which is essential as an attempt to fix the "failed to reattach"
error on Windows. If file_cache is enabled, Opcache will
automaticaly switch to file_cache_only mode in the case a process
failed to map the shared segment at the required address. The
important small part of the SHM will still be mapped, which
allows information exchange between normal processes using SHM
and those using the fallback mechanism.
This is based on Dmitry's, Matt's and mine ideas. So many thanks for
support!
PHP should be configured and built with --enable-huge-code-pages, OS should be configured to provide huge pages.
It's possible to enable/disable this future in php.ini through opcache.enable_huge_code_pages=0/1.
The feature was tested on Linux and provided 2% improvement on real-life apps, because of 2-3 times reduction in number of iTLB misses.
Squashed commit of the following:
commit d96eab8d79
Author: Francois Laupretre <francois@tekwire.net>
Date: Fri Jun 26 01:23:31 2015 +0200
Use the new 'ZSTR' macros in the rest of the code.
Does not change anything to the generated code (thanks to compat macros) but cleaner.
commit b352643910
Author: Francois Laupretre <francois@tekwire.net>
Date: Thu Jun 25 13:45:06 2015 +0200
Improve zend_string API
Add missing methods
Removed HashTable->arHash (reduced memory consumption). Now hash slots may be accessed using HT_HASH() macro.
Hash slotas are allocated together with Buckets (before them) and lay in reverse order from HashTable->arData base address (see comments in Zend/zend_types.h)
Indexes in hash table and conflict resolution chains (Z_NEXT) may be stored as indeces or offsets in bytes, depending on system (32 or 64-bit).
HashTable data filelds are reordered to keep the most useful for zend_hash_find() data in the same CPU cache line.
At least in apache, killing a locker with SIGKILL from inside opcache prevents
the child process from emitting a corresponding log event to the access or
error logs, therefore requests are termed with no record. This makes is
incredibly difficult to debug problems espically on systems with high counts
of cache restarts. Since INFO level is far too noisy for production use,
switch the killing to a higher level so we can see it easier.
Now each HashTable is also zend_array, so it's refcounted and may be a subject for Copy on Write
zend_array_dup() was changed to allocate and return HashTable, instead of taking preallocated HashTable as argument.
