diff --git a/bswap.h b/bswap.h
index 180cc638f9..37fb04ed97 100644
--- a/bswap.h
+++ b/bswap.h
@@ -133,6 +133,9 @@ CPU_CONVERT(le, 64, uint64_t)
 #define le16_to_cpupu(p) le16_to_cpup(p)
 #define le32_to_cpupu(p) le32_to_cpup(p)
 
+#define cpu_to_be16wu(p, v) cpu_to_be16w(p, v)
+#define cpu_to_be32wu(p, v) cpu_to_be32w(p, v)
+
 #else
 
 static inline void cpu_to_le16wu(uint16_t *p, uint16_t v)
@@ -165,6 +168,30 @@ static inline uint32_t le32_to_cpupu(const uint32_t *p)
     return p1[0] | (p1[1] << 8) | (p1[2] << 16) | (p1[3] << 24);
 }
 
+static inline void cpu_to_be16wu(uint16_t *p, uint16_t v)
+{
+    uint8_t *p1 = (uint8_t *)p;
+
+    p1[0] = v >> 8;
+    p1[1] = v;
+}
+
+static inline void cpu_to_be32wu(uint32_t *p, uint32_t v)
+{
+    uint8_t *p1 = (uint8_t *)p;
+
+    p1[0] = v >> 24;
+    p1[1] = v >> 16;
+    p1[2] = v >> 8;
+    p1[3] = v;
+}
+
+#endif
+
+#ifdef WORDS_BIGENDIAN
+#define cpu_to_32wu cpu_to_be32wu
+#else
+#define cpu_to_32wu cpu_to_le32wu
 #endif
 
 #undef le_bswap
diff --git a/hw/vga.c b/hw/vga.c
index 2a0a9c3d01..2660b04695 100644
--- a/hw/vga.c
+++ b/hw/vga.c
@@ -824,8 +824,7 @@ typedef void vga_draw_line_func(VGAState *s1, uint8_t *d,
 
 static inline unsigned int rgb_to_pixel8(unsigned int r, unsigned int g, unsigned b)
 {
-    /* XXX: TODO */
-    return 0;
+    return ((r >> 5) << 5) | ((g >> 5) << 2) | (b >> 6);
 }
 
 static inline unsigned int rgb_to_pixel15(unsigned int r, unsigned int g, unsigned b)
diff --git a/hw/vga_template.h b/hw/vga_template.h
index 51302b3495..909571ebb3 100644
--- a/hw/vga_template.h
+++ b/hw/vga_template.h
@@ -44,7 +44,7 @@ static inline void glue(vga_draw_glyph_line_, DEPTH)(uint8_t *d,
 {
 #if BPP == 1
         ((uint32_t *)d)[0] = (dmask16[(font_data >> 4)] & xorcol) ^ bgcol;
-        ((uint32_t *)d)[3] = (dmask16[(font_data >> 0) & 0xf] & xorcol) ^ bgcol;
+        ((uint32_t *)d)[1] = (dmask16[(font_data >> 0) & 0xf] & xorcol) ^ bgcol;
 #elif BPP == 2
         ((uint32_t *)d)[0] = (dmask4[(font_data >> 6)] & xorcol) ^ bgcol;
         ((uint32_t *)d)[1] = (dmask4[(font_data >> 4) & 3] & xorcol) ^ bgcol;
@@ -106,22 +106,21 @@ static void glue(vga_draw_glyph9_, DEPTH)(uint8_t *d, int linesize,
     xorcol = bgcol ^ fgcol;
     do {
         font_data = font_ptr[0];
-        /* XXX: unaligned accesses are done */
 #if BPP == 1
-        ((uint32_t *)d)[0] = (dmask16[(font_data >> 4)] & xorcol) ^ bgcol;
+        cpu_to_32wu((uint32_t *)d, (dmask16[(font_data >> 4)] & xorcol) ^ bgcol);
         v = (dmask16[(font_data >> 0) & 0xf] & xorcol) ^ bgcol;
-        ((uint32_t *)d)[3] = v;
+        cpu_to_32wu(((uint32_t *)d)+1, v);
         if (dup9)
             ((uint8_t *)d)[8] = v >> (24 * (1 - BIG));
         else
             ((uint8_t *)d)[8] = bgcol;
         
 #elif BPP == 2
-        ((uint32_t *)d)[0] = (dmask4[(font_data >> 6)] & xorcol) ^ bgcol;
-        ((uint32_t *)d)[1] = (dmask4[(font_data >> 4) & 3] & xorcol) ^ bgcol;
-        ((uint32_t *)d)[2] = (dmask4[(font_data >> 2) & 3] & xorcol) ^ bgcol;
+        cpu_to_32wu(((uint32_t *)d)+0, (dmask4[(font_data >> 6)] & xorcol) ^ bgcol);
+        cpu_to_32wu(((uint32_t *)d)+1, (dmask4[(font_data >> 4) & 3] & xorcol) ^ bgcol);
+        cpu_to_32wu(((uint32_t *)d)+2, (dmask4[(font_data >> 2) & 3] & xorcol) ^ bgcol);
         v = (dmask4[(font_data >> 0) & 3] & xorcol) ^ bgcol;
-        ((uint32_t *)d)[3] = v;
+        cpu_to_32wu(((uint32_t *)d)+3, v);
         if (dup9)
             ((uint16_t *)d)[8] = v >> (16 * (1 - BIG));
         else