This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: [PATCH -tip v5 03/10] kprobes: Introduce kprobes jump optimization
- From: Masami Hiramatsu <mhiramat at redhat dot com>
- To: Frederic Weisbecker <fweisbec at gmail dot com>
- Cc: Ingo Molnar <mingo at elte dot hu>, Ananth N Mavinakayanahalli <ananth at in dot ibm dot com>, lkml <linux-kernel at vger dot kernel dot org>, systemtap <systemtap at sources dot redhat dot com>, DLE <dle-develop at lists dot sourceforge dot net>, Jim Keniston <jkenisto at us dot ibm dot com>, Srikar Dronamraju <srikar at linux dot vnet dot ibm dot com>, Christoph Hellwig <hch at infradead dot org>, Steven Rostedt <rostedt at goodmis dot org>, "H. Peter Anvin" <hpa at zytor dot com>, Anders Kaseorg <andersk at ksplice dot com>, Tim Abbott <tabbott at ksplice dot com>, Andi Kleen <andi at firstfloor dot org>, Jason Baron <jbaron at redhat dot com>, Mathieu Desnoyers <mathieu dot desnoyers at polymtl dot ca>
- Date: Tue, 24 Nov 2009 10:34:08 -0500
- Subject: Re: [PATCH -tip v5 03/10] kprobes: Introduce kprobes jump optimization
- References: <20091123232115.22071.71558.stgit@dhcp-100-2-132.bos.redhat.com> <20091123232141.22071.53317.stgit@dhcp-100-2-132.bos.redhat.com> <20091124024417.GA6752@nowhere>
Hi Frederic,
Frederic Weisbecker wrote:
> On Mon, Nov 23, 2009 at 06:21:41PM -0500, Masami Hiramatsu wrote:
>> +config OPTPROBES
>> + bool "Kprobes jump optimization support (EXPERIMENTAL)"
>> + default y
>> + depends on KPROBES
>> + depends on !PREEMPT
>
>
> Why does it depends on !PREEMPT?
Oh, because it has not supported preemptive kernel yet.
(I'd like to tell you why in another mail)
>> @@ -301,6 +302,31 @@ void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
>> __free_insn_slot(&kprobe_insn_slots, slot, dirty);
>> mutex_unlock(&kprobe_insn_mutex);
>> }
>> +#ifdef CONFIG_OPTPROBES
>> +/* For optimized_kprobe buffer */
>> +static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */
>> +static struct kprobe_insn_cache kprobe_optinsn_slots = {
>> + .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
>> + /* .insn_size is initialized later */
>> + .nr_garbage = 0,
>> +};
>> +/* Get a slot for optimized_kprobe buffer */
>> +kprobe_opcode_t __kprobes *get_optinsn_slot(void)
>> +{
>> + kprobe_opcode_t *ret = NULL;
>> + mutex_lock(&kprobe_optinsn_mutex);
>> + ret = __get_insn_slot(&kprobe_optinsn_slots);
>> + mutex_unlock(&kprobe_optinsn_mutex);
>> + return ret;
>> +}
>
>
>
> Just a small nano-neat: could you add a line between variable
> declarations and the rest? And also just before the return?
> It makes the code a bit easier to review.
Sure :-)
>> +static void kprobe_optimizer(struct work_struct *work);
>> +static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
>> +#define OPTIMIZE_DELAY 5
>> +
>> +/* Kprobe jump optimizer */
>> +static __kprobes void kprobe_optimizer(struct work_struct *work)
>> +{
>> + struct optimized_kprobe *op, *tmp;
>> +
>> + /* Lock modules while optimizing kprobes */
>> + mutex_lock(&module_mutex);
>> + mutex_lock(&kprobe_mutex);
>> + if (kprobes_all_disarmed)
>> + goto end;
>> +
>> + /* Wait quiesence period for ensuring all interrupts are done */
>> + synchronize_sched();
>
>
>
> It's not clear to me why you are doing that.
> Is this waiting for pending int 3 kprobes handlers
> to complete? If so, why, and what does that prevent?
>
> Also, why is it a delayed work? I'm not sure what we are
> waiting for here.
[...]
> Again, I think this dance with live patching protected
> by int 3 only, which patching is in turn a necessary
> stage before, is a complicated sequence that could be
> simplified by choosing only one patching in stop_machine()
> time.
There is a reason why we have to wait here and it's excuse
why it hasn't supported preemption yet too, I'll tell you
in next mail :-)
>> +
>> + get_online_cpus(); /* Use online_cpus while optimizing */
>
>
>
> And this comment doesn't tell us much what this brings us.
> The changelog tells it stands to avoid a text_mutex deadlock.
> I'm not sure why we would deadlock without it.
As Mathieu and I discussed on LKML (http://lkml.org/lkml/2009/11/21/187)
text_mutex will be locked on the way of cpu-hotplug.
Since kprobes locks text_mutex too and stop_machine() refers online_cpus,
it will cause a dead-lock. So, I decided to use get_online_cpus() to
locking hotplug while optimizing/unoptimizng.
Thank you,
--
Masami Hiramatsu
Software Engineer
Hitachi Computer Products (America), Inc.
Software Solutions Division
e-mail: mhiramat@redhat.com