12:15 tomeu: phh: hmm, wonder why yours and Jasbir's demos fail to get a completion IRQ on my setup
12:15 tomeu: rockchip
12:15 tomeu: 's demo does
12:16 tomeu: what kernel driver were you testing with?
13:22 phh: Idk, vendor 5.10 for Android
13:22 phh: I'll try my demo because it's not impossible it's broken, I'll have to test
13:23 tomeu: cool, thanks
16:53 phh: tomeu: k my demo fails too
16:54 phh: I should have gitted it, I think I had it working at some point and I broke it when trying to clean it...
17:01 phh: tomeu: well I do have a better result than you
17:01 phh: [ 1394.993998] RKNPU: invalid irq status: 0x4, raw status: 0xc0000055, require mask: 0x30c, task counter: 0x0
17:02 phh: I do receive an IRQ
17:02 phh: it's just misrouted somehow
17:04 phh: k I got it to work
17:04 phh: tomeu: https://github.com/phhusson/rknpu-reverse-engineering/blob/main/hello2.c this samples succeeds on my device
17:06 phh: I got `tasks[0].int_mask = 0xf;` value by reading the log, saying irq status 0x5, and checked rknpu's rknpu_fuzz_status to see what is 0x5 replaced with
17:07 tomeu: ah, so the relevant change is:
17:07 tomeu: - tasks[0].int_mask = 0x300; // 0x300 = wait for DPU to finish
17:07 tomeu: + tasks[0].int_mask = 0xf; // Ask for any interrupt at all...?
17:07 tomeu: ?
17:08 tomeu: I was using 0x300 as well, as that what the blob sends
17:09 tomeu: but changing it to 0xf doesn't seem to make any difference
17:14 tomeu: wonder if I broke the kernel when forward porting it to 6.10, but the blob works...
18:20 phh: if you don't get the "invalid irq status" like me, your issue is different than what I had
18:20 phh: but anyway, you tried taking my latest hello2.c building it and running it?
18:52 tomeu: yeah, and I don't get any irq, but the ssd demo binary does
21:23 phh: do you want me to try your code on my vendor kernel?