Re: [PATCH v2] trace: Add x86 irq vector entry/exit tracepoints

[David Sharp <dhsharp@google.com>]

On Thu, Jul 7, 2011 at 2:57 AM, Ingo Molnar <mingo@elte.hu> wrote:
>
> * H. Peter Anvin <hpa@zytor.com> wrote:
>
>> Yes, it was much more of a generic concern.  However, it is very
>> important that people have a correct idea about what the stability
>> of something like tracepoint is -- or we'll end up in a situation
>> where we can never change the kernel because anything is suddenly
>> "user space visible."
>
> We've transitioned even ABI-assuming tracepoints in the past, so it's
> not a big issue in practice. The reason is that this is an atypical
> type of ABI: information is read-only exported, for observation
> purposes.
>
> If the kernel changes in a fundamental way that removes a tracepoint
> altogether, then there's nothing left to observe - so apps don't
> break per se.
>
> So i've yet to see a single example of the kernel 'never being able
> to change' due to a tracepoint. The worst we've seen in practice is
> the inability to change a specific tracepoint (not the surrounding
> kernel code - while preserving the information that is exposed) - so
> the worst effect was limited to tracing itself - never to the
> subsystem that it traces.
>
> Note that even in that (single known) example we were able to resolve
> the problem (which was limited to the tracing subsystem) by adding
> new tracepoints and thus phasing out the old ones.
>
> Thanks,
>
>        Ingo
>

Thanks all for your thorough review. :) It sounds like there is some
agreement now. I think this Steve is waiting for an Acked-by from an
x86 maintainer to apply this patch. Are there any further objections
or comments on the patch?

d#