10:47 <wens> I see that the power domain controllers in rockchip and mediatek device trees have child nodes describing the tree structure of all the power domains, and resources for each individual domain, and all of them have "#power-domain-cells = <X>"; however in the driver, none of these child nodes have of_genpd_add_provider_* called on them.

10:47 <wens> wonder if this was cargo-culted somehow and no one bothered to check whether it actually works

10:52 <wens> consumers just reference the top level power controller

11:30 <robmur01> ...but since #power-domain-cells is nonzero, that reference includes an ID, and that ID is (presumably) then matched against a child's reg value to ultimately link it back to the right resources

11:34 <robmur01> I guess you could equally have #power-domain-cells = 0 and have consumers point their phandle directly at the relevant child, but I wouldn't say either scheme is necessarily right or wrong

12:01 <geertu> Yeah, that looks strange: #power-domain-cells = <1> in the top level power controller, and #power-domain-cells = <0> in the children

12:02 <geertu> Even more weird: top level and first child level use <1>, second child level uses <0>

12:02 <geertu> Wow, even 4 nested levels with <1>

12:03 <geertu> I can understand you want to put some topology in DT (for the clocks), but if there are no references to these subnodes, you don't need #power-domain-cells

12:17 <robmur01> yeah, pick one pattern or the other, but mixing both in one binding seems silly

12:23 <wens> for the rockchip ones, the sub-nodes' names would have to be changed to something that doesn't match the common power-domain binding

12:46 <milkylainen> Hmm. can one dirty a specific set of pages in the mtd subsystem? Or in a specific fs containing an ftl, like jffs2, ubi(?

12:46 <milkylainen> ubifs(ubi) even

12:46 <milkylainen> I want the hw gc, software ftl or whatever to move pages.

20:13 <hanetzer> marcan: sorry for the rando ping, but are you the marcan of asahi linux?