Bug 92982

I should add that this problem does not occur when booting with `nomodeset`. It then uses the vesa framebuffer instead of radeondrmfb.

Does it hog the CPU while it's hanging? If so, can you get a CPU profile with sysprof or perf or oprofile?

Otherwise, can you try narrowing down where the time is spent in the ioctl, e.g. by adding debugging output?

Created attachment 119928 [details]
perf stat

Created attachment 119929 [details]
perf report

I wrote the following small program:

#include <stropts.h>
#include <linux/kd.h>

void main() {
	ioctl(0, KDSETMODE, KD_GRAPHICS);
	ioctl(0, KDSETMODE, KD_TEXT);
}

and benchmarked this using perf stat and perf report (see attachments). Is this what you were looking for?

That doesn't look like it's CPU bound to me, so you need another approach to isolate where the time is spent.

Wouldn't that be cfb_imageblit as reported by perf report?

`45.27%  test     [kernel.vmlinux]  [k] cfb_imageblit`

If not, can you explain how I can narrow down where the time is spent in the ioctl by adding debugging output? Do I have to compile my own kernel with added printfs?

I did some logging using drm.debug=14. I switched back and forth between X and the tty, the output is in the attachment. Can you see if something is odd in the output? (maybe the dmesg timestamps are useful?)

Created attachment 120094 [details]
Dmesg output drm.debug=14

(In reply to royvandervegt from comment #7)
> Do I have to compile my own kernel with added printfs?

I'm afraid so.

It turns out the problems are caused by DPM. Without DPM (using radeon.dpm=0) everything works as it should.

I have tried tweaking the DPM parameters using https://wiki.archlinux.org/index.php/ATI#Dynamic_power_management, but changing dpm_state to `performance` and power_dpm_force_performance_level to `high` doesn't solve the issues (in fact, changing these parameters seems to have no effect at all?).

Obviously disabling DPM isn't ideal, so it would be nice if there was a workaround/fix.

Interestingly enough, it seems to have little to do with the clock speeds. 

With dpm turned on and radeon_dpm_force_performance_level set to high (doesn't switch properly):

uvd    vclk: 0 dclk: 0
power level 2    sclk: 75000 mclk: 80000 vddc: 1120 vddci: 0

With radeon.dpm=0 (WORKS, switches quickly):
default engine clock: 750000 kHz
current engine clock: 749980 kHz
default memory clock: 800000 kHz
current memory clock: 799870 kHz
voltage: 1120 mV
PCIE lanes: 16

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/drm/amd/issues/661.