Don't call emscripten_tls_init from static constructor. NFC #14981
Conversation
sbc100
force-pushed
the
tls_init_no_ctor
branch
4 times, most recently
from
5623ed5 to
14b3503
Compare
This change makes handling of `emscripten_tls_init` the same on all threads, including the main thread and/or library loading thread. This is precursor to a change that makes use of the return value of emscripten_tls_init to do TLS relocations so making the call explicit is required. I've also removed the use of EM_ASM from library_pthread.c.
sbc100
force-pushed
the
tls_init_no_ctor
branch
from
14b3503 to
f8679e0
Compare
| #if PTHREADS_PROFILING | ||
| PThread.createProfilerBlock(tb); | ||
| PThread.setThreadName(tb, "Browser main thread"); | ||
| PThread.setThreadStatus(tb, {{{ cDefine('EM_THREAD_STATUS_RUNNING') }}}); |
There was a problem hiding this comment.
Why don't we need to set the thread status any more down in __emscripten_init_main_thread_js?
There was a problem hiding this comment.
I've added call to PThread.threadInit which takes case of setThreadStatus. Before threadInit was only called on threads. Now its called on the main thread too (during __emscripten_init_main_thread_js).
| PThread.tlsInitFunctions.push(initTLS); | ||
| if (runtimeInitialized) { | ||
| initTLS(); | ||
| } |
There was a problem hiding this comment.
It looks like this pushes and calls initTLS whether or not we are in a pthread, but previously both only happened when not in a pthread. Why is this change correct?
There was a problem hiding this comment.
Before the main thread would run its TLS inits from the static ctor, but the threads would need to run this explicitly. The point of this change is to remove this difference and always run init TLS explicitly.
This change makes handling of
emscripten_tls_initthesame on all threads, including the main thread and/or library
loading thread.
This is precursor to a change that makes use of the return
value of
emscripten_tls_initto do TLS relocations so makingthe call explicit is required.