danitool has quit [Remote host closed the connection]
floof58 is now known as Guest1997
floof58 has joined #openwrt-devel
Guest1997 has quit [Ping timeout: 480 seconds]
<Mangix>
how do i configure cookie timeout on luci?
floof58 is now known as Guest2000
floof58 has joined #openwrt-devel
Guest2000 has quit [Ping timeout: 480 seconds]
<russell-->
neggles: i found the u-link nand videos on youtube, but he never identifies the TSSOP48 clip thingie or the adapter board that fits in the u-link nand (but i think i want one, especially if i can run on linux)
<neggles>
russell--: I don't know if it works on linux, I suspect not, the software used to be open-source but that just resulted in the market getting flooded with cheap chinese clones
<neggles>
the guy who makes it is very helpful and replies to emails etc. very quickly
<gch981213>
oliv3r[m]: In a traditional MIPS processor, when a branch instruction is decoded, the next instruction has already been fetched. To simplify the hardware, MIPS decided that the instruction after a branch is always executed no matter whether the branch is taken or not.
<gch981213>
oliv3r[m]: The assembly starts with .set noreorder, so compiler won't fill the delay slot for us, and we need to find a harmless (It's better if it's useful) instruction to be put after the branch. If there's no noreorder option, GCC can deal with the delay slot and find something to put there, and we don't need to take care of that.
<oliv3r[m]>
ahh, ok; something worth a comment :p but thanks for that information
<oliv3r[m]>
so, I have to be extra careful when looking at branches (not jumps?) and assume that while I read branch -> goto here; I should read it as; exec instruction after branch; then branch -> goto here
<gch981213>
oliv3r[m]: Correct.
<oliv3r[m]>
so ignoring the 'useful operand optimization' best is to always set 'nop'
<oliv3r[m]>
(talking about -O0 readability here)
<oliv3r[m]>
is there a name for this behavior? 'instruction after branch always executed'?
<gch981213>
I think linux removes noreorder and let compilers take care of all the delay slots.
<oliv3r[m]>
and does it execute it before, during or just after the branch? (not that it matters that much
<gch981213>
oliv3r[m]: It's called 'delay slot' :)
<oliv3r[m]>
perfect, I can add that as comment to the code :)
<oliv3r[m]>
so what about, if I have a branch, and the if the branch is not taken, basically, the 'else'. which would be considered the logical conclusion of your code. The first instruction after your branch would this be executed in both cases? That is illogical and very confusing :p
<gch981213>
oliv3r[m]: For the execution order: From the software perspective we can say the delay slot is executed after the branch instruction. It's a result of pipelined architecture so there execution overlaps a bit.
<gch981213>
oliv3r[m]: "The first instruction after your branch would this be executed in both cases?" Correct.
<oliv3r[m]>
I wonder if the original author of this .S file handcrafted it, or wrote it in C, let gcc compile it, and then hand-optimized it
<gch981213>
I believe that's why Linux removes the use of noreorder a long time ago.
<oliv3r[m]>
I think it's set to noreorder here, because it's a very tight loop/function
<gch981213>
oliv3r[m]: "branches (not jumps?) " Jumps are included.
<oliv3r[m]>
so that's why you see 'jr $ra; nop'
<gch981213>
oliv3r[m]: Yes.
<oliv3r[m]>
thank you for that lesson in mips optimizations :)
<oliv3r[m]>
that was very useful
<oliv3r[m]>
btw, do you know if the magic values for e.g. CP0_CONFIG (from mipsreg.h) are defined in the kernel? the register has a define, the content doesn't seem so
<gch981213>
you are welcome :)
<gch981213>
oliv3r[m]: No idea about that
<oliv3r[m]>
bummer, i tried grepping ... i have 'disable branch and return prediction' mtc00xc, CP0_CONFIG, 7 so I'd rather use defines for 0xc
<Ansuel>
the shell error was in macos version detection but it's not the cause
<Ansuel>
it looks like the thing is trying to prompt for a missing config not set
<Ansuel>
that loops to hell until the thing fail to build
<dhewg>
cc=xcode clang?
<Ansuel>
mhh where is that?
<dhewg>
i just see `cc`, is that xcode'c clang?
<dhewg>
sometimes its `cc`, others `gcc`, dunno if that's a hint?
<mrkiko>
dhewg: hi! Sorry for the delay. With original firmware in 7530 I get sync speed of kinda 17.2 MB/s, with openwrt/vrx511 I get around 15.5 or so, don't have exact number so may be wrong, but lower. Was wondering if you can tell me where I can look, where I might poke to try to get "back" the performance of original firmware... if at all feasible
<dhewg>
Ansuel: uboot has .azure-pipelines.yml with a tools-only macos builder, so I guess it's supposed to work? Are you using `gmake` instead of plain `make`?
<dhewg>
mrkiko: use the avm firmware? :P
<Ansuel>
dhewg the way we compile mkimage didn't change from master to 22.03 and this thing on master works correctly...
<dhewg>
I only use master, maybe it's a mkimage bump that fixed it?
<dhewg>
mrkiko: but no, I don't RE, so firmware is a blackbox to me
<Ansuel>
dhewg yes it probably is
<dhewg>
mrkiko: acoording to my notes I seem to be using:
<dhewg>
Extracted from firmware-speedport-pro-plus-v120141-5-0-018-1.img:
<slh>
mrkiko: I would let it run continuously for at least a week, before talking about performance deltas, when I was still VDSL from DTAG, every modem reboot was penalized (as the DSLAM thought it to be cable problems and reduced throughput to counter that) - just upgrading the modem firmware (so a single reboot) meant -15% for just over a week, until it recovered to full speed
hexagonwin has quit [Remote host closed the connection]
<mrkiko>
slh: thanks!!
<slh>
some ISPs are really sensitive when it comes to that (SRA/ seamless rate adaption and DLM/ Dynamic Line Management)
<slh>
for me, it usually too 12-13 days to get fully back to full speed again
<Habbie>
i guess breakage causes more support calls than 15% slowdowns
srslypascal has joined #openwrt-devel
<slh>
probably, but it's still weird if you do a modem firmware upgrade, just to come back to severely reduced link rate values (and worse, if you downgrade again, in the hope that it was just a problem with this firmware)
<philipp64>
Do all Coreboot-based machines support GPT, and if so, can I easily convert an MBR to GPT and reboot, or are there other steps?
<slh>
and 10+ days is a looong time to wait
<mrkiko>
slh: fully agree
<Ansuel>
slh tim put you on a recovery vdsl2 profile only after various disconnect and that can be recovered by calling the call center so it's not really a problem
<Ansuel>
fastweb is one isp in italy that does that sort of thing with stupid auto tune thing
<slh>
Ansuel: with DTAG, it's not really a recovery profile (that would be 16 MBit/s VDSL without vectoring), it's just a reduction of the link rate (the more frequent your disconnects, the more the penalty), but DLM is quite sensitive with DTAG
<Ansuel>
here they put you on snr 12 and have fun with it with big ginp values
<Ansuel>
but normally they also permits meme snr like 3 (some chipset can force that by faking the data sent to the dslam)
<Ansuel>
(also vectoring is not that common)
<slh>
over here in D, vectoring (profile 17b, up to 100/40 MBit/s) is very common in deployments since ~2016 (and they've invested a lot into outdoor DSLAMs), super-vectoring (profile 35b, up to 250/40 MBit/s) is less common (only in very densely populated areas)
<Mangix>
Vdsl2 is still used?
<slh>
...but I'm on ftth for 2.5 years now and yould put VDSL to rest in mid 2021
<Ansuel>
here it's almost everywhere 35b and only in some city/place with vectoring
<Ansuel>
Mangix a lot
<slh>
yep
<Ansuel>
italy did a big jump from 10m adsl2+ to vdsl2+ by just making each telephon cabinet to a minidslam
<Ansuel>
and bam everyone is at 100/200m for most of the place
<slh>
yep, similar here, since 2016
<Mangix>
only 1 ISP locally does vdsl2
<Mangix>
two others are docsis
<Ansuel>
docsis i hate them
<slh>
there is no TV cable in the ground, only phone cables and (since 2020) fibre
<Ansuel>
same here in italy tv is with antenna
<Mangix>
what's wrong with docsis?
<slh>
cable TV ends ~900m away from here (only the older town centre)
<Ansuel>
Mangix i see them as too restrictive as you can use proprietary device
<Mangix>
you mean modem?
<Ansuel>
yes
<Mangix>
here, vdsl2 is the restrictive thing. moden/router with no bridge mode
<Mangix>
docsis modems at least have bridge mode
<Ansuel>
here we made some law to stop isp not providing configuration data
<slh>
VDSL2: own modem (you can rent it, but can supply your own, dual-stack with public IPv4 and dynamic (always changing) /56 IPv6 prefix
<Mangix>
the one here uses 6in4 IIRC
<slh>
ftth: ISP GPON ONT (by law, they are required to allow you changing it, but that's ...difficult), router whatever you want (they offer you a rented device, but are fine with anything, MAC locked for ~6-10 hours), cgNAT IPv4 address, semi-static /56 IPv6 prefix
<Mangix>
my ISP has native
<Ansuel>
cgnat ouch
<Ansuel>
here ftth is really isp taste... they can give you sfp module but normally they just put ont and attach whatever you want as router
<slh>
yeah, but I can get along quite well with IPv6 for everything where it matters (wireguard over IPv6)
<slh>
well, my ftth ISP has only 16'384 IPv4 addresses for their 1.5 million customers...
<slh>
the curse of all new ISPs, which came (too) late to the party, so cable- and fibre ones in particular
<Ansuel>
the situation will get worse and worse with isp start using 4in6 LOL
<slh>
I'm quite happy that all mobile ISPs have IPv6 over LTE and 5g, so the only problem would be private or commercial WLANs without IPv6 - other than that, I do get along with IPv6-only for anthing incoming
<slh>
it was a hard decision to take ftth with cgNAT, but it mostly works out (thanks to IPv6)
<Ansuel>
/bin/sh: line 0: [: too many arguments
<Ansuel>
FKING HELLLLLLLLLLLLLLLLLLL 2 HOURS I'M TRYING TO UNDERSTAND WHY
<Mangix>
what is that for?
<Ansuel>
darwin condition for uboot 2022.01
<Ansuel>
os_x_before is bugged
<Mangix>
link?
<Ansuel>
this is also present in recent version i assume
<Ansuel>
i should really sort the thing and just hardcode everything to openwrt container
<hurricos>
For broadcasting service records, does anyone have an alternative to avahi-daemon/
<hurricos>
Avahi is only publishing my printer once
<hurricos>
refuses to do so again
<hurricos>
I might be miunderstanding -- Avahi has a reflector option and it might just want the service to ask it to re-send the broadcast when needed
<Habbie>
i am also interested as i've run into avahi doing the wrong thing six times this month
<hurricos>
on the other hand, avahi supports service files, and I've given it one, and it published it once.
aiyion_ has joined #openwrt-devel
<Habbie>
(a different thing from yours, to be clear)
<hurricos>
I'm happy restarting avahi-daemon every 30 seconds if I need to
aiyion has quit [Remote host closed the connection]