Re: [PATCH 2/6] uprobes: introduce is_swbp_at_addr_fast()

[Peter Zijlstra]

On Mon, 2012-04-16 at 16:44 +0200, Oleg Nesterov wrote:
> On 04/16, Peter Zijlstra wrote:
> >
> > On Fri, 2012-04-06 at 00:21 +0200, Oleg Nesterov wrote:
> > > +int __weak is_swbp_at_addr_fast(unsigned long vaddr)
> > > +{
> > > +       uprobe_opcode_t opcode;
> > > +       int fault;
> > > +
> > > +       pagefault_disable();
> > > +       fault = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
> > > +                                                       sizeof(opcode));
> > > +       pagefault_enable();
> > > +
> > > +       if (unlikely(fault)) {
> > > +               /*
> > > +                * XXX: read_opcode() lacks FOLL_FORCE, it can fail if
> > > +                * we race with another thread which does mprotect(NONE)
> > > +                * after we hit bp.
> > > +                */
> > > +               if (read_opcode(current->mm, vaddr, &opcode))
> > > +                       return -EFAULT;
> > > +       }
> > > +
> > > +       return is_swbp_insn(&opcode);
> > > +}
> >
> > Why bother with the pagefault_disable() and unlikely fault case and not
> > simply do copy_from_user() and have it deal with the fault if its needed
> > anyway?
> 
> But we can't do this under down_read(mmap_sem) ?
> 
> If another thread waits for down_write() then do_page_fault() can't take
> this lock, right?

Ah, indeed, I thought read_opcode() would do the fault, but that's
get_user_pages() which requires the caller to hold mmap_sem instead.

Can't we 'optimize' read_opcode() by doing the pagefault_disable() +
__copy_from_user_inatomic() optimistically before going down the whole
gup()+lock+kmap path?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/