<oliv3r>
as far as I can see, the function gets called in print_cpuinfo and as far as I can tell we don't have the BSS available then yet
apritzel has joined #linux-sunxi
<oliv3r>
though it supprises me a little as BSS is only for zero initialized statics i thought, so using non-zero should work, but the u-boot README claims differently
IgorPec10 has quit [Ping timeout: 250 seconds]
<wens>
where does u-boot README claim differently?
<oliv3r>
while it probably should read BSS and DATA segments are not available, if the claim of not able to use static/globals is true
<oliv3r>
as far as I know, BSS only stores zero initialized static/globals
Mr__Anderson has quit [Remote host closed the connection]
<oliv3r>
but i'm not super knowledgable abou tthese low details :)
xalius has quit [Ping timeout: 250 seconds]
<oliv3r>
the symptoms i'm personally seeing is, that when i print a static var in print_cpuinfo, it is quite ... random and not as expected. e.g. static int i = 0; printf(%d\n", i++);
<wens>
it says that globa data will be relocated after board_init_f
<wens>
heard gossip that marvell is exiting the cpu business
<KotCzarny>
wens: sad day if its true
Mr__Anderson has joined #linux-sunxi
ricardocrudo has joined #linux-sunxi
kasper__ has quit [Ping timeout: 258 seconds]
kasper has joined #linux-sunxi
mosterta has quit [Read error: No route to host]
premoboss has joined #linux-sunxi
mosterta has joined #linux-sunxi
caog has joined #linux-sunxi
<montjoie>
even with NAPI for TX and RX, the performance are not what I expected:(
<montjoie>
380/280Mbit/s
alain_ has joined #linux-sunxi
<alain_>
montjoie: how much can the hardware deliver with the BSP driver?
<alain_>
montjoie: at least good news is that I've been running your driver for 10+ days on my OPiPlus and that it is stable, no issues to report.
<montjoie>
alain_: with BSP KotCzarny give me results of 600Mbit/s
<montjoie>
it is why I am not happy
<alain_>
what is the BSP driver doing differently from yours to get that kind of performance?
<montjoie>
good question
<alain_>
Is it a case of not including the same level of safeguards and taking shortcuts?
<montjoie>
they do less memory barriers
<montjoie>
but I do not think that the problem is that
afaerber has joined #linux-sunxi
IgorPec has quit [Ping timeout: 264 seconds]
<montjoie>
KotCzarny: could you retry with only one CPU under BSP ?
popolon has joined #linux-sunxi
afaerber has quit [Ping timeout: 252 seconds]
afaerber has joined #linux-sunxi
<oliv3r>
wens: what i ment was, BSS is not available, so you cannot use global/static variables, only stack variables and global_data
<oliv3r>
which i interpreted as no global/static variables, which matches with the random values i'm seeing
<wens>
oliv3r: yup, looks like a bug indeed
<oliv3r>
wens: in the text or how it actually works? :)
<oliv3r>
wens: or my previous linked line with the return optimization if guard
<oliv3r>
i'm guessing the last
<oliv3r>
i'll add a patch to fix that to my patchqueue as it is related
alain_ has quit [Ping timeout: 250 seconds]
<oliv3r>
speaking of u-boot, https://github.com/linux-sunxi/u-boot-sunxi/blob/mirror/next/arch/arm/mach-sunxi/cpu_info.c#L110 what is the purpose of the (unsigned long) cast here? The commit says to be compatible with both arm32 and arm64; afaik ulong is always 32 bits, so uint32_t would be a more clear cast, or doese ulong change depending on arch and can it switch sizes, in that case I would imageine this to
<wens>
she was in the allwinner video (banana pi rep)
matthias_bgg has joined #linux-sunxi
<tkaiser>
KotCzarny: And at least they started recently to support community by donating some boards a few sun8i devs :)
<KotCzarny>
tkaiser, yeah, that's a plus
<KotCzarny>
still, better would be actually getting proper technical info for the devs
lemonzest has joined #linux-sunxi
<oliv3r>
KotCzarny: but they can't as they likley don't even have it themselves :)
<KotCzarny>
:)
<KotCzarny>
but they can ASK
premoboss has quit [Remote host closed the connection]
Gerwin_J has joined #linux-sunxi
alexxy has quit [Ping timeout: 264 seconds]
alexxy has joined #linux-sunxi
Andy-D has joined #linux-sunxi
<wens>
from what i heard, allwinner isn't very responsive to them either :(
<KotCzarny>
then we should stop bashing vendors and bash allwinner instead?
<wens>
depends on what we're bashing people over?
IgorPec10 has joined #linux-sunxi
* jelle
missed something
<KotCzarny>
wens, support, information
ht91109 has joined #linux-sunxi
<tkaiser>
KotCzarny: Allwinner most likely doesn't care about linux-sunxi or any Linux activity (or considers this even dangerous). And in the meantime I would believe Foxconn/SinoVoip do not get the idea what's wrong with selling paperweights. I never saw they would think from a customer's perspective and don't differentiate between 50k sold BPi M3 and 49K collecting dust since no working OS images are available.
kasper_ has joined #linux-sunxi
kasper has quit [Ping timeout: 260 seconds]
<KotCzarny>
then what needs to be said to that nora lee is that linux-sunxi is not android and is of no interest to them
IgorPec10 has quit [Ping timeout: 240 seconds]
<MoeIcenowy>
I viewed some comments on the fake Pine64 on Taobao
<MoeIcenowy>
lots of people use it just as a set-top box
<MoeIcenowy>
such a waste
<jelle>
:-(
<MoeIcenowy>
They're saying "Performing very well to connect to TV to watch videos"
premoboss has joined #linux-sunxi
premoboss has quit [Read error: Connection reset by peer]
premoboss has joined #linux-sunxi
<Wizzup>
Is it at all plausible to have a custom BROM? (I am thinking of some verified boot)
<Wizzup>
I realise it may require a hw mod
<Wizzup>
Wiki says it's integrated or non replaceable
Amit_T has quit [Ping timeout: 240 seconds]
The_Loko has joined #linux-sunxi
<apritzel>
Wizzup: the BROM? That's part of the silicon, etched at the fab producing the chips
reev has quit [Ping timeout: 276 seconds]
kaspter has joined #linux-sunxi
<Wizzup>
I see. And the next part that is executed is already the sd card or usb fel
nixdork has quit [Remote host closed the connection]
<oliv3r>
tkaiser: the biggest problem is they have 'working' software
<tkaiser>
oliv3r: 'Working'? What do you mean?
kaspter has quit [Ping timeout: 260 seconds]
DullTube has quit [Quit: Leaving]
JohnDoe_71Rus has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org]
nixdork has joined #linux-sunxi
matthias_bgg has quit [Quit: Leaving]
matthias_bgg has joined #linux-sunxi
<ssvb>
Wizzup: in terms of parity with the desktop computers, https://linux-sunxi.org/Bootable_SPI_flash may be treated as a rough equivalent of the PC BIOS (except that we have boot priority quirks to deal with)
cnxsoft has quit [Quit: cnxsoft]
<ssvb>
Wizzup: now we need to wait until development boards with on-board SPI flash become available, but you can do a "hw mod" even now :)
<lennyraposo>
hey ssvb
<lennyraposo>
just to let you know allwinner should be getting their custom xorg today
<lennyraposo>
then it's wait and see
<lennyraposo>
I will see if I can get them to release what they have for us here
<lennyraposo>
two heads are better than one ;)
ht91109 has quit [Remote host closed the connection]
WhiteKIBA has left #linux-sunxi ["WeeChat 1.5"]
<speakman>
Hi folks. I've got a device driver which utilizes external interrupt on an GPIO pin (A20). But I can't find any information how to simply set a single gpio pin in EINT mode using device tree.
<speakman>
Do you have any good docs for doing this?
nove has joined #linux-sunxi
<wens>
does the binding take a gpio or interrupt?
premoboss has quit [Quit: Sto andando via]
<speakman>
hm? not sure what you mean by that wens. Can you elaborate?
<speakman>
the device driver first gpio_request(pin, "name"); and then request_irq(gpio_to_irq(pin), ...);
<Wizzup>
ssvb: Thank you! Most welcome.
<Wizzup>
(That info is most welcome*)
<wens>
speakman: uh, are you on mainline with dts?
<speakman>
wens: 4.3 and cloned the sun7i-a20-olinuxino-micro.dts for adding modifications
<wens>
speakman: so you can use "name-gpios = <...>" in the device node
<wens>
the pinctrl nodes are there because the pinctrl/gpio driver doesn't do proper locking
<wens>
linux considers pinctrl and gpio separate things, while they are combined for sunxi hardware
<speakman>
Will I need to change anything in the DBS at all then?
<oliv3r>
ssvb: i think the best contender would be a modded lime without flash
<oliv3r>
ssvb: i wonder if you can make a pcb, with SPI flash ontop and then solder it where the flash goes, kinda like the realtek wifi is solderd into place
<oliv3r>
tkaiser: well they have some android image that works with the SDK i guess, so they are done they have 'working software'
formruga has quit [Ping timeout: 244 seconds]
<speakman>
wens: Looking into /proc/interrupts there do seem to be interrupt allocated for the driver
<speakman>
78: 0 sunxi_pio_edge 22 Edge magstripe
<speakman>
(magstripe is my test driver)
<speakman>
This was working with the old 3.4 kernel, but back then I set the PIN mode in the FEX file.
<speakman>
Maybe it's easier to try a more established driver for this use. I'm just about to connect this one:
<tkaiser>
ssvb: But you don't want me to understand that?! :)
<ssvb>
tkaiser: well, I'm now trying to understand the answer myself, I'm not a statistician after all :) but there are some really skilled and experienced people there
<tkaiser>
ssvb: Please ignore my ignorance, I tried to understand as much as I can but one questin pops up immediately: How many samples do we need to do anything useful with the test results?
<KotCzarny>
he said 12 is enough
<ssvb>
tkaiser: yes, that's one of the things that I'm trying to figure out
<KotCzarny>
but magic number is 42
<wens>
speakman: as i said, the gpio driver will set the pin mode for you
<KotCzarny>
and for distro needs i think you can ignore all probabilities lower than 1%
<wens>
speakman: as pin mode and gpio are tied together
formruga has joined #linux-sunxi
<wens>
speakman: later on you can add a pinctrl node
<wens>
like for mmc_cd
<speakman>
wens: ok, then I don't have to worry about that pin mode stuff :)
<speakman>
wens: thanks for helping out! leaving office now, but tomorrow I will try to connect the egalax_ts driver using Device Tree.
<oliv3r>
ssvb: i can help you get more test samples for lime2 boards
<oliv3r>
ssvb: i quite a lot here :)
<oliv3r>
ssvb: as long as it can be done without costing me (much) time
<oliv3r>
e.g. insert sd card, run, copy/paste output :)
<tkaiser>
oliv3r: FEL boot rocks, just get most recent sunxi-tools, a beefy x86 machine with plenty of USB ports and test 20 Lime in parallel! :)
<tkaiser>
ssvb: oliv3r: Not kidding but maybe a fleet of Lime2 would really be helpful for the whole approach?
<oliv3r>
tkaiser: i'm not going to run all over the place to do that though :)
<tkaiser>
oliv3r: ;) But you don't even need an SD card to do the tests. The only device that showed weird behaviour when running lima-memtester through FEL boot was BPi M2+ (and that's nothing to care about)
<tkaiser>
Mini-USB to USB cable and another Linux host and you're done
<oliv3r>
tkaiser: my boards are built into devices
<oliv3r>
:)
<oliv3r>
so i can do enter, run things, remove
<oliv3r>
that's quick and esy
<oliv3r>
fel takes too much oliv3r time :)
<oliv3r>
i'll have to lug my laptop to our test room etc
<oliv3r>
not convinient :)
<oliv3r>
and i bet uploading lima-memtest + cpuburn + kernel etc takes a bit of time?
<tkaiser>
oliv3r: Ah, understood. Do they have a HDMI port exposed (since at least testing with one Lime2 whether you get the _correct_ visual feedback is essential)
<oliv3r>
ssvb: but since we have some very strange crashing issue, i can get our test team to run the tester on each of our test machines :)
<oliv3r>
tkaiser: i can test 4 or 5 lime's myself with the hdmi connected
<oliv3r>
but the other devices will operate without display
<tkaiser>
oliv3r: Doesn't take that long (max. 2 minutes) and then you can disconnect the laptop. Maybe faster than SD card
<KotCzarny>
well, one could setup one of the limes to be the host running those fel booting too
<tkaiser>
oliv3r: On the other devices monitoring AXP209 temperature is then enough. But we had an issue recently with the 2GB Orange Pis that lima-memtester was just a memtester without lima (so results worthless)
<KotCzarny>
so run lime2, make it felboot another lime, done
<KotCzarny>
(smaller than laptop ;)
<tkaiser>
oliv3r: But ssvb might correct me since there's also one situation (glowing red background) that already indicates failure and might require a connected display.
<oliv3r>
tkaiser: taht's WAY to long :p
<oliv3r>
2 minutes * 20 machines = 40 minutes
<oliv3r>
if you can explain that to my boss, why im not working for almost an hour, then sure no prob :)
<oliv3r>
swapping an SD card takes 5 seconds :)
IgorPec has joined #linux-sunxi
<oliv3r>
KotCzarny: that takes time to setup however :)
<oliv3r>
also, where are the results going?
<oliv3r>
sd you can store them :)
<oliv3r>
store on itself, power off when done
<oliv3r>
next machine
<oliv3r>
i can get a testguy to do that for 20 or 30 some machines
Netlynx has joined #linux-sunxi
Netlynx has quit [Changing host]
Netlynx has joined #linux-sunxi
reinforce has quit [Read error: Connection reset by peer]
<tkaiser>
oliv3r: Hehe, waiting for an answer from ssvb since this is currently not implemented. The test in its current form relies more or less on a monkey in front of the device watching closely a spinning cube on the display (at least that was how I felt last week ;) )
reinforce has joined #linux-sunxi
<TheLinuxBug>
oliv3r: how much 1 hour of your time cost?
<KotCzarny>
oliv3r: they arent networked?
<KotCzarny>
keep in mind usual test requires ~30 minutes
<tkaiser>
KotCzarny: One hour and better over-night instead
<ssvb>
btw, this e-mail had been posted before the Orange Pi PC board became popular and before we got nice DRAM reliability statistics from it
sherpa has quit [Ping timeout: 250 seconds]
<ssvb>
big thanks to tkaiser for providing guidance to the users and making it possible!
<TheLinuxBug>
:D
<ssvb>
now looks like oliv3r is representing a device maker company and has access to a sufficient number of board samples, this might be the next big step for us
<ssvb>
by using tuned DRAM parameters instead of the default ones, we can probably increase the DRAM clock speed from 408 MHz to 480 MHz
<ssvb>
but it just means that the new parameters are supposed to be more reliable compared to the default ones
<ssvb>
we still don't know if 480 MHz is universally good for pcduino2 or we would need to reduce it by one more step
<ssvb>
the number of board samples is still relatively small, and probably we will never get any additional data for this board (it is already discontinued)
jernej has joined #linux-sunxi
<ssvb>
tkaiser, oliv3r: as I mentioned in the e-mail post, we can make everything more automated, so that "Anyone who is interested in participating, would only need to use a special bootable SD card and just let the device out in the Internet..."
<tkaiser>
ssvb: An unattended mode is IMO a requirement to get a relevant sample count.
<ssvb>
tkaiser: yes, but still it first needs some work to get implemented, which basically means my unpaid (?) time
<ssvb>
tkaiser: second, some people may be freaked out by the fact that it needs Internet access and automatically reports the results to the "cloud"
<ssvb>
tkaiser: nobody has commented on this idea so far, that's why we are not moving anywhere
<ssvb>
so now you have a semi-automated procedure, which is not too bad, but still could be improved
pmattern has quit [Quit: Genug für heute.]
IgorPec has quit [Read error: Connection reset by peer]
<ssvb>
in addition, we initially needed a better controlled test drive for it :) so that we could identify the lima-memtester problem on opi+2e or anything similar
IgorPec has joined #linux-sunxi
<tkaiser>
ssvb: Sure, but to be honest: I only care about reliability, so if we now know that 672 MHz isn't ok with current settings on all H3 devices, use 624 MHz and you're done. People doing memory intensive stuff can then test on their own.
<tkaiser>
The whole testing has been insanely time consuming already.
<GeneralStupid>
tkaiser++
<ssvb>
tkaiser: it was time consuming because we were investigating and learning things in the process
<ssvb>
tkaiser: for example, your weird unreliable banana case
<ssvb>
tkaiser: with all this knowledge, now we can introduce additional automation of the process
<GeneralStupid>
i mean, if you buy a device which is overclocked by default and you need high performance you boght the wrong device...
<tkaiser>
ssvb: the next surprise is just around the corner ;)
<tkaiser>
GeneralStupid: 'overclocked by default' doesn't matter. I'm interested in the settings we use in Armbian
<ssvb>
GeneralStupid: it can't be helped, we can't trust the board manufacturers to do a sane job on the software configuration front :)
<TheLinuxBug>
ssvb: maybe it would be worth while to take some donations for linux-sunxi devs to help pay for the time needed to test these things. I am by no means rich, but if what tyou guys are doing is furthering the usefullnes of these boards, I am hapy to contribute 20-30$ or something. I think some others would too, you guys do a lot of awesome work for the community.
<GeneralStupid>
i did a lot of work on my own to get debian running. and nowadays you can just install armbian. that is a really big step forward
<ssvb>
TheLinuxBug: IMHO that's where the board manufacturers need to step in, because collecting donations from individual users may be problematic or even not quite legal from the point of view of the tax office
<TheLinuxBug>
I see
<TheLinuxBug>
well because you are a legal entity
<TheLinuxBug>
I can understand that
<TheLinuxBug>
what needs to happen is someon fill out the org paper work and regierter you guys as a non-profit in US and you could then take the donations, but probably more time in that then actuallly doing the testing :Z
<TheLinuxBug>
are not*
<TheLinuxBug>
All the manufacturers are just in it for $$ only unfortunately though it seems, so likelyhood of them actually coming forward to help? Must be very small chance?
<ssvb>
TheLinuxBug: the linux-sunxi community is not any kind of legal entity, it's just a bunch of people hanging around here, also including you :)
IgorPec has quit [Ping timeout: 240 seconds]
<ssvb>
also Red Hat and Free Electrons are working on Allwinner SoCs support in the mainline kernel, partially because that's a way for them to advertising themselves, and partially because of doing some paid contracted job (at least Free Electrons does that, not sure about Red Hat)
<GeneralStupid>
armbian <3 they really did a nice job there :D
<GeneralStupid>
jernej: your work looks great too -.-
<GeneralStupid>
i just wanted to say
<oliv3r>
KotCzarny: they are networked :)
staplr has quit [Ping timeout: 250 seconds]
<oliv3r>
tkaiser: yeah we can run it over night/weekend, but it has to run fully stand-alone
<oliv3r>
ssvb: i can help with generating a bootable image, we can then just add the binaries and let all that run from a boot script. but we need input/output (either by storing results/reporting it online)
<oliv3r>
ssvb: reporting to the cloud is not a really big problem for me (they are currently not networked)
<TheLinuxBug>
KotCzarny, jrg: I made my own case for my OPi will have to post a pic, lol its a quick hack job
<oliv3r>
ssvb: by testing fleet will expand from ~30 to about 50 or even 75
<NiteHawk>
Could be worse. Will keep an eye on it for a while, if they learn that their stuff doesn't last long, they'll move on...
<KotCzarny>
nitehawk, nice job
<KotCzarny>
what did you use to drill those holes?
<KotCzarny>
knife?
<NiteHawk>
rellla, turl, libv: i took the liberty to promote myself to "admin" for this (which is a bit of a security hole for "beaurocrats") - hope you don't mind
<TheLinuxBug>
a drill and a knife and boredem ;p
<KotCzarny>
:>
<TheLinuxBug>
was at one time a small first aid kit I believe
<KotCzarny>
i prefer high speed drill
<TheLinuxBug>
made a perfect little case though
IgorPec has quit [Ping timeout: 260 seconds]
<KotCzarny>
yup
<KotCzarny>
recycling!
<KotCzarny>
;)
<TheLinuxBug>
actualy
<TheLinuxBug>
shoudl show you my A10
<TheLinuxBug>
you would get a kick out of its case
<KotCzarny>
:>
<KotCzarny>
my bpi-m1 has a box from french sweets
<KotCzarny>
even has little eiffel tower on the side ;)
<KotCzarny>
btw. you can still paint that box to some color/pattern
<KotCzarny>
white isnt nice with black cables sticking out
Mr__Anderson has quit [Remote host closed the connection]
<dave0x6d>
there's no accelerated crypto on the Allwinner H3 (A7), is there? I see 'half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt' listed as the CPU features.
<dave0x6d>
looks like the A7 is not going to be usable with LUKS.
<dave0x6d>
well, maybe it's okay. Seems like multithreaded AES can get up to 90 MB/s according to GeekBench
<igraltist>
this is a lot
fredy has quit [Excess Flood]
jstein has quit [Remote host closed the connection]
xalius has joined #linux-sunxi
fredy has joined #linux-sunxi
al1o has quit [Quit: My MacBook Pro has gone to sleep. ZZZzzz…]