ChanServ changed the topic of #linux-rockchip to: Rockchip development discussion
<Jmabsd>
Tenkawa: which SBC???
lurchi_ is now known as lurchi__
vagrantc has quit [Quit: leaving]
lurchi_ has joined #linux-rockchip
lurchi__ has quit [Ping timeout: 250 seconds]
rektide has quit [Quit: Lost terminal]
rektide has joined #linux-rockchip
stikonas has joined #linux-rockchip
hramrach has quit [Ping timeout: 272 seconds]
hramrach has joined #linux-rockchip
Whistler has joined #linux-rockchip
stikonas has quit [Ping timeout: 272 seconds]
camus has joined #linux-rockchip
<maz>
robmur01: last time I argued with the RK folks, they were using the GICv3 MBI frame (GICv2m-style) to provide MSIs without a working ITS. Back to the future...
matthias_bgg has joined #linux-rockchip
lopsided98_ has quit [Ping timeout: 246 seconds]
<Jmabsd>
Does anyone have any idea what RK3588 SBC Tenkawa says he will receive soon, a Radxa?
<mps>
Jmabsd: from the head, odroid iirc
<Jmabsd>
Odroid makes an RK3588? =o
<Jmabsd>
that would bring back attention to 'which are the RK3588 SBC makers' :-)
<wens>
check cnxsoft for related articles?
<robmur01>
maz: yay! Who doesn't love spoofing interrupts via /dev/mem?
<maz>
robmur01: I hear userspace interrupts are the future, hereby proving that UNIX signals were the right concept the first place... ;-)
<robmur01>
shush, you PDP-hugger :P
<maz>
unfortunately not! I only go as far back as the (micro)VAX!
<maz>
but if you have a PDP11 stash to share, I'm game!
dliviu has quit [Ping timeout: 248 seconds]
dliviu has joined #linux-rockchip
<robmur01>
it does bug me a bit that we clearly have a spec hole in the area of GIC coherency - AFAICS nothing actually tells integrators what they can, should, or can't do :/
<robmur01>
slightly worried that if I get an RK3588 it might make me feel more compelled to do something about that...
<maz>
robmur01: I've been kicking that can for years, and it went nowhere. not it is your time to try and tackle it! :)
<robmur01>
ah, but now you're in a position to convince our customers that pKVM needs GIC-600 so more of them hit the problem and kick up a stink :D
kevery has quit [Ping timeout: 250 seconds]
<maz>
robmur01: I'm sure this is all going to work really well...
lopsided98 has joined #linux-rockchip
lopsided98 has quit [Quit: No Ping reply in 180 seconds.]
lopsided98 has joined #linux-rockchip
lurchi_ has quit [Ping timeout: 248 seconds]
Tenkawa has joined #linux-rockchip
kevery has joined #linux-rockchip
<Jmabsd>
Tenkawa: Which RK3588 board did you order =o
<Tenkawa>
the Rock5
<Tenkawa>
going to be waiting a while though
<Jmabsd>
ic
<Jmabsd>
Tenkawa: What is the price of the Radxa Rock5 actually, can you tell me the price page
<Tenkawa>
sure
<Tenkawa>
just a sec
<Tenkawa>
I think I started at the 4gb page.. they have 4/8/16gb variants there
<Jmabsd>
Tenkawa: so that is the primary, prefered way?
<Tenkawa>
it is here in the states
<Jmabsd>
ok
kevery has quit [Ping timeout: 256 seconds]
tlwoerner has quit [Read error: Connection reset by peer]
vagrantc has joined #linux-rockchip
robmur01 has quit [Read error: Connection reset by peer]
robmur01 has joined #linux-rockchip
<macc24>
can anyone test if dsi displays work on rk3399 on 5.17?
<macc24>
mmind00?
<vagrantc>
i'm running 5.17 on a pinebook pro ... not sure if it uses dsi
<macc24>
it does not
<mmind00>
macc24: I don't think I have a rk3399 with dsi around ... though px30 with dsi I do have ... but realistically I'm not sure if I get to that before easter
<macc24>
:/
<Tenkawa>
I have a 3399 but nothing dsi to hook up to it unfortunately
<Tenkawa>
and it does have 5.17.2 on it atm
matthias_bgg has quit [Quit: Leaving]
Tenkawa has quit [Quit: Was I really ever here?]
anarsoul|2 has joined #linux-rockchip
wens_ has joined #linux-rockchip
mriesch has quit [*.net *.split]
wens has quit [*.net *.split]
anarsoul has quit [*.net *.split]
mriesch has joined #linux-rockchip
tlwoerner has joined #linux-rockchip
Tenkawa has joined #linux-rockchip
stikonas has joined #linux-rockchip
SallyAhaj has quit [Remote host closed the connection]
SallyAhaj has joined #linux-rockchip
<macc24>
ok so
<macc24>
turns out
<macc24>
you can't have hdmi and mipi_dsi enabled at the same time in dts
<Tenkawa>
oh well... I more want the main core sbc than the ancillary io
<Tenkawa>
I already was running my CM4 on PCI-E\
<Tenkawa>
already need a few more drives
<Jmabsd>
Tenkawa: So Rock5 model B, what is the difference with model A actually
<Jmabsd>
Tenkawa: so Rock5 model B has an RK3588S and not RK3588 ?
<Jmabsd>
sigh, indeed various pages presenting the Rock5 have been unclear - the specs page here https://wiki.radxa.com/Rock5/CM5 say clearly it's two PCIe v2 and its' RK3588S.
<Jmabsd>
Tenkawa,Psydroid,*: We can guess the answer about IO on the Rock5 model B then: It's one PCIe v2 lane to the lower M.2 and one PCIe v2 lane to the upper M.2. This means the SSD will have 625MB/sec theoretical cap, and also any 10gbps ethernet will have a 5gbps cap. Wonder why they chose the RK3588S? :-o
<Jmabsd>
Wait wait, facepalmers, nono it's the Rock5 "Compute Module" https://wiki.radxa.com/Rock5/CM5 that uses the RK3588S .. uuh. Soooo uh
<Tenkawa>
that would be a neat cm* competitor someday
<Jmabsd>
Tenkawa: Okay so apologies for my confusion above about Rock5 Compute Module vs Rock5 SBC: Indeed the Rock5 SBC uses the RK3588 or at least so it seems. Waiting for their confirmation. Emailed them.
<Jmabsd>
Tenkawa: Is there any "Rock5 model A"??
<Tenkawa>
I haven't seen one... that would not seem to be a selling point for that soc
Tenkawa has quit [Quit: Was I really ever here?]
<Jmabsd>
Tenkawa: Selling point, which?
<Jmabsd>
Why do they start with a "model B" if no "model A" yet exists????
<Jmabsd>
Tenkawa: It's on Radxa's table to clarify now. They have email/moderator message to clarify the PCIe etc.
<Jmabsd>
Tenkawa,*: I would think the Rock5 SBC is actually RK3588 and *not* RK3588S. Then indeed it has 5x PCIe lanes total, that's awesome. And then I would think the lower side has the 4x PCIe v3 lanes, and the upper side has the 1x PCIe v2 lane.
stikonas has quit [Remote host closed the connection]
<Jmabsd>
Tenkawa: I think the most likely situation would be then that the lower M.2 is indeed 4x PCIe v3 lanes. That means the top M.2 is too slow to do 10gbps ethernet from. If so, the lower M.2 should be bifurcated to do SSD on 2 lanes and ethernet on 2 lanes, that means 1250MB/sec PCIe bandwidth each for SSD and ethernet respectively - really good. ;-) Of course a PCIe switch chip could help too
<jakllsch>
the Pine64 people aren't so optimistic about Rockchip's PCIe implementation..
<Jmabsd>
jakllsch: actually how much unoptimistic are they, what kind of issues
<jakllsch>
there are coherency and other problems with the SoC integration
<jakllsch>
something about coherency both with DMA, and somehow the GIC is involved
<Jmabsd>
I'm a bit surprised that the available info for the Radxa Rock5 SBC was so unclear. However, I think my own guess presented just now here, is the correct answer.
<Jmabsd>
jakllsch: Oh. dGPU as in AMDGPU / Radeon card? =o
<jakllsch>
yeah
<Jmabsd>
Hmmmm. Well maybe that is OK maybe, for server use as you won't have a GPU then
<Jmabsd>
But okay that would be a bummer hah. I wonder if it can be fixed in software.
<jakllsch>
i'm of the opinion it's a driver problem, but upstream isn't
<Jmabsd>
jakllsch: Very interesting. I see. Wonder how OpenBSD would look at the matter.
<Jmabsd>
Thanks for mentioning the PCIe issue. Oh btw, for a M.2 PCIe v3 4 lane connector, do you think it can be bifurcated easy??
<Jmabsd>
Or need a PCIe switch chip?
<jakllsch>
like NetBSD, they probably don't have the personpower to fix the dri/drm drivers they get from Linux
<Jmabsd>
E.g. connect both one M.2 NVMe SSD to it and one M.2 10gbps ethernet adaptor :)
<Jmabsd>
what do you mean fix
<jakllsch>
you can work around DMA cache-incoherency in the DRI/DRM drivers if you know where to put the flushes/invalidations
<jakllsch>
the upstream code just assumes coherency
<Jmabsd>
Tenkawa,Psydroid,*: Btw, Radxa charge 25 USD per 4GB RAM. This would leave the price point for the Rock5 SBC with 32GB RAM at 320USD. That is not bad!
<Jmabsd>
I see. That is beyond my knowledge.
<Jmabsd>
jakllsch: Do you know in-depth how PCI and DMA work? I don't yet heh
<jakllsch>
I see no reason the GIC issue can't be worked around in a *BSD though
<jakllsch>
Jmabsd: yeah, i do
<Jmabsd>
Cool. Wait, of what consequence were GIC issues?
<Jmabsd>
jakllsch: Cool! Another day I may query you about it. Great you know.
<Jmabsd>
Are you into low level software or also hardware?
<jakllsch>
the GIC ITS tables are limited to a 32-bit physical address
<Jmabsd>
On the RK3588? Ok. And that has what consequence?
<jakllsch>
despite the chip supporting more ram
<jakllsch>
linux is refusing to add a work-around without a formal eratta from Rockchip, and Rockchip doesn't want to
<jakllsch>
if the GIC driver knows not to place the ITS table above 4GiB it'll work
<Jmabsd>
Sigh. Ok I suggest you are right here Linux should fix it
<Jmabsd>
Oh, the Firefly RK3588 SBC apparently has idle power of 1 Watt, that is way below Raspberry Pi 4 which has idle around 4 watt was it, and Mac M1 which is around 7W - "Idle: ≈1.35W (12V/110mA" https://download.t-firefly.com/产品规格文档/行业主板/ITX-3588J - Specification.pdf?v=1649978761
<Jmabsd>
Btw, looking at this same PDF page 9, it looks like they admit it's 5 PCIe slots - they have one ordinary PCIe 4 lane slot, and then one miniPCIe slot for modem.