this is in the malloc initialisation path- but pthread_atfork calls malloc() in
musl (doesn't in glibc), so it causes an infinite spin on the spinlock taken in
this path.

there's no real good way to work around this, but musl 1.2.4 contains a
commit to use internal __libc_malloc for pthread_atfork instead of calling
the externally defined malloc and causing a loop here, so this won't be needed
then.

(also surprising how this doesn't break anything)
--
--- a/base/allocator/partition_allocator/partition_root.cc
+++ b/base/allocator/partition_allocator/partition_root.cc
@@ -248,9 +248,9 @@
   // However, no perfect solution really exists to make threads + fork()
   // cooperate, but deadlocks are real (and fork() is used in DEATH_TEST()s),
   // and other malloc() implementations use the same techniques.
-  int err =
-      pthread_atfork(BeforeForkInParent, AfterForkInParent, AfterForkInChild);
-  PA_CHECK(err == 0);
+  //int err =
+  //    pthread_atfork(BeforeForkInParent, AfterForkInParent, AfterForkInChild);
+  //PA_CHECK(err == 0);
 #endif  // BUILDFLAG(IS_LINUX) || BUILDFLAG(IS_CHROMEOS)
 }