Re: [PATCH -tip 5/4] Expands irq-off region in text_poke()

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Masami Hiramatsu (mhiramat@redhat.com) wrote:
> 
> 
> Mathieu Desnoyers wrote:
> > * Masami Hiramatsu (mhiramat@redhat.com) wrote:
> >> Ingo Molnar wrote:
> >>> * Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> wrote:
> >>>
> >>>> * Ingo Molnar (mingo@elte.hu) wrote:
> >>>>> * Masami Hiramatsu <mhiramat@redhat.com> wrote:
> >>>>>
> >>>>>> @@ -523,14 +526,17 @@ void *__kprobes text_poke(void *addr, co
> >>>>>>  		pages[1] = virt_to_page(addr + PAGE_SIZE);
> >>>>>>  	}
> >>>>>>  	BUG_ON(!pages[0]);
> >>>>>> -	if (!pages[1])
> >>>>>> -		nr_pages = 1;
> >>>>>> -	vaddr = vmap(pages, nr_pages, VM_MAP, PAGE_KERNEL);
> >>>>>> -	BUG_ON(!vaddr);
> >>>>>> -	local_irq_disable();
> >>>>>> +	local_irq_save(flags);
> >>>>>> +	set_fixmap(FIX_TEXT_POKE0, page_to_phys(pages[0]));
> >>>>>> +	if (pages[1])
> >>>>>> +		set_fixmap(FIX_TEXT_POKE1, page_to_phys(pages[1]));
> >>>>>> +	vaddr = (char *)fix_to_virt(FIX_TEXT_POKE0);
> >>>>>>  	memcpy(&vaddr[(unsigned long)addr & ~PAGE_MASK], opcode, len);
> >>>>>> -	local_irq_enable();
> >>>>>> -	vunmap(vaddr);
> >>>>>> +	clear_fixmap(FIX_TEXT_POKE0);
> >>>>>> +	if (pages[1])
> >>>>>> +		clear_fixmap(FIX_TEXT_POKE1);
> >>>>>> +	local_flush_tlb();
> >>>>>> +	local_irq_restore(flags);
> >>>>>>  	sync_core();
> >>>>> I'm not sure at all about this widening of the irq-atomic 
> >>>>> section and the idea of allowing non-locked access on single-CPU 
> >>>>> situations - we dont really want to micro-optimize any of this 
> >>>>> on such a level, holding the text lock is a robust rule all code 
> >>>>> should be listening to. (Creating locking assymetry always 
> >>>>> inserts a certain amount of fragility - adding to an already 
> >>>>> fragile concept here.)
> >>>>>
> >>>>> And note that there's no reason why text_poke could not be used 
> >>>>> in stop_machine_run() - the stop_machine_run() handler must not 
> >>>>> take the text_lock of course - but outside code calling 
> >>>>> stop_machine_run() can do it and can hence serialize properly.
> >>>>>
> >>>>> Note that even if we did this then your v2 patch is not fully 
> >>>>> correct: you need to move the sync_core() at the end of the 
> >>>>> sequence inside the critical section too. (right now this is 
> >>>>> mostly harmless because the INVLPG inside the clear_fixmap() 
> >>>>> happens to be serializing so it has an implicit sync_core() 
> >>>>> property - but nevertheless we better do this straight away to 
> >>>>> not cause problems later down the line.)
> >>>>>
> >>>>> 	Ingo
> >>>> Agreed. The alternatives_smp_lock/alternatives_smp_unlock 
> >>>> specific case does not bring us much if it has no perceivable 
> >>>> performance impact. It's better to keep a standard interface 
> >>>> and clear requirements.
> >>> Note that i dont object to another aspect of this same change: 
> >>> the fact that it makes the whole sequence more atomic and more 
> >>> defensive [which is never bad of fragile interfaces].
> >>>
> >>> I only got worried about the "lets use this without the text 
> >>> lock" ideas.
> >>>
> >>> So if Masami-san sends a delta patch with a different changelog 
> >>> and with the sync_core() bit moved inside the critical section, 
> >>> i'll apply that too.
> >> OK, here is the delta patch.
> >>
> >> Expand irq-atomic region to cover fixmap using code and sync_core.
> >>
> >> Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
> >> Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
> >> Cc: Ingo Molnar <mingo@elte.hu>
> >> ---
> >>  arch/x86/kernel/alternative.c |    4 ++--
> >>  1 file changed, 2 insertions(+), 2 deletions(-)
> >>
> >> Index: linux-2.6-tip/arch/x86/kernel/alternative.c
> >> ===================================================================
> >> --- linux-2.6-tip.orig/arch/x86/kernel/alternative.c
> >> +++ linux-2.6-tip/arch/x86/kernel/alternative.c
> >> @@ -526,13 +526,12 @@ void *__kprobes text_poke(void *addr, co
> >>  		pages[1] = virt_to_page(addr + PAGE_SIZE);
> >>  	}
> >>  	BUG_ON(!pages[0]);
> >> +	local_irq_save(flags);
> >>  	set_fixmap(FIX_TEXT_POKE0, page_to_phys(pages[0]));
> >>  	if (pages[1])
> >>  		set_fixmap(FIX_TEXT_POKE1, page_to_phys(pages[1]));
> >>  	vaddr = (char *)fix_to_virt(FIX_TEXT_POKE0);
> >> -	local_irq_save(flags);
> >>  	memcpy(&vaddr[(unsigned long)addr & ~PAGE_MASK], opcode, len);
> >> -	local_irq_restore(flags);
> >>  	clear_fixmap(FIX_TEXT_POKE0);
> >>  	if (pages[1])
> >>  		clear_fixmap(FIX_TEXT_POKE1);
> >> @@ -540,6 +539,7 @@ void *__kprobes text_poke(void *addr, co
> >>  	sync_core();
> >>  	/* Could also do a CLFLUSH here to speed up CPU recovery; but
> >>  	   that causes hangs on some VIA CPUs. */
> >> +	local_irq_restore(flags);
> >>  	for (i = 0; i < len; i++)
> >>  		BUG_ON(((char *)addr)[i] != ((char *)opcode)[i]);
> > 
> > I think irq off should cover the BUG_ON too. This safety check assumes
> > we are the only ones modifying "addr".
> 
> I think others don't change without text_mutex, don't it?
> 

They shouldn't, but given we decided to grow the irq off region to
contain all the code that needs to be executed atomically, it should
also contain the BUG_ON, because it is expected to be as atomic as the
rest of the code.

Mathieu

> Thank you,
> 
> -- 
> Masami Hiramatsu
> 
> Software Engineer
> Hitachi Computer Products (America) Inc.
> Software Solutions Division
> 
> e-mail: mhiramat@redhat.com
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68