[mythtv] LFireDev Error: Starting A/V streaming: Cannot allocate memory

[Bruce Markey]

Kelly Campbell wrote:
> I think I found the cause but haven't been able to confirm the fix yet.
> 
> https://bugs.launchpad.net/bugs/148854

Yeah, this seems to be headed in the right direction.

I've been using firewire captures for awhile. Having had a
leaning curve to overcome or workaround a lot of gotchas,
my impression was it is just a little too flaky to be
trusted for day to day use.

I don't recall "eno: Cannot allocate memory (12)" as being
a problem when I first started but this has been a common
occurrence over the past several weeks. Also, I had an earlier
impression that I needed to run "firewire_tester -p -n 0 -R"
after rebooting and after the connection was reset once, the
connection would remain good until the next reboot. This too
has not been true over the past several weeks but I was unaware
how or if these were related.

In some brief stabs at looking through the code, it appeared
that the errno 12 wasn't from myth or libiec61883 but from
libraw1394 where it was sizing the reads relative to the vm
page size. However, it didn't make sense why it would throw
this error. What I hadn't looked for was the very conspicuous
"dma_region_alloc: vmalloc_32() failed" messages in syslog.
These were enlightening.

Not only did I find these at times when recordings failed with
errno 12 but there were also clusters of errors grouped together
every two seconds. There were the result of this:

#!/bin/sh
#
# fw: script to attempt to reset firewire connections

firewire_tester -p -n 0 -r 3
if [ $? == 1 ]
then
    firewire_tester -p -n 0 -R
    firewire_tester -p -n 0 -r 3
fi

I'd found that running "-R" when the connection was good would
result in breaking it again so this just resets and retests if
it fails. I wasn't aware, however, that firewire_tester could
fail due to the exact same vmalloc issue. These failures are not
a network issue with the STB but false negatives due to the
kernel bug.

Oct  3 22:42:26 bigboy kernel: dma_region_alloc: vmalloc_32() failed
Oct  3 22:42:28 bigboy kernel: dma_region_alloc: vmalloc_32() failed
Oct  3 22:42:30 bigboy kernel: dma_region_alloc: vmalloc_32() failed
Oct  3 22:42:32 bigboy kernel: dma_region_alloc: vmalloc_32() failed
Oct  3 22:42:34 bigboy kernel: dma_region_alloc: vmalloc_32() failed
Oct  3 22:42:36 bigboy kernel: dma_region_alloc: vmalloc_32() failed

These were the result of my firewire_tester script failing.

Oct  3 22:59:56 bigboy kernel: dma_region_alloc: vmalloc_32() failed

This was mythbackend trying to start SportsCenter

Oct  3 23:00:59 bigboy kernel: dma_region_alloc: vmalloc_32() failed

This was my restart of mythbackend and it failed to record a
second time. I restarted again and the recording worked on the
third attempt.

So, I rebooted a kernel last night with this patch from the
original author at " http://lkml.org/lkml/2007/7/18/34 ". I
have done several firewire captures including back to back
recordings without fail. I've also run my "fw" many times
and it never throws a vmalloc error.

: root at bigboy \# ; grep -c dma_region_alloc /var/log/syslog.0
28
: root at bigboy \# ; grep -c dma_region_alloc /var/log/syslog
0

--  bjm