Re: [RFC v3 08/19] arch: um: add shim to trap to allow installing a fault catcher for tests

[Luis Chamberlain <mcgrof-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

On Mon, Dec 03, 2018 at 03:34:57PM -0800, Brendan Higgins wrote:
> On Thu, Nov 29, 2018 at 7:34 PM Luis Chamberlain <mcgrof-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> >
> > On Wed, Nov 28, 2018 at 11:36:25AM -0800, Brendan Higgins wrote:
> > > diff --git a/arch/um/kernel/trap.c b/arch/um/kernel/trap.c
> > > index cced829460427..bf90e678b3d71 100644
> > > --- a/arch/um/kernel/trap.c
> > > +++ b/arch/um/kernel/trap.c
> > > @@ -201,6 +201,12 @@ void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
> > >       segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
> > >  }
> > >
> > > +static void segv_run_catcher(jmp_buf *catcher, void *fault_addr)
> > > +{
> > > +     current->thread.fault_addr = fault_addr;
> > > +     UML_LONGJMP(catcher, 1);
> > > +}
> > > +
> > >  /*
> > >   * We give a *copy* of the faultinfo in the regs to segv.
> > >   * This must be done, since nesting SEGVs could overwrite
> > > @@ -219,7 +225,10 @@ unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
> > >       if (!is_user && regs)
> > >               current->thread.segv_regs = container_of(regs, struct pt_regs, regs);
> > >
> > > -     if (!is_user && (address >= start_vm) && (address < end_vm)) {
> > > +     catcher = current->thread.fault_catcher;
> >
> > This and..
> >
> > > +     if (catcher && current->thread.is_running_test)
> > > +             segv_run_catcher(catcher, (void *) address);
> > > +     else if (!is_user && (address >= start_vm) && (address < end_vm)) {
> > >               flush_tlb_kernel_vm();
> > >               goto out;
> > >       }
> >
> > *not this*
> 
> I don't understand. Are you saying the previous block of code is good
> and this one is bad?

No, I was saying that the above block of code is a functional change,
but I was also pointing out other areas which were not and could be
folded into a separate atomic patch where no functionality changes.

> > > @@ -246,12 +255,10 @@ unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
> > >               address = 0;
> > >       }
> > >
> > > -     catcher = current->thread.fault_catcher;
> > >       if (!err)
> > >               goto out;
> > >       else if (catcher != NULL) {
> > > -             current->thread.fault_addr = (void *) address;
> > > -             UML_LONGJMP(catcher, 1);
> > > +             segv_run_catcher(catcher, (void *) address);
> > >       }
> > >       else if (current->thread.fault_addr != NULL)
> > >               panic("fault_addr set but no fault catcher");
> >
> > But with this seems one atomic change which should be submitted
> > separately, its just a helper. Think it would make the actual
> > change needed easier to review, ie, your needed changes would
> > be smaller and clearer for what you need.
> 
> Are you suggesting that I pull out the bits needed to implement abort
> in the next patch and squash it into this one?

No, I'm suggesting you can probably split this patch in 2, one which
wraps things with no functional changes, and another which adds your
changes.

  Luis