#
✕ nick:kettenis
#asahi · 2022-08-01 14:32
kettenisyes, the extlinux.con suggestion uses the legacy bootpath
#asahi · 2022-08-01 14:30
kettenislet me remind folks that the UEFI bootpath is the only one that is fully supported
#asahi · 2022-07-27 21:14
kettenisunlikely to be relevant
#asahi · 2022-07-27 20:44
kettenisthat is just broken
#asahi · 2022-07-27 20:44
kettenisso fedora is shipping xorg compiled with bti/pauth enabled but mesa with bti/pauth disabled?
#asahi · 2022-07-27 19:36
kettenisSIGILL?
#asahi · 2022-07-27 19:07
ketteniswould be good if someone could check it isn't an issue with u-boot
#asahi · 2022-07-27 19:06
kettenisdon't immediately see why efifb wouldn't work with r10b10g10x2
#asahi · 2022-07-27 19:00
kettenisbut even efifb should just work...
#asahi · 2022-07-27 15:34
kettenisI'm unlikely to look into this unless I run into this issue myself
#asahi · 2022-07-27 15:33
kettenisscanning the usb bus seems to take a long time for you
#asahi · 2022-07-27 15:32
kettenisGlanzmann: I suspect this is a particular issue with the u-boot usb code not playing well with one of your usb devices
#asahi · 2022-07-22 11:10
kettenisI believe rpi is whole different level of pain as the PCIe controller has issues with 64-bit transactions on the bus it is connected to
#asahi · 2022-07-22 11:07
kettenisanyay, (sadly) this means that the nnvidia blob drivers have a fair chance on working as an eGPU on M1
#asahi · 2022-07-22 11:04
kettenismaz: the left hand doesn't know what the right hand is doing ;)
#asahi · 2022-07-22 10:54
ketteniswhich suggests that they do map video RAM as DEVICE_GRE (or DEVICE_nGnRE if the kernel doesn't support DEVICE_GRE)
#asahi · 2022-07-22 10:53
kettenishttps://github.com/NVIDIA/open-gpu-kernel-modules/blob/88636e701daad77ae77a0bd861bbb0ac28a31763/kernel-open/nvidia/nv-mmap.c#L325
#asahi · 2022-07-22 10:53
kettenisand
#asahi · 2022-07-22 10:52
kettenishttps://github.com/NVIDIA/open-gpu-kernel-modules/blob/88636e701daad77ae77a0bd861bbb0ac28a31763/kernel-open/common/inc/nv-pgprot.h#L80
#asahi · 2022-07-22 10:52
kettenismarcan: not sure what you mean with HMM, but NVIDIAs "open source" kernel driver has some interesting hints about how they implement write combining on arm64:
#asahi · 2022-07-19 21:10
kettenissorry, but your description of what you did is too vague
#asahi · 2022-07-19 21:05
kettenisthe "unknown" partitions are the various apfs partitions
#asahi · 2022-07-19 21:03
kettenisyes, that looks ok
#asahi · 2022-07-16 12:50
ketteniscool
#asahi · 2022-07-16 12:42
kettenisjannau: speaking about upstreaming, is the dart series now derailed because of M2 support?
#asahi · 2022-07-16 12:26
kettenisthat's going to be interesting, although I don't think there is anything fundamental that prevents us from loading the firmware from the ESP in u-boot
#asahi · 2022-07-16 12:25
kettenishmm, so that means if we want to support tke keyboard in u-boot, we'd have to load it there
#asahi · 2022-07-16 12:22
kettenisjannau: so can't you delay the load until the root filesystem is mounted and load it from there?
#asahi · 2022-07-16 12:21
kettenisjannau: the firmware is only needed for the touchpad isn't it?
#asahi · 2022-07-15 21:22
kettenisprobie_: I think you could copy BOOTAA64.EFI and bsd.rd to the ESP and boot that with some manual intervention
#asahi · 2022-06-19 11:09
kettenisthe answer is pretty much no for linux (either passthrough or native)
#asahi · 2022-06-19 07:55
kettenisso that means that all pci drivers that use ioremap_wc() to map a prefetchable BAR won't work on these machines
#asahi · 2022-06-19 07:26
kettenisbut even then, it is probably not worth it spending time on making eGPUs work
#asahi · 2022-06-19 07:23
kettenis" PCI bars
#asahi · 2022-06-19 07:23
kettenissince G is what you really care about for mapping "prefetchable
#asahi · 2022-06-19 07:23
ketteniswonder wether GRE or GnRE works
#asahi · 2022-06-19 07:22
ketteniscompatible with x86 ;)
#asahi · 2022-06-19 07:21
ketteniswith nGnRnE you simply don't get the benefits of posting at the CPU level
#asahi · 2022-06-19 07:04
kettenisit'd be interesting to see what an Apple Silicon Mac Pro looks like and whether Apple is going to support PCIe GPUs in those
#asahi · 2022-06-19 07:02
ketteniswhich defenitely uses memcpy and assumes that non-aligned access works
#asahi · 2022-06-19 07:01
kettenisand usually these discussions happen in the context of the open-source (Mesa) graphics stack
#asahi · 2022-06-19 07:01
kettenisah, well, KVM throws in a whole other complication
#asahi · 2022-06-19 06:58
kettenisI'm not familliar enough with the various graphics APIs, but at least some of the OpenGL functionality to directly map stuff like that from VRAM is an extension that drivers don't have to implement
#asahi · 2022-06-19 06:54
kettenisbut yeah, not really a GPU
#asahi · 2022-06-19 06:53
kettenisI was thinking of displaylink as an example of something that doesn't expose a VRAM framebuffer
#asahi · 2022-06-19 06:51
kettenisthe usb ones certainly don't ;)
#asahi · 2022-06-19 06:50
kettenisor is that an Intel/AMD extension?
#asahi · 2022-06-19 06:50
ketteniswell, does OpenGL/Vulkan/CUDA allow mapping VRAM directly into userland
#asahi · 2022-06-19 06:48
kettenis(I mean unaligned accesses to VRAM from userland)
#asahi · 2022-06-19 06:47
kettenisand presumably NVIDIA's userland code doesn't do this

plus ›