13:44 karolherbst: ohh, cool, now I am close fixing the mt issues
13:53 RSpliet: karolherbst: \o/
13:53 karolherbst: now helgrind starts to annoy me about mesa race conditons :D
13:54 karolherbst: okay.. cool, the remaining stuff seems to be within libdrm... that's probably some bufctx shit going wrong
13:55 RSpliet: I'm quite convinced (despite not backing it up :-P) that mt is at the root of VirtualBox hanging my machine when I enable 3D guest acceleration for Windows 8 or 10 guests. Hope I can soon confirm it was O:-)
13:55 karolherbst: yeah, probably
13:55 karolherbst: RSpliet: try this branch: https://github.com/karolherbst/mesa/commits/mt_fixes_take2
13:56 karolherbst: well.. it only works for nvc0+
13:56 karolherbst: nv50/nv30 are broken right now
13:56 karolherbst:had to add more locks than he hoped he would have to
13:57 karolherbst: but I doubt I can get rid of those two without doing a big rework on how we track hw state
13:57 RSpliet: karolherbst: I don't really use a Windows VM very often... last time I booted my Win8 VM was a few weeks ago, discovered that it was missing updates since 2015
13:58 karolherbst: right.. but what did you had to do to get the crash?
13:58 karolherbst: maybe just idling the VM is enough :p
13:58 RSpliet: so it might be a while before I'm compelled to update my mesa branch and actually go through the works, esp with that PhD deadline shifting closer
13:58 karolherbst: ahh, I see
13:58 RSpliet: karolherbst: really, booting was all it takes.
13:58 karolherbst: ohh, okay
13:58 RSpliet: Don't think I even had to log in to the guest VM
13:58 karolherbst: how relible was the crash?
13:58 RSpliet: 100%
13:58 karolherbst: ups
13:59 RSpliet: only worked around by disabling 3D accel, which for a while prevented me from upgrading to Win10 in the first place.
13:59 RSpliet: Apparently these days you may ignore the "your non-3D accelerated GPU is incompatible with Win10" upgrade error
14:00 karolherbst: :)
14:00 karolherbst: it was worse with 8.1
14:00 karolherbst: but these days they have a good enough fallback path
14:00 RSpliet: w/o 3D accel it's slow as heck though
14:00 karolherbst: they need it for their server anyway
14:00 karolherbst: ohh
14:00 karolherbst: just switch to the server UI
14:01 RSpliet: I switched to my Windows XP VM. Slow too, but not unbearable
14:01 karolherbst: :D
14:01 RSpliet: All I needed it for was to try and hack my car anyway, those tools tend to be ancient
14:01 karolherbst: but with your XP one, you get one trojan per second... not sure if that's desirable
14:02 RSpliet: There's at least one layer of NAT in between (the VMs network cards)
14:02 karolherbst: doesn't matter, XP is just fundamentally broken :p
14:02 karolherbst: I wouldn't be surprised if you it gets infected by checking for updates alone
14:03 RSpliet: "Your computer got infected with Windows XP"
14:03 karolherbst: like all the ATMs are :p
14:03 karolherbst: well... not all
14:06 karolherbst: ehh: https://gist.githubusercontent.com/karolherbst/efdafb63abb47f9ed498125870078345/raw/93e896550995255dc7844be6642ff43164f3a0dc/gistfile1.txt
14:06 karolherbst: I really hate that we need a pushbuffer for the screen
14:06 karolherbst: that causes that much of a headache
14:07 karolherbst: mhh, although nvc0_miptree_transfer_map _has_ a context reference
14:07 karolherbst: weird
14:07 karolherbst: ufff
14:08 karolherbst: we set dev->vram_limit and dev->gart_limit on every submit
14:08 karolherbst: yeah.. that's not gonna to end well
14:09 karolherbst: at least that's the only common race condition left.. oh well