On Thu, Jan 12, 2023 at 11:38:17AM +0000, Mark Rutland wrote: > On Wed, Jan 11, 2023 at 06:27:53PM +0000, Mark Rutland wrote: > > On Mon, Jan 09, 2023 at 03:43:16PM +0100, Miguel Ojeda wrote: > > > On Mon, Jan 9, 2023 at 2:58 PM Mark Rutland <mark.rutland@xxxxxxx> wrote: > > > > > > > > As far as I can tell, GCC doesn't respect '-falign-functions=N': > > > > > > > > * When the __weak__ attribute is used > > > > > > > > GCC seems to forget the alignment specified by '-falign-functions=N', > > > > but will respect the '__aligned__(N)' function attribute. Thus, we can > > > > work around this by explciitly setting the alignment for weak > > > > functions. > > > > > > > > * When the __cold__ attribute is used > > > > > > > > GCC seems to forget the alignment specified by '-falign-functions=N', > > > > and also doesn't seem to respect the '__aligned__(N)' function > > > > attribute. The only way to work around this is to not use the __cold__ > > > > attibute. > > > > > > If you happen to have a reduced case, then it would be nice to link it > > > in the commit. A bug report to GCC would also be nice. > > > > > > I gave it a very quick try in Compiler Explorer, but I couldn't > > > reproduce it, so I guess it depends on flags, non-trivial functions or > > > something else. > > > > So having spent today coming up with tests, it turns out it's not quite as I > > described above, but in a sense worse. I'm posting a summary here for > > posterity; I'll try to get this to compiler folk shortly. > > I've added the cold bits to an existing ticket: > > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345 > > I have not been able to reproduce the issue with __weak__, so I'll go dig into > that some more; it's likely I was mistaken there. It turns out that was a red herring; GCC is actually implicitly marking the abort() function as cold, and as Linux's implementation happened to be marked as weak I assumed that was the culprit. I'll drop the changes to weak and update our abort implementation specifically, with a comment. I'll also go update the ticket above. Thanks, Mark.
