Re: [PATCH 5/9] jump label: Addregister_jump_label_key/unregister_jump_label_key

From: Peter Zijlstra
Date: Mon Oct 18 2010 - 10:08:20 EST


On Mon, 2010-10-18 at 10:03 -0400, Jason Baron wrote:
> On Mon, Oct 18, 2010 at 02:05:21PM +0200, Peter Zijlstra wrote:
> > On Fri, 2010-10-15 at 23:13 +0200, Peter Zijlstra wrote:
> > > On Fri, 2010-10-15 at 17:09 -0400, Steven Rostedt wrote:
> > > > On Fri, 2010-10-15 at 23:03 +0200, Peter Zijlstra wrote:
> > > >
> > > > > Urgh, this sucks.. :-(
> > > > >
> > > > > So now we have to actually track all JUMP_LABEL() sites and call
> > > > > register muck on them.. even though we already track them through the
> > > > > special data section.
> > > > >
> > > > > Is there really no way around this?
> > > >
> > > > I'll take a look to see if we can monkey with magic and automate it.
> > >
> > > So the problem is something like:
> > >
> > > core kernel:
> > >
> > > jump_label_enable()
> > >
> > > module:
> > >
> > > JUMP_LABEL()
> > >
> > > And then because we don't have a proper __jump_table section, the
> > > jump_label_enable() won't properly work?
> > >
> > > Why not let jump_label_enable() add a dummy entry with the enabled bit
> > > and once you load the module merge the real entry into it.
> >
> > Or actually use the value of the key pointer.. it would mean either
> > standardizing the size (int/atomic_t would work), or using a version of
> > the fallback JUMP_LABEL implementation to sort out the type issues.
>
>
> So I initially implmented this as 'jump_label_enable()' would add a new
> entry for the key, if it didn't already exist. However, I was concerned
> about the case where module 'a' defined the key variable, and then
> module 'b' did the enable/disable, and then module 'a' was removed and
> thus the key value could be re-used, and module's 'b' key would mean
> something different.
>
> However, I'm not sure that is possible - since module 'b' would have
> symbol dependency on module 'a', and thus module 'a' could not be
> unloaded before module 'b'.

Right, if the key variable is part of A and B uses it, then A should not
be allowed to be unloaded.

> Thus, when a module is freed, I think we can scan all the keys and check
> if any key is contained within the text section of the module that is
> about to be freed. If so, we simply remove that key entry. does this
> make sense?

Yep.
--
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/