https://bugs.exim.org/show_bug.cgi?id=2334
--- Comment #8 from Philip Hazel <ph10@???> ---
Meanwhile, I have found this text in
http://read.mlook.mobi/read/2699/page/6
"To run faster, bytecode interpreters using JIT determine what machine code the
bytecode is trying to run, write the machine code to a buffer, mark it
executable, and then execute it with the actual processor. With typical iOS
code signing, this is impossible. To allow JIT, but still keep much of the
security of the original code signing scheme, Apple chose a compromise. It
would allow only certain processes (for example, MobileSafari) to make a memory
region that was writable and executable to perform their JIT work. Furthermore,
the process could make exactly one such region. Any attempts to make additional
writable and executable regions would fail."
This is *exactly* the phenomenon we saw: ONE call with MAP_JIT works. The issue
is that, when PCRE2 is compiled with more than one code unit size, the
pcre2_jit_test program does things more than once, which hits the problem.
The problem isn't entirely confined to PCRE2. The sljit compiler is an
independent entity. Zoltan and I were considering adding some kind of
user-specified option. Further discussion needed....
As for the heap limit issue: I will investigate and do experiments. Thanks for
the testing.
--
You are receiving this mail because:
You are on the CC list for the bug.
