Progress on Raspberry Pi

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

Progress on Raspberry Pi

George Mitchell
My AdaFruit serial cable arrived, so now I can gather useful information
when my RPi crashes (which is still happening fairly often as I try to
build print/cups, having started five days ago and still not having
gotten through ghostscript).

I built my image with crochet on an amd64 running r255361, and the image
itself is based on that version.  (However, in sys/arm/arm I updated
pmap-v6.c to 255612 and stdatomic.c to 255613.  I see they have been
updated again within the last day; perhaps I should try again.)

/etc/src.conf:
MALLOC_PRODUCTION=yes

/etc/make.conf:
WITH_PKGNG=yes
MALLOC_PRODUCTION=yes
# added by use.perl 2013-08-20 10:11:49
PERL_VERSION=5.14.4

I commented out the five syscons-related lines in sys/arm/conf/RPI-B so
I could use the serial port.  My SDHC card is apparently on the hairy
edge of working, so I definitely needed /boot/loader.conf to contain:
hw.bcm2835.sdhci.hs=0.  Before I added that, I ended up with the dreaded
error 19 and a manual mountroot prompt, at which point mountroot would
accept a character from the serial input about 1/4 of the time.  Other
serial input works without a problem.

Before I remembered to comment out the /dev/ttyvN lines in /etc/tty, I
naturally saw getty errors on the nonexistent devices, but I would also
randomly get an "Interrupted system call" on /dev/ttyu0, followed by a
long pause, followed (finally) by a login prompt.  After I commented out
the ttyvN lines, the "Interrupted system call" error on /dev/ttyu0 went
away.

Questions:

1. Do we have any sort of a memory test for the RPi?  It seems unlikely
that I have a hardware problem, but I would like to rule it out.

2. Now that I can do something when I get the prefetch abort or panic,
what should I do at the "db>" prompt to help debug it?  It's happening
about five or six times a day.

Again, my thanks to all the developers!                     -- George
_______________________________________________
[hidden email] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-arm
To unsubscribe, send any mail to "[hidden email]"
Reply | Threaded
Open this post in threaded view
|

Re: Progress on Raspberry Pi

Luiz Otavio O Souza-4
On 20 September 2013 21:49, George Mitchell <[hidden email]> wrote:

> My AdaFruit serial cable arrived, so now I can gather useful information
> when my RPi crashes (which is still happening fairly often as I try to
> build print/cups, having started five days ago and still not having
> gotten through ghostscript).
>
> I built my image with crochet on an amd64 running r255361, and the image
> itself is based on that version.  (However, in sys/arm/arm I updated
> pmap-v6.c to 255612 and stdatomic.c to 255613.  I see they have been
> updated again within the last day; perhaps I should try again.)
>
> /etc/src.conf:
> MALLOC_PRODUCTION=yes
>

MALLOC_PRODUCTION=yes is just a workaround for the RPi problem, in a ideal
world (where everything works) we  wouldn't need to set it.


>
> /etc/make.conf:
> WITH_PKGNG=yes
> MALLOC_PRODUCTION=yes
> # added by use.perl 2013-08-20 10:11:49
> PERL_VERSION=5.14.4
>
> I commented out the five syscons-related lines in sys/arm/conf/RPI-B so
> I could use the serial port.  My SDHC card is apparently on the hairy
> edge of working, so I definitely needed /boot/loader.conf to contain:
> hw.bcm2835.sdhci.hs=0.  Before I added that, I ended up with the dreaded
> error 19 and a manual mountroot prompt, at which point mountroot would
> accept a character from the serial input about 1/4 of the time.  Other
> serial input works without a problem.
>

Can you check if the patch on arm/179688 (
http://www.freebsd.org/cgi/query-pr.cgi?pr=arm/179688) fix the mountroot
input problem for you ?


>
> Before I remembered to comment out the /dev/ttyvN lines in /etc/tty, I
> naturally saw getty errors on the nonexistent devices, but I would also
> randomly get an "Interrupted system call" on /dev/ttyu0, followed by a
> long pause, followed (finally) by a login prompt.  After I commented out
> the ttyvN lines, the "Interrupted system call" error on /dev/ttyu0 went
> away.
>

This is added by crochet, isn't ? (boards/RaspberryPi/overlay/etc/ttys)


>
> Questions:
>
> 1. Do we have any sort of a memory test for the RPi?  It seems unlikely
> that I have a hardware problem, but I would like to rule it out.


> 2. Now that I can do something when I get the prefetch abort or panic,
> what should I do at the "db>" prompt to help debug it?  It's happening
> about five or six times a day.
>
>
We have a known bug that triggers under heavy load (during builds):
http://www.freebsd.org/cgi/query-pr.cgi?pr=arm/182060

We need to check (now that gcc builds should be working again for arm) if
there is any difference between the builds with clang and gcc, oabi and
eabi.

Thanks,
Luiz
_______________________________________________
[hidden email] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-arm
To unsubscribe, send any mail to "[hidden email]"
Reply | Threaded
Open this post in threaded view
|

Re: Progress on Raspberry Pi

George Mitchell
On 09/20/13 21:16, Luiz Otavio O Souza wrote:

> On 20 September 2013 21:49, George Mitchell <[hidden email]
> <mailto:[hidden email]>> wrote:
>
>     My AdaFruit serial cable arrived, so now I can gather useful information
>     when my RPi crashes (which is still happening fairly often as I try to
>     build print/cups, having started five days ago and still not having
>     gotten through ghostscript).
>
>     I built my image with crochet on an amd64 running r255361, and the image
>     itself is based on that version.  (However, in sys/arm/arm I updated
>     pmap-v6.c to 255612 and stdatomic.c to 255613.  I see they have been
>     updated again within the last day; perhaps I should try again.)
>
>     /etc/src.conf:
>     MALLOC_PRODUCTION=yes
>
>
> MALLOC_PRODUCTION=yes is just a workaround for the RPi problem, in a
> ideal world (where everything works) we  wouldn't need to set it.
>
>
>     /etc/make.conf:
>     WITH_PKGNG=yes
>     MALLOC_PRODUCTION=yes
>     # added by use.perl 2013-08-20 10:11:49
>     PERL_VERSION=5.14.4
>
>     I commented out the five syscons-related lines in sys/arm/conf/RPI-B so
>     I could use the serial port.  My SDHC card is apparently on the hairy
>     edge of working, so I definitely needed /boot/loader.conf to contain:
>     hw.bcm2835.sdhci.hs=0.  Before I added that, I ended up with the dreaded
>     error 19 and a manual mountroot prompt, at which point mountroot would
>     accept a character from the serial input about 1/4 of the time.  Other
>     serial input works without a problem.
>
>
> Can you check if the patch on arm/179688
> (http://www.freebsd.org/cgi/query-pr.cgi?pr=arm/179688) fix the
> mountroot input problem for you ?

I'll try that.

>
>
>     Before I remembered to comment out the /dev/ttyvN lines in /etc/tty, I
>     naturally saw getty errors on the nonexistent devices, but I would also
>     randomly get an "Interrupted system call" on /dev/ttyu0, followed by a
>     long pause, followed (finally) by a login prompt.  After I commented out
>     the ttyvN lines, the "Interrupted system call" error on /dev/ttyu0 went
>     away.
>
>
> This is added by crochet, isn't ? (boards/RaspberryPi/overlay/etc/ttys)
>
>
>     Questions:
>
>     1. Do we have any sort of a memory test for the RPi?  It seems unlikely
>     that I have a hardware problem, but I would like to rule it out.
>
>
>     2. Now that I can do something when I get the prefetch abort or panic,
>     what should I do at the "db>" prompt to help debug it?  It's happening
>     about five or six times a day.
>
>
> We have a known bug that triggers under heavy load (during builds):
> http://www.freebsd.org/cgi/query-pr.cgi?pr=arm/182060
>
> We need to check (now that gcc builds should be working again for arm)
> if there is any difference between the builds with clang and gcc, oabi
> and eabi.
>
> Thanks,
> Luiz

Back in January, gcc worked fine.  More recently, not so fine, and I
finally gave up and plunged into the clang/EABI world.  It mostly works
aside from my crashes.                                      -- George

_______________________________________________
[hidden email] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-arm
To unsubscribe, send any mail to "[hidden email]"
Reply | Threaded
Open this post in threaded view
|

Re: Progress on Raspberry Pi

Tim Kientzle-2
In reply to this post by George Mitchell

On Sep 20, 2013, at 5:49 PM, George Mitchell <[hidden email]> wrote:

> My AdaFruit serial cable arrived, so now I can gather useful information
> when my RPi crashes…

Excellent!

> I built my image with crochet on an amd64 running r255361, and the image
> itself is based on that version.  (However, in sys/arm/arm I updated
> pmap-v6.c to 255612 and stdatomic.c to 255613.  I see they have been
> updated again within the last day; perhaps I should try again.)
>
> /etc/src.conf:
> MALLOC_PRODUCTION=yes
>
> /etc/make.conf:
> WITH_PKGNG=yes
> MALLOC_PRODUCTION=yes
> # added by use.perl 2013-08-20 10:11:49
> PERL_VERSION=5.14.4

By default, Crochet ignores the host /etc/src.conf
and /etc/make.conf on the assumption that what
is right for the host system is not necessarily correct
for the target system.

If you specifically want to use them, you can add

SRCCONF=/etc/src.conf
__MAKE_CONF=/etc/make.conf

to your configuration file.  Of course, you can also put
src.conf and make.conf files elsewhere for Crochet to use.

> I commented out the five syscons-related lines in sys/arm/conf/RPI-B so
> I could use the serial port.  My SDHC card is apparently on the hairy
> edge of working, so I definitely needed /boot/loader.conf to contain:
> hw.bcm2835.sdhci.hs=0.  Before I added that, I ended up with the dreaded
> error 19 and a manual mountroot prompt, at which point mountroot would
> accept a character from the serial input about 1/4 of the time.  Other
> serial input works without a problem.

<sigh>  Keyboard input to mount root seems to
be broken on a lot of FreeBSD systems.

> Before I remembered to comment out the /dev/ttyvN lines in /etc/tty,…

Should Crochet's RaspberryPi configuration disable those?

> 2. Now that I can do something when I get the prefetch abort or panic,
> what should I do at the "db>" prompt to help debug it?  It's happening
> about five or six times a day.

Type "bt" to get a backtrace.  That's the single most informative
thing.  Note:  The first few stack frames displayed are from inside
the debugger itself and aren't particularly interesting.  The important
frames are a little later where we see *why* we ended up in the
debugger.

Tim

_______________________________________________
[hidden email] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-arm
To unsubscribe, send any mail to "[hidden email]"
Reply | Threaded
Open this post in threaded view
|

Re: Progress on Raspberry Pi

George Mitchell
On 09/21/13 14:25, Tim Kientzle wrote:

>
> On Sep 20, 2013, at 5:49 PM, George Mitchell <[hidden email]> wrote:
>
>> My AdaFruit serial cable arrived, so now I can gather useful information
>> when my RPi crashes…
>
> Excellent!
>
>> I built my image with crochet on an amd64 running r255361, and the image
>> itself is based on that version.  (However, in sys/arm/arm I updated
>> pmap-v6.c to 255612 and stdatomic.c to 255613.  I see they have been
>> updated again within the last day; perhaps I should try again.)
>>
>> /etc/src.conf:
>> MALLOC_PRODUCTION=yes
>>
>> /etc/make.conf:
>> WITH_PKGNG=yes
>> MALLOC_PRODUCTION=yes
>> # added by use.perl 2013-08-20 10:11:49
>> PERL_VERSION=5.14.4
>
> By default, Crochet ignores the host /etc/src.conf
> and /etc/make.conf on the assumption that what
> is right for the host system is not necessarily correct
> for the target system.
>
> If you specifically want to use them, you can add
>
> SRCCONF=/etc/src.conf
> __MAKE_CONF=/etc/make.conf
>
> to your configuration file.  Of course, you can also put
> src.conf and make.conf files elsewhere for Crochet to use.

Thanks!  I put that in my crochet config file just a couple of weeks
ago when you first mentioned it.

>
>> I commented out the five syscons-related lines in sys/arm/conf/RPI-B so
>> I could use the serial port.  My SDHC card is apparently on the hairy
>> edge of working, so I definitely needed /boot/loader.conf to contain:
>> hw.bcm2835.sdhci.hs=0.  Before I added that, I ended up with the dreaded
>> error 19 and a manual mountroot prompt, at which point mountroot would
>> accept a character from the serial input about 1/4 of the time.  Other
>> serial input works without a problem.
>
> <sigh>  Keyboard input to mount root seems to
> be broken on a lot of FreeBSD systems.
>
>> Before I remembered to comment out the /dev/ttyvN lines in /etc/tty,…
>
> Should Crochet's RaspberryPi configuration disable those?

Not sure.  Before I got the serial cable hooked up, it was nice to be
able to use the USB keyboard to log into the system.

>
>> 2. Now that I can do something when I get the prefetch abort or panic,
>> what should I do at the "db>" prompt to help debug it?  It's happening
>> about five or six times a day.
>
> Type "bt" to get a backtrace.  That's the single most informative
> thing.  Note:  The first few stack frames displayed are from inside
> the debugger itself and aren't particularly interesting.  The important
> frames are a little later where we see *why* we ended up in the
> debugger.
>
> Tim
>

Okay, will do.  I'll try to collect a few and maybe post them in the
few days.

Any predictions of what will happen if I try to compile and install
x11/xorg on the RPi?                                     -- George
_______________________________________________
[hidden email] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-arm
To unsubscribe, send any mail to "[hidden email]"
Reply | Threaded
Open this post in threaded view
|

Re: Progress on Raspberry Pi

George Mitchell
On 09/21/13 15:01, George Mitchell wrote:

> On 09/21/13 14:25, Tim Kientzle wrote:
>>[...]
>>> 2. Now that I can do something when I get the prefetch abort or panic,
>>> what should I do at the "db>" prompt to help debug it?  It's happening
>>> about five or six times a day.
>>
>> Type "bt" to get a backtrace.  That's the single most informative
>> thing.  Note:  The first few stack frames displayed are from inside
>> the debugger itself and aren't particularly interesting.  The important
>> frames are a little later where we see *why* we ended up in the
>> debugger.
>>
>> Tim
>>
>
> Okay, will do.  I'll try to collect a few and maybe post them in the
> few days.
>
> Any predictions of what will happen if I try to compile and install
> x11/xorg on the RPi?                                     -- George

It appears that "t" is the command to use here.  The first one is in:

http://www.m5p.com/~george/pi/trace1.txt

along with kernel.tbz, which includes my /boot/kernel/kernel.symbols,
/boot/kernel/kernel, and kernel/kernel.gz.tramp (don't know what that
is, but my tar command picked it up).  Perhaps that will help decipher
the addresses in the backtrace.  This one occurred while I was
re-running "portmaster -BDg devel/glib20" under the script command to
see why the whole "gio" include directory isn't getting installed.
Using MALLOC_PRODUCTION=yes doesn't seem to have cured this problem
-- George
_______________________________________________
[hidden email] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-arm
To unsubscribe, send any mail to "[hidden email]"