#asahi-re · 2021-08-23 06:05
marcanah, right, clocks#asahi-re · 2021-08-23 06:05
svenclocks, i2c. i think j_ey did some work on gpio/pinctrl but i dunno how far along he is#asahi-re · 2021-08-23 06:05
marcanok, I'm going to stream a bit today, what was on the nobody's doing it list? i2c and pinctrl was it?#asahi-re · 2021-08-23 06:04
marcanyeah#asahi-re · 2021-08-23 06:04
marcan*on#asahi-re · 2021-08-23 06:04
svenyeah, i have two variants that i implemented: one just uses the poll mode as-as. the other one convinces it to keep sending unless it gets EBUSY and request the irq to continue then#asahi-re · 2021-08-23 06:04
marcanwhatever we decide one needs to be fairly future-proof#asahi-re · 2021-08-23 06:04
marcanoh, also: the DT bindings need some thought. I want to make sure that e.g. if we decide the mailbox framework sucks one day, that we don't run into issues with the bindings#asahi-re · 2021-08-23 06:03
marcanso we can also just defer that to whenever we run into perf issues#asahi-re · 2021-08-23 06:03
marcanbut it might not matter in practice for now#asahi-re · 2021-08-23 06:03
marcanso that it can keep sending#asahi-re · 2021-08-23 06:03
marcanjust once#asahi-re · 2021-08-23 06:03
marcanit would need a flag to poll after a send#asahi-re · 2021-08-23 06:03
marcanyeah that too, the queuing#asahi-re · 2021-08-23 06:02
svenbut that's a small thing#asahi-re · 2021-08-23 06:02
svenyeah. the core needs a small change anyway unless we want to wait for an interrupt for every single message we sent#asahi-re · 2021-08-23 06:02
marcangiven we're building monolithic drivers anyway#asahi-re · 2021-08-23 06:02
marcanbut... not worth it, I think#asahi-re · 2021-08-23 06:02
marcanthough it could be adapted to ours by adding a mode to mailbox where all channels share a busy status#asahi-re · 2021-08-23 06:01
marcanyeah, exactly#asahi-re · 2021-08-23 06:01
svenwhich... we don't#asahi-re · 2021-08-23 06:01
sventhe whole channel concepts really seems to be more about a single mailbox block having multiple FIFOs#asahi-re · 2021-08-23 06:01
svenyeah#asahi-re · 2021-08-23 06:01
marcanthe rtkit lib would of course be a module to be shared by all the drivers#asahi-re · 2021-08-23 06:01
marcanand deal with the endpoints on our rtkit stuff#asahi-re · 2021-08-23 06:01
marcanlet's just treat it as a single-channel mailbox#asahi-re · 2021-08-23 06:00
svenso just a simple mailbox driver + rtkit shared library sounds good then#asahi-re · 2021-08-23 06:00
marcanI took a good look at mailbox the other day and I don't like that we can't use the channel concept, but given this situation, it's not worth making changes to that#asahi-re · 2021-08-23 06:00
svenyeah#asahi-re · 2021-08-23 06:00
marcanyeah, different story#asahi-re · 2021-08-23 06:00
sveni briefly looked into remoteproc but that doesn't fit all that well either#asahi-re · 2021-08-23 06:00
marcantrying to split it up into modules... would just add overhead for no reason#asahi-re · 2021-08-23 06:00
svenyeah#asahi-re · 2021-08-23 06:00
marcanand if the thing crashes the whole thing crashes#asahi-re · 2021-08-23 06:00
marcanafter all whoever wants to run these things will need the whole driver anyway#asahi-re · 2021-08-23 05:59
marcanyeah, tbh I don't see modularity really helping us here#asahi-re · 2021-08-23 05:56
svenmarcan: yeah, so if we can really just handle every single co-processor with a single driver i think that shared rtkit library you suggested makes sense#asahi-re · 2021-08-23 03:52
marcanso we can probably just drive them all from a single DCP driver eventually#asahi-re · 2021-08-23 03:52
marcansven: yeah, I was thinking that DCP is probably the only case where those drivers are so broken out in macos, but I think for linux those things tend to be all monolithic#asahi-re · 2021-08-21 15:31
svenhm, might only be important to decide how to handle all the common rtkit stuff for the kernel#asahi-re · 2021-08-21 15:02
alyssamarcan: Woof#asahi-re · 2021-08-21 14:45
marcanI'm not sure how much we have to care at this point#asahi-re · 2021-08-21 14:45
marcansemi-autonomously afaik#asahi-re · 2021-08-21 14:45
marcanalyssa: those are driven by different drivers#asahi-re · 2021-08-21 14:31
svenand i suspect the pagetable format is going to be very similar#asahi-re · 2021-08-21 14:30
alyssa(for 16k pages. certainly there's a point for 2MB pages ha)#asahi-re · 2021-08-21 14:30
svenit'll certainly be capable of creating IOVA -> physaddr mappings just like DART can#asahi-re · 2021-08-21 14:30
alyssa not much point for graphics ...#asahi-re · 2021-08-21 14:30
svenbut afaik DART has stuff like subpage-protection. no idea if GART has that as well#asahi-re · 2021-08-21 14:29
alyssasven: I mean correct me if I'm wrong but DART is made to be simple