Re: [PATCH v3] trace: Add x86 irq vector entry/exit tracepoints

From: Frederic Weisbecker <fweisbec@gmail.com>
To: Vaibhav Nagarnaik <vnagarnaik@google.com>
Cc: David Sharp <dhsharp@google.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	Michael Rubin <mrubin@google.com>,
	linux-kernel@vger.kernel.org, x86@kernel.org,
	Jiaying Zhang <jiayingz@google.com>
Subject: Re: [PATCH v3] trace: Add x86 irq vector entry/exit tracepoints
Date: Wed, 13 Jul 2011 16:11:34 +0200	[thread overview]
Message-ID: <20110713141129.GG9201@somewhere> (raw)
In-Reply-To: <CAL26m8LsEwEEE1e4D9dWn2Gm5_cvtNXMRod4d2hsD=qHGG1Nig@mail.gmail.com>

On Tue, Jul 12, 2011 at 03:08:48PM -0700, Vaibhav Nagarnaik wrote:
> Breaking this patch up in different small ones makes sense. Can you
> comment on this proposal for the following trace events?
> 
> For tracepoints in generic IRQ handlers:
> 1. trace_timer_vector - takes an enum for BROADCAST, HRTIMER, ONESHOT,
>    PERIODIC and NOHZ.

Have you read my previous email entirely? 

Where I explained we shouldn't use that vector naming outside of low
level archs, and that those enums break tracing granularity?

> 2. trace_irq_work_vector - for IRQ_WORK_VECTOR

trace_irq_work_run() should be enough. And people can add
trace_irq_work_queue() if they need to one day.

> 3. trace_reschedule_vector - for RESCHEDULE_IPI vector

trace_reschedule_interrupt()

> 4. trace_call_function_vector - takes an enum for CALL_FUNCTION and
>    CALL_FUNCTION_SINGLE

Two seperate tracepoints should be better.

> Another trace event for arch-specific IRQ vectors which don't have
> generic event handlers:
> 5. trace_platform_irq_vector - takes an enum, which is defined in
>    asm/irq.h for each platform. This is traced in arch-specific files
>    only.

Yep. For this perhaps an enum can make sense.

Thanks.