14:03 tagr: anholt: typically the latest U-Boot is what I'd try on any device, really, unless if it doesn't work, in which case some recent release should do the trick
14:03 tagr: we used to test this fairly exhaustively, but the infrastructure that we used for that was at some point recycled, so there's no regression testing for U-Boot at the moment, I think
14:04 tagr: that's perhaps something that we should add back at some point, but it's difficult to find the resources for that kind of work =(
14:06 tagr: anholt: regarding that boot log you posted for Jetson Nano, that doesn't look like an IOMMU issue, though
14:08 tagr: that "Failed to set up IOMMU for device" error is actually expected right now (that's actually from the DMA/IOMMU glue which we currently can't yet initialize, otherwise the bootsplash will cause the boot log to get spammed with IOMMU fault messages)
14:08 tagr: an IOMMU related issue in Nouveau would typically show up as a slew of DATA errors of some sort
14:09 tagr: these timeouts I vaguely recall seeing very early on and it might have been related to problems with VPR/WPR
14:14 tagr: looks like that 0x137000 register is GPCPLL_CFG, but I have no idea why reads to that would timeout
15:33 RSpliet: tagr: I've never looked into this proper, but high level ARM experience stupid questions then: I presume that on Tegra clocks come from the SoC-wide clock generator. Depending on the gen these PLL regs may not actually exist on tegra? Or.. is the external clock even enabled?
15:42 tagr: RSpliet: yeah, the clock ultimately comes from the SoC's CAR (clock and reset) block as far as I know, but the GPCPLL_CFG register still seems to exist
15:42 tagr: at least I don't see a reason why acourbot would've added the code all these years ago if these weren't relevant
16:00 karolherbst: yeah.. and that code also works, just nothing of it is really stable
16:01 karolherbst: tagr: we might end up reading too early though
16:51 anholt: tagr: "nouveau 57000000.gpu: bus: MMIO read of 00000000 FAULT at 137000 [ TIMEOUT ]" is the problem.
16:52 anholt: so maybe I've been calling it the wrong thing, but it's this fault during power on.
23:41 pabs: tagr: do the Linaro folks no longer do u-boot testing on their LAVA instance?