Re: [PATCH] Add a text_poke syscall v2

[Oleg Nesterov <oleg@redhat.com>]

On 11/30, Jiri Kosina wrote:
>
> On Fri, 29 Nov 2013, Oleg Nesterov wrote:
>
> > > > Andi, et al. I am going to discuss the things I do not really
> > > > understand, probably this can't make any sense, but...
> > >
> > > I think it's enough to set the dirty bit in the underlying
> > > struct page, no need to play games with the PTE.
> >
> > Ah, sorry for confusion, I guess you misunderstood.
> >
> > I meant, perhaps sys_text_poke() doesn't the in-kernel text_poke
> > machinery altogether?
> >
> > Can't we invalidate pte (so that any user will stuck in page fault),
> > update the page(s), restore the pte and drop the locks?
>
> Do you think this'd be faster than the int3-based aproach?

No.

And more, I simply do not know if it would be slower or faster, and how
much. Just I hope that this won't be "much" slower.

OTOH, this is obviously more scalable, and this way sys_text_poke() won't
block, say, jump_label or kprobes. Not sure this actually matters though.

> We have moved from using stop_machine() to int3-based patching exactly
> because it's much more lightweight.

Oh, I do not think it makes sense to compare stop_machine() with this
approach...

Oleg.