diff --git a/ports/avr/atomport-asm.s b/ports/avr/atomport-asm.s
index a4525c4..f7a55a3 100644
--- a/ports/avr/atomport-asm.s
+++ b/ports/avr/atomport-asm.s
@@ -113,6 +113,28 @@ archContextSwitch:
     push r28
     push r29
 
+    /**
+     * On devices with large program space we also save RAMPZ, EIND.
+     * Note that GCC 4.3 and later to actually save RAMPZ when called
+     * via an interrupt handler, which means that we end up stacking
+     * RAMPZ twice. However we do need to save RAMPZ for cooperative
+     * context switches where we are called via a function call rather
+     * than an ISR (at this time GCC does not save RAMPZ before function
+     * calls). This has the added benefit that we continue to support
+     * GCC < 4.3, but with the added overhead of the double-stacking for
+     * newer versions of GCC.
+     *
+     * An alternative method that would work for GCC >= 4.3 only would be
+     * to detect whether we were called from an interrupt handler and not
+     * save RAMPZ under those circumstances.
+     */
+#ifdef __AVR_3_BYTE_PC__
+    in r0,_SFR_IO_ADDR(RAMPZ)
+    push r0
+    in r0,_SFR_IO_ADDR(EIND)
+    push r0
+#endif
+
     /**
      * Save the final stack pointer to the TCB. The parameter pointing to
      * the old TCB is still untouched in R25-R24. We have saved R16/R17
@@ -145,12 +167,15 @@ archContextSwitch:
      *    <R17>
      *    <R28>
      *    <R29>
+     *    <RAMPZ>     (Only certain devices)
+     *    <EIND>      (Only certain devices)
      *
      * The stack frame if this was a preemptive switch looks as follows:
      *
      *   <R1>    // saved by ISR
      *   <R0>    //
      *   <SREG>  //
+     *   <RAMPZ> //   (Only certain devices)
      *   <R18>   //
      *   <R19>   //
      *    ||     //
@@ -167,6 +192,8 @@ archContextSwitch:
      *   <R17>
      *   <R28>
      *   <R29>
+     *   <RAMPZ>     (Only certain devices)
+     *   <EIND>      (Only certain devices)
      *
      *
      * In addition, the thread's stack pointer (after context-save) is
@@ -203,6 +230,17 @@ archContextSwitch:
     out _SFR_IO_ADDR(SPL),r16  /* Set our stack pointer to the new thread's */
     out _SFR_IO_ADDR(SPH),r17  /* stack pointer, from its TCB. */
 
+    /**
+     * On devices with large program space we also restore RAMPZ, EIND.
+     */
+#ifdef __AVR_3_BYTE_PC__
+    pop r0
+    in r0,_SFR_IO_ADDR(EIND)
+    pop r0
+    in r0,_SFR_IO_ADDR(RAMPZ)
+    push r0
+#endif
+
     /**
      * Restore registers R2-R17, R28-R29.
      */
@@ -321,6 +359,17 @@ archFirstThreadRestore:
     out _SFR_IO_ADDR(SPL),r16  /* Set our stack pointer to the new thread's */
     out _SFR_IO_ADDR(SPH),r17  /* stack pointer, from its TCB. */
 
+    /**
+     * On devices with large program space we also restore RAMPZ, EIND.
+     */
+#ifdef __AVR_3_BYTE_PC__
+    pop r0
+    in r0,_SFR_IO_ADDR(EIND)
+    pop r0
+    in r0,_SFR_IO_ADDR(RAMPZ)
+    push r0
+#endif
+
     /**
      * Restore registers R2-R17, R28-R29.
      */
diff --git a/ports/avr/atomport.c b/ports/avr/atomport.c
index 86e67fa..9ae5555 100644
--- a/ports/avr/atomport.c
+++ b/ports/avr/atomport.c
@@ -250,6 +250,14 @@ void archThreadContextInit (ATOM_TCB *tcb_ptr, void *stack_top, void (*entry_poi
     *stack_ptr-- = 0x00;    /* R28 */
     *stack_ptr-- = 0x00;    /* R29 */
 
+    /**
+     * On devices with large program space we also context switch RAMPZ, EIND.
+     */
+#ifdef __AVR_3_BYTE_PC__
+    *stack_ptr-- = 0x00;    /* RAMPZ */
+    *stack_ptr-- = 0x00;    /* EIND */
+#endif
+
     /**
      * All thread context has now been initialised. Save the current
      * stack pointer to the thread's TCB so it knows where to start
@@ -344,4 +352,4 @@ ISR (TIMER1_COMPA_vect)
 ISR (BADISR_vect)
 {
     /* Empty */
-}
\ No newline at end of file
+}