diff --git a/sys/amd64/include/proc.h b/sys/amd64/include/proc.h
--- a/sys/amd64/include/proc.h
+++ b/sys/amd64/include/proc.h
@@ -97,15 +97,6 @@
 
 #ifdef	_KERNEL
 
-/* Get the current kernel thread stack usage. */
-#define GET_STACK_USAGE(total, used) do {				\
-	struct thread	*td = curthread;				\
-	(total) = td->td_kstack_pages * PAGE_SIZE;			\
-	(used) = (char *)td->td_kstack +				\
-	    td->td_kstack_pages * PAGE_SIZE -				\
-	    (char *)&td;						\
-} while (0)
-
 struct proc_ldt *user_ldt_alloc(struct proc *, int);
 void user_ldt_free(struct thread *);
 struct sysarch_args;
diff --git a/sys/amd64/include/stack.h b/sys/amd64/include/stack.h
--- a/sys/amd64/include/stack.h
+++ b/sys/amd64/include/stack.h
@@ -3,4 +3,28 @@
  */
 /* $FreeBSD$ */
 
+#ifndef _MACHINE_STACK_H_
+#define	_MACHINE_STACK_H_
+
 #include <x86/stack.h>
+
+#ifdef _SYS_PROC_H_
+
+/* Get the current kernel thread stack usage. */
+#define GET_STACK_USAGE(total, used) do {				\
+	struct thread	*td = curthread;				\
+	(total) = td->td_kstack_pages * PAGE_SIZE;			\
+	(used) = (char *)td->td_kstack +				\
+	    td->td_kstack_pages * PAGE_SIZE -				\
+	    (char *)&td;						\
+} while (0)
+
+static __inline bool
+kstack_contains(struct thread *td, vm_offset_t va, size_t len)
+{
+	return (va >= td->td_kstack && va + len >= va &&
+	    va + len <= td->td_kstack + td->td_kstack_pages * PAGE_SIZE);
+}
+#endif	/* _SYS_PROC_H_ */
+
+#endif
diff --git a/sys/arm/arm/ptrace_machdep.c b/sys/arm/arm/ptrace_machdep.c
--- a/sys/arm/arm/ptrace_machdep.c
+++ b/sys/arm/arm/ptrace_machdep.c
@@ -32,6 +32,7 @@
 #include <sys/proc.h>
 #include <sys/ptrace.h>
 #include <sys/reg.h>
+#include <machine/pcb.h>
 #ifdef VFP
 #include <machine/vfp.h>
 #endif
diff --git a/sys/arm/include/proc.h b/sys/arm/include/proc.h
--- a/sys/arm/include/proc.h
+++ b/sys/arm/include/proc.h
@@ -56,15 +56,4 @@
 
 #define	KINFO_PROC_SIZE 816
 
-#ifdef _KERNEL
-#include <machine/pcb.h>
-
-/* Get the current kernel thread stack usage. */
-#define	GET_STACK_USAGE(total, used) do {				\
-	struct thread *td = curthread;					\
-	(total) = td->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb);	\
-	(used) = td->td_kstack + (total) - (vm_offset_t)&td;		\
-} while (0)
-
-#endif  /* _KERNEL */
 #endif /* !_MACHINE_PROC_H_ */
diff --git a/sys/arm/include/stack.h b/sys/arm/include/stack.h
--- a/sys/arm/include/stack.h
+++ b/sys/arm/include/stack.h
@@ -63,6 +63,25 @@
 void unwind_module_loaded(struct linker_file *);
 void unwind_module_unloaded(struct linker_file *);
 
+#ifdef _SYS_PROC_H_
+
+#include <machine/pcb.h>
+
+/* Get the current kernel thread stack usage. */
+#define	GET_STACK_USAGE(total, used) do {				\
+	struct thread *td = curthread;					\
+	(total) = td->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb);	\
+	(used) = td->td_kstack + (total) - (vm_offset_t)&td;		\
+} while (0)
+
+static __inline bool
+kstack_contains(struct thread *td, vm_offset_t va, size_t len)
+{
+	return (va >= td->td_kstack && va + len >= va &&
+	    va + len <= td->td_kstack + td->td_kstack_pages * PAGE_SIZE);
+}
+#endif	/* _SYS_PROC_H_ */
+
 #endif
 
 #endif /* !_MACHINE_STACK_H_ */
diff --git a/sys/arm64/arm64/debug_monitor.c b/sys/arm64/arm64/debug_monitor.c
--- a/sys/arm64/arm64/debug_monitor.c
+++ b/sys/arm64/arm64/debug_monitor.c
@@ -44,6 +44,7 @@
 #include <machine/cpu.h>
 #include <machine/debug_monitor.h>
 #include <machine/kdb.h>
+#include <machine/pcb.h>
 
 #ifdef DDB
 #include <ddb/ddb.h>
diff --git a/sys/arm64/arm64/elf32_machdep.c b/sys/arm64/arm64/elf32_machdep.c
--- a/sys/arm64/arm64/elf32_machdep.c
+++ b/sys/arm64/arm64/elf32_machdep.c
@@ -52,6 +52,7 @@
 #include <sys/vnode.h>
 
 #include <machine/elf.h>
+#include <machine/pcb.h>
 #ifdef VFP
 #include <machine/vfp.h>
 #endif
diff --git a/sys/arm64/arm64/freebsd32_machdep.c b/sys/arm64/arm64/freebsd32_machdep.c
--- a/sys/arm64/arm64/freebsd32_machdep.c
+++ b/sys/arm64/arm64/freebsd32_machdep.c
@@ -37,6 +37,7 @@
 #include <sys/sysent.h>
 #include <sys/sysproto.h>
 #include <machine/armreg.h>
+#include <machine/pcb.h>
 #ifdef VFP
 #include <machine/vfp.h>
 #endif
diff --git a/sys/arm64/arm64/ptrace_machdep.c b/sys/arm64/arm64/ptrace_machdep.c
--- a/sys/arm64/arm64/ptrace_machdep.c
+++ b/sys/arm64/arm64/ptrace_machdep.c
@@ -48,6 +48,7 @@
 #include <sys/ucontext.h>
 
 #include <machine/armreg.h>
+#include <machine/pcb.h>
 
 /* Only used to get/set 32bits VFP regs */
 int
diff --git a/sys/arm64/include/proc.h b/sys/arm64/include/proc.h
--- a/sys/arm64/include/proc.h
+++ b/sys/arm64/include/proc.h
@@ -72,16 +72,4 @@
 #define	KINFO_PROC_SIZE	1088
 #define	KINFO_PROC32_SIZE 816
 
-#ifdef _KERNEL
-
-#include <machine/pcb.h>
-
-#define	GET_STACK_USAGE(total, used) do {				\
-	struct thread *td = curthread;					\
-	(total) = td->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb);	\
-	(used) = td->td_kstack + (total) - (vm_offset_t)&td;		\
-} while (0)
-
-#endif
-
 #endif /* !_MACHINE_PROC_H_ */
diff --git a/sys/arm64/include/stack.h b/sys/arm64/include/stack.h
--- a/sys/arm64/include/stack.h
+++ b/sys/arm64/include/stack.h
@@ -39,4 +39,22 @@
 
 bool unwind_frame(struct thread *, struct unwind_state *);
 
+#ifdef _SYS_PROC_H_
+
+#include <machine/pcb.h>
+
+#define	GET_STACK_USAGE(total, used) do {				\
+	struct thread *td = curthread;					\
+	(total) = td->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb);	\
+	(used) = td->td_kstack + (total) - (vm_offset_t)&td;		\
+} while (0)
+
+static __inline bool
+kstack_contains(struct thread *td, vm_offset_t va, size_t len)
+{
+	return (va >= td->td_kstack && va + len >= va &&
+	    va + len <= td->td_kstack + td->td_kstack_pages * PAGE_SIZE);
+}
+#endif	/* _SYS_PROC_H_ */
+
 #endif /* !_MACHINE_STACK_H_ */
diff --git a/sys/arm64/linux/linux_sysvec.c b/sys/arm64/linux/linux_sysvec.c
--- a/sys/arm64/linux/linux_sysvec.c
+++ b/sys/arm64/linux/linux_sysvec.c
@@ -71,7 +71,7 @@
 #include <arm64/linux/linux_sigframe.h>
 
 #include <machine/md_var.h>
-
+#include <machine/pcb.h>
 #ifdef VFP
 #include <machine/vfp.h>
 #endif
diff --git a/sys/ddb/db_ps.c b/sys/ddb/db_ps.c
--- a/sys/ddb/db_ps.c
+++ b/sys/ddb/db_ps.c
@@ -49,6 +49,8 @@
 
 #include <ddb/ddb.h>
 
+#include <machine/stack.h>
+
 #define PRINT_NONE	0
 #define PRINT_ARGS	1
 
diff --git a/sys/geom/geom_io.c b/sys/geom/geom_io.c
--- a/sys/geom/geom_io.c
+++ b/sys/geom/geom_io.c
@@ -53,6 +53,7 @@
 #include <sys/stack.h>
 #include <sys/sysctl.h>
 #include <sys/vmem.h>
+#include <machine/stack.h>
 #include <machine/stdarg.h>
 
 #include <sys/errno.h>
diff --git a/sys/i386/include/proc.h b/sys/i386/include/proc.h
--- a/sys/i386/include/proc.h
+++ b/sys/i386/include/proc.h
@@ -66,13 +66,6 @@
 
 #include <machine/md_var.h>
 
-/* Get the current kernel thread stack usage. */
-#define GET_STACK_USAGE(total, used) do {				\
-	struct thread	*td = curthread;				\
-	(total) = (vm_offset_t)get_pcb_td(td) - td->td_kstack;		\
-	(used) = (vm_offset_t)get_pcb_td(td) - (vm_offset_t)&td;	\
-} while (0)
-
 void 	set_user_ldt(struct mdproc *);
 struct 	proc_ldt *user_ldt_alloc(struct mdproc *, int);
 void 	user_ldt_free(struct thread *);
diff --git a/sys/i386/include/stack.h b/sys/i386/include/stack.h
--- a/sys/i386/include/stack.h
+++ b/sys/i386/include/stack.h
@@ -3,4 +3,27 @@
  */
 /* $FreeBSD$ */
 
+#ifndef _MACHINE_STACK_H_
+#define	_MACHINE_STACK_H_
+
 #include <x86/stack.h>
+
+#ifdef _SYS_PROC_H_
+
+/* Get the current kernel thread stack usage. */
+#define GET_STACK_USAGE(total, used) do {				\
+	struct thread	*td = curthread;				\
+	(total) = (vm_offset_t)get_pcb_td(td) - td->td_kstack;		\
+	(used) = (vm_offset_t)get_pcb_td(td) - (vm_offset_t)&td;	\
+} while (0)
+
+static __inline bool
+kstack_contains(struct thread *td, vm_offset_t va, size_t len)
+{
+	return (va >= td->td_kstack && va + len >= va &&
+	    va + len <= td->td_kstack + td->td_kstack_pages * PAGE_SIZE);
+}
+
+#endif	/* _SYS_PROC_H_ */
+
+#endif
diff --git a/sys/kern/subr_epoch.c b/sys/kern/subr_epoch.c
--- a/sys/kern/subr_epoch.c
+++ b/sys/kern/subr_epoch.c
@@ -56,6 +56,8 @@
 #include <vm/vm_kern.h>
 #include <vm/uma.h>
 
+#include <machine/stack.h>
+
 #include <ck_epoch.h>
 
 #ifdef __amd64__
diff --git a/sys/netgraph/ng_base.c b/sys/netgraph/ng_base.c
--- a/sys/netgraph/ng_base.c
+++ b/sys/netgraph/ng_base.c
@@ -66,6 +66,8 @@
 #include <machine/cpu.h>
 #include <vm/uma.h>
 
+#include <machine/stack.h>
+
 #include <net/netisr.h>
 #include <net/vnet.h>
 
diff --git a/sys/powerpc/include/proc.h b/sys/powerpc/include/proc.h
--- a/sys/powerpc/include/proc.h
+++ b/sys/powerpc/include/proc.h
@@ -59,16 +59,4 @@
 #define	KINFO_PROC_SIZE 816
 #endif
 
-#ifdef _KERNEL
-
-#include <machine/pcb.h>
-
-/* Get the current kernel thread stack usage. */
-#define	GET_STACK_USAGE(total, used) do {				\
-	struct thread *td = curthread;					\
-	(total) = td->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb);	\
-	(used) = td->td_kstack + (total) - (vm_offset_t)&td;		\
-} while (0)
-#endif
-
 #endif /* !_MACHINE_PROC_H_ */
diff --git a/sys/powerpc/include/stack.h b/sys/powerpc/include/stack.h
--- a/sys/powerpc/include/stack.h
+++ b/sys/powerpc/include/stack.h
@@ -33,4 +33,23 @@
 extern int asttrapexit[];
 extern int end[];
 
+#ifdef _SYS_PROC_H_
+
+#include <machine/pcb.h>
+
+/* Get the current kernel thread stack usage. */
+#define	GET_STACK_USAGE(total, used) do {				\
+	struct thread *td = curthread;					\
+	(total) = td->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb);	\
+	(used) = td->td_kstack + (total) - (vm_offset_t)&td;		\
+} while (0)
+
+static __inline bool
+kstack_contains(struct thread *td, vm_offset_t va, size_t len)
+{
+	return (va >= td->td_kstack && va + len >= va &&
+	    va + len <= td->td_kstack + td->td_kstack_pages * PAGE_SIZE);
+}
+#endif	/* _SYS_PROC_H_ */
+
 #endif /* !_MACHINE_STACK_H_ */
diff --git a/sys/riscv/include/proc.h b/sys/riscv/include/proc.h
--- a/sys/riscv/include/proc.h
+++ b/sys/riscv/include/proc.h
@@ -45,15 +45,4 @@
 
 #define	KINFO_PROC_SIZE	1088
 
-#ifdef _KERNEL
-#include <machine/pcb.h>
-
-/* Get the current kernel thread stack usage. */
-#define	GET_STACK_USAGE(total, used) do {				\
-	struct thread *td = curthread;					\
-	(total) = td->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb);	\
-	(used) = td->td_kstack + (total) - (vm_offset_t)&td;		\
-} while (0)
-
-#endif  /* _KERNEL */
 #endif /* !_MACHINE_PROC_H_ */
diff --git a/sys/riscv/include/stack.h b/sys/riscv/include/stack.h
--- a/sys/riscv/include/stack.h
+++ b/sys/riscv/include/stack.h
@@ -48,4 +48,23 @@
 
 bool unwind_frame(struct thread *, struct unwind_state *);
 
+#ifdef _SYS_PROC_H_
+
+#include <machine/pcb.h>
+
+/* Get the current kernel thread stack usage. */
+#define	GET_STACK_USAGE(total, used) do {				\
+	struct thread *td = curthread;					\
+	(total) = td->td_kstack_pages * PAGE_SIZE - sizeof(struct pcb);	\
+	(used) = td->td_kstack + (total) - (vm_offset_t)&td;		\
+} while (0)
+
+static __inline bool
+kstack_contains(struct thread *td, vm_offset_t va, size_t len)
+{
+	return (va >= td->td_kstack && va + len >= va &&
+	    va + len <= td->td_kstack + td->td_kstack_pages * PAGE_SIZE);
+}
+#endif	/* _SYS_PROC_H_ */
+
 #endif /* !_MACHINE_STACK_H_ */
diff --git a/sys/sys/proc.h b/sys/sys/proc.h
--- a/sys/sys/proc.h
+++ b/sys/sys/proc.h
@@ -1324,13 +1324,6 @@
 	curthread->td_pflags2 &= save;
 }
 
-static __inline bool
-kstack_contains(struct thread *td, vm_offset_t va, size_t len)
-{
-	return (va >= td->td_kstack && va + len >= va &&
-	    va + len <= td->td_kstack + td->td_kstack_pages * PAGE_SIZE);
-}
-
 static __inline __pure2 struct td_sched *
 td_get_sched(struct thread *td)
 {
diff --git a/sys/x86/x86/stack_machdep.c b/sys/x86/x86/stack_machdep.c
--- a/sys/x86/x86/stack_machdep.c
+++ b/sys/x86/x86/stack_machdep.c
@@ -45,7 +45,7 @@
 #include <vm/vm_param.h>
 #include <vm/pmap.h>
 
-#include <x86/stack.h>
+#include <machine/stack.h>
 
 #ifdef __i386__
 #define	PCB_FP(pcb)	((pcb)->pcb_ebp)