Filesystem corruption and bad perfomance with SRCS16 and PAE ( raid 5 2TB)

[Scott Long]

John Baldwin wrote:
> On Wednesday 07 February 2007 04:58:17 am Daniel O'Connor wrote:
>> On Wednesday 07 February 2007 19:23, Artem Kuchin wrote:
>>> FreeBSD/amd64 is a very young platform on FreeBSD. While the core FreeBSD
>>> kernel and base system components are generally fairly robust, there are
>>> likely to still be rough edges, particularly with third party packages."
>>>
>>> scares me. Do you really think it is better than PAE?
>> PAE is quite young as well, I think it was committed to the tree around March 
>> 2003. The earliest AMD64 commit I could find was May 2003 although repo 
>> copying makes it confusing..
>>
>> I think you'll find the list of drivers incompatible with PAE to be much 
>> longer with amd64.
> 
> Err, amd64 and PAE are the same problem for drivers (dealing with 64-bit
> physical addresses for DMA which can require bounce buffering if your
> hardware only supports 32-bit physical addresses).  The fix is to use
> the bus_dma abstraction in the driver instead of directly using vtophys()
> and a driver needs that fix for both PAE and amd64.  amr(4) should work
> fine with both PAE and amd64 with > 4GB of RAM.
> 

Actually, they are different problems.  There are all sorts of fun ways
for a driver to be 64-bit unclean in ways that will make it work with
amd64 but not with PAE.

Scott