Index: firmware/target/arm/ata-sd-pp.c
===================================================================
--- firmware/target/arm/ata-sd-pp.c	(Revision 17715)
+++ firmware/target/arm/ata-sd-pp.c	(Arbeitskopie)
@@ -464,18 +464,49 @@
     );
 }
 
-/* Writes have to be kept slow for now */
 static inline void copy_write_sectors(const unsigned char** buf)
 {
+    if ( (((intptr_t)*buf & 3 ) == 0) )
+    {
+        asm volatile (
+            "ldmia  %[buf]!, { r2, r4, r6, r8 } \r\n"
+            "mov    r3, r2, lsr #16             \r\n" 
+            "mov    r5, r4, lsr #16             \r\n" 
+            "mov    r7, r6, lsr #16             \r\n" 
+            "mov    r9, r8, lsr #16             \r\n" 
+            "stmia  %[data], { r2-r9 }          \r\n"
+            "ldmia  %[buf]!, { r2, r4, r6, r8 } \r\n"
+            "mov    r3, r2, lsr #16             \r\n" 
+            "mov    r5, r4, lsr #16             \r\n" 
+            "mov    r7, r6, lsr #16             \r\n" 
+            "mov    r9, r8, lsr #16             \r\n" 
+            "stmia  %[data], { r2-r9 }          \r\n"
+            : [buf]"+&r"(*buf)
+            : [data]"r"(&DATA_REG)
+            : "r2", "r3", "r4", "r5", "r6", "r7", "r8", "r9"
+        );
+    }
+    else
+    {
+        /* Do we ever have an unaligned buffer? */
+
+#ifdef SANSA_E200
+        /* On the card slot we need another udelay here to prevent the 'two bytes inserted' bug. */
+        /* But on the internal flash, there must be no udelay! */
+        if (currcard == &card_info[1])
+            udelay(2);
+#endif
+
     int cnt = FIFO_LEN;
     unsigned t;
 
     do
-    {
-        t  = *(*buf)++;
-        t |= *(*buf)++ << 8;
-        DATA_REG = t;
-    } while (--cnt > 0); /* tail loop is faster */
+        {
+            t  = *(*buf)++;
+            t |= *(*buf)++ << 8;
+            DATA_REG = t;
+        } while (--cnt > 0); /* tail loop is faster */
+    }
 }
 
 static int sd_select_bank(unsigned char bank)
@@ -983,11 +1014,14 @@
             SD_STATE_REG = PRG;
         }
 
-        udelay(2); /* needed here (loop is too fast :-) */
-
         /* Wait for the FIFO to empty */
         if (sd_poll_status(FIFO_EMPTY, 0x80000))
         {
+
+            udelay(3); /* Needed on e200. Maybe has to be adjusted for c200 and GoGear. */
+                       /* Too short delay will cause lockup or corruption, */
+                       /* too long delay will raise the 'two bytes inserted' bug. */
+
             copy_write_sectors(&buf); /* Copy one chunk of 16 words */
             /* TODO: Switch bank if necessary */
             continue;