Hi Viresh,
(1) Recently I have done a lot of testing for the new RCU functionality (new RCU FNOHZ implementation, updated RCU no-callback implementation) which is scheduled for 3.11/3.12 mainline. In my tests it helped to keep the A15's quiet and showed power savings for some workloads.
During last Linaro Connect people started to ask me if we can bring these features in already earlier.
Would it be possible for you to merge in 'idlenocb.2013.03.18b' (based on v3.9-rc2) (or newer) from 'git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git' into your big.LITTLE MP tree? I talked to Paul McKenney about this and he would like to have this additional test coverage.
I have performed the following tests: System comes up and pm-qa (cpufreq, cpuidle, cpuhotplug) on nano and FTS (mpbasic, mpcore, mpextended, mploadbalance, mpregression, pmump) on Android FS look good.
(2) On top of the merge, we must assure that the TC2 default config file gets changed:
Change the config file: * Set CONFIG_RCU_FAST_NO_HZ=y * Set CONFIG_RCU_NOCB_CPU=y * Set CONFIG_RCU_NOCB_CPU_ALL=y
diff -Naur config_orig .config --- config_orig 2013-03-13 14:15:10.574248125 +0000 +++ .config 2013-03-13 14:15:34.334248346 +0000 @@ -93,9 +93,12 @@ CONFIG_RCU_FANOUT=32 CONFIG_RCU_FANOUT_LEAF=16 # CONFIG_RCU_FANOUT_EXACT is not set -# CONFIG_RCU_FAST_NO_HZ is not set +CONFIG_RCU_FAST_NO_HZ=y # CONFIG_TREE_RCU_TRACE is not set -# CONFIG_RCU_NOCB_CPU is not set +CONFIG_RCU_NOCB_CPU=y +# CONFIG_RCU_NOCB_CPU_NONE is not set +# CONFIG_RCU_NOCB_CPU_ZERO is not set +CONFIG_RCU_NOCB_CPU_ALL=y CONFIG_IKCONFIG=y CONFIG_IKCONFIG_PROC=y CONFIG_LOG_BUF_SHIFT=16
Cheers,
-- Dietmar
-- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
On 26 March 2013 22:52, Dietmar Eggemann dietmar.eggemann@arm.com wrote:
Would it be possible for you to merge in 'idlenocb.2013.03.18b' (based on v3.9-rc2) (or newer) from 'git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git' into your big.LITTLE MP tree? I talked to Paul McKenney about this and he would like to have this additional test coverage.
13.03 release is already over and this would be available after that.
(2) On top of the merge, we must assure that the TC2 default config file gets changed:
Change the config file:
- Set CONFIG_RCU_FAST_NO_HZ=y
- Set CONFIG_RCU_NOCB_CPU=y
- Set CONFIG_RCU_NOCB_CPU_ALL=y
@Mathieu/Nicolas: you see any issues with these? I know we have been working on some issues with/without them for some time.
On 13-03-26 11:27 AM, Viresh Kumar wrote:
On 26 March 2013 22:52, Dietmar Eggemann dietmar.eggemann@arm.com wrote:
Would it be possible for you to merge in 'idlenocb.2013.03.18b' (based on v3.9-rc2) (or newer) from 'git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git' into your big.LITTLE MP tree? I talked to Paul McKenney about this and he would like to have this additional test coverage.
13.03 release is already over and this would be available after that.
(2) On top of the merge, we must assure that the TC2 default config file gets changed:
Change the config file:
- Set CONFIG_RCU_FAST_NO_HZ=y
- Set CONFIG_RCU_NOCB_CPU=y
- Set CONFIG_RCU_NOCB_CPU_ALL=y
@Mathieu/Nicolas: you see any issues with these? I know we have been working on some issues with/without them for some time.
The CONFIG_RCU_FAST_NO_HZ option is known to increase the likely hood of hitting a tight race condition in the cpu hotplug and smpboot thread area. The condition doesn't disappear when the option is off but it decreases the failure rate.
Mathieu.
On 13-03-26 12:31 PM, Mathieu Poirier wrote:
On 13-03-26 11:27 AM, Viresh Kumar wrote:
On 26 March 2013 22:52, Dietmar Eggemann dietmar.eggemann@arm.com wrote:
Would it be possible for you to merge in 'idlenocb.2013.03.18b' (based on v3.9-rc2) (or newer) from 'git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git' into your big.LITTLE MP tree? I talked to Paul McKenney about this and he would like to have this additional test coverage.
13.03 release is already over and this would be available after that.
(2) On top of the merge, we must assure that the TC2 default config file gets changed:
Change the config file:
- Set CONFIG_RCU_FAST_NO_HZ=y
- Set CONFIG_RCU_NOCB_CPU=y
- Set CONFIG_RCU_NOCB_CPU_ALL=y
@Mathieu/Nicolas: you see any issues with these? I know we have been working on some issues with/without them for some time.
The CONFIG_RCU_FAST_NO_HZ option is known to increase the likely hood of hitting a tight race condition in the cpu hotplug and smpboot thread area. The condition doesn't disappear when the option is off but it decreases the failure rate.
Mathieu.
What we could do is take it in 'next' and see how it behaves. After all the worst that can happen is to still observe the issue...
On 26/03/13 18:34, Mathieu Poirier wrote:
On 13-03-26 12:31 PM, Mathieu Poirier wrote:
On 13-03-26 11:27 AM, Viresh Kumar wrote:
On 26 March 2013 22:52, Dietmar Eggemann dietmar.eggemann@arm.com wrote:
Would it be possible for you to merge in 'idlenocb.2013.03.18b' (based on v3.9-rc2) (or newer) from 'git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git' into your big.LITTLE MP tree? I talked to Paul McKenney about this and he would like to have this additional test coverage.
13.03 release is already over and this would be available after that.
(2) On top of the merge, we must assure that the TC2 default config file gets changed:
Change the config file:
- Set CONFIG_RCU_FAST_NO_HZ=y
- Set CONFIG_RCU_NOCB_CPU=y
- Set CONFIG_RCU_NOCB_CPU_ALL=y
@Mathieu/Nicolas: you see any issues with these? I know we have been working on some issues with/without them for some time.
The CONFIG_RCU_FAST_NO_HZ option is known to increase the likely hood of hitting a tight race condition in the cpu hotplug and smpboot thread area. The condition doesn't disappear when the option is off but it decreases the failure rate.
Mathieu.
Hi Mathieu,
I ran an MP configuration based on the latest ll-20130325.0 branch from Linaro plus linux_paulmck_rcu/idlenocb.2013.03.26b with the RCU kernel configuration mentioned above.
I used pm-qa cpuhotplug_08.sh test against this system
# for i in $(seq 1000); do ./cpuhotplug_08.sh; done
and haven't encountered any issues. BTW, I'm not doing any IKS related tests.
-- Dietmar
What we could do is take it in 'next' and see how it behaves. After all the worst that can happen is to still observe the issue...
-- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
On 13-04-02 08:45 AM, Dietmar Eggemann wrote:
On 26/03/13 18:34, Mathieu Poirier wrote:
On 13-03-26 12:31 PM, Mathieu Poirier wrote:
On 13-03-26 11:27 AM, Viresh Kumar wrote:
On 26 March 2013 22:52, Dietmar Eggemann dietmar.eggemann@arm.com wrote:
Would it be possible for you to merge in 'idlenocb.2013.03.18b' (based on v3.9-rc2) (or newer) from 'git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git' into your big.LITTLE MP tree? I talked to Paul McKenney about this and he would like to have this additional test coverage.
13.03 release is already over and this would be available after that.
(2) On top of the merge, we must assure that the TC2 default config file gets changed:
Change the config file:
- Set CONFIG_RCU_FAST_NO_HZ=y
- Set CONFIG_RCU_NOCB_CPU=y
- Set CONFIG_RCU_NOCB_CPU_ALL=y
@Mathieu/Nicolas: you see any issues with these? I know we have been working on some issues with/without them for some time.
The CONFIG_RCU_FAST_NO_HZ option is known to increase the likely hood of hitting a tight race condition in the cpu hotplug and smpboot thread area. The condition doesn't disappear when the option is off but it decreases the failure rate.
Mathieu.
Hi Mathieu,
I ran an MP configuration based on the latest ll-20130325.0 branch from Linaro plus linux_paulmck_rcu/idlenocb.2013.03.26b with the RCU kernel configuration mentioned above.
I used pm-qa cpuhotplug_08.sh test against this system
# for i in $(seq 1000); do ./cpuhotplug_08.sh; done
and haven't encountered any issues. BTW, I'm not doing any IKS related tests.
-- Dietmar
Good morning,
That is very interesting information - I'll try this on my side and will get back to you.
Thanks, Mathieu.
What we could do is take it in 'next' and see how it behaves. After all the worst that can happen is to still observe the issue...
-- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
On 13-04-02 08:45 AM, Dietmar Eggemann wrote:
On 26/03/13 18:34, Mathieu Poirier wrote:
On 13-03-26 12:31 PM, Mathieu Poirier wrote:
On 13-03-26 11:27 AM, Viresh Kumar wrote:
On 26 March 2013 22:52, Dietmar Eggemann dietmar.eggemann@arm.com wrote:
Would it be possible for you to merge in 'idlenocb.2013.03.18b' (based on v3.9-rc2) (or newer) from 'git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git' into your big.LITTLE MP tree? I talked to Paul McKenney about this and he would like to have this additional test coverage.
13.03 release is already over and this would be available after that.
(2) On top of the merge, we must assure that the TC2 default config file gets changed:
Change the config file:
- Set CONFIG_RCU_FAST_NO_HZ=y
- Set CONFIG_RCU_NOCB_CPU=y
- Set CONFIG_RCU_NOCB_CPU_ALL=y
@Mathieu/Nicolas: you see any issues with these? I know we have been working on some issues with/without them for some time.
The CONFIG_RCU_FAST_NO_HZ option is known to increase the likely hood of hitting a tight race condition in the cpu hotplug and smpboot thread area. The condition doesn't disappear when the option is off but it decreases the failure rate.
Mathieu.
Hi Mathieu,
I ran an MP configuration based on the latest ll-20130325.0 branch from Linaro plus linux_paulmck_rcu/idlenocb.2013.03.26b with the RCU kernel configuration mentioned above.
I used pm-qa cpuhotplug_08.sh test against this system
# for i in $(seq 1000); do ./cpuhotplug_08.sh; done
and haven't encountered any issues. BTW, I'm not doing any IKS related tests.
Good morning Dietmar,
I tested this over night and wasn't able to trigger the problem - that is very good news. Today I'll redo my test with ll-20130325.0 only. If I hit the condition then the fix is definitely in the 15 patches on top of idlenocb.2013.03.26.
Thanks for the tip, Mathieu.
-- Dietmar
What we could do is take it in 'next' and see how it behaves. After all the worst that can happen is to still observe the issue...
-- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
-
Dietmar Eggemann -
Mathieu Poirier -
Viresh Kumar