Re: [RFC][PATCH] x86, fpu: trace points

[Ingo Molnar <mingo@kernel.org>]

* Dave Hansen <dave@sr71.net> wrote:

> 
> From: Dave Hansen <dave.hansen@linux.intel.com>
> 
> I've been carrying this patch around for a bit and it's helped me
> solve at least a couple FPU-related issues.  It's a _bit_ of a
> hack and probably too indiscriminate for mainline.
> 
> But, I'd be really interested to get something similar in to
> mainline.
> 
> How do folks feel about this as it stands?  Could we do something
> more structured?
> 
> Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
> ---
> 
>  b/arch/x86/include/asm/fpu/internal.h |    5 +
>  b/arch/x86/include/asm/trace/fpu.h    |  115 ++++++++++++++++++++++++++++++++++
>  b/arch/x86/kernel/fpu/core.c          |   18 +++++
>  b/arch/x86/kernel/fpu/signal.c        |    2 
>  b/arch/x86/kvm/x86.c                  |    6 -
>  5 files changed, 143 insertions(+), 3 deletions(-)

It certainly looks good to me!

Which bit do you consider a hack? It's a pretty straightforward set of 
tracepoints.

> +DECLARE_EVENT_CLASS(fpu_state_event,
> +
> +	TP_PROTO(struct fpu *fpu),
> +	TP_ARGS(fpu),
> +
> +	TP_STRUCT__entry(
> +		__field(struct fpu *, fpu)
> +		__field(bool, fpregs_active)
> +		__field(bool, fpstate_active)
> +		__field(int, counter)
> +		__field(u64, xfeatures)
> +		__field(u64, xcomp_bv)
> +	),

The only detail I'd change is that I'd make the tracepoint names explicitly 
x86-ish, i.e. I'd rename the event class to 'x86_fpu'. (No need to put 
'state_event' into the class name, all tracepoints are events and we obviously 
trace some sort of state.)

Likewise I'd name the tracepoints themselves along the 'x86_fpu_*' pattern.

> +	TP_fast_assign(
> +		__entry->fpu = fpu;
> +		__entry->fpregs_active = fpu->fpregs_active;
> +		__entry->fpstate_active = fpu->fpstate_active;
> +		__entry->counter = fpu->counter;

Nit: my pet peeve about vertically aligning initializations, like you did it just 
a bit further down:

> +		if (cpu_has_xsave) {
> +			__entry->xfeatures = fpu->state.xsave.header.xfeatures;
> +			__entry->xcomp_bv  = fpu->state.xsave.header.xcomp_bv;
> +		}
> +	),
> +	TP_printk("fpu: %p fpregs_active: %d fpstate_active: %d counter: %d xfeatures: %llx xcomp_bv: %llx",

... and here I'd make the trace message "x86/fpu: ", to make it obvious and easy 
to parse. The events are very x86 specific in any case.

Thanks,

	Ingo