[mythtv-users] DMA Read errors and inter chipset
Michael T. Dean
mtdean at thirdcontact.com
Wed Aug 18 20:22:57 EDT 2004
On 08/18/2004 06:36 PM, MythTV wrote:
>Well Murphy came calling today after my 10+ gig recording marathon yesterday
>(recorded constantly from 11am til 6pm) Now today I can not even record..
>damn I thought I had this licked...
>I now get:
>Aug 18 04:19:12 mythtv kernel: APIC error on CPU1: 08(08)
>Aug 18 04:19:13 mythtv kernel: APIC error on CPU0: 04(04)
>which is why everything is freaking out again.. well back to the drawing
>board. (until I can actually buy a new motherboard)
Oh, you have an SMP machine...
The fix that almost worked was to disable ACPI, Local APIC, and IO-APIC
(acpi=off nolapic noapic). However, an SMP kernel always uses the local
APIC and IO-APIC in SMP mode, so it's probably ignoring your apic flags
(check dmesg boot logs to verify)--which would explain an APIC error
with APIC disabled.
APIC error 08 is "3: Receive accept error" and 04 is "Send accept error"
(see smp_error_interrupt() in arch/i386/kernel/apic.c in your kernel
source--the function just below the comment "This interrupt should never
happen with our APIC/SMP architecture" :). Error bits 0-3 (i.e. errors
0 through 8) normally indicate hardware problems (i.e. poor board
design, inadequate power supply, overheating, running SMP with
processors that aren't validated for SMP, etc.).
Unfortunately, you really have a multi-processor machine, and disabling
a processor would be a pretty big hit to the system. If you had a
uni-processor (UP) machine or you had a PIV with SMT (I refuse to use
Intel's trademarked name for a technology that had been used by the
industry for years before Intel started using it, but in the interest of
clarity, I'm talking about "Hyped-up (simultaneous-multi-) Threading"),
and you didn't need SMT support (i.e. you're not doing high-def), I
would have recommended trying a UP kernel (or booting with the "nosmp"
Since you pretty much need SMP support (to prevent wasting half your
machine), things get more difficult. In the short term, I would try
booting with "nosmp" (which means you'll only use one processor) and see
how that helps stability. For long term, I suggest you start reading
http://www.tldp.org/HOWTO/SMP-HOWTO.html , and especially see if any of
the workarounds in http://www.tldp.org/HOWTO/SMP-HOWTO-4.html apply.
More information about the mythtv-users