ChanServ changed the topic of #dri-devel to: <ajax> nothing involved with X should ever be unable to find a bar
columbarius has joined #dri-devel
quantum5 has joined #dri-devel
xyene has joined #dri-devel
co1umbarius has quit [Ping timeout: 480 seconds]
mbrost has joined #dri-devel
Company has quit [Quit: Leaving]
fxkamd has quit [Remote host closed the connection]
fxkamd has joined #dri-devel
camus has joined #dri-devel
camus1 has quit [Ping timeout: 480 seconds]
nchery has quit [Quit: Leaving]
pendingchaos_ has joined #dri-devel
pendingchaos has quit [Ping timeout: 480 seconds]
tarceri has joined #dri-devel
pendingchaos_ is now known as pendingchaos
shashank1202 has joined #dri-devel
sukualam has joined #dri-devel
sukualam has quit [Quit: Quit]
sukualam has joined #dri-devel
sukualam has quit [Ping timeout: 480 seconds]
<graphitemaster> So, if the AMDGPU KMD supports MCBP mainly for virtualization, what's preventing the other KMDs from supporting it too. I feel like once you have this part done, it's just a matter of handling the user-space side of things inside Mesa.
camus has quit [Remote host closed the connection]
camus has joined #dri-devel
alyssa has joined #dri-devel
<alyssa> does drm_kms_helper_hotplug_event necessarily cause an atomic_flush()?
<alyssa> it looks like it does for fbdev and mutter/xorg, but not for plain xorg
<alyssa> (and doesn't for sway I think)
* alyssa needs a flush to get something on screen after hotplug
mbrost has quit [Remote host closed the connection]
<alyssa> thought userspace would always kick that, but no..
<alyssa> I can't just call atomic_flush myself since I don't have the drm_atomic_state
<imirkin> relying on userpsace to do things is always a losing proposition
<alyssa> yeah, srsly..
<alyssa> esp if the userspace in question is xorg..
<imirkin> xorg is actually pretty reasonable...
<imirkin> if you're having trouble with xorg, your expectations are off
<imirkin> it's sorta the definition of correctness, for better or for worse, for the "legacy" kms api usage
<alyssa> nod
<imirkin> (i.e. non-atomic)
<imirkin> [there were attempts to make it use atomic kms, but they were pretty half-hearted and deleted]
* alyssa trying to understand how this possibly works for any other driver ..
<airlied> alyssa: for xorg things should happen via atomic emulation I thought
<alyssa> (then again id believe it if this exact use case is broken in some arm drivers)
<imirkin> yeah, X doesn't call atomic_anything
<alyssa> airlied: Emulation is working as promised
<imirkin> if you're talking about exposure, X doesn't provide any exposure info
<airlied> so the kernel should implement the legacy APIs via the atomic helpers stuff
<airlied> alyssa: are you just running a bare X server?
<alyssa> airlied: yeah, that's what's hitting this
<airlied> alyssa: you need to run something on the X server like a DE
<alyssa> (case of "send a hotplug to userspace but then nothing happens ===> black screen unless the user runs xrandr")
<alyssa> xorg+i3
<airlied> X doesn't deal with hotplug
<alyssa> so this is a known broken wontfix situation that i've been struggling to debug for hours? lovely!
<imirkin> X deals with a bit of hotplug...
<imirkin> it'll pick up new connectors and stuff
<imirkin> but it won't do anything other than making it available
<imirkin> up to userspace to request stuff
<airlied> Displayport is a bit too much for poor X
<airlied> so X provides them to the clients
<imirkin> some DE's will listen to those events and make use of the new items
orbea1 has joined #dri-devel
<alyssa> airlied: fwiw "hotplug" in this case includes "power button on the monitor"
<airlied> clients will get randr events
<airlied> alyssa: yeah displayport will do that to you
<alyssa> ah right yes this is a displayport upstream ugh
<imirkin> airlied: you added the dynamic DP stuff to modesetting a while back though ... should work OK
* airlied remembers keithp did some xrandr hacks
<airlied> imirkin: the problem is with link negotiation etc
<imirkin> and i copied it to nouveau :)
<imirkin> hm?
<airlied> like it works as long as you have a DE running
<imirkin> or you run xrandr to tell Xorg what you want
<airlied> but if you don't have a DE running and link training fails or other wierness
orbea has quit [Ping timeout: 480 seconds]
<airlied> then you won't get an image
<imirkin> yea of course
<imirkin> why would you expect an image?
<airlied> because it used to work with VGA :-P
<imirkin> oh, if the primary is DP, and link training fails?
<imirkin> then that's a very sad panda
<alyssa> displayport is kinda sad
<alyssa> for my understanding of the stack
<imirkin> on the bright side, you can have hubs and loops
<airlied> spanning tree, no loops for you :-P
<alyssa> k it's definitely sleep o'clock
<airlied> alyssa: I see a thing called autorander that might do it
<alyssa> airlied: thanks for digging me out of the rabbit hole
* alyssa opens gnome up again and goes back to not caring
<airlied> I think I trolled keithp once about making xrandr --daemon be a thing for legacy users
<mareko> anholt: not really, there's "umr -t" (v), which shows GPU memory usage for all processes, but all of them are labeled as Xorg, which is useless
<zmike> jenatali: ping
<alyssa> well, that's one more bug I can cross off my "blockers to sending the RFC to dri-devel" mental todo list
<zmike> jenatali: in vs, how do bitfields in unions work re: sizing? https://gitlab.freedesktop.org/mesa/mesa/-/jobs/14657023
<jenatali> Uh I'd expect them to work sanely
<jenatali> Give me a minute, lemme take a look at the change
<zmike> ty
orbea1 has quit []
<zmike> not sure if it's the bitfields or what, but need to get this fixed up and merged tonight
orbea has joined #dri-devel
<jenatali> zmike: At first glance, I *suspect* the problem is the mixed types in the bitfield struct
<jenatali> Anywhere I've seen bitfields used to pad out, they all have the same base type I think
<zmike> you mean using uint32_t for all the members would be more likely to pass?
<jenatali> Let me see if I've got the radv build environment set up locally... I think I did at one point
<jenatali> Yeah that's what I mean
<zmike> would be great if you could test locally before I waste more ci time :)
<jenatali> Yeah gimme a minute
<jenatali> zmike: I don't have the AMD native LLVM stuff in my local LLVM build to be able to build RADV :(
<zmike> :/
<jenatali> Let me fight with godbolt a bit
<zmike> I'm sure this will be fine, I'll just tell the compiler you said it was supposed to work
fxkamd has quit []
<airlied> I seem to remember it needing to be the same type
<jenatali> zmike: Yeah, godbolt says that union is 20 bytes on MSVC
<zmike> whoa
<zmike> so same type fixes that, huh
<jenatali> Play around with it there instead of wasting CI's time ;)
<zmike> this is a good idea
* zmike has too many tools to keep track of
<jenatali> IIRC it's a matter of changing alignment, switching between native 8-bit types (1-byte alignment) and bitfields of 32-bit types with 4-byte alignment breaks the bitfield packing
<zmike> hm looks like changing the types gives it a size of 8?
<jenatali> zmike: https://godbolt.org/z/fsxraW16T works
<zmike> huh weird
<zmike> uint32_t doesn't work
<jenatali> Using it consistently does
<jenatali> https://godbolt.org/z/qx8v89eq3 works too
<zmike> wat
<zmike> I have the exact same thing here and it says 8
<jenatali> Link? Share button in the top-right
<jenatali> zmike: Bitfield size on key_size
<jenatali> You didn't add : 8
<zmike> oh
<zmike> obviously
<jenatali> :)
<zmike> this is what happens when I try to code after pub
<zmike> thanks for the help!
<jenatali> Anytime
kts has quit [Ping timeout: 480 seconds]
<HdkR> zmike: Do you need a struct packing verifier that works across architectures? :P
<HdkR> zmike: https://github.com/FEX-Emu/FEX/blob/main/Scripts/StructPackVerifier.py This is the garbage thing I whipped together for FEX. Has a Windows/Mingw verifier (which I currently don't use)
<HdkR> Uses libclang for verification
<HdkR> Needs some struct annotating and setup, but has saved me from many headaches so far
<zmike> not this time, just trying to smash this radv MR in before branchpoint
<zmike> usually I just use pahole
<HdkR> Oh, interesting. Does that work across arch and ABI differences?
<zmike> shrug
<zmike> works for me :)
<HdkR> Neat, maybe I'll need to look at it
<HdkR> Didn't know about that tool
<airlied> jenatali: not convinced it'll be a neat path forward or not, clover is kinda convoluted in this area :-P
<jenatali> airlied: yeah I've been meaning to make more progress on that front in spare time
<jenatali> I'll take a closer look later
<airlied> jenatali: the logging and then c++ vs c strings are the pain points so far
<airlied> just giving it a test_basic run now
shashank1202 has quit [Quit: Connection closed for inactivity]
muhomor has quit [Ping timeout: 480 seconds]
camus1 has joined #dri-devel
camus has quit [Ping timeout: 480 seconds]
<airlied> jenatali: ah spirv version mismatch is the other one, I'ev left it in a branch clover-wip-use-clc for now
<jenatali> airlied: what mismatch?
<airlied> just between how the clc interface defines them and how clover has them
<airlied> have to just line it up, likely pretty trivial, I just hardcoded spirv max in my wup
<airlied> wip
<jenatali> Oh ok, not too bad
* airlied is also experimenting with -fdeclare-opencl-builtins
<airlied> and using opencl-c-base.h instead of opencl-c.h
<jenatali> airlied: it's still broken for generic address space for cl3.0
<jenatali> You fixed the headers but not the builtins
<airlied> yeah I wanted to see what was left to do there
<jenatali> Cool. I'd love to have to bundle one less header
<airlied> it seems to speed up compile as well
<jenatali> Yeah makes sense
<jenatali> I wish I had so much more time to devote to these kinds of things
<airlied> karolherbst: I rebased your images branch
aravind has joined #dri-devel
camus has joined #dri-devel
camus1 has quit [Read error: Connection reset by peer]
Duke`` has joined #dri-devel
shashank_sharma has quit [Remote host closed the connection]
shashanks has joined #dri-devel
shashanks has quit [Remote host closed the connection]
sdutt has quit [Read error: Connection reset by peer]
camus has quit [Read error: Connection reset by peer]
camus has joined #dri-devel
rpigott has quit [Read error: Connection reset by peer]
rpigott has joined #dri-devel
macromorgan is now known as Guest2666
Guest2666 has quit [Read error: Connection reset by peer]
macromorgan has joined #dri-devel
itoral has joined #dri-devel
tzimmermann has joined #dri-devel
macromorgan has quit [Read error: Connection reset by peer]
pnowack has joined #dri-devel
mattrope has quit [Read error: Connection reset by peer]
Duke`` has quit [Ping timeout: 480 seconds]
mlankhorst has joined #dri-devel
danvet has joined #dri-devel
Danct12 has quit [Ping timeout: 480 seconds]
camus1 has joined #dri-devel
itoral has quit [Remote host closed the connection]
camus has quit [Ping timeout: 480 seconds]
itoral has joined #dri-devel
<danvet> karolherbst, don't we just need a marge instance with an s/Part-of:/Link:/ and done?
<danvet> but imo even without that paint applied it's good already
<danvet> airlied, ^^
<danvet> iirc you'd need a separate marge instance for that bikeshed, and that seems a bit overkill
<danvet> or you do the bikeshed as part of the "push to drm-misc"
<danvet> at least for now
lemonzest has joined #dri-devel
itoral has quit [Remote host closed the connection]
Danct12 has joined #dri-devel
itoral has joined #dri-devel
lemonzest has quit [Remote host closed the connection]
lemonzest has joined #dri-devel
bbrezillon has quit [Ping timeout: 480 seconds]
mripard has quit [Ping timeout: 480 seconds]
itoral has quit [Remote host closed the connection]
itoral has joined #dri-devel
itoral has quit [Remote host closed the connection]
itoral has joined #dri-devel
camus has joined #dri-devel
camus1 has quit [Read error: Connection reset by peer]
lemonzest has quit [Read error: No route to host]
frieder has joined #dri-devel
ddevault has quit [Quit: Why do I even put this quit message in if I never quit]
ddevault has joined #dri-devel
rasterman has joined #dri-devel
vivijim has joined #dri-devel
calebccff has quit [Quit: ZNC 1.8.2 - https://znc.in]
FLHerne has quit [Quit: There's a real world out here!]
Daanct12 has joined #dri-devel
Plagman has quit [Remote host closed the connection]
samuelig has quit [Quit: Bye!]
pepp has quit [Remote host closed the connection]
xxmitsu has quit [Remote host closed the connection]
dj-death has quit [Remote host closed the connection]
Danct12 has quit [Ping timeout: 480 seconds]
Ziemas has quit [Remote host closed the connection]
samuelig has joined #dri-devel
javierm has quit [Quit: leaving]
javierm has joined #dri-devel
javierm has quit []
javierm has joined #dri-devel
Company has joined #dri-devel
FLHerne has joined #dri-devel
muhomor has joined #dri-devel
pushqrdx has quit [Ping timeout: 480 seconds]
camus1 has joined #dri-devel
pushqrdx has joined #dri-devel
camus has quit [Ping timeout: 480 seconds]
tursulin has joined #dri-devel
lynxeye has joined #dri-devel
gawin has joined #dri-devel
Ziemas has joined #dri-devel
calebccff has joined #dri-devel
xxmitsu has joined #dri-devel
mripard has joined #dri-devel
bbrezillon has joined #dri-devel
dj-death has joined #dri-devel
pepp has joined #dri-devel
Daanct12 has quit [Remote host closed the connection]
gpuman has joined #dri-devel
camus has joined #dri-devel
Ahuj has joined #dri-devel
camus1 has quit [Ping timeout: 480 seconds]
gpuman_ has joined #dri-devel
gpuman has quit [Ping timeout: 480 seconds]
shadeslayer has quit [Quit: The Lounge - https://thelounge.chat]
xexaxo_ has quit [Ping timeout: 480 seconds]
<pq> alyssa, why would you want to put something on the screen after hotplug, if userspace is not telling you to put something on that new screen? I mean, you don't. Or is that connector already lit up by userspace explicitly before it was connected?
gpuman has joined #dri-devel
gpuman_ has quit [Ping timeout: 480 seconds]
heat has joined #dri-devel
<gawin> anholt: have you thought about enabling this code again, but only for older gpus (pre glsl 1.30)? https://gitlab.freedesktop.org/mesa/mesa/-/commit/e21b9f1f19d2345026a7fbe095a776d0b64557ec
<gawin> not sure how about i915g, but for r300 saving any potential temporary is beneficial
lemonzest has joined #dri-devel
gpuman_ has joined #dri-devel
Danct12 has joined #dri-devel
gpuman has quit [Ping timeout: 480 seconds]
xexaxo_ has joined #dri-devel
heat has quit [Remote host closed the connection]
gpuman has joined #dri-devel
gpuman_ has quit [Ping timeout: 480 seconds]
JTL1 has joined #dri-devel
JTL is now known as Guest2710
JTL1 is now known as JTL
Guest2710 has quit [Ping timeout: 480 seconds]
JTL has quit []
JTL has joined #dri-devel
anujp has quit [Ping timeout: 480 seconds]
<vsyrjala> mripard: re: xdc talk igt dependencies. what big deps you mean? i think cairo and glib are the highest level ones
<vsyrjala> honestly can't even remember why glib is required. maybe someone didn't want to hand roll a main loop or something
<vsyrjala> so that could perhaps be nuked
JTL is now known as JTL1
<pinchartl> if it's just for an event loop, libevent is nice
JTL1 has quit []
flacks has quit [Quit: Quitter]
JTL has joined #dri-devel
<pinchartl> we use it in a libcamera demo application
<pinchartl> (libcamera has a custom event loop though, maybe we could move that to libevent too)
flacks has joined #dri-devel
lplc has quit [Ping timeout: 480 seconds]
anujp has joined #dri-devel
<vsyrjala> g_hash_table,g_key_file,g_base64,g_regex,g_main_loop/g_io_channel so a bit more usage than i thought
<vsyrjala> but i don't think glib should have a huge dependency list itself
pcercuei has joined #dri-devel
<vsyrjala> or maybe i'm just out of date and it's implemented on top of a browser these days
<ccr> wasm with emscripten, it is the modern way
lplc has joined #dri-devel
gpuman_ has joined #dri-devel
gpuman has quit [Ping timeout: 480 seconds]
<mripard> vsyrjala: pixman is in there as well, python3 to some extent too
<danvet> tzimmermann, [PATCH] fbdev: Garbage collect fbdev scrolling acceleration, part 1 (from TODO list) <- do you plan to merge this?
<danvet> probably good if we either merge right now, so that there's still enough soak time in linux-next before the merge window
<mripard> vsyrjala: (even though pixman is probably a dependency of cairo already)
<danvet> or wait after -rc6
<vsyrjala> well, either you use cairo+pixman or you hand roll all software rendering
<vsyrjala> pretty sure python should be totally optional
<danvet> seanpaul, vsyrjala catching up on mails, but what did go boom with the DRM_MODESET_LOCK_ALL_BEGIN series?
<mripard> danvet: I'm not sure I got the deal with the vc4 patches that got dropped in drm-misc-fixes, should I just reapply them or?
<danvet> mripard, well they conflicted with linus' reverts
<danvet> so direct reapply wont work
<danvet> that's why they got dropped, since this was holding up everything else
<danvet> plus some falling-through-cracks meant 1 month with no drm-misc-fixes pull, which isn't good
<vsyrjala> danvet: wrong ww ctx used was the obvious fail. the other thing is no one reviewed things sufficiently to make sure dropping mode_config.mutex was ok
<vsyrjala> also didn't go through ci, didn't get an ack from i915 maintainers
<vsyrjala> ci caught the fail only after it got merged
<danvet> ah yeah that sounds bad enough
<danvet> I guess seanpaul got rusty :-)
<mripard> danvet: ack, I'll reapply them and try to fix the conflicts then
<jani> danvet: you reviewed v1
<danvet> yeah the Kconfig changes looks all good
<danvet> jani, you checked that CI picked it up and enabled?
<jani> danvet: no :]
<danvet> jani, maybe do that, since I don't trust my own Kconfig simulator in the head :-)
<danvet> CONFIG_DRM_DEBUG_MODESET_LOCK=y
<danvet> jani, ^^ you're good, push it!
<danvet> vsyrjala, I guess rip out the screaming macro and replace it with your magic block thing?
<danvet> convice seanpaul, and ship it
gpuman has joined #dri-devel
<danvet> imo consistency in this definitely trumps my bikeshed
<danvet> consistency throughout drm I mean
gpuman_ has quit [Ping timeout: 480 seconds]
<tzimmermann> danvet, i thought it was already
<tzimmermann> but i can do
<danvet> tzimmermann, tbh I haven't checked
<danvet> looks like it's not there
<jani> danvet: thanks
itoral has quit []
<karolherbst> danvet: I think my idea was more to be able to play around with it without breaking mesa
<danvet> karolherbst, yeah for that a local patch (or maybe even for dim) to s/Part-of:/Link:/ should do the job
<karolherbst> yeah, not too much concerened about this part, but more about adding s-b-o tags from the person assigning to marge to merge it, etc...
<karolherbst> I have no idea where we will arrive with this :)
<danvet> karolherbst, hm right that sob sounds like a good idea
<danvet> otoh it might be a bit much to do that implicitly like that with a button push
<danvet> karolherbst, maybe just an acked-by tag
<danvet> like Acked-by: danvet # approved merging
<karolherbst> danvet: the issue I see is, that after merging somebodcy has to do the manual labour of adding those tags
<danvet> sob it rather too formal maybe
<karolherbst> yeah... but
<danvet> karolherbst, yeah we'd need to hack up marge :-)
<karolherbst> I mean.. today we add it through dim
<karolherbst> at least for mboxes applied
muhomor has quit [Remote host closed the connection]
muhomor has joined #dri-devel
macromorgan has joined #dri-devel
xlei has quit [Ping timeout: 480 seconds]
jcline has quit [Quit: Bye.]
jcline has joined #dri-devel
muhomor has quit [Remote host closed the connection]
<alyssa> pq: I.... guess I wouldn't...? It seems really weird to have the cables plugged in and working but have the "No signal" message on screen because userspace didn't do any mode sets
<pq> alyssa, but that's how it works. Policy belongs in userspace.
<pq> alyssa, what would you even display there, anyway?
<vsyrjala> penguins!
<pq> can't steal an FB from another CRTC
<pq> alyssa, why would it say "no signal" though? Shouldn't the monitor go/stay in powersave/off?
yoslin has quit [Ping timeout: 480 seconds]
<alyssa> uhhhhhh
<pq> btw. I find the "power-off on DisplayPort looks like unplug" a blessing, much easier to test hotplug than ripping cables. :-)
<alyssa> the reproducer is "using X over HDMI-over-DisplayPort, hit the monitor power off button, get some coffee, hit the monitor power on button"
<alyssa> Yeah, truth
<pq> alyssa, so the monitor was showing a good picture before it was unplugged and then hotplugged?
<alyssa> Yeah
<pq> ok, that's a quite different use case
<pq> I thought this was about hotplug with nothing on the connector before.
<alyssa> that use case works fine for Xorg with a DE, or with sway, or with Xorg if you type `xrandr --output HDMI-1 --mode 1920x1080` after powering on
<vsyrjala> and then the monitor's firmware gets tangled up as usual and you need to pull out the power cord
<pq> vsyrjala, that has never happened to me (until I got that HDR monitor for testing; it never survives a reboot of the *computer*).
<pq> alyssa, right, The DE reacts to unplug and disables the CRTC/connector. Then it reacts to hotplug and enables CRTC/connector. When you don't have a DE reacting to things, it gets... annoying.
<alyssa> nod
<pq> alyssa, just like airlied said, from userspace perspective the CRTC/connector stays on after unplug if userspace does not deliberately turn it off.
<pq> physically that probably cannot happen with DP? and HDMI?
<alyssa> wait so do I need to fix this up in kernel space or not?
<pq> (this is a long story...)
<alyssa> (Is that Xorg-without-a-DE displayport hotplug case a bug in my kernel driver or Xorg breaking as intended?)
<pq> When the monitor then comes back, the link training needs to happen again if it is going to display a picture again.
<pq> who is responsible for triggering that is a good question
<vsyrjala> hpd
<pq> I suspect PC gfx card drivers see that the userspace state of the CRTC is still on, and seeing the hotplug (hpd) they try to re-traing the link and set everything up like it was.
<pq> to make it work like it did in the VGA era
* alyssa isn't really interested in making her shiny M1 work like the VGA era..
gpuman_ has joined #dri-devel
<vsyrjala> though if you're lazy you could try the 'set link status prop=bad->fire uevent->let userspace trigger a full modeset"
<pq> however, I could argue that setting link-status property to failed and not doing anything more automatically in the driver is just as good
<pq> it may not fix the Xorg without DE use case, but who cares
<vsyrjala> the ddx should handle it
<pq> it does?
<alyssa> so I should be using the link-status property here? even though the DCP firmware does all the link training transparently and the physical connector is an HDMI port because of course it is?
gpuman has quit [Ping timeout: 480 seconds]
<vsyrjala> presuambly you should get some notification from the firmware that your link has dropped/come back
<pq> alyssa, so what do you need atomic_flush for? :-)
<pq> alyssa, listen to vsyrjala anyway, I'm completely on the other side of the UAPI.
<alyssa> vsyrjala: yes, I have that notification, it's called "hotplug" :)
<alyssa> (Amusingly the power off button doesn't generate any event, but powering on again generates hotplug(OFF) and then a moment later hotplug(ON) because DisplayPort..)
xlei has joined #dri-devel
<daniels> yeah, both hotplug and link-status are a hint to the KMS user that it might want to do something about it
fxkamd has joined #dri-devel
sdutt has joined #dri-devel
adjtm has quit [Remote host closed the connection]
adjtm has joined #dri-devel
tzimmermann has quit [Quit: Leaving]
Ahuj has quit [Ping timeout: 480 seconds]
el0y has joined #dri-devel
<danvet> mlankhorst, will you review the i915 dma_resv_iter patches from könig?
ybogdano has quit [Read error: Connection reset by peer]
yoslin has joined #dri-devel
mattrope has joined #dri-devel
f11f12 has quit [Remote host closed the connection]
f11f12 has joined #dri-devel
sdutt has quit []
sdutt has joined #dri-devel
sdutt has quit []
sdutt has joined #dri-devel
angular_mike__ has joined #dri-devel
<mlankhorst> danvet: thought it was merged already?
<mlankhorst> oh i915
<danvet> yeah i915
<danvet> I stamped all the others
<mlankhorst> Yeah I'll look
<mlankhorst> considering I had something similar
shashanks has joined #dri-devel
tarceri_ has joined #dri-devel
soreau has quit [Read error: Connection reset by peer]
Erandir has joined #dri-devel
soreau has joined #dri-devel
Duke`` has joined #dri-devel
tarceri has quit [Ping timeout: 480 seconds]
zack__ has left #dri-devel [#dri-devel]
zackr has joined #dri-devel
f11f12 has quit [Remote host closed the connection]
f11f12 has joined #dri-devel
gpuman_ has quit [Read error: Connection reset by peer]
gpuman has joined #dri-devel
ddevault has quit [Remote host closed the connection]
ddevault has joined #dri-devel
gpuman_ has joined #dri-devel
gpuman has quit [Ping timeout: 480 seconds]
nchery has joined #dri-devel
mattst88 has quit [Ping timeout: 480 seconds]
FireBurn has joined #dri-devel
gpuman has joined #dri-devel
gpuman_ has quit [Ping timeout: 480 seconds]
ddevault has quit [Ping timeout: 480 seconds]
f11f12 has quit [Remote host closed the connection]
f11f12 has joined #dri-devel
<FireBurn> @agd5f any success in figureing out what's up with the new detection code?
rgallaispou has quit [Ping timeout: 480 seconds]
gawin has quit [Ping timeout: 480 seconds]
* alyssa wonders who has to be poked for the drm/cma-helper commit to be applied
aravind has quit [Ping timeout: 480 seconds]
anujp has quit [Ping timeout: 480 seconds]
Lucretia has quit []
join_subline has quit [Ping timeout: 480 seconds]
ddevault has joined #dri-devel
ddevault has quit [Remote host closed the connection]
Lucretia has joined #dri-devel
anujp has joined #dri-devel
ybogdano has joined #dri-devel
join_subline has joined #dri-devel
mattst88 has joined #dri-devel
gouchi has joined #dri-devel
<agd5f> FireBurn, it's being investigated
frieder has quit [Remote host closed the connection]
<sravn> danvet: there was push-back on the garbage collect fb scrolling due to all the old HW that only supports fbdev. Do we ignore the pushback?
<sravn> danvet: I see tzimmermann (offline) already applied it - sigh
gpuman_ has joined #dri-devel
ddevault has joined #dri-devel
gpuman has quit [Ping timeout: 480 seconds]
ybogdano has quit [Remote host closed the connection]
mbrost has joined #dri-devel
ddevault has quit [Ping timeout: 480 seconds]
gawin has joined #dri-devel
macromorgan has quit [Remote host closed the connection]
macromorgan has joined #dri-devel
moa is now known as bluebugs
<danvet> sravn, there was also complaining on the patch which nuked scrolling, and it's not been reverted
<danvet> so *shrug*
<danvet> pinchartl, [PATCH v2] drm/cma-helper: Set VM_DONTEXPAND for mmap <- I thought you have one of these display blocks too?
gpuman has joined #dri-devel
gpuman_ has quit [Ping timeout: 480 seconds]
<pinchartl> IOMMU yes. not sure what VM_DONTEXPAND does though :-)
<pinchartl> it's been a while since I last look at the IOMMU side, I've forgotten most of it
<pinchartl> (still have to decide if that's a good or bad thing)
tobiasjakobi has joined #dri-devel
macromorgan has quit [Quit: Leaving]
macromorgan has joined #dri-devel
igor has joined #dri-devel
jewins has joined #dri-devel
<alyssa> danvet: the issue is with IOMMU *and* CMA *and* not overriding functions
<alyssa> the normal CMA drivers don't override but also don't have IOMMU
<alyssa> the IOMMU drivers (rockchip, exynos, ...) override so don't hit this I think
<anarsoul> why do you need CMA if you have IOMMU?
<alyssa> anarsoul: what would you use instead?
<alyssa> (see #dri-devel discussion from August about this.)
<anarsoul> OK
<danvet> pinchartl, apparently it splats somewhere
<anarsoul> (I thought that you can construct virtually contiguous region with IOMMU...)
<alyssa> anarsoul: the summary turned out to be "the C in CMA refers to device address space contiguous, not necessarily physically contiguous"
<alyssa> and since the DMA API will use the IOMMU API internally instead of a carveout, everything works out
<danvet> alyssa, rcar-du also splatting?
<alyssa> though the docs/naming is confusing as hell
<alyssa> danvet: rcar-du?
<danvet> alyssa, that's pinchartl's driver
<anarsoul> alyssa: interesting
<danvet> I'm just wondering why no one complained about this before, your driver shouldn't be that special
<danvet> alyssa, also what check goes boom?
<alyssa> danvet: IIRC the check in drm_gem_mmap_obj goes boom `WARN_ON(!(vma->vm_flags & VM_DONTEXPAND))`
<alyssa> Robin found and fixed this independently for HDLCD
<alyssa> I guess the "why has no one complained before" is that the popular drivers aren't affected
<danvet> alyssa, ah I guess that explains maybe why no one is screaming
<danvet> that is somewhat recent
<danvet> so add Fixes: line and good?
<alyssa> sure..
<danvet> iirc that check is from this year
<danvet> maybe pinchartl should look at what his driver does more :-)
<alyssa> minor (i.e. driver still works but dmesg goes boom) regressions only affecting obscure hardware take time
<alyssa> ^for people to start complaining
<danvet> alyssa, and when you have the Fixes: pls push it to drm-misc-fixes
<danvet> dim fixes even formats the line correctly
* alyssa doesn't have push
<alyssa> git blame says the WARN_ON dates to Oct 2019
<alyssa> but Arm hasn't done any work on hdlcd since May 2019 and I'm not convinced anyone outside Arm actually uses hdlcd so there's that ;)
<alyssa> rcar-du looks like it might hit this boom
<pinchartl> CMA is a very bad name here, we don't explicitly use CMA as such
<pinchartl> CMA is the framework that made it possible to allocate physically contiguous memory
<pinchartl> but it's all hidden by the DMA mapping API
<pinchartl> we'll get DMA contiguous memory if there's an IOMMU, and physically contiguous memory if there isn't
<pinchartl> and even in the latter case, there's no guarantee that CMA will actually be used in the backend
ddevault has joined #dri-devel
<pinchartl> we shouldn't have named the helpers gem cma
mattst88_ has joined #dri-devel
<alyssa> 💯
FireBurnUK has joined #dri-devel
<pinchartl> I haven't tested rcar-du + IOMMU in quite some time as, as far as I know, IOMMU support isn't enabled upstream on that platform
<alyssa> danvet: ^ there you go then
FireBurn has quit [Read error: Connection reset by peer]
<pinchartl> in theory it's an important use case :-)
mattst88 has quit [Ping timeout: 480 seconds]
gpuman_ has joined #dri-devel
gpuman has quit [Ping timeout: 480 seconds]
<alyssa> in theory, theory and practice are the same ;)
mattst88_ has quit []
mattst88 has joined #dri-devel
<danvet> alyssa, hm I had no idea it was that old
<danvet> but yeah if you add that as fixes: feel free to whack my ack on top and push
tobiasjakobi has quit [Remote host closed the connection]
<alyssa> thanks :)
* alyssa still doesn't have push but she can bother bbrezillon about that or something :p
<agd5f> alyssa, you can get push if you want it
<alyssa> sweaty guy with red button.jpg
<jekstrand> tarceri_: What did you think of my suggested comment on !13166
gpuman has joined #dri-devel
xexaxo_ has quit [Ping timeout: 480 seconds]
gpuman_ has quit [Ping timeout: 480 seconds]
gawin has quit [Remote host closed the connection]
<danvet> alyssa, yeah go get that drm-misc commit thing
<alyssa> 17:49 < alyssa> in theory, theory and practice are the same ;)
<alyssa> er
lemonzest has quit [Quit: WeeChat 3.2]
yoslin has quit [Remote host closed the connection]
yoslin has joined #dri-devel
ngcortes has joined #dri-devel
igor has quit []
lynxeye has quit []
Plagman has joined #dri-devel
boistordu has joined #dri-devel
xexaxo_ has joined #dri-devel
ngcortes has quit [Quit: Leaving]
ella-0_ has joined #dri-devel
ngcortes has joined #dri-devel
i-garrison has quit []
ella-0 has quit [Ping timeout: 480 seconds]
i-garrison has joined #dri-devel
f11f13 has joined #dri-devel
f11f12 has quit [Ping timeout: 480 seconds]
<airlied> we branched yet, I want to start landing disruptive stuff :-P
Ahuj has joined #dri-devel
<airlied> ah I got rebased over a branchpoint :)
iive has joined #dri-devel
jljusten has quit [Quit: WeeChat 3.2.1]
jljusten has joined #dri-devel
alanc has quit [Remote host closed the connection]
alanc has joined #dri-devel
dongwonk has joined #dri-devel
<airlied> yay finally got cl cts to pass a test with 128/64 images
sagar_ has quit [Read error: Connection reset by peer]
<zmike> oh no
sagar_ has joined #dri-devel
mbrost has quit [Ping timeout: 480 seconds]
<karolherbst> airlied: nice
<eric_engestrom> airlied: yup, branchpoint done, -rc1 release incoming but you can land your disruptive stuff now :P
<eric_engestrom> also yeah, sorry about !13328, I thought I could trick marge, but I feel like she's not doing anything anymore, I may have done something I shouldn't have :(
gouchi has quit [Remote host closed the connection]
<eric_engestrom> or maybe she's just busy on some non-mesa repo?
<eric_engestrom> daniels: if you're around, could you have a look just to be sure she's not stuck?
<eric_engestrom> (ah, sorry about the ping, I just saw on #freedesktop you said you're on holiday this week, I didn't mean to bother you :)
igrtrrnt has joined #dri-devel
mbrost has joined #dri-devel
<kisak> eric_engestrom: you can see the bot's activity log at https://gitlab.freedesktop.org/marge-bot . I've got a hunch that it'll unstuck itself after about an hour
<eric_engestrom> yeah, but I'm not seeing any activity anymore, that's my point :(
gouchi has joined #dri-devel
<airlied> yeah it does seem to be in the woods
danvet has quit [Ping timeout: 480 seconds]
gouchi has quit [Remote host closed the connection]
pnowack has quit [Quit: pnowack]
Bennett has joined #dri-devel
Ahuj has quit [Ping timeout: 480 seconds]
<anholt> marge should be back now
<anholt> but x86 runners seem to be in short supply
<eric_engestrom> I think it's because I just created two branches and two tags and it depleted the supply, but they should be back shortly
<anholt> this whole week's been rough in my experience.
Duke`` has quit [Ping timeout: 480 seconds]
elongbug has quit [Ping timeout: 480 seconds]
<pinchartl> eric_engestrom: that was my Monday :-)
pushqrdx has quit [Ping timeout: 480 seconds]
pushqrdx has joined #dri-devel
ngcortes_ has joined #dri-devel
ngcortes has quit [Ping timeout: 480 seconds]
<anholt> oof. spending 13% of gtest runtime on FilterTests when specifying each test by name in gtest_filter. check out PatternMatchesString for some entertainment.
<cmarcelo> anholt: :-/ wonder if at least trying to shortcut patterns that don't have * and ? would be worth. (might need to split on : first for convenience)
ngcortes_ has quit []
ngcortes has joined #dri-devel
mbrost has quit [Ping timeout: 480 seconds]
<anholt> yeah, I expect that that would help a ton. but also the chain of "fix gtest -> get it into consumers" sounds not worth my time (yay C).
<jekstrand> mareko: Does radeonsi use lower_atomics_to_ssbo or do you hook up the "real" HW atomics?
<airlied> jekstrand: r600 has real hw atomics, don't think radeonsi does
<jekstrand> airlied: Ok. r600 also doesn't use NIR yet. So maybe we can punt this problem a little bit?
<jekstrand> hrm....
<jekstrand> cmarcelo: ^^
<jekstrand> cmarcelo: We may need nir_var_atomic
<airlied> jekstrand: it doesn't use nir by default, it does have a working nir backend in the tree
* jekstrand doesn't really like it but oh, well.
<jekstrand> airlied: Right. Probably don't want to break it. :)
<anholt> yeah, I think Gert was saying the r600 nir stuff was pretty close at this point?
<airlied> I'm sure dschuermann would love to have a VLIW vs of aco :-
<airlied> :-P
mbrost has joined #dri-devel
<cmarcelo> jekstrand: maybe we'll need that...
<jekstrand> cmarcelo: I doubt it'd be nearly as much of a pain as nir_var_mem_image
<jekstrand> cmarcelo: Another option would be to make it a lowering pass and run it after atomics_as_ssbo
<Lyude> agd5f, hwentlan - at least one of you two said that you didn't have any issue with breaking things in radeon.ko's (most likely already-broken) MST support correct?
<hwentlan> Wouldn't have been me as I don't really deal with radeon.ko. You'd want agd5f's opinion on that
<Lyude> gotcha
<Lyude> yeah, now that I have time to look at things again I'm thinking about killing off as much of the non-atomic stuff in mst as I can
<Lyude> also - marking things as deprecated (as in, using gcc's actual attributes for marking deprecated functions) isn't much of a thing in the kernel source is it?
<airlied> nope not a thing at all
vivijim has quit [Ping timeout: 480 seconds]
<Lyude> darn
<Lyude> would like to do something to make sure people don't use older MST VCPI functions until I have a chance to remove them
<airlied> Lyude: does anyone use them now other than radeon?
<Lyude> nope
<airlied> it's not like we are merging non-atomic drivers
<Lyude> i just never really had the time to actually evaluate doing such a conversion before
<airlied> atomic radeon would be fun :-P
mlankhorst has quit [Ping timeout: 480 seconds]
<Lyude> (and if anyone does use those helpers, that's a bug)
<Lyude> airlied: someday
<Lyude> i still haven't even finished my hpd irq cleanup though
<airlied> Lyude: there are a fair few places on the net saying to pass radeon.mst=1 to make somee things work
<airlied> so it might not be great if it break
<Lyude> airlied: tbh, I think I'm willing to take that risk as long as I can confirm that mst is as broken on radeon as we all think it is
<airlied> Lyude: it used to work up the point of turning on dual-panel monitors, but often the second panel had different color levels
<Lyude> not having to manage the mess of non-atomic and atomic state in the MST helpers would really simplify a bunch of code
<airlied> you could fork into radeon the non-atomic and leave it die
<Lyude> oooh, good point
<Lyude> well, for the time being I think I'm happy with that plan - and will at least stop worrying about updating the non-atomic bits for the time being
* airlied isn't sure I've got any MST hw to plug into a radeon at home
<Lyude> I've got the MST hardware, but only a few radeons at home
<Lyude> shouldn't be hard to find more though.
* airlied only has docking stations, which I can't plug into radeon
<vsyrjala> find the right pins and solder some wires
Kayden has quit [Quit: time to get off IRC for now]
<vsyrjala> well, maybe there are easier ways to get mst hw
pushqrdx has quit [Remote host closed the connection]
pcercuei has quit [Quit: dodo]
Bennett has quit [Remote host closed the connection]
jani has quit [Remote host closed the connection]
jani has joined #dri-devel
__giggi__ has quit [Remote host closed the connection]
rasterman has quit [Quit: Gettin' stinky!]
igrtrrnt has quit [Ping timeout: 480 seconds]
iive has quit []
sadlerap has joined #dri-devel
<agd5f> Lyude, I don't recall ever getting radeon mst working, but it's been a while