<jow>
so statelessness means you're able to export a relevant part of the rootfs without it being tainted by volatile things (like it would be the case with /)
<rsalvaterra>
Yep, pretty much.
<neggles>
and it's one directory, with everything that matters in it
<jow>
I was confused by its meaning in this context
<neggles>
it also eventually removes the need for the `#!/usr/bin/env python3` trick
<neggles>
well, not for python, because pythong
<neggles>
er, s/g//
<jow>
python will find new ways to break existing deployments
<neggles>
yeah, bad example. `#!/usr/bin/env bash` though
<jow>
/usr/bin/env in shebangs somehow became second nature for me
<neggles>
cause *usually* bash is /bin/bash, but if you're on a Mac or [a couple distros I forget] or you've built something custom...
robimarko has joined #openwrt-devel
<jow>
right, Mac OS, other weird deployments
<neggles>
I've never understood what the difference is meant to be between /bin and /usr/bin really
<jow>
iirc it once meant to be essentials vs. supplemental
<neggles>
/usr/local/bin makes sense, effectively "things that didn't come from a package manager"
<jow>
/bin is stuff for booting the core system, /usr is "user software" you can mount in via nfs etc.
<neggles>
it's more "PDP-11 disks were small"
<neggles>
that email is the first time i'd actually seen it spelled out as "/bin is for things you might need before your rootfs gets mounted"
danitool has quit [Quit: Cubum autem in duos cubos, aut quadratoquadratum in duos quadratoquadratos]
<neggles>
but that stuff's all in initrd nowadays...
<rsalvaterra>
neggles: Exactly. We have this separation today due to problems that have arisen 50 years ago. :)
<neggles>
yup
<neggles>
IMO given that 1) this is the way every major distro is going & 2) most of the embedded things I've cracked open in the last year have had merged /usr, it would probably make sense for openwrt to do it as well
danitool has joined #openwrt-devel
<rsalvaterra>
Probably a topic for the next meeting…? :P
<neggles>
robimarko: oh hey! how do we feel about the IPQ8174
<robimarko>
In which sense?
<neggles>
well, some QCA SoCs are more of a nightmare to handle than others
<robimarko>
I wish it was only QCA SoC-s
<rsalvaterra>
Ugh… yet *another* A53 designh? What year is this?
<rsalvaterra>
*design
<neggles>
why upgrade the CPU when you can just put in more NPU you dont have to pay licensing fees for? :P
<neggles>
it's kind of surprising they've not put much effort into the "linux on HEXAGON" project
<neggles>
then they could build router SoCs with no ARM at all :P
<robimarko>
Its an upgrade compared to IPQ40xx or IPQ806x anyway
<neggles>
made it all the way into mainline...
<robimarko>
neggles: Why bother?
<robimarko>
They got the licenses anyway
<rsalvaterra>
neggles: (Ugh…)²
<robimarko>
Whats wrong with A53?
<robimarko>
Mediatek is using it as well in the brand new SoC
<neggles>
A55 is ~30%? faster for less power and die space, but it also requires a newer process node
<rsalvaterra>
robimarko: State-of-2012 art. In-order, not very power-efficient compared to current alternatives…
pepe2k has joined #openwrt-devel
<robimarko>
Well, this is market where the price is only what matters
<robimarko>
They aint gonna use high end nodes if they dont have to
<neggles>
i suspect A53 is as good as it gets with what 45nm?
<neggles>
and yeah, nobody cares if their router pulls 15w or 12w and the lion's share of the power draw will be the radios anyway
<rsalvaterra>
robimarko: I'm not question marketing decisions, I'm sure they make perfect sense, financially… but I'm a technical guy. :P
<rsalvaterra>
I like new and shiny toys. :P
<neggles>
I was hoping for A55 too :(
<robimarko>
rsalvaterra: Me as well
<robimarko>
negles: The thing is that IPQ807x is kind of couple of years old
<neggles>
robimarko: perhaps I expect too much of you :P trying to decide between grabbing an 8072A device or possibly an 8174A device to mess with
<robimarko>
Dont go for the IPQ8174A series
<robimarko>
Those OAK parts have been barelly supported by QCA
<neggles>
or maybe i should just go right to IPQ6018 since that's where the new unifi things are & will probably become the IPQ4018/19 of the wifi 6 generation
<robimarko>
8072A is kind of the best buy
<robimarko>
The thing is that IPQ6018 is not cheap enough
<neggles>
*yet*
<robimarko>
More like it wont be ever
<robimarko>
Cause you can get IPQ8072A for 99USD all the time
<rsalvaterra>
Stuff like this just makes me want to say "fsck it" and go with the brute force of x86 hardware…
<neggles>
when they were first released about 9 months? a year? ago, that SoM was US$149 and the clearfog base was about $220
<neggles>
now $218 and $301 :(
<robimarko>
Its basically just Armada 7040 with upgraded core
<robimarko>
DDR4 support etc
<neggles>
it's not though, it kind of is, but it's got *way* more bandwidth, lemme see if i can find the chart i had
<robimarko>
Its a direct replacement for 7040
<robimarko>
And they made sure to not make it too good
<robimarko>
Hence the modular CP115 design so they just attach more of those for 9131 etc
<rsalvaterra>
neggles: 64-bit DDR4, 2400 MT/s. Not bad…
<robimarko>
Yeah, they updated the outdated stuff
<neggles>
bluh I can't find the doc
<neggles>
but it was from pensando, comparing their chip to the 9130, 8040, and 7040
<neggles>
9130 beats 8040 by a solid 30%
<robimarko>
Thats probably from the various Octeon related offloading engine stuff they crammed in along the MVPP2
<robimarko>
And it uses like dozen FW blobs
<neggles>
*shrug* they're in mainline and fairly well characterized, i don't mind firmware blobs as long as they're not going to break your ability to upgrade and they're actually being used by more than five people
<neggles>
but the 9130 also has quite a bit more in the way of IO than the 8040, even without the expander chips
<neggles>
oh and it's DPDK-supported
<neggles>
though it is interesting that the 8040 has disappeared off marvell's products page, but the 7040 and 9130 are still there
<neggles>
bluh maybe the 9130 doesn't have significantly better IO, can't really tell from the public info. meh.
<robimarko>
It doesnt have better I/O than 8040
<robimarko>
9130 only has 5 SERDES lanes
<neggles>
six
<robimarko>
8040 has 10 though
<robimarko>
So you have to upgrade to 9131 at least
<neggles>
datasheets i have are showing 6x HS SERDES + 2x ICI + 4x SERDES for 9130, 12x HS SERDES for 8040
<robimarko>
7040 is there because there is a to of large customers using it
<neggles>
which i think for the 9130 translates to 6x SERDES + 4x PCIe lanes
<neggles>
at least the latest rev of the DS has this explicitly laid out
<neggles>
so a 9130 with one 88F8215 has ~the same IO as an 8040
<robimarko>
Yep, and they are selling that as 9131
<neggles>
so this is the chip they want in everyone's midrange firewall appliances then
<neggles>
one chip covering three HW SKUs and six models
<neggles>
from that perspective it's a rather clever design really
ldir has joined #openwrt-devel
<neggles>
heh. the more you look at these block diagrams and datasheets, and considering the 8040 has mysteriously vanished from the product pages, the more it looks like the 9130 is an 8040 with half the SERDES lanes restricted to only do MoChi
<neggles>
(and incidentally already supported in openwrt)
<robimarko>
I wanted to get those to "experiment" but the boss wasnt really keen since we already have M801-s
<robimarko>
But these are way more practical
<neggles>
the 901 is a pretty sweet little box really. shame they cost a fortune
<robimarko>
Managed to get two CN9130 ref boards though
<neggles>
the price isn't unreasonable, it's just more than i can justify :P
<neggles>
oh nice, solidrun actually put their "two honeycomb LX2 in a 1U chassis" concept into production
<rsalvaterra>
Why do they call them "software-defined routers"? Aren't *all* OpenWrt routers software-defined? :P
<neggles>
because you can program the packet acceleration engines with DPDK and various other things, so you can put your own DPI/IDS/IPS/etc algorithms into the accel hardware instead of it being fixed-function :P
<rsalvaterra>
That sounds like the antithesis of "software-defined", having a hardware accelerator (albeit programmable).
<rsalvaterra>
When I think of "software-defined", I think of everything being done by a general purpose CPU.
<neggles>
well in comparison to, say, qualcomm's NPUs or mediatek hw-nat accelerators
<neggles>
where it does one of a half dozen things and not a lot else
<neggles>
perhaps a better parallel would be early days of GPUs when we went from having several fixed function blocks (transform, lighting, texturing etc) to the "gigantic puddle of shaders you shove tiny programs into"
<rsalvaterra>
Heh. I'll probably just throw x86 hardware at the problem. Much cheaper. :)
<neggles>
[various firewall appliance vendors] beg to disagree :P
<neggles>
sophos's latest generation of hardware consists of either a 913x, or a ryzen embedded and various SKUs of octeon TX2
<neggles>
same goes for a couple other vendors, and in a month or so i'll be cracking open a palo alto PA-440 which I would put money on being a CN913x
<neggles>
this XGS2100 claims to be a CN83xx 16-core
<neggles>
huh yep, Oct TX not TX2, guess that explains the 4.14 kernel
<neggles>
robimarko: this might be the 4.14 kernel from the github repo
kopijaheaseli has quit [Read error: Connection reset by peer]
<robimarko>
That would be the first one in the wild then
<neggles>
i've got four of these deployed, and the rest of the XGS lineup are all either <ARM octeon of some kind> or <an intel chip> + <big angry ARM octeon of some kind>
<neggles>
on the octeon, Linux version 4.14.207-10.3.6.0-1 (jenkins@bsp-build-00) (gcc version 7.3.0 (Marvell Inc. Version: Marvell GCC7 build 249.0)) #1 SMP PREEMPT Tue Oct 19 13:14:45 EDT 2021
<neggles>
on the Ryzen R1505G side, Linux localhost 4.14.38 #2 SMP Thu Nov 11 19:44:04 CET 2021 x86_64 GNU/Linux
<neggles>
4.14.38? oof, FIPS certification I guess...
<neggles>
ok back to trying to work out how this devicetree needs to look
bluew has quit [Quit: Leaving]
Namidairo has joined #openwrt-devel
srslypascal is now known as Guest1914
srslypascal has joined #openwrt-devel
srslypascal is now known as Guest1915
srslypascal has joined #openwrt-devel
<neggles>
I would just like you all to know that sophos's firewall OS appears to be built using the openwrt buildsystem
<jow>
cool
<jow>
not that it benefits us in any way
rua has quit [Quit: Leaving.]
Guest1914 has quit [Ping timeout: 480 seconds]
<neggles>
true. they're *very* diligent with their GPL releases, though - these have documentation! and a sane layout! and build working* code!
<neggles>
and have things such as 'sdk-ext-sophos-release-SDK-10.3.6.0-PR1'
<neggles>
hello, octeon SDK 10
Guest1915 has quit [Ping timeout: 480 seconds]
<neggles>
yeah, no mips stuff in here at all, not shocked
<robimarko>
SDK10 is universal Armada and TX2 SDK
<neggles>
yeah
<neggles>
it's a lot cleaner than the old mips one
<neggles>
well that answers that question, the sophos XGS 107/116 are CN9130, the 126/136 are 9131
<russell-->
i did a review of deployed ER-X's (I had 7 in the field, now just 6). All of them had some degree of squashfs file corruption, recovers on sysupgrade.
<russell-->
tested with: find /rom -type f | xargs sha256sum
<Habbie>
can a package makefile download patches in addition to the source tarball?
<Habbie>
I don't immediately find any examples of this in the openwrt-21.02 packages tree
rua has quit [Ping timeout: 480 seconds]
rua has joined #openwrt-devel
arthur_melo has joined #openwrt-devel
Grommish has joined #openwrt-devel
rua has quit [Ping timeout: 480 seconds]
rua has joined #openwrt-devel
<mrkiko>
Habbie: I guess it can, but in general the package tends to ship with itself the needed patches; downloading them from an outside source is considered sometimes considered a "problematic" practice, for reproducibility for example.
<fpsusername[m]>
Habbie: did you ever continue tinkering with the kpn extender?
ekathva has quit [Remote host closed the connection]
<ynezz>
f00b4r0: indeed, we don't want to backport all that nvmem stuff into 21.02
<f00b4r0>
figured as much :)
<Slimey>
yello
dedeckeh has quit [Remote host closed the connection]
rsalvaterra has quit []
rsalvaterra has joined #openwrt-devel
rua has quit [Ping timeout: 480 seconds]
<Habbie>
mrkiko, ok - it would include a hash, of course, just like for the tarball, but if generally we ship inside the packages tree, that's what i'll do
<Habbie>
fpsusername[m], hey! I've tinkered with many devices since, but not yet with that one
<Habbie>
fpsusername[m], however, all that has given me enough confidence to solder serial to it some day
rua has joined #openwrt-devel
<fpsusername[m]>
Lmao, it's just a 4 pin header, of which you actually need two, for TX and RX. Ground connecting might not be necessary
<Habbie>
always connect ground!
<Habbie>
otherwise you're going to have a bad time
<Habbie>
and, it's just, etc., but until a few months ago, the most expensive thing i had soldered on was 5 bucks, and that was two years ago
<Habbie>
'square' TX, followed by RX/GND/VCC in that order, is that a super common layout?
rua has quit [Ping timeout: 480 seconds]
<svanheule>
Habbie: for TP-Link devices, yes
<Habbie>
svanheule, right, i have a non-tp-link here that happened to work too
<Habbie>
hence me wondering if this was common in non-tplink too
<Habbie>
of course, this could be a rebranded tplink
<Habbie>
(sitecom wl-330)
<svanheule>
outside of TP-Link, RX/TX are usually pins 2-3 in my experience
<Habbie>
2-3 where the square is 0? or 1?
<svanheule>
square is 1 in that case; board design doesn't usually start counting from zero
<dwfreed>
yeah, usually VCC or GND is pin 1, then RX/TX in the middle, then the other of VCC and GND on the other side
<dwfreed>
everybody's different, though
<Habbie>
right - with the voltmeter, i got 3V3 on 1 and 4 and 0V on 2 and 3
rua has joined #openwrt-devel
<Habbie>
ok, TX is on 4
<Habbie>
not 1 :)
rua has quit [Ping timeout: 480 seconds]
rua has joined #openwrt-devel
cbeznea has quit [Quit: Leaving.]
<Habbie>
fpsusername[m], my other thrift store find apparently is 4MB flash, 16MB RAM :D
<fpsusername[m]>
Lol, how ancient is that stuff? 2.4GHz B/G only?
<Habbie>
oh yes
<Habbie>
sitecom wl-330
<rsalvaterra>
Habbie: That's cute! Are you adding support for it? :)
<Habbie>
rsalvaterra, yeah, i'm porting backfire 10.03.2 right now :D
<Habbie>
Linux version 2.6.21 (root@localhost.localdomain) (gcc version 3.4.2) #61 Thu Jan 15 11:10:13 CST 2009
<rsalvaterra>
Aw… I think I could shove a 5.15 kernel in there. *Very* cut down, of course.
<Habbie>
(the paste is real, the port comment is obviously a joke :D )
<Habbie>
rsalvaterra, haha
<Habbie>
the real joke is, i paid 5 bucks for these two devices (which i do not regret - extra b/g APs will come in handy here) and then paid 13 bucks for one power supply to power either of them :D
<Habbie>
but a 12V supply with switchable plugs will surely be useful at other times too
<rsalvaterra>
Hah! Didn't you have an old compatible PSU lying around?
* rsalvaterra
has too many to count
<Habbie>
old PC PSU? i probably do
<Habbie>
but not the right plugs ;)
<Habbie>
the plugs might be the best part of those 13 bucks in the end
<rsalvaterra>
Not a PC one, a standard 12 V PSU with DC connector.
<Habbie>
right, i had several of those, but all the jacks were too big
<Habbie>
s/had/have/
<rsalvaterra>
Unfortunate. :/
<Habbie>
yeah
<Slimey>
heh
robimarko has quit [Quit: Leaving]
dansan has quit [Read error: Connection reset by peer]
dansan has joined #openwrt-devel
<Grommish>
Habbie: If you have access to something like a weather-pack set, you could make interchangable ends
<Habbie>
Grommish, what's a weather-pack set?
<Habbie>
Grommish, oh i see it
<Habbie>
Grommish, i have duponts, alligator clips, etc.
<Grommish>
Habbie: used in wiring harnesses for automobiles, they are designed for crimping pins into houses
<Habbie>
right
<Habbie>
for connecting to real plugs
<Grommish>
Well, it's a 2 piece, so you cut the end off the PSU, wire in the connector, then take the other side to the various output pins you need
<Grommish>
Like a quick-connect for Pnumo or Liquid lines
<Habbie>
indeed, i see that now after viewing a few more pages
<Habbie>
that's neat
<Habbie>
thanks!
<Grommish>
I'm sure they have small/less expensive versions for just electronics. I just have experience with those particular ones
<Habbie>
yeah
<Habbie>
i don't need waterproof
<Habbie>
but it looks semi-universal
<Habbie>
it is, at least, inspiration :)
c0sm1cSlug has quit [Remote host closed the connection]