Linux 4.6 Will Fix A Bug Where Some Laptops Are Always Throttled With Bad Performance

[duby229]

Well, I've seen laptops reporting their critical trip points of 105C or so, that's how vendors are filling ACPI tables. Which is waaaaaaay to much. Even if CPU can withstand it (some CPUs are rated to 90C only!), it warrants extremely hot device, close to plastic parts melting point. I've also seem some ARMs reporting similar trip points. Needless to say, board goes unstable way before 105C is reached by CPU and throttle activates, making it nearly useless. Some vendors clearly overrate their abilities or do not give a fuck about filling tables/defaults properly.

The point wasn't about cleaning laptop or something. The point is that thermal protection isn't what I would call protective in many cases. It supposed to kick in in such cases, keeping system operations stable and saving IC from fatal overheating. Yet it hard to admit thermal protection always achieves these goals. Half of time it just missing the point, targeting really bizarre trip points.

No I do understand exactly. But as a repairman I can't fix that, I can only fix the source of the overheating problem and that is usually caused by a clogged up heatsink. A clogged heatsink is an extremely common problem and so it's extremely common to see Intel laptops running 100+C. I do agree they should throttle at a lower temp, but that's just not how it is.

[SystemCrasher]

To be fair, the critical trip point is never supposed to be reached. When the system hits that critical point it is supposed to shut down. Immediately. With the bare minimum of doing a filesystem sync.

Well, I've just took random laptop to re-check what's going on. It has got 100000 as "passive" trip point and 103000 as "critical". That is it, 100C and 103C. I would not dare to see what would happen with this thing at 100C and I guess 100C and 103C are fairly close, so I wouldn't count on "should never happen". Btw, when it comes to filesystem sync, it could be quite lengthy. On slow storage it could take several MINUTES to complete. Randomly picked ARM board has been a bit more sane, 85C for passive and 105C for critical. TBH I would prefer something more sensible, like 70-80C passive and 85-90C critical. I'm not really sure this hardware would actulally survive 100C, and not exactly sure if some plastic parts around wouldn't melt. 100C is water boiling boint. So one can boil an egg on CPU at this point. Literally.

Far before that point at about 80°C, the CPU is supposed to start clock throttling.

I do not see where this assumption listed. "Passive" trip point could be equally stupid, and particular HW may or may not have "other" abilities e.g, via SMM BIOS, etc, which is entirely missing on, say, ARM, and varies wildly on x86's, could be equally moron. I've seen some SMM BIOS things which were raising false alarms, being set up to alert at 70-75C and rather yelling into PC buzzer at something like 55C, which is perfectly okay for full system load, lol.

That 105° mark indicates something seriously wrong such as a complete cooling failure.

Sure thing, trip points are not supposed to be reached under normal course of actions, but it does not excuses weird values (as well as SMBIOS mischief).

[SystemCrasher]

No I do understand exactly. But as a repairman I can't fix that, I can only fix the source of the overheating problem and that is usually caused by a clogged up heatsink. A clogged heatsink is an extremely common problem and so it's extremely common to see Intel laptops running 100+C. I do agree they should throttle at a lower temp, but that's just not how it is.

Many china ARM boards often overestimate their abilities. Like these brand new H3 things, where it is pretended their CPU can handle 1.6GHz quad-core run. Which isn't exactly case in long run, especially if system runs without fairly large heatsink installed. Needless to say stupid thermal settings could be detrimental for system stability. So as system engineer I would welcome some standard way to place overrides. Maybe I'm missing something, but I'm not aware on how to do something like this so far, so it seems "reverse" kind of problem still exists, where one may want linux thermal subsystem to kick in much earlier than it would actually do.

[DeepDayze]

Will this patch work with HP 8460p (with i7 2620M)

[Veerappan]

Where did you see that this patch would benefit the T440p? (I'm asking because I have the same laptop)

I didn't. And honestly I'm not that sure it'll help my specific model as I pay pretty good attention to the performance and power states of it when I'm docked at work... But I am definitely playing around with the panel self-refresh and framebuffer compression options that have been enabled by default in 4.6 in an attempt to further improve my battery life.

[Alex Atkin UK]

Yet this issue still hasn't been fixed on Windows 10.

I'd rather use Linux but its just not there when it comes to usability with convertible touch-screen devices.

[triangle]

I am using a lenovo V570 (sandy bridge)
Here is my sensors output:

$ sensors
acpitz-virtual-0
Adapter: Virtual device
temp1: +49.0°C (crit = +98.0°C)
temp2: +44.0°C (crit = +126.0°C)

coretemp-isa-0000
Adapter: ISA adapter
Physical id 0: +49.0°C (high = +86.0°C, crit = +100.0°C)
Core 0: +47.0°C (high = +86.0°C, crit = +100.0°C)
Core 1: +48.0°C (high = +86.0°C, crit = +100.0°C)

temp2 looks funny, 126.0°C is a bit high!
Mite I be effected by this bug? (don't know if sensors tells me everything)

[Alex Atkin UK]

I'm not sure if sensors output is relevant but I can certainly confirm that my laptop also has the same temp2 as yours and definitely is affected by the issue.

In my case, it throttles the CPU to minimum once the battery hits 33%, so at least not quite as bad as people who are being throttled all the time.

On Windows people are bypassing the issue with ThrottleStop and disabling BD_PROCHOT.

[Stebs]

Mite I be effected by this bug? (don't know if sensors tells me everything)

My Lenovo S400 Ideapad is affected, drove me crazy until I found out the cause/correct bug some days ago. Especially since it "suddenly" occured (from kernel 4.4.3 -> 4.4.4). Trying with an up-to-date 4.1 kernel did not change anything, so I thought, it were not kernel-related. But it is, the Bug has been "backported" to a lot of kernels!
I can't see anything special with sensors, all fine there.
But you can see the Bug with:

cpupower frequency-info

Look at this line:

[I]current policy: frequency should be within xxx GHz and yyy GHz[/I]

Upon cold boot, all is fine and yyy is at max (2.6 Ghz for me)
But after a resume, yyy goes down, sometimes first to 2 Ghz, after more resumes, it comes down to 1000 Mhz.
So after suspend/resume, my laptop only ran max. at 1 Ghz.
Currently using kernel 3.14.64-1, all is fine there, after reading this article, i will test 3.18 (until new kernels with patch are out).

[edmon]

My Lenovo S400 Ideapad is affected, drove me crazy until I found out the cause/correct bug some days ago. Especially since it "suddenly" occured (from kernel 4.4.3 -> 4.4.4). Trying with an up-to-date 4.1 kernel did not change anything, so I thought, it were not kernel-related. But it is, the Bug has been "backported" to a lot of kernels!
I can't see anything special with sensors, all fine there.
But you can see the Bug with:

cpupower frequency-info

Look at this line:

[I]current policy: frequency should be within xxx GHz and yyy GHz[/I]

Upon cold boot, all is fine and yyy is at max (2.6 Ghz for me)
But after a resume, yyy goes down, sometimes first to 2 Ghz, after more resumes, it comes down to 1000 Mhz.
So after suspend/resume, my laptop only ran max. at 1 Ghz.
Currently using kernel 3.14.64-1, all is fine there, after reading this article, i will test 3.18 (until new kernels with patch are out).

try:

 #rmmod thermal;modprobe thermal

with 4.4.4 kernel