<calris_>
Why doesn't git clone git://github.com/linux-sunxi/u-boot-sunxi.git bring in all the branches?
<calris_>
I only get sunxi, no sunxi-current or lichee-dev :(
<ssvb>
calris_: that's how git works, you can get the list of remote branches and tags with 'git ls-remote' and checkout the needed branch
<calris_>
git checkout -b sunxi-current origin/sunxi-current should do the trick I imagine?
vinifm has joined #linux-sunxi
vinifm has quit [Client Quit]
<calris_>
Hmmm, running Eclipse on an A10 is painfully slow - any ideas on a lightweight IDE (please don't say vi - I'm a big fan, but need a few more features)
<oliv3r>
wingrime: A1x doesn't go 'all the way' and on top of that, if the DDR routing on the PCB is done poorly, it won't reach the A1x's max. You are always free to experiment and find the highest stable setting for your board of course :)
<oliv3r>
i wonder what a reliable memory benchmark is that notices corruptions
<hramrach>
something like memtest that loads from bootloader directly
<oliv3r>
note though, i only know what i found in the datasheet, i have no clue how all this works
<hramrach>
also the question is if we can use the SRAM for something useful under normal operation
<wingrime>
olvi3r
<oliv3r>
if I would have to guess what sram is and how it works; i would have said it's die memory, that is available during boot and used as cache during normal operation (what is left over)
<hramrach>
if not the suspend code can just stay there indefinitely, sure
<wingrime>
oliv3r look at page 82
<wingrime>
oliv3r simply not all SRAM are can be used
<oliv3r>
wingrime: yes, some undocumented bits :) and the mapping
<oliv3r>
wingrime: yep, SOME is available
<wingrime>
oliv3r some devices have own zones
<oliv3r>
calaris and I actually talked about using SRAM-B for SPL :)
<wingrime>
oliv3r AREA A can be used by us
<oliv3r>
he tested it in uboot, and it could read/write to sram-B without problems
<wingrime>
oliv3r other can be used but not same time with HW
<oliv3r>
wingrime: what about SRAM_A3_A4_MAP? 'map to AHB' (though the documetnation sucks, no doubt about it
<oliv3r>
wingrime: but you are far far far more smarter about this then I am; i just read the datasheet :D
<wingrime>
AHB - it global bus
<oliv3r>
only way to find out, is to 'test' and see what happens
<oliv3r>
wingrime: yeah, but you map that area either to cpu/dma (bit 5:4) or ... 'undefined'
<wingrime>
oliv3r: SRAM have not controller at all, it simplest ram ever
<oliv3r>
I guess you can best compare SRAM with 'regular ram' on a microcontroller'
<wingrime>
oliv3r: some devices have direct connection to it
<oliv3r>
like an AVR has about 512 bytes of ram, 'sram' :)
<wingrime>
oliv3r: or it can be access using general bus
<oliv3r>
and if i'd connect ddram to one of the pins (theoretically) that'd be 'normal' ram
<oliv3r>
correct?
<wingrime>
oli3r: we can place some hi-performance stuff to it
<wingrime>
oli3r: some old IC like 8051 can ability to external sram
<oliv3r>
i may have done copy/paste/typo on the addresses
<oliv3r>
and then made the mistake of wrongfully doing some math :)
<oliv3r>
btw, the i, D and L2 cache are allready all sram i just ntice :)
<oliv3r>
is it unreasonable to assume, if you 'map' a region to CPU/DMA it's used as L2 automagically?
<ssvb>
wingrime: SRAM is not always fast, you can't take this as granted
<wingrime>
ssvb: we have L2 in SRAM
<oliv3r>
I and D cahce is also in sram
<wingrime>
oliv34: a1-a4 100% usable
<wingrime>
C1 c3 also possible usable at boot
<hramrach>
hmm, interesting
<oliv3r>
wingrime: so we can 'extend' l2 cache with a1 and a4?
<paulk-desktop>
am I the only one to have issues with volume keys detection?
<oliv3r>
wingrime: B is useable from u-boot/spl
<paulk-desktop>
ie: I must press a long time to have an event create
<hramrach>
the standby.bin uses different calling convention from rest of kernel
<paulk-desktop>
created*
<hramrach>
probably to save stack space
<paulk-desktop>
with the sun4i-keyboard driver
<wingrime>
oliv3r: I don't think we can do that
<oliv3r>
paulk-desktop: if i'm not mistaken, volume keys are hooked up to LRADC
<paulk-desktop>
yep
<paulk-desktop>
so I'm wondering whether I should rewrite the driver or nor
<paulk-desktop>
or not*
<oliv3r>
the driver probably needs proper re-writing anyway
<oliv3r>
so nothing lost there :)
<paulk-desktop>
yay
<paulk-desktop>
so I'll do that
<oliv3r>
i haven't looked much into LRADC, but from what I understand, you could either hookup 2 buttons OR an analog 'dial' to it to control 'volume' (or whatever you want to, since it's an input)
<oliv3r>
you could even turn it into a 'scrollwheel' i guess :)
<paulk-desktop>
ok
<wingrime>
oliv3r: I looked for lradc registers
<wingrime>
oliv3r: It simple
<wingrime>
oliv3r: but don't forget for drop standby.bin we must configure lradc as irq source for wakeup on suspend
<ssvb>
wingrime: SRAM can be fast, but this is not always the case, in embedded systems you may have SRAM primarily optimized for low power consumption
<wingrime>
ssvb: may be , but it must be faster than dram
<ssvb>
wingrime: the real performance depends on how this SRAM block is connected to the CPU (it may have low clock speed and narrow bus width)
<oliv3r>
in the case of A1x, it seems like it's used for both L2 cache and other tasks, so it could be fast :)
<wingrime>
ssvb: not sure
<wingrime>
ssvb: we have blacks
<ssvb>
wingrime: not really, DRAM supports burst reads and writes, it is really good for sequential accesses but has high latency
<wingrime>
ssvb: also SRAM not need commands qeues
<ssvb>
yes, SRAM has lower latency
<wingrime>
sram can be accessed indepently but I can't say in aur case
<ssvb>
but bandwidth needs to be verified
<wingrime>
ssvb: you can try
<ssvb>
yes, I can :)
<ssvb>
so can you
<wingrime>
ssvb: if I looked to verilog files for a1x.....
<wingrime>
but it defently NDA.....
rellla has joined #linux-sunxi
<wingrime>
oliv3r: a10 have BIST ?
<wingrime>
Built-in self-test
<wingrime>
?
<oliv3r>
wingrime: possibly, but we know nothing of it
<wingrime>
C3 must be usable also
<oliv3r>
wingrime: partially, for sure
<oliv3r>
btw, there's also a 'coresight debug module'
<wingrime>
I thing SRAM can direct connection with some devices avoiding general soc bus
<ssvb>
wingrime: one of the possible tests is something like "time DISPLAY=:0 xterm +j -maximized -e cat some-large-text-file.txt", this is scrolling the text one line at a time without skipping
<oliv3r>
why, just why. drivers/video/sunxi/disp/disp_tv.c: script_parser_fetch("audio_para", "audio_pa_ctrl",
<oliv3r>
what does TV code have to do with the 'speaker' of a tablet
<wingrime>
oliv3r: have a13 "B" and "C2" section ?
<oliv3r>
wingrime: absolutly, no Idea.
<oliv3r>
you can test with u-boot console, 'mtest' i belive?
<oliv3r>
wingrime: but i'm assuming ALL arm cores have 'B' section as its used for trustzone so arm probably says its mandatory
<oliv3r>
even if you'd never use it
<wingrime>
oliv3r: you wan't use sram for what?
<wingrime>
ssvb: are you tested neno opt for any region size?
<hramrach>
well, you can DMA to sram which is presumably fast
<hramrach>
but sram is small buffer and DRAM is generally good for the kind of stuff dma does
<wingrime>
ssvb:maybe there is some cases when old routine better
<hramrach>
so for fast connecting devices it does not seem that awesome
<hramrach>
putting some cache/page table management coe there would be cool
<hramrach>
no page misses on that code, ever
<hramrach>
but it's probably too large in Linux
<wingrime>
what do you want to move to SRAM ?
<hramrach>
can't really think of anything that would bebefit from such small piece of memory
<hramrach>
even if it's fast
<hramrach>
how large is the usable area?
<wingrime>
40kb
<hramrach>
that was the size of system ram on zx spectrum ..
<hramrach>
you could run hte whole OS in that ;-)
<hramrach>
heh, 'os'
<wingrime>
hramrach: more defenetly
<hramrach>
well, pages are 4k so you caould perhaps fit top level page table in there or something. if it's even possible for the CPU to read it from there
<wingrime>
48 kb "A" blank
<wingrime>
ssvb: a10 have 256 L2 a13 have 128 L2 ?
<wingrime>
hramrach 48kb generaly usable
<wingrime>
but we have some interesing mem
<wingrime>
ISP ???? have 64 kb
<wingrime>
VE (video) 512 kb
<wingrime>
but it still unclear
rellla has quit [Remote host closed the connection]
<wingrime>
ssvb: can you gix x11 log spam
<hramrach>
the cmap spam?
<wingrime>
FBIOPUTCMAP:Invalid argument
<hramrach>
that's upstream problem
<hramrach>
can be fixed probably
<hramrach>
the 512kb from video is interesting for headless kernels I guess
<hramrach>
any reasonable way to make the kernel build system compile a file twice?
<hramrach>
I just #included them but that's not exactly reasonable
<wingrime>
hramrach: test, are you still have problems with gadget
<wingrime>
?
<hramrach>
did not test recently
<hramrach>
presumably it still fails after unload /reload module
<wingrime>
try agian
<wingrime>
It must be in stage
<hramrach>
will need to build new kernel
<hramrach>
currently have the android gadget which does not cause kernel panic because it cannot be unloaded
<wingrime>
hramrach: and works good?
<hramrach>
etherenet works when only ethernet is enabled
<wingrime>
hramrach: cam you send updated a10/a13 configs with gadget enabled
<hramrach>
well, was looking into configs
<hramrach>
but can only test a10
<hramrach>
did not even get to testing the a10 one
<ssvb>
SRAM does not need periodic refresh (which is very good for power consumption), but it is definitely not as fast as L1 cache or even L2 cache
<hramrach>
and compared to dram?
<ssvb>
it has better latency, but worse bandwidth
<ssvb>
wingrime: yes, the NEON code in xorg driver still needs a bit of tuning to make sure that it is always a performance win (especially on Cortex-A15 chromebook)
<ssvb>
hramrach: SRAM is one of many buzzword, and a lot of people just assume that it is so obviously guaranteed to be a good solution that no verification is even ever needed :)
<hramrach>
is it fixed mapped or does it get paged too?
rz2k has joined #linux-sunxi
<oliv3r>
wingrime: we want to see if we can use SRAM B (which looks really good) for SPL so we can load fat/ext partitions with the extra ram :)
<oliv3r>
wingrime: linux-sunxi probably lies about the memory sizes, i could have made a mistake when writing those, 512kb sram for VE sounds a bit too much
Turl has quit [Ping timeout: 246 seconds]
<wingrime>
olib3r: 128kb L2 on a13 and 256 on a10 can be true?
hipboi has joined #linux-sunxi
<ssvb>
wingrime: both A10 and A13 have 256K L2
<wingrime>
ssvb: a13 manual say that 128k
<wingrime>
82 page
<ssvb>
I'm almost sure it is a typo, every other source says 256K
<hramrach>
these small caches
<hramrach>
even Pentium II used to have 512kb iirc
<oliv3r>
256 L2 isn't that odd; even modern CPU's only have that sometimes
<hramrach>
modern lame cpus like celerons
<ssvb>
Pentium II was a high end "server" processor, a lot of users who cared about price/performance actually had Celerons
<hramrach>
well, this is Chinese chip so what would you expect ..
Turl has joined #linux-sunxi
<ssvb>
Only Samsung and Apple used 512K L2 for Cortex-A8
<hramrach>
well, back then the first celerons were more powerful than the server CPUs because celerons had the cache clocked at teh cpu speed but PIIs at half .. Intel blunder
<ssvb>
well, Cortex-A8 has >=10 cycles latency for L2 cache, Cortex-A9 is even worse because it uses external L2 cache
<ssvb>
half clock speed is not necessarily too bad, it is still fast enough and large
<ssvb>
cache misses for smaller cache are significantly slower
<hramrach>
it's bad compated to full-clock
<hramrach>
and some media decoders might fit into small cache
<hramrach>
small working set or predictable access pattern
<ssvb>
well, we have L1 / L2 / L3 caches hierarchy, each next level is slower but can have much larger size
<ssvb>
L1 cache must be fast, this is really critical
<ssvb>
hramrach: placing page table in SRAM might be not a bad idea and could provide performance improvement on really memory heavy workloads when the caches are thrashed badly
<ssvb>
we could try to test it
<ssvb>
SRAM has lower latency than DRAM, so page table walk should be a bit faster
fra79Wii has joined #linux-sunxi
<hramrach>
hmm, so I have like 3 ethernet and 2 serial gadgets or so which I can load
<hramrach>
and audio
<hramrach>
and printer
<hramrach>
others don't load at all
<fra79Wii>
Hi guys.. I'd like to try to good a 3.4 kernel from nand.. to actually boot android.
<fra79Wii>
…to boot not good… I have tried u-boot from lichee-dev but it doesn't seems to boot the uImage, but keeping booting /dev/block/nandc
<hramrach>
that's bootloader configuration
<hramrach>
the default boot script probably does that
<fra79Wii>
I thought I could boot fom SD and then place the boot arguments to boot band.. like boot=/dev/block/nandd… but I was wondering if I have to boot on nandd or nandb, where the boot env is
<hramrach>
wingrime: I ca now unload/reload the g_ether driver without kernel panic \o/
<fra79Wii>
or where the real root is.. so nandd.. I'm trying booth :P..starting with nandb
<hramrach>
fra79Wii: can't you look on the nand?
<fra79Wii>
I'm not sure which way is the right "android" way to boot...
<fra79Wii>
if it were linux I'd boot on nandd..
<hramrach>
to boot from the andriod partitions
<hramrach>
so to copy the kernel to the partition where it usually is
<fra79Wii>
there is nandb which is unmountable.. I didn't figured out which file system is...
<fra79Wii>
agh, ok.. that's doesn't works...
<hramrach>
it might be raw kenel image or boot loader config
<fra79Wii>
I would like to boot a 3.4 and I haven't understand I could i do it
<hramrach>
I don't use android so can't really help you wit that
<fra79Wii>
so when I create a boot.img with 3.4 to flash the partition it simply doesn't boot
<fra79Wii>
I was trying to find some workaround to somehow boot android with 4.2
<hramrach>
do you have serial console?
<fra79Wii>
3.4…I mean.. nope, I have nothing else then a sd and a USB cable :P…. sitting on a couch..
<fra79Wii>
zero equipment here
<fra79Wii>
ok I try out to bypass the bootloader booting from sd_to_nand
<hramrach>
really hard to ttell what fails then
<wingrime>
hramrach: test cable
<hramrach>
did the same test which caused kernel panic previously
<paulk-desktop>
mhh, it seems that the LRADC is highly unreliable
<wingrime>
hramrach: no panic when you remove cable ?
<hramrach>
no panic
<wingrime>
mnemoc: ^
vinifm has joined #linux-sunxi
<hramrach>
but there is some routing problem
<hramrach>
when I unplud the USB cable I can still ping the board
ibrah has joined #linux-sunxi
<hramrach>
dd, when I unplug the usb the ping are routed elsewhere
<hramrach>
but when the cable is plugged in they appear to go through usb
<hramrach>
with lower round trip
ibrah has quit [Ping timeout: 245 seconds]
<paulk-desktop>
oliv3r, so in the end, my issue was about the LRDAC freq being too low
<paulk-desktop>
so I refactored the code and all for nothing in the end
<paulk-desktop>
would it still be worth submitting to the mailing list?
ibrah has joined #linux-sunxi
ibrah has quit [Read error: No route to host]
ibrah has joined #linux-sunxi
<wingrime>
olib3r: do you think we can drop SPL and boot uboot directly ?
wingrime has quit [Read error: No route to host]
ibrah has quit [Ping timeout: 264 seconds]
hramrach has quit [Remote host closed the connection]
shineworld has quit [Ping timeout: 264 seconds]
shineworld has joined #linux-sunxi
<oliv3r>
paulk-desktop: if you rewrote the driver that fast, sure :p
<oliv3r>
wingrime noway; but we can make better SPL
bsdfox has joined #linux-sunxi
fra79Wii has quit [Quit: This computer has gone to sleep]
ganbold_ has joined #linux-sunxi
wingrime has joined #linux-sunxi
ganbold__ has quit [Remote host closed the connection]
ibrah has joined #linux-sunxi
hipboi has quit [Quit: Leaving]
eebrah has quit [Ping timeout: 256 seconds]
ibrah has quit [Ping timeout: 255 seconds]
techn_ has joined #linux-sunxi
<wingrime>
mnemoc: eLIBs_CleanFlushDCacheRegion
<wingrime>
it realy required for dma?
<wingrime>
strange that only sun3i have it
<wingrime>
oh.. I don't right here
Skaag has quit [Read error: Connection reset by peer]
Skaag has joined #linux-sunxi
<wingrime>
hramrach: do you think we need some limit when send on dma
<wingrime>
I mean small pieces can be sended using pio mode
<wingrime>
now driver do it over dma always
<wingrime>
oliv3r: ssvb: same to you
<paulk-desktop>
oliv3r, well I didn't really change much to the original design of the driver, I just refactored it
<wingrime>
paulk-desktop: thanks a lot
<wingrime>
paulk-desktop: are you make check-patch ?
<paulk-desktop>
then I have a second patch to make the frequence up to 250Hz
<paulk-desktop>
which is required to make fast presses detectable
<wingrime>
please consider that freq means powerconsume
<wingrime>
but i saw simular fix from aw in dev branch
<wingrime>
so I think all god
<wingrime>
*good
<paulk-desktop>
there is nothing to question there, it's that or unusable
<paulk-desktop>
with CWM recovery, where the interface navigation relies on volume keys, it's unusable without 250Hz
<paulk-desktop>
I'll try 125Hz to see if it would be fast enough
<paulk-desktop>
125Hz is borderline: regular button press works but as soon as I want to go a bit faster, it blocks
<paulk-desktop>
so I guess 250Hz is the way to go
<wingrime>
paulk-desktop: I still wait CM for a13
<paulk-desktop>
wingrime, I'll be porting Replicant, a fully free CyanogenMod derivative
<paulk-desktop>
I guess CM would be doable as well
<paulk-desktop>
I sent the patches, I'm not sure it worked
<paulk-desktop>
my smtp server says ok, we'll see
<paulk-desktop>
also I saw there is fedora for a13 devices -- is it gnome-shell?
drachensun has quit [Quit: Leaving]
<wingrime>
I use debian
<wingrime>
and linaro
paulk-desktop has quit [Quit: Ex-Chat]
paulk-desktop has joined #linux-sunxi
<wingrime>
there many problems at all
<vinifm>
i used both, linaro is good
<wingrime>
debian xorg have problems with touch
<wingrime>
hung on suspend when xorg running
<vinifm>
I had problems with wpa on Debian
<vinifm>
wpasupplicant
<paulk-desktop>
I see
leowt has joined #linux-sunxi
<leowt>
how are the mali-400 x11 drivers?
<leowt>
still slower than fb?
eebrah_ has quit [Read error: Connection reset by peer]
eebrah_ has joined #linux-sunxi
<oliv3r>
paulk-desktop: fedora 18 image is xfce4
<oliv3r>
paulk-desktop: as for replicant vs cyanogenmod isn't it basically the same, except for replicant trying to replace all blobs with oss components or leave them out? In that case, it doesn't matter which you use. Mali is needed for both ;) and turl is working on CM support (http://github.com/allwinner-dev-team)
<paulk-desktop>
oliv3r, yeah that's it, pretty much
<eebrah_>
#lima
<paulk-desktop>
though it should run fine with only free software
<paulk-desktop>
since mali is fast enough without blobs
<paulk-desktop>
and there is free audio and likely camera is doable too
calris has quit [Remote host closed the connection]
<shineworld>
yeah, but I don't understand how can a tablet maker to put all them at 1.5ghz without problems
<Dreadlish>
maybe he's lying ;d
<Dreadlish>
or manufacturer add governor to cpu, which can manage that
bfree has quit [Quit: Konversation terminated!]
<oliv3r>
paulk-desktop: it won't matter which rootfs you use for android, you'll be stuck with our kernel and that's besides cedarx and mali 100% free. and thus both replicant and cm10 will be (besides those parts) be 100% free. and i dont know what replicant does exactly (i thoght they clone cm git) you should have near identical system. granted having replicant on board wold be awesome and replicant sure could use a system as open
<oliv3r>
paulk-desktop: are ou subscribed to the list? i haven't seen it yet so looks like it didn't come through. should be near instant
<oliv3r>
shineworld: they all lie. most of the time what they do, is 1.1 GHz for arcore +375 MHz for mali core = almost 1.5 GHz
<oliv3r>
:)
<paulk-desktop>
oliv3r, yeah, so the difference with CM will be minimal
<paulk-desktop>
yes I subscribed to the list
<paulk-desktop>
it seems that the message didn't go through
<paulk-desktop>
I used git send-email
<paulk-desktop>
I'll remove my relay host and see
eebrah_ has joined #linux-sunxi
<oliv3r>
i had to setup my mail client specifically for git
<paulk-desktop>
Well it does say:
<shineworld>
by my last test I'm satisfied with mali performances on cubieboard