#asahi · 2022-03-09 19:31
arndj`ey: ok, that's why I didn't see that commit in mainline ;-)#asahi · 2022-03-09 19:31
arndah right, you can't have it both ways like this#asahi · 2022-03-09 19:30
arndthere should also be handover from efi_earlycon to efifb, but I assume nobody here is actually using efi_earlycon (despite it being really useful for debugging problems in early boot)#asahi · 2022-03-09 19:29
arnd I can imagine that there is no handover between efifb and simpledrm because both expect to be the early drivers, not the real ones#asahi · 2022-03-09 19:28
arndI think there should also be some handover code, which ensures that you can boot with a built-in simpledrm or simplefb or efifb but later load the real driver that then owns the device after the simple drivers give it up.#asahi · 2022-03-09 19:26
arndI would expect it to treat the backlight as optional: if no driver is found, it would continue without backlight#asahi · 2022-03-09 19:26
arndGlanzmann: if the lack of the backlight-gpio driver causes the probe to fail, that might be a bug in the simpledrm driver, but I don't see where it references the backlight at all#asahi · 2022-03-09 19:25
arndGlanzmann: I think that message is the opposite: efi support is in the kernel but not the boot loader#asahi · 2022-03-09 19:20
arndGlanzmann: what happens when the gpio backlight driver is not there? No backlight or probe failure?#asahi · 2022-03-09 18:27
arndmaz: ok, I see it now, the 2-3 weeks delivery is only for 32GB RAM, which is not available with the Ultra#asahi · 2022-03-09 18:25
arndmaz: oh, it still works in apple.com/uk/: 2-3 weeks vs 8-9 weeks#asahi · 2022-03-09 18:24
arnd(in apple.com/de/)#asahi · 2022-03-09 18:23
arndbut now it seems to all be up to eight weeks#asahi · 2022-03-09 18:23
arndmaz: yes, you have to pick the 10-core model first, and then upgrade to the larger CPU but keep the small SSD, jannau first noticed that doing this also gives you four week delivery instead of eight weeks for identical hardware config#asahi · 2022-03-09 18:16
arndmaz: 20 cores, 48GPU/64RAM/512SSD#asahi · 2022-03-09 18:04
arndjannau: I ordered mine now with expected delivery Apr 6-13, but now the product page says May 6-13 regardless of the model. Either they the order system it, or enough people did the same thing until the dates lined up#asahi · 2022-03-09 14:16
arndso you need to manually enable CONFIG_FB=y in order to use the FBDEV emulation#asahi · 2022-03-09 14:15
arndmps: the problem that a lot of users reported with CONFIG_FB is that DRM_FBDEV_EMULATION no longer forces CONFIG_FB=y, but instead gets turned off when you have DRM_KMS_HELPER=y and FB!=y#asahi · 2022-03-09 07:54
arndjannau: good to know. I don't expect it to actually make a difference. It seems another difference is that doing it this way you can pick a 512GB SSD, while the other one starts at 1TB for an extra €230. For the difference, I would probably get a 2TB m.2 drive and an external TB enclosure.#asahi · 2022-03-01 12:22
arndproper SSDs tend to be fast enough with any file system that you wouldn't notice the downsides of f2fs unless you do real measurements, for the low-end stuff, the benefits of f2fs should be very noticeable#asahi · 2022-03-01 12:21
arndI would expect the overhead of doing double wear leveling and GC between f2fs and the SSD to cost more than it can possibly gain#asahi · 2022-03-01 12:20
arndmps: ok, that's fine, it definitely helps on SD cards and USB sticks as well as lower-end eMMC, not so sure about proper SSDs (regardless of NVMe/SATA/USB attachment)#asahi · 2022-03-01 12:19
arndand later would have new enough cores then#asahi · 2022-03-01 12:19
arndmt8192#asahi · 2022-03-01 12:18
arndj`ey: oh, you are right, sorry#asahi · 2022-03-01 12:18
arndmps: supporting larger page sizes in f2fs definitely makes sense, as it already optimizes for writing 64KB+ sized chunks. Not sure why you care though, do you have a storage device that benefits from using f2fs over ext4?#asahi · 2022-03-01 12:16
arndrk35xx and mt8183 do support 16KB pages and >4GB RAM#asahi · 2022-03-01 12:13
arndmps: both are Cortex-A72, but are effectively limited to 4GB, so you definitely don't want 64KB pages#asahi · 2022-03-01 12:10
arnd(technically also A64fx wants 64K pages, but I think if you have one of those, you want a custom kernel anyway)#asahi · 2022-03-01 12:08
arndeverything later can do 16K pages anyway#asahi · 2022-03-01 12:08
arndI think there are very few systems with a Cortex-A57 or A72 that have enough RAM to benefit from 64K pages, and no A53 ones#asahi · 2022-03-01 12:06
arndso presumably showing some actual numbers may make them change it#asahi · 2022-03-01 12:06
arndThe main downside of the 64K kernel what happens when you get into low memory conditions and start swapping, it pretty much becomes unusable then, 16K pages are much closer to the behavior you get with 4K, but give most of the same benefits#asahi · 2022-03-01 12:05
arndit looks like the testing done on Ubuntu was rather minimal: https://lists.ubuntu.com/archives/kernel-team/2020-September/113596.html#asahi · 2022-03-01 12:01
arndmps: ok, good (in general, not necessarily for asahi)#asahi · 2022-03-01 12:01
arndUbuntu already ships both 4K and 64K kernels, it's possible that they could be convinced to either add a third one, or replace the 64K one, as there are only very few machines that benefit from larger pages but can't do 16K#asahi · 2022-03-01 11:59
arndFedora and Debian are traditionally trying to limit the number of kernels as much as possible, Armbian and Arch OTOH seem to be fine with shipping board specific kernels#asahi · 2021-12-23 20:51
arndj`ey: you'd do this for drivers that are absolutely required to boot the platform, but can be compile-tested#asahi · 2021-12-23 20:50
arnda) is 'tristate "some driver" if COMPILE_TEST && !ARCH_APPLE; depends on THIS_SUBSYSTEM; depends on ARCH_APPLE || COMPILE_TEST ; default ARCH_APPLE'#asahi · 2021-12-23 20:49
arndj`ey: actually I just got that wrong myself, I mixed up the case where the driver should be hidden but enabled and the case where it should be visible#asahi · 2021-12-23 20:48
arndj`ey it depends a bit on the subsystem maintainer. I generally like the 'tristate "some driver" if ARCH_APPLE ; depends on ARCH_APPLE || COMPILE_TEST ; default ARCH_APPLE' logic, but it's a bit wordy#asahi · 2021-11-04 11:47
arndamw: the corellium spi driver seems fine for the most part, someone could just submit that. This driver is fairly independent of everything else, so there are not a lot of things one could do differently. The main missing bits are documentation for the DT binding, and properly hooking it up to the clk and pinctrl drivers. In the corellium dts the spi controller links to two nodes handled by their clk driver, but that is not upstream#asahi · 2021-10-27 12:44
arnd"b1111358e1e8 ARM: dts: add SKOV imx6q and imx6dl based boards" has nine #asahi · 2021-10-27 12:43
arndI think sfr pointed me to a script before, but I never integrated that#asahi · 2021-10-27 12:42
arnd"git log --no-merges" makes it a bit faster#asahi · 2021-10-27 12:41
arnd863f302d181f has 16 S-o-B#asahi · 2021-10-27 12:40
arndj_ey: git log --oneline | while read a b ; do echo `git show -s $a | grep -c Signed-off-by ` $a $b | grep -vw '^[0123]' ; done | sort -n#asahi · 2021-10-27 12:26
arndwhich means that marcan forwarded a patch by j_ey to arnd, who applied it to the soc tree with the earlier ack still being meaningful#asahi · 2021-10-27 12:25
arndSigned-off-by: arnd#asahi · 2021-10-27 12:25
arndSigned-off-by: marcan