On 6 December 2012 15:36, Viresh Kumar viresh.kumar@linaro.org wrote:
On 6 December 2012 19:36, Amit Kucheria amit.kucheria@linaro.org wrote:
Even if that is the case, I'm afraid I don't quite like the way this was done. IMHO, you shouldn't just revert bits of another author's patches that you don't agree with.
If there are issues regarding the the patches from Vincent, I'd do the following things in order of priority:
- Prove to him that the race exists, preferably with a reproducible test
case 2. Give him a chance to convince you otherwise 3. Share test results that show bad things happen as a result of some code 4. Ask _him_ to separate out that bit from the original patch so you can only pick the bits you like
I haven't seen this happen. All I've seen is one side claim it can happen and the other claim that it can't. *shrug*
Viresh, as an experienced maintainer, I hope you see the value of this approach rather than just pull in the tree.
I realise we're all under pressure here. So let's take a deep breath, step back and do it the right way.
First of all i must admit, i haven't followed the discussion closely, as this part of kernel is still rocket science for me :)
Secondly, what you said is correct Amit. But, i must say there has been a long time since the last time release happened and all this must have been sorted out in that time both from Linaro and ARM side. And i didn't saw any effort on that. Only when i came back to sort out issues in my tree, this issue is still highlighted.
Now, getting so close to release and making a big change, that will eventually affect the core part we are working on is not a great idea.
IMHO, we should keep sched-pack-small-tasks-v1-fixed instead of sched-pack-small-tasks-v1-arm - packing small task is disable with sched-pack-small-tasks-v1-arm. - sched: secure access to other CPU statistics ensure value coherency. This patch solves perhaps not all use cases and I have to study this point but it solves most of racing accesses for sure.We will just enlarge the breach by removing it
But we still have some time for a meaningful discussion to happen and one party to agree. Both can't be correct. I need to send the pull request by Monday and so whatever is required to be done, must be done by tomorrow evening.
-- viresh