That Peculiar Linux 3.18 Kernel Bug Might Be Closed Soon

[phoronix]

Phoronix: That Peculiar Linux 3.18 Kernel Bug Might Be Close To Being Closed

For the past month there's been kernel developers investigating "a big unknown worry in a regression" that have left many key kernel developers -- including Linus Torvalds -- puzzled. It looks like that investigation is finally being close to being resolved...

http://www.phoronix.com/vr.php?view=MTg3Mzk

[bitman]

Except here it happens on amd 990fx chipset motherboard. Wth..

[Adarion]

Was flying over the Wikipedia page on HPET... "designed by MS and Intel" set off my alarms. Then flying over Thomas Gleixner's message I read "First of all the HPET design is a complete failure.". Oh, well. I better set tsc as my clocksource then. And I had thought HPET was meant to be an improvement.

And merry belated xmas/chanuka.

[GreatEmerald]

Hah, I told you it might be hardware-specific!

[Azrael5]

Good news, I red issues on HPET long ago. It was introduced to reduce the time of response between devices increasing the frequencies of request; but this technology apparently was abandoned after Vista. I?m waiting for next releas just for these kernel improvements.

Good kernel makes good devices.

[AJSB]

When compiling a Kernel, can we disable HPET option safely ?
i.e. does all MoBos have a 8254 timer legacy mode ?

[Azrael5]

When compiling a Kernel, can we disable HPET option safely ?
i.e. does all MoBos have a 8254 timer legacy mode ?

How can work hpet features without hpet chip and without enabling option on bios?

[steveriley]

If I'm interpreting this correctly, it looks like HPET is the clocksource early during boot, but then it gets switched to TSC.

Kubuntu 14.10:

teve@t520:~$ [B]dmesg |egrep -i 'hpet|tsc'[/B]
[    0.000000] ACPI: HPET 0x00000000DAFE7000 000038 (v01 LENOVO TP-8A    00001440 PTL  00000002)
[    0.000000] ACPI: HPET id: 0x8086a301 base: 0xfed00000
[    0.000000] hpet clockevent registered
[    0.000000] tsc: Fast TSC calibration using PIT
[    0.000000] tsc: Detected 2691.296 MHz processor
[    0.031422] HPET id 0 under DRHD base 0xfed91000
[    0.031424] HPET id 0 under DRHD base 0xfed91000
[    0.031426] HPET id 0 under DRHD base 0xfed91000
[    0.031427] HPET id 0 under DRHD base 0xfed91000
[    0.031429] HPET id 0 under DRHD base 0xfed91000
[    0.031431] HPET id 0 under DRHD base 0xfed91000
[    0.031432] HPET id 0 under DRHD base 0xfed91000
[    0.031434] HPET id 0 under DRHD base 0xfed91000
[    0.071733] TSC deadline timer enabled
[    0.192350] hpet0: at MMIO 0xfed00000, IRQs 2, 8, 0, 0, 0, 0, 0, 0
[    0.192356] hpet0: 8 comparators, 64-bit 14.318180 MHz counter
[    0.194393] Switched to clocksource hpet
[    0.571652] rtc_cmos 00:02: alarms up to one month, y3k, 114 bytes nvram, hpet irqs
[    1.496840] tsc: Refined TSC clocksource calibration: 2691.257 MHz
[    2.495526] Switched to clocksource tsc
[    4.694699] vboxdrv: TSC mode is 'synchronous', kernel timer mode is 'normal'.

steve@t520:~$ [B]cat /sys/devices/system/clocksource/clocksource0/current_clocksource[/B]
tsc

Debian 7.7:

root@m92p:~# [B]dmesg |egrep -i 'hpet|tsc'[/B]
[    0.000000] ACPI: HPET 0x00000000D4615920 000038 (v01 LENOVO TC-9S    00001460 AMI. 00000005)
[    0.000000] ACPI: HPET id: 0x8086a701 base: 0xfed00000
[    0.000000] hpet clockevent registered
[    0.000000] tsc: Fast TSC calibration using PIT
[    0.000000] tsc: Detected 2893.416 MHz processor
[    0.025331] HPET id 0 under DRHD base 0xfed91000
[    0.066313] TSC deadline timer enabled
[    0.207724] hpet0: at MMIO 0xfed00000, IRQs 2, 8, 0, 0, 0, 0, 0, 0
[    0.208163] hpet0: 8 comparators, 64-bit 14.318180 MHz counter
[    0.210251] Switched to clocksource hpet
[    0.483350] rtc_cmos 00:02: alarms up to one month, y3k, 242 bytes nvram, hpet irqs
[    1.474632] tsc: Refined TSC clocksource calibration: 2893.426 MHz
[    2.475064] Switched to clocksource tsc

root@m92p:~# [B]cat /sys/devices/system/clocksource/clocksource0/current_clocksource[/B]
tsc

[halo9en]

If I'm interpreting this correctly, it looks like HPET is the clocksource early during boot, but then it gets switched to TSC.

And apparently it cannot be avoided when using nohz mode. Using clocksource=tsc I get: "Override clocksource tsc is not HRT compatible. Cannot switch while in HRT/NOHZ mode".

[Tobu]

And apparently it cannot be avoided when using nohz mode. Using clocksource=tsc I get: "Override clocksource tsc is not HRT compatible. Cannot switch while in HRT/NOHZ mode".

grep . /sys/devices/system/clocksource/clocksource*/*_clocksource

will show you the sources that still work in NoHz.
Looking at the code, that means high-precision, continuous, not unstable.