Hi folks...

I have a change to UVM that I need tested on a system that requires
explicit I-cache frobbing, such as the MIPS or Alpha 21164 (I can't
easily test this on a 21164 myself right now).

The deal here is that uvm_pagermapin() has been mapping pages with
VM_PROT_EXECUTE, which causes an I-cache flush when the mapping is
entered and when it is removed.  However, if the page needs execute
permissions, those will be added when the page is mapped into the
proper address space.

So, on the surface, this VM_PROT_EXECUTE in uvm_pagermapin() seems
redundant.  But I would like to verify that that is the case.

To test this, you need a Very Up To Date sys/uvm (including uvm_pager.c
rev 1.29).

Any takers?

-- 
        -- Jason R. Thorpe <thorpej@zembu.com>

Index: uvm_pager.c
===================================================================
RCS file: /cvsroot/syssrc/sys/uvm/uvm_pager.c,v
retrieving revision 1.29
diff -c -r1.29 uvm_pager.c
*** uvm_pager.c	2000/05/19 03:45:04	1.29
--- uvm_pager.c	2000/05/19 05:40:28
***************
*** 180,198 ****
  		if ((pp->flags & PG_BUSY) == 0)
  			panic("uvm_pagermapin: page not busy");
  #endif
- 
- 		/*
- 		 * XXX We used to use VM_PROT_DEFAULT here, but
- 		 * XXX we don't since we know the direction of
- 		 * XXX the I/O now.  However, VM_PROT_DEFAULT
- 		 * XXX included VM_PROT_EXECUTE.  While that could
- 		 * XXX lead to unnecessary I-cache flushes, something
- 		 * XXX in the path might rely on that being done,
- 		 * XXX so we still include it, for now.
- 		 * XXX DOUBLE CHECK THIS!
- 		 */
  		pmap_enter(vm_map_pmap(pager_map), cva, VM_PAGE_TO_PHYS(pp),
! 		    prot | VM_PROT_EXECUTE, PMAP_WIRED | prot);
  	}
  
  	UVMHIST_LOG(maphist, "<- done (KVA=0x%x)", kva,0,0,0);
--- 180,187 ----
  		if ((pp->flags & PG_BUSY) == 0)
  			panic("uvm_pagermapin: page not busy");
  #endif
  		pmap_enter(vm_map_pmap(pager_map), cva, VM_PAGE_TO_PHYS(pp),
! 		    prot, PMAP_WIRED | prot);
  	}
  
  	UVMHIST_LOG(maphist, "<- done (KVA=0x%x)", kva,0,0,0);