Janne Grunau / is on Nostr

#asahilinux r/e of apple silicon devices, video codecs and arm assembler/simd optimisazions for #ffmpeg, #x264 and #dav1d

Last Notes

2024-07-16 04:40:07

- reply

using GL.host with GL.default installed works now / for me

2024-07-15 23:19:40

- reply

would that problem go away if we used GL.host as extension and get rid of the override env variable?

2024-07-15 20:32:50

- reply

there are two reason why this it not the preferred solution from our side. This is only temporarily necessary. Merging the driver / uapi takes longer than anticipated/wished but eventually it will be merged.
The DRM linux maintainers are uneasy about the use of unstable DRM uapi in the wider eco system. That’s the reason why the driver in upstream mesa is intentionally broken. A local package in distributions which already ship the driver in the base system is a much smaller footprint.

2024-07-09 18:20:12

- reply

Mark and me were CC-ed and Mark brought issues with multiple ESPs up but we failed to follow up on it to see it resolved.

2024-07-09 13:36:20

- reply

there is https://github.com/rhboot/efivar/pull/267
I initially just checked efibootmgr as obvious user space tool

2024-07-09 12:45:51

- reply

u-boot 2024.07 supports memory / disk backed EFI variables with runtime write support: https://lore.kernel.org/u-boot/20240418125456.50944-1-ilias.apalodimas@linaro.org/
I’m not sure if the userspace component to sync the mem stored variables to u-boot’s file storage already exists

2024-07-09 12:12:25

- reply

I would be also wary of adding nvram write support to u-boot a hand over control of the mtd spi device to u-boot while running a full OS beneath. Something like that would be required for runtime writable EFI variables.
I don’t think we necessarily need separate variables for each install (requires a mapping from ESP to apfs container). u-boot could update the boot container and reboot based on BOOT* entry.
Overall not a solution I would like to rely on

2024-07-09 09:01:39

- reply

we have nvram read/write support in linux but there is some concern about breaking iBoot by adding custom variables. Adding certain variables (iirc “autoboot=0”) which were supported on intel macs made apple silicon devices unbootable and required a dfu revive.

2024-07-09 06:13:20

- reply

one solution would be to prevent bringup of complex devices before kexec. i.e. if we only bring-up the same hardware as u-boot it should be possible to kexec.
Not that this is a great solution as the only way to prevent device init is to make ensure modules are only present after boot selection. It doesn’t fit into the appeal of the proposal (“just use linux”) but it shouldn’t inttoduce functional regressions.
We still would have u-boot as bootloader and the lack of runtime setable EFI variables will probably cause problems as well.

2024-06-17 06:20:33

- reply

but only for installation, booting from it later is fine.
I can’t think of many explanations for that. Maybe they use the dfu port internally for provision/activation of new installations. Strange that that would leak outside though.

2024-05-22 11:43:51

- reply

I wonder if this is the reason for US_FLAG_IGNORE_UAS (at least for some chipsets/devices)

2024-05-02 06:54:09

- reply

120Hz will be at least from my side somewhat complicated as I would like to add it only when the driver supports VRR. The latter could be uncomplicated or a nightmare due to design mismatches between drm/kms and dcp firmware (I haven’t looked yet)

2024-04-15 21:28:24

- reply

the mismatched timer freq nonsense was a mismatched chicken bit between performance and efficiency cores. The responsible bit has been identified. 24 MHz is still the native frequency and 1GHz is only upscaled

2024-04-15 20:45:51

- reply

I doubt M2 laptops will ever get dual external display support with macOS. I’m not surprised that M3 Macbook Airs don’t advertise Thunderbolt 4. Adding it would make the tech specs more complicated (shared with M2 Air) and I think it’s not unreasonable to say that dual display out means in every situation and just when the lid is closed. I wouldn’t have wanted to discuss this with Intel for certification.
Resolution requirements are 4k@60 so not a problem for dcp.