[lkml]   [2014]   [Apr]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v2 6/9] uprobes/x86: Introduce uprobe_xol_ops and arch_uprobe->ops
On 04/08, Masami Hiramatsu wrote:
> (2014/04/05 3:51), Oleg Nesterov wrote:
> >
> > TODO: An error from adjust_ret_addr() shouldn't be silently ignored,
> > we should teach arch_uprobe_post_xol() or handle_singlestep() paths
> > to restart the probed insn in this case. And probably "adjust" can
> > be simplified and turned into set_ret_addr(). It seems that we do
> > not really need copy_from_user(), we can always calculate the value
> > we need to write into *regs->sp.
> It seems that you fixed this in 8/9, we don't need the TODO list in
> the description.

Well, OK, I'll update the changelog and remove the "error ... ignored"
part. Although to be honest, I do not understand why do you think it
is bad to document the other problems you found while you were writing
the patch.

> > + if (auprobe->ops->emulate)
> > + return auprobe->ops->emulate(auprobe, regs);
> > +
> > + /* TODO: move this code into ->emulate() hook */
> If you think this can move into the emulate(),

Yes sure,

> you should do in this
> patch.

No, sorry, I strongly disagree, this should come as a separate change,
and only after "Emulate jmp's".

> Since the following code runs by default, there should be
> no problem to do that.

Hmm. Not sure I understand "by default". If you meant that this should
go into default_emulate_op() (which we do not have), then I strongly
disagree again.

It should not, if nothing else we need to record insn->length somewhere,
this should go into ttt_emulate_op() we add later. And it simply looks
much more natural to handle jmp's and nop's together.

Masami, this time I simply can't understand your objections, please



 \ /
  Last update: 2014-04-08 18:41    [W:0.081 / U:1.268 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site