Garrett D'Amore wrote:
> [EMAIL PROTECTED] wrote:
>> Well, after messing about with it a bit, the best I can figure is the 
>> thing is not deterministic.  I was trying to see if between the PnP 
>> and Vzn settings in the BIOS, I could get it to hang.  It's booting 
>> every time now.  I did select HW defaults at one point from the BIOS, 
>> so maybe that changed something else I didn't spot.
>
> The switching back and forth in PnP is *meaningless*.  Because, there 
> is some bug in the Toshiba BIOS were changing the value doesn't really 
> take effect unless a full power down is done.  This includes removing 
> the battery physically from the system.  The Toshiba M9 BIOS is kinda 
> funky here, and I've had a heard time qualifying this properly, but 
> lets just say that after switching to All, then back to PnP OS, the 
> problem I had with my particular device's BIOS settings went away.

Well, this may explain what I'm seeing then.  I initially had hangs 
during xVM dom0 boot, but ever since messing with the BIOS it boots.  
One other Sun colleague, George Asaad, still has one in a hanging 
state.  His is hanging like mine was, loading that ether_mac module.  
Now that I know xVM doesn't support MSI (thanks to Stuart's email), it's 
probably trying to initialize the device and never hearing back.

Why it should go further after messing with the BIOS, I don't know.  
Maybe the device is further initialized to some point?  We're at the 
edges of my understanding of how such things work...

> Until I allowed the battery to drain fully.  Then mysteriously it 
> reappeared, causing me about 3 days of painful debugging trying to 
> find a problem I thought was in my driver (SDcard), but which 
> ultimately boiled down to this weird BIOS thing.  Setting the value 
> back to "All" cleared my problems up.

Had you raised this issue with Toshiba?  I think I know some folks who 
might be able to do so if you've not yet, but I don't want to cause any 
unnecessary noise (unless you think that's warranted as well).


- Matt

p.s.: I couldn't get my SDCard to work in either case, but I didn't mess 
with it that much.  I probably missed something simple.

>
>    -- Garrett
>>
>> virt-install prompts for HVM if Vzn is turned on -OR- off in the 
>> BIOS.  I don't know what's going on there.
>>
>> More below...
>>
>> Juergen Keil wrote:
>>> Matt Ingenthron wrote:
>>>  
>>>> Garrett D'Amore wrote:
>>>>   
>>>>> Matt Ingenthron wrote:
>>>>>     
>>>>>> I'll do a live upgrade to 82, and hopefully that'll "just work" 
>>>>>> for me too.  Thanks for looking into it Garrett.
>>>>>>
>>>>>> If you happen to think of any modifications you've made to the 
>>>>>> BIOS or anything like that, please let me know.  I assume there 
>>>>>> are none and it's just bugs that have been squashed post build 79b.
>>>>>>         
>>>>> The one thing that I've done is change the BIOS PnP OS setting so 
>>>>> that BIOS configures all devices.  That is necessary for the 
>>>>> SDcard stuff.  I don't think it makes a difference otherwise, but 
>>>>> maybe its something that Xen^WxVM is sensitive to.
>>>>>       
>>>> Is that the "Device Configuration", where the choices are "Setup by 
>>>> OS" or "All"?  Why can't they just call it PnP?
>>>>
>>>> I just changed that, and sure enough, I've booted under xVM.
>>>>     
>>>
>>> Hmm,  is there any other driver sharing the interrupt vector with
>>> the e1000g device?  Run mdb -k and have a look at the ::interrupts
>>> output, when not booted under xVM and with the BIOS PnP setting
>>> reverted:
>>>
>>>   echo ::interrupts | mdb -k
>>>   echo ::interrupts -d | mdb -k
>>>   
>>
>> I get "NOTICE: IRQ xxx shared" during boot under xVM, even though I'm 
>> not booting verbose so there are some buggy things going on there.
>>>
>>> Maybe e1000g is sharing the interrupt vector with another piece of 
>>> hardware,
>>> and that other hardware has a pending interrupt during xVM dom0 
>>> boot.  e1000g driver is the first driver that is loaded, installs 
>>> it's interrupt handler, the vector is unmasked, and we immediatelly get
>>> interrupts on that vector, but they are not from e1000g.  There's
>>> no driver to clear the interrupt condition => the system hangs.
>>>   
>>
>> Hmm, this could be.
>>
>> For those I remember the name of above, the interrupts under xVM show 
>> hci1394 and uhci/ehci.
>>
>> Does xVM handle such things well?  How many of these get passed 
>> through to Solaris.  If there are any pointers to a better technical 
>> doc on how xVM works with dom0's, I'd appreciate it.  Most of the 
>> docs I see are pretty high level-- it's hard to see how much xVM does 
>> and how much is left for the dom0 to do.
>>
>> At this point, I'm tempted to leave it alone since it's working, but 
>> I'm happy to investigate interrupts if the data is valuable to someone.
>>
>>> My Tecra S1 had such an issue with the ehci and ipw drivers
>>> 6353812 "ipw driver must disable interrupts at shutdown; tecra s1 
>>> hangs on reboot"
>>>    http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6353812
>>>
>>>  
>>>> I also changed a Virtualization enabled/disabled setting which I 
>>>> did not see before.  That would seem to be important for xVM.  :)
>>>>     
>>>
>>> Yes, probable helps with HVM domUs...
>>>
>>>  
>>>> The NVidia driver is working just fine as well.
>>>>
>>>> Let me go through the matrix to see what we can figure out.  
>>>> There's either a bug or a release note to be created here I think.
>>>>     
>>>
>>>
>>>   
>>
>>
>>
>


-- 
Matt Ingenthron - Web Infrastructure Solutions Architect
Sun Microsystems, Inc. - Global Systems Practice
http://blogs.sun.com/mingenthron/
email: [EMAIL PROTECTED]             Phone: 310-242-6439

_______________________________________________
xen-discuss mailing list
[email protected]

Reply via email to