11:46 <bbrezillon> robmur01: Any opinion on the approach suggested by Jason in the custom io-pgtable allocator patch series?

11:47 <bbrezillon> I'd really like to make sure IOMMU maintainers are okay with that before giving this solution a try...

11:48 <bbrezillon> (will require some changes to panthor's uAPI too)

11:48 <bbrezillon> s/some//

12:09 <karolherbst> alyssa: is anybody testing rusticl+panfrost? :D

12:09 <karolherbst> (that reminds me....)

12:36 <mntirc> bbrezillon: if i want to build a kernel today for rk3588 testing with pancsf, which is the best branch to start with? https://gitlab.freedesktop.org/bbrezillon/linux/-/commits/panthor%2Bfdinfo ?

12:45 <bbrezillon> mntirc: https://gitlab.freedesktop.org/panfrost/linux/-/tree/panthor-v3+rk3588-evb1?ref_type=heads

12:47 <bbrezillon> and https://gitlab.freedesktop.org/bbrezillon/mesa/-/tree/v10+panthor-v3?ref_type=heads for mesa

12:59 <mntirc> bbrezillon: thank you!

13:06 <mntirc> is panthor a new name for pancsf or...?

13:15 <bbrezillon> mntirc: yes

13:16 <bbrezillon> the panthor -> pancsf rename happened in v2 of the patchset

13:16 <mntirc> ah cool

13:16 <bbrezillon> sorry, pancsf -> panthor

13:26 <alyssa> karolherbst: dunno

13:26 <alyssa> i don't even have the hw anymore :~)

13:27 <mntirc> bbrezillon: the current tip of that branch gives me > drivers/gpu/drm/rockchip/rockchip_drm_drv.c:1351:16: error: ‘struct drm_device’ has no member named ‘irq_enabled’

13:30 <karolherbst> nice.... :) though somebody promised to send me hardware, I should ping them

13:36 <mntirc> there's also a bunch of drivers using "to_rockchip_encoder" which doesn't seem to be defined anywhere

13:37 <mntirc> disabling these drivers for now

13:53 <bbrezillon> mntirc: yeah, you have to disable VOP support (the rk display port is just a gigantic hack)

13:54 <mntirc> ok

13:54 <mntirc> bbrezillon: which display output(s) is/are supported by that kernel? hdmi?

13:55 <bbrezillon> might want to use rk3588_panthor_debug_defconfig (make ARCH=arm64 rk3588_panthor_debug_defconfig)

13:55 <bbrezillon> mntirc: only HDMI, yes

13:59 <mntirc> ok, thx

17:42 <HdkR> karolherbst: You should have let me know, I could have brought an RK3588 to XDC :P

17:43 <karolherbst> don't worry, others already promised me to give me stuff :D

18:09 <robclark> bbrezillon: I'm not sure Jason's suggestion for pre-reserving iova actually solves anything.. the # of pages needed for map/unmap (unmap can in theory require new pages if it splits a larger page) can depend on what physical pages you are mapping.. like you might have queued up UNMAP followed by a MAP of different pages to the same iova which increases the # of pages you need compared to the state before the UNMAP

18:16 <robmur01> yeah, I've only skimmed it but it sounds more like a solution to some totally different problem that nobody raised :/

23:05 <mntirc> bbrezillon: just a sanity check: are usb and sdmmc working for you on your kernel/branch?

23:07 <mntirc> i hacked together a devicetree for firefly icore-rk3588 and got to console with networking, but no usb devices nor sd card detected (and cpu stall if i try to activate xhci 1)