* JIT/AArch64: [macos][ZTS] Support fast path for tlv_get_addr
Access to TLV(thread local variable) in macOS is in "dynamic" form and
function tlv_get_addr() is invoked to resolve the address. See the
example in [1].
Note there is one struct TLVDescriptor [2] for each TLV. The first
member holds the address of function tlv_get_addr(), and the other two
members, "key" and "offset", would be used inside tlv_get_addr().
The disassembly code for function tlv_get_addr() is shown in [3]. With
the value from system register, i.e. tpidrro_el0, together with "key"
and "offset", the TLV address can be obtained.
Note that the value from tpidrro_el0 varies for different threads, and
unique address for TLV is resolved.
It's worth noting that slow path would be executed, i.e. function
tlv_allocate_and_initialize_for_key(), for the first time of TLV access.
In this patch:
1. "_tsrm_ls_cache" is guaranteed to be accessed before any VM/JIT code
during the request startup, e.g. in init_executor(), therefore, slow
path can be avoided.
2. As TLVDecriptor is immutable and zend_jit_setup() executes once, we
get this structure in tsrm_get_ls_cache_tcb_offset(). Note the 'ldr'
instruction would be patched to 'add' by the linker.
3. Only fast path for tlv_get_addr() is implemented in macro
LOAD_TSRM_CACHE.
With this patch, all ~4k test cases can pass for ZTS+CALL in macOS on
Apple silicon.
[1] https://gist.github.com/shqking/4aab67e0105f7c1f2c549d57d5799f94
[2]
https://opensource.apple.com/source/dyld/dyld-195.6/src/threadLocalVariables.c.auto.html
[3] https://gist.github.com/shqking/329d7712c26bad49786ab0a544a4af43
Change-Id: I613e9c37e3ff2ecc3fab0f53f1e48a0246e12ee3
We're not checking the return value and the NTS version of this
generates warnings. If we want to handle lock failures, we should
do a hard abort inside tsrm_env_lock() itself.
This patch removes the so called local variables defined per
file basis for certain editors to properly show tab width, and
similar settings. These are mainly used by Vim and Emacs editors
yet with recent changes the once working definitions don't work
anymore in Vim without custom plugins or additional configuration.
Neither are these settings synced across the PHP code base.
A simpler and better approach is EditorConfig and fixing code
using some code style fixing tools in the future instead.
This patch also removes the so called modelines for Vim. Modelines
allow Vim editor specifically to set some editor configuration such as
syntax highlighting, indentation style and tab width to be set in the
first line or the last 5 lines per file basis. Since the php test
files have syntax highlighting already set in most editors properly and
EditorConfig takes care of the indentation settings, this patch removes
these as well for the Vim 6.0 and newer versions.
With the removal of local variables for certain editors such as
Emacs and Vim, the footer is also probably not needed anymore when
creating extensions using ext_skel.php script.
Additionally, Vim modelines for setting php syntax and some editor
settings has been removed from some *.phpt files. All these are
mostly not relevant for phpt files neither work properly in the
middle of the file.
This patch however does not drop support for the BeOS compatible variant, Haiku, see Github PR #2697 which is currently a WiP
I intentionally left out some fragments for BeOS in the build system for that seems to be bundles
Hereby, interned strings are supported in thread safe PHP. The patch
implements two types of interned strings
- interning per process, strings are not freed till process end
- interning per request, strings are freed at request end
There is no runtime interning.
With Opcache, all the permanent iterned strings are copied into SHM on
startup, additional copying into SHM might happen on demand.
If this does not break the Unix system somehow, I'll be amazed. This should get most of it out, apologies for any errors this may cause on non-Windows ends which I cannot test atm.
TLS is already used in TSRM, the way exporting the tsrm cache through
a thread local variable is not portable. Additionally, the current
patch suffers from bugs which are hard to find, but prevent it to
be worky with apache. What is done here is mainly uses the idea
from the RFC patch, but
- __thread variable is removed
- offset math and declarations are removed
- extra macros and definitions are removed
What is done merely is
- use an inline function to access the tsrm cache. The function uses
the portable tsrm_tls_get macro which is cheap
- all the TSRM_* macros are set to placebo. Thus this opens the way
remove them later
Except that, the logic is old. TSRMLS_FETCH will have to be done once
per thread, then tsrm_get_ls_cache() can be used. Things seeming to be
worky are cli, cli server and apache. I also tried to enable bz2
shared and it has worked out of the box. The change is yet minimal
diffing to the current master bus is a worky start, IMHO. Though will
have to recheck the other previously done SAPIs - embed and cgi.
The offsets can be added to the tsrm_resource_type struct, then
it'll not be needed to declare them in the userspace. Even the
"done" member type can be changed to int16 or smaller, then adding
the offset as int16 will not change the struct size. As well on the
todo might be removing the hashed storage, thread_id != thread_id and
linked list logic in favour of the explicit TLS operations.
