding a fast path for
> the single-threaded case. To reduce the number of memory accesses for
> locking/unlocking, only increment the recursion counter if the lock
> is already taken. 
> 
> On Neoverse V1, a microbenchmark with many small freads improved by
> 2.9 times. Multithreaded performance improved by 2%.

Strangely this seems to have broken the glibc-debian-ppc64 buildbot:
https://builder.sourceware.org/buildbot/#/builders/glibc-debian-ppc64

I don't see how this commit can cause a failure just on debian-ppc64
(all other distro/arches are fine).

And the corresponding bunsen test results don't really show why.
https://builder.sourceware.org/testrun/498f51f327afdd7030516455b709a31a0038b432
Most of the .out files are empty, but some indicate an segmentation fault.

Comparing to the build before only shows test result diffs, no
configuration differences.
https://builder.sourceware.org/testrun-diffs?commitish=498f51f327afdd7030516455b709a31a0038b432&commitish=58fd9d63b078b6bbfdba45135c4021038f33534e

I don't have access to the buildbot, so cannot easily investigate more.

Tom, could you have a look and see if you can find out more? Does just
reverting commit c51c483d2b8ae66fe31a12509aedae02a6982ced make things
OK again? Or did something else on the buildbot worker change?

Note that the buildbot only does a make subdirs=elf check, not a full
make check.

Thanks,

Mark