linux-kernel - Re: 'global' rq->clock

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Sat, 3 May 2008 12:10:16 +0200
From:	Ingo Molnar <mingo@...e.hu>
To:	David Miller <davem@...emloft.net>
Cc:	a.p.zijlstra@...llo.nl, efault@....de, elendil@...net.nl,
	parag.warudkar@...il.com, linux-kernel@...r.kernel.org,
	guichaz@...oo.fr, andi@...stfloor.org
Subject: Re: 'global' rq->clock

* David Miller <davem@...emloft.net> wrote:

> > there's already such a mechanism in sched-devel.git (and has been 
> > there for a week or two): an architecture can set time_sync_thresh 
> > to -1LL during early bootup and essentially disable all the 
> > synchronization logic.
> 
> Does it remove all of the code too? :-)
> 
> Please give us a config boolean.  The only platform for which a 
> run-time knob is even necessary is x86.

yeah, will try something like that too.

the thing is, core kernel folks have to resist such pressures 
_somewhat_.

Architecture maintainers will put pressure on us from two directions: if 
a particular generic feature only helps _another_ architecture, they 
find it a nuisance and want it to be gone as much as possible.

If a particular feature helps them, they want it supported and 
default-enabled as much as possible. In fact, complaints come if a 
generic-looking feature shows up in one architecture only. (recent 
example: inlining optimizations ;-)

And you are totally right about sched_clock() being dead on accurate an 
globally synchronous on sparc64 - and you are right to find _any_ issue 
about it a nuisance. I totally envy you that sparc64's sched_clock() is 
so simple - it should have been like that on x86 years ago, if hw 
designers werent so impotent about it.

( although please note that the growing generalization that goes on
  _did_ find a subtle nohz problem on sparc64 early in the merge window,
  so it's not like these changes are totally useless to you. )

but it all goes in the other direction as well: many folks find 
endianness problems a nuisance on x86, many folks find TLB and explicit 
cache coherence complications a nuisance on x86 as well. The bus-to-phys 
complication which is an identity on x86. Etc. etc.

But the core kernel is a conscious and intelligent union of all 
architecture's needs, and often we maintain complications even if they 
make no sense on the most popular platform. I think it makes strategic 
sense because it keep the kernel truly generic and truly clean. It also 
keeps architectures honest: even today the x86 architecture is still not 
as clean as sparc64 for example.

so please be patient, we are working on it :)

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/