07:58 curro: karolherbst: ah. it would certainly be reasonable to do that lowering of compute calls too large for the hardware within the state tracker. but i don't see how that means that the global offset handling is broken in clover. it just doesn't attempt to lower such calls currently, under the assumption that the driver can do it transparently if needed
09:54 dschuermann: does someone know off hand which optimization removes deref_cast(undef)?
09:58 dschuermann: or rather tex using that
11:15 olesalscheider: I got it to boot. It seems that the CPU supports encrypted memory (and I had enabled that in the kernel config). But the iGPU does not and so there was nothing to see on the screen.
11:15 olesalscheider: Sorry, wrong channel.
15:18 DPA: Only the first device which opens a /dev/dri/cardX can call drmSetMaster to become drm master, right?
15:18 DPA: Is there a way to open such a device without preventing other devices from becoming drm master?
15:27 pinchartl: DPA: you can call drmDropMaster() to drop master permissions, allowing something else to open the device and get master permissions
15:28 DPA: Ok, thanks!
15:30 pinchartl: DPA: beside this, there's a lease mechanism that allows giving permissions on part of the device to other processes
16:27 DPA: Is there anything a render node can do that a regular card can't after dropping drm master?
16:27 DPA: If I drop it here: https://projects.dpa.li/git/?p=xserver.git;a=blob;f=hw/xfree86/drivers/modesetting/driver.c;h=58a7b7e489e4cd114cd9701e0ff4f80326d80ee2;hb=bf63e3d4874f8b7c7aba0d2bd6d5a1045ec07c0b#l973
16:27 DPA: Then with the etnaviv render node, glxgears works, but using the etnaviv card, I get:
16:27 DPA: > libGL error: failed to authenticate magic 1
16:27 DPA: > libGL error: failed to load driver: etnaviv
16:27 DPA: And then it falls back to llvmpipe.
16:30 DPA: But only if I drop drm master.