--- ioscale.tex.orig	2004-06-10 09:01:14.425378903 -0700
+++ ioscale.tex	2004-06-10 09:02:32.683415493 -0700
@@ -51,12 +51,12 @@
 explicit flush of DMA write data after each PIO read.
 
 For the cases in which strong ordering of PIO read responses and DMA
-writes are not necessary, a new API
-was needed so that drivers could communicate that a given PIO read
-response could used relaxed ordering with respect to prior DMA writes.  The
-\emph{read\_relaxed} API was added early in the 2.6 series for this
-purpose, and mirrors the normal read routines, which have variants for
-various sized reads.
+writes are not necessary, a new API was needed so that drivers could
+communicate that a given PIO read response could used relaxed ordering
+with respect to prior DMA writes.  The \emph{read\_relaxed} API
+\cite{relaxed} was added early in the 2.6 series for this purpose, and
+mirrors the normal read routines, which have variants for various
+sized reads.
 
 The results below show how expensive a normal PIO read transaction can
 be, especially on a system doing a lot of I/O (and thus DMA).
@@ -132,9 +132,9 @@
 \end{table}
 
 Adding this API to Linux (i.e. making it non-sn2 specific) was
-discussed some time ago, and may need to be raised again, since it
-does appear to be useful on sn2, and is probably similarly useful on
-other platforms.
+discussed some time ago \cite{mmiob}, and may need to be raised again,
+since it does appear to be useful on sn2, and is probably similarly
+useful on other platforms.
 
 \subsection*{Local allocation of consistent DMA mappings}
 
--- biblio.tex.orig	2004-06-10 09:02:54.232357765 -0700
+++ biblio.tex	2004-06-10 09:09:33.890074178 -0700
@@ -25,4 +25,8 @@
 
 \bibitem{SPECOMP} \url{http://www.spec.org/omp/results/ompl2001.html}
 
+\bibitem{relaxed} \url{http://linux.bkbits.net:8080/linux-2.5/cset%4040213ca0d3eIznHTPAR_kLCsMZI9VQ?nav=index.html|ChangeSet@-1d}
+
+\bibitem{mmiob} \url{http://www.cs.helsinki.fi/linux/linux-kernel/2002-01/1540.html}
+
 \end{thebibliography}