Subject: Re: Fault type values for uvm_fault()
To: Jason Thorpe <>
From: Matthias Drochner <>
List: port-arm
Date: 02/23/2006 14:54:09
This is a multipart MIME message.

Content-Type: text/plain; charset=us-ascii said:
> If we're not going to bother to pass the correct fault type values to
>  uvm_fault(), then we should simply eliminate that argument to
> uvm_fault() altogether. 

Yes -- that argument is just used internally if uvm_fault() is called
by uvm_fault_wire(), to pass the WIRE/WIREMAX flags.
It doesn't look too hard to rearrange this and kill vm_fault_t

While we are here: Could someone check whether the appended patch
is correct? This would eliminate the last use of uvm_fault_wire()
in md code.

best regards

Content-Type: text/plain ; name="armidlestack.txt"; charset=us-ascii
Content-Description: armidlestack.txt
Content-Disposition: attachment; filename="armidlestack.txt"

--- cpu.c.~1.61.~	Sun Jan 22 16:35:38 2006
+++ cpu.c	Thu Feb 23 14:32:54 2006
@@ -599,18 +599,14 @@ cpu_alloc_idlepcb(struct cpu_info *ci)
 	vaddr_t uaddr;
 	struct pcb *pcb;
 	struct trapframe *tf;
-	int error;
 	 * Generate a kernel stack and PCB (in essence, a u-area) for the
 	 * new CPU.
-	if (uvm_uarea_alloc(&uaddr)) {
-		error = uvm_fault_wire(kernel_map, uaddr, uaddr + USPACE,
-		if (error)
-			return error;
-	}
+	uaddr = uvm_km_alloc(kernel_map, USPACE, 0, UVM_KMF_WIRED);
+	if (!uaddr)
+		return ENOMEM;
 	ci->ci_idlepcb = pcb = (struct pcb *)uaddr;