MAXPHYS and physical memory (Was: Re: siis/atacam/ata/gmirror 8.0-BETA3 disk performance)

[Alexander Motin]

Mel Flynn wrote:
> On Wednesday 02 September 2009 17:46:46 Alexander Motin wrote:
>> Mel Flynn wrote:
>>> On Wednesday 02 September 2009 16:51:35 Alexander Motin wrote:
>>>> For maximum linear I/O performance you may want to build kernel with
>>>> options 	MAXPHYS=(1024*1024)
>>> I've found that just doubling the default MAXPHYS already panics-on-boot
>>> a 1.5GB i386 system. Is there any reasonable conversion table for MAXPHYS
>>> to physical memory, since various memory related kernel setups are
>>> derived from or calculated with MAXPHYS?
>> What especially your panic was about? It could be bug in ATA(4) or some
>> other code, that does not handle MAXPHYS correctly. I don't think that
>> you could reach memory limit during simple system boot because of that.
>> I am successfully running my testing Pentium-75 with 64MB RAM with 1MB
>> MAXPHYS.
>>
>> Could you show your panic message?
> 
> It's been a while since I last tried, during 7.1-STABLE. Two different 
> machines wouldn't boot, both 1.5G; loading kernel.old, then commenting the 
> MAXPHYS change, rebuilding kernel made things work, so I stopped pursuing this 
> at the time.
> I remember the panic message quite early page fault, one of the first drivers 
> loaded (ahci or acpi). Since this is supposed to work, I'm going to pursue it 
> further and see if I can still reproduce it.

I have fixed MAXPHYS issue in ata-ahci in RELENG_8, but I haven't merged 
it lower as code is quite different. So it would be nice if you can 
check it on 8.x.

-- 
Alexander Motin