[-mm][PATCH 0/6] (yet another) kprobes x86 code unification and boosters

[-mm][PATCH 0/6] (yet another) kprobes x86 code unification and boosters

[Masami Hiramatsu]

Hello all,

I developed a series of patches which unifies kprobes code on x86
and introduces boosters on x86-64. These patches can be applied to 2.6.24-rc4-mm1.

The purpose of this patchset is unifying kprobes_[32|64].[c|h] to kprobes.[c|h]
for simplifying code maintenance.

I know these patches are conflicting with Harvey's patch.
We need to solve that.

kprobe-booster and kretprobe-booster were explained in:
http://marc.info/?l=linux-kernel&m=113862526017068&w=2

Currently, these patches do unification as following order.
1. Clean up and fix bugs in kprobes[1/6, 2/6].
2. Introduce kprobe-booster and kretprobe-booster for x86-64[3/6, 4/6].
   (x86-32 kprobes already has same functionalities)
3. Prepare unification[5/6].
4. Unify kprobes code[6/6].

If you have any comment, please let me know.


Best Regards,

-- 
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America) Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com, masami.hiramatsu.pt@hitachi.com

Re: [-mm][PATCH 0/6] (yet another) kprobes x86 code unification and boosters

[Ingo Molnar]

* Masami Hiramatsu <mhiramat@redhat.com> wrote:

> Hello all,
> 
> I developed a series of patches which unifies kprobes code on x86 and 
> introduces boosters on x86-64. These patches can be applied to 
> 2.6.24-rc4-mm1.
> 
> The purpose of this patchset is unifying kprobes_[32|64].[c|h] to 
> kprobes.[c|h] for simplifying code maintenance.
> 
> I know these patches are conflicting with Harvey's patch. We need to 
> solve that.

your series fixes the 64-bit crash that i was seeing, so i've picked it 
up. Please work it out with Harvey which cleanups of him are not 
included yet.

	Ingo

Re: [-mm][PATCH 0/6] (yet another) kprobes x86 code unification and boosters

[Harvey Harrison]

On Mon, 2007-12-17 at 17:33 +0100, Ingo Molnar wrote:
> * Masami Hiramatsu <mhiramat@redhat.com> wrote:
> 
> > Hello all,
> > 
> > I developed a series of patches which unifies kprobes code on x86 and 
> > introduces boosters on x86-64. These patches can be applied to 
> > 2.6.24-rc4-mm1.
> > 
> > The purpose of this patchset is unifying kprobes_[32|64].[c|h] to 
> > kprobes.[c|h] for simplifying code maintenance.
> > 
> > I know these patches are conflicting with Harvey's patch. We need to 
> > solve that.
> 
> your series fixes the 64-bit crash that i was seeing, so i've picked it 
> up. Please work it out with Harvey which cleanups of him are not 
> included yet.
> 

I thought the jprobe_saved_sp looked fishy on X86_64 in my unified set
but I went bug-for-bug compatible I guess :-)  If you look at my cleaned
up set this was one of the main source of remaining ifdefs.

Harvey

Re: [-mm][PATCH 0/6] (yet another) kprobes x86 code unification and boosters

[Masami Hiramatsu]

Hi,

Ingo Molnar wrote:
> * Masami Hiramatsu <mhiramat@redhat.com> wrote:
> 
>> Hello all,
>>
>> I developed a series of patches which unifies kprobes code on x86 and 
>> introduces boosters on x86-64. These patches can be applied to 
>> 2.6.24-rc4-mm1.
>>
>> The purpose of this patchset is unifying kprobes_[32|64].[c|h] to 
>> kprobes.[c|h] for simplifying code maintenance.
>>
>> I know these patches are conflicting with Harvey's patch. We need to 
>> solve that.
> 
> your series fixes the 64-bit crash that i was seeing, so i've picked it 
> up. Please work it out with Harvey which cleanups of him are not 
> included yet.

Absolutely sure.
I compared my patch and Harvey's.
These directions are almost same.

Harvey, I have found some differences and I'd like to fix that with you.
I think following comments and style cleanups in your patch are good to me.

> @@ -156,7 +157,7 @@ twobyte_has_modrm[256 / sizeof(unsigned long)] = {
>  #undef RF
>  
>  /* insert a jmp code */
> -static __always_inline void set_jmp_op(void *from, void *to)
> +static inline void set_jmp_op(void *from, void *to)
>  {
>  	struct __arch_jmp_op {
>  		char op;
> @@ -170,7 +171,7 @@ static __always_inline void set_jmp_op(void *from, void *to)
>  /*
>   * returns non-zero if opcodes can be boosted.
>   */
> -static __always_inline int can_boost(kprobe_opcode_t *opcodes)
> +static inline int can_boost(kprobe_opcode_t *opcodes)
>  {
>  	kprobe_opcode_t opcode;
>  	kprobe_opcode_t *orig_opcodes = opcodes;


> @@ -734,7 +740,7 @@ static int __kprobes post_kprobe_handler(struct pt_regs *regs)
>  	regs->flags |= kcb->kprobe_saved_flags;
>  	trace_hardirqs_fixup_flags(regs->flags);
>  
> -	/*Restore back the original saved kprobes variables and continue. */
> +	/* Restore the original saved kprobes variables and continue. */
>  	if (kcb->kprobe_status == KPROBE_REENTER) {
>  		restore_previous_kprobe(kcb);
>  		goto out;
> @@ -860,7 +866,7 @@ int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
>  	addr = (unsigned long)(kcb->jprobe_saved_sp);
>  
>  	/*
> -	 * TBD: As Linus pointed out, gcc assumes that the callee
> +	 * As Linus pointed out, gcc assumes that the callee
>  	 * owns the argument space and could overwrite it, e.g.
>  	 * tailcall optimization. So, to be absolutely safe
>  	 * we also save and restore enough stack bytes to cover

And also, if you can unify x86/mm/extable_*.c and introduce fixup_exception() to 64-bit,
it is very helpful to remove ifdefs from kprobe_fault_handler().

Thank you,

-- 
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America) Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com, masami.hiramatsu.pt@hitachi.com