#asahi · 2021-11-16 09:20
marcanmini: it has error reporting#asahi · 2021-11-16 09:13
marcanthe idea that we can stick billions of bits in chunks of silicon and somehow have zero errors is completely insane#asahi · 2021-11-16 09:13
marcanit is completely ridiculous that *every other storage technology* has used ECC for decades, and ECC RAM is still not standard#asahi · 2021-11-16 09:13
marcanfirefox317: for not getting random bit errors because RAM bits are too small these days#asahi · 2021-11-16 09:10
marcan(on neither M1 nor M1 Pro)#asahi · 2021-11-16 09:10
marcan(there's an ECC disable bit in AMCC, and it seems to not be set...)#asahi · 2021-11-16 09:09
marcanif so, about time! (and now I don't want to use any other consumer machine)#asahi · 2021-11-16 09:09
marcanI think these machines might all have ECC RAM?!#asahi · 2021-11-16 09:09
marcanhuh, I was wondering after someone mentioned rowhammer on twitter...#asahi · 2021-11-16 08:40
marcanthen he stopped talking to me and 2 days later announced they'd "ported" Linux to the M1#asahi · 2021-11-16 08:40
marcanthen he got mad at me and I mentioned that in that state it would only be useful at docs#asahi · 2021-11-16 08:40
marcanI mentioned the old sandcastle tree was not upstreamable on a reply somewhere (it wasn't even signed off...)#asahi · 2021-11-16 08:40
marcan< tmlind> usable mostly as hardware documentation, just like any other arm vendor trees.. <- this is what got Chris mad at me, btw#asahi · 2021-11-16 08:38
marcan(and this is why I was saying we've tried on the USB issue, believe me :p)#asahi · 2021-11-16 08:38
marcanfound some interesting things in the new interrupt controller, like CPU cluster priority and delivery delays/timeouts (not 100% certain on everything yet, but it's nice to know these features exist)#asahi · 2021-11-16 08:37
marcanthis is why I poke around registers to see what I find, I don't just do what macOS does#asahi · 2021-11-16 08:37
marcanyes :)#asahi · 2021-11-16 08:37
marcanbut when you're upstreaming a port... it actually matters that you understand the hardware and drive it properly#asahi · 2021-11-16 08:37
marcanand for a virtualization product... that's kind of all you need in the end, as long as macOS runs (the m1n1 hypervisor doesn't take a very different approach there, given it's a research tool)#asahi · 2021-11-16 08:36
marcanI know that approach well because I've been there; hack on it until it works, make the PoC#asahi · 2021-11-16 08:35
marcanand they also had confused sysreg names for different CPU generations in their attempt at porting over the chicken bits (which was dangerously close to the original xnu asm for the same)#asahi · 2021-11-16 08:35
marcan:p#asahi · 2021-11-16 08:35
marcanthey had some GPIO_CONFIG_DONE bit that was actually GPIO_INPUT_BUFFER_ENABLE#asahi · 2021-11-16 08:35
marcanyeah, but some of it is outright wrong :-)#asahi · 2021-11-16 08:34
marcanmight work for virtualizing iOS, but not great for linux#asahi · 2021-11-16 08:34
marcanbut I think they're taking the approach that if it runs it's good enough#asahi · 2021-11-16 08:34
marcangiven their entire business is a hypervisor you'd think they'd spend time figuring out register definitions properly...#asahi · 2021-11-16 08:34
marcanshrug, given the quality of their port I'm not too worried#asahi · 2021-11-16 08:33
marcanthat's how we know :)#asahi · 2021-11-16 08:33
marcanthere's one chicken bit set with a comment above it (back when apple was releasing source for this) saying it was found on linux#asahi · 2021-11-16 08:33
marcanthey do#asahi · 2021-11-16 08:33
marcanbut that would go directly against their business interests#asahi · 2021-11-16 08:32
marcanif they really wanted to participate they'd share whatever internal hardware documentation they have#asahi · 2021-11-16 08:32
marcanit kind of goes against their business model so I'm not surprised#asahi · 2021-11-16 08:28
marcanthey've been CCed on a ton of the submissions too and never said a word#asahi · 2021-11-16 08:28
marcan(much less maintain it)#asahi · 2021-11-16 08:28
marcanso it seems the whole releasing it and saying they'll upstream it was pure PR talk and they never had any intention to do that#asahi · 2021-11-16 08:28
marcansame reason Apple has an internal Linux port to these things#asahi · 2021-11-16 08:28
marcantmlind: they admitted in their BlackHat talk that the whole thing was just intended as a hypervisor validation platform#asahi · 2021-11-16 08:14
marcanI should commit this, I cleaned up a bunch of the pmgr stuff yesterday#asahi · 2021-11-16 08:14
marcanI have a script to convert the ADT device list to pwrstate DT nodes, but it still needs manual munging because I don't fully understand/agree with how apple decides what should be powered on and off on boot (they have a bit for that, but they set it on stuff that isn't necessary like UARTs, but they also set it on stuff that absolutely is like SPMI...)#asahi · 2021-11-16 08:13
marcanso anything else from pmgr is going to get separate nodes#asahi · 2021-11-16 08:13
marcan(well, to 16k probably works better)#asahi · 2021-11-16 08:12
marcanI've also decided the syscon nodes are going to be strictly for this, and sized after the pstate reg blocks (so highest reg rounded up to 0x100)#asahi · 2021-11-16 08:12
marcanon the plus side I was right, zero changes in t6000 there that we need to worry about#asahi · 2021-11-16 08:12
marcansven: for the next pmgr submission I'm thinking of actually dumping in the entire set of pwrstate nodes (like I tested yesterday for t6000), wonder what they'll say :D#asahi · 2021-11-16 08:10
marcanso that's entirely plausible#asahi · 2021-11-16 08:10
marcanah, but AVD at least does claim to power gate itself in the pstate reg#asahi · 2021-11-16 08:10
marcanI don't see a power domain that screams AVD/MMX though#asahi · 2021-11-16 08:08
marcanI wonder if this is partially because that firmware is lost if the AVD is power gated; wouldn't surprise me