Re: Blade 2000 stability

Martin Husemann wrote:
> On Mon, Jul 15, 2013 at 03:23:41PM +0200, BERTRAND Joël wrote:
>> 	Hello,
>>
>> 	I have installed NetBSD (from daily snapshot) on a Blade 2000. I
>> 	have seen that Blade 2000 (with USIII) was stable, but I have seen a lot of
>> stability issues. Randomly, this server hangs without any message (log
>> or console). Kernel doesn't panic, but seems to enter in a deadlock
>> (console doesn't respond anymore, network connections are closed...).
>
> There are some PCI root bridge issues still unresolved, but they usually
> cause the machine to reset instead of hanging.

	I haven't seen any PCI root bridge issues (with USIII+ CPU, I don't 
have USIIIi).

> You could try a kernel with DIAGNOSTIC/DEBUG/LOCK_DEBUG added (slow!), and
> see if this catches anything. Do you have a special workload to trigger the
> hang?

	I have investigated since my first mail and I think that memory was 
defective. This server ran with 8 * 1 GB. When I have replaced this 
memory by 8 * 256 MB, it runs fine.

>> 	Other question : does NetBSD adjust CPU frequency ? I have found
>> information about acpi drivers, but nothing about a sparc64 driver.
>
> No, is it even possible on that hardware?

	It is possible (Linux can slow down CPU when it is idle). CPU can run 
at nominal frequency, half and 1/32 frequency. I'm trying to patch 
kernel to add cpufreq data in cpu_info structure.

> Also make sure you have Julian's latest changes to dev/tda.c, it will make
> an audible difference on Blade 1000/2000.

	Yes I know ;-)

	Regards,

	JKB