<roh>
setting up dma is quite many accesses usually. it only gains if you can do cpu work which does neither need ram or bus. (because dma is blocking that)
<mth>
if you're doing video, accelerated YUV to RGB conversion would be useful though
<whitequark>
mth: the ipu is on my to-do list too. not sure if it's same in jz4725 and jz4740
<roh>
thats done in the userspace component usually (if the hw supports it)
<whitequark>
i think that existing implementations add some ioctl to framebuffer driver
<whitequark>
*ioctls
<mth>
whitequark: please keep us informed; we'd like to have IPU support in OpenDingux as well (Dingoo A320 is JZ4740 based)
<mth>
Ingenic does that, but they always invent custom controls
<mth>
there are more standard ways to do it
<mth>
maybe via v4l2?
<whitequark>
mth: my processor is jz4725 (jz4750 core), some stuff is compatible with jz4760, but lars may have more to say about latter
<roh>
whitequark: ioctls to the fbdev are a bad idea
<roh>
a) nonstandard
<roh>
b) slow.
<roh>
you have a context change per call. dont do it.
<mth>
whitequark: afaik 4725 is like 4740 and 4755 is like 4760
<mth>
but there is more difference between 4755 and 4760
<whitequark>
tcu is almost completely different, and there are more differences
<whitequark>
ECC, NAND, and CODEC are absolutely new
<mth>
but isn't the nanonote using the 4725?
<roh>
whitequark: what would be cool is ingenic gpu support for blitting and video in directfb.
<roh>
it should be doable very similar to pre-agp cards. the mga driver should be a good example for where to do what.
<whitequark>
mth: looking at the site, it is 4720, and I've once talked with lars about that. He says that jz4720 is stripped-down version of 4740 (or maybe vice-versa, I don't quite remember, but they are similar that way)
<whitequark>
anyway the branch he is working on is jz4740
<whitequark>
roh: do you mean the integrated one, the ipu? I don't see how it may accelerate blitting
<whitequark>
and I doubt that old pre-agp cards have accelerated video (but i may be wrong)
<roh>
sure they do
<roh>
back that that was all mmio to set uo
<roh>
now these units are gone and you need to do all via commandchannel
<roh>
basically assemble a bitstream of commands and data and burst that over to the gpu via agp
<roh>
a radeon>9000
<mth>
the 4740 IPU can only convert YUV to RGB
<mth>
the 4755 can convert RGB to RGB as well
<whitequark>
mth: i've checked the manuals, and sadly, 4740 ipu is much much simpler that 4725 one
<roh>
a radeon later than the 9000 series has no video unit anymore for example. you need to use a shader for that now etc.. so its 'easy' compared to desktops on embedded ;)
<whitequark>
you won't be able to use my driver, if i ever write one
<whitequark>
on the other side, writing a driver for that ipu is much simpler
<mth>
4725 programming manual does state IPU input format as YUV/YCbCr though
<whitequark>
it can do rgb too
<mth>
are you sure it is 4725 and not 4755?
<whitequark>
well, i can send you a photo of the chip :)
<whitequark>
that may sound strange, but i actually know the name of soc i'm developing drivers for
<mth>
or do they have two different chips both labeled 4725?
<mth>
it seems to be a substantial upgrade from the original 4725
<whitequark>
probably. it's actually a stripped-down version of 4750 (no new or changed blocks, but removed usb host and much more; also there is a lot of leftover 4750 references all over the manual)
<mth>
there once was a 4740-based 4725, but all references to it seem to have disappeared from their FTP
<whitequark>
roh: vala looks like a good live project; maybe for my task lite will be sufficient too. so I'll stick with directfb and gtk+vala or lite
<whitequark>
mth: maybe they have decided to let it fall into oblivion
<whitequark>
roh: can you tell a bit more about the right way of doing accelerated video through ipu?
<whitequark>
are you suggesting to mmap ipu registers and then call them directly from mplayer? I don't think that 30 more context switches per second will matter a lot
<whitequark>
and they're somewhat cheap in mips, if i understand it correctly
<roh>
whitequark: nope. mmap the ipu register from directfb and leth that abstract that for mplayer
<roh>
also one can compile directfb for multiple app use. so more than one app can run and directfb takes care of 'muxing'
<roh>
in the end 'the first process to use dfb' gets the 'driver' (usually something like a menu or 'minimal window manager' and all others communicate with it via shm/fusion.
<roh>
fusion is the 'ipc-layer' dfb uses for multi-app stuff.
<roh>
i think fusion needs a kernel driver. not sure if that is still true
<whitequark>
afaik it is
<whitequark>
that looks like a right way
<whitequark>
could I add support for OSD with it?
<roh>
sure. usually there are either layers for yuf rendering.. if not you basically do a blit with colorkeying
<roh>
or alpha... depending of what colorformats are possible with the hw
<aw_>
where you can drop/search  U9 JS28F256J3F105
<xiangfu>
aw_: thanks.
<wpwrak>
(accelerated video) that's been around since the days of the ISA bus ! and back then you needed it the most ;-) just compare an ET4000 with an S3 801 :)
<wpwrak>
kristianpaul: the "volatile" means that something else but the program may change the value. thus, the compiler may not cache the value, eliminate redundant access, or reorder accesses.
<wpwrak>
kristianpaul: otherwise, things like *(uint8_t *) uart = 'h'; *(uint8_t *) uart = 'i'; could just become *(uint8_t *) uart = 'i';
<wpwrak>
(4720) the 4720 has the same die as the 4740, but does not connect all the signals
<wpwrak>
steve|m: i also have a mini-usb connector in ben-wpan: components/mini_usb_b.lib and modules/mini-usb.fpd
<wpwrak>
steve|m: (copper layer tested in prototype. the other layers unknown so far)
<wpwrak>
aw_: (the first has "cowboy legs" while the second has straight legs)
<aw_>
wpwrak, i'd prefer the first one with "cowboy legs" which can be mounted strongly even u/p is much than second one.
<wpwrak>
aw_: ah, good :) that happens to be the one i've specified so far. we've been worried about the difficulty of smt'ing it
<wpwrak>
aw: (btw, the pcb is 0.8 mm, which is less than what they recommend for the connector. but unfortunately, there don't seem to be any USB A plugs that are designed for thin boards. we'll just have to live with this.)
<aw_>
wpwrak, i see, with thickness 0.8mm, yup..no more parts manufacturer will produce for you to meet a short "cowboy legs" except you are A**LE. :-)
<wpwrak>
aw_: yeah, we still have to work on those *huge* production runs ;-)
<valhalla>
is there some promotional material for the nanonote / free hardware ready to be printed? I will help man a boot on linux at a local electronics fair and of course I will bring my NN
<wolfspraul>
it's a bit messy, but thanks a lot for your initiative!
<valhalla>
wolfspraul: thanks. I don't have much hope for that fair (it used to cather to radio enthusiasts, but it has gone a bit too mainstream), but it would be a shame not to bring anything about free hardware
<wpwrak>
valhalla: maybe mention ben-wpan ?
<wpwrak>
valhalla: also people who know their way around microwave circuits and could contribute towards improving things would be hotly welcome :)
<lars_>
i know my way around the mircowave oven, is that enough?
<qi-bot>
[commit] Werner Almesberger: install/: renamed READMEs and added (non-working) setup instructions http://qi-hw.com/p/ben-wpan/0cfd277
<wpwrak>
(the second shows a slightly older board)
<wpwrak>
the technology is IEEE 802.15.4 (the thing underneath ZigBee and 6LoWPAN), at 2.4 GHz
<valhalla>
so the project is a free hardware board for IEEE 802.15.4, with usb and/or sdio connection?
<wpwrak>
with USB or 8:10 card. it's not an SDIO device, although the physical interface is the same. i.e., you have to be able to bit-bang the signals (which is the case in the ben)
<tuxbrain>
wpwrak: (izchat) that means you can send/recive something writing in one NN and reading on the other? can you make a video of this?
<qi-bot>
[commit] Ayla: Changed an error message, that was displaying "Error:" twice. Plus, it now more meaningful. http://qi-hw.com/p/gmenu2x/ec000e0
<tuxbrain>
roh, wpwrak any more thoughs on atben atusb envelope? I like the idea on atusb of an already manufactured plastic box, but we must be sure before production that it will fit to modify pcb if needed. If we start now we can have the first batch ready by the mid/end May.
<tuxbrain>
regarding atben, the sucru aproach is taking strength but victor, more skilled on plastics and such will take a look and alternatives and costs
<roh>
tuxbrain: didnt have time.. too many things right now.. i think it would make more sense taking a atben and walking over akihaba markets in tokyo
<roh>
but they have atomic end-time warfare there so i do not dare to bother people over there
<wpwrak>
tuxbrain: (izchat) it's a simple two-way communication program that sends its messages in IEEE 802.15.4 MAC frames. (without IP or such)
<wpwrak>
tuxbrain: before that, i also had communication between bens, but i used "empty" frames for communication, i.e., without IEEE 802.15.4 MAC header or addresses
<wpwrak>
tuxbrain: (video) it's not very photogenic yet - one of the two Bens is a naked board without display
<tuxbrain>
wpwrak: (naked ben) mmm thechi pr0n!
<wpwrak>
roh: (japan) if you know someone there, they may appreciate a bit of normality
<wpwrak>
tuxbrain: (geekporn) yeah, raw and naked :)
<tuxbrain>
wpwrak: (usb connector) finally they will use the usb connector you specified, no problems on smt process with the cowboy legs, confirmed by smt vendor.
<jlamothe>
Does anyone know if there's a way of changing the addribute of a pin in a gEDA footprint file so that it has no pad on it (for mounting holes)?  I haven't really been able to find a straight-forward answer from Google.
<wpwrak>
tuxbrain: whee ! :)
<tuxbrain>
wpwrak: Tuxbrain will go ahead with production I'm writting a mail with my final points and "organizing" things a bit, now I must leave so I will end it at night, but basically is a go go go , so are final gerbers ready? I so give me the links to start PCB part.
<wpwrak>
tuxbrain: let make generate a final package. i'll add the USB hole to the README, bump the version numbers, and check if i should enlarge one mounting hole a little (that hole gave me some troubles when mounting the usb connector)
<wpwrak>
tuxbrain: (hole) whether i really need to enlarge it depends on the actual size my mill produces
<tuxbrain>
wpwrak: ok, waiting for you command sir!
<wpwrak>
tuxbrain: (may) so 2 weeks for the pcbs and 4 weeks for smt ?
<tuxbrain>
really smt should be 2/3 weeks but we have Eastern holidays in the middle :(
<wpwrak>
bargh. evil holidays
<tuxbrain>
we (smt vendor and I) will try to minimize this by making a firm order to start sourcing and such but you know ... here in spain "Semana Santa" is sacred ... damn religions (sigh)
<wpwrak>
sourcing should take about 48-72 hours if you order from digi-key :)
<tuxbrain>
wpwrak: spain... remember this is spain.... damn germanic :P
<wpwrak>
tuxbrain: you can also order at digi-key from spain ;-) in fact, it's a good idea to order as early as possible. because if anything goes wrong, then you have a) time to look for alternative sources, or b) (worst case - if you need to pick a not 100% compatible replacement) change the design
<wpwrak>
tuxbrain: luckily, in this case, b) is unlikely
<tuxbrain>
ok but as I point In my mail not sended yet, I preffer than they do the source, if there is any mistake I can reclaim, I fail on anything like packaging or whatever I will end with a lot of components and a delay...
<tuxbrain>
but yes once I have the pcb started I will tell them to start sourcing
<wpwrak>
tuxbrain: (they do the sourcing) yes, i understand. but it also makes sense if they start early. that way, there's less risk. particularly with the unstable situation in japan.
<tuxbrain>
wpwrak: agree
<tuxbrain>
I must leave
<tuxbrain>
read you later
<wpwrak>
tuxbrain: cya !
<kristianpaul>
is waiting for the sacred "semana santa" too
<wpwrak>
tuxbrain: ah, and i also need to know whether the hole for the USB connector should be part of the gerbers or if it's enough if i specify it in the README
<wpwrak>
kristianpaul: looking forward to it or disliking it ?
<kristianpaul>
wpwrak: i dont work on semana santa :-)
<wpwrak>
kristianpaul: i hate all those holidays that mess up the weekly routing. of course, i take my days off whenever i please :)
<kristianpaul>
i need more time for qi ;)
<wpwrak>
s/routing/routine/
<kristianpaul>
ah, you tell me, my town is flood of tourist the whole week
<kristianpaul>
so, another excuse to stay at home ;-)
<wpwrak>
kristianpaul: tourists coming to worship the place where the famed gps developer lives ? :)
<kristianpaul>
they came because the "Basilica del Se~os de los Milagros"
<kristianpaul>
Se~nor*
<kristianpaul>
will consider for the next year go back to the farm
<kristianpaul>
as soon as i get internet ther ;)
<wpwrak>
kristianpaul: full retreat ;-)
<kristianpaul>
wpwrak: (volatile) thats tricky, it seems to be used to do MMIO in  serveral rtems drivers for milkymist
<kristianpaul>
well seems i neeed read more about it, before get the DMA thing for SiGE to work in software..
<wpwrak>
kristianpaul: (volatile) yes, forgetting to use volatile can produce bugs that are very entertaining to chase
<whitequark>
can't beliveve my eyes. so are you just planning to make a batch of some device you've invented and tested?
<whitequark>
is it really that easy over there?
<wpwrak>
whitequark: what further complications would you expect ? ;-)
<whitequark>
wpwrak: well, I don't have enough spare money to realy try to manufacture one of mine useless devices, but I always thought that's a process full of its very own crap and numerous fuckups
<whitequark>
*really
<wpwrak>
whitequark: well, in a way it is. but there's no way to learn other than taking the risk and trying it :)
<wpwrak>
whitequark: and the PCB and SMT companies also advise us about potential problems. so it's not too bad. (i hope :)
<whitequark>
pcb production isn't that hard even in russia
<whitequark>
you should have chosen black solder mask for that board. it'd look much cooler with it ;)
<wpwrak>
whitequark: yeah, probably. didn't want to make things complicated, though. step by step :)
<whitequark>
(besides that... 5 euros for a small pcb. if I'd make that at home with photoresist, it'd cost maybe... 0.5$?)
<wpwrak>
whitequark: oh, that cost is almost entirely logistics. in terms of cost vs. overhead, it's quite stupid to order a single board. yet a lot of people seem to want that.
<wpwrak>
whitequark: even for ten pieces it's silly. prices get better for 50 and more units.
<wpwrak>
whitequark: also, for a fair comparison with DIY, you should consider the value of the work, not just the materials. that usually changes the picture quite dramatically :)
<whitequark>
wpwrak: diy'ing that takes from half to a hour for me, including exposure and etching
<wpwrak>
whitequark: now tell me your skilled work is worth less than 4 EUR/hour :)
<wpwrak>
whitequark: also, do you have a CNC mill to do the cutting ? this board is a little tricky
<whitequark>
wpwrak: oh yes, right. I have some kind of chinese multi-machine, it'd work for a board or two, and I'm looking at building a simple cnc myself: it's not as hard as it looks
<whitequark>
but this particluar board can be cut with scissors, dremel and a steady hand
<wpwrak>
whitequark: yup, it can be cut manually. it has a predecessor that was actually done that way. just a bit more work.
<whitequark>
the sad thing for diy'ing are vias. I've never managed to do them correctly. not only all my drills are too big, but the small ones break after just several holes, and, which is the biggest problem, they can't be fixated in the collet of my dremel
<whitequark>
I hate them.
<wpwrak>
whitequark: (ubb) of course, if you need just one piece and you're already set up to fabricate such things, then it doesn't make sense to buy that one. but there's a surprisingly large number of hobbyists who consider the whole pcb-making process as the purest evil magic. they're happy if they don't have to do that :)
<wpwrak>
whitequark: (vias) oh yes, i hate them too
<whitequark>
that's the reason I've not yet done my most loved idea, a diyed (open hardware probably) music player with graphical display
<wpwrak>
whitequark: (and there are three more of each kind i had made, plus various other prototypes with slightly different designs)
<whitequark>
anything I found was either ugly or using nano-sized smt components (which are generally good) and the damned vias
<whitequark>
aha, that pattern of non-uniform solder distribution is very familiar for me :)
<whitequark>
that's amazing btw
<whitequark>
do they work?
<wpwrak>
whitequark: (vias) my approach for them is now to drill them with my CNC machine. no more broken drills, all holes are vertical, etc. then i stick a piece of wire through them, and solder.
<whitequark>
which method have you used? the laser-printer-and-iron one or with the photoresist?
<wpwrak>
whitequark: i used toner transfer
<wpwrak>
(nano-sized components) you mean BGAs and other chips with contacts only at the bottom but nothing on the sides ?
<whitequark>
I've never managed to make a board with toner transfer which has all its connections non-broken
<whitequark>
per my experience, toner transfer may (sometimes) be easier, but it produces unstable results
<whitequark>
contrary to that, photoresist is stable: if you've got a crappy board once, you'll get the exactly same crap each time you'll try with same conditions
<wpwrak>
(toner transfer) my boards usually need a little bit of correcting. but it's not too bad. and i'm not sure how well photo-coated boards would like the CNC step anyway.
<wpwrak>
yes, toner transfer is a bit less dependable. i use a laminator for the actual transfer process, which helps.
<whitequark>
while photoresist is not exposured, it's something like a, hmm, maybe glue on stickers from market. anyway, you can freely drill through it, and it would be good
<whitequark>
I always do drilling after etching, through
<whitequark>
after the exposure it becomes fragile, so it would not be possible
<whitequark>
also, to stick the resist to the board properly, laminator is required too
<wpwrak>
(like glue) yeah, that wouldn't go well with CNC. lots of dust.
<whitequark>
it's coated with a PET film
<whitequark>
if you're interested, I may make a test once I'll be at home
<wpwrak>
so you do the coating yourself ? okay, then you could cut them first and then coat.
<whitequark>
sure
<whitequark>
hmm, I see you've used atrf series
<whitequark>
they are somehow absent in russia
<whitequark>
I've found some TI's CC2500 instead, they look very promising, and also it's TI
<whitequark>
how did you calculated the antenna?
<wpwrak>
don't ti have export restrictions on their "modern" chips (with AES) ?
<whitequark>
cc2500 don't have aes
<whitequark>
at most, they have CRC and some kind of FEC
<whitequark>
I really like the simplicity of the integrated 2.4g transciever which only needs two or three caps
<whitequark>
hm, so the board thickness matters a lot
<wpwrak>
microwave is picky about the board thickness
<whitequark>
well, you have some nice equipment (what's that at the bottom of photo?)
<whitequark>
I wonder how expensive it is
<wpwrak>
ahm the cc2500 isn't ieee 802.15.4, is it ?
<wpwrak>
(bottom) that's an usrp2 with a xcvr2450 board. about USD 2000 before customs fees and taxes. that's actually quite cheap for a device with these capabilities.
<wpwrak>
they'll release a new daughterboard that's even better next month. the xcvr2450 does the 2.4 and 5 GHz ISM bands, but only half-duplex. the new board will do 400 MHz-4 GHz, full-duplex.
<whitequark>
oh, the usrp
<whitequark>
but, 2000usd.
<whitequark>
(If I only had lived in the same country with someone who has such devices...)
<wpwrak>
"professional" vector signal analyzers/generators cost something like 10-100 times that price. it's really a bargain if you think of it :)
<whitequark>
I prefer do not think about it at all
<wpwrak>
;-))
<whitequark>
AA2500 is a simple radio-uart with a bunch of interesting modulation and packet transmission modes, and other features like fec and crc
<whitequark>
it's quite powerful as-is, and has a nice frontend, which is simple too
<whitequark>
some russian developers had a (somewhat) successful use stories, so looks like I may get something workable
<whitequark>
wpwrak, can I ask you when my boards won't work again? :)
<wpwrak>
(cc2500) i'm not so fond of non-standard modulation techniques. with ieee 802.15.4, we have at least something reasonably well defined and interoperable. otherwise, it's just the same old vendor lock-in all over again.
<wpwrak>
whitequark: hehe, i'm still an RF noob, but i can at least contribute some war stories ;)
<whitequark>
wpwrak, not just about RF. I have troubles in (all) other areas as well. e.g. switch-mode supplies and, to my shame, even cp2102 hasn't worked
<whitequark>
well, about the latter, I think that's because of the flux I mistakenly bought thinking it may be something good
<whitequark>
turns out that it is actually not so bad flux, but unless you're heating your entire board thoroughly, it would remain in active, and conductive (~tens of ohms) state
<whitequark>
I've thought I've killed one of my ben-like boards, but once an idea came and (after a three weeks spent on a shelf with very active flux) I've heated the board for ~5min, and it worked perfectly
<wpwrak>
(cp2102) aww. well, i had my share of fun with their competitor, ftdi. but i think all the flaws i found were actually part of the design ...
<whitequark>
oh, ftdi
<whitequark>
ft232 is 6 times more expensive than cp2102
<whitequark>
($10 for a chip in most cheap place in Moscow)
<wpwrak>
(flux) tens of ohms ? wow !! even tens of kOhms flux is a royal pain
<whitequark>
heh, that's interesting flux. after it was heated for about 200 deg. C for a minute or two, it deactivates completely and everything works
<whitequark>
and it is really really good as a flux
<whitequark>
but every board has some small hidden place which does not get heated enough
<whitequark>
and then fun begins.
<wpwrak>
(ftdi) naw, the FT232RQ isn't so expensive
<whitequark>
IIRC I've used FT232RL
<whitequark>
the one in case with pins on two sides
<whitequark>
like soic, but more dense
<wpwrak>
something like 25% more expensive than the silabs part. and theoretically more capable. alas, all the extra capabilities are unreliable/buggy
<whitequark>
heh
<whitequark>
cp2102 costs like $1.5 here
<whitequark>
so, even if I kill half of chips I buy, it's still cheaper than using fts
<whitequark>
(btw, the single board I've did with ft worked until I've managed to overload it somehow, and it died)
<whitequark>
I've double-checked the price. currently at terraelectronica (they're technically a wholesale shop, but you may very well buy one or two parts, it's perfectly good for them) it costs 48 RUR
<whitequark>
that's $1.7 according to google
<whitequark>
I mean, 48RUR for one part. if I'd buy more than 10, they would be even cheaper
<whitequark>
I can send you a bag of them, heh
<wpwrak>
hmm. they must get rather interesting discounts. or maybe someone diverted a container full of chips ;-)
<whitequark>
they have absolutely best prices in moscow, maybe in the whole russia
<wpwrak>
pretty impressive.
<whitequark>
and a rather wide range of different elements. not only semiconductors, they also have e.g. inductors too
<whitequark>
yeah
<kyak>
kristianpaul: i feel your breath :)
<whitequark>
anyway, thanks for the info about ftdi crappiness. now i don't think they're better than silabs chip at all
<kristianpaul>
ha
<kristianpaul>
oops
<whitequark>
has someone used laser cutting services? they look like a right way to prototype a good case
<wpwrak>
whitequark: the problems with ftdi are: lousy documentation (they like to keep things secret), ergo buggy linux drivers, and flaws in things like synchronous bit-banging
<wpwrak>
whitequark: (ftdi) as long as you just need a basic UART, you're probably fine. but try to use the A-bus for something more demanding, and you're screwed. likewise, you can't use the C-bus reliably due to incomplete documentation.
<whitequark>
hmmm... A-bus? C-bus? I've never heard of that
<wpwrak>
whitequark: (ft) if you try to flash the eeprom, you may also end up with garbage due to buggy linux drivers because of missing documentation
<wpwrak>
whitequark: (ft) of course, if you use ftdi's closed source drivers, things may work better
<wpwrak>
whitequark: (a-bus) a-bus is the first uart. b-bus is the 2nd uart in the ft2232 devices. c-bus are a few pins available for status signals/leds or bit-banging.
<wpwrak>
whitequark: (laser cutting) roh is our expert for this
<whitequark>
aha, thanks
<wpwrak>
whitequark: i think he sleeps next a laser cutter :)
<wpwrak>
s/next/next to/
<whitequark>
looking at all that boards, I think I now have a reason to buy ben.
<whitequark>
what variant works better, atusb one or sd-bitbanged one?
<whitequark>
are they rf-compatible?
<wpwrak>
they're compatible. the atben (bitbanged) goes into the ben. atusb goes into your pc.
<wpwrak>
20110306 were unsuccessful attempts to make atusb less noisy
<wpwrak>
(txrx) the at86rf231
<whitequark>
kyak: moscow
<kyak>
whitequark: nice :) we could've done a group buy couple of weeks ago!
<kyak>
(though the chances of receiving two Ben might be lower)
<whitequark>
kyak: i didn't knew about sharism and everything a couple of weeks ago. anyway, there sure will be more nice things to buy
<whitequark>
wpwrak: so, atmel parts all the way. I know they're good at uC's, are their RF parts equally good?
<wpwrak>
whitequark: (rf parts) so far, they're behaving :)
<wpwrak>
whitequark: what's not so nice about them is that you only have a single buffer for incoming and outgoing frames. so you can't receive while transferring a frame from/to the buffer. a bit like the good old 3c501 ethernet card.
<wpwrak>
(transferring) however, if you want to squeeze out the last few milliseconds, you can start retrieving an incoming frame while it's still being received.
<whitequark>
wpwrak: that cards falled out of use a couple of years later I've beginned to work with networks