Index: amd64/amd64/mp_machdep.c
===================================================================
--- amd64/amd64/mp_machdep.c	(revision 191739)
+++ amd64/amd64/mp_machdep.c	(working copy)
@@ -91,9 +91,10 @@
 /* Free these after use */
 void *bootstacks[MAXCPU];
 
-/* Temporary holder for double fault stack */
+/* Temporary variables for init_secondary()  */
 char *doublefault_stack;
 char *nmi_stack;
+uintptr_t dpcpu;
 
 /* Hotwire a 0->4MB V==P mapping */
 extern pt_entry_t *KPTphys;
@@ -601,6 +602,9 @@
 	pc->pc_ldt = (struct system_segment_descriptor *)&gdt[NGDT * cpu +
 	    GUSERLDT_SEL];
 
+	pc->pc_dynamic = dpcpu;
+	dpcpu_init(pc, DPCPU_SIZE);
+
 	/* Save the per-cpu pointer for use by the NMI handler. */
 	np->np_pcpu = (register_t) pc;
 
@@ -881,6 +885,8 @@
 		bootstacks[cpu] = (void *)kmem_alloc(kernel_map, KSTACK_PAGES * PAGE_SIZE);
 		doublefault_stack = (char *)kmem_alloc(kernel_map, PAGE_SIZE);
 		nmi_stack = (char *)kmem_alloc(kernel_map, PAGE_SIZE);
+		dpcpu = kmem_alloc(kernel_map,
+		    roundup2(DPCPU_SIZE, PAGE_SIZE));
 
 		bootSTK = (char *)bootstacks[cpu] + KSTACK_PAGES * PAGE_SIZE - 8;
 		bootAP = cpu;
Index: amd64/amd64/machdep.c
===================================================================
--- amd64/amd64/machdep.c	(revision 191739)
+++ amd64/amd64/machdep.c	(working copy)
@@ -1368,6 +1368,7 @@
 	struct pcpu *pc;
 	struct nmi_pcpu *np;
 	u_int64_t msr;
+	uintptr_t dpcpu;
 	char *env;
 
 	thread0.td_kstack = physfree + KERNBASE;
@@ -1428,6 +1429,10 @@
 	PCPU_SET(ldt, (struct system_segment_descriptor *)&gdt[GUSERLDT_SEL]);
 	PCPU_SET(fs32p, &gdt[GUFS32_SEL]);
 	PCPU_SET(gs32p, &gdt[GUGS32_SEL]);
+	dpcpu = physfree + KERNBASE;
+	physfree += roundup2(DPCPU_SIZE, PAGE_SIZE);
+	PCPU_SET(dynamic, dpcpu);
+	dpcpu_init(pc, DPCPU_SIZE);
 
 	/*
 	 * Initialize mutexes.
Index: sys/pcpu.h
===================================================================
--- sys/pcpu.h	(revision 191739)
+++ sys/pcpu.h	(working copy)
@@ -45,6 +45,61 @@
 struct pcb;
 struct thread;
 
+/*
+ * Define a set for pcpu data.
+ * 
+ * We don't use SET_DECLARE because it defines the set as 'a' when we
+ * want 'aw'.  GCC considers uninitialized data in a seperate section
+ * writable and there is no generic zero initializer that works for
+ * structs and scalars.
+ */
+extern uintptr_t *__start_set_pcpu;
+extern uintptr_t *__stop_set_pcpu;
+__asm__(".section set_pcpu, \"aw\", @progbits");
+__asm__(".previous");
+
+/*
+ * Array of dynamic pcpu base values.  Indexed by id.
+ */
+extern uintptr_t dpcpu_ptrs[];
+
+/*
+ * Convenience defines.
+ */
+#define	DPCPU_NAME(n)		pcpu_entry_##n
+#define	DPCPU_START		(uintptr_t)&__start_set_pcpu
+#define	DPCPU_STOP		(uintptr_t)&__stop_set_pcpu
+#define	DPCPU_SIZE		(DPCPU_STOP - DPCPU_START)
+#define	DPCPU_OFFSET(n)		((uintptr_t)&DPCPU_NAME(n) - DPCPU_START)
+
+/*
+ * Declaration and definition.
+ */
+#define	DPCPU_DECLARE(t, n)	extern t DPCPU_NAME(n)
+#define	DPCPU_DEFINE(t, n)	t DPCPU_NAME(n) __section("set_pcpu") __used
+
+/*
+ * Accessors with a given base.
+ */
+#define	_DPCPU_PTR(b, n)						\
+    (__typeof(DPCPU_NAME(n))*)((b) + DPCPU_OFFSET(n))
+#define	_DPCPU_GET(b, n)	(*_DPCPU_PTR(b, n))
+#define	_DPCPU_SET(b, n, v)	(*_DPCPU_PTR(b, n) = v)
+
+/*
+ * Accessors for the current cpu.
+ */
+#define	DPCPU_PTR(n)		_DPCPU_PTR(PCPU_GET(dynamic), n)
+#define	DPCPU_GET(n)		(*DPCPU_PTR(n))
+#define	DPCPU_SET(n, v)		(*DPCPU_PTR(n) = v)
+
+/*
+ * Accessors for remote cpus.
+ */
+#define	DPCPU_ID_PTR(i, n)	_DPCPU_PTR(dpcpu_ptrs[(i)], n)
+#define	DPCPU_ID_GET(i, n)	(*DPCPU_ID_PTR(i, n))
+#define	DPCPU_ID_SET(i, n, v)	(*DPCPU_ID_PTR(i, n) = v)
+
 /* 
  * XXXUPS remove as soon as we have per cpu variable
  * linker sets and  can define rm_queue in _rm_lock.h
@@ -96,6 +151,11 @@
 	struct rm_queue  pc_rm_queue; 
 
 	/*
+	 * Dynamic per-cpu data area.
+	 */
+	uintptr_t	pc_dynamic;
+
+	/*
 	 * Keep MD fields last, so that CPU-specific variations on a
 	 * single architecture don't result in offset variations of
 	 * the machine-independent fields of the pcpu. Even though
@@ -136,6 +196,7 @@
 void	pcpu_destroy(struct pcpu *pcpu);
 struct	pcpu *pcpu_find(u_int cpuid);
 void	pcpu_init(struct pcpu *pcpu, int cpuid, size_t size);
+void	dpcpu_init(struct pcpu *pcpu, size_t size);
 
 #endif	/* _KERNEL */
 
Index: kern/subr_pcpu.c
===================================================================
--- kern/subr_pcpu.c	(revision 191739)
+++ kern/subr_pcpu.c	(working copy)
@@ -56,6 +56,9 @@
 #include <sys/smp.h>
 #include <ddb/ddb.h>
 
+DPCPU_DEFINE(int, id);
+
+uintptr_t dpcpu_ptrs[MAXCPU];
 struct pcpu *cpuid_to_pcpu[MAXCPU];
 struct cpuhead cpuhead = SLIST_HEAD_INITIALIZER(cpuhead);
 
@@ -82,6 +85,23 @@
 
 }
 
+void
+dpcpu_init(struct pcpu *pcpu, size_t size)
+{
+
+	/*
+	 * Initialize defaults from our linker section.
+	 */
+	memcpy((void *)pcpu->pc_dynamic, (void *)DPCPU_START, DPCPU_SIZE);
+
+	/*
+	 * Place it in the global pcpu array.
+	 */
+	dpcpu_ptrs[pcpu->pc_cpuid] = (uintptr_t)pcpu->pc_dynamic;
+
+	DPCPU_ID_SET(pcpu->pc_cpuid, id, pcpu->pc_cpuid);
+}
+
 /*
  * Destroy a struct pcpu.
  */
@@ -91,6 +111,7 @@
 
 	SLIST_REMOVE(&cpuhead, pcpu, pcpu, pc_allcpu);
 	cpuid_to_pcpu[pcpu->pc_cpuid] = NULL;
+	dpcpu_ptrs[pcpu->pc_cpuid] = 0;
 }
 
 /*
@@ -110,7 +131,9 @@
 {
 	struct thread *td;
 
-	db_printf("cpuid        = %d\n", pc->pc_cpuid);
+	db_printf("cpuid        = %d (%d)\n",
+	    pc->pc_cpuid, DPCPU_ID_GET(pc->pc_cpuid, id));
+	db_printf("curcpuid	= %d\n", DPCPU_GET(id));
 	db_printf("curthread    = ");
 	td = pc->pc_curthread;
 	if (td != NULL)