In article <20050320191346.GA27918@spathi.chuq.com>
chuq@chuq.com wrote:

> >  		if (pmap_extract(upmap, uva, &pa) == FALSE)
> >  			panic("vmapbuf: null page frame");
> >  		pmap_enter(vm_map_pmap(phys_map), kva, pa,
> > -		    VM_PROT_READ | VM_PROT_WRITE, PMAP_WIRED);
> > +		    VM_PROT_READ | VM_PROT_WRITE,
> > +		    VM_PROT_READ | VM_PROT_WRITE | PMAP_WIRED);
 :
> we shouldn't pass VM_PROT_WRITE in the pmap_enter() flags (the last arg)
> if the user memory isn't going to be modified.  the buffer's read/write flags
> are from the perspective of the device, so the user memory is only modified
> if B_READ is set.

How about this one? (taken from powerpc/vm_machdep.c)

Precisely, is it worth to pass VM_PROT_READ (and VM_PROT_WRITE on B_READ)
to pmap_enter(9) flags in vmapbuf() on mips?
And should we pass VM_PROT_WRITE in pmap_enter() prot arg only
on B_READ too?
---
Izumi Tsutsui


Index: mips/vm_machdep.c
===================================================================
RCS file: /cvsroot/src/sys/arch/mips/mips/vm_machdep.c,v
retrieving revision 1.105
diff -u -r1.105 vm_machdep.c
--- mips/vm_machdep.c	1 Jan 2005 03:25:46 -0000	1.105
+++ mips/vm_machdep.c	26 Mar 2005 02:40:27 -0000
@@ -318,6 +306,7 @@
 	vaddr_t kva;	/* Kernel VA (new to) */
 	paddr_t pa;	/* physical address */
 	vsize_t off;
+	int prot;
 
 	if ((bp->b_flags & B_PHYS) == 0)
 		panic("vmapbuf");
@@ -329,11 +318,12 @@
 	bp->b_data = (caddr_t)(kva + off);
 
 	upmap = vm_map_pmap(&bp->b_proc->p_vmspace->vm_map);
+	prot = VM_PROT_READ | ((bp->b_flags & B_READ) ? VM_PROT_WRITE : 0);
 	do {
 		if (pmap_extract(upmap, uva, &pa) == FALSE)
 			panic("vmapbuf: null page frame");
 		pmap_enter(vm_map_pmap(phys_map), kva, pa,
-		    VM_PROT_READ | VM_PROT_WRITE, PMAP_WIRED);
+		    prot, prot | PMAP_WIRED);
 		uva += PAGE_SIZE;
 		kva += PAGE_SIZE;
 		len -= PAGE_SIZE;