Re: [PATCH] newlib/configure.host: Turn off sti/cli for i386 setjmp for i386-rtems*

[Joel Sherrill <joel dot sherrill at oarcorp dot com>]

On 2/26/2016 4:44 AM, Corinna Vinschen wrote:
> On Feb 25 14:44, Joel Sherrill wrote:
> > When running RTEMS paravirtualized in user space, the cli/sti instructions
> > can't be used. But I can't see why they are needed in either user or
> > supervisor state. Turning the use of them off for RTEMS.
> >
> > This code is unmodified since the beginning of the repository, so I have
> > no insight into why it was done. I may even have been the one to add them.
> > Absolutely no idea except they aren't necessary.
> >
> > ---
> >   newlib/configure.host | 2 ++
> >   1 file changed, 2 insertions(+)
> >
> > diff --git a/newlib/configure.host b/newlib/configure.host
> > index 5fa37d4..f6abb5c 100644
> > --- a/newlib/configure.host
> > +++ b/newlib/configure.host
> > @@ -592,6 +592,8 @@ case "${host}" in
> >   	newlib_cflags="${newlib_cflags} -DMALLOC_PROVIDED -DEXIT_PROVIDED -DSIGNAL_PROVIDED -DREENTRANT_SYSCALLS_PROVIDED -DHAVE_NANOSLEEP -DHAVE_BLKSIZE -DHAVE_FCNTL -DHAVE_ASSERT_FUNC"
> >           # turn off unsupported items in posix directory
> >   	newlib_cflags="${newlib_cflags} -D_NO_GETLOGIN -D_NO_GETPWENT -D_NO_GETUT -D_NO_GETPASS -D_NO_SIGSET -D_NO_WORDEXP -D_NO_POPEN -D_NO_POSIX_SPAWN"
> > +        # turn off using cli/sti in i386 setjmp/longjmp
> > +	newlib_cflags="${newlib_cflags} -D_I386MACH_ALLOW_HW_INTERRUPTS"
> >   	;;
> >   # VxWorks supplies its own version of malloc, and the newlib one
> >   # doesn't work because VxWorks does not have sbrk.
>
> Patch is ok, please apply.
>
> Would it make sense to enable _I386MACH_ALLOW_HW_INTERRUPTS by default
> on x86/x86_64?

I can't see any downside. I spot checked other ports and no other port seemed to have comparable code. I would be OK with that. Where is it defaulted?

Also FWIW, I would like to see this resolved before the next newlib snapshot
so the RTEMS tools can include this. We are due for a tool update and I don't
want to impact users too much in a single month. :)


 
> Thanks,
> Corinna

--joel