Requests for -current (14.2-->15.0)

GazL · 10-20-2021, 08:21 AM

Quote:

Originally Posted by Daedra

According to this post, it looks like it does default to full if nothing is specified at boot time.

Which seems at odds with the quote from the help text LC quoted:

Quote:

This option allows to define the preemption model on the kernel command line parameter and thus override the default preemption model defined during compile time.

... which would imply that if you have (as slackware does)

Code:

# CONFIG_PREEMPT_NONE is not set
CONFIG_PREEMPT_VOLUNTARY=y
# CONFIG_PREEMPT is not set
CONFIG_HAVE_PREEMPT_DYNAMIC=y
CONFIG_PREEMPT_NOTIFIERS=y
CONFIG_DRM_I915_PREEMPT_TIMEOUT=640
# CONFIG_PREEMPTIRQ_DELAY_TEST is not set

then it would default to voluntary but one can override it.

Much of the reading on this seems unclear and contradictory.

ctrlaltca · 10-20-2021, 08:31 AM

It defaults to full when no option is specified explicitly:

https://git.kernel.org/pub/scm/linux...d/core.c#n6594

If you define a preemption model, this is the function that changes the mode and prints the line in the kernel log:

https://git.kernel.org/pub/scm/linux...d/core.c#n6610

Daedra · 10-20-2021, 08:38 AM

Quote:

Originally Posted by franzen

Happy birthday Pat!

Thanks to Daedra for the preempt links, i've now running a preempt kernel
which i can change at boottime to none/voluntary/full at runtime.

Code:

> uname -a
Linux xxx 5.14.13 #1 SMP PREEMPT Wed Oct 20 06:15:56 CEST
2021 x86_64 Intel(R) Core(TM)2 Duo CPU     T6670  @ 2.20GHz GenuineIntel
GNU/Linux

lilo.conf

Code:

append=" preempt=voluntary "

reboot

Code:

# dmesg | grep Preempt:
[    0.060149] Dynamic Preempt: voluntary

lilo.conf

Code:

append=" preempt=full "

reboot

Code:

# dmesg | grep Preempt:
[    0.059655] Dynamic Preempt: full

If there's no need for preemption, this kernelconfig may be still more
benefitical than the one on -current, e.g. set preempt=none on servers.

Thanks for the config. It's nice to see someone has tested this. I plan on giving it a go this weekend when I have more time.

Daedra · 10-20-2021, 08:56 AM

Quote:

Originally Posted by GazL

Which seems at odds with the quote from the help text LC quoted:

... which would imply that if you have (as slackware does)

Code:

# CONFIG_PREEMPT_NONE is not set
CONFIG_PREEMPT_VOLUNTARY=y
# CONFIG_PREEMPT is not set
CONFIG_HAVE_PREEMPT_DYNAMIC=y
CONFIG_PREEMPT_NOTIFIERS=y
CONFIG_DRM_I915_PREEMPT_TIMEOUT=640
# CONFIG_PREEMPTIRQ_DELAY_TEST is not set

then it would default to voluntary but one can override it.

Much of the reading on this seems unclear and contradictory.

The way I am understanding it is if you set CONFIG_PREEMPT_VOLUNTARY=y then that would be static and voluntary is the only option. In order for dynamic to work the arch has to support dynamic which can be checked in the config, the line would be CONFIG_HAVE_PREEMPT_DYNAMIC. Then you have to change CONFIG_PREEMPT_VOLUNTARY=y to CONFIG_PREEMPT=y for dynamic to even be an option.

Code:

On Mon, Apr 26, 2021 at 3:31 AM Chris Murphy <lists(a)colorremedies.com&gt;
wrote:

...

CONFIG_HAVE_PREEMPT_DYNAMIC only says the architecture supports the
feature. This is hard-coded by arch/x86/Kconfig.
To actually enable the feature, CONFIG_PREEMPT=y would have to be set. The
combination of CONFIG_PREEMPT and CONFIG_HAVE_PREEMPT_DYNAMIC then selects
CONFIG_PREEMPT_DYNAMIC. See kernel/Kconfig.preempt.

Michal

GazL · 10-20-2021, 09:31 AM

Yes, you're right daedra. I was under a misconception that CONFIG_DYNAMIC_PREEMPT was a separate option one could select in addition to the 3 normal choices, but that's not how they've done it.

Daedra · 10-20-2021, 09:35 AM

I do find it somewhat odd that the kernel devs decided to make full the default if nothing is specified. It seems that voluntary would of been a saner default since its a good one size fits all solution.

saxa · 10-20-2021, 12:32 PM

vte-0.66.0
https://download.gnome.org/sources/v...-0.66.0.tar.xz

marav · 10-20-2021, 01:08 PM

LVM 2.03.14

Code:

Version 2.03.14 - 20th October 2021
===================================
- Device scanning is skipping directories on different filesystems.
- Print info message with too many or too large archived files.
- Reduce metadata readings during scanning phase.
- Optimize computation of crc32 check sum with multiple PVs.
- Enhance recover path on cache creation failure.
- Filter out unsupported MQ/SMQ cache policy setting.
- Fix memleak in mpath filter.
- Support newer location for VDO statistics.
- Add support for VDO async-unsafe write policy.
- Improve lvm_import_vdo script.
- Support VDO LV with lvcreate -ky.
- Fix lvconvert for VDO LV bigger then 2T.
- Create VDO LVs automatically without zeroing.
- Rename vdoimport to lvm_import_vdo.

https://sourceware.org/pub/lvm2/LVM2.2.03.14.tgz

marav · 10-20-2021, 01:24 PM

Quote:

Originally Posted by Roman Dyaba

i was do build linux-5.14.13-realtime for my system with make CC=clang ,
but developers later let have well through with build linux-5.15 , because it tune for CLANG
compilator, i.e. via NVIDIA and other com yesterday switch to LLVM for crossplatform.
CLang is part of GCC community too.

in morning today i was do build linux-5.14.13-realtime kernel with make CC=clang, and it work fine !
it was use LLVM-13.0, from Slackware-current 64-bit updates, what i see in changelog.

This changes now open to Cross-Platform for standard Slackware.

https://llvm.org

"Clang is clearly doing a *HORRIBLE* job with stack usage." - L. Torvalds - Sept. 2021

volkerdi · 10-20-2021, 03:13 PM

Quote:

Originally Posted by marav

LVM 2.03.14

Code:

Version 2.03.14 - 20th October 2021
===================================
- Device scanning is skipping directories on different filesystems.
- Print info message with too many or too large archived files.
- Reduce metadata readings during scanning phase.
- Optimize computation of crc32 check sum with multiple PVs.
- Enhance recover path on cache creation failure.
- Filter out unsupported MQ/SMQ cache policy setting.
- Fix memleak in mpath filter.
- Support newer location for VDO statistics.
- Add support for VDO async-unsafe write policy.
- Improve lvm_import_vdo script.
- Support VDO LV with lvcreate -ky.
- Fix lvconvert for VDO LV bigger then 2T.
- Create VDO LVs automatically without zeroing.
- Rename vdoimport to lvm_import_vdo.

https://sourceware.org/pub/lvm2/LVM2.2.03.14.tgz

I think they forgot to mention the changes to the udev rules that seem to be trying to call out to systemd-run.

GazL · 10-20-2021, 04:01 PM

Quote:

Originally Posted by volkerdi

I think they forgot to mention the changes to the udev rules that seem to be trying to call out to systemd-run.

All part of the game-plan to force the hold-outs to surrender and migrate over I expect.

drumz · 10-20-2021, 04:46 PM

Quote:

Originally Posted by volkerdi

I think they forgot to mention the changes to the udev rules that seem to be trying to call out to systemd-run.

Ah, this must be the reason one of my partitions is no longer mounted at boot. I have LVM on top of LUKS on top of a raw partition. After boot I have to manually run

Code:

vgchange -ay --ignorelockingfailure

and then "mount" works.

I separately have a LUKS on top of LVM on top of multiple disks which correctly gets mounted at boot.

Relevant dmesg:

Code:

[   39.637923] udevd[2000]: failed to execute '/usr/bin/systemd-run' '/usr/bin/systemd-run -r --no-block --property DefaultDependencies=no --unit lvm-activate-fastcryptvg lvm vgchange -aay --nohints fastcryptvg': No such file or directory

I'll try editing /lib/udev/rules.d/69-dm-lvm.rules to remove systemd to see if I can restore previous behavior...

drumz · 10-20-2021, 05:06 PM

Changing line 83 of /lib/udev/rules.d/69-dm-lvm.rules from:

Code:

ENV{LVM_VG_NAME_COMPLETE}=="?*", RUN+="/usr/bin/systemd-run -r --no-block --property DefaultDependencies=no --unit lvm-activate-$env{LVM_VG_NAME_COMPLETE} lvm vgchange -aay --nohints $env{LVM_VG_NAME_COMPLETE}"

to

Code:

RUN+="/sbin/lvm pvscan --background --cache --activate ay --major $major --minor $minor", ENV{LVM_SCANNED}="1"

Fixes my problem. (That is, my partition is now auto-mounted at boot time again, which stock lvm2-2.03.14 broke.)

These comments from the file (from the previous working version of lvm2) are enlightening:

Code:

# The method for invoking pvscan is selected at build time with the option
# --(enable|disable)-udev-systemd-background-jobs to "configure".
# On modern distributions with recent systemd, it's "systemd_background";
# on others, "direct_pvscan".
GOTO="direct_pvscan"
...
# FIXME: this mode is not used and should be removed.
LABEL="direct_pvscan"

I don't know if my fix is "proper". I just copied the relevant line from the old (lvm2-2.03.13) package.

garpu · 10-20-2021, 05:41 PM

So hold off on updating LVM? Or safe to do so?

marav · 10-20-2021, 06:26 PM

Quote:

Originally Posted by volkerdi

I think they forgot to mention the changes to the udev rules that seem to be trying to call out to systemd-run.

it's not the kind of gift you were expecting I guess ...