finish amd64 port of DragonFly GSoC project

Matthew Dillon dillon at apollo.backplane.com
Tue May 26 13:04:56 PDT 2009


:Matthew Dillon wrote:
:>     It's looking very good so far.  After I recompiled the world the
:>     segmentation faults went away.  It looks like you implemented the
:>     entire 128TB user VM space and a 512G kernel VM space in the new
:>     amd64 pmap code?
:>   
:The constants in sys/platform/pc64/include/pmap.h indicate that 128TB is 
:available for user VM space and at most 512GB is available for the 
:kernel, but I'm not sure what is the actual limit for the kernel VM space

    It will be 64TB or so as space must be set aside for the direct physical
    memory map and device spaces.  But the 512G limit is more convenient for
    the kernel limit because it means we don't have to cycle through all
    the PML4 tables to add a new PDP for the kernel (since every user pmap
    is allocating its own PML4 table).  512G is fine, FreeBSD is doing the
    same thing.

    I wrote a little user program to test the 128TB user space, it was able
    to mmap the space fine.  Once you get the reference counting fixed up
    so the stuff can be properly freed I think we will be in real good shape.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>





More information about the Kernel mailing list