20:51 agd5f: vliaskov, karolherbst: there is nothing AMD specific that needs to happen for d3cold. It comes down to what each dGPU does to support it. With AMD dGPUs, we control the dGPU power via a GPIO which is controlled via the ACPI _PR3 method. There is not some magic chipset control for this.
20:53 karolherbst: in theory yes, in practise no
20:53 karolherbst: nvidia has a list of workarounds they apply depending on the bridge controller/root port
20:53 karolherbst: we need something simlar for Intel as well
20:54 karolherbst: No idea what's actually broken, but it's something on the firmware/hardware level
20:56 karolherbst: so what happens is, that the hardware/firmware fails to bring up the GPU and it appears to be completely dead after resuming from d3cold
20:56 agd5f: karolherbst, probably depends on what the _PR3 method actually does. AFAIK, each dGPU vendor provides them for the platform.
20:57 karolherbst: sure, but it's broken
20:57 agd5f: it might to something like send a signal to the nvidia driver rather than actually powering down the dGPU or something like that.
20:58 karolherbst: I asked nvidia as well, and they also said "it should just work, nothing needed" and then I saw their huge lists of PCI quirks
20:58 agd5f: ah
20:58 karolherbst: that's our workaround for intel which seems to actually work: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpu/drm/nouveau/nouveau_drm.c?h=v6.1-rc2#n699
20:59 karolherbst: but yeah.. I suspect we might be missing _something_ but nobody tells us anything, soo.....
20:59 karolherbst: others spend months/years debugging this on an ACPI level even
21:55 jekstrand: skeggsb_: What do I need for firmware to use your 01-gsp-rm branch?
21:56 jekstrand: skeggsb_: 2060 if it matters
23:30 skeggsb: jekstrand: 515.48.07
23:35 jekstrand: skeggsb: Got it!