[PATCH] D46459: [asan] Make GetCurrentThread RTEMS-friendly

[Walter Lee via Phabricator via llvm-commits]

waltl added inline comments.


================
Comment at: compiler-rt/lib/asan/asan_thread.cc:394
+    return nullptr;
+
   AsanThreadContext *context =
----------------
waltl wrote:
> alekseyshl wrote:
> > vitalybuka wrote:
> > > alekseyshl wrote:
> > > > vitalybuka wrote:
> > > > > What if you just put here and we avoid platform specific branch
> > > > > ```
> > > > > if (UNLIKELY(!asan_inited))
> > > > >   AsanInitFromRtl();
> > > > > ```
> > > > > 
> > > > > !asan_inited is not expected here on any other platform
> > > > But it still be a check, still some code executed, even if it is unlikely. With SANITIZER_RTEMS, it will be removed on all platforms, but RTEMS.
> > > > 
> > > > The semantics is also wrong, this way it creates an impression that it can happen on all platforms.
> > > With first part I agree, we don't want additional check here as this can be called quite frequently.  So it makes sense to move AsanInitFromRtl() into __asan_handle_no_return
> > > 
> > > For semantics, I'd rather have as much as possible same function on all platform, even if it handles cases which are not necessary on some of them.
> > From the patch comment it can be called "after ASan run-time has been destroyed". Why would we want to call AsanInitFromRtl again?
> Unfortunately it's not safe to call AsanInitFromRtl() under these conditions.  It could be before pthreads have been initialized.
> 
And yes, the issue also applies after run-time destruction.



Repository:
  rL LLVM

https://reviews.llvm.org/D46459