GH-9370: Fix opcache jit protection bits.

[devnexen]

Allow mprotect to occur, pthread_jit_write_np are just to allow mprotect
flags to be used.

[cmb69]

Thank you! Looks good to me.

[dstogov]

I cannot test this, but according to https://keith.github.io/xcode-man-pages/pthread_jit_write_protect_np.3.html pthread_jit_write_protect_np() should work without mprotect().

May be mprotect() is necessary, because we used it before in zend_accel_shared_protect(). Also note that we seem don't use MAP_JIT flag and this may make difference.

[devnexen]

@jdp1024 would the new changes work for you ?

[jdp1024]

@jdp1024 would the new changes work for you ?

It doesn't work in M1. Call to mprotect seems necessary.

[devnexen]

Ok thx, I m going to put those back.

Could you try again once you get a chance pls ?

[jdp1024]

Ok thx, I m going to put those back.

Could you try again once you get a chance pls ?

mmap() with MAP_JIT along with the removal of two returns worked.

[devnexen]

Ok waiting for @dstogov approval.

[dstogov]

mmap() with MAP_JIT along with the removal of two returns worked.

Does PHP work without the patch (with returns)?

[jdp1024]

mmap() with MAP_JIT along with the removal of two returns worked.

Does PHP work without the patch (with returns)?

It works after the removal of the two returns (without MAP_JIT).

It does not work with MAP_JIT while keeping the two returns.

So it seems that the mprotect calls are necessary in M1 machines.

[dstogov]

OK. Merge this.