Messages in this thread | | | Date | Wed, 7 May 2008 20:49:21 +0200 | From | Ingo Molnar <> | Subject | Re: AIM7 40% regression with 2.6.26-rc1 |
| |
* Linus Torvalds <torvalds@linux-foundation.org> wrote:
> .. Hmm ... Time passes. Linus looks at git history. > > It does look like "cond_resched()" has not worked with the BKL since > 2005, and hasn't taken the BKL into account. Commit 5bbcfd9000: > > [PATCH] cond_resched(): fix bogus might_sleep() warning > > + if (unlikely(preempt_count())) > + return; > > which talks about the BKS, ie it only took the *semaphore* > implementation into account. Never the spinlock-with-preemption-count > one. > > Or am I blind?
hm, i think you are right.
most latency reduction was concentrated on the PREEMPT+PREEMPT_BKL case, and not getting proper cond_resched() behavior in case of !PREEMPT_BKL would certainly not be noticed by distros or users.
We made CONFIG_PREEMPT_BKL=y the default on SMP in v2.6.8, in this post-2.6.7 commit that introduced the feature:
| commit fb8f6499abc6a847109d9602b797aa6afd2d5a3d | Author: Ingo Molnar <mingo@elte.hu> | Date: Fri Jan 7 21:59:57 2005 -0800 | | [PATCH] remove the BKL by turning it into a semaphore
There was constant trouble around all these variations of preemptability and their combination with debugging helpers. (So i was rather happy to get rid of !PREEMPT_BKL - in the (apparently wrong) assumption that no tears will be shed.)
Ingo
| |