4.14.0-3 kernel dropped my CPU speed?

[sej7278]

Just upgraded the kernel to 4.14.0-3 in Sid and noticed "grep MHz /proc/cpuinfo" and lscpu are showing various values from 1GHz to 3.3GHz depending on load, as you'd expect on a laptop using cpufreq governors or somesuch.

The thing is, this is a desktop Xeon X3470 2.9GHz overclocked to 3.3GHz with TurboBoost disabled, its locked to 3.3GHz all the time in the BIOS, hell the CPU doesn't even support thermal monitoring!

4.14.0-2 doesn't show this behaviour, nor do any earlier kernels I've ever used.

I noticed there's some stuff about thermals and governors in dmesg, and turbostat shows the correct TSC_MHZ and also reports no such file or directory for the cpufreq driver:

Code: Select all

# dmesg |grep -i cpu
[    0.000000] smpboot: Allowing 8 CPUs, 0 hotplug CPUs
[    0.000000] setup_percpu: NR_CPUS:512 nr_cpumask_bits:512 nr_cpu_ids:8 nr_node_ids:1
[    0.000000] percpu: Embedded 44 pages/cpu @ffff88151fc00000 s142232 r8192 d29800 u262144
[    0.000000] pcpu-alloc: s142232 r8192 d29800 u262144 alloc=1*2097152
[    0.000000] pcpu-alloc: [0] 0 1 2 3 4 5 6 7 
[    0.000000] 	RCU restricting CPUs from NR_CPUS=512 to nr_cpu_ids=8.
[    0.000000] RCU: Adjusting geometry for rcu_fanout_leaf=16, nr_cpu_ids=8
[    0.013560] CPU: Physical Processor ID: 0
[    0.013560] CPU: Processor Core ID: 0
[    0.013566] mce: CPU supports 9 MCE banks
[    0.013573] CPU0: Thermal monitoring enabled (TM1)
[    0.060000] smpboot: CPU0: Intel(R) Xeon(R) CPU           X3470  @ 2.93GHz (family: 0x6, model: 0x1e, stepping: 0x5)
[    0.060000] core: CPU erratum AAJ80 worked around
[    0.060000] core: CPUID marked event: 'bus cycles' unavailable
[    0.060000] smp: Bringing up secondary CPUs ...
[    0.060000] .... node  #0, CPUs:      #1
[    0.074565] smp: Brought up 1 node, 8 CPUs
[    0.077156] cpuidle: using governor ladder
[    0.077156] cpuidle: using governor menu
[    0.128059] HPET: 8 timers in total, 5 timers will be used for per-cpu timer
[    0.871529] ledtrig-cpu: registered to indicate activity on CPUs

# turbostat 
turbostat version 17.06.23 - Len Brown <lenb@kernel.org>
CPUID(0): GenuineIntel 11 CPUID levels; family:model:stepping 0x6:1e:5 (6:30:5)
CPUID(1): SSE3 MONITOR SMX EIST TM2 TSC MSR ACPI-TM TM
CPUID(6): APERF, No-TURBO, DTS, No-PTM, No-HWP, No-HWPnotify, No-HWPwindow, No-HWPepp, No-HWPpkg, No-EPB
cpu3: MSR_IA32_MISC_ENABLE: 0x4000850089 (TCC EIST No-MWAIT PREFETCH No-TURBO)
CPUID(7): No-SGX
cpu3: MSR_MISC_PWR_MGMT: 0x00000000 (ENable-EIST_Coordination DISable-EPB DISable-OOB)
cpu3: MSR_PLATFORM_INFO: 0x90040031600
9 * 133.3 = 1200.0 MHz max efficiency frequency
22 * 133.3 = 2933.3 MHz base frequency
cpu3: MSR_IA32_POWER_CTL: 0x00000001 (C1E auto-promotion: DISabled)
cpu3: MSR_TURBO_RATIO_LIMIT: 0x18181a1b
24 * 133.3 = 3199.9 MHz max turbo 4 active cores
24 * 133.3 = 3199.9 MHz max turbo 3 active cores
26 * 133.3 = 3466.6 MHz max turbo 2 active cores
27 * 133.3 = 3599.9 MHz max turbo 1 active cores
cpu3: MSR_PKG_CST_CONFIG_CONTROL: 0x00000403 (UNlocked: pkg-cstate-limit=3: pc6)
cpu3: POLL: CPUIDLE CORE POLL IDLE
cpu3: C1: MWAIT 0x00
cpu3: C1E: MWAIT 0x01
cpu3: C3: MWAIT 0x10
cpu3: C6: MWAIT 0x20
NSFOD /sys/devices/system/cpu/cpu3/cpufreq/scaling_driver
cpu0: MSR_IA32_TEMPERATURE_TARGET: 0x00631000 (99 C)
Core	CPU	Avg_MHz	Busy%	Bzy_MHz	TSC_MHz	IRQ	SMI	C1	C1E	C3	C6	C1%	C1E%	C3%	C6%	CPU%c1	CPU%c3	CPU%c6	CoreTmp	Pkg%pc3	Pkg%pc6
-	-	92	2.90	2824	3300	5896	0	3379	289	483	2655	0.16	0.08	0.63	96.28	3.52	0.69	92.88	33	0.00	0.00
0	0	39	1.25	2816	3300	276	0	0	2	2	182	0.00	0.00	0.02	98.77	1.53	0.15	97.07	32	0.00	0.00
0	4	44	1.38	2824	3300	136	0	0	0	12	177	0.00	0.00	0.16	98.49	1.40
1	1	54	1.70	2804	3300	110	0	0	6	111	472	0.00	0.03	1.16	97.15	5.62	0.44	92.24	33
1	5	130	4.23	2723	3300	4317	0	3367	215	85	448	0.99	0.26	0.44	94.14	3.09
2	2	13	0.50	2291	3300	115	0	0	0	31	296	0.00	0.00	1.17	98.38	12.40	1.63	85.48	33
2	6	386	11.85	2899	3300	193	0	11	21	165	357	0.29	0.05	1.44	86.43	1.04
3	3	46	1.48	2767	3300	430	0	0	0	21	351	0.00	0.00	0.06	98.50	1.21	0.56	96.74	30
3	7	25	0.80	2745	3300	319	0	1	45	56	372	0.00	0.34	0.58	98.36	1.89

But when I run "stress --cpu 4" cpuinfo etc. report 3.2GHz, as if a governor is enabled and it goes to full tilt under load.

Any ideas - is this some new software governor that ignores the hardware's capabilities, and if so how do I disable it?

I checked its not the new microcode, rolling back to 3.20171117.1 from 3.20180108.1 makes no difference, its definitely the kernel.

reportbug doesn't seem to be working, some problem with the website causes it to crash, but reporting a kernel issue to Debian seems a bit pointless anyway.

[sej7278]

kernel parameters seem to fix it, but still annoying the kernel overrides the bios....

Code: Select all

processor.max_cstate=1 intel_idle.max_cstate=1

[rtimai]

As I am running Debian Stretch 9.3 on kernel 4.9.0-5-amd64, I don't share your observations, but you may want to search the web for "Meltdown Spectre Intel CPU vulnerabilities" for which patches have been released in more recent kernels. This is a hardware issue being mitigated by software, namely kernel patches, and a CPU performance penalty has been acknowledged (depending on tasks being performed.) I have not heard of any microcode changes. The vulnerable hardware is widespread, and mitigation appears complex. The WeLiveSecurity site has information in accessible form for consumers. Good luck.

[sej7278]

no its nothing to do with the meltdown patch (which is both kernel and microcode) this is throttling not just a performance drop.

it seems the latest kernel implements bios-overriding power-saving, which could have latency issues but should save on the power bill.