Bizarre clone attempt failures on Raspberry Pi2...

[Karl Denninger]

On 7/15/2016 09:53, Warner Losh wrote:
> I saw your failok PR. That's a good idea.
>
> A better idea, though, is to implement true dual-console mode where
> both the kernel *AND* /etc/rc run on both the serial and the video
> consoles. I've poked at this in the past, but have never had time to
> fully implement it due to other things being more pressing.
>
> Warner
Agreed.  The best option is the one that exposes what is going on in a
way that is *usable* to those in any reasonable set of expected "base"
configurations.  But for now "failok" is better than nothing.

BTW I've run into some very nasty microSD card failure modes of late on
these devices and suspect that atime is thrashing them in terms of write
endurance, resulting in the destruction of several cards after
approximately 9 months of "always on" use.  The symptom is that you wind
up with a hard error in a given location on the card and there's nothing
you can do about it; this results in repeated panics.  This has now
destroyed three Sandisk 32Gb cards here in the last two weeks, all on
different Pi2s doing different things, two of which were on
UPS-protected power and thus hadn't taken unprotected shutdowns.

As a result I am now mounting root "noatime" which should materially
decrease the write load.

For most users with these devices in common configurations that isn't
going to include a serial console.  Hell, I don't usually have one
hooked up, although I *can*, because it's just flat-out simpler to plug
in a monitor and USB keyboard.  For the common person who buys a Pi2 to
screw around they will likely not have a serial console available to
them at all.

Indeed the opposite situation exists in the Intel world; it's a royal
PITA to run serial as a primary console that "just works" on servers (in
fact I've never found the "recipe" for doing so that doesn't leave me
with an exposure where a particular mode of failure might leave me with
no way to get in and resolve whatever is broken, including BIOS access),
which is why built-in IPMI KVMs are such a big deal on those boards --
they "just work" for the "expected" circumstance that "mirrors" what the
common user expects and has.

BTW uboot is allegedly able to support a USB keyboard.  I don't know if
the version we are currently using does, but it allegedly is capable of
doing so in the current code that is out there.

In addition, how does one change the 10 second boot delay in the current
kernel to something shorter? /boot/loader.conf is ignored on ARM systems
since they're loaded with uboot, yes?

-- 
Karl Denninger
karl at denninger.net <mailto:karl at denninger.net>
/The Market Ticker/
/[S/MIME encrypted email preferred]/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 2996 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.freebsd.org/pipermail/freebsd-arm/attachments/20160715/9866410c/attachment.bin>