08:22 <sven> don‘t think so, I never got around to write any documentation

15:47 <kevans91> alright, another silly question; what exactly is this reg? is this one of those 'reg describes the PCI RID' things? https://github.com/u-boot/u-boot/blob/master/arch/arm/dts/t8103-j274.dts#L53 (../port02/ethernet0)

16:17 <kettenis> kevans91: that reg encodes the "address" of the device on the PCI bus

16:19 <kettenis> see: https://www.openfirmware.info/data/docs/bus.pci.pdf

16:21 <kettenis> it basically tells you that the local-mac-address property applies to the device at bus 3, device 0, function 0, which is the built-in bge(4)

16:29 <jannau> sven: are you sure that the missing "dma-coherent" makes no difference? casual reading of dma-iommu suggests it might.

16:30 <jannau> I'm looking into using it for the t6000 dart uncached PTE bit

16:31 <kevans91> kettenis: ok, cool, thanks. we're having a heck of a time with pcie, and I'm afraid I really don't understand pcie well enough to be helpful yet... working on it, though

16:32 <sven> jannau: no

16:32 <sven> but iirc uncached mapping is just something like IOMMU_CACHE

16:35 <sven> I 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

16:35 <jannau> yes, I'm want to use IOMMU_CACHE to control the uncached bit. As far as I can see IOMMU_CACHE is controlled by the dev->dma_coherent which is set by the dma-coherent dt property

16:36 <sven> ah, maybe it didn’t make a difference because the current dart driver ignores that :D

16:38 <jannau> shouldn't it make a difference for iommu_dma_sync_*

16:39 <sven> I think that’s only used for streaming mappings

16:40 <sven> (which I wrongly assumed to be almost never used back then iirc)

16:41 <sven> either way, adding dma-coherent to the dt is probably something we should do