1 |
Thanks Adam, |
2 |
|
3 |
On Monday, 1 July 2019 00:03:21 BST Adam Carter wrote: |
4 |
|
5 |
> IIRC boost states work by boosting a single core when the other cores are |
6 |
> idle, so you wouldn't expect it to kick in on a -jN emerge. |
7 |
|
8 |
I have observed on an Intel i7 with a schedutil governor, cores kicking in |
9 |
individually while compiling or running something like cpuburn, while the rest |
10 |
of the cores remain above idle. I understand these days boost states are |
11 |
regularly used whenever the thermal budget of the CPU allows and the CPU load |
12 |
demands it. |
13 |
|
14 |
|
15 |
> You could try |
16 |
> closing all other programs and temporarily running -j1 as an experiment to |
17 |
> see if it kicks in at all. If it doesn't work then i'd look at BIOS |
18 |
> updates. Have you included the microcode in the kernel? |
19 |
|
20 |
I've tried both single and multiple jobs loading as many as 8 threads, |
21 |
switching between performance, userspace, schedutil governors. As far as / |
22 |
proc/cpuinfo or 'cpupower monitor' indicated, the frequency never switched |
23 |
into boost and the 'cpupower frequency-info' did not show the boost state as |
24 |
active. This is in contrast to the minimal-CD, which revs all the way up to |
25 |
3500MHz on all cores and shows the boost state as active. |
26 |
|
27 |
The laptop has been flashed with the latest UEFI firmware available by the |
28 |
OEM. I am also loading the microcode in the kernel, with: |
29 |
|
30 |
CONFIG_EXTRA_FIRMWARE="amd-ucode/microcode_amd_fam15h.bin ....." |
31 |
|
32 |
dmesg shows the following: |
33 |
|
34 |
$ dmesg | grep -i micro |
35 |
[ 0.622453] [drm] Loading ARUBA Microcode |
36 |
[ 5.759940] [drm] Loading hainan Microcode |
37 |
[ 6.713011] microcode: CPU0: patch_level=0x06001119 |
38 |
[ 6.714206] microcode: CPU1: patch_level=0x06001119 |
39 |
[ 6.715311] microcode: CPU2: patch_level=0x06001119 |
40 |
[ 6.716330] microcode: CPU3: patch_level=0x06001119 |
41 |
[ 6.717333] microcode: Microcode Update Driver: v2.2. |
42 |
|
43 |
You'll notice it doesn't show "microcode: updated early to new patch_level" |
44 |
unlike what some older Intel systems of mine show right at the start of dmesg; |
45 |
mind you an AMD Kaveri system I'm running is the same. Perhaps there is no |
46 |
later microcode for these CPU-APUs. |
47 |
|
48 |
Here is what I suspect. The minimal-CD is not loading any radeon firwmare, or |
49 |
running Xorg. It may be radeon is interfering with the CPU, holding back from |
50 |
boost states in case the APU graphics needs it later on. I could be a bug - |
51 |
I'm not sure. I would have thought since this an old CPU, by now it should be |
52 |
working without any tweaks and any bugs would have been ironed out in the |
53 |
kernel. :-/ |
54 |
|
55 |
I attach the diff between the minimal-CD kernel config and mine in case anyone |
56 |
can spot something in there around the options under CPU_FREQ and scaling |
57 |
drivers. |
58 |
-- |
59 |
Regards, |
60 |
|
61 |
Mick |