summaryrefslogtreecommitdiff
path: root/guides/sysadmin/machines/amdahl30/killing-time.org
blob: f781582d4c2152fb697fc6c7659d46e25de306a4 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
* Failure
On November 19 2024, LDLC's off-brand SSD died on me.  RIP.
Re-installed Tumbleweed on the replacement (Kingston SA400S3) on
November 28.  Since then…
** Performance loss
Getting uncannily reproducible frame drops (60 ↘ 40±10, movement
visibly choppy) in Hades Ⅱ when moving toward effects/particles-heavy
areas.  No idea WTF, those areas ran fine before.

- "High" graphics setting at native 1920×1080 resolution.
  - Tried "Low" graphics, lowered resolution, disabled vsync: symptoms
    persist.
- Not forcing any "compatibility tool" version, assuming this yields
  "Proton Experimental".
  - Tried a couple of old Proton versions: symptoms persist.
- Reinstalled game & nuked everything under
  - =~/.cache/mesa_shader_cache*=
  - =~/.cache/radv_builtin_shaders*=
  - =~/.config/unity3d=
  - =~/.local/share/Steam=
  - =~/.local/share/vulkan/=
  - =~/.steam*=
  in case "stale shaders" were to blame or something.
- Tumbleweed/Plasma/Wayland session.
  - Tried X11: symptoms persist.
- Reducing noise with =balooctl6 suspend=, =swapoff -a= (RAM nowhere
  near exhausted).

Well then.
*** CPU frequency scaling?
Started by noticing that the Plasma "Power Management" tray widget
says "Power Profile" is "Not available".  Not 100% sure whether that
was the case with the old installation; maybe I had had something
configured or installed to enable this?

Internet says "install and enable power-profiles-daemon", except
that's on:

#+begin_example
$ systemctl status power-profiles-daemon.service
● power-profiles-daemon.service - Power Profiles daemon
     Loaded: loaded (/usr/lib/systemd/system/power-profiles-daemon.service; disabled; preset: disabled)
     Active: active (running) since Sun 2024-12-01 11:46:32 CET; 45min ago
 Invocation: b2545a02bc9642b7aeb5f370e8b50e7c
   Main PID: 2289 (power-profiles-)
      Tasks: 4 (limit: 18320)
        CPU: 52ms
     CGroup: /system.slice/power-profiles-daemon.service
             └─2289 /usr/libexec/power-profiles-daemon
#+end_example

But:

#+begin_example
$ powerprofilesctl
,* balanced:
    PlatformDriver:     placeholder

  power-saver:
    PlatformDriver:     placeholder
#+end_example

Internet says I am missing the right scaling driver, and seems very
keen on enabling =amd_pstate=, which I do not seem to have available:

#+begin_example
$ cpupower frequency-info
analyzing CPU 5:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 5
  CPUs which need to have their frequency coordinated by software: 5
  maximum transition latency:  Cannot determine or is not supported.
  hardware limits: 1.40 GHz - 3.70 GHz
  available frequency steps:  3.70 GHz, 1.70 GHz, 1.40 GHz
  available cpufreq governors: ondemand performance schedutil
  current policy: frequency should be within 1.40 GHz and 3.70 GHz.
                  The governor "schedutil" may decide which speed to use
                  within this range.
  current CPU frequency: Unable to call hardware
  current CPU frequency: 3.30 GHz (asserted by call to kernel)
  boost state support:
    Supported: yes
    Active: no

$ zcat /proc/config.gz | grep -i pstate
CONFIG_X86_INTEL_PSTATE=y
CONFIG_X86_AMD_PSTATE=y
CONFIG_X86_AMD_PSTATE_DEFAULT_MODE=3
# CONFIG_X86_AMD_PSTATE_UT is not set
#+end_example

=/proc/config.gz= suggests the kernel configuration supports it, but
=cpupower= does not seem to know about it.  =dmesg= offers:

#+begin_example
$ sudo dmesg -H
[…] amd_pstate: the _CPC object is not present in SBIOS or ACPI disabled
#+end_example

Though:

#+begin_example
$ lscpu | grep -i cppc
Flags:                                […] cppc […]
#+end_example

So ACPI problem?  Lots of posts mentioning =amd_= parameters on the
kernel command-line but AFAIU those are stale with newer kernels (6.11
here) which automatically (attempt to) load the =amd_pstate= driver.

Went through the UEFI menu and found nothing related to ACPI or
[[https://forum.level1techs.com/t/amd-p-state-driver/197885/24][X2APIC]].  Skeptical UEFI settings anyway, since I did not change them
between the old and new installations.

/Some time later/

Probably not ACPI, =dmesg= is choke full of ACPI noise.  OTOH, using
some diagnosis methods from [[https://bugzilla.kernel.org/show_bug.cgi?id=218171][this kernel bug report]]:

#+begin_example
$ find /sys/devices -name '*cppc*'
🦗
#+end_example

(=acpidump ; acpixtract ; iasl ; grep -i cpc *.dsl= also yields 🦗,
but =iasl= complains about "unresolved" "control methods", so 🤷)

/Some time later/

[[https://wiki.archlinux.org/title/CPU_frequency_scaling#amd_pstate][ArchWiki]] does say "Change /Enable CPPC/ […] from /Auto/ to /Enabled/".
My UEFI menu tucks that under /Overclocking → Advanced CPU
Configuration → AMD CBS → CPPC CTRL/.  That change *does* convince
Linux to enable =amd_pstate=; going over the previous tests in reverse
order:

#+begin_example
$ [… acpidump && acpixtract && iasl … ] && grep -i cpc *.dsl
ssdt1.dsl:        Name (_CPC, Package (0x17)  // _CPC: Continuous Performance Control
[… repeats 12 times …]

$ find /sys/devices -name '*cppc*' -o -name '*pstate*' | tr -s '[:digit:]' N | sort -u
/sys/devices/system/cpu/amd_pstate
/sys/devices/system/cpu/cpufreq/policyN/amd_pstate_highest_perf
/sys/devices/system/cpu/cpufreq/policyN/amd_pstate_hw_prefcore
/sys/devices/system/cpu/cpufreq/policyN/amd_pstate_lowest_nonlinear_freq
/sys/devices/system/cpu/cpufreq/policyN/amd_pstate_max_freq
/sys/devices/system/cpu/cpufreq/policyN/amd_pstate_prefcore_ranking
/sys/devices/system/cpu/cpuN/acpi_cppc

$ sudo dmesg -H
[… ominous silence about amd_pstate …]

$ cpupower frequency-info
analyzing CPU 1:
  driver: amd-pstate-epp
  CPUs which run at the same hardware frequency: 1
  CPUs which need to have their frequency coordinated by software: 1
  maximum transition latency:  Cannot determine or is not supported.
  hardware limits: 400 MHz - 4.31 GHz
  available cpufreq governors: performance powersave
  current policy: frequency should be within 2.38 GHz and 4.31 GHz.
                  The governor "powersave" may decide which speed to use
                  within this range.
  current CPU frequency: Unable to call hardware
  current CPU frequency: 3.57 GHz (asserted by call to kernel)
  boost state support:
    Supported: yes
    Active: yes
    AMD PSTATE Highest Performance: 255. Maximum Frequency: 4.31 GHz.
    AMD PSTATE Nominal Performance: 219. Nominal Frequency: 3.70 GHz.
    AMD PSTATE Lowest Non-linear Performance: 141. Lowest Non-linear Frequency: 2.38 GHz.
    AMD PSTATE Lowest Performance: 24. Lowest Frequency: 400 MHz.

$ powerprofilesctl
  performance:
    CpuDriver:	amd_pstate
    Degraded:   no

,* balanced:
    CpuDriver:	amd_pstate
    PlatformDriver:	placeholder

  power-saver:
    CpuDriver:	amd_pstate
    PlatformDriver:	placeholder
#+end_example

And lo, the 🍃↔🚀 slider appears in the Power Management tray widget.

Nervous about entering the "Overclocking" UEFI zone tho, and concerned
about these "Maximum frequencies".

/And does it even help with the game?/

🥁

No.  No it does not; no discernible difference in FPS nor vibes.

Will assume this new baseline cannot hurt - OT1H "overclocking" is
scary, OTOH Linux now has a finer handle on the CPU and hopefully will
not overwork it to death?
*** Sᴇᴠᴇʀᴀʟ Wᴇᴇᴋꜱ Lᴀᴛᴇʀ
- [[https://www.gamingonlinux.com/forum/topic/5475/page=1/][ridge reports]] "bad frame pacing on ADMGPU",
  - when vsync is turned off: a non-factor in my testing,
  - lots of useful information in that thread tho and
    interesting-sounding pointers,
  - [[https://www.gamingonlinux.com/forum/topic/5475/page=2/#r42519][Shmerl]] says:
    - games can cause stutter by underloading the GPU, causing it to
      drop out of "high performance mode",
      - (=amdgpu_top= and =radeontop= do confirm that lag spikes
        correlate with GPU usage drop)
    - see [[https://gitlab.freedesktop.org/drm/amd/-/issues/1500][drm/amd#1500]]:
      - /lots/ of sysfs noodling there; unfortunately, none of the
        suggested settings for =power_dpm_force_performance_level= &
        =pp_power_profile_mode= change the symptoms.

- In [[https://gitlab.freedesktop.org/drm/amd/-/issues/3618#note_2689087][this drm/amd#3618 thread]], @agd5f suggests "6.11 stable kernels"
  include a fix for the issue at hand there and a further rework "was
  submitted to 6.13"; @mattipulkkinen reports happy results with
  6.13-rc2 (FTR, symptoms persist here with 6.12.8).

- Piggybacked onto [[https://gitlab.freedesktop.org/mesa/mesa/-/issues/11300][mesa/mesa#11300]]:
  - common: Hades Ⅱ, iGPU, recent kernel & Mesa, Proton Experimental,
  - differences: Fedora, GNOME, X11,
  - noteworthy: good performance on Windows,
  - suggestion by @Venemo: downgrade & bisect Mesa;
    - tempting, though scared of bricking graphical sessions and/or
      ending up with a frankensystem (intalling binaries under a
      prefix is probably easy, but then keeping track of config tweaks
      and cache artifacts sounds fraught).

- In [[https://gitlab.freedesktop.org/upower/power-profiles-daemon/-/issues/164][upower/power-profiles-daemon#164]], @Nyan reports problematic iGPU
  capping; not convinced this is applicable though, given the reported
  symptoms (video playback is fine here).

- Seen reports of Variable Refresh Rate causing problems:
  - searched high and low to understand why VRR appears nowhere in
    Plasma settings, despite the start menu turning up "Display
    Configuration" when searching for "VRR",
  - mystery solved by ~kscreen-doctor -o~: =Vrr: incapable= 🤷

- [[https://www.techpowerup.com/forums/threads/what-fixed-stuttering-and-random-framerate-spikes-in-games-for-me.327264/][aska33j proclaims]] that /disabling CPPC/ "fixed stuttering and random
  framerate spikes in games for [them]" so…  roundtrip to UEFI,
  disabling that.  The =amd_pstate= warning is back; the "Power
  Profile" slider is no longer accessible in the systray widget; no
  discernible effect in-game anyway.

- Looking at Steam forums, [[https://steamcommunity.com/app/1145350/discussions/1/596260472619121965/][some folks]] do report FPS drops /shortly
  after the update/:
  #+begin_quote
  it started fine after the major update, now suddenly im stuck with 40~50 fps with micro sutters
  — December 6 2024
  #+end_quote

- After AMD drivers & Mesa, figured I could look at vkd3d's issue
  tracker.  [[https://github.com/doitsujin/dxvk/issues/4436][doitsujin/dxvk#4436]] and
  [[ValveSoftware/steam-for-linux#11446]] looked somewhat promising:
  reports of lag on "KDE Tumbleweed Wayland", reported not long before
  my symptoms began (November 2024)); alas, ~LD_PRELOAD=~ does not
  help.
  -
    #+begin_quote
    Alternatively, remove the offending line in =/usr/share/drirc.d/00-radv-defaults.conf=
    #+end_quote

    /discovers [[https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/util/00-radv-defaults.conf][=/usr/share/drirc.d/=]]/

    Computers were a mistake.

- Peeked at [[https://github.com/HansKristian-Work/vkd3d-proton/blob/master/.github/ISSUE_TEMPLATE/bug_report.md][vkd3d-proton's issue template]] and idly ran with
  ~PROTON_LOG=1~.  Over the course of 30 seconds or so, the log file
  gets flooded with 3MB's worth of =trace:unwind:dump_unwind_info= 🤨
*** This is insane
Selected subset of moving parts; "testability" considering ease of
clean reverts:

| Part         | Testability                                                                         |
|--------------+-------------------------------------------------------------------------------------|
| Linux kernel | 🫣 [[https://en.opensuse.org/SDB:InstallNewerKernel][some distro documentation]]; afraid of side-effects                                |
| AMD drivers  | 🤷 no clue; maybe inextricable from kernel?                                         |
| Mesa         | 😬 easy to recompile; hard to control transient state in cache & config folders     |
| Steam        | 🫥 under Steam's control                                                            |
| Wine         | 🫥 under Steam's control                                                            |
| Proton       | 👌 as long as I stick to versions under Steam's control; have not considered GE yet |
| vkd3d-proton | 🫥 under Steam's control                                                            |
| Hades Ⅱ     | 🫥 under Steam's control                                                            |

That's looking at software packages as individual blackboxes;
config-wise, worth noting:

| Part       | Testability       |
|------------+-------------------|
| AMD pstate | 😬 UEFI roundtrip |
| sysfs      | OK                |

Let's throw in:

| Part          | Testability                       |
|---------------+-----------------------------------|
| Mobo firmware | 🔥 reports of nuked boot settings |