Message-ID: <20091003123900.GA22046@Krystal>
Date:	Sat, 3 Oct 2009 08:39:00 -0400
From:	Mathieu Desnoyers <mathieu.desnoyers@...ymtl.ca>
To:	Ingo Molnar <mingo@...e.hu>
Cc:	Jason Baron <jbaron@...hat.com>, linux-kernel@...r.kernel.org,
	tglx@...utronix.de, rostedt@...dmis.org, ak@...e.de,
	roland@...hat.com, rth@...hat.com, mhiramat@...hat.com
Subject: Re: [PATCH 1/4] jump label - make init_kernel_text() global

* Ingo Molnar (mingo@...e.hu) wrote:
> 
> * Jason Baron <jbaron@...hat.com> wrote:
> 
> > On Thu, Oct 01, 2009 at 01:20:03PM +0200, Ingo Molnar wrote:
> > > * Jason Baron <jbaron@...hat.com> wrote:
> > > 
> > > > allow usage of init_kernel_text - we need this in jump labeling to 
> > > > avoid attemtpting to patch code that has been freed as in the __init 
> > > > sections
> > > 
> > > s/attemtpting/attempting
> > > 
> > > > Signed-off-by: Jason Baron <jbaron@...hat.com>
> > > > ---
> > > >  include/linux/kernel.h |    1 +
> > > >  kernel/extable.c       |    2 +-
> > > >  2 files changed, 2 insertions(+), 1 deletions(-)
> > > > 
> > > > diff --git a/include/linux/kernel.h b/include/linux/kernel.h
> > > > index f61039e..9d3419f 100644
> > > > --- a/include/linux/kernel.h
> > > > +++ b/include/linux/kernel.h
> > > > @@ -295,6 +295,7 @@ extern int get_option(char **str, int *pint);
> > > >  extern char *get_options(const char *str, int nints, int *ints);
> > > >  extern unsigned long long memparse(const char *ptr, char **retptr);
> > > >  
> > > > +extern int init_kernel_text(unsigned long addr);
> > > >  extern int core_kernel_text(unsigned long addr);
> > > >  extern int __kernel_text_address(unsigned long addr);
> > > >  extern int kernel_text_address(unsigned long addr);
> > > > diff --git a/kernel/extable.c b/kernel/extable.c
> > > > index 7f8f263..f6893ad 100644
> > > > --- a/kernel/extable.c
> > > > +++ b/kernel/extable.c
> > > > @@ -52,7 +52,7 @@ const struct exception_table_entry *search_exception_tables(unsigned long addr)
> > > >  	return e;
> > > >  }
> > > >  
> > > > -static inline int init_kernel_text(unsigned long addr)
> > > > +int init_kernel_text(unsigned long addr)
> > > >  {
> > > >  	if (addr >= (unsigned long)_sinittext &&
> > > >  	    addr <= (unsigned long)_einittext)
> > > 
> > > i'm confused. Later on jump_label_update() does:
> > > 
> > > +                       if (!(system_state == SYSTEM_RUNNING &&
> > > +                                       (init_kernel_text(iter->code))))
> > > +                               jump_label_transform(iter, type);
> > > 
> > > which is:
> > > 
> > > +                       if (system_state != SYSTEM_RUNNING ||
> > > +                                       !init_kernel_text(iter->code)))
> > > +                               jump_label_transform(iter, type);
> > > 
> > > What is the logic behind that? System going into SYSTEM_RUNNING does not 
> > > coincide with free_initmem() precisely.
> > > 
> > 
> > The specific case I hit was in modifying code in arch_kdebugfs_init()
> > which is '__init' after the system was up and running. The tracepoint is
> > in 'kmalloc()' which is marked as __always_inline.
> > 
> > 
> > > Also, do we ever want to patch init-text tracepoints? I think we want to 
> > > stay away from them as much as possible.
> > 
> > I was trying to make sure that tracepoints in init-text were honored.
> > 
> > > 
> > > It appears to me that what we want here is a straight:
> > > 
> > >                        if (kernel_text(iter->code))
> > >                                jump_label_transform(iter, type);
> > > 
> > > Also, maybe a WARN_ONCE(!kernel_text()) - we should never even attempt 
> > > to transform non-patchable code. If yes then we want to know about that 
> > > in a noisy way and not skip it silently.
> > > 
> > 
> > hmmm....indeed, kernel_text_address() does do what I want here (I must 
> > have mis-read its definition). Although, I'm not sure there isn't a 
> > race here betweeen freeing the init sections and possibly updating 
> > them. For modules, there is no race since the module init free code 
> > takes the module_mutex, and I do as well in this code...
> > 
> > I've now also tested this code on 32-bit x86 system, and it seems to 
> > perform nicely. I'm seeing a 15 cycle improvement per tracepoint.
> > 
> > I've based the text section updating on text_poke_fixup(), which has 
> > recently come into question about safety of cross modifying code. I 
> > could rebase my patches back to use stop_machine()? I guess I'm 
> > looking for some advice on how to proceed here.
> 
> I think this very limited form of code patching that you are using here 
> (patching a JMP) _should_ be safe - so we can avoid stop_machine().
> 

I might be missing a bit of context here, I just want to make sure we
are on the same page: patching a jmp instruction is safe on UP, safe
with stop_machine(), is very likely safe with the breakpoint-ipi
approach (but we need the confirmation from Intel, which hpa is trying
to get), but is definitely _not_ safe if neither of these methods are
used on a SMP system. If a non-aligned multi-word jump is modified while
another CPU is fetching the instruction, bad things could happen.

BTW, patching kernel and module init sections can be done without
sop_machine(), because only one CPU is ever accessing the init code.

But again, I might be missing some context. If so, sorry for the noise.

Thanks,

Mathieu

> 	Ingo

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists