[RFC] ptrace: add generic SET_SYSCALL request

[will.deacon@arm.com (Will Deacon)]

On Fri, Nov 07, 2014 at 12:44:07PM +0000, Arnd Bergmann wrote:
> On Friday 07 November 2014 12:11:19 Russell King - ARM Linux wrote:
> > On Fri, Nov 07, 2014 at 01:03:00PM +0100, Arnd Bergmann wrote:
> > > On Friday 07 November 2014 11:55:51 Will Deacon wrote:
> > > > We need this for arm64 and, since all architectures seem to have a mechanism
> > > > for setting a system call via ptrace, moving it to generic code should make
> > > > sense for new architectures too, no?
> > > 
> > > It makes a little more sense now, but I still don't understand why you
> > > need to set the system call number via ptrace. What is this used for,
> > > and why doesn't any other architecture have this?
> > 
> > All other architectures have a way.  x86, for example, you set orig_eax
> > (or orig_rax) to change the syscall number.  On ARM, that doesn't work
> > because we don't always pass the syscall number in a register.
> > 
> 
> Sorry for being slow today, but why can't we use the same interface that
> s390 has on arm64:
> 
> static int s390_system_call_get(struct task_struct *target,
>                                 const struct user_regset *regset,
>                                 unsigned int pos, unsigned int count,
>                                 void *kbuf, void __user *ubuf)
> {
>         unsigned int *data = &task_thread_info(target)->system_call;
>         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
>                                    data, 0, sizeof(unsigned int));
> }
> 
> static int s390_system_call_set(struct task_struct *target,
>                                 const struct user_regset *regset,
>                                 unsigned int pos, unsigned int count,
>                                 const void *kbuf, const void __user *ubuf)
> {
>         unsigned int *data = &task_thread_info(target)->system_call;
>         return user_regset_copyin(&pos, &count, &kbuf, &ubuf,
>                                   data, 0, sizeof(unsigned int));
> }
> 
> static const struct user_regset s390_regsets[] = {
> 	...
>         {
>                 .core_note_type = NT_S390_SYSTEM_CALL,
>                 .n = 1,
>                 .size = sizeof(unsigned int),
>                 .align = sizeof(unsigned int),
>                 .get = s390_system_call_get,
>                 .set = s390_system_call_set,
>         },
> 	...
> };
> 
> Is it just preference for being consistent with ARM32, or is there a
> reason this won't work?

Interesting, I hadn't considered a unit-length regset.

> It's not that I care strongly about the interface, my main point is
> that the changelog doesn't describe why one interface was used instead
> the other.

I suspect the current approach was taken because it follows the same scheme
as 32-bit ARM. If both methods are sufficient (Kees would have a better idea
than me on that), then I don't have a strong preference.

Will