[Bloat] cake 'conservative' option and panics

[Bloat] cake 'conservative' option and panics

[Kevin Darbyshire-Bryant]

[-- Attachment #1: Type: text/plain, Size: 567 bytes --]

Hi All,

I had a 'fun' moment a couple of days ago where I tried the
'conservative' option in the sqm-scripts advanced options and found my
router went into crash/reboot cycle.  What I should have done was fiddle
with 'conservative' live, so lesson learned there.  Anyway, anyone seen
this?  OpenWrt trunk on Archer C7 v2, with tc-adv, kmod-sched-pie &
kmod-sched-cake from Dave's ceropackages repo (I *think* that's the
latest/current one but.......)

Something like 'tc qdisc change dev eth0 cake conservative' makes my
router go bye bye :-)

Kevin


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4764 bytes --]

Re: [Bloat] cake 'conservative' option and panics

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 394 bytes --]

I tried to get OpenWRT on my router last week, but ran into some
roadblocks, so I'm not in an ideal position to help debug it.  I'm pretty
sure it works properly on desktop hardware.

However, the configuration interface should be robust enough to deal with
mismatches between userspace and module versions.  I'm having difficulty
imagining why you're seeing such a problem.

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 471 bytes --]

Re: [Bloat] cake 'conservative' option and panics

[Kevin Darbyshire-Bryant]

[-- Attachment #1: Type: text/plain, Size: 2217 bytes --]

On 10/08/15 13:16, Jonathan Morton wrote:
>
> I tried to get OpenWRT on my router last week, but ran into some
> roadblocks, so I'm not in an ideal position to help debug it.  I'm
> pretty sure it works properly on desktop hardware.
>
I'd be interested to know what issues you hit?  Was it anything related
to the instructions I knocked up here:
http://www.bufferbloat.net/projects/codel/wiki/Cake#Installing-CAKE-out-of-tree-on-OpenWrt-rough-instructions

I'm quite concerned if these instructions contain an error as I don't
want others falling into traps I've created.
>
> However, the configuration interface should be robust enough to deal
> with mismatches between userspace and module versions.  I'm having
> difficulty imagining why you're seeing such a problem.
>
Ha ha :-)  I'm not so surprised, the application of a suitably placed
idiot (ie. me) can reveal all sorts. I'm sure it's something I'm doing.

Thoughts/observations: Archer C7 Openwrt has recently bumped to 4.1.4
kernel.  No idea if that affects the module interface (I rebuild
everything and tell openwrt to include everything in the image rather
than using opkg to install additional packages, so I don't have a build
date/package date mismatch)

I'm picking up 'tc-adv' & 'kmod-sched-cake' from Dave's ceropackages
https://github.com/dtaht/ceropackages-3.10.  These point to heads of
https://github.com/dtaht/tc-adv and  https://github.com/dtaht/sch_cake
respectively.  I'm assuming these are the best places to go for
'current' versions of patched tc & cake, am I wrong?

I noticed that sch-cake has a recent commit
https://github.com/dtaht/sch_cake/commit/7a8d833f9174a9662a5d95baac870eebd9b9d94f
which exposes the number of active flows into
struct tc_cake_xstats.  tc-adv hasn't been changed to report that
value...so is this up to date?  Fortunately the structure size doesn't
change but I'm curious as to how pkt_sched.h should be maintained
between kmod-sched-cake & tc-adv?

If someone can give me a nudge/help into what to do I'll happily give
things a go.

It would be good if I could get a crash dump or even console log out of
the router....hmm no idea how to do that!

Kevin




[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4764 bytes --]

Re: [Bloat] cake 'conservative' option and panics

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 257 bytes --]

I have an OpenWRT image built, but the device refuses to boot even the
official pre-built image.  Luckily it retains the stock firmware, so it's
still usable.

It is the Buffalo WBMR-...-G300H, if you're interested in armchair
debugging.

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 325 bytes --]

Re: [Bloat] cake 'conservative' option and panics

[Kevin Darbyshire-Bryant]

[-- Attachment #1: Type: text/plain, Size: 899 bytes --]

On 10/08/15 20:28, Jonathan Morton wrote:
>
> I have an OpenWRT image built, but the device refuses to boot even the
> official pre-built image.  Luckily it retains the stock firmware, so
> it's still usable.
>
That suggests the image isn't being flashed completely/properly.  Hmm!
>
> It is the Buffalo WBMR-...-G300H, if you're interested in armchair
> debugging.
>
Urgh, from http://wiki.openwrt.org/toh/buffalo/wbmr-hp-g300h I notice it
has a non-straightforward flashing procedure to get the first load of
openwrt on it, something about encrypted firmware.  There's talk of
'option 1' which uses dd-wrt as a stepping stone to command line access
to then overwrite with an openwrt image, I have to say that strikes me
as easier than 'option 2' which involves tftp servers, various button
presses, no doubt some magic incantations....uuurgh.




> - Jonathan Morton
>



[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4764 bytes --]

Re: [Bloat] cake 'conservative' option and panics

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 269 bytes --]

Actually, option 2 seemed cleaner to me, so that's what I was using.  The
first firmware image just goes into RAM and serves as a stable environment
from which to flash the full image.  In theory.  In fact it just turned all
its lights off and hung.

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 318 bytes --]