ktifhfl has quit [Quit: Ping timeout (120 seconds)]
ktifhfl has joined #openwrt-devel
<csharper2005>
djfe: device has dualboot but ubi1 is empty because fw version is 1.00 and we have no new fw version from Mercusys at the time. First U-Boot at least check ubi0 format and requires all 3 volumes with the right names.
<hgl>
Ansuel: I've run tested the refactor build, and also sent the LuCI fix for the uwsgi package. I mentioned a few pending issues I discovered in the refactor PR. Please take a look when you're available.
<schmars[m]>
that's what sysupgrade has looked like for me for a long time
<oliv3r[m]>
robimarko: was just looking at that realtek dump; but that's what it is :( I just hope we can get some magic values 'demagiced'
<oliv3r[m]>
and if you look at how they are using regmaps, it'll make you cry a little on the inside
<schmars[m]>
lost count of how many people i've told that "command failed" means "command is succeeding" in this particular case :)
<Ansuel>
oliv3r happy to talk in private ?
<oliv3r[m]>
Ansuel: no problem!
<oliv3r[m]>
robimarko: in any case, I think having a dialog with realtek is super valueable, and hopefully we can pull them into the void a little
<oliv3r[m]>
but one decision openwrt has to make; is 'okay to accept some 'lesser quality' to keep the dialog going, or stick to our guns, and value quality more
<f00b4r0>
schmars[m]: darn. Makes it impossible to assert success in scripts :(
<oliv3r[m]>
but look at this for giggles: "clk_regmap_update(&clkm->clkr, clkm->pll_ofs, 0x1f << sft, val << sft);" instead of just doing 'proper' regmaps :(
tidalf_ has quit [Remote host closed the connection]
tidalf has joined #openwrt-devel
<oliv3r[m]>
robimarko_: they just want to check a box; 'upstream openwrt support'
<robimarko>
oliv3r[m]: I get that, but what then happens is they just vanish and then we are stuck with bumping the kernel until its gets impossible and it gets removed
<robimarko>
Its supposed to be a 2 way street
<oliv3r[m]>
yeah, but that's not how it works :p
<oliv3r[m]>
they have gotten a new job by then :)
<robimarko>
Well, that is why we cant just merge stuff like realtek target was
<robimarko>
Cause we are left with bad upstream support and no docs
<Ansuel>
well providing docs would be real good
<Ansuel>
even an NDA agreemen or some sort...
<robimarko>
I think we all know that is not going to happen
<robimarko>
I wish vendors would offer docs even under NDA to certain community members
Borromini has joined #openwrt-devel
<oliv3r[m]>
i still fail to understand what the advantage is of keeping your docs secret, but offering at the very least some code :)
<Ansuel>
even the reg design is industry secret
<Ansuel>
the dma descriptor... if you provide docs with how to enable some offload feature there goes qcom selling qsdk support :D
<robimarko>
Same here, especially with old SoC designs
<robimarko>
Ansuel: QCA doesnt even sell QSDK support, its free if you buy the silicon
<f00b4r0>
robimarko: isn't that what Mediatek does (providing docs)? I thought they were the better player here
<robimarko>
f00b4r0: I dont think they are providing docs to anybody other than Felix
<robimarko>
All of the Filogic work that Daniel did is based of the SDK code as far as I am aware
<f00b4r0>
robimarko: that's still one more person than, well, every other vendor? :)
<robimarko>
f00b4r0: I dont know what relationship Felix has with MTK
<robimarko>
But I give MTK credits for really improving when it comes to upstreaming
<robimarko>
They are now contributing even to U-boot
<f00b4r0>
*nod*
<robimarko>
So, better than other vendors I know
<robimarko>
QCA is really hit and miss, they seem to start upstreaming certain SoC and then just stop
<robimarko>
Their mobile SoC-s on the other hand are quite active
<Ansuel>
android and google
<f00b4r0>
^
<Ansuel>
ok lets put android on ipq soc
<Ansuel>
LOL
<oliv3r[m]>
so frustrating and makes you feel helpless though
<oliv3r[m]>
you'd think that they have learned in the last 10 years; that upstreaming has benefits
<oliv3r[m]>
especially with all the hot discussions on 'longer security reports'
<f00b4r0>
oliv3r[m]: legal teams don't care.
<robimarko>
Nah, I feel like they just start turning around and then legal or any other senior management gets replaced and we are at square one
<oliv3r[m]>
yeah; far more likely
<Ansuel>
my idea is that hw team is peak and superior... also who design stuff.. but the rest of the sw part is rushed till the thing barely work
<robimarko>
Well, that happens when you dont have time
<oliv3r[m]>
i see that at $work too; hw gets pushed out; without (much) software involvement; hardware is validated; this alll happens in days to a few months; the next few years, software gets to dance; and always under pressure
<robimarko>
Not to mention all of the fun HW bugs and erratas
<oliv3r[m]>
yep; but lets face it; hw design is generally a lot 'cheaper' (in terms of time/skills)
<robimarko>
Oh yeah
<oliv3r[m]>
sure, doing an AMD CPU is complex; but i've done FPGA dev; and for the most part; just click some components together in your IDE, which generates vhdl for you; write a bit of code for your modules
<oliv3r[m]>
not even to speak of 'general' PCB design
<f00b4r0>
errr
<f00b4r0>
that's not my understanding. Custom silicon design is *not* cheap. Skills required are usually very high, the design tools are very pricey, not to mention time, prototyping, validation etc
<f00b4r0>
could be wrong though, I lost touch with that venue decades ago
<Ansuel>
new chiplet tech improved the situation a lot but still it's magic
<f00b4r0>
*nod*
<f00b4r0>
usually sw is the "cheap" side of things, that can typically be outsourced too. At least from mgmt perspective.
<Ansuel>
and i bet that is the main problem... they outsource to the wrong people
<Ansuel>
and they go cheap on the outsourcing
<f00b4r0>
the high barrier to entry to hw design is pretty much the very reason why there are so few players
<f00b4r0>
oliv3r[m]: in fact, FPGA dev is *not* hw design: the hw is already there for you, it's merely "reconfigurable". That's exactly what makes it easy ;)
<oliv3r[m]>
I know; I graduated even on FPGA design :p but this is of course what most of these SOC vendors do
<oliv3r[m]>
mind you, 'reconfiguring' it in pure code (VHDL or what not) is still complex and hard :D
<oliv3r[m]>
doing a custom asic (one without (many) standard IP blocks; that's pretty hard and expensive
<oliv3r[m]>
but a soc; is "easy"
<f00b4r0>
I respectfully disagree :)
<robimarko>
Even the big players use most of the peripherals off the shelf
<robimarko>
But it aint easy
<f00b4r0>
and I know for sure that if I had pursued a career in hw design when I ended my studies, I'd make quite a lot more money than I did as i swe :)
<robimarko>
Doing silicon design yes, but I have friends doing automotive embedded and its less money than SW
<f00b4r0>
automotive pays jacks shit, though? :)
<f00b4r0>
regardless of field
<robimarko>
Well not really, I have friends doing SW for the same automotive company and they are payed better
<f00b4r0>
i see
Borromini has quit [Ping timeout: 480 seconds]
<f00b4r0>
so that's why modern cars have all sorts of electronic failures ;D
<robimarko>
So HW department is getting a big angry recently as they froze hiring HW engineers but are hiring and paying more in the SW department
<robimarko>
I think we can thank the bean counters for wonders of modern cars
<Ansuel>
f00b4r0 it's bad considering a wrong design and your battery catch fire
<f00b4r0>
robimarko: *nod* :)
<f00b4r0>
Ansuel: I'm lead to believe that BEVs catching fire is a feature, not a bug xD
<robimarko>
Considering all of the modern electric cars have cooled batteries almost all fires are when the battery is pierced externally
<f00b4r0>
well LiPo is inherently unsafe. At least the typically used chemistries
<robimarko>
Yeah
<hurricos>
robimarko: nice to see realtek contributing indeed, but yeah, driver porting is :(
<oliv3r[m]>
a 'simple' SoC is easy, not a super high-performance soc; as i said, you drag some off-the-shelf peripherals into your IDE, and you have a SoC.
<robimarko>
hurricos: yeah, nice to see any vendor contributing, but dumping code full of magic values doesnt help
<hurricos>
truly not. I'm now seeing the LIST of commits T_T
<oliv3r[m]>
whats worse; the current state of rtl83xx/rtl93xx makes it look like what they share is 'okay'
<robimarko>
And with the claim of upstreaming since kernel 5.8 I am not holding my breath for this getting upstreamed ever
<robimarko>
oliv3r[m]: I wouldnt measure anything against those targets, cause even QCA-s broken crap looks good compared to that
<oliv3r[m]>
i checked the 5.8 code; it's just some dts with basic arm stuff (not upstreamed by them btw)
<oliv3r[m]>
only some off-the-shelf compatibles are there (CPU, UART etc) and 1 driver (wdt) which is like 50 loc
<robimarko>
Hehe, so somebody was enthusiastic and then just gave up when the "fun" stuff started
<Ansuel>
or probably fired
<Ansuel>
or was too valuable and moved to other project
<hurricos>
:(
<robimarko>
Well, looks like that initial upstreaming was a contract job with some SUSE dev
<robimarko>
Last DTS change if we dont count the generic UART node rename was in 202
<robimarko>
*2020
<hurricos>
I should also know better to trust realtek to make good hardware. It's a combination of community work to decipher register documentation, and afaict fairly simple featureset that has made rtl83xx and rtl93xx stuff work so far
<hurricos>
this is the realtek of rtl8812au fame ....
<hurricos>
rtl88[1234][1234][abcde][ue] fame =_=
<Ansuel>
hauke (no idea if you notice but send you a pm)
<robimarko>
hurricos: Well, they are basically just dumping all of the not fully standard HW drivers in a PR
danitool has joined #openwrt-devel
<guidosarducci>
Ansuel: thanks for committing the IO_URING PR#12482. Any news on your 6.1 PR?
<Ansuel>
will take care once we branch... but it won't take a lot to refresh the 6.1 pr
<oliv3r[m]>
but, again; now this is politically sensitive; push them away with 'refactor more' (a few times) and potentially never hear back from them and have them give up; or olive branch things
<robimarko>
oliv3r[m]: How to olive branch this?
<oliv3r[m]>
Ansuel: eta for 6.1? I really want to move away from 5.15 asap :p
<oliv3r[m]>
robimarko: that's the political sensitive task; just saying 'its not good enough' won't help here either
<oliv3r[m]>
I do wonder where they are with that 'we are busy upsteraming, but it takes a long time' when did that start, where are those commits? what's the flak they are getting there
<robimarko>
oliv3r[m]: Well, that is my issue
<robimarko>
They are not upstreaming at all
Borromini has joined #openwrt-devel
<guidosarducci>
Ansuel: no, not too hard I agree. I just updated and tested malta for 6.1, found upstream bugs as usual :-(.
<robimarko>
If they were actually sending stuff upstream then I would be happy, but nothing changed since 2020 upstream
<f00b4r0>
robimarko: oh but they are upstreaming... to openwrt. And it's hard ;^}
* f00b4r0
hides
<Ansuel>
guido i already god fun with ipq806x...
<Ansuel>
nand driver was broken
<Ansuel>
pcie was broken...
<Ansuel>
gpio broken...
<robimarko>
f00b4r0: You can look at it that way
<Ansuel>
but was quick enough to send fixes even before 6.1 was outside rc
<robimarko>
Ansuel: That is the fun stuff
<Ansuel>
oh no i have to recover working on the LED patch... someone help me
<guidosarducci>
Ansuel: <sigh> my "fun" with 6.1 malta was confirming long-standing suspicion that current MIPS32 BPF JIT is broken and basic tests weren't run.
MaxSoniX has quit [Quit: Konversation terminated!]
<guidosarducci>
Ansuel: irony is I wrote, tested and submitted my own JIT upstream in 2021, but it wasn't used.
gladiac has quit [Quit: k thx bye]
cbeznea has joined #openwrt-devel
PtitGNU has quit [Remote host closed the connection]
PtitGNU has joined #openwrt-devel
Borromini has quit [Quit: Lost terminal]
minimal has joined #openwrt-devel
<domon>
Hi, maybe somebody knows, does exist of the dump/clone of codeaura?
<slh>
realtek has gone more active in terms of mainlining wireless drivers and reviewing changes for drivers, including those that aren't really in their prime focus, still a long way to go, of course - but some kind of (slow) change is visible
<slh>
compared to throwing a GPL tarball over the fence and going into hiding
djfe_ is now known as djfe
rmilecki has joined #openwrt-devel
rmilecki has quit [Read error: Connection reset by peer]