#asahi-dev · 2022-02-06 17:15
jannauis there a way to make kernel_addr_r dynamic on the mem map? the distro bootcmds will fail otherwise#asahi-dev · 2022-02-06 16:08
jannauon t600x two pmgr nodes are required but only one gets u-boot,dm-pre-reloc#asahi-dev · 2022-02-06 16:07
jannauit enabled one pmgr at the end of the recursion and on m1 only the main pmgr is needed#asahi-dev · 2022-02-06 16:03
jannaukeyboard is not working, input over the uart doesn't work either#asahi-dev · 2022-02-06 16:01
jannauhttps://github.com/AsahiLinux/m1n1/pull/161#asahi-dev · 2022-02-06 15:55
jannaukettenis: nvme works after fixing the u-boot,dm-pre-reloc problem#asahi-dev · 2022-02-06 15:25
jannauno, usb and keyboard don't work either, I'm currently looking why the u-boot,dm-pre-reloc changes in m1n1 do not work#asahi-dev · 2022-02-06 15:13
jannauyes, works with power-domains commented in the dts#asahi-dev · 2022-02-06 15:07
jannauyes, under the HV I now get 'No serial driver found' but Samsung S5P UART is enabled#asahi-dev · 2022-02-06 15:03
jannaukettenis: does not work at all, no output even with early debug uart#asahi-dev · 2022-02-06 14:05
jannaualigning everything to 4k still works, but only the fb size was unaligned to begin with#asahi-dev · 2022-02-06 13:59
jannauyes, 4k works as well#asahi-dev · 2022-02-06 13:50
jannaukettenis: no, the only problem is that the frame buffer size is not page aligned#asahi-dev · 2022-02-06 13:50
jannaumostly thanks to kettenis, annoying that I didn't find https://github.com/kettenis/u-boot/pull/2/files when I tried the first time#asahi-dev · 2022-02-06 13:48
jannaubasic t600x support is working in a single u-boot binary#asahi-dev · 2022-02-06 13:31
jannaukettenis: sigh, the problem is annoyingly simple. add_map() is looping endlessly due to a non page aligned framebuffer size#asahi-dev · 2022-02-06 13:01
jannauwithout it I get both frame buffer and console prompt, albeit slowly#asahi-dev · 2022-02-06 13:00
jannaukettenis: dcache_enable() breaks u-boot on the t600x#asahi-dev · 2022-02-06 12:05
jannauunfortunately not from my side, I keep being distracted by other stuff. I was planning to work on that this weekend but haven't done anything yet#asahi-dev · 2022-02-06 11:53
jannaukettenis: yes, dart-dcp is locked, dart-disp0 has the framebuffer mapped and resetting it breaks the fb#asahi-dev · 2022-02-06 11:52
jannaufor the locked dcp dart all thatis needed is to ignore locked darts#asahi-dev · 2022-02-06 11:52
jannaufor the disp0 dartI have currently "apple,reset-mask" but that is probably going to change#asahi-dev · 2022-02-06 11:48
jannauMy fear is that the memset in alloc_page_table takes effect before I can remap all reserved regions#asahi-dev · 2022-02-06 11:43
jannauI not sure if I'm not just lucky. will apple_dart_map_pages cause a tlb_invalidate()? I didn't check thouroughly. if it does apple_dart_setup_reserved_regions() is obviously a race for dcp with more than 1 reserved region#asahi-dev · 2022-02-06 09:29
jannausure, look at it when you have time#asahi-dev · 2022-02-06 09:28
jannauI think it will blow up on device detach#asahi-dev · 2022-02-06 09:28
jannaudart reserved memory is in https://github.com/jannau/linux/commit/09e83ed9f30b999c00aea08726c6d91d32a6408b#asahi-dev · 2022-02-06 09:27
jannauI went for now with remapping the locked ttbr, mostly since Alyssa already did that. changes as far as io-pgtable-arm is concerned are limited to https://github.com/jannau/linux/commit/d813575ca19d122ce4f371fabdd1d1a3143a65ec#asahi-dev · 2022-02-06 09:25
jannausven: I pushed reserved memory / locked dart changes to https://github.com/jannau/linux/tree/bits/200-dcp#asahi-dev · 2022-02-06 07:52
jannaunote I gaven't looked at the code#asahi-dev · 2022-02-06 07:51
jannaueven if it quite unlikely that the code will ever run on a big-endian cpu#asahi-dev · 2022-02-06 07:50
jannauit type checks the input. you could use swab64() directly but the operation we want is to convert a 64-bit big-endian value to the cpu's endianess#asahi-dev · 2022-02-06 07:41
jannauit probably should use be64_to_cpu() (effecively using __builtin_bswap64)#asahi-dev · 2022-02-05 19:14
jannauI'm also not quite sure if the reserved regions are still necessary when the they are properly mapped#asahi-dev · 2022-02-05 19:11
jannauok, I see much smaller maximal VAs and I suppose only the framebuffer itself will differ depending on display resolution#asahi-dev · 2022-02-05 18:55
jannauwhere are the reserved regions for the locked darts in apple_dart_get_resv_regions() coming from?#asahi-dev · 2022-02-05 18:54
jannaualyssa: thanks. I suppose you don't have outstanding commit for dcp. I plan to rebase the tree to remove commits not longer needed#asahi-dev · 2022-02-05 18:28
jannauI can clear the pagetable there and recreate the mappings from reserved regions in apple_dart_setup_translation()#asahi-dev · 2022-02-05 18:27
jannauI have dcp working without with just memremapping the ttbr of the locked dcp dart in io-pgtable-arm.c#asahi-dev · 2022-02-04 06:58
jannauI'm working on a proposal that would work for dart/dcp: using reserved-memory and define how it should be mapped in the iommu node#asahi-dev · 2022-02-04 06:56
jannauthe only problem I see with having the memory-regions/mappings in the dtb is that there are no bindings to express that#asahi-dev · 2022-02-04 06:40
jannaudecoupling locked form inherit-mapping would not be a problem#asahi-dev · 2022-02-04 06:39
jannauthis currently is controlled by the dart being locked, the disp0 dart is soft-locked by a property in the dtb#asahi-dev · 2022-02-04 06:37
jannauI'm under the impression that sven thinks upstream will not like/accept it#asahi-dev · 2022-02-04 06:35
jannaumarcan: that's more or less what alyssa already did for dcp see https://github.com/jannau/linux/commit/e4ec146a8fc6fb565a18e8fb6c501b4f7f8bc7fb#asahi-dev · 2022-02-03 20:32
jannauI'm writing that as proposal in reserved-memory dt-bindings thread#asahi-dev · 2022-02-03 20:31
jannauthe dart driver knows then which streams it must not reset and the device can still look up the region via its iommus phandle#asahi-dev · 2022-02-03 20:30
jannauwe could also just move the reserved regions from the device to the iommu node#asahi-dev · 2022-02-03 20:14
jannauyes, it kills the framebuffer and dcp#asahi-dev · 2022-02-03 19:38
jannauthe dart (dart-dips0) used to map the frame buffer is not locked, only dart-dcp is locked