Re: Should the kernel perf interface be available on autobuilders?

[Petter Reinholdtsen <pere@hungry.com>]

[Lluís Vilanova]
> AFAIK, a paranoid level of 2 for perf is enough to not make coz's
> tests break (level 3 or above seems to effectively disable the perf
> interface). Also, it seems that build machines for all other
> architectures have a perf paranoid level of 2 or lower.

Yes.  And I guess the fundamental question here is this: Should a
profiler package in Debian be able to verify its own functionallity
during build (ie run its test suite), even if it depend on the linux
kernel perf interface for its operation.

If I understand Ben correctly, he recommend that the answer is no, as
the perf interface is so insecure that it should only be enabled on a
developers machine when the developer find the risk acceptable.

Of course I would prefer the answer to be yes, because I want to keep
broken packages out of the archive and running the test suite during
built is a very good way to do that.

I do not know the risk involved, and trust the opinion of Ben on the
risk involved.  But is level 2 just as dangerious (or perhaps that risk
is acceptable), or is level 3 required?  If the perf interface at any
level below 3 is too dangerous, perhaps it should be disabled on the
current autobuilders too?

-- 
Happy hacking
Petter Reinholdtsen