Microfreezes - FreeBSD 15.1 (bfr.13)

I put the main kernel 14.0, but the graphics mode did not boot (amdgpu) - it did not work to install drivers and stuff for X.
I didn't notice any microfreezes in the console.
 
Guys, just curious, what was the last recommended settings (e.g. kern.sched.steal_thresh=0) you used to mitigate scheduling stutters? Thank you!
 
IMHO Microfreezes return for FreeBSD 15.1 (mb and for before ...)
FreeBSD now have new model SMP - ULE
# sysctl kern.sched.ule.steal_thresh; sysctl kern.sched.ule.preempt_thresh; sysctl machdep.idle

New variable names and default values:
kern.sched.ule.steal_thresh: 2
kern.sched.ule.preempt_thresh: 80
machdep.idle: acpi

I set any values:
kern.sched.ule.steal_thresh: 2 <-> 1 <-> 0
kern.sched.ule.preempt_thresh: 80 <-> 40 <-> 224 - 224 I find on Internet space... but not remember place
machdep.idle: mwait <-> acpi

Try entering a long line of characters in any text editor and simply moving the cursor back and forth — you will eventually observe a freeze. As another variant: just grab a window with the mouse and move it around.

=== System ===
FreeBSD wfwc-001 15.1-RELEASE-p5 FreeBSD 15.1-RELEASE-p5 releng/15.1-n283639-893c7c99b60d GENERIC amd64
hw.model: AMD Ryzen 5 3600 6-Core Processor

=== kern.sched.ule ===
kern.sched.ule.always_steal: 0
kern.sched.ule.trysteal_limit: 2
kern.sched.ule.steal_thresh: 2
kern.sched.ule.steal_idle: 1
kern.sched.ule.balance_interval: 127
kern.sched.ule.balance: 1
kern.sched.ule.affinity: 1
kern.sched.ule.idlespinthresh: 157
kern.sched.ule.idlespins: 10000
kern.sched.ule.static_boost: 115
kern.sched.ule.preempt_thresh: 224
kern.sched.ule.interact: 30
kern.sched.ule.slice: 12
kern.sched.ule.quantum: 94488

=== machdep.idle ===
machdep.idle: mwait

=== idle test instructions ===
Before: mwait
Run: sysctl machdep.idle=spin
Then reproduce: drag a window with the mouse / move cursor in editor
Record: did freezes disappear, become rarer, or stay the same?
Restore: sysctl machdep.idle=mwait
sysctl hw.model
hw.model: AMD Ryzen 5 3600 6-Core Processor

pciconf -lv | grep -B3 display

vgapci0@pci0:13:0:0: class=0x030000 rev=0xc1 hdr=0x00 vendor=0x1002 device=0x73ef subvendor=0x1da2 subdevice=0x452e
vendor = 'Advanced Micro Devices, Inc. [AMD/ATI]'
device = 'Navi 23 [Radeon RX 6650 XT / 6700S / 6800S]'
class = display
=== DRM package ===
drm-612-kmod-6.12.110.1501000_3 Direct Rendering Manager (DRM) GPU drivers
drm-kmod-20260508 Direct Rendering Manager (DRM) GPU drivers metaport
gpu-firmware-kmod-20260519,1 Firmware modules for the drm-kmod drivers

=== Mesa packages ===
mesa-dri-26.2.2_1 OpenGL hardware acceleration drivers for DRI2+
mesa-libs-26.2.2_1 OpenGL libraries that support GLX and EGL clients

=== Loaded GPU kernel modules ===
32 1 0xffffffff83c00000 8c1e28 amdgpu.ko
39 1 0xffffffff83ae2000 356d0 amdgpu_dimgrey_cavefish_sos_bin.ko
40 1 0xffffffff83b18000 415e0 amdgpu_dimgrey_cavefish_ta_bin.ko
41 1 0xffffffff83b5a000 3dd88 amdgpu_dimgrey_cavefish_smc_bin.ko
42 1 0xffffffff83b98000 1de58 amdgpu_dimgrey_cavefish_dmcub_bin.ko
43 1 0xffffffff83bb6000 425e0 amdgpu_dimgrey_cavefish_pfp_bin.ko
44 1 0xffffffff844c2000 425e0 amdgpu_dimgrey_cavefish_me_bin.ko
45 1 0xffffffff84505000 42560 amdgpu_dimgrey_cavefish_ce_bin.ko
46 1 0xffffffff84548000 238f0 amdgpu_dimgrey_cavefish_rlc_bin.ko
47 1 0xffffffff8456c000 43860 amdgpu_dimgrey_cavefish_mec_bin.ko
48 1 0xffffffff845b0000 43860 amdgpu_dimgrey_cavefish_mec2_bin.ko
49 1 0xffffffff845f4000 a5e0 amdgpu_dimgrey_cavefish_sdma_bin.ko
50 1 0xffffffff845ff000 8fb70 amdgpu_dimgrey_cavefish_vcn_bin.ko

=== Mesa / OpenGL runtime info ===
OpenGL vendor string: AMD
OpenGL renderer string: AMD Radeon RX 6650 XT (radeonsi, navi23, ACO, DRM 3.61, 15.1-RELEASE-p5)
OpenGL version string: 4.6 (Compatibility Profile) Mesa 26.2.2

Now I set:
# sysctl kern.sched.ule.steal_thresh; sysctl kern.sched.ule.preempt_thresh; sysctl machdep.idle
kern.sched.ule.steal_thresh: 2
kern.sched.ule.preempt_thresh: 224 ?????
machdep.idle: mwait
 
Last edited:
Yes, bugzilla. And also consider perhaps updating the title of the post and remove "13" , so this thread gets re-surfaced. I tried out the "moving window around repeteadly" and have not been able to reproduce the stall on a i915 gpu. This was on x11 , no picom. Is this something specific to the amd drm pipeline perhaps? Can others with diffrent GPUs reproduce this?
 
siladi said:
Yes, bugzilla. And also consider perhaps updating the title of the post and remove "13" , so this thread gets re-surfaced. I tried out the "moving window around repeteadly" and have not been able to reproduce the stall on a i915 gpu. This was on x11 , no picom. Is this something specific to the amd drm pipeline perhaps? Can others with diffrent GPUs reproduce this?
Click to expand...
I don't known ... Mb drm. Any forums say what drm drivers but before very simple freezes depend on AMD CPU. I don't find now needed info...
 
Question:
It sounds like OP WCSN has modified scheduling parameters. Is that correct? If so, have you tried the behavior with the defaults?
Asking because I think there have been changes that made defaults "better"
 
I'd try 4BSD with a 3600 and disable SMT in BIOS, but I think GPU crash/errors might be unrelated (I'd try oldest drm-kmod available; either 515 or 61)

loader.conf:
Code: 
kern.hz="137438"
kern.eventtimer.periodic="1"
kern.eventtimer.idletick="1"
kern.sched.name="4BSD"

sysctl.conf (I'd disable mitigations for testing and maybe back on/1 later, Ryzen might have more mitigations; afaik hlt/halt's better-perf than mwait)
Code: 
machdep.mitigations.zenbleed.enable="0"
machdep.mitigations.rngds.enable="0"
machdep.mitigations.flush_rsb_ctxsw="0"
machdep.idle="hlt"
machdep.idle_mwait="0"
kern.sched.4bsd.ipiwakeup.enabled="0"
kern.sched.4bsd.ipiwakeup.useloop="1"
kern.sched.4bsd.ipiwakeup.usemask="0"

Not sure if this thread was referencing FreeBSD 14 before, but it feels like something changed with desktop interactions and random stutter with 15 (Warcraft 3's audio was stuttery-mess without most of those settings 15.1/16.0 on no-HT Skylake; I read ULE's more-tuned for SMT/logical multicore on modern CPUs, but logical's never been consistent on desktop hence 4BSD + SMT off :p)
 
You must log in or register to reply here.
Back
Top