softirq: avoid softirq_work_list for SCHED_SOFTIRQ when sent remotely
in [Kernel]
|
Prev: [ANNOUNCE] BFS CPU scheduler v0.318 for 2.6.34
Next: softirq: avoid softirq_work_list for SCHED_SOFTIRQ when sent remotely
From: David Miller on 20 May 2010 04:20 From: Peter Zijlstra <peterz(a)infradead.org> Date: Thu, 20 May 2010 10:12:39 +0200 > Most odd all that, Dave, Jens, what happened to all that remote_softirq > stuff? Nobody ended up using this remote softirq infrastructure, it can be completely deleted. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Peter Zijlstra on 20 May 2010 04:20 On Mon, 2010-05-17 at 11:27 -0700, Suresh Siddha wrote: > plain text document attachment > (fix_local_trigger_remote_softirq_for_sched.patch) > There is no need to add the SCHED_SOFTIRQ work to the softirq_work_list > when sent remotely. This is because any pending work associated with > SCHED_SOFTIRQ need not be migrated to a new cpu when the current cpu is > going down etc. > > Also I am not sure how this softirq_work_list works for other softirqs. > I don't see anyone removing the list entries from the softirq_work_list > after it is handled on a remote cpu. Most odd all that, Dave, Jens, what happened to all that remote_softirq stuff? -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Peter Zijlstra on 20 May 2010 04:40 On Thu, 2010-05-20 at 10:23 +0200, Jens Axboe wrote: > On Thu, May 20 2010, David Miller wrote: > > From: Peter Zijlstra <peterz(a)infradead.org> > > Date: Thu, 20 May 2010 10:12:39 +0200 > > > > > Most odd all that, Dave, Jens, what happened to all that remote_softirq > > > stuff? > > > > Nobody ended up using this remote softirq infrastructure, it can be > > completely deleted. > > What parts of it? There has been reworks of parts of the code since I > added it, the block layer only uses __smp_call_function_single() to > trigger remote softirqs. Right, so __smp_call_function_single() is the normal remote ipi stuff from kernel/smp.c, send_remote_softirq() is the pile of code in kernel/softirq.c that uses that to tickle remote softirqs. The 'problem' seems to be that the remote softirq code (not the ipi bits) doesn't have any users and is quite complex and apparently incomplete. Suresh, would it make sense to do as Jens does and simply use __smp_call_function_single() or do you think it still makes sense to have a small wrapper and maybe share that with Jens? -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: David Miller on 20 May 2010 05:20
From: Jens Axboe <jens.axboe(a)oracle.com> Date: Thu, 20 May 2010 10:23:09 +0200 > On Thu, May 20 2010, David Miller wrote: >> From: Peter Zijlstra <peterz(a)infradead.org> >> Date: Thu, 20 May 2010 10:12:39 +0200 >> >> > Most odd all that, Dave, Jens, what happened to all that remote_softirq >> > stuff? >> >> Nobody ended up using this remote softirq infrastructure, it can be >> completely deleted. > > What parts of it? There has been reworks of parts of the code since I > added it, the block layer only uses __smp_call_function_single() to > trigger remote softirqs. And that's what the networking is using for RPS/RFS too. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |