08:07 <maz> sven: I wonder what this Intel guy is on about...

08:12 <sven> i think he dislikes that i have to attach a device to a domain and then maybe detach it directly afterwards which, to be fair, is awkward.

08:12 <sven> and that's required because before attaching the first device to an empty domain i might have no idea what the actual i/o pagesize(s) will be. and that's really caused by the original design decision that a single bus can only have a single iommu type.

08:14 <sven> apple had a previous iphone SoC where there were some DARTs that could only do 4k pages and other that could only to 16k pages on the same bus

08:14 <sven> and i think arm's smmu also only decides the actual pagesize after the first device is attached, but i'd have to read that code again to confirm that

08:15 <maz> sven: sure, the domain part is a bit of a quirk. but keeping the status quo of "the driver knows better" is counter productive.

08:15 <sven> yeah, agreed

08:15 <sven> imho this has to happen on the iommu level and not the driver level

08:18 <maz> I guess that x86 IOMMU is still stuck outside of the iommu framework.

08:22 <sven> could be. i know that the drivers for those are surprisingly complex

08:23 <maz> yeah, and they have ugly hooks in things like GPU drivers.

08:25 <j_ey> sven: .pgsize_bitmap = -1UL, /* Restricted during device attach */

08:26 <j_ey> from drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c

08:30 <sven> ah, yes. but anyway, if the only complaint is about that attach-check-detach quirk in the end i can work around that. either i just introduce something like get_pgsize_for_dev to the iommu driver ops or i just limit the DART driver to 16k for now and kick the issue down the road if apple ever decides to bring back 4k DARTs

14:05 <alyssa> Sorry for the mailbox issue. I've reverted the offending commit

14:05 <alyssa> Pushed 20211022

15:00 <sven> kettenis: i think BIT(31) in DART_PARAMS1 is set when the DART can do region protection

15:00 <sven> still haven't tested for the pcie darts though

15:22 <kettenis> sven: same here; still polishing my wifi diff

15:24 <sven> kettenis: i just tested it with a bit of a hack: i only cleared BIT(1) in the PTE and i got the "region protection" exception

15:24 <sven> so i think it should work with those as well

15:24 <sven> (region protection failure is BIT(7) in the error register)

15:35 <kettenis> cool

19:57 <jeffmiw> alyssa: thanks, mu-one branch 20211022 boots now without the mailbox issue.

19:59 <jeffmiw> I have a mba, is there anything you need to be tested with this branch ?

20:32 <alyssa> er does 20211022 boot on mba?

20:33 <alyssa> that's nice. I forgot if I fixed the mba issues.

20:43 <jeffmiw> yes it does

20:54 <alyssa> woo :-)

21:08 <alyssa> sven: let me know when you have an updated mailbox patch to apply btw

21:08 <alyssa> though I guess I might as well just wait for it to show up in linux-next