<pointfree>
azonenberg: How does your timing characterization system work? Do you plot routes and measure time from A to B? Do you also subtract overlapping portions between routes to isolate timing for individual segments?
digshadow has joined ##openfpga
digshadow has quit [Ping timeout: 246 seconds]
http_GK1wmSU has joined ##openfpga
http_GK1wmSU has left ##openfpga [##openfpga]
specing has quit [Ping timeout: 260 seconds]
digshadow has joined ##openfpga
amclain has quit [Quit: Leaving]
amclain has joined ##openfpga
digshadow has quit [Ping timeout: 260 seconds]
digshadow has joined ##openfpga
digshadow has quit [Ping timeout: 260 seconds]
scrts has quit [Ping timeout: 240 seconds]
amclain has quit [Quit: Leaving]
scrts has joined ##openfpga
scrts has quit [Ping timeout: 260 seconds]
scrts has joined ##openfpga
<azonenberg>
pointfree: As of now I find all paths and don't care about overlapping segments
<azonenberg>
I need to work on better handling of synchronous stuff
<azonenberg>
before i do anything else
<azonenberg>
And i also have to gather more data :p
<awygle>
has anybody looked at firrtl?
<awygle>
(chisel's RTLIL)
<rqou>
yosys supports it to some extent
<rqou>
that's all i know
<awygle>
oh does it really? *googles*
<cr1901_modern>
Finite Impulse Response RTL
<rqou>
what if i synthesize an infinite impulse response filter into FIRRTL? :P
<awygle>
oh bummer, it works on the backend.much less interesting.
<rqou>
feel free to add it to the frontend?
<rqou>
patches welcome(TM)
<awygle>
:P
<rqou>
i mean, it works towards the goal of turning yosys into "general purpose circuit playground"
<awygle>
i would not characterize that as _my_ goal, exactly, though it is certainly _a_ goal
<rqou>
right
<rqou>
it's more like azonenberg+my goal
<awygle>
i need to get back to work on place+route pronto, twitter keeps trying to nerd-snipe me
scrts has quit [Ping timeout: 255 seconds]
scrts has joined ##openfpga
digshadow has joined ##openfpga
scrts has quit [Ping timeout: 276 seconds]
X-Scale has quit [Read error: Connection reset by peer]
scrts has joined ##openfpga
scrts has quit [Ping timeout: 255 seconds]
<cyrozap>
pointfree: Thanks for that! The PSoC-to-Verilog stuff doesn't really exist yet, other than the UDB-to-Verilog stuff I did last year. Right now, it's just the old stuff, but copy-pasted to a new directory and modified to support multiple UDBs (but it doesn't actually build yet because I stopped work on it to work on a smaller project).
<cyrozap>
Turns out it doesn't actually support USB 3.0--it's a 2.0 device with a 3.0 micro-B connector.
scrts has joined ##openfpga
<balrog>
cyrozap: hah...
<cyrozap>
And it's not even a UVC-compliant device, so now I have to go RE the Windows drivers.
<balrog>
*sigh*
<balrog>
lol the top review
<balrog>
It only work with the software provided (HD VIDEO BOX); It does not work with Wirecast...; It does not work with Pinnacle Studio
<rqou>
btw has anybody other than me been having difficulty with amazon's shipping "guarantees" lately?
<balrog>
rqou: nah, not me
<balrog>
everything seems to ship reasonably on time
<rqou>
hmm
<cyrozap>
rqou: Their crowdsourced shipping service sucks.
<cyrozap>
balrog: Yeah, I knew what I was getting into, I just wasn't sure how bad it was.
<rqou>
ugh, i really want to build a "just do it why is this so complicated" video capture device
<rqou>
but i don't have time for that any time in the near future
<rqou>
thanks azonenberg :P
<rqou>
(at defcon azonenberg and i roughly created a task list for the next 10 months or so)
jhol has quit [Quit: Coyote finally caught me]
<cyrozap>
rqou: "why is this so complicated" USB, that's why :P
<rqou>
mine wasn't intending to use USB
<cyrozap>
Oh, then in that case the complexity will be in the "how to turn 10.2 Gbps HDMI into 1 Gbps Ethernet (or whatever) without any/with minimal quality loss".
<rqou>
1080p is only around 3gbps
<rqou>
so my goal was just to turn <10gbps HDMI into 10gbps ethernet
<rqou>
the most "fun" is capturing analog video
<rqou>
apparently everybody has to use some shitty asic
<rqou>
as opposed to just taking a giant ADC
<rqou>
and implementing NTSC/PAL decode in software
scrts has quit [Ping timeout: 260 seconds]
<azonenberg>
That's what i'd do if i was doing NTSC input
<azonenberg>
But i wont because NTSC is obsolete :P
<rqou>
but retro! :P :P
scrts has joined ##openfpga
cr1901_modern has quit [Ping timeout: 246 seconds]
scrts has quit [Ping timeout: 240 seconds]
scrts has joined ##openfpga
pie_ has quit [Ping timeout: 246 seconds]
pie_ has joined ##openfpga
digshadow has quit [Ping timeout: 276 seconds]
pie_ has quit [Ping timeout: 260 seconds]
scrts has quit [Ping timeout: 240 seconds]
scrts has joined ##openfpga
teepee has quit [Ping timeout: 258 seconds]
teepee has joined ##openfpga
teepee has quit [Ping timeout: 260 seconds]
teepee has joined ##openfpga
mifune has joined ##openfpga
teepee has quit [Ping timeout: 255 seconds]
scrts has quit [Ping timeout: 276 seconds]
teepee has joined ##openfpga
Hootch has joined ##openfpga
scrts has joined ##openfpga
massi has joined ##openfpga
scrts has quit [Ping timeout: 260 seconds]
scrts has joined ##openfpga
qu1j0t3 has quit [Ping timeout: 240 seconds]
scrts has quit [Ping timeout: 276 seconds]
scrts has joined ##openfpga
scrts has quit [Ping timeout: 260 seconds]
scrts has joined ##openfpga
qu1j0t3 has joined ##openfpga
teepee has quit [Ping timeout: 245 seconds]
teepee has joined ##openfpga
m_t has joined ##openfpga
scrts has quit [Ping timeout: 260 seconds]
scrts has joined ##openfpga
mifune has quit [Ping timeout: 268 seconds]
mifune has joined ##openfpga
mifune has joined ##openfpga
scrts has quit [Ping timeout: 240 seconds]
scrts has joined ##openfpga
mifune has quit [Read error: Connection reset by peer]
<rqou>
why is batch-compiling bitstreams so difficult?
digshadow has joined ##openfpga
<azonenberg>
rqou: just wait till you try doing it in vivado
* azonenberg
tcl's rqou until he gives up trying
pie_ has quit [Remote host closed the connection]
pie_ has joined ##openfpga
<rqou>
wtf, now lxc is broken
<rqou>
alright, i guess since i'm having so much computer difficulty today will be sysadmin day
mifune has quit [Read error: Connection reset by peer]
mifune has joined ##openfpga
scrts has quit [Ping timeout: 248 seconds]
* pie_
pats rqou
<rqou>
now apt is being stupid
<rqou>
i kinda understand why the non-debian distros decided they wanted a SAT solver just to figure out how to install packages
mifune has quit [Quit: Leaving]
<rqou>
apt is insisting it needs to uninstall virt-manager
<pie_>
oh shit lol, i just thought of this and its a thing istrumpimpeachedyet.com
<rqou>
hey, how about HasTheEUExplodedYet.com? :P
<pie_>
why would the eu explode
<pie_>
like...wat?
scrts has joined ##openfpga
<pie_>
or do you mean fall apart
<rqou>
yeah
<pie_>
oh ok
Hootch has joined ##openfpga
<rqou>
goddammit, debian sid seems to have a bunch of uninstallability right now
<azonenberg>
it is called "unstable" for a reason
<rqou>
would you rather have debian stale (stable)?
<azonenberg>
Yes
<azonenberg>
That's my primary OS these days
<azonenberg>
in fact i have some of my VMs still on oldstable but i should realyl fi that
<azonenberg>
fix*
<rqou>
i'm increasingly tempted to move to the weeaboo distro
<azonenberg>
which is that?
<rqou>
arch
<pie_>
true otakus use nixos
<pie_>
are you not a true otaku rqou? ARE YOU NOY?
<pie_>
>:C *guilting rqou into nixos*
test123456 has quit [Quit: Leaving]
cr1901 has joined ##openfpga
<rqou>
cr1901: *gasp* you're using ipv6?
<pie_>
rqou, he cant hear your gasp because theres too much space
<cr1901>
Not intentionally... I'm just using whatever the household I'm petsitting is using
scrts has quit [Ping timeout: 240 seconds]
<cr1901>
That being said, ifconfig says I'm using both a ipv4 and ipv6
<pie_>
ipv6 is doing that just to placate you
<cr1901>
Idk if an interface is supposed to be able to do that, nor do I know how the kernel is choosing which IP to actually use
<pie_>
dont let it take advantage of you
<rqou>
yeah, it's supposed to work like that
<pie_>
(probably routing tables)
<rqou>
it's called native dual stack, and i am running that too
<pie_>
(probably dns)
<pie_>
id guess if you try to access something via ipv6 it uses that
<rqou>
yeah
<pie_>
otherwise ipv4
<rqou>
although there's also a crazy prioritization system to pick the right source address
<pie_>
now, how applications choose to use ipv6 idk
<rqou>
many applications will do it automagically if dns gives an AAAA record
<cr1901>
pie_: Routing tables?
<pie_>
cr1901, yeah nevermind about routing tables
<azonenberg>
Routing tables work for one protocol at a time
<rqou>
ah cr1901 you're on the comcast native dual stack network
* azonenberg
drools
<azonenberg>
i want that
<azonenberg>
My comcast business pipe will give me one /64 DHCP i think
<rqou>
you can get it if you drop business/static-IP
<azonenberg>
but not a static /56 or /48
<rqou>
consumer can get a /60
<lain>
at&t gave us native ipv6 here at my parents
<azonenberg>
rqou: yeah i need static with rDNS control
<lain>
... but the modem+router they issued has a bug that means we can't use it unless we request a motorola modem+router
<rqou>
because at&t's network is dumb
<rqou>
at&t loves shitty CPE boxes
<rqou>
they're practically obsessed with them
<lain>
because apparently 2wire/pace can't be arsed to fix the fact that they borked a config file (although pace has said they issued a fix and at&t is the one who refuses to push it out, and at&t fired back saying that the update borked something else, etc)
<cr1901>
azonenberg: The ppl I'm housesitting for... are comfortable $$-wise. But I don't imagine they chose a dual stack conciously :P
<rqou>
even things that don't need CPE boxes (like GPON) get CPE boxes because at&t loves them so much
<lain>
rqou: yes
<rqou>
cr1901: comcast is rolling out native dual stack for all consumer connections
<lain>
rqou: though I've had surprising success talking directly to at&t techs, which is something I've never achieved on, say, comcast
<lain>
it's just a question of how much I care to actually get someone on the phone
<rqou>
now if only the at&t gpon offering that i keep seeing billboards for would actually exist here
<rqou>
(and if only someone would find the obligatory exploit to pull the .1x certs off of the shitty CPE box)
<lain>
like one time way back when 2mbit/s dsl was the shiznit, ours was running at about 20kbit/s lolol, so I called up at&t and the tech was like "you need to run this speed test" and I'm like "the fucking modem says 20kbit/s link rate despite superb SNR" but this is tier 1 tech so it's script all the way
scrts has joined ##openfpga
<lain>
so I'm playing the game at first, I go to run the speed test, and lo, it requires java. well I didn't have java installed on this machine, and do you know how long it takes to download JRE at 20kbit/s?
<lain>
days.
<cr1901>
How did my interface know to assign an ipv6 addr when connecting to the new router?
<rqou>
slaac
<cr1901>
It doesn't do this at home...
<rqou>
ipv6 is magic
<rqou>
(when it works at least)
<lain>
so I requested to escalate to a tier 2 technician, and the request was granted, and the guy who answered was remarkably frank with me lol, he was very obviously a scruffy network engineer...
<rqou>
too bad it isn't really safe to not have inbound firewalling on the gateway for most home networks, so the goal of end-to-end-routability still has some holes
<lain>
so he confirms the modem's SNR and such with me and he's like "yeah there's no symbol errors, SNR is good, and you've tried power cycling the modem, so it's the DSLAM end that's stuck. how much do you like your neighbors?"
<pie_>
lol
<lain>
and I'm like "uhh.. they're pretty alright, why?" "because I can't reboot a single endpoint remotely, I'm taking down the whole DSLAM riiiiiiiiiight /now/ *click*" lol
<lain>
few minutes later after everything linked back up, we had our glorious 2mbit/s once again :P
<lain>
he and I shared misc. stories while we waited, he was great :3
<rqou>
hey btw, are there any sane ways to easily keep installed packages synchronized between two computers?
<pie_>
rqou, use nixos
<rqou>
f*ck you i'm not using nixos
<pie_>
:D
<rqou>
i am really tempted to move to arch because debian sid is still ridiculously slow
<rqou>
at updating things
<lain>
meanwhile when a comcast business line we used had issues, they would have a tech on site within 30 minutes every single time... but they didn't actually find the problem for MONTHS. in their defense, it was a hilariously hard to find problem, but obvious in hindsight.
<azonenberg>
lain: i usually had to wait a day or so to get a tech on site
<cr1901>
ipv6 looks like a mess... is slaac like dhcp?
<lain>
but even after they figured out the problem and put in a work order to fix it, nobody fixed it for like 3-4 months. the only reason it got fixed then was the construction crew (unrelated roadwork) /ripped the line clean off the pole/ so they had to do a whole new run
<azonenberg>
cr1901: no, it's stateless
<pie_>
lain, was there a fly impaled on the coax line catching radio signals?
<azonenberg>
basically the router sends out a "here i am" message
<azonenberg>
your client just takes the subnet from that packet, concatenates your MAC address, and that's the IP
scrts has quit [Ping timeout: 260 seconds]
<azonenberg>
you never have to even send any packets (although to get online faster you can broadcast a "are there any routers here?" message)
<cr1901>
all on its own? Without any prompting by an external event (except a timer)?
<rqou>
hmm, i wonder how much effort it would be to get comcast to take a look at my "my signal levels are too low to work reliably" problem
<azonenberg>
Correct
<rqou>
which i've currently worked around by removing the STB (and corresponding splitter)
<lain>
pie_: so we're sitting there at our desks working and the tech is back down a dark hallway scoping the line, and it's unusually quiet. normally we have loud music blaring all night... and we hear this train go by (train tracks in the back yard of the building, RIGHT outside), and the tech yells "HOLY SHIT"
<azonenberg>
If you get a SLAAC message on an interface on most modern OSes, you'll get an IP on that subnet, and set the appropriate router as your next hop
<rqou>
because apparently 1/8th of the signal coming from the drop isn't very much :P
<lain>
pie_: and in that instant, all three of us knew exactly what was going on. there was a loose connector somewhere along the tracks. and when the train vibrated it, the SNR went to shit
scrts has joined ##openfpga
<azonenberg>
cr1901: All of my recent FPGA-based instruments work like this
Hootch has quit [Quit: Leaving]
<azonenberg>
they don't even support IPv4
<rqou>
meanwhile my parents' house has an amplifier in the pedestal right outside the house
<azonenberg>
the MAC is printed on the device, you just find its IP by concatenating your subnet to that
<lain>
the modem would lose the link, because long trains are long, and then the train would pass and eventually the modem would re-link :P
<pie_>
xD
<rqou>
azonenberg: why slaac and not just a link-local address?
<cr1901>
azonenberg: dhcp is provided by a daemon, right? slaac is provided by the driver code?
<azonenberg>
rqou: Because link-local only works for point to point
<rqou>
slaac is provided by a daemon on linux
<rqou>
also f*ck texlive
<azonenberg>
rqou: The idea is you just plug it into a LAN drop and everyone on the network can access it
<rqou>
it's always the biggest part (by number of bytes) of every single upgrade
<lain>
pie_: on the bright side, the worker who eventually re-strung the line (after the construction guys ripped it off the pole) said he saw the work order from before about the loose connector, and laughed as he explained that he redid ALL the connectors for the entire run just to be sure, and he strung it as high as he's legally allowed to for the area, to keep it out of reach of the construction crew XD
<azonenberg>
In other news i'm kicking myself for not putting a mac address eeprom on the starshipraider prototype
<rqou>
azonenberg: um, link-local addresses are _link_ local
<azonenberg>
I have some code that auto-generates a random mac based on the fpga die serial number
<rqou>
it works on the entire lan
<lain>
I guess they had never sent anybody the work order, until the line was actually broken. because comcast.
<azonenberg>
rqou: It also is not routable
<rqou>
sure
<azonenberg>
Which means i cant use it over a VPN etc
<azonenberg>
SLAAC is an internet standard, pretty much any ipv6 network should support it
<rqou>
you can use IPv6 NAT :P :P
<azonenberg>
i have never seen one that didn't except manually configured point to point routes between routers or something
* lain
smacks rqou
<lain>
:P
<rqou>
hey, at least this one is stateless
<cr1901>
azonenberg: Well for your devices it's "easier"? i.e. no sleep modes?
<lain>
there is dhcpv6 but then you get into fun situations like "lol my machine obtained both a dhcpv6 and slaac address"
<rqou>
yeah
<rqou>
this is where the weird source address selection rules apply
<azonenberg>
cr1901: if i sleep i just remember the last subnet
<rqou>
ipv6 is _fun_
<azonenberg>
as long as the PHY keeps a link up, nothing changes
<rqou>
um, slaac addresses expire
<azonenberg>
if the PHY is power down, upon wake you just wait for the next router advertisement
<azonenberg>
lain: easy solution, i don't implement DHCPv6 or static
<lain>
azonenberg: yeah
<azonenberg>
And i only support one slaac address
<rqou>
no ULAs?
<azonenberg>
My cores have one, and only one, SLAAC-assigned address
<azonenberg>
as the sole address for the stack
<cr1901>
why is the PHY involved in slaac (I must be missing something)?
<azonenberg>
cr1901: if the PHY loses link
<azonenberg>
you're possibly not on the same subnet when it comes back up
<azonenberg>
So you have to reset the slaac address and wait for another
<cr1901>
oh, that's logical
<rqou>
er, it seems you're not _really_ following the spec
<azonenberg>
I'm assuming nobody is fudging vlans under you
<cr1901>
basically my q is "I never really thought about it, but how does the interface get a new IP on my netbook after it sleeps and wakes up" Ditto w/ any one-off such as your FPGA designs
<azonenberg>
and changing you from subnet to subnet without losing the address
<azonenberg>
without losing the link*
<rqou>
aren't you required to have a link-local address?
<azonenberg>
rqou: I'm supporting the core of the spec required for functionality without cruft
<cr1901>
"and changing you from subnet to subnet without losing the address" how is this possible?
<rqou>
but your instrument can't be e.g. multihomed
<azonenberg>
rqou: It only has one physical interface
<azonenberg>
so...
<azonenberg>
If you have multiple interfaces, you'd run one stack per physical interface and it's up to the rest of the SoC on whether to support routing between the physical interfaces or not
<rqou>
but it can be connected to a router with two upstreams
<rqou>
if you want the instrument accessible to the entire internet :P
<rqou>
that sounds safe :P :P
<azonenberg>
rqou: lol yeeeah
<azonenberg>
I cant rule out this changing down the road but it's pretty rare for something like this to need >1 IP
<cr1901>
"but it can be connected to a router with two upstreams" <--- eeeh I'm lost. TWO upstreams?!
<lain>
>running ipv6 without a firewall
<lain>
let us not forget that firewalls originally worked without NAT, and the two concepts are entirely orthogonal
<rqou>
hey, not any worse than connecting a scada HMI to the internet :P :P :P
<rqou>
er, my parents' router blocks all inbound ipv6 connections and can't be disabled
<lain>
LOL
<azonenberg>
lol
<rqou>
meanwhile my router has one subnet with firewalling and one subnet without
<azonenberg>
I'm excited to purge my cisco gear from the network once i finish that open FPGA based switch
<rqou>
the default is the one with firewalling
<lain>
I went through several consumer routers a few years ago and they either had totally buggy useless ipv6 stacks, or they had no ipv6 firewall whatsoever
<cr1901>
azonenberg: How can you change subnet without losing the link?
<pie_>
cr1901, failover?
<azonenberg>
my EOL'd cisco switches, i think, are the only things left on my network that do not support ipv6
<rqou>
i gave up on consumer routers years ago
<lain>
then I moved to ubnt edgerouters and my life regained meaning
<rqou>
i just use a linux box
<azonenberg>
cr1901: conf t / int g0/8 / sw ac vl 3
<lain>
ubnt edgerouter is basically a linux box designed to be awesome at being a router
<rqou>
(which i still need to go update the microcode on; that's going to be _fun_)
<lain>
and they're /cheap/
<azonenberg>
now you're on vlan 3 which is probably another subnet
<lain>
$50 for the er-x
<azonenberg>
but the PHY never relinked
<lain>
it's 5-port, any of them can be wan/lan, switched or routed, etc
<lain>
</ad>
<rqou>
heh, my router is a 2-port router :P :P
<azonenberg>
It's an obscure enough case i'm not concerned about it, if the box doesn't work after such an operation you can just link down and up
<rqou>
1gbase-t wan port, 10gbase-x lan port
<rqou>
this is a very azonenberg approach it seems :P
<azonenberg>
rqou: please tell me it's at least an optical interface?
<azonenberg>
or well a SFP
<rqou>
nah, twinax DAC to the switch right below it
<azonenberg>
not necessarily optical sfp
<azonenberg>
That counts
<azonenberg>
i meant, not baseT
<rqou>
no, nobody uses 10gbase-t
<cr1901>
"conf t / int g0/8 / sw ac vl 3" Is this a cisco router command or something?
<cr1901>
rqou: How can a router have two upstreams w/ one connection to the interenet?
<rqou>
no, two connections to the internet
<rqou>
you can do this if you have lots of $$$
<azonenberg>
cr1901: yes
<azonenberg>
rqou: see, if i was going to multihome
<cr1901>
rqou: And what does it accomplish? Load balancing or something?
<azonenberg>
i'd have one upstream on each of two separate subnets
<rqou>
usually failover
<azonenberg>
with two switches
<azonenberg>
then have each host have two physical interfaces
<azonenberg>
But any one socket would only work on one physical interface, since presumably the two connections would have different IPs
<cr1901>
rqou: Oh, they're connected to two different ISPs?
<azonenberg>
The alternative, if i had $$$$
<rqou>
btw since it seems nobody noticed my comment about microcode, my "router" is running a kaby lake with the "AH bug"
<azonenberg>
would be to have PI space and run BGP to the internet
<rqou>
i wonder how much latent UB is happening right now
<azonenberg>
In which case each box only has one IP
<azonenberg>
and the router just picks one physical interface or the other to hit the internet from
<cr1901>
or two different "upstreams" from the same IP, so if one link goes down, the other takes over?
<rqou>
at least btrfs has checksums, and i'm not running ocaml :P
<azonenberg>
cr1901: that makes no sense
<azonenberg>
since generally same ISP = same physical layer, same cable plant, same head end equipment
<rqou>
updating the bios on this machine is quite hard because there's no gpu
<azonenberg>
Unless this is like an enterprise ISP that actually has two routers for you
<cr1901>
so failover is two different ISPs typically?
<azonenberg>
and gives you one pipe to each over two separate cable plants
<cr1901>
lol
<azonenberg>
If i were to multihome here, for example
<lain>
rqou: the AH bug.. which one is that? the one where a "specific set of complex architectural conditions" can cause UB?
<azonenberg>
i'd have one CenturyLink ADSL pipe over POTS
<rqou>
azonenberg: you'd be screwed either way because you're on an island :P
<azonenberg>
and one Comcast DOCSIS pipe over coax cable
<azonenberg>
Or maybe Comcast + LTE
<azonenberg>
Then just have each box get two IPs
scrts has quit [Ping timeout: 255 seconds]
<lain>
comcast + dsl + lte + satellite
<lain>
:P
<azonenberg>
lain: lol
<rqou>
i'd laugh if comcast+centurylink just both turn into two dwdm wavelengths on the same strand of fiber leaving the island
<lain>
rqou: very likely tbh
<azonenberg>
rqou: Totally possible, yeah
<cr1901>
azonenberg: How would a single interface on each box know which IP to choose
<azonenberg>
cr1901: It wouldn't, you'd have one interface per IP (or well, one subinterface if you had vlans trunked to it)
<azonenberg>
But that wouldn't give you failover to switch/cable plant issues internally
<cr1901>
Then what's the point ._.?
<azonenberg>
Full redundancy would mean each host has two lan ports, one to each of two switches
<azonenberg>
each switch goes to a different routr and then to a different ISP upstream
<azonenberg>
you can now lose any cable, router, or switch and maintain internet access
<rqou>
oh btw china made dwdm equipment so ridiculously cheap
<azonenberg>
although if you lose one, the server is unreachable via that IP and accessible by the other
<rqou>
fs.com is amazing
<lain>
yes
<azonenberg>
And any sockets to the down IP will be dead obviously
<cr1901>
azonenberg: That's assuming your OS knows to swap interfaces automatically?
<azonenberg>
cr1901: you just bind your socket to ::0
<azonenberg>
and it'll accept connections on either IP
scrts has joined ##openfpga
<azonenberg>
As a client, you'd need to poke the routing table to prefer one, and detect if the path was unroutable and fail over to the other
<cr1901>
hmmm
<azonenberg>
once the socket is open, you'll just keep using that interface
<azonenberg>
rqou: do you have a good source for cheap(ish) 10g host adapters?
<azonenberg>
pcie 1- or 2-port SFP
<rqou>
lain: anyways, the bug i was thinking of involved somehow using AH/BH/CH/DH in a loop with HT enabled
<rqou>
i currently use expensive intel x520s
<cr1901>
"And any sockets to the down IP will be dead obviously" <-- you mean remote sockets to the interface that just went down, correct? (well they'll be dead in the other direction too, but still)
<rqou>
there are some mellanox/chelsio cards that are really cheap on ebay, but i've never tested them
<rqou>
the intel cards are unfortunately on the order of $100
<rqou>
old mellanox/chelsio cards are about $20
<rqou>
btw, the xH registers on x86 were a mistake :P
<rqou>
they mess up store-to-load forwarding and cause all kinds of bugs
<rqou>
<old joke>don't make me LAHF</old joke>
<cr1901>
They weren't a mistake in 1978...
<rqou>
yeah yeah 8080 compatibility
<cr1901>
azonenberg: Last q (warning: pedantic), when you said "And any sockets _to_ the down IP..." doesn't a socket describe a to/from relationship :P?
<rqou>
but the layers of legacy lead to fun things like "mov ah, sil" being impossible
<cr1901>
sil?
<rqou>
x86_64 added sil/dil/bpl/spl
<rqou>
for the low 8 bits of rsi/rdi/rbp/rsp
<cr1901>
oh ffs...
<rqou>
but the REX prefix is needed to access them
<rqou>
and the REX prefix loses access to AH/BH/CH/DH
<cr1901>
azonenberg: or did you deliberately omit the "from the down IP" part?
<cr1901>
x64 encoding is just... so utterly horrible
<rqou>
which reminds me that i don't think i've ever seen a compiler actually output code that uses some of the weirder things like r8w
<cr1901>
(tbf 8086 isn't fun either, of course much easier, but still sucks)
<rqou>
and i've rarely seen a compiler use xH either
<rqou>
hilariously, i _have_ seen sil/dil get used
<rqou>
cr1901: you mean you don't like things like the VEX/REX prefixes?
<rqou>
or modrm?
<cr1901>
modrm sucks on 8086 and by the transitive property it sucks on x64
<rqou>
what about stacking segment overrides :P
<cr1901>
Ehhh, less bothered by that, you get used to it
<cr1901>
Not that you should be doing it anyway
<rqou>
even 0x64 0x64 0x64 0x64 0x90? :P
<rqou>
fs:fs:fs:fs:nop
<rqou>
(at least i think this is supposed to work)
<cr1901>
that... doesn't do anything? Add 12 more prefixes for maximum entertainment.
shiyer has joined ##openfpga
<azonenberg>
cr1901: i meant any socket with the down IP on either end
<azonenberg>
will end up timing out
<cr1901>
azonenberg: I figured as much. But knowing me, there's some $SUPER_COMPLEX_EDGE_CASE that I was missing since I'm not a certified TCP/IP expert (tm)
<cr1901>
(read: I never had to implement it myself, so I'm only familiar with the parts I've had to debug in my own applications)
<rqou>
whee, debian fixed the radeon driver not loading and broke my xserver/lxdm launching
<rqou>
i am seriously considering migrating to arch
<rqou>
like, right now
<rqou>
sorry azonenberg, but it looks like your .jed->.json converter is getting delayed
<azonenberg>
rqou: stop using unstable
<azonenberg>
unstable is going to have issues like this
<cr1901>
I've been fairly satisfied w/ Ubuntu. Easy to get alternate repos to upgrade LTS packages when I need extra features (e.g. geany for Rust). Pity dpkg is deeply unpleasant package format
<azonenberg>
use testing (at best)
<azonenberg>
or stable (which has been very stable for me)
<rqou>
stable has been way too stale for me
<azonenberg>
What do you need newer for?
<rqou>
testing is supposedly the worst of both worlds
<azonenberg>
i just build from source for kicad
<rqou>
gpu f*ckery
<azonenberg>
nvidia blobs have always run fine for me on stabkle
<azonenberg>
stable*
<cr1901>
azonenberg: Don't they provide repos for newer versions (kicad)?
<azonenberg>
no experience with ati
<azonenberg>
cr1901: no idea but i do dev sometimes so i always build from source
<rqou>
also, i've been burned by stable releases far too many times at this point
<rqou>
every time i've upgraded, things massively fell apart
<cr1901>
I'd love to build from source on this Intel Atom netbook, but...
<cr1901>
I also want to be able to use my netbook
<azonenberg>
rqou: on debian stable?
<rqou>
no, with ubuntu
<azonenberg>
i have not had that experience at all
<azonenberg>
ubuntu is a trainwreck
<azonenberg>
its based on debian unstable :p
<rqou>
i've never had a ubuntu upgrade work
<azonenberg>
or testing? eiuther way, too old
<azonenberg>
i mean, too unstable
<azonenberg>
gaah
<azonenberg>
Debian just works
<azonenberg>
its not the latest and greatest, but it gets the job done
<rqou>
not when you want to update mesa
<azonenberg>
only issues i've had with it have been on recent laptops with an old debian version
<azonenberg>
where i had to upgrade to a more recent kernel to get some driver
<azonenberg>
But that's totally understandable and not a problem once you get it set up
<rqou>
the debian community also really gives me the feeling of "you're not in the loop? too bad, we're ignoring you"
* cr1901
better go and feed the fur and fangs
<azonenberg>
i have no real involvement with the community there
<azonenberg>
i just download the OS and use it
* pie_
whispers something in rqou's ear
<rqou>
ugh, the problem is that the nvidia driver is missing
GenTooMan has joined ##openfpga
<rqou>
so i'm futzing around, and loading the amdgpu kernel module causes iommu errors
<rqou>
that's definitely a good sign
<rqou>
/s
scrts has quit [Ping timeout: 276 seconds]
<pie_>
yay?
<pie_>
so does that give you a privesc vuln in the fabric of the universe or does that still need a little work?
<rqou>
nah, the iommu blocks it so everything is fine
<rqou>
but my xorg is still broken because debian did something inexplicable
scrts has joined ##openfpga
<pie_>
aw
<pie_>
guess we cant have catgirls yet
<jn__>
catgirls?
* jn__
is confused
<azonenberg>
jn__: me too, lol
<azonenberg>
In other news, yay for reading language standards
<azonenberg>
:p
<jn__>
[still off-topic, but differently] btw, if someone needs a new anime, i can really recommend Steins Gate (and i don't usually watch anime)
<pie_>
obligatory ghost in the shell original movie mention
<pie_>
kthx
<pie_>
for the rare that may not have seen it
<pie_>
:P
scrts has quit [Ping timeout: 276 seconds]
scrts has joined ##openfpga
pie_ has quit [Read error: Connection reset by peer]