#
#asahi-dev · 2025-09-06 10:04
svenbut i guess there's some state setup by iboot that needs to survive
#asahi-dev · 2025-09-06 10:04
svenweird
#asahi-dev · 2025-09-06 10:03
jannauon plug-in after suspend
#asahi-dev · 2025-09-06 10:03
jannaudwc3 502280000.usb: failed to reinitialize core
#asahi-dev · 2025-09-06 10:03
jannaudwc3 502280000.usb: DWC3 controller soft reset failed.
#asahi-dev · 2025-09-06 10:03
jannauat least downstream. on t8103:
#asahi-dev · 2025-09-06 10:02
jannauwe need ps_atcN_usb_aon for dwc3 to survive suspend
#asahi-dev · 2025-09-06 10:01
svenoh, duh, i messed up the tunables and forgot to set -?sz
#asahi-dev · 2025-09-06 09:59
svenugh.. why is that happening again now?!
#asahi-dev · 2025-09-06 09:59
sven[ 37.126452] phy-apple-atc 503000000.phy: Pipehandler lock not acked.
#asahi-dev · 2025-09-06 09:48
svenor there's some additional init that we're still missing
#asahi-dev · 2025-09-06 09:47
svenhuh, could be
#asahi-dev · 2025-09-06 09:47
jannaumaybe related to the omitted fuses on t602x
#asahi-dev · 2025-09-06 09:46
jannauon t602x we need ps_atcN_usb_aon. without it we end up with "dwc3 b02280000.usb: this is not a DesignWare USB3 DRD Core"
#asahi-dev · 2025-09-06 09:29
sveni updated the commit description
#asahi-dev · 2025-09-06 09:26
nickchanI mean SARTv0
#asahi-dev · 2025-09-06 09:26
nickchanso... I guess the SARTv1 commit is staying as-is?
#asahi-dev · 2025-09-06 09:14
svenyeah, i'm gonna give it a try
#asahi-dev · 2025-09-06 08:54
jannauif you want to change it I can reply to the patch
#asahi-dev · 2025-09-06 08:54
jannaunot sure that that would be a better solution but it would be more aligned with kernel style
#asahi-dev · 2025-09-06 08:53
svenyeah, that's what i'm thinking about right now. and then just drop the devm_apple_tunable_free thing entirely
#asahi-dev · 2025-09-06 08:53
jannauit looks weird. I think more in line with kernel code would be to use a flexible array memeber for values and allocate the whole struct apple_tunable with devm_*alloc
#asahi-dev · 2025-09-06 08:49
svenhm, let me take a look at the tunables.c again. it's been so long ago that i don't even remember writing that code
#asahi-dev · 2025-09-06 08:48
jannauotherwise looks good to and dropping the RFC seems fine
#asahi-dev · 2025-09-06 08:47
jannauthe devm_kcalloc in the hardware tunable trips me up but it can understand why did it like this
#asahi-dev · 2025-09-06 08:46
sveni guess i'll remove RFC as well and send this as v2
#asahi-dev · 2025-09-06 08:44
svenyeah
#asahi-dev · 2025-09-06 08:44
jannauextending the glue layer should be mentioned in the changelog
#asahi-dev · 2025-09-06 08:44
sventrue
#asahi-dev · 2025-09-06 08:43
jannausven: in the cover letter "Removed most quirks added to dwc3" isn't true if you don't count skip_core_init_mode as quirk but extending the glue layer
#asahi-dev · 2025-09-06 08:14
svenhappy with that otherwise though
#asahi-dev · 2025-09-06 08:14
svenhttps://git.kernel.org/pub/scm/linux/kernel/git/sven/linux.git/log/?h=wip/atcphy still has the 5 seconds issue :(
#asahi-dev · 2025-09-05 16:20
svenyeah, this has been supported since dart was upstream specifically because usb needs this as well
#asahi-dev · 2025-09-05 14:44
jannausee dwc3_0_dart_0 / dwc3_0_dart_1 for another example
#asahi-dev · 2025-09-05 14:43
jannaujust add "iommus = <&ane_dart_0 0>, <&ane_dart_1 0>, <&ane_dart_2 0>;" to the device tree and defien those free dart instances
#asahi-dev · 2025-09-05 14:41
chaos_princessyes, if i remember correctly, you just add all three and the iommu code will automatically replicate everything
#asahi-dev · 2025-09-05 14:40
jannauyou should not need to do anything specific
#asahi-dev · 2025-09-05 14:40
chaos_princessonly replicate dealing with darts, the rest of the isp's iommu code - you probably shouldn't
#asahi-dev · 2025-09-05 14:39
alexalteaI had no idea about the linked darts, that helps. I'll try to replicate the isp code, thanks again
#asahi-dev · 2025-09-05 14:38
chaos_princessbasically, copy what isp already does :P
#asahi-dev · 2025-09-05 14:37
chaos_princessnow i _think_, you just add all 3 darts into the device tree and it does the right thing
#asahi-dev · 2025-09-05 14:37
chaos_princessand isp/ane were originally made, we didn't have support for dealing with it all sanely
#asahi-dev · 2025-09-05 14:37
jannaufeel free to ask more questions
#asahi-dev · 2025-09-05 14:37
chaos_princessso, some devices like ane and isp have "linked darts" where you need all darts set to the same thing
#asahi-dev · 2025-09-05 14:36
chaos_princessright, that
#asahi-dev · 2025-09-05 14:36
alexalteaalso, it seems it uses actually dma/gem for dart0, most of the direct accesses are for dart1 and dart2
#asahi-dev · 2025-09-05 14:34
alexalteathanks jannau
#asahi-dev · 2025-09-05 14:33
jannauI forgot that ane did this
#asahi-dev · 2025-09-05 14:32
jannauhttps://github.com/AsahiLinux/linux/blob/asahi/drivers/accel/ivpu/ivpu_gem.c could be a good reference
#asahi-dev · 2025-09-05 14:31
jannauin first approximation it shouldn't access dart directly but use the dma api and gem-shmem helpers

plus ›