#
✕ nick:kettenis
#asahi-dev · 2025-04-07 22:38
kettenisfaulting even
#asahi-dev · 2025-04-07 22:37
kettenisfaulring address is probably 0x1800170180
#asahi-dev · 2025-04-07 22:37
kettenisthe upper address bits are typically garbage
#asahi-dev · 2025-04-07 22:35
kettenis(s3_3_c15_c8_0 through s3_3_c15_c10_0)
#asahi-dev · 2025-04-07 22:34
kettenisthe l2c_err_xxx registers might provide some hints
#asahi-dev · 2025-04-07 22:20
kettenisjannau: 0xbe000000 >> 26 = 0x2f which is SError
#asahi-dev · 2025-04-07 21:56
kettenisbut IIRC you get an SError for Device-nGnRnE too, so I expect the hardware really only works with Device-nGnRE
#asahi-dev · 2025-04-07 21:54
ketteniswould be interesting to see whether Device-GRE works
#asahi-dev · 2025-04-07 21:52
kettenisat least Apple throws an SError instead of failing to produce the right results in random ways
#asahi-dev · 2025-04-07 21:52
kettenisbut a lot of the interconnects found on arm64 SoCs don't support that
#asahi-dev · 2025-04-07 21:51
kettenisthe only thing that matches really x86 WC is Norma-NC
#asahi-dev · 2025-04-07 21:51
kettenisthis is another variant of the usual problem with GPUs on arm64
#asahi-dev · 2025-04-07 21:50
kettenisbet m1n1 is using device-something
#asahi-dev · 2025-04-07 21:46
kettenisjannau: how are those BARs mapped? Normal-NC or Device-something?
#asahi-dev · 2025-04-04 17:59
kettenisI believe that is the preferred way to power on PCI devices that are soldered onto a system board
#asahi-dev · 2025-04-04 17:58
kettenismaz, robher: there also is CONFIG_PCI_PWRCTL now
#asahi-dev · 2025-03-20 14:12
kettenisopen source OSes (and applications) already have to deal with the frequency varying from machine to machine
#asahi-dev · 2025-03-20 13:54
kettenisbut might as well pick 1 GHz since ARM has declared that's the future
#asahi-dev · 2025-03-20 13:54
kettenisI don't think it really matters as long as we configure things consistent across cores
#asahi-dev · 2025-03-20 13:36
kettenismore like fixed at 24 MHz if < M3, configurable between 24 MHz and 1 GHz when >= M3
#asahi-dev · 2025-03-20 12:51
kettenisbest if that happens first before I try to upstream the u-boot bits
#asahi-dev · 2025-03-20 12:50
ketteniskernel
#asahi-dev · 2025-03-20 12:50
kettenisis upstreaming the dockchannel DT bindings on someone's radar?
#asahi-dev · 2025-03-20 12:27
kettenisjannau: thanks for doing that u-boot test
#asahi-dev · 2025-03-20 12:20
kettenischadmed: I fear that means you need to convert that to YAML first
#asahi-dev · 2025-02-06 12:59
kettenisI don't think so; but I guess I should consult our rust expert to see if there is a way forward
#asahi-dev · 2025-02-06 12:42
kettenisand it suggests using rustup or cargo -Zbuild-std but the latter doesn't work since I'm on the `stable` channel
#asahi-dev · 2025-02-06 12:39
ketteniserror[E0463]: can't find crate for `core`
#asahi-dev · 2025-02-06 12:39
kettenisso yes the aarch64-unknown-none-softfloat target is a problem:
#asahi-dev · 2025-02-06 11:13
kettenisI'll take a look again and come back here if I get stuck ;)
#asahi-dev · 2025-02-06 11:06
kettenisI'm a rust n00b so any help is appreciated ;)
#asahi-dev · 2025-02-06 11:03
kettenishmm, so maybe it is just the build instructions that depend on it?
#asahi-dev · 2025-02-06 10:57
kettenisa way to build m1n1 with the standard rustc compiler would be helpful
#asahi-dev · 2025-02-06 10:57
kettenisanother problem is that there is no rustup for OpenBSD
#asahi-dev · 2024-12-16 16:41
kettenisso going to far back might not make a lot of sense
#asahi-dev · 2024-12-16 16:41
kettenisAlso, note that there is talk about removing ARMv4/ARMv4T/ARMv5/ARMv6 from Linux at some point in the near future
#asahi-dev · 2024-12-16 16:36
kettenisthose early SoCs were much more Samsung-like, so maybe they're compatible with something that is already supported by Linux?
#asahi-dev · 2024-12-16 16:34
ketteniswell, just give them a more specific compatible
#asahi-dev · 2024-10-28 21:22
kettenismaybe; I probably missed this because of vacation followed by too much work travel
#asahi-dev · 2024-10-28 21:03
kettenisat this point it is porbably easiest to just bring back the old lmb code with a different name
#asahi-dev · 2024-10-28 21:02
kettenisyeah, with lots of stuff piled on top of it
#asahi-dev · 2024-10-28 20:57
kettenisjannau: that is different breakage from that dev_get_priv() thing though isn't it?
#asahi-dev · 2024-10-28 20:45
kettenisah, I was afraid it would break things
#asahi-dev · 2024-10-28 20:40
kettenisI can look into that, but it might take me a few days
#asahi-dev · 2024-10-28 20:37
kettenisso maybe implementing .connect() and a new .disconnect() and drop the remove
#asahi-dev · 2024-10-28 20:22
ketteniscould keep track of the mappings and hope the drivers properly remove them
#asahi-dev · 2024-10-28 20:12
kettenismaybe the dart driver needs to return -EPROBE_DEFER if there are still active users
#asahi-dev · 2024-10-28 20:11
kettenisand the iommu references go sideways in the tree
#asahi-dev · 2024-10-28 20:10
kettenisI think the u-boot device model is weaker than the linux version
#asahi-dev · 2024-10-28 20:04
kettenismaybe apple_dart_remove() gets called before xhci_{dwc3|pci}_remove()?

plus ›