#asahi-dev · 2022-04-29 17:15
sveni'm mostly curious which of the hardware keys are the same#asahi-dev · 2022-04-29 17:15
svenm1 mini results in this https://f.svpe.de/baf349300264335f2b5605096161113d1b649270c5b1603e162e2b336797bd17_aes_m1_mini_sven.txt#asahi-dev · 2022-04-29 17:15
sveni somehow got curious about that aes engine, can anyone run this test and paste the results somewhere? https://f.svpe.de/4fda9a85379eecdf4e53140607a54b828404c1174c65013579364e7e3ed88a97_aes.py#asahi-dev · 2022-04-28 10:08
svenI think I have to send another pull instead of send-email to the nvme maintainers anyway to make sure the commit hashes stay the same#asahi-dev · 2022-04-28 10:06
svenI guess I can just sign that tag as well#asahi-dev · 2022-04-28 10:01
svenok, i'll prepare that setup just in case then :-)#asahi-dev · 2022-04-28 10:00
svenI think both thunderbolt and DCP likely won't be ready for 5.19 anyway#asahi-dev · 2022-04-28 09:59
svenmarcan: do you intend to submit SMC for 5.19? if yes i'll have to look into preparing that git setup where we can merge rtkit through multiple trees. if you'll wait for the next cycle anyway I guess we can just take rtkit+sart through the nvme tree (if arnd agrees)#asahi-dev · 2022-04-28 09:56
svenI think the problem is that fastmail signs too many headers (at least "Sender" which will be changed to the list)#asahi-dev · 2022-04-28 08:36
svenhrm, looks like b4 doesn't complain for mails sent to lists from e.g. kernel.org but it does complain about those sent from my domain. time to open a ticket with fastmail and/or figure out how to get one of those @kernel.org things otherwise :/#asahi-dev · 2022-04-28 08:24
svenhrm, or maybe it's always been broken with mailing lists? :/#asahi-dev · 2022-04-28 08:14
svenI hate email. Apparently at least the nvme series went to spam on gmail for some people because for some reason dkim/spf/dmarc broke *sigh*#asahi-dev · 2022-04-27 16:01
svenrotating across darts would’ve been very strange #asahi-dev · 2022-04-27 04:51
svenDcow[m]1: thanks, looks like it’ll be v4 ;)#asahi-dev · 2022-04-26 20:23
svenhttps://lore.kernel.org/linux-nvme/20220426201539.12829-1-sven@svenpeter.dev/T/#t and there's nvme-v3 :-)#asahi-dev · 2022-04-26 19:43
svenit essentially has a command fifo, wouldn’t surprise be if you need to reset it if a command in there gets stuck#asahi-dev · 2022-04-26 19:25
svenI never tried that old script with translation because I didn’t know about that stream_enable register so there could be also some other issue#asahi-dev · 2022-04-26 19:20
svenPretty sure that’s still missing from that old code#asahi-dev · 2022-04-26 19:20
svenyou need to write to that stream enable register as well#asahi-dev · 2022-04-26 17:54
svenrtkit had no functional changes from v2 fwiw, just added _ULL to the GEMASK defines #asahi-dev · 2022-04-26 17:53
svenawesome, thanks!#asahi-dev · 2022-04-26 17:38
svenerm, s/64bit/64 SID/#asahi-dev · 2022-04-26 17:37
sveneither it's possible to get the number of SIDs from one of the early DART registers, or we check how many TCRs we can access, or we just add another compatible for the 64bit DART i#asahi-dev · 2022-04-26 17:36
sventhey just pass 16 or 64 as a custom DT property IIRC which is not how you do it#asahi-dev · 2022-04-26 17:36
svensee, that's why I recommend to not look at that code because it can confuse you#asahi-dev · 2022-04-26 17:29
svenanyone with a t6k here who can give https://github.com/AsahiLinux/linux/tree/nvme-v3 a brief test?#asahi-dev · 2022-04-26 17:29
sveni think marcan's policy is to treat it as if you were reading decompiled XNU assembly though#asahi-dev · 2022-04-26 17:28
sveni haven't seen anything problematic from a copyright standpoint inside corellium's code but I generally avoid looking at it because it's just a PoC and can give you the wrong idea about how to write a sane driver#asahi-dev · 2022-04-26 17:27
svenfor usb XNU just mirrors the writes and pagetables for both DART instances, might just be the same for ISP#asahi-dev · 2022-04-26 16:15
svenand iommu-maps is then consulted to figure out which pci device number is supposed to be mapped to which sid#asahi-dev · 2022-04-26 16:14
svenIm willing to be the sid count is also encoded in one of the DART_PARAM regs at the start though #asahi-dev · 2022-04-26 16:13
svenwith DART I did the same but within Python and 16 is hard coded for now#asahi-dev · 2022-04-26 16:13
svenPCIe figures out 16 vs 64 (or whatever number) inside the rid2sid map by testing which regs are writable #asahi-dev · 2022-04-24 17:47
svenjannau: I can’t find it anymore, but iirc there is a bit in one of the early registers that indicates if the dart hw supports subpage protection#asahi-dev · 2022-04-24 13:54
sveni vaguely remember that there was also some bit somewhere to indicate support#asahi-dev · 2022-04-24 13:52
svenohh.. maybe those just don't support subpage protection#asahi-dev · 2022-04-24 13:16
svenso it’s actually not mandatory on t6k?#asahi-dev · 2022-04-24 13:16
svenhuh#asahi-dev · 2022-04-24 09:43
svenonly one way to figure it out ;)#asahi-dev · 2022-04-23 16:41
sveneither way, adding dma-coherent to the dt is probably something we should do #asahi-dev · 2022-04-23 16:40
sven(which I wrongly assumed to be almost never used back then iirc)#asahi-dev · 2022-04-23 16:39
svenI think that’s only used for streaming mappings #asahi-dev · 2022-04-23 16:36
svenah, maybe it didn’t make a difference because the current dart driver ignores that :D#asahi-dev · 2022-04-23 16:35
svenI vaguely remember that it didn’t make a different for dma_alloc_coherent. But I looked into this long before I even understood how half of the dma api worked so there’s a good chance I’m wrong #asahi-dev · 2022-04-23 16:32
svenbut iirc uncached mapping is just something like IOMMU_CACHE#asahi-dev · 2022-04-23 16:32
svenjannau: no#asahi-dev · 2022-04-23 08:22
svendon‘t think so, I never got around to write any documentation #asahi-dev · 2022-04-22 20:04
svenhuh, and this breaks dwc3 apparently *sigh*