#asahi · 2021-04-07 08:49
mazarnd: I'll just mention the driver in my PR to tglx.#asahi · 2021-04-07 08:48
mazarnd: tglx usually isn't that interested in the non-x86 irqchip stuff, which is why I'm dealing with it.#asahi · 2021-04-07 08:30
mazsven: I think kettenis_ pointed me to this a while ago (though I can't find the original source): https://pastebin.ubuntu.com/p/sBTGfSzj3p/#asahi · 2021-04-06 15:24
mazor it has a ROM of some sort.#asahi · 2021-04-06 15:24
mazI guess something gets fed to it early on.#asahi · 2021-04-06 15:23
mazno idea.#asahi · 2021-04-06 15:23
mazI'd be surprised if tg3 didn't need some firmware too.#asahi · 2021-04-06 14:36
mazkettenis_: yeah, some bog-standard ASIX adapter.#asahi · 2021-03-29 13:19
mazkettenis1: but is that how you configure that MSI split?#asahi · 2021-03-29 13:17
mazkettenis1: ah, that's directly in the port BARs?#asahi · 2021-03-29 13:16
mazarnd: I think XHCI prefers MSI over MSI-X, for bizare reasons. I remember hacking it to test things on Chello.#asahi · 2021-03-29 13:15
mazkettenis1: are these registerse "documented" anywhere?#asahi · 2021-03-29 13:13
mazkettenis1: how are things configured in u-boot at the moment?#asahi · 2021-03-29 13:01
mazthat's what I do too. I don't see a need for extra slicing.#asahi · 2021-03-29 12:57
mazthe first one is odd, because they leave 8 MSIs on the table...#asahi · 2021-03-29 12:55
mazcorellium splits the MSI space for each controller in 3, though I'm not sure that's actually enforced by the HW.#asahi · 2021-03-29 12:39
mazbut the non-translated doorbell is what gives me a strong indication that this all lives in the RC, and not in the AIC.#asahi · 2021-03-29 12:38
mazarnd: as for using the generic PCI MSI controller, that's almost what I have (just a shim to allocate the various IRQ bits).#asahi · 2021-03-29 12:38
mazarnd: it looks like the 32 MSIs are global.#asahi · 2021-03-29 12:17
mazit really looks like the doorbell is upstream of the IOMMU.#asahi · 2021-03-29 12:17
mazarnd: if the MSI doorbell was translated, I'd agree with you. but that's not how the HW seems to behave.#asahi · 2021-03-29 11:10
mazs/yet/yes/#asahi · 2021-03-29 11:09
mazwinocm: yet, kettenis1 has done an awesome job at an initial port, with working PCI (which allows me to netboot the sucker).#asahi · 2021-03-29 11:08
mazbut everything else can be derived from the context.#asahi · 2021-03-29 11:08
mazas you can see, this is as minimal as it gets (base interrupt, number of interrupts).#asahi · 2021-03-29 11:07
mazkettenis1: so far, I'm running with this (directly derived from your u-boot version): https://paste.debian.net/1191492/#asahi · 2021-03-29 10:11
mazkettenis1: and of course, I need to integrate the DART stuff to do anything useful.#asahi · 2021-03-29 10:11
mazkettenis1: got probing working, need to test MSI (because the xhci driver really want interrupts... seems overrated to me... ;-)#asahi · 2021-03-29 08:15
mazmarcan: feel free to keep one of the early patches around for context.#asahi · 2021-03-29 08:14
mazmarcan: for these two VHE-specific issues, let's keep the discussion separate for now. It is likely that they would directly go into the arm64 tree anyway.#asahi · 2021-03-28 12:09
mazmarcan: tested on top of v4, all working now.#asahi · 2021-03-28 11:58
mazmarcan: apologies for dragging you on the wrong path!#asahi · 2021-03-28 11:58
mazmarcan: local hack fixed it indeed.#asahi · 2021-03-28 11:43
mazlet me fix that locally and see what happens...#asahi · 2021-03-28 11:41
mazand we do that irrespective of the masking.#asahi · 2021-03-28 11:40
mazif the mask is actually a mask, and that the guest timer is pending, we are still going to process the interrupt.#asahi · 2021-03-28 11:38
mazI wonder if I'm being tricked by the FIQ handler dealing with both HV and guest timers *in that order*.#asahi · 2021-03-28 11:31
mazotherwise I wouldn't be able to make forward progress.#asahi · 2021-03-28 11:30
mazright. but they fire *once*, not as a level.#asahi · 2021-03-28 11:29
mazmarcan: and you don't see the timers firing when the mask is set to 0?#asahi · 2021-03-28 11:16
mazit stay active.#asahi · 2021-03-28 11:16
maznothing in the interrupt handler clears the timer.#asahi · 2021-03-28 11:15
mazand it doesn't.#asahi · 2021-03-28 11:15
mazmarcan: we do restore the timers with TGE=1, always. but if it was a level, that interrupt would keep firing forever, because I'm not using this mask *at all*.#asahi · 2021-03-28 11:01
mazunless I've messed up my experiment (not unlikely), it looks like it.#asahi · 2021-03-28 10:58
mazI'm talking about the HV registers here (though the guest regs behave the exact same way).#asahi · 2021-03-28 10:57
mazregisters*#asahi · 2021-03-28 10:57
mazthat's not the point. FIQ acts as a edge triggered interrupt, and the ack is the read of the CNT registerse.#asahi · 2021-03-28 10:56
mazTGE=1, I'm on the host.#asahi · 2021-03-28 10:56
mazread ISR before reading CNTP_CTL, F is set. read it again, it is clear.