I'v added a hack to TLB::checkCacheability() that sets the request flags to 
Request::UNCACHEABLE if the Paddr is in a given range.
My test program works though I'm not positive it's because of the hack or if 
the read/writes to the memory mapped region are getting flushed from the cache 
for some other reason.

I also tried placing my device at a physical address where the 
PAddrUncachedBit43 is set but this bit is lost TlbEntry.

I guess I'm looking for a graceful solution that keeps me from changing m5 
code. At this point my change set to m5 is small and I'd like to keep that way.

I haven't figured out how to enable/configure Write Combining. I've search the 
code and didn't find much of anything.

Mike

On Mar 24, 2011, at 11:31 AM, Steve Reinhardt wrote:


> SE mode has a page table and TLBs, so it shouldn't be too hard to put
> per-page uncacheable bits in there that are handled the same as in FS
> mode.  The page table structure is defined in src/mem/page_table.*.
> At runtime it's managed via the Process class, in src/sim/process.*.
> 
> Steve
> 
> On Thu, Mar 24, 2011 at 10:07 AM, Michael Levenhagen <[email protected]> 
> wrote:
>> Hello,
>> 
>> I've incorporated parts of m5 into Sandia Lab's Structural Simulation 
>> Toolkit (SST).
>> I can run a simple alpha executable in system emulation mode within SST.
>> I've also created a simple io_device and mapped it into my test apps memory 
>> space.
>> 
>> My question is how would I correctly set a region of physical memory to 
>> UNCACHEABLE with write combining?
>> I understand that given I'm running in system emulation mode what I want to 
>> do is probably not supported.
>> I'm just looking for an idea of where to start looking in the code.
>> 
>> Mike
>> _______________________________________________
>> m5-users mailing list
>> [email protected]
>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>> 
> 

_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to