19:19 Lyude: karolherbst: think we should cc those pascal rpm fixes to stable? I should probably go test it a little more thoroughly on the laptops that we've got lying around, but I think if we can verify this is a safe change (and I'd be surprised if it wasn't tbh) it's probably stable material
19:19 karolherbst: probably
19:20 karolherbst: but again, I kind of prefer a fix inside the pci subsystem :p
19:20 Lyude: mm, once i've finished looking through all of these I'll do a quick peek to see if I can figure out where, if anywhere, the pci subsystem can/should do this
19:20 karolherbst: there is obviously something wrong on the pcie level
19:21 karolherbst: and most likely the GPU and the controller can't really communicate
19:21 karolherbst: Lyude: there might be something to reprobe the connection... but I didn't find anything to do that
19:21 karolherbst: maybe the hardware can, but it's not wired up?
19:21 karolherbst: dunno
19:23 Lyude: karolherbst:looking at the ode in drm/nouveau/nvkm/subdev/pci/gk104.c, I think whether or not this is a bug in the pcie subsystem might depend on what the registers we use in gk104_pcie_max_speed() and gk104_pcie_set_link_speed() actually are (e.g. if they're part of standard PCIe config space, or really just some random mmio bits)
19:24 karolherbst: starting with kepler those are more or less random mmio bits
19:24 karolherbst: this was different with pre kepler
19:25 karolherbst: some still are within the pci config space though
19:25 karolherbst: Lyude: the pci config space is mapped at 0x88000
19:25 Lyude: mmmm yeah, I think you might be right thwen
19:26 karolherbst: anyway, that's what nvidia is doing as well
19:26 karolherbst: and with disabled reclocking we don't even touch those except for the v1 -> v2 transition
19:26 karolherbst: and this is only important for tesla boards
19:26 karolherbst: might have seen it on fermi as well
19:26 karolherbst: (and on kepler, if you mess with the device prior loading nvidia)
19:28 Lyude: ...hm, does the pcie subsystem even have functions to change the link rate?
19:28 Lyude: I see some to query it at least
19:30 karolherbst: dunno
19:30 Lyude: doesn't look like it does
19:30 karolherbst: I think it has
19:30 Lyude: or I'm not finding it with grep anyway
19:31 Lyude: karolherbst: what i'm the most curious about here is whether or not there actually is a standard way to change this, because if there isn't that's probably why the PCI subsystem doesn't do it on it's own, maybe there's some device-specific magic here
19:32 karolherbst: oh, but the speed isn't the actual problem here
19:32 karolherbst: I think something like this happens: controller assumes the link is at 2.5, because that's what the device used before suspending
19:32 karolherbst: now the device comes up and assumes 8.0
19:32 karolherbst: controller still thinks 2.5
19:32 karolherbst: => bad things happen
19:33 Lyude: karolherbst: mm, makes sense
19:33 karolherbst: anyway, I am no expert with PCIe, so that's why I would rather ask some PCIe people about that
19:33 Lyude: yeah, agreed
19:33 karolherbst: Lyude: the big issue is, that the GPU itself change the link
19:34 karolherbst: there is no driver interaction (except my workaround)
19:34 Lyude: karolherbst: that makes me curious of something
19:34 karolherbst: we have the signed devinit script inside the vbios
19:34 karolherbst: and a signed pmu image with a devinit script parser
19:34 Lyude: karolherbst: have you seen drivers/pci/pcie/bw_notification.c
19:34 karolherbst: and we launch that while doing devinit
19:34 karolherbst: Lyude: nope
19:36 karolherbst: Lyude: looks like it is new?
19:36 Lyude: karolherbst: well anyway, let's cc linux-pci@vger.kernel.org and Bjorn Helgaas <helgaas@kernel.org> on the next respin of this
19:37 karolherbst: yeah
19:39 karolherbst: Lyude: ahh, found the upstream bug again: https://bugzilla.kernel.org/show_bug.cgi?id=156341
19:39 karolherbst: the cc list alone :D
19:40 Lyude: phew
19:45 Lyude: karolherbst: alright-feel free to respin and add ccs
20:05 karolherbst: Lyude: checkpatch didn't complain about the bool field
20:05 karolherbst: maybe somebody was ranting on the mailing list without being able to back that up :p
20:10 Lyude: karolherbst:oh good
20:10 Lyude: karolherbst: because I was getting sick of using u8
20:10 karolherbst: :D
20:12 karolherbst: sent
22:16 karolherbst: Lyude: actually, I think the issue has to be the controller. Because you know, GPUs can kind of detect if the controller does't support a certain PCIe link speed and they automatically fall back to a lower one
22:16 karolherbst: allthough the mechanism here might be a different one
22:17 Lyude: karolherbst: should be pretty easy to check, but is it actually likely that all of these different laptops are using basically the same PCI controller?
22:18 karolherbst: probably :D
22:19 Lyude: mhm, yeah
22:19 karolherbst: I mean they use the one from the Intel CPU
22:19 karolherbst: no?
22:19 karolherbst: that's usually where the GPU is connected with
22:19 Lyude: ah right, probably
22:20 Lyude: karolherbst:https://paste.fedoraproject.org/paste/WcNgVtncegWOlNGOFlvv3g
22:21 karolherbst: well, in the past only the CPUs PCIe controller provided PCIe 3.0
22:21 karolherbst: so GPUs were connected to that one
22:21 karolherbst: even on desktops
22:22 HdkR: Lyude: I'm jelly of your Titan Ridge controller
22:22 Lyude: HdkR: hehe, i've got plenty more!
22:22 Lyude:has a _lot_ of laptops at her disposal at work :)
22:22 HdkR: This Razer kept using Alpine Ridge for w/e reason :|
22:23 Lyude: seriously?
22:23 Lyude: really wish I still had contacts at razer :|
22:24 HdkR: They also use the worst Type-C physical connectors in the world of course. So many things to complain about
22:24 Lyude: HdkR: do they? I never had issues with my 2016 model
22:25 HdkR: It's an issue with this particular 13" model. They are so loose that it refuses to charge sometmes
22:25 Lyude: yeesh
22:26 HdkR: Also the friggin keyboard make me drop keys due to inconsistent key presses
22:26 Lyude: oh double yeesh, I don't think my keyboard ever had issues with that
22:26 Lyude: the worst I've had is one key on here is slightly broken, but it still works well enough
22:27 HdkR: It's just disappointing more than anything
22:27 karolherbst: Razor is spending 70% on the brand anyway :p
22:28 karolherbst: uhm, *Razer
22:28 HdkR: Sure, but it was a comparison of like 15 different laptops and it came out on top :P
22:29 HdkR: mixture of features and price was right
22:32 Lyude: karolherbst: tbh-I would usually agree, but as dumb as this laptop is sometimes I really like my razer machine
22:32 Lyude: this one at least is build pretty solid
22:32 Lyude: *built
22:32 karolherbst: at least something
22:32 Lyude: also they put effort into the touchpad
22:33 HdkR: Yea, build quality is nice on this one as well, Touchpad is almost as good as Surface Book, screen is decent
22:33 karolherbst: it's kind of sad, but there aren't that many "high quality" laptop models out there :/
22:33 Lyude: yeah :\
22:33 karolherbst: Lenovo is super weak the last ten years
22:33 karolherbst: and Razer has this weirdo situation
22:33 Lyude: and synapse
22:34 Lyude: as bizarre as this might be, right now I think some f the zbooks are the most solid machines I've touched in a while
22:34 Lyude: the very recent ones at least
22:34 karolherbst: I am quite happy with my dell xps, allthough that one is like expensive overall :/ and it has this stupid camera :D
22:34 karolherbst: Lyude: the HP ones?
22:34 Lyude: karolherbst:mhm, note: HP is garbage outside of enterprise
22:34 karolherbst: sure
22:34 karolherbst: same as Dell :p
22:34 HdkR: Stupid camera placement killed about half my options when looking at things
22:35 karolherbst: Lyude: I had a very nice Asus laptop once, but that was a 1k+ model as well
22:35 HdkR: Turns out when you're doing remote meetings, staring at someone's fingers and up their nose isn't good...
22:35 karolherbst: HdkR: I have it disabled all the time
22:36 karolherbst: but yeah...
22:36 karolherbst: I like how "small" the XPS is
22:36 HdkR: I used to do like three remote meetings a week :D
22:36 karolherbst: the border around the display is really small
22:36 karolherbst: HdkR: only three?
22:36 karolherbst: :p
22:36 HdkR: On average, there we many more a lot of weeks :)
22:36 HdkR: Dumb meeting heavy company
22:37 karolherbst: still, I'd be happy to have only three on avarage :p
22:37 Lyude: HdkR:dumb beurocratic everything heavy company, honestly lol
22:37 HdkR: :D
22:37 karolherbst: HdkR: I have 4.75 reoccuring ones over the month each week
22:37 karolherbst: 4 each week, 1 bi weekly, 1 monthly
22:37 HdkR: Nice
22:37 karolherbst: yeah, well ...
22:37 karolherbst: even though they are kind of good meetings in general
22:38 Lyude: yikes o_O, I have 2 bi weekly, and 1 weekly
22:38 karolherbst: Lyude: good for you :p
22:38 Lyude: and i choose to go to one of those bi weekly meetings
22:38 HdkR: I now average 0 per month
22:38 Lyude: :p
22:39 karolherbst: Lyude: this week is like 5 hours of meetings, and I have one public holiday :(
22:41 Lyude: oh hey! I didn't know we had a holiday this month :)
22:41 Lyude: (in the US apparently as well)
22:41 karolherbst: yeah, people celebrate they fought of the nazis :p
22:42 Lyude: hell yeah
22:42 HdkR: What is a holiday? :P
22:43 karolherbst: HdkR: :p
22:43 karolherbst: in the US you are so punished anyway :p
22:43 karolherbst: I have like 14 public holidays over the year
22:44 HdkR: I typically work holidays even when not self-employed so the tend not to matter :P
22:44 karolherbst: I have the issue that I have to take all my PTOs in a year
22:44 karolherbst: by law
22:45 karolherbst: and I already fall behind again :( current_day_of_the_year / days_within_the_year > ptos_taken/ total_ptos
22:45 HdkR: oh no
22:46 karolherbst: yeah, you wouldn't know this problem :p
22:46 HdkR: So PTO for a month, come back and have forgotten everything you were working on? :D
22:46 karolherbst: heh?
22:46 karolherbst: a month wouldn't be enough
22:46 karolherbst: allthough, maybe it would..
22:47 karolherbst: ohh, right, only 20 days of pto left :D
22:47 HdkR: Three weeks of lounging
22:47 karolherbst: I usually take one or two weeks before/around christmas
22:47 karolherbst: and use the other days for whatever comes up