Re: [PATCH v3] Introduce page fault tracepoint

["H. Peter Anvin" <hpa@zytor.com>]

On 09/09/2013 02:55 PM, Seiji Aguchi wrote:
> Change from v2
>  - Print entry->ip instead of entry->regs->ip to avoid kernel crash.
>  - Use %pf instead of 0x%lx to print address and ip.
> 
> This patch introduces page fault tracepoints to x86 architecture
> by switching IDT.
> 
> [Use case of page fault events]
> 
>   Two events, for user and kernel spaces, are introduced at the beginning of
>   page fault handler.
> 
>   - User space event
>     There is a request of page fault event for user space as below.
> 
>     http://marc.info/?l=linux-mm&m=136807959830182&w=2
>     http://marc.info/?l=linux-mm&m=136807959130175&w=2
> 

For permanence, please use links of the form:

	http://lkml.kernel.org/r/message-id

(Yes, they currently point to marc.info, but can be redirected to point
to any archive.)

>   - Kernel space event:
>     Overhead in kernel space is measurable by enabling it.
> 
> [Creating IDT]
> 
>  A way to create IDT is as below.
> 
>  - Introduce set_intr_gate_raw() to register just non-trace handler to IDT.
>    This is used at boot time which tracing is disabled.
>  - Make set_intr_gate() macro so that it can register trace handler to
>    trace IDT and non-trace handler to normal IDT.
> 

This is needlessly confusing, which is apart of why reviewing this patch
took a lot more time than it should.

Please break this patch into two: one which sets up the tracing IDT and
one to create the #PF tracepoint.  The assumption is, I am assuming,
there will be more.

	-hpa