Fix failing VirtualProtect() calls

[cmb69]

If JIT is disabled due to incompatible zend_execute_ex, only
JIT_G(enabled) is set to zero, but JIT_G(on) keeps its value.
However, calling VirtualProtect() with NULL fails with
ERROR_INVALID_PARAMETER. Thus, we check for JIT_G(enabled) as
well, before (un)protecting the SHM.

---

For reference: the diff of the failing test case.

However, with this patch, the test is still marked as borked:

=====================================================================
BORKED TEST SUMMARY
---------------------------------------------------------------------
Warning: JIT is incompatible with third party extensions that override zend_execute_ex(). JIT disabled. in Unknown on line 0 [%s\ext\opcache\tests\jit\bug426.phpt]
=====================================================================

[cmb69]

Maybe the test is borked only in debug mode; apparently it succeeds now in CI.

[nikic]

I think instead we should change

php-src/ext/opcache/ZendAccelerator.c

Line 3044 in 0040e59

if (JIT_G(buffer_size) == 0

to check for jit_size instead of JIT_G(buffer_size), and let it set enabled and on to 0 there.

[cmb69]

The problem is that this code is not exectuted, because JIT_G(enabled) is already false in that case. When looking at d5a82e2, I wonder whether JIT_G(on) should also be set to 0 when JIT_G(enabled) is set to 0. Frankly, I don't understand the difference between these globals.

[nikic]

The problem is that this code is not exectuted, because JIT_G(enabled) is already false in that case. When looking at d5a82e2, I wonder whether JIT_G(on) should also be set to 0 when JIT_G(enabled) is set to 0. Frankly, I don't understand the difference between these globals.

Ah, good point! Yes, I think that code should set both enabled and on to 0.

The difference between the two options is (I believe) that one determines whether the JIT is enabled at all, while the other determines whether it's currently active.

[cmb69]

Thanks!