On Thu 03 Sep 2026 at 11:21:03 -0700, Jeff Frasca wrote: > On Thu, Jul 2, 2026 at 10:25?AM Rhialto <rhialto%falu.nl@localhost> wrote: > > > [ 4022.877] (II) RADEON: Driver for ATI/AMD Radeon chipsets: > > > > I am curious if you also get the corruptions with the "radeon" driver. I > > had to switch to the integrated Intel graphics, but that lacks support > > for whatever x11/redshift does, and also I can't use mpv in gpu mode, > > but I have to use "vo=xv" in ~/.config/mpv/mpv.con. Likewise, wip/golly > > doesn't display its life world. So I'd rather switch back. > > I know I'm doing a bit of thread necromancy here (I missed this 2 > months ago because of personal life stuff). > > What does the corruption look like? Is it random grey bars on the > screen that are approximately 16px wide and 1px tall? For me the corruption is different. It shows up usually only after a few weeks of uptime, and it affects fonts. I suspect that there is a bunch of font glyps stored in video memory, and it looks like they get partially overwritten. It mostly affects bitmap fonts. The text in my xterms gets mostly unreadable. (It seems like the data that overwrites the fonts is more zeros than ones, since mostly font pixels disappear). Programs like Firefox which use scalable fonts are usually not affected when this happens. So I typically notice when I switch from a workspace to another with xterms. I have to add that since I wrote the quote above, I could use mpv with gpu accelleration again in 10.0. I suspect some update of MesaLib did this. It still works in 11.0 which I have installed now. On -11, I have seen (different) corruptions in some cases, where the kernel logs that the GPU is stuck. This may be the same root cause, such as "incorrect memory is being overwritten", or it may be different. - Firefox, which now uses accellerated WebGL if you let it. The GL Aquarium triggered it after some time and fiddling with the settings on the page: https://webglsamples.org/aquarium/aquarium.html - print/foliate when I opened several windows - multimedia/handbrake These last two programs have a gtk4 GUI, so I'm wary of gtk4 now. I am now running xcompmgr (although that has some bugs of its own; I guess I should try x11/picom which is mentioned in https://wiki.netbsd.org/laptops/) and I have not tried the gtk4 programs again. (I should try it on my testing machine; possibly using a compositor changes the usage of accelleration enough that it doesn't trigger the bug) > Oddly enough, when I had my machine setup to run NetBSD under QEMU > with PCI passthrough for the 5450, it worked flawlessly, so there's > something in there with how the video card memory is getting mapped, > and I suspect something is writing some 64-byte structure over memory > that should only be getting used for the GPU, but moving the location > of the card on the PCI bus fixes it. "The wrong memory being overwritten" is not inconsistent with my own suspicions. I wonder if there is some way to get a similar change in PCI usage but without running my whole session in Qemu. -Olaf. -- ___ Olaf 'Rhialto' Seibert <rhialto/at/falu.nl> \X/ There is no AI. There is just someone else's work. --I. Rose
Attachment:
signature.asc
Description: PGP signature