diff --git a/arch/arm/include/arm/irq.h b/arch/arm/include/arm/irq.h
index 11c2bb0463..98335c23f5 100644
--- a/arch/arm/include/arm/irq.h
+++ b/arch/arm/include/arm/irq.h
@@ -103,6 +103,12 @@
 #define REG_LR              REG_R14
 #define REG_PC              REG_R15
 
+/* The PIC register is usually R10. It can be R9 is stack checking is enabled
+ * or if the user changes it with -mpic-register on the GCC command line.
+ */
+
+#define REG_PIC             REG_R10
+
 /****************************************************************************
  * Public Types
  ****************************************************************************/
diff --git a/arch/arm/include/cortexm3/irq.h b/arch/arm/include/cortexm3/irq.h
index ce126748a6..b439ec42fa 100644
--- a/arch/arm/include/cortexm3/irq.h
+++ b/arch/arm/include/cortexm3/irq.h
@@ -109,6 +109,12 @@
 #define REG_LR              REG_R14
 #define REG_PC              REG_R15
 
+/* The PIC register is usually R10. It can be R9 is stack checking is enabled
+ * or if the user changes it with -mpic-register on the GCC command line.
+ */
+
+#define REG_PIC             REG_R10
+
 /****************************************************************************
  * Public Types
  ****************************************************************************/
diff --git a/arch/arm/src/arm/up_initialstate.c b/arch/arm/src/arm/up_initialstate.c
index 2ba6834454..c3a5f8a83f 100644
--- a/arch/arm/src/arm/up_initialstate.c
+++ b/arch/arm/src/arm/up_initialstate.c
@@ -93,6 +93,21 @@ void up_initial_state(_TCB *tcb)
 
   xcp->regs[REG_PC]      = (uint32)tcb->start;
 
+  /* If this task is running PIC, then set the PIC base register to the
+   * address of the allocated D-Space region.
+   */
+
+#ifdef CONFIG_PIC
+  if (tcb->dspace != NULL)
+    {
+      /* Set the PIC base register (probably R10) to the address of the
+       * alloacated D-Space region.
+       */
+
+      xcp->regs[REG_PIC] = (uint32)tcb->dspace->region;
+    }
+#endif
+
   /* Enable or disable interrupts, based on user configuration */
 
 # ifdef CONFIG_SUPPRESS_INTERRUPTS
diff --git a/arch/arm/src/cortexm3/up_initialstate.c b/arch/arm/src/cortexm3/up_initialstate.c
index c529064d46..9bf65bf57a 100644
--- a/arch/arm/src/cortexm3/up_initialstate.c
+++ b/arch/arm/src/cortexm3/up_initialstate.c
@@ -97,6 +97,21 @@ void up_initial_state(_TCB *tcb)
 
   xcp->regs[REG_XPSR]    = CORTEXM3_XPSR_T;
 
+  /* If this task is running PIC, then set the PIC base register to the
+   * address of the allocated D-Space region.
+   */
+
+#ifdef CONFIG_PIC
+  if (tcb->dspace != NULL)
+    {
+      /* Set the PIC base register (probably R10) to the address of the
+       * alloacated D-Space region.
+       */
+
+      xcp->regs[REG_PIC] = (uint32)tcb->dspace->region;
+    }
+#endif
+
   /* Enable or disable interrupts, based on user configuration */
 
 # ifdef CONFIG_SUPPRESS_INTERRUPTS
diff --git a/binfmt/binfmt_execmodule.c b/binfmt/binfmt_execmodule.c
index de65c15cf2..21fe559874 100644
--- a/binfmt/binfmt_execmodule.c
+++ b/binfmt/binfmt_execmodule.c
@@ -145,7 +145,7 @@ int exec_module(FAR const struct binary_s *bin, int priority)
   /* Add the DSpace address as the PIC base address */
 
 #ifdef CONFIG_PIC
-  tcb->picbase = bin->dspace;
+  tcb->dspace = bin->dspace;
 
   /* Re-initialize the task's initial state to account for the new PIC base */
 
diff --git a/binfmt/libnxflat/libnxflat_load.c b/binfmt/libnxflat/libnxflat_load.c
index aa00f50c75..b27dca9356 100644
--- a/binfmt/libnxflat/libnxflat_load.c
+++ b/binfmt/libnxflat/libnxflat_load.c
@@ -118,7 +118,7 @@ static void nxflat_reloc(struct nxflat_loadinfo_s *loadinfo, uint32 rl)
        * DSpace to hold information needed by ld.so at run time.
        */
 
-      datastart = loadinfo->dspace;
+      datastart = (uint32)loadinfo->dspace->region;
 
       /* Get a pointer to the value that needs relocation in
        * DSpace.
@@ -257,22 +257,22 @@ int nxflat_load(struct nxflat_loadinfo_s *loadinfo)
    * uninitialized ISpace memory.
    */
 
-  loadinfo->dspace = (uint32)malloc(loadinfo->dsize);
+  loadinfo->dspace = (struct dspace_s *)malloc(SIZEOF_DSPACE_S(loadinfo->dsize));
   if (loadinfo->dspace == 0)
     {
       bdbg("Failed to allocate DSpace\n");
       ret = -ENOMEM;
       goto errout;
     }
+  loadinfo->dspace->crefs = 1;
 
-  bvdbg("Allocated DSpace (%d bytes) at %08x\n",
-      loadinfo->dsize, loadinfo->dspace);
+  bvdbg("Allocated DSpace (%d bytes) at %p\n", loadinfo->dsize, loadinfo->dspace);
 
   /* Now, read the data into allocated DSpace at doffset into the
    * allocated DSpace memory.
    */
 
-  ret = nxflat_read(loadinfo, (char*)loadinfo->dspace, dreadsize, doffset);
+  ret = nxflat_read(loadinfo, (char*)loadinfo->dspace->region, dreadsize, doffset);
   if (ret < 0)
     {
       bdbg("Failed to read .data section: %d\n", ret);
@@ -285,10 +285,10 @@ int nxflat_load(struct nxflat_loadinfo_s *loadinfo)
   /* Resolve the address of the relocation table.  In the file, the
    * relocations should lie at the same offset as BSS.  The current
    * value of relocstart is the offset from the beginning of the file.
-   * The following adjustment will convert it to an address in DSpace.
+   * The following adjustment will convert it to an address in dspace->
    */
 
-  reloctab = (uint32*)(loadinfo->relocstart + loadinfo->dspace - loadinfo->isize);
+  reloctab = (uint32*)(loadinfo->relocstart + (uint32)loadinfo->dspace->region - loadinfo->isize);
 
   bvdbg("Relocation table at 0x%p, reloccount=%d\n",
       reloctab, loadinfo->reloccount);
@@ -304,7 +304,7 @@ int nxflat_load(struct nxflat_loadinfo_s *loadinfo)
    * in the file.
    */
 
-  memset((void*)(loadinfo->dspace + loadinfo->datasize),
+  memset((void*)(loadinfo->dspace->region + loadinfo->datasize),
 	          0, loadinfo->bsssize);
   return OK;
 
diff --git a/examples/nxflat/tests/Make.defs b/examples/nxflat/tests/Make.defs
index c4ec82c52a..3b6400730e 100644
--- a/examples/nxflat/tests/Make.defs
+++ b/examples/nxflat/tests/Make.defs
@@ -41,5 +41,5 @@ NXFLATCC = $(CC)
 NXFLATCFLAGS = $(CFLAGS)
 
 NXFLATLD = $(LD)
-NXFLATLDFLAGS =
+NXFLATLDFLAGS = -e main
 
diff --git a/include/nuttx/binfmt.h b/include/nuttx/binfmt.h
index 855cc3359f..0c4238989d 100644
--- a/include/nuttx/binfmt.h
+++ b/include/nuttx/binfmt.h
@@ -43,6 +43,7 @@
 #include <nuttx/config.h>
 #include <sys/types.h>
 #include <nxflat.h>
+#include <nuttx/sched.h>
 
 /****************************************************************************
  * Pre-processor Definitions
@@ -65,7 +66,7 @@ struct binary_s
 
   main_t       entrypt;              /* Entry point into a program module */
   FAR void    *ispace;               /* Memory-mapped, I-space (.text) address */
-  FAR void    *dspace;               /* Address of the allocated .data/.bss space */
+  FAR struct dspace_s *dspace;       /* Address of the allocated .data/.bss space */
   size_t       isize;                /* Size of the I-space region (needed for munmap) */
   size_t       stacksize;            /* Size of the stack in bytes (unallocated) */
 };
diff --git a/include/nuttx/nxflat.h b/include/nuttx/nxflat.h
index f78c79dcd4..543f00e6d3 100644
--- a/include/nuttx/nxflat.h
+++ b/include/nuttx/nxflat.h
@@ -40,7 +40,9 @@
  * Included Files
  ****************************************************************************/
 
+#include <nuttx/config.h>
 #include <nxflat.h>
+#include <nuttx/sched.h>
 
 /****************************************************************************
  * Pre-processor Definitions
@@ -61,29 +63,29 @@ struct nxflat_loadinfo_s
    * text section instance in the system for each module.
    */
 
-  uint32 ispace;       /* Address where hdr/text is loaded */
-  uint32 entryoffs;    /* Offset from ispace to entry point */
-  uint32 isize;        /* Size of ispace. */
+  uint32 ispace;           /* Address where hdr/text is loaded */
+  uint32 entryoffs;        /* Offset from ispace to entry point */
+  uint32 isize;            /* Size of ispace. */
 
   /* Data Space (DSpace): This region contains all information that in referenced
    * as data (other than the stack which is separately allocated).  There will be
    * a unique instance of DSpace (and stack) for each instance of a process.
    */
 
-  uint32 dspace;       /* Address where data/bss/etc. is loaded */
-  uint32 datasize;     /* Size of data segment in dspace */
-  uint32 bsssize;      /* Size of bss segment in dspace */
-  uint32 stacksize;    /* Size of stack (not allocated) */
-  uint32 dsize;        /* Size of dspace (may be large than parts) */
+  struct dspace_s *dspace; /* Allocated D-Space (data/bss/etc) */
+  uint32 datasize;         /* Size of data segment in dspace */
+  uint32 bsssize;          /* Size of bss segment in dspace */
+  uint32 stacksize;        /* Size of stack (not allocated) */
+  uint32 dsize;            /* Size of dspace (may be large than parts) */
 
   /* This is temporary memory where relocation records will be loaded. */
 
-  uint32 relocstart;   /* Start of array of struct flat_reloc */
-  uint32 reloccount;   /* Number of elements in reloc array */
+  uint32 relocstart;       /* Start of array of struct flat_reloc */
+  uint32 reloccount;       /* Number of elements in reloc array */
 
   /* File descriptors */
 
-  int    filfd;        /* Descriptor for the file being loaded */
+  int    filfd;            /* Descriptor for the file being loaded */
 
   const struct nxflat_hdr_s  *header; /* A reference to the flat file header */
 };
diff --git a/include/nuttx/sched.h b/include/nuttx/sched.h
index 520d73047d..0c55dc1f87 100644
--- a/include/nuttx/sched.h
+++ b/include/nuttx/sched.h
@@ -161,6 +161,18 @@ typedef struct environ_s environ_t;
 # define SIZEOF_ENVIRON_T(alloc) (sizeof(environ_t) + alloc - 1)
 #endif
 
+/* This structure describes a reference counted D-Space region */
+
+struct dspace_s
+{
+  uint32 crefs;               /* This is the number of pthreads that shared the
+                               * the same D-Space.
+                               */
+  ubyte  region[1];           /* Beginning of the allocated region */
+};
+
+#define SIZEOF_DSPACE_S(n) (sizeof(struct dspace_s) - 1 + (n))
+
 /* This is the task control block (TCB) */
 
 struct _TCB
@@ -217,7 +229,7 @@ struct _TCB
   /* External Module Support ****************************************************/
 
 #ifdef CONFIG_PIC
-  FAR void *picbase;                     /* Allocated area for .bss and .data   */
+  FAR struct dspace_s *dspace;           /* Allocated area for .bss and .data   */
 #endif
 
   /* POSIX Thread Specific Data *************************************************/
diff --git a/sched/sched_releasetcb.c b/sched/sched_releasetcb.c
index 567db20211..cd8e9c8c32 100644
--- a/sched/sched_releasetcb.c
+++ b/sched/sched_releasetcb.c
@@ -138,9 +138,16 @@ int sched_releasetcb(FAR _TCB *tcb)
       /* Delete the tasks's allocated DSpace region (external modules only) */
 
 #ifdef CONFIG_PIC
-      if (tcb->picbase)
+      if (tcb->dspace)
         {
-          sched_free(tcb->picbase);
+          if (tcb->dspace->crefs <= 1)
+            {
+              sched_free(tcb->dspace);
+            }
+          else
+            {
+              tcb->dspace->crefs--;
+            }
         }
 #endif
 
diff --git a/sched/task_setup.c b/sched/task_setup.c
index 25a85394a9..83bae570b1 100644
--- a/sched/task_setup.c
+++ b/sched/task_setup.c
@@ -142,6 +142,43 @@ static STATUS task_assignpid(FAR _TCB *tcb)
   return ERROR;
 }
 
+/****************************************************************************
+ * Function: task_dupdspace
+ *
+ * Description:
+ *   When a new task or thread is created from a PIC module, then that
+ *   module (probably) intends the task or thread to execute in the same
+ *   D-Space.  This function will duplicate the D-Space for that purpose.
+ *
+ * Parameters:
+ *   tcb - The TCB of the new task.
+ *
+ * Returned Value:
+ *   None
+ *
+ * Assumptions:
+ *
+ ****************************************************************************/
+
+#ifdef CONFIG_PIC
+static inline void task_dupdspace(FAR _TCB *tcb)
+{
+  FAR _TCB *rtcb = (FAR _TCB*)g_readytorun.head;
+  if (rtcb->dspace != NULL)
+    {
+      /* Copy the D-Space structure reference and increment the reference
+       * count on the memory.  The D-Space memory will persist until the
+       * last thread exits (see sched_releasetcb()).
+       */
+
+      tcb->dspace = rtcb->dspace;
+      tcb->dspace->crefs++;
+    }
+}
+#else
+#  define task_dupdspace(tcb)
+#endif
+
 /****************************************************************************
  * Public Functions
  ****************************************************************************/
@@ -150,11 +187,11 @@ static STATUS task_assignpid(FAR _TCB *tcb)
  * Name: task_schedsetup
  *
  * Description:
- *   This functions initializes a Task Control Block (TCB)
- *   in preparation for starting a new thread.
+ *   This functions initializes a Task Control Block (TCB) in preparation
+ *   for starting a new thread.
  *
- *   task_schedsetup() is called from task_init(),
- *   task_start(), and pthread_create();
+ *   task_schedsetup() is called from task_init(), task_start(), and
+ *   pthread_create();
  *
  * Input Parameters:
  *   tcb        - Address of the new task's TCB
@@ -165,8 +202,8 @@ static STATUS task_assignpid(FAR _TCB *tcb)
  * Return Value:
  *   OK on success; ERROR on failure.
  *
- *   This function can only failure is it is unable to assign
- *   a new, unique task ID to the TCB (errno is not set).
+ *   This function can only failure is it is unable to assign a new, unique
+ *   task ID to the TCB (errno is not set).
  *
  ****************************************************************************/
 
@@ -204,6 +241,13 @@ STATUS task_schedsetup(FAR _TCB *tcb, int priority,
 
       tcb->task_state   = TSTATE_TASK_INVALID;
 
+      /* Clone the parent tasks D-Space (if it was running PIC).  This
+       * must be done before calling up_initial_state() so that the
+       * state setup will take the PIC address base into account.
+       */
+
+      task_dupdspace(tcb);
+
       /* Initialize the processor-specific portion of the TCB */
 
       up_initial_state(tcb);
diff --git a/sched/wd_start.c b/sched/wd_start.c
index ea9117459b..a406b3f30c 100644
--- a/sched/wd_start.c
+++ b/sched/wd_start.c
@@ -158,9 +158,7 @@ STATUS wd_start(WDOG_ID wdog, int delay, wdentry_t wdentry,  int argc, ...)
   /* Save the data in the watchdog structure */
 
   wdog->func = wdentry;         /* Function to execute when delay expires */
-#ifdef CONFIG_PIC
   up_getpicbase(&wdog->picbase);
-#endif
   wdog->argc = argc;
 
   va_start(ap, argc);