#asahi-dev · 2026-08-19 13:10
chadmedor is it just the metadata #asahi-dev · 2026-08-19 13:10
chadmedthe tiles are bigger for y too right#asahi-dev · 2026-08-19 13:10
sofus32 for the y plane and 8 for uv#asahi-dev · 2026-08-19 13:09
sofusi based my name on what the tracer called it, so metadata per tile. And thats what i think needs to change (at least it does for avd)#asahi-dev · 2026-08-19 13:08
chadmedmeta_bytes is the total metadata size of the image#asahi-dev · 2026-08-19 13:07
chadmedsorry, yeah thats what the 8 is#asahi-dev · 2026-08-19 13:07
chadmedmeta_bytes is the bytes of metadata per tile#asahi-dev · 2026-08-19 13:07
sofushttps://github.com/chadmed/linux/blob/fda11f8f39a7331b195213c95746375c521d1365/drivers/gpu/drm/apple/plane.c#L45#asahi-dev · 2026-08-19 13:06
sofus(the naming might be a bit confusing)#asahi-dev · 2026-08-19 13:05
sofusright now it looks like dcp uses a static meta_bytes=8#asahi-dev · 2026-08-19 13:05
chadmedthat seems right, same as agx and dcp#asahi-dev · 2026-08-19 13:03
sofushttps://github.com/sofus13/linux/blob/64ad7ba7548fd2afa1ea27ca4a88dd1f2ddc1327/drivers/media/platform/apple/avd/avd-drv.c#L36#asahi-dev · 2026-08-19 13:03
sofuschadmed: for interchange i think need to change the meta_bytes calculation, in the avd driver, it looks like this:#asahi-dev · 2026-08-19 10:53
sventhat should be easier than either DP or PCIe #asahi-dev · 2026-08-19 10:53
svenI wonder how hard it is to get usb4 tunneling to work #asahi-dev · 2026-08-19 09:49
svenack #asahi-dev · 2026-08-19 09:47
jannauit also requires https://lore.kernel.org/asahi/20260819-arm64-dts-apple-t6021-dma-ranges-v1-1-a00e109640f6@jannau.net/T/#u#asahi-dev · 2026-08-19 09:46
jannauwill be for the 7.2 rebase of the downstream kernel. I have no motivation to backport the thunderbolt PCIe decoupling changes#asahi-dev · 2026-08-19 09:34
svenI need to fix m1n1 to write to both _ and - for now and then after a bit more testing of the new cable VSE nonsense we can also start to think about shipping this in the downstream kernel #asahi-dev · 2026-08-19 09:32
sven*them#asahi-dev · 2026-08-19 09:32
svenack, will merge then this evening after $work and pull your series in as well #asahi-dev · 2026-08-19 09:31
jannausven: with https://github.com/jannau/linux/commit/c34304366738041af38ee95704559877c56cd9e8 and https://github.com/jannau/linux/commit/6856fd4cb0922e59c2506f94434d1b9ea00012ab as fixes for tbt-reset-wip#asahi-dev · 2026-08-19 09:25
svenmuch better than that old property :)#asahi-dev · 2026-08-19 09:24
svennice, thanks!#asahi-dev · 2026-08-19 09:21
jannauworks: https://github.com/jannau/linux/commits/b4/iommu-apple-dart-aperture/ I'll send this later today#asahi-dev · 2026-08-19 08:12
jannaumy intel notebook generates the same fff… xdomain uuid suffix#asahi-dev · 2026-08-19 08:05
svenprobably just random then so the fff… fallback linux uses is fine then I guess #asahi-dev · 2026-08-19 08:04
svenlooks like the xdomain uuid macOS used changes each boot #asahi-dev · 2026-08-19 06:40
svennice!#asahi-dev · 2026-08-19 06:39
jannauI've started implementing it#asahi-dev · 2026-08-19 06:25
svenand at least it’s on the correct node #asahi-dev · 2026-08-19 06:25
svencould still be worth a try #asahi-dev · 2026-08-18 21:39
jannauthere is no device tree with #iommu-cells > 2 though#asahi-dev · 2026-08-18 21:36
svenoh nice #asahi-dev · 2026-08-18 21:32
jannauDocumentation/devicetree/bindings/iommu/iommu.txt has an example with #iommu-cells = <4> to specify a DMA window which would map to the iommu aperture#asahi-dev · 2026-08-18 21:02
jannauI guess it works in the same way as my funky dma_mask by allocation top-down#asahi-dev · 2026-08-18 20:58
sventhen i must've missed some code somewhere because to me it looked like especially the lower bound would just be thrown away#asahi-dev · 2026-08-18 20:57
svenoh, nice!#asahi-dev · 2026-08-18 20:57
jannau'dma-ranges = <0x100 0x0 0x100 0x0 0xffff0000>;' in the parent node works but #address-cells needs to be increased to 2#asahi-dev · 2026-08-18 20:53
svenugh, though i think probe would be too late to set that up#asahi-dev · 2026-08-18 20:52
sveni wonder if some iommu_dma_set_aperture would be accepted upstream#asahi-dev · 2026-08-18 20:47
svenah true, i never increase it in the driver#asahi-dev · 2026-08-18 20:46
jannaucan't use more anyway at the moment since the defailt dma_mask is 32-bit#asahi-dev · 2026-08-18 20:45
jannau#address-cells can stay at 1, size as well if 4GB are enough#asahi-dev · 2026-08-18 20:43
svenoh well.. maybe v1 will just be base M1/M2 then#asahi-dev · 2026-08-18 20:42
svenand from reading the code it will only enforce the upper limit. maybe.#asahi-dev · 2026-08-18 20:42
sveni think we'd have to bump back the *-cells to 2#asahi-dev · 2026-08-18 20:40
jannau"dma-range" instead of "dma-ranges" is not going to work and my values were wrong as well#asahi-dev · 2026-08-18 20:36
svenhmmm... :/#asahi-dev · 2026-08-18 20:23
jannauit works with and without "dma-ranges" in the parent node. I used 'dma-range = <0x100 0x0 0x1 0x0>;'