[sparc32] Copy full soft PMD in fault handler

[sparc32] Copy full soft PMD in fault handler

[Keith M Wesolowski]

sparc_do_fault assumes that pmd_val(*pmd) = pmd_val(*pmd_k) will
duplicate a PMD entry.  With large soft-PMDs this doesn't work.

Is this ok?  Specifically I'm wondering whether it's necessary to use
swap here is in pmd_set.  The comment in the i386 handler "Do _not_
use "tsk" here. We might be inside an interrupt in the middle of a
task switch.." doesn't make me feel too good about this function's
health and well-being either, since we use tsk.

# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
#	           ChangeSet	1.1551  -> 1.1552 
#	arch/sparc/mm/fault.c	1.15    -> 1.16   
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 04/02/02	wesolows@foobazco.org	1.1552
# Copy all 16 sub-pmd-entries, not just the first one.
# --------------------------------------------
#
diff -Nru a/arch/sparc/mm/fault.c b/arch/sparc/mm/fault.c
--- a/arch/sparc/mm/fault.c	Mon Feb  2 01:25:54 2004
+++ b/arch/sparc/mm/fault.c	Mon Feb  2 01:25:54 2004
@@ -392,7 +392,7 @@
 
 		if (pmd_present(*pmd) || !pmd_present(*pmd_k))
 			goto bad_area_nosemaphore;
-		pmd_val(*pmd) = pmd_val(*pmd_k);
+		*pmd = *pmd_k;
 		return;
 	}
 }


-- 
Keith M Wesolowski

Re: [sparc32] Copy full soft PMD in fault handler

[David S. Miller]

On Mon, 2 Feb 2004 01:43:47 -0800
Keith M Wesolowski <wesolows@foobazco.org> wrote:

> sparc_do_fault assumes that pmd_val(*pmd) = pmd_val(*pmd_k) will
> duplicate a PMD entry.  With large soft-PMDs this doesn't work.
> 
> Is this ok?

I think this patch is just fine.

> The comment in the i386 handler "Do _not_
> use "tsk" here. We might be inside an interrupt in the middle of a
> task switch.." doesn't make me feel too good about this function's
> health and well-being either, since we use tsk.

I think it does not apply to sparc, on sparc we disable interrupts during
the context switch so that it "appears" atomic.  But actually you should
change it over to use init_mm.pgd just like other platforms merely for
consistency.