[n9005][2014/DEC/01][TWKK] faux123/Yank555.lu kernel 002b3

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby yank555 » 16 May 2014, 07:12

Julian Jeremiah wrote:Yank,

I realized that after every reboot hardlimit_screen_off, wakeup_kick_delay and wakeup_kick_freq will always revert to its default values. Any advice how to make it stick? No luck even after changed permissions.


Everything in /sys are virtual files giving access to kernel variable, so RAM, as such they can never survive a reboot ;)

You need to make and run a script on boot which sets them.

JP.

Send from my 'proudly eFused' Note 3 (n9005) running Temasek v75 on Yank555.lu htle v1.0a-beta4 (3.4.90) kernel.
Nexus 5 (aka. Hammerhead) 32Gb
Note 3 SM-N9005 32Gb Proudly eFused
SGS3 GT-I9300 32Gb
HTC Sensation XE
HTC HD2
TF300TG 32Gb

Image
Credits FAdrums !
User avatar
yank555
-----------
 
Posts: 3520
Joined: 11 Mar 2013, 20:51
Has thanked: 924 times
Been thanked: 4608 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby Julian Jeremiah » 16 May 2014, 07:39

Ahhhhh that's quite a hassle :cry: anyway thanks a lot

Mind if I ask what's the difference between your hardlimit screen off and Imoseyon's max screen off? Logically both serves the same purpose, correct me if I'm wrong :D

yank555 wrote:
Julian Jeremiah wrote:Yank,

I realized that after every reboot hardlimit_screen_off, wakeup_kick_delay and wakeup_kick_freq will always revert to its default values. Any advice how to make it stick? No luck even after changed permissions.


Everything in /sys are virtual files giving access to kernel variable, so RAM, as such they can never survive a reboot ;)

You need to make and run a script on boot which sets them.

JP.

Send from my 'proudly eFused' Note 3 (n9005) running Temasek v75 on Yank555.lu htle v1.0a-beta4 (3.4.90) kernel.
Julian Jeremiah
 
Posts: 43
Joined: 12 May 2014, 14:57
Has thanked: 21 times
Been thanked: 6 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby yank555 » 16 May 2014, 10:06

Julian Jeremiah wrote:Mind if I ask what's the difference between your hardlimit screen off and Imoseyon's max screen off? Logically both serves the same purpose, correct me if I'm wrong :D


Mine is a driver located at cpufreq, which works for any governor used, from what I gather, Imo's is implemented in InteractivX governor, but I may be wrong, you know his kernel certainly better than I do ;)

Also, my CPU Hardlimit does more than just hardlimit screen off, it does hardlimit (screen on), wakekick and touchboost freqs.

JP.
Nexus 5 (aka. Hammerhead) 32Gb
Note 3 SM-N9005 32Gb Proudly eFused
SGS3 GT-I9300 32Gb
HTC Sensation XE
HTC HD2
TF300TG 32Gb

Image
Credits FAdrums !
User avatar
yank555
-----------
 
Posts: 3520
Joined: 11 Mar 2013, 20:51
Has thanked: 924 times
Been thanked: 4608 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby Julian Jeremiah » 16 May 2014, 10:56

Yes certainly your implementation of hardlimit does so much more :D

You're right it's only interactiveX governor which has max screen off profile.


yank555 wrote:
Julian Jeremiah wrote:Mind if I ask what's the difference between your hardlimit screen off and Imoseyon's max screen off? Logically both serves the same purpose, correct me if I'm wrong :D


Mine is a driver located at cpufreq, which works for any governor used, from what I gather, Imo's is implemented in InteractivX governor, but I may be wrong, you know his kernel certainly better than I do ;)

Also, my CPU Hardlimit does more than just hardlimit screen off, it does hardlimit (screen on), wakekick and touchboost freqs.

JP.
Julian Jeremiah
 
Posts: 43
Joined: 12 May 2014, 14:57
Has thanked: 21 times
Been thanked: 6 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby yank555 » 16 May 2014, 11:08

Julian Jeremiah wrote:Yes certainly your implementation of hardlimit does so much more :D

You're right it's only interactiveX governor which has max screen off profile.


I wouldn't say it does more, it's just a different thing, it sits deeper, below the governor, while screen off max in a governor is one more tunable for that governor ;)

If you want, governors run on top of cpufreq, so no matter what a governor would ask, cpufreq can or cannot allow it, essentially a governor is an agent the cpufreq driver has to let it know how to scale cpu frequency, like some kind of consultant who tells it when more or less power seems to be needed. The actual scaling is done by cpufreq based on the governor's suggestion, and I regulate what the consultant suggest ;)

JP.
Nexus 5 (aka. Hammerhead) 32Gb
Note 3 SM-N9005 32Gb Proudly eFused
SGS3 GT-I9300 32Gb
HTC Sensation XE
HTC HD2
TF300TG 32Gb

Image
Credits FAdrums !
User avatar
yank555
-----------
 
Posts: 3520
Joined: 11 Mar 2013, 20:51
Has thanked: 924 times
Been thanked: 4608 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby Julian Jeremiah » 16 May 2014, 13:48

Thanks Master! ;)

Btw did you ever try "screen on time" on each governor after you've tweaked its parameters? Sorry if Ive phrased my question wrongly :D

What I'm trying to ask is how long can the % of battery lasts (says 90%) before it descends to 98% while browsing on web. I've tried almost all kernels which is available, finding the one best fits my needs, using different governors to test, tweaking its parameters, etc and on the worst scenario it could lasts is 1% : 2-3 mins (ratio); at the best is 1% : 5 1/2 mins (estimated). I'm unable to squeeze more juice out of it to lasts longer :lol:

yank555 wrote:
Julian Jeremiah wrote:Yes certainly your implementation of hardlimit does so much more :D

You're right it's only interactiveX governor which has max screen off profile.


I wouldn't say it does more, it's just a different thing, it sits deeper, below the governor, while screen off max in a governor is one more tunable for that governor ;)

If you want, governors run on top of cpufreq, so no matter what a governor would ask, cpufreq can or cannot allow it, essentially a governor is an agent the cpufreq driver has to let it know how to scale cpu frequency, like some kind of consultant who tells it when more or less power seems to be needed. The actual scaling is done by cpufreq based on the governor's suggestion, and I regulate what the consultant suggest ;)

JP.
Julian Jeremiah
 
Posts: 43
Joined: 12 May 2014, 14:57
Has thanked: 21 times
Been thanked: 6 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby yank555 » 16 May 2014, 15:15

Julian Jeremiah wrote:Thanks Master! ;)

Btw did you ever try "screen on time" on each governor after you've tweaked its parameters? Sorry if Ive phrased my question wrongly :D

What I'm trying to ask is how long can the % of battery lasts (says 90%) before it descends to 98% while browsing on web. I've tried almost all kernels which is available, finding the one best fits my needs, using different governors to test, tweaking its parameters, etc and on the worst scenario it could lasts is 1% : 2-3 mins (ratio); at the best is 1% : 5 1/2 mins (estimated). I'm unable to squeeze more juice out of it to lasts longer :lol:


Hard to analyse it that way because of a lot of things :

1) Battery % is an estimate calculated based on battery voltage, that is not precise and not linear, meaning you will lose 1% sooner at high battery level compared to lower battery level, so you need to calculate the average over a longer period, like 1 charge cycle.

2) Governors have an impact on CPU freq. which in turn has an impact on voltage applied which has an impact on the current drawn (that's what you want to analyse).

3) Hotplugging will on/offline cores, so basically the current drawn will be 1-4x higher, 1 core at 1GHz draws 4 times less than 4 cores at 1GHz, so hotplug is even more important than frequency.

4) With screen on, all of the above don't matter that much, because they represent peanuts compared to the display panel ! We have an Amoled panel, which means the more pixels are lit and the brighter they are lit the more current they will draw and that will be what determines actual battery life. Keep it dark and you will go further.

----

I tend to chose a governor I find ok as base, tweak it towards battery until I feel it's 'presence' (tiny lags here and there) and then bump it one nod up again, and stick with that.

Where all this really makes a difference, that's during screen off, as then the only thing really left drawing battery are the CPUs and the radio (gsm, 3g, wifi, bt...), since the 'big guy' (display) is turned off and not drawing any current at all.

Not much you can do for the radio (except turning it partly off), the CPU handling is what will make the biggest difference, that's why there's focus on that in kernels.

On the Note 3 influencing the hotplug driver, as in getting rid of mpdecision, is a way to get proper control over how many CPU should be online, and when to take them on/offline. With mpdecision, those choices have been made for us by Qualcomm.

So at some point that will be something I'll be looking into...

JP.

Send from my 'proudly eFused' Note 3 (n9005) running Temasek v75 on Yank555.lu htle v1.0a-beta4 (3.4.90) kernel.

Edit :

Also, with Samsung's default touchboost freq., governors don't really matter anymore, as the first touch will bump the freq. to 1.2GHz and after several repetitve touches 1.7GHz, so the governor has but 1.7 - 2.3 to play with (without OC that is) when you actively use your device.
Nexus 5 (aka. Hammerhead) 32Gb
Note 3 SM-N9005 32Gb Proudly eFused
SGS3 GT-I9300 32Gb
HTC Sensation XE
HTC HD2
TF300TG 32Gb

Image
Credits FAdrums !
User avatar
yank555
-----------
 
Posts: 3520
Joined: 11 Mar 2013, 20:51
Has thanked: 924 times
Been thanked: 4608 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby Julian Jeremiah » 16 May 2014, 19:02

Is it the reason why many of them are now switching over to the latest msm_hotplug? I've heard it yields better result.

yank555 wrote:Hard to analyse it that way because of a lot of things :

1) Battery % is an estimate calculated based on battery voltage, that is not precise and not linear, meaning you will lose 1% sooner at high battery level compared to lower battery level, so you need to calculate the average over a longer period, like 1 charge cycle.

2) Governors have an impact on CPU freq. which in turn has an impact on voltage applied which has an impact on the current drawn (that's what you want to analyse).

3) Hotplugging will on/offline cores, so basically the current drawn will be 1-4x higher, 1 core at 1GHz draws 4 times less than 4 cores at 1GHz, so hotplug is even more important than frequency.

4) With screen on, all of the above don't matter that much, because they represent peanuts compared to the display panel ! We have an Amoled panel, which means the more pixels are lit and the brighter they are lit the more current they will draw and that will be what determines actual battery life. Keep it dark and you will go further.

----

I tend to chose a governor I find ok as base, tweak it towards battery until I feel it's 'presence' (tiny lags here and there) and then bump it one nod up again, and stick with that.

Where all this really makes a difference, that's during screen off, as then the only thing really left drawing battery are the CPUs and the radio (gsm, 3g, wifi, bt...), since the 'big guy' (display) is turned off and not drawing any current at all.

Not much you can do for the radio (except turning it partly off), the CPU handling is what will make the biggest difference, that's why there's focus on that in kernels.

On the Note 3 influencing the hotplug driver, as in getting rid of mpdecision, is a way to get proper control over how many CPU should be online, and when to take them on/offline. With mpdecision, those choices have been made for us by Qualcomm.

So at some point that will be something I'll be looking into...

JP.

Send from my 'proudly eFused' Note 3 (n9005) running Temasek v75 on Yank555.lu htle v1.0a-beta4 (3.4.90) kernel.

Edit :

Also, with Samsung's default touchboost freq., governors don't really matter anymore, as the first touch will bump the freq. to 1.2GHz and after several repetitve touches 1.7GHz, so the governor has but 1.7 - 2.3 to play with (without OC that is) when you actively use your device.
Julian Jeremiah
 
Posts: 43
Joined: 12 May 2014, 14:57
Has thanked: 21 times
Been thanked: 6 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby yank555 » 16 May 2014, 19:08

Julian Jeremiah wrote:Is it the reason why many of them are now switching over to the latest msm_hotplug? I've heard it yields better result.


Whether the results are better or not, it will allow us to work on it, whereas we have no sources for mpdecision.

So it's one solution, intelliplug would be an alternative to it. I'd need to look at msm_hotplug code to see if it will fit my expectations...

I'll look into that, but I need time, and the next weeks will be loaded again, so not now.

And mpdecision isn't doing that bad a job for me to rush me into his ;)

JP.

Send from my 'proudly eFused' Note 3 (n9005) running Temasek v75 on Yank555.lu htle v1.0a-beta4 (3.4.90) kernel.
Nexus 5 (aka. Hammerhead) 32Gb
Note 3 SM-N9005 32Gb Proudly eFused
SGS3 GT-I9300 32Gb
HTC Sensation XE
HTC HD2
TF300TG 32Gb

Image
Credits FAdrums !
User avatar
yank555
-----------
 
Posts: 3520
Joined: 11 Mar 2013, 20:51
Has thanked: 924 times
Been thanked: 4608 times

Re: [n9005][2014/MAY/13][TWKK] faux123/Yank555.lu kernel 001

Postby Julian Jeremiah » 16 May 2014, 19:22

Thanks.. We will all be waiting for you to bring this kernel to the next level of supremacy :D

yank555 wrote:
Julian Jeremiah wrote:Is it the reason why many of them are now switching over to the latest msm_hotplug? I've heard it yields better result.


Whether the results are better or not, it will allow us to work on it, whereas we have no sources for mpdecision.

So it's one solution, intelliplug would be an alternative to it. I'd need to look at msm_hotplug code to see if it will fit my expectations...

I'll look into that, but I need time, and the next weeks will be loaded again, so not now.

And mpdecision isn't doing that bad a job for me to rush me into his ;)

JP.

Send from my 'proudly eFused' Note 3 (n9005) running Temasek v75 on Yank555.lu htle v1.0a-beta4 (3.4.90) kernel.
Julian Jeremiah
 
Posts: 43
Joined: 12 May 2014, 14:57
Has thanked: 21 times
Been thanked: 6 times

PreviousNext

Return to Kernels

Who is online

Users browsing this forum: No registered users and 12 guests

cron