[raw v1 0/4] Replace __this_cpu ops with raw_cpu_ops where necessary

[raw v1 0/4] Replace __this_cpu ops with raw_cpu_ops where necessary

[Christoph Lameter]

The kernel can detect __this_cpu ops in preemptible contexts after
the preemption check patchset that I posted last week.

Here is a selection of patches to fix false positives caused by
the additional checks. So far I have found no bug. Amazing.
Looks like there was some prior work done to shake these things out?

I ran an upstream kernel using Ubuntu 13.04 + the preempt checking +
the patches below on my desktop and with the following patches no
__this_cpu ops triggered messages. The kernel had the default Ubuntu
desktop configuration (+ CONFIG_PREEMPT + CONFIG_DEBUG_THIS_CPU_OPERATIONS).

Re: [raw v1 0/4] Replace __this_cpu ops with raw_cpu_ops where necessary

[Ingo Molnar]

* Christoph Lameter <cl@linux.com> wrote:

> The kernel can detect __this_cpu ops in preemptible contexts after
> the preemption check patchset that I posted last week.
> 
> Here is a selection of patches to fix false positives caused by
> the additional checks. So far I have found no bug. Amazing.
> Looks like there was some prior work done to shake these things out?
> 
> I ran an upstream kernel using Ubuntu 13.04 + the preempt checking +
> the patches below on my desktop and with the following patches no
> __this_cpu ops triggered messages. The kernel had the default Ubuntu
> desktop configuration (+ CONFIG_PREEMPT + CONFIG_DEBUG_THIS_CPU_OPERATIONS).

Here are the current list of technical problems that still plague your 
submission:

 - see the technical questions I raised about the snmp patch

 - broken threading resulting in patches arriving out of order. (You
   ignored peterz's helpful suggestion to upgrade Quilt which would allow
   you to fix this.)

 - broken subject lines - no 'PATCH' for patches

 - inconsistent patch titles: broken, inconsistent subsystem tags, etc.

 - incomplete changelogs: you fail to quote the false positive that led 
   you to a change. In cases this results in a totally context-free
   changelog which loses information the moment it's committed upstream.

In particular the last 4 items were mentioned to you for your last 
submission - which you failed to fix. You should wait with your next 
submission until you have enough time to fix *all* problems that were 
reported to you, not just some.

All in one, still a sloppy, incomplete submission.

Thanks,

	Ingo

Re: [raw v1 0/4] Replace __this_cpu ops with raw_cpu_ops where necessary

[Ingo Molnar]

* Christoph Lameter <cl@linux.com> wrote:

> The kernel can detect __this_cpu ops in preemptible contexts after the 
> preemption check patchset that I posted last week.
> 
> Here is a selection of patches to fix false positives caused by the 
> additional checks. So far I have found no bug. Amazing. Looks like there 
> was some prior work done to shake these things out?

You can probably thank -rt/PREEMPT_RT for such efforts - I think Thomas in 
particular has hit a couple of genuine bugs with __this_cpu APIs the hard 
way and then fixed them.

This series, once finished, will allow us to detect such problems cheaper 
and earlier.

Thanks,

	Ingo