<rqou>
i got an irs scammer from (202) 335 4243, have fun
<rqou>
*4242
<rqou>
azonenberg: any of your friends enjoy scambaiting?
<Zorix>
how do you know its real? they spoof caller id
<Zorix>
call back number? heh
<rqou>
yeah, this is the call back number
<rqou>
remember to scambait responsibly
soylentyellow has joined ##openfpga
<Zorix>
nice
digshadow has quit [Quit: Leaving.]
digshadow has joined ##openfpga
Dolu has quit [Ping timeout: 246 seconds]
<awygle>
tinyfpga: your bootloader works on ecp5s, right?
[X-Scale] has joined ##openfpga
X-Scale has quit [Ping timeout: 260 seconds]
[X-Scale] is now known as X-Scale
genii has quit [Quit: GO LEAFS GO !!]
digshadow has quit [Quit: Leaving.]
digshadow has joined ##openfpga
digshadow has quit [Ping timeout: 240 seconds]
pie__ has quit [Ping timeout: 240 seconds]
unixb0y has quit [Ping timeout: 240 seconds]
unixb0y has joined ##openfpga
digshadow has joined ##openfpga
<tinyfpga>
awygle: yup! I have it working on my TinyFPGA EX boards right now
<awygle>
tinyfpga: is there anything "fancy" about programming the bitfile? Like a CRC to be calculated or anything?
<awygle>
Or can you just dunk it into the flash and call it a day?
<tinyfpga>
awygle: the bitstream file has everything it needs already, you just write it to the flash as is
<awygle>
tinyfpga: hm okay thanks, must be something wrong with my programming algorithm then
<tinyfpga>
awygle: what board are you using? Which bitstream file format? The plain binary format or the .mcs file?
<awygle>
tinyfpga: the .bit file
<awygle>
tinyfpga: it's a custom board
<tinyfpga>
awygle: if you have a JTAG connection you can read the status register from the FPGA and it has several bits that describe any errors in loading the bitstream from SPI flash
<tinyfpga>
it’s very useful
<awygle>
tinyfpga: interesting, can I do that through Diamond?
<tinyfpga>
awygle: yes, I just used the diamond programmer tool
<rqou>
do note that i'm not at all in tune with popular culture
<rqou>
homoclinic orbits are a fun example of nonlinear systems trolling you :P
<awygle>
Nonlinear is one of those words that college treats as "don't even bother" but which is actually very tractable in a huge number of circumstances
<rqou>
lol yeah
<awygle>
But yes, sometimes it trolls you :-P
* rqou
is currently taking ee222
<rqou>
decent fun
<rqou>
would recommend
<rqou>
it's great to be at a school that's pretty well known for controls
<rqou>
goddammit
<rqou>
i just found in the lecture notes the missing theorem i needed for the previous problem set
<rqou>
awygle: "you can never have enough linear algebra" <-- agree or disagree? :P
<rqou>
also azonenberg, agree or disagree?
<awygle>
linalg is great
<awygle>
9/10 mathematical discipline
<rqou>
awygle: btw i've been told the new ee16a/ee16b (replaces 20/40) are much heavier on the "fancy school" approach of "turn circuits into linalg, now it's 'just' math"
<azonenberg>
rqou: it's good to know, but honestly i rarely use it
<azonenberg>
When i need it, i really need it
<azonenberg>
:p
<awygle>
I got my first job in part because I solved a linear system in like four ways
<awygle>
Laplace, direct integration, linear algebra, and "oh I've seen this before"
<rqou>
was maharbiz teaching ee40 at that time?
<awygle>
mmmmaybe? I mostly remember the girl I sat next to in lectures tbh
<rqou>
lool
<awygle>
I definitely didn't do linear algebra in 40
<rqou>
i did some
<rqou>
i did a crapton in 20
<awygle>
I did a lot in 20 but I didn't recognize it as such, really
<awygle>
I took 20 my first semester, I was still very "ooo college"
<rqou>
nooooo bad idea
<rqou>
:P
<awygle>
It went fine
<rqou>
well, that's good
<rqou>
i was told that the grade distribution for that class was always bimodal
<rqou>
with those who were good at linalg in one half and those who weren't in the other
<awygle>
I think that's pretty fair yeah
<awygle>
One of those "abstract reasoning" things. "how good are you at generalizing?"
<rqou>
yeah
<awygle>
I found 40 much harder because it was very slap dash and didn't have a strong theoretical/mathematical foundation
<rqou>
did you do the eeg project?
<awygle>
Nor any particular motivation for learning
<awygle>
"this is a transistor! It's important" "if you say so dude"
<rqou>
did you do the eeg project in 40?
<awygle>
No. We did a PCB and a bunch of random op amp labs
<rqou>
what was the pcb?
<awygle>
So many damn op amps in 40, and I've never used one since
<rqou>
lol same
<awygle>
Msp430, coin cell, ldo, cap touch, 7-seg, and a button I think
<rqou>
hrm
<awygle>
It's hanging on the wall in the office but I can't see it from here :-P
<azonenberg>
awygle: you havent used an opamp since? o_O
<rqou>
probably not maharbiz, or at least an "early beta"
<rqou>
i'm really sad they killed the eeg lab
<rqou>
*project
<awygle>
azonenberg: nope. Not in anger
<rqou>
it was quite fun watching people debugging the eeg project (basically just a high-gain instrumentation amplifier)
<rqou>
you could immediately tell which boards were done by people who "did this before"
<awygle>
"You need ground hooked up. It's kind of important."
<rqou>
lool
<azonenberg>
rqou: lol
<rqou>
i saw >1 board with all the resistors lined up in a neat row, and then all of the capacitors lined up in another row
<awygle>
I once "inverted a signal" on homework by just connecting VCC to GND and GND to VCC
<rqou>
the noise level wasn't nearly as good as desired :P
<azonenberg>
rqou: o_O
<azonenberg>
wuut
<azonenberg>
i sometimes forget just how awful some folks are at pcb design
<rqou>
oh, this was a PTH board
<awygle>
I had a systems engineer who always submitted comments like that in pcb reviews....
<awygle>
"can you make this symmetrical?"
<sorear>
is that good or bad
<rqou>
awygle: like steve jobs and the ram array? :P :P
<awygle>
sorear: it's not *inherently* bad, but it's stupid, especially on analog boards where the physical layout actually affects performance
<awygle>
Symmetry can be a design criteria but for most things it should be at the bottom of the list, well after "shortest loop" and similar
<rqou>
they tried to get people to make loops shorter by giving more extra credit to smaller boards
<rqou>
this again only worked against people who had already done this before
<azonenberg>
rqou: if i was getting extra credit for smaller boards
<sorear>
against?
<azonenberg>
i would annoy the fsck out of the TA
<rqou>
as in, those were the only people affected
<awygle>
"make the tiniest board" is a really fun design challenge
<azonenberg>
Because my board would be all 0201s and WLCSPs
<rqou>
azonenberg: restricted to PTH only, and kit parts only
<awygle>
azonenberg: pth only
<azonenberg>
awww where's the fun in that?
<rqou>
i made it smaller by making the resistors stand up vertically
<awygle>
Constraint is the soul of creativity
<rqou>
also, iirc single-side load only
<rqou>
so no weird tricks with staggering DIPs
<awygle>
We did double sided
<azonenberg>
rqou: i'd probably still make a smt version because i could
<rqou>
hmm, maybe that was allowed
<azonenberg>
then a pth version for credit
<awygle>
But parts on only one
<azonenberg>
i dont think i have ever built a 100% PTH board, now that i think about it
<rqou>
awygle: yeah, double sided traces, single side load
<azonenberg>
i jumped from breadboarded dips straight to smt pcbs
<rqou>
yeah, me too
<rqou>
although for extra credit my board had an smd atmega on it
<awygle>
I did one as an intern
<awygle>
Awful awful board
<rqou>
connected to the output of the instrumentation amp (which i screwed up and had to add an airwire for)
<azonenberg>
and these days i dont even use soics or qfps uinless i cannot get the part any other way
<awygle>
That's also hanging on my wall to keep me humble :-P
<rqou>
and also connected to a pair of hobby servo connectors
<awygle>
QFNs are the worst package, I will die on this hill
<rqou>
so that the servos would turn and actuate some very-shittily-constructed triangular-shaped pieces of paper when you moved your eyes
<azonenberg>
awygle: dual row qfns are bad
<rqou>
no points for guessing what this was supposed to be an imitation of :P
<azonenberg>
rqou: LOL
<rqou>
azonenberg: i assume you don't like Atheros/QCA wireless parts? :P :P
<azonenberg>
rqou: never used them
<azonenberg>
but single row qfn? i actually use that as my *prefered* package
<azonenberg>
when picking parts on digikey i search qfn first
<azonenberg>
then expand out if i dont see anything that meets spec
<awygle>
azonenberg: I know, but you're wrong :-P
<rqou>
the mips wifi socs in those cheap travel routers are pretty infamously dual-row qfn
<rqou>
also azonenberg why "LOL?"
<awygle>
QFP far superior unless doing RF or very high speed digital
<awygle>
BGA also superior
<azonenberg>
rqou: bci cat earsw
<azonenberg>
ears*
<azonenberg>
right?
<azonenberg>
awygle: 1mm bga is ideal
<azonenberg>
but not a lot of stuff comes in that other than fpgas
<rqou>
did you know about this product before i mentioned it?
<azonenberg>
rqou: yes
<rqou>
O_o how? $WIFE?
<azonenberg>
rqou: i went to a tech school
<rqou>
lol ok
<azonenberg>
the guy who sat next to me in comp org was a furry
<azonenberg>
if you were not a weeb, you were a minority
<rqou>
btw in case it's not abundantly clear, i also own a set :P
<azonenberg>
sooo...
<awygle>
Yeah "servo plus triangles" is an easy one
<azonenberg>
awygle: QFP is much more prone to shorting than qfn in my experience
<azonenberg>
0.5mm qfp self-aligns poorly and bridges at the slightest provocation
<azonenberg>
0.8mm qfp is pretty forgiving, but it's massive
<azonenberg>
soic is even worse
<azonenberg>
tssop has all the downsides of tqfp but takes up more space
<awygle>
QFP rarely shorts and if it does I can fix it with an iron. QFN shorts all the time, or voids, or... And can't be reworked without hot air which I don't have
<rqou>
you should really get a hot air station
<awygle>
Yeah I should
<rqou>
imho hot air > iron for tiny parts
<awygle>
I only use my iron for rework these days, basically
<azonenberg>
yeah see, i reflow everything
<awygle>
I have an LNA board that's like seven parts and I still couldn't be bothered to iron it
<azonenberg>
i'm actually looking at moving to pin-in-paste connector reflow at some point
<azonenberg>
so i dont have to use the iron to put on jtag connectors etc
<azonenberg>
i can and do rework QFN with an iron
<rqou>
O_o you can do that?
<azonenberg>
the only time it bridges is excess solder
<awygle>
It sounds like you might be using bad stencil or solder mask rules for your qfps tbh
<azonenberg>
awygle: no its mostly misalignment
<rqou>
i always thought pth connectors were soldered with a row of chinese ladies :P
<azonenberg>
rqou: if the connector is reflow compatible thermally speaking
<azonenberg>
you can stencil paste right into the holes, stick the connector in, then oven it
<rqou>
that works?
<awygle>
Yeah
<azonenberg>
you may have to add extra solder afterwards
<awygle>
Not uncommon
<azonenberg>
they sell little 0402 sized blocks of solder you can PnP onto the paste
<awygle>
Careful you're getting real connectors tho
<azonenberg>
to supplement the paste
<sorear>
Is "row of chinese ladies" a term of art or do you literally mean a manned assembly line?
<rqou>
how come many of the devices i've disassembled still show signs that pth connectors are hand-soldered
<azonenberg>
rqou: becaise ot'
<azonenberg>
because it's probably cheaper when labor is... chinese prices
<awygle>
sorear: he means an assembly line
<azonenberg>
also, many connectors are not reflow compatible, thermally
<azonenberg>
especially if made by the lowest shenzhen bidder
<awygle>
Many connectors are not iron compatible either :-P
<rqou>
like mine that don't even take solder reliably? :P
<awygle>
My cheap 100mil headers melt the plastic block away from the pin when soldering
<rqou>
wait awygle did you also do what i did and buy a bag of the cheapest headers from aliexpress?
<awygle>
Yup. Except Amazon.
<azonenberg>
This is why i always buy parts from digikey
<azonenberg>
they may cost 5x as much but they're going to work
<awygle>
I usually don't populate headers tho
<rqou>
literally an order of magnitude more expensive
<rqou>
even adafruit has cheaper headers than digikey
<awygle>
I have some press fit ones that I use for programming because i don't have any pogo pin stuff
<rqou>
(those work ok too btw, adafruit's sourcing team must be pretty good)
<azonenberg>
rqou: so, the way i see it is
<azonenberg>
if your project's budget is significantly affected by the cost of headers
<azonenberg>
your project is boring :p
<azonenberg>
and needs more FPGAs and GaAs RF parts
<rqou>
sure, but i have a bunch of boring "hack together some uC stuff" projects :P
<rqou>
ugh i hate time-varying systems
<awygle>
GaAs is not that expensive tbh
<rqou>
breaks all my intuition
<awygle>
GaN is, of course
<azonenberg>
awygle: yeah actually the most expensive analog part i've used was SiGe
<awygle>
Connectors are usually a significant part of my costs (although that's usually SMAs)
<azonenberg>
a $40 comparator
<azonenberg>
With bandwidth to 10 GHz :D
<awygle>
azonenberg: I remember the blog post :-P
<rqou>
btw the other day i discovered another fun thing that's expensive: SiC
<azonenberg>
I havent used that yet
<azonenberg>
awygle: well SMAs are a whole other can of worms
<rqou>
yeah, probably not your "type" of design
<azonenberg>
good ones are like $10 a pop
<azonenberg>
rqou: thats mostly power right?
<rqou>
afaik SiC is mostly used for power electronics
<awygle>
Actually the most expensive parts I tend to use are PbZrTi
<azonenberg>
awygle: o_O
<azonenberg>
wtf is that
<awygle>
SiC is cool
<rqou>
FRAM?
<awygle>
High voltage
<azonenberg>
lead and zirconium...
<awygle>
rqou: ding ding ding
<sorear>
O3?
<azonenberg>
I know i put like a $5 MLCC on a design once
<azonenberg>
so you effectively double your core power
<sorear>
I'm wondering how they arrived at that number given that there's no listed derating
<rqou>
O_o new raspi
<rqou>
now with a transferred-to-cypress wifi chip and slightly less crappy ethernet
<rqou>
nvm, apparently cypress got the previous wifi chip too
<rqou>
so i guess the main improvement is less crappy lan
* azonenberg
still thinks octavo is the way to go if you want a "linux in a black box" right now
<azonenberg>
and dont need high performance
<rqou>
i would agree except i don't particularly like the soc
<azonenberg>
yes, i'd love a i.mx som
<rqou>
someone convince them to make an imx6 sip
<azonenberg>
or sip
<azonenberg>
but for my case, as a sshd-in-a-chip
<azonenberg>
it's fine
<rqou>
btw azonenberg afaik there's no iommu in that chip
<rqou>
although afaik also there's _no_ mobile soc with actual proper iommu for the gpu
<rqou>
i don't understand why this is so friggin hard
<rqou>
iiui arm even _has_ an iommu ip core available
<sorear>
how do we decide which bus masters get iommus?
<rqou>
easy: all of them :P
<azonenberg>
Or go full antikernel and have every device on the bus enforce its own acls
<sorear>
can you usefully configure an iommu on a core?
<azonenberg>
per page
<azonenberg>
i am increasingly convinced thats the only way to go
<rqou>
azonenberg stop making things harder to hack!
<rqou>
correction: please make my/our devices harder to hack and ninty/other people's devices still easy to hack :P :P
<whitequark>
don't worry only nerds will use azonenberg's architecture anyway
<rqou>
true enough
<rqou>
x86 trashfire is still Good Enough(TM)
<sorear>
if your gpu is expected to run opencl it needs to be able to access the memory of random user processes
<whitequark>
for everyone else it's "we need to ship yesterday and ~security~ is what's preventing that??"
<sorear>
which seems to be leaving the domain of traditional iommus
<rqou>
er, why?
<azonenberg>
sorear: no, it needs to be able to access memory you've specifically asked it to
<rqou>
the driver just needs to poke the iommu when you submit new work to the gpu
<azonenberg>
rqou: in any case i am not super concerned about the gpu on that chip
<azonenberg>
as i wont be running X or a gui whatsoever
<sorear>
rqou: the shader can access host memory using host *virtual* addresses
<azonenberg>
and i wont have any ability to run third party code on it
<rqou>
sorear: wait _what_
<rqou>
since when?
<rqou>
also, how does that work even without an iommu?
* azonenberg
remembers the golden days of CUDA when you had to explicitly cuMemcpy to push data to/from the gpu
<azonenberg>
call me weird but i like explicit NUMA
<azonenberg>
rather than random pointers that could go anywhere
<whitequark>
IOMMU is page granularity
<rqou>
azonenberg do you know anything about Vulkan?
<sorear>
rqou: the GPU makes bus requests on behalf of a specific PID, the host IOMMU then walks page tables to translate the address
<whitequark>
no matter what weirdness the userspace submits you can set the mappings up
<rqou>
also allows sony to f*ck up and open the stack :P
<sorear>
I uh am not super familiar with the history of this area, I read a slide deck on AMD's HSA work once
<rqou>
hrm, not sure how that might work, but if you just make the gpu iommu pages equal to the usermode process iommu pages, you're already done
<sorear>
but CPU cores and GPU cores have differently constructed ALUs but increasingly convergent views of the memory system
<rqou>
can't pwn the kernel anyways because the userspace process can't access the kernel
<azonenberg>
sorear: meanwhile i take the exact opposite viewpoint
<azonenberg>
You should mostly work in physical addresses
<azonenberg>
And virtual addresses are a convenient fiction that exists within the scope of one thread so they can do things like creating arrays >2KB in size without jumping around a linked list of pages manually
<azonenberg>
but any time you hand off a pointer to somebody it should be physical
<whitequark>
that breaks down if you hand off a pointer to something large
<rqou>
hrm, but you can't aslr that
<azonenberg>
rqou: physical aslr?
<azonenberg>
whitequark: message passing normally has a fixed MTU
<rqou>
hey, nintendo ended up forced to do it :P
<azonenberg>
you can always send a message containing a list of page addresses
<azonenberg>
Like a multilevel page table
<rqou>
btw i wonder how long it will be until nintendo/sony just end up in every textbook of how not to build a cryptosystem? :P
<azonenberg>
And you can easily just have a libc api that says "create descriptor for this block of virtual memory"
<azonenberg>
and it'll make an array of each page in there
<azonenberg>
rqou: i found client code not long ago that used hard-coded constants plus the current time to generate crypto keys
<azonenberg>
so, they have competition
<rqou>
still better than sony
<rqou>
by a tiny tiny bit
<rqou>
also, session keys or ecdsa rands?
<azonenberg>
Session keys
<rqou>
eh, much better than sony :P
<sorear>
i need a better account of why larrabee failed, because larrabee-but-better is the direction I see things going
<rqou>
at least it doesn't leak the private key that way
<azonenberg>
Well i mean there were easier ways to root the thing
<sorear>
the gpu isn't a device, it's a parallel set of cores optimized for latency-insensitive vector work and fixed-function graphics hardware which uses the same kernel scheduler
<azonenberg>
and it returns a PID or a signature error
<azonenberg>
There's currently no support, or plans, to do inter-core migration of threads
<azonenberg>
But i could plausibly do that down the road once i implement context serialization to ram in antikernel v0.2
<rqou>
azonenberg: i'm waiting for the day you actually make an antikernel NoC and then i pwn it immediately by glitching it :P
<azonenberg>
rqou: It's not intended to be physically tamper resistant
<azonenberg>
so i'm sure that will be totally possible
<rqou>
although i've been having no luck with glitching so far
<azonenberg>
the primary threat model is, bad guy has arbitrary code exec within one thread context and/or a rogue IP core
<rqou>
other than glitching the ice40 instead of the victim
<azonenberg>
and wants to escalate laterally to another subsystem
<sorear>
azonenberg: ISA heterogenity is a bridge that will need to be crossed, yes, although you probably know what I work on and my thoughts on the viability of a common core
<rqou>
when are we going to get a ZoidbergHSM? :P :P :P
<azonenberg>
sorear: i dont actually
<sorear>
(I came here from #riscv)
<azonenberg>
ah ok
<azonenberg>
antikernel v0.2 is going to be riscv instead of mips for the cpu
<azonenberg>
basic integer isa only, with no privileged insns
<azonenberg>
i may steal the syscall insn from the privileged isa to trap to hardware
<azonenberg>
rqou: That would be powered by a pair of supercaps
<azonenberg>
and live inside a sealed glass ampoule with metal sputtered on the inside
<azonenberg>
powered by a fiber optic laser
<azonenberg>
alternating running off one cap and charging the other
<azonenberg>
nice big toslink fiber beaming a few mW in and out, then a smaller data fiber
<azonenberg>
zero direct coupling of external power to internal stuff
<rqou>
how does the glass ampoule help you?
<azonenberg>
rqou: emi shield, also tamper resistance
<azonenberg>
you seal it with gas at an unknown pressure inside
<azonenberg>
then put a pressure and temp sensor inside
<rqou>
O_o wtf
<azonenberg>
if the glass breaks, the pressure changes withotu a corresponding temp change
<rqou>
that's an interesting idea
<azonenberg>
only way to defeat that, that i know of
<rqou>
i don't think i've ever heard of an idea like that
<azonenberg>
is to use ultrasound through the glass to try to estimate the pressure inside
<azonenberg>
then open inside a calibrated glove box / vacuum chamber
<azonenberg>
and do all your invasive stuff in that
<azonenberg>
Have fun :p
<azonenberg>
(at least until you defeat the senosr and replace it with a stand-in)
<awygle>
if you know the gas you should be able to get pressure from temperature and weight, no?
<azonenberg>
awygle: you'd have to know the mass of the whole HSM
<azonenberg>
before adding gas
<awygle>
Maybe even delta-T
<azonenberg>
you could always add random weights to it
<awygle>
Ehh no maybe not
<azonenberg>
you could also mix e.g. helium and sf6
<azonenberg>
to create an unknown gas density
<azonenberg>
(in random fractions)
<rqou>
hmm, i wonder if the pressure causes a measurable optical effect in the glass?
<awygle>
Freeze the gas?
<azonenberg>
awygle: temp sensors in the hsm, it wipes if it gets too hot/cold
<azonenberg>
rqou: again, figuring out the pressure isnt the big challenge
<sorear>
use a laser to drill a <1µm hole in the glass, measure the egress rate and composition, then plug the hole
<azonenberg>
it's creating a vacuum chamber at EXACTLY the right pressure
<azonenberg>
then breaking the ampoule without causing a pressure spike
<azonenberg>
The goal is to take a simple invasive attack and, using low-tech defenses, massively increase the difficulty and expense
<azonenberg>
you can put a light sensor on the pcb too
<azonenberg>
to require the attack to occur in complete darkness
<sorear>
azonenberg: is this thing going to wipe itself if I have a loud conversation next to it
<rqou>
hrm, i wonder if an extremely skilled glassworker could fill another much larger ampoule with gas at the same pressure and fuse it with the victim ampoule
<azonenberg>
rqou: but you still have to do your invasive attack inside that
<awygle>
Conformal coat in transparent silicone rubber, then shatter glass
<rqou>
yes
<rqou>
hmm awygle's idea is better
<azonenberg>
awygle: pressure spike when it breaks, no?
<azonenberg>
in any case i'd love to build one then red-team it
<rqou>
i don't think so?
<azonenberg>
these attacks are all hypothetical
<awygle>
(or a superior rubber that's airtight, I know nothing about rubbers)
<rqou>
hmm, i actually want to build this now
<azonenberg>
this has been on my wishlist for years
<azonenberg>
i came up with it in a chat with monochroma
<rqou>
other than an accurate pressure, sealing a circuit in an ampoule isn't _that_ hard
<awygle>
azonenberg: depends how you break it and what your sensor can handle. It's not hard to break glass e.g. thermally
<azonenberg>
also, for extra fun
<azonenberg>
have two spring contacts pressing against the sputtered metal film on the glass
<azonenberg>
and measure resistance :)
<awygle>
Magnetic field phreaking
<whitequark>
you can likely still glitch it with EMI
<rqou>
i was thinking coat in conformal coating but somehow inject HF without allowing air to leak out
<azonenberg>
whitequark: even inside what's basically a faraday cage, with optics as the only interface?
<awygle>
Yeah, specifically M will be hard to shield with sputtered film
<whitequark>
or a radiation source
<azonenberg>
M yes, hmm
<whitequark>
azonenberg: a thin layer of metal won't work very well as a faraday cage
<azonenberg>
you could probably add sensors to wipe if either was seen
<azonenberg>
a coil + peak detector
<azonenberg>
then say a big dram with a really low refresh rate
<azonenberg>
look for seus
<sorear>
I mean hsms are a pretty mature field
<awygle>
I can't remember my RFI math right now but shielding is proportional to depth for sure
<azonenberg>
sorear: yeah, and the plastic wraps are pretty solid
<azonenberg>
in general for a hsm the goal isnt to survive a glitch untourched
<whitequark>
azonenberg: that will mean your HSM dies from cosmic rays :P
<azonenberg>
it's to wipe/panic if you see one
<azonenberg>
whitequark: you'd set a threshold
<azonenberg>
more than X seus per hour = problem
<sorear>
and you're also starting to sound a lot like the drexler-fan-community
<awygle>
High Z incident particle, glitch with the secondaries
<awygle>
Your Faraday cage is now my linac target
<sorear>
I have one real question though
<azonenberg>
good luck not wrecking it :p
<sorear>
Why 'Zoidberg'
<azonenberg>
sorear: my last name sounds kinda like zoidberg
<azonenberg>
back in grad school, everyone in the electronics club called me that
<awygle>
and they're both doctors
<azonenberg>
i adopted the persona and would greet classmates by scuttling sideways woopwoop'ing and clicking my claws
<whitequark>
LOL
<azonenberg>
(those who were in on it, at least)
<azonenberg>
the running joke was i was only getting the PhD so that when people called me Dr. Zoidberg
<azonenberg>
i'd have the paper to go with the name
<sorear>
not John, although that can be fixed
<awygle>
azonenberg: I could absolutely avoid "wrecking" it. I'd have a hard time glitching usefully but that's also true of, like, my printer - I don't actually know how to do that
<azonenberg>
awygle: when i say wrecking, i mean introducing glitches that produced a desired effect without also introducing glitches into my sensors and triggering the panic
<azonenberg>
radiation is a pretty hard glitching mechanism to control, no?
<azonenberg>
you can get SEUs but not targeted
<whitequark>
you can focus the beam
<awygle>
Depends entirely
<whitequark>
if you know where the dram is
<azonenberg>
so as long as you have a SEU sensor in an e.g. unused block ram
<awygle>
Linacs are pretty focused
<azonenberg>
Put them on die with your crypto core on the fpga then
<azonenberg>
use a bram as a seu detector
<awygle>
Z sputter is harder, depends on the geometry etc
<sorear>
beam focusing for colliders is ridiculous
<awygle>
(I'm imagining a medical linac because that's all I've used)
<azonenberg>
also keep in mind you're hitting an opaque object
<azonenberg>
and you dont know exactly where inside it the pcb, fpga, etc are
<whitequark>
just xray it
<awygle>
azonenberg: you can make one of anything secure. I would buy two, and then I would know that.
<whitequark>
use CT to avoid triggering the defenses
<whitequark>
the dose with CT is very low
<azonenberg>
awygle: randomized assembly?
<whitequark>
lol
<azonenberg>
more realistically, random PAR
<awygle>
whitequark: won't the thin shield actually work better on x ray than lower frequency?
<azonenberg>
each bitstream is different
<azonenberg>
different fpga resources used for each
* sorear
just tried to look up the width of the LHC beam at an interaction virtex
<whitequark>
every new defense you introduce means that your devices get less reliable
<azonenberg>
whitequark: actually that was a fun experiment i awnted to do
<awygle>
random PAR is interesting
<whitequark>
at some point they become useless for their primary goal, assuming that is not "frustrate red team"
<azonenberg>
mail xray dosimeters around to various people
<azonenberg>
in various countries
<azonenberg>
and states
<awygle>
Yeah this is obviously just a goof
<azonenberg>
Try to map out what packages get xrayed
<azonenberg>
conjecture: most if not all cross-border parcels are xrayed, probably not letters
<azonenberg>
but most domestic mail is not
<azonenberg>
they may spot-check things that seem fishy
<awygle>
azonenberg should make a honey pot hsm
<awygle>
Whose primary goal is "convince red team it stores useful data, and frustrate them"
<azonenberg>
lool
<awygle>
Post it note saying "DEFINITELY NOT MY PRIVATE KEYS"
<awygle>
Speaking of CT, I have to remember to get the scan of my foot...
<whitequark>
the slowest broadband uplink I can get is 300 Mbps and I don't need that much
<rqou>
O_o
<rqou>
HK or RU?
<whitequark>
HK
<whitequark>
but RU doesn't trail far behind
<rqou>
wow, HK's upgraded
<azonenberg>
lol wut
<azonenberg>
meanwhile, i'm on 50x10
<azonenberg>
i can get all the way up to gigabit down at my new house
<azonenberg>
but the upstream is stupidly slow
<whitequark>
azonenberg: in HK I can get up to 10G
<azonenberg>
i think it's gig down x 50 up?
<rqou>
apparently all countries except the US actually care about infrastructure
<azonenberg>
Or something equally ludicrous
<azonenberg>
and this is a business connection
<whitequark>
I have no fucking idea what can people do with 10G at home
<rqou>
whitequark: are you still in the middle of nowhere near HKUST?
<azonenberg>
getting upstreasm bandwidth is HARD
<whitequark>
rqou: yes
<rqou>
O_o O_o
<rqou>
HK definitely upgraded
<whitequark>
netvigator finally rolled out FTTH on not-island
<azonenberg>
whitequark: 10G to the internet?
<whitequark>
azonenberg: yes
<rqou>
yes
<rqou>
hk actually is fast
<azonenberg>
i have no idea what i'd do with that because 99% of servers cant keep up
<azonenberg>
even with gig
<rqou>
cn is somehow really limited with international b/w
<rqou>
even if you're not using vpn
<azonenberg>
rqou: the GFW can't keep up? :p
<whitequark>
azonenberg: it says 10G "On a shared basis with other PCCW Global customers" because they only have 4T of of backbone bandwidth
<azonenberg>
whitequark: lool
<rqou>
anyways, sonic has been really really aggressively advertising that they are rolling out symmetric gigabit here
<azonenberg>
realistically, i dont know what i'd do with gig to the internet
<rqou>
but they don't actually have it yet and won't have it until i move out
<azonenberg>
40G from my desktop to the lab? i'd use that
<rqou>
maybe it'd be done faster if they wasted less money flyering everywhere
<rqou>
:P
<whitequark>
I'm not sure what I'd do with 300M
<azonenberg>
anything past ~100 symmetric is a waste to me right now
<azonenberg>
simply because the other end is my limiting factor
<whitequark>
like I can download anime in BDRemux 1080p
<whitequark>
but the seeders are usually slower than that
<azonenberg>
exactly
<azonenberg>
the other end of the link is the bottleneck
<azonenberg>
so there's no point
<whitequark>
no, it's worse
<rqou>
i've managed to saturate a gig on the UCB network
<whitequark>
my *hard drive* becomes the bottleneck
<azonenberg>
rqou: i've saturated gig too
<azonenberg>
i downloaded a debian iso in 7 seconds once
<whitequark>
I actually managed to saturate the 6TB WD drives with mostly-sequential writes
<azonenberg>
whitequark: lol
<azonenberg>
my plan, now that i have a box with a 10g nic in it
<azonenberg>
is to set up a nas with a 10g or 40g nic on it
<azonenberg>
And move as much stuff as i can to it
<azonenberg>
because it will be faster than local storage :p
<whitequark>
azonenberg: oh you're right the upstream on 10G plans is "only" 2.5G
<rqou>
whitequark: what about paying for anime? :P :P
<whitequark>
which is still an absurd amount of bandwidth
<azonenberg>
whitequark: yeah my interest in bandwidth is mostly upstream b/c i do offsite backups
<whitequark>
rqou: paying whom
<azonenberg>
that, and updating vivado, are the main things i need a lot for
<rqou>
the usual answer here is crunchyroll, but they're probably us-only as usual
<rqou>
so i guess nobody :P
<whitequark>
crunchyroll "isn't available in your region", and besides, 80% of my interest is in obscure shit crunchyroll doesn't even have
<whitequark>
crunchyroll is great if you're interested in the latest slice
<azonenberg>
ally crunches rolls all the time
<azonenberg>
i think she gets some shows elsewhere
<whitequark>
(which will get taken down from it in two years so you still have to pirate to have a backup)
<whitequark>
there's a bunch of series i like that are so obscure that they don't even appear to have english subtitles anymore (as in they fell off the earth)
<sorear>
what's the state of the art in mobile links these days
<whitequark>
fortunately, rutracker.org has really good archivists
<whitequark>
sorear: define mobile
<whitequark>
there's 802.11ac which this laptop has
<rqou>
btw the joke is that cruncyroll was built by all the weebs in berkeley HKN :P
<whitequark>
a two-antenna STA on 802.11ac can get 1.69G
<rqou>
i've never gotten anything even close
<sorear>
whitequark: able to connect over a large fraction of HK land area without "excessive" reconfiguration (interpret as you will)
<whitequark>
... at 256QAM, which you'll never get with the existing state of 2.4G spectrum
<whitequark>
sorear: well there's LTE
<sorear>
s/land area/urban land area/
<whitequark>
actually, there USED to be LTE, it ... stopped working sometime between when I last came and this time
<whitequark>
now I only get HSPA+
<whitequark>
anyway the most I ever got with LTE was around 12 Mbps
<whitequark>
and some pitiful upload, maybe around 1-2 Mbps
<rqou>
i've never gotten more that a few hundred mbps even with wifi on 5G
<rqou>
also, i was "that asshole" running HT40 on 2.4G (no, it doesn't help much)
<whitequark>
oh sorry, 802.11ac is 5G-only
<rqou>
oh right, VHT80
<rqou>
HT40 = n
<whitequark>
and it has channels of up to 160 MHz
<rqou>
yeah, i don't think any shipping silicon can do that?
<sorear>
still a very large gap between 10G ftth and .012G lte
<whitequark>
sorear: well... duh?
<whitequark>
it's fiber.
<whitequark>
you can't argue with fiber.
<whitequark>
i figure they just rolled out the most recent CPE they had and hope this will suffice for the next 20 years.
<whitequark>
not a bad strategy.
<rqou>
is it actual FTTH?
<rqou>
not the stupid at&t fiber to the curb with shitty vdsl?
<whitequark>
given that previously I had shitty ADSL that barely got me 8M down, I imagine they previously employed the same strategy
<whitequark>
the website says FTTH
<whitequark>
I'm only registering for it now
<rqou>
interesting
<whitequark>
but their coverage map says I'm covered
<rqou>
btw, fun observation from the mainland: all the ftth i saw in the mainland was blue UPC connectors, not green APC
<rqou>
i guess they don't care that much about the reflections?
<whitequark>
azonenberg: oh have i mentioned the costs
<whitequark>
their 2x1000M plan is $55/mo
<rqou>
US sucks at infra, news at 11
<sorear>
can't argue with fiber but a utility-scale ac or ad -type system could be a thing
<whitequark>
sorear: not in high-density housing
<rqou>
it's called LTE :P
<sorear>
i doubt mobile terabit is physically possible but mobile gigabit should be
<whitequark>
it's definitely physically possible
<whitequark>
the question is in which conditions
<whitequark>
there's shit like 16384-QAM
<rqou>
also, troll idea: free space optical links
<whitequark>
if you put a BTS on an island, and a man with a phone on an island, I am sure it will work
<sorear>
I don't think fsoc counts as mobile in a wall-rich urban environment
<sorear>
60ghz barely does
<sorear>
given that wall penetration limits you to around 5ghz ~ 100mm and that mimo antennas much closer than λ are not useful, I do think we're practically limited to 10G or so for laptop-sized devices in existing built environments
<sorear>
maybe if walls start being built with fsoc repeaters in them that could change but that's a whole nother level of troll idea
<whitequark>
you could just have a switch per room
<sorear>
increasing QAM order scales data rate *logarithmically* so I consider it a non-answer akin to "analog computers are infinitely fast"
<rqou>
i'm expecting both the "wifi gives you cancer" and "free energy" conspiracy nuts to both cream themselves upon hearing that proposal :P :P
<rqou>
(the "repeaters in walls" idea)
<sorear>
there are probably considerations related to quantization of the EM field which intrinsically limit QAM order but I still do not actually understand QFT
* azonenberg
continues to dream of laptops with SFP+ cages on the
<azonenberg>
them*
<whitequark>
we don't even have ethernet on laptops anymore
<rqou>
yeah, way way too big
<azonenberg>
whitequark: who's "we"?
<rqou>
azonenberg: you're not thinking Different(TM) enough :P
<whitequark>
look at most modern laptops
* azonenberg
has a stinkpad
<azonenberg>
it still has vga
<azonenberg>
no rs232 but i think the docking station may have it
<whitequark>
azonenberg: i think the X series doesn't have ethernet anymore
<whitequark>
dual thunderbolt 3 though
<whitequark>
which is completely useless
<rqou>
yeah, someone please please RE the TB physical layer
<rqou>
physical/link
<sorear>
it's not documented? gross
<rqou>
nope
<rqou>
there are rumors it might be opened, but hasn't happened yet
<rqou>
remember to think Different(TM) :P
<sorear>
tb is intel, not apple
<sorear>
although they're the most famous user
<rqou>
wasn't it a collaboration?
<sorear>
looks like yes
futarisIRCcloud has quit [Quit: Connection closed for inactivity]
rohitksingh_wor1 has joined ##openfpga
rohitksingh_work has quit [Ping timeout: 256 seconds]
Dolu has joined ##openfpga
pie__ has joined ##openfpga
<rqou>
goddammit, even if i put firefox in a cgroup to make it not able to consume as much memory, it will eventually get to a point where the master process leaks so much that content processes oom instantly
<jn__>
browsers :/
<rqou>
pretty much
Dolu has quit [Ping timeout: 240 seconds]
rohitksingh_wor1 has quit [Read error: Connection reset by peer]
<pie_>
Now for some intelligent commentary on the situation at RPI. Please don't sue.
<pie_>
Get Your Apathy On
clifford has joined ##openfpga
soylentyellow has quit [Ping timeout: 246 seconds]
eduardo_ has joined ##openfpga
eduardo__ has quit [Ping timeout: 240 seconds]
<azonenberg>
rqou: i put firefox in a vm to cap it
<azonenberg>
it gets oomkilled a few times a week
<azonenberg>
but at least it doesnt drive my system into swap / use ram i could use for other things
<azonenberg>
In other news, comcast has a gigabit plan available at my new house
<azonenberg>
it's the most asymmetric connection i've seen in my life
<azonenberg>
Gig down, 35 M up
<azonenberg>
is that even enough for the tcp acks? :p
<balrog>
wow
<balrog>
oh yeah I went to verizon's website and it claims that the bill would be cut in half if we upgrade to gigabit
<balrog>
I wonder how true that is :D
<rqou>
azonenberg: why not a cgroup?
<azonenberg>
i dont even need a full symmetric connection, but can i maybe get less than 28:1?
<azonenberg>
rqou: because i also use the vm to provide some security isolation etc
<azonenberg>
not perfect but a lot better than userspace lockdown
<rqou>
waiting for a pwnZoidberg2own contest :P
<azonenberg>
i have one vm for irc, one for social media, one for email
<azonenberg>
one for general web browsing
<azonenberg>
one for online banking
<azonenberg>
etc
<azonenberg>
The goal isnt to be impenetrable, it's to be a harder target than the next guy
<azonenberg>
i dont think my online banking info is worth burning a browser 0day AND a vmware escape on
<rqou>
i wonder if e.g. governments hoard 0days for pwning "infosec"
<rqou>
or is pwning an administrative assistant still sufficient? :P
<azonenberg>
my conjecture for a long time was that pentesters are a high risk high return target
<azonenberg>
we're a lot more likely to throw your payload in IDA if we catch it
<azonenberg>
But imagine the value you get
<azonenberg>
a steady stream of -1 day bugs
<azonenberg>
that are about to get patched and you can use recklessly b/c their days are numbered anyway
<rqou>
hey azonenberg, very drama question:
<rqou>
are the guns you own AR-15 derivatives? how do you feel about it during the current discourse?
* qu1j0t3
isn't here for gun talk
<azonenberg>
rqou: anyone trying to ban any kind of rifle hasn't studied the data
<azonenberg>
they're used in somewhere around 2% of firearm-related crime
digshadow has quit [Ping timeout: 260 seconds]
<azonenberg>
I have zero patience for any politician introducing laws that they know full well will have minimal to zero impact on $ISSUE but make them look good to constituents who aren't super familiar with the issue
<azonenberg>
Regardless of what issue is involved
<azonenberg>
it's as absurd as a law banning the practice of getting an abortion while standing on your head and wearing neon green tights
<azonenberg>
no matter which side of the issue you're on, you should be able to see there's no point
<azonenberg>
And if there's one thing i can't stand it's hypocrites
genii has joined ##openfpga
<rqou>
but all the prominent cases involve rifles. gang/drug violence only affects poor people :P
<azonenberg>
Like i said hypocrisy
<azonenberg>
also, the logical fallacy (i forget the name) about rare but media-hyped events being more probable
<azonenberg>
Data-driven politics are extremely unlikely to ever take hold, though, because all of the lobbyists are terrified that the data might not support their side
<azonenberg>
Also because collecting data on many of the subjects is extremely difficult
<rqou>
thanks NRA
<azonenberg>
I'm talking in general here, not specifically regarding firearms
<azonenberg>
that's just one of the more prominent cases in which both sides are massively, willfully uninformed
* azonenberg
giggles at the though of a government that taxed energy based on the calculated externalities of the generation method
user10032 has joined ##openfpga
<azonenberg>
and/or went all-in on nuclear and banned construction of new fossil fuel plants on public safety grounds
<balrog>
azonenberg: require gun registration and liability insurance :P
<balrog>
both are required to own a 3500lb metal machine that is also pretty dangerous when misused
<azonenberg>
another case in which people are terrified of the extremely improbable scenario of a meltodwn
<azonenberg>
But the data doesnt support the probability of this happening being significant compared to the reduction in risk of cancer etc from fossil fuel air pollusion over decades of operation
<balrog>
azonenberg: yup...
<azonenberg>
But of course politicians can't sell people on "oh look, over the course of this coal plant's operation 100K people will die from pollution-related illnesses"
<azonenberg>
They're afraid of 50 people getting cancer from a nuke accident that might not even happen
<balrog>
because it doesn't hit as hard and quick
<azonenberg>
Because cumulative air pollution isn't sexy and doesnt make national news
<balrog>
no one notices being slowly poisoned over time
<azonenberg>
Can you imagine if media attention on various dangers was proportional to the chance of it happening?
<balrog>
as opposed to a large explosion that makes people sick in a short time period
<azonenberg>
front page would be full of MVAs all the time
<azonenberg>
people would be demanding safer cars by day three
<balrog>
yep, but those are seen as "normal"
<azonenberg>
it doesnt matter
<azonenberg>
imagine if we did what happens when somebody dies in a plane crash
<azonenberg>
have front-page gory photos of the smashed up car even if the accident was three states away
<azonenberg>
loop it on prime-time news
<azonenberg>
profile the history of the victims and all the family they left behind
<azonenberg>
now repeat for heart attacks
<pie_>
"xyz has had a preventable heart attack ..."
<azonenberg>
aside from being massively depressing, you know what that would do to people's gut impression of risk? lol
<azonenberg>
nobody would worry about plane crashes anymore
<azonenberg>
i remember seeing a fun study saying that post-9/11 the national death rate went slightly up
<azonenberg>
because more people were driving instead of flying places
<pie_>
on that note i need to start eating better ...
<azonenberg>
And crashing more frequently as a result
<pie_>
huh.
<azonenberg>
anyway, circling back to the original topic
<azonenberg>
if people actually had an accurate impression of risk, people wouldn't be campaigning to restrict rifles
<azonenberg>
They'd be focusing on suicide prevention
<azonenberg>
because for every person who is shot by a third party for any reason, at least two (i forget the exact ratio) pull the trigger themself
<azonenberg>
And that doesn't even count those who use other means
<azonenberg>
But five people intentionally ODing in separate incidents won't make headlines the way one guy shooting three people will
digshadow has joined ##openfpga
<azonenberg>
On topic... Ethernet bit ordering is annoying
<azonenberg>
i kinda get why they do it, but seriously
<azonenberg>
having the leftmost byte in a word be numbered 0?
<azonenberg>
But the rightmost bit in a byte?
<azonenberg>
at least it's not 1-based numbering like SONET
<jn__>
eww, IBM bit order
<jn__>
wait no, i misread
<awygle>
CCSDS has the worst bit order explanations
<rqou>
my father hit a case where they were layering <"internet" protocol A> over <"telco" protocol B> and the bit order conventions were the opposite
<rqou>
they did it right, but <big vendor> that shipped first got the ordering backwards
<rqou>
"In 2009, the NMPA and RIAA reached a settlement over mechanical licenses that record companies had not paid to songwriters — likely for the same bureaucratic reasons that Spotify is currently struggling with. The irony there is that the three biggest members of both the NMPA and the RIAA are the same companies: Sony, Warner, and Universal. The fight over mechanical licensing is the left hand and the right hand
<rqou>
amazing
clifford has joined ##openfpga
<fouric>
Would this be an appropriate place to ask my question about HDL synthesizability?
<awygle>
fouric: depends on the question. probably ##fpga is more strictly on topic but nobody here will actually complain. shoot
<fouric>
My professor said that, in the general case, HDLs such as Verilog and VHDL are not sythesizable, and that you have to go read the documentation of individual vendors' toolchains in order to figure out exactly what you can synthesize.
<fouric>
...or, at least, that's what I *think* he said.
<fouric>
How true is this?
<fouric>
...and if it's at all true, *why*? Is it really that hard to define a simple language that can programmatically be transformed into simple digital logic?
<awygle>
Not sure about "read every vendor's tools" but yes, many constructs are not synthesizable
<awygle>
The reason is that both languages started as simulation languages
<mumptai>
and there are differences in the synthesizable subset between tools and versions
<awygle>
Then someone was like "oh hey if we limit ourselves somewhat we could actually build this!"
<awygle>
Verilog has a standardized synthesizable subset (which yes may be supported at different levels by different vendors)
<awygle>
My impression is that the variance is less now than it was when I was in school, at least
<fouric>
Hm, OK.
<fouric>
That makes sense.
<fouric>
...and I can't easily think of a better way of doing it.