Prefer unaligned accesses over copying the rx buffer on all archs that can

handle it, not only i386.  ok brad@

diff --git a/sys/dev/pci/if_devar.h b/sys/dev/pci/if_devar.h
index 891169f..716d456 100644 (file)
--- a/sys/dev/pci/if_devar.h
+++ b/sys/dev/pci/if_devar.h
@@ -1,4 +1,4 @@
-/*     $OpenBSD: if_devar.h,v 1.28 2007/09/03 00:47:37 krw Exp $       */
+/*     $OpenBSD: if_devar.h,v 1.29 2008/08/15 15:49:08 naddy Exp $     */
 /*     $NetBSD: if_devar.h,v 1.13 1997/06/08 18:46:36 thorpej Exp $    */
 
 /*-
@@ -111,7 +111,7 @@ typedef struct {
  *
  * The receive space MUST ALWAYS be a multiple of the page size.
  * And the number of receive descriptors multiplied by the size
- * of the receive buffers must equal the recevive space.  This
+ * of the receive buffers must equal the receive space.  This
  * is so that we can manipulate the page tables so that even if a
  * packet wraps around the end of the receive space, we can
  * treat it as virtually contiguous.
@@ -121,7 +121,7 @@ typedef struct {
  * architecture which can't handle unaligned accesses) because with
  * 100Mb/s cards the copying is just too much of a hit.
  */
-#if !defined(__i386__)
+#ifdef __STRICT_ALIGNMENT
 #define        TULIP_COPY_RXDATA       1
 #endif