Re: [PATCH] ia64: enable HAVE_COPY_THREAD_TLS, switch to kernel_clone_args

[Christian Brauner <christian.brauner@ubuntu.com>]

On Thu, May 14, 2020 at 12:15:42PM +0200, Christian Brauner wrote:
> On Thu, May 14, 2020 at 12:08:27PM +0200, John Paul Adrian Glaubitz wrote:
> > On 5/14/20 12:04 PM, Christian Brauner wrote:
> > > So that's interesting because systemd doesn't support itanium correctly afaict.
> > > They have a raw_clone() function that they use which is not tailored to
> > > ia64. __NR_clone should be defined as so to hit clone2() but they don't
> > > pass a stack size argument down which is invalid on ia64:
> > Ah, I wasn't aware of this limitation. I'm surprised the systemd testsuite passes
> > then on ia64.
> > 
> > On sparc64, Michael Karcher provided the hand-written assembly you quoted to
> > fix the raw_clone() on this architecture.
> > 
> > I assume we could do the same on ia64?
> 
> I think the following should be correct:
> 
> (Probably worth checking that __NR_clone and __NR_clone2 are the same
> for ia64. Also note, that __NR_clone was _never_ supported by glibc on
> ia64. They even have this comment:
> 
> /* clone is not supported under Linux/ia64, use clone2. */
> )
> 
> static inline pid_t raw_clone(unsigned long flags) {
>         pid_t ret;
> 
>         assert((flags & (CLONE_VM|CLONE_PARENT_SETTID|CLONE_CHILD_SETTID|
>                          CLONE_CHILD_CLEARTID|CLONE_SETTLS)) == 0);
> #if defined(__s390x__) || defined(__s390__) || defined(__CRIS__)
>         /* On s390/s390x and cris the order of the first and second arguments
>          * of the raw clone() system call is reversed. */
>         ret = (pid_t) syscall(__NR_clone, NULL, flags);
> #elif defined(__sparc__)
>         {
>                 /**
>                  * sparc always returns the other process id in %o0, and
>                  * a boolean flag whether this is the child or the parent in
>                  * %o1. Inline assembly is needed to get the flag returned
>                  * in %o1.
>                  */
>                 int in_child, child_pid, error;
> 
>                 asm volatile("mov %3, %%g1\n\t"
>                              "mov %4, %%o0\n\t"
>                              "mov 0 , %%o1\n\t"
> #if defined(__arch64__)
>                              "t 0x6d\n\t"
> #else
>                              "t 0x10\n\t"
> #endif
>                              "addx %%g0, 0, %2\n\t"
>                              "mov %%o1, %0\n\t"
>                              "mov %%o0, %1" :
>                              "=r"(in_child), "=r"(child_pid), "=r"(error) :
>                              "i"(__NR_clone), "r"(flags) :
>                              "%o1", "%o0", "%g1", "cc" );
> 
>                 if (error) {
>                         errno = child_pid;
>                         ret = -1;
>                 } else
>                         ret = in_child ? 0 : child_pid;
>         }
> +#elif defined(__ia64__)
> +	/* On ia64 the stack and stack size are passed as separate arguments. */
> +	return (pid_t)syscall(__NR_clone, flags, NULL, 0);
> +#else
> +	return (pid_t)syscall(__NR_clone, flags, NULL);
> +#endif

Scratch that. It's even worse. On ia64 it is _invalid_ to pass a NULL
stack. That's at least what the glibc assembly assumes:

	cmp.eq p6,p0=0,in0
	cmp.eq p7,p0=0,in1
	mov r8=EINVAL
	mov out0=in3		/* Flags are first syscall argument.	*/
	mov out1=in1		/* Stack address.			*/
(p6)	br.cond.spnt.many __syscall_error	/* no NULL function pointers */
(p7)	br.cond.spnt.many __syscall_error	/* no NULL stack pointers */
	;;
	mov out2=in2		/* Stack size.				*/

so newer systemd just works by accident on ia64 if at all correctly
afaict.