#asahi-dev · 2022-03-26 07:16
jannauI started investigating numa annotations would make sense the t6002. at least for sequential reads it doesn't seem to be the case#asahi-dev · 2022-03-26 07:13
jannaumarcan: do you still have your m1n1 memory benchmarks? it looks like you never pushed them#asahi-dev · 2022-03-25 19:04
jannauah, the powerdomains for the second die are broken, I wonder why linux was working#asahi-dev · 2022-03-25 18:57
jannauu-boot doesn't like the devices on the second die, looks like it fails to power them on#asahi-dev · 2022-03-25 18:09
jannaunvme and the two front facing usb-c ports work as well without kernel code changes#asahi-dev · 2022-03-25 12:05
jannauthat is afaik the only place where the part numners are used#asahi-dev · 2022-03-25 12:05
jannaumaz: same model as the m1 max. the m1 pro uses 0x24/0x25 init for the high performnce cores differs. see https://github.com/AsahiLinux/m1n1/blob/main/src/chickens.c#asahi-dev · 2022-03-25 12:00
jannaumarcan will probably redo the devicetree anyway#asahi-dev · 2022-03-25 12:00
jannaumaz: thanks, fixed#asahi-dev · 2022-03-24 21:27
jannauhe mentioned that it out of scope here and I agree#asahi-dev · 2022-03-24 21:13
jannauapple claims it is ASM3142 though so most likely the same part and firmware problem as on the imac#asahi-dev · 2022-03-24 21:12
jannaurkjnsn: yes, according to the template ADT the m1 max mac studio has the same pcie usb controller#asahi-dev · 2022-03-24 18:43
jannaudistributes IO over the whole chip area, maybe the interposer is in the way#asahi-dev · 2022-03-24 18:36
jannauyes, minimal effort since the cpu looks mostly just like two m1 max' and good preparation from marcan#asahi-dev · 2022-03-24 18:29
jannauhttps://github.com/jannau/linux/commit/c8dbd1edf6af164d18e1e2bab2d93cb7b1028c8a#asahi-dev · 2022-03-24 18:28
jannauprobably if m1n1 would have been told to ignore the cores on the second die. without the changes it tries to start all cores and fails#asahi-dev · 2022-03-24 18:17
jannauASM3142 on the imac with 4 ports so it might be missing firmware #asahi-dev · 2022-03-24 18:01
jannauI haven't added the ones from the second die yet#asahi-dev · 2022-03-24 18:01
jannauyes, the four dwc3/atcphy ports on the first die work as expected as far as I can tell #asahi-dev · 2022-03-24 17:58
jannauit has a smc power enable gpio in the ADT but adding to dt hasn't changed anything. I assume it's the same controller as in the mini#asahi-dev · 2022-03-24 17:43
jannauyes, roughly evenly distributed over all cores#asahi-dev · 2022-03-24 17:42
jannauI haven't added devices on th second die yet#asahi-dev · 2022-03-24 17:37
jannauI'm a little surprised that it worked out of the box (ignoring simple errors in the devicetree)#asahi-dev · 2022-03-24 17:33
jannauyes, native defconfig#asahi-dev · 2022-03-24 17:29
jannauusing https://gitlab.com/knurd42/kcbench#asahi-dev · 2022-03-24 17:29
jannaucompiles 5.15 defconfig in 78 seconds #asahi-dev · 2022-03-24 17:28
jannaumaz: https://www.jannau.net/2022-03-24_m1_ultra_asahi_dmesg_clean.txt#asahi-dev · 2022-03-24 17:22
jannauM1 Ultr doubles that#asahi-dev · 2022-03-24 17:22
jannauM1 Pro/Max has 1 cluster with 2 efficiency cores and 2 clusters with 4 performance cores each#asahi-dev · 2022-03-24 17:16
jannau+n#asahi-dev · 2022-03-24 17:15
jannau2 efficiency sockets and 4 performace sockets#asahi-dev · 2022-03-24 17:15
jannaulscpu seems to count the clusters as sockets so it has 6 sockets#asahi-dev · 2022-03-24 17:13
jannauyeah: "CPU(s): 20" https://paste.debian.net/1235490/#asahi-dev · 2022-03-24 11:53
jannauchadmed: it should not matter on which die a specific function is. we should have no problem using them on either die#asahi-dev · 2022-03-24 11:48
jannauchadmed: most HW is on die 0, ans/nvme, 2 usb-c/thunderbolt ports and 1 gpio controller is used from die 1#asahi-dev · 2022-03-24 08:07
jannauand 40% of that is just mapping the frame buffer#asahi-dev · 2022-03-24 08:07
jannau 7 files changed, 116 insertions(+), 21 deletions(-)#asahi-dev · 2022-03-24 07:55
jannaustill rough edges and probably incomplete but enough for usb proxy, chainloading and running m1n1 under the HV#asahi-dev · 2022-03-24 07:54
jannaucurrent state of t6002 bringup for m1n1: https://github.com/jannau/m1n1/tree/t6002-bringup#asahi-dev · 2022-03-24 07:01
jannaumakes sense. I'll try to verify that the second die works and wait otherwise. I should work on spi_hid anyway#asahi-dev · 2022-03-24 06:29
jannaumarcan: did you spent more thought of how to avoid the manual duplication? I will probably start with a minimal dts with duplication#asahi-dev · 2022-03-24 06:24
jannauI guess that means I have to write a devicetree tonight#asahi-dev · 2022-03-24 06:24
jannauGlanzmann: FYI linux boots without obvious crashes on M1 Ultra (using the j314c dtb, that means it uses HW of the first die)#asahi-dev · 2022-03-23 18:54
jannauapparently over a genesys logic usb hub#asahi-dev · 2022-03-23 18:29
jannauand then there's ausb-usb3phy with ausb-phy[01]/ausb-drd[01]#asahi-dev · 2022-03-23 18:28
jannauinteresting, there are two additional usb ports in the M1 ultra ADT. 6 ATC type-c/thunderbolt ports, 2 asmedia pcie-based ones (presumably usb type-a as on the mac mini)#asahi-dev · 2022-03-23 07:54
jannauit looks though if apple just shoved the regs from second die into the single pmgr adt node#asahi-dev · 2022-03-23 07:53
jannaunot yet, not much sense in doing so without being able to turn anything on on the second die#asahi-dev · 2022-03-23 06:35
jannauusb-c ports from the second die don't work since I don't know how to power them on#asahi-dev · 2022-03-23 06:34
jannauGlanzmann: basic m1n1 functionality works. display + chainloading over usb-c