Bug 194966
| Summary: | Radeon KMS driver lack power and clocks management support | ||||||
|---|---|---|---|---|---|---|---|
| Product: | Base System | Reporter: | zuborg | ||||
| Component: | kern | Assignee: | freebsd-bugs (Nobody) <bugs> | ||||
| Status: | Closed Overcome By Events | ||||||
| Severity: | Affects Many People | CC: | dumbbell, emaste, james122333, sasamotikomi, uzsolt, x11, zeising | ||||
| Priority: | --- | ||||||
| Version: | 10.1-STABLE | ||||||
| Hardware: | Any | ||||||
| OS: | Any | ||||||
| Attachments: |
|
||||||
Updated 10.1-BETA and 10.1-RC versioned bugs to 10.1-STABLE. Can you verify if this problem persists on 10.2-PRERELEASE? (In reply to Glen Barber from comment #2) Yes, the problem with overheating still persist in 10.2-PRERELEASE r285308. Last one power management patch: https://people.freebsd.org/~avg/radeon-pm-profile.diff Discussion: https://lists.freebsd.org/pipermail/freebsd-x11/2016-March/017355.html a10-5700(apu) not work(not show real temperature:1 degree)... Is there any news about this issue? Is this an issue with an updated driver from drm-kmod-next? (In reply to Niclas Zeising from comment #7) I can't test it because my old laptop (with radeon card) went awry and I bought a new one (with intel). Closing this, since the original hardware is not around any more, and it was a long time ago. Please re-open this or create a new PR if the issue occurs again. (In reply to Niclas Zeising from comment #9) My PC was broken before...so I rebought it few days ago... Is there a "dynpm" feature(radeon) in drm-kmod-next? If so...I'll give it a try later... |
Created attachment 149319 [details] Patch /sys/dev/drm2/. radeonkms driver change device sclk and mclk when initializing according to drm.force_sclk and drm.force_mclk loader.conf tuneables Radeon KMS driver initialize hardware in default 'performance' mode, which cause significant noise and power consumption. Following patch adds rudimentary support of engine/memory clocks management. Immediately after initialization driver looks at value of drm.force_sclk and drm.force_mclk variables (tuneable via loader.conf). If non-zero value specified driver call radeon_set_engine_clock()/radeon_set_memory_clock() to adjust clocks setting. Boundary checks are not performed, it should be used with care! Actually it would be significantly better to add direct support of clocks/voltage values by `sysctl` on the fly, not only via loader.conf as in proposed patch. Patch was tested, values (in MHz) drm.force_sclk=250 drm.force_mclk=250 reduce noise of card fan very noticeable.