performance problems after kernel/microcode update ?

List overview All Threads
Download

newer

older

Re: [LAU] retire_capture_urb: xxxx...

Recommended USB Audio Interface

Niklas Reppel

14 Sep 2018 14 Sep '18

1:34 a.m.

Hi, i recently updated my Arch linux system, to to the following version: Linux 4.18.6-arch1-1-ARCH #1 SMP PREEMPT Afterwards, i noticed that something has changed regarding the realtime permissions and that i had to install a new package called 'realtime-priviliges', which i never needed before, and subsequently add my main user to the 'realtime' group. In that wake, i updated the processor microcode, which i didn't use at all before (there always was some error message during bootup, but nothing else). After being able to start JACK in realtime again, i felt that there was a significant loss of performance. Things that didn't cause any dropouts before now start to crackle and gristle (as a sidenote, i mostly use SuperCollider to make sound). Also, and this puzzled me especially, there were dropouts, but no XRuns ... normally, when i push the system to the limits, i get XRuns, that'd be expected. But now, it's dropouts without, and they start way earlier. My machine has an Intel(R) Core(TM) i7-4600U CPU @ 2.10GHz Processor if that info is of user for anyone ... Anybody else noticed something like that recently ? Best, nik -- Niklas Reppel www.parkellipsen.de

Show replies by date

Michael Jarosch

14 Sep 14 Sep

2:58 a.m.

Am Donnerstag, den 13.09.2018, 23:26 +0200 schrieb Niklas Reppel:

...

After being able to start JACK in realtime again, i felt that there was a significant loss of performance.

Maybe it has nothing to do with a/the new realtime management, but with spectre mitigation and stuff. Phoronix listed all the kernel parameters to get rid of these workarounds. I don't find the exact side anymore, but these are mainly the parameters you have to add to your /etc/default/grub - only use them, if you don't care about your system's security :D : pti=off spectre_v2=off l1tf=off nospec_store_bypass_disable no_stf_barrier Greets! Mitsch

Tim

3:13 a.m.

On 09/13/2018 05:26 PM, Niklas Reppel wrote:

...

Hello, I always reply when I hear of crackles, pops, and dropouts because it took me several months to figure out my problem with that. Maybe it will help. Here is the latest time I replied, a month ago, please scroll down: http://linux-audio.4202.n7.nabble.com/M-Audio-Fast-Track-Pro-unreliable-dis… Long story short: Try running with only one processor core active. It completely cured my problem, and another user's as well. In my case the device is a PCI M-Audio delta1010, and in the other user's case it was a similar ice1712-based card. However in the above Fast-Track-Pro case, that's USB and we don't yet know if the trick solved that user's problem... Be patient: If the number of CPU cores is the cause it may take several minutes for the crackling to appear since I believe what happens is that some low-level 'counters/pointers' which should always be in sync slowly approach and 'pass' each other, and that's when the noise occurs. In my case the very same symptom occurred in Windows. That proved beyond doubt that it was a hardware issue. Hope that helps. Tim.

David W. Jones

5:51 a.m.

On September 13, 2018 1:13:12 PM HST, Tim <termtech(a)rogers.com> wrote:

...

On 09/13/2018 05:26 PM, Niklas Reppel wrote:

Hi, i recently updated my Arch linux system, to to the following

version:

Linux 4.18.6-arch1-1-ARCH #1 SMP PREEMPT Afterwards, i noticed that something has changed regarding the

realtime

permissions and that i had to install a new package called

'realtime-priviliges',

which i never needed before, and subsequently add my main user to the 'realtime' group. In that wake, i updated the processor microcode, which i didn't use

all before (there always was some error message during bootup, but nothing else). After being able to start JACK in realtime again, i felt that there

was

a significant loss of performance. Things that didn't cause any dropouts before now start to crackle

and

gristle (as a sidenote, i mostly use SuperCollider to make sound). Also, and this puzzled me especially, there were dropouts, but no

XRuns

... normally, when i push the system to the limits, i get XRuns, that'd be expected. But now, it's

dropouts without, and they start way earlier. My machine has an Intel(R) Core(TM) i7-4600U CPU @ 2.10GHz Processor

that info is of user for anyone ... Anybody else noticed something like that recently ? Best, nik

While I, on the other hand, use all cores (real and hyperthread) on my i7 with no crackles or distorted recordings and such at all. Same for the 4 real cores on my AMD desktop system. In my opinion, most PC motherboards weren't designed with any audio thought beyond including an on-board audio chip. -- David W. Jones gnome(a)hawaii.rr.com authenticity, honesty, community http://dancingtreefrog.com Sent from my Android device with F/LOSS K-9 Mail.

Will Godfrey

10:06 a.m.

On Thu, 13 Sep 2018 15:48:11 -1000 "David W. Jones" <gnome(a)hawaii.rr.com> wrote:

...

On September 13, 2018 1:13:12 PM HST, Tim <termtech(a)rogers.com> wrote:

On 09/13/2018 05:26 PM, Niklas Reppel wrote:

Hi, i recently updated my Arch linux system, to to the following

version:

Linux 4.18.6-arch1-1-ARCH #1 SMP PREEMPT Afterwards, i noticed that something has changed regarding the

realtime

permissions and that i had to install a new package called

'realtime-priviliges',

which i never needed before, and subsequently add my main user to the 'realtime' group. In that wake, i updated the processor microcode, which i didn't use

all before (there always was some error message during bootup, but nothing else). After being able to start JACK in realtime again, i felt that there

was

a significant loss of performance. Things that didn't cause any dropouts before now start to crackle

and

gristle (as a sidenote, i mostly use SuperCollider to make sound). Also, and this puzzled me especially, there were dropouts, but no

XRuns

... normally, when i push the system to the limits, i get XRuns, that'd be expected. But now, it's

dropouts without, and they start way earlier. My machine has an Intel(R) Core(TM) i7-4600U CPU @ 2.10GHz Processor

that info is of user for anyone ... Anybody else noticed something like that recently ? Best, nik

Is this only Intel processors that are doing this? I've been running an AMD Ryzen since March and had no problems at all. -- Will J Godfrey http://www.musically.me.uk Say you have a poem and I have a tune. Exchange them and we can both have a poem, a tune, and a song.

david

10:49 a.m.

On 09/13/2018 08:05 PM, Will Godfrey wrote:

...

On Thu, 13 Sep 2018 15:48:11 -1000 "David W. Jones" <gnome(a)hawaii.rr.com> wrote:

On September 13, 2018 1:13:12 PM HST, Tim <termtech(a)rogers.com> wrote:

On 09/13/2018 05:26 PM, Niklas Reppel wrote:

Hi, i recently updated my Arch linux system, to to the following

version:

Linux 4.18.6-arch1-1-ARCH #1 SMP PREEMPT Afterwards, i noticed that something has changed regarding the

realtime

permissions and that i had to install a new package called

'realtime-priviliges',

which i never needed before, and subsequently add my main user to the 'realtime' group. In that wake, i updated the processor microcode, which i didn't use

all before (there always was some error message during bootup, but nothing else). After being able to start JACK in realtime again, i felt that there

was

a significant loss of performance. Things that didn't cause any dropouts before now start to crackle

and

gristle (as a sidenote, i mostly use SuperCollider to make sound). Also, and this puzzled me especially, there were dropouts, but no

XRuns

... normally, when i push the system to the limits, i get XRuns, that'd be expected. But now, it's

dropouts without, and they start way earlier. My machine has an Intel(R) Core(TM) i7-4600U CPU @ 2.10GHz Processor

that info is of user for anyone ... Anybody else noticed something like that recently ? Best, nik

Is this only Intel processors that are doing this? I've been running an AMD Ryzen since March and had no problems at all.

You mean for the original poster, or me? I have no such problems on either of my systems. But my AMD isn't a Ryzen, it's a Phenom II. -- David W. Jones gnome(a)hawaii.rr.com authenticity, honesty, community http://dancingtreefrog.com

Will Godfrey

3:13 p.m.

On Thu, 13 Sep 2018 20:49:31 -1000 david <gnome(a)hawaii.rr.com> wrote:

...

On 09/13/2018 08:05 PM, Will Godfrey wrote: > On Thu, 13 Sep 2018 15:48:11 -1000

<snip>

...

Is this only Intel processors that are doing this? I've been running an AMD Ryzen since March and had no problems at all.

You mean for the original poster, or me? I have no such problems on either of my systems. But my AMD isn't a Ryzen, it's a Phenom II.

I was referring to the O/P -- Will J Godfrey http://www.musically.me.uk Say you have a poem and I have a tune. Exchange them and we can both have a poem, a tune, and a song.

David Runge

11:59 a.m.

Hi Niklas, On 2018-09-13 23:26:30 (+0200), Niklas Reppel wrote:

...

Please subscribe to the arch-proaudio mailing list. The change was announced there [1].

...

In that wake, i updated the processor microcode, which i didn't use at all before (there always was some error message during bootup, but nothing else). After being able to start JACK in realtime again, i felt that there was a significant loss of performance. Things that didn't cause any dropouts before now start to crackle and gristle (as a sidenote, i mostly use SuperCollider to make sound). Also, and this puzzled me especially, there were dropouts, but no XRuns ... normally, when i push the system to the limits, i get XRuns, that'd be expected. But now, it's dropouts without, and they start way earlier. My machine has an Intel(R) Core(TM) i7-4600U CPU @ 2.10GHz Processor if that info is of user for anyone ... Anybody else noticed something like that recently ?

It is very likely that the changes introduced to the Linux Kernel in response to the spectre and meltdown security issues are causing this in combination with (known to be) performance degrading Intel ucode. However, I would not advice you to disable mitigations such as page table isolation (PTI), unless you know about the implied risks! Best, David [1] https://lists.archlinux.org/pipermail/arch-proaudio/2018-August/000191.html -- https://sleepmap.de

Ralf Mardorf

15 Sep 15 Sep

3:17 a.m.

On Fri, 14 Sep 2018 09:59:25 +0200, David Runge wrote:

...

However, I would not advice you to disable mitigations such as page table isolation (PTI), unless you know about the implied risks!

Regarding Meltdown, loosely speaking, if the audio workstation is connected to the Internet, we could boot without disabling PTI. If the audio workstation is used for audio productions, we could boot with disabling PTI and without connecting to the Internet. IIRC the microcode for my CPU not only is required for Spectre mitigation, but also to make TSC available and perhaps to fix other CPU issues, too, so there might be no option to boot without the microcode at all. The '/sys/devices/system/cpu/vulnerabilities/*' "list" is increasing ;). [rocketmouse@archlinux ~]$ uname -a Linux archlinux 4.18.7-rt5-1-rt-securityink #1 SMP PREEMPT RT Thu Sep 13 08:01:15 CEST 2018 x86_64 GNU/Linux [rocketmouse@archlinux ~]$ dmesg | grep micro [ 0.000000] microcode: microcode updated early to revision 0x25, date = 2018-04-02 [ 0.457030] microcode: sig=0x306c3, pf=0x2, revision=0x25 [ 0.457083] microcode: Microcode Update Driver: v2.2. [rocketmouse@archlinux ~]$ ls -l /sys/devices/system/cpu/vulnerabilities/* -r--r--r-- 1 root root 4096 Sep 15 01:12 /sys/devices/system/cpu/vulnerabilities/l1tf -r--r--r-- 1 root root 4096 Sep 15 01:12 /sys/devices/system/cpu/vulnerabilities/meltdown -r--r--r-- 1 root root 4096 Sep 15 01:12 /sys/devices/system/cpu/vulnerabilities/spec_store_bypass -r--r--r-- 1 root root 4096 Sep 15 01:12 /sys/devices/system/cpu/vulnerabilities/spectre_v1 -r--r--r-- 1 root root 4096 Sep 15 01:12 /sys/devices/system/cpu/vulnerabilities/spectre_v2 [rocketmouse@archlinux ~]$ cat /sys/devices/system/cpu/vulnerabilities/* Mitigation: PTE Inversion; VMX: conditional cache flushes, SMT disabled Mitigation: PTI Mitigation: Speculative Store Bypass disabled via prctl and seccomp Mitigation: __user pointer sanitization Mitigation: Full generic retpoline, IBPB, IBRS_FW

Ralf Mardorf

4:41 a.m.

PS: The microcode has not that much impact regarding Spectre mitigation, as I thought. [rocketmouse@archlinux ~]$ grep 'no micro' /boot/syslinux/syslinux.cfg -A4 MENU LABEL Arch Linux Rt Securityink no micro LINUX ../vmlinuz-linux-rt-securityink APPEND root=LABEL=s3.archlinux ro INITRD ../initramfs-linux-rt-securityink.img [rocketmouse@archlinux ~]$ ls -l /sys/devices/system/cpu/vulnerabilities/* -r--r--r-- 1 root root 4096 Sep 15 02:28 /sys/devices/system/cpu/vulnerabilities/l1tf -r--r--r-- 1 root root 4096 Sep 15 02:28 /sys/devices/system/cpu/vulnerabilities/meltdown -r--r--r-- 1 root root 4096 Sep 15 02:28 /sys/devices/system/cpu/vulnerabilities/spec_store_bypass -r--r--r-- 1 root root 4096 Sep 15 02:28 /sys/devices/system/cpu/vulnerabilities/spectre_v1 -r--r--r-- 1 root root 4096 Sep 15 02:28 /sys/devices/system/cpu/vulnerabilities/spectre_v2 [rocketmouse@archlinux ~]$ cat /sys/devices/system/cpu/vulnerabilities/* Mitigation: PTE Inversion; VMX: conditional cache flushes, SMT disabled Mitigation: PTI Vulnerable Mitigation: __user pointer sanitization Mitigation: Full generic retpoline ################################################################################################## [rocketmouse@archlinux ~]$ grep 'Securityink nopt^i' /boot/syslinux/syslinux.cfg -A4 MENU LABEL Arch Linux Rt Securityink nopt^i LINUX ../vmlinuz-linux-rt-securityink APPEND root=LABEL=s3.archlinux ro nopti INITRD ../intel-ucode.img,../initramfs-linux-rt-securityink.img [rocketmouse@archlinux ~]$ ls -l /sys/devices/system/cpu/vulnerabilities/* -r--r--r-- 1 root root 4096 Sep 15 02:33 /sys/devices/system/cpu/vulnerabilities/l1tf -r--r--r-- 1 root root 4096 Sep 15 02:33 /sys/devices/system/cpu/vulnerabilities/meltdown -r--r--r-- 1 root root 4096 Sep 15 02:33 /sys/devices/system/cpu/vulnerabilities/spec_store_bypass -r--r--r-- 1 root root 4096 Sep 15 02:33 /sys/devices/system/cpu/vulnerabilities/spectre_v1 -r--r--r-- 1 root root 4096 Sep 15 02:33 /sys/devices/system/cpu/vulnerabilities/spectre_v2 [rocketmouse@archlinux ~]$ cat /sys/devices/system/cpu/vulnerabilities/* Mitigation: PTE Inversion; VMX: conditional cache flushes, SMT disabled Vulnerable Mitigation: Speculative Store Bypass disabled via prctl and seccomp Mitigation: __user pointer sanitization Mitigation: Full generic retpoline, IBPB, IBRS_FW