diff --git a/backend.c b/backend.c
index c397eb1..5fc0fc8 100644
--- a/backend.c
+++ b/backend.c
@@ -1205,6 +1205,9 @@
 	}
 #endif
 
+	if (fio_pin_memory(td))
+		goto err;
+
 	/*
 	 * May alter parameters that init_io_u() will use, so we need to
 	 * do this first.
@@ -1330,6 +1333,8 @@
 	td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE];
 	td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM];
 
+	fio_unpin_memory(td);
+
 	fio_mutex_down(writeout_mutex);
 	if (td->bw_log) {
 		if (td->o.bw_log_file) {
@@ -1545,9 +1550,6 @@
 	unsigned long spent;
 	unsigned int i, todo, nr_running, m_rate, t_rate, nr_started;
 
-	if (fio_pin_memory())
-		return;
-
 	if (fio_gtod_offload && fio_start_gtod_thread())
 		return;
 	
@@ -1780,7 +1782,6 @@
 	fio_idle_prof_stop();
 
 	update_io_ticks();
-	fio_unpin_memory();
 }
 
 void wait_for_disk_thread_exit(void)
diff --git a/fio.h b/fio.h
index e55413a..2c8d904 100644
--- a/fio.h
+++ b/fio.h
@@ -308,6 +308,8 @@
 	 */
 	struct prof_io_ops prof_io_ops;
 	void *prof_data;
+
+	void *pinned_mem;
 };
 
 /*
@@ -348,7 +350,6 @@
 extern int groupid;
 extern int output_format;
 extern int temp_stall_ts;
-extern unsigned long long mlock_size;
 extern uintptr_t page_mask, page_size;
 extern int read_only;
 extern int eta_print;
@@ -481,8 +482,8 @@
 /*
  * Memory helpers
  */
-extern int __must_check fio_pin_memory(void);
-extern void fio_unpin_memory(void);
+extern int __must_check fio_pin_memory(struct thread_data *);
+extern void fio_unpin_memory(struct thread_data *);
 extern int __must_check allocate_io_mem(struct thread_data *);
 extern void free_io_mem(struct thread_data *);
 
diff --git a/init.c b/init.c
index 44c74e3..9e2d113 100644
--- a/init.c
+++ b/init.c
@@ -45,7 +45,6 @@
 int output_format = FIO_OUTPUT_NORMAL;
 int eta_print = FIO_ETA_AUTO;
 int eta_new_line = 0;
-unsigned long long mlock_size = 0;
 FILE *f_out = NULL;
 FILE *f_err = NULL;
 char **job_sections = NULL;
diff --git a/memory.c b/memory.c
index 443d71d..9e751be 100644
--- a/memory.c
+++ b/memory.c
@@ -12,51 +12,49 @@
 #include <sys/shm.h>
 #endif
 
-static void *pinned_mem;
-
-void fio_unpin_memory(void)
+void fio_unpin_memory(struct thread_data *td)
 {
-	if (pinned_mem) {
-		dprint(FD_MEM, "unpinning %llu bytes\n", mlock_size);
-		if (munlock(pinned_mem, mlock_size) < 0)
+	if (td->pinned_mem) {
+		dprint(FD_MEM, "unpinning %llu bytes\n", td->o.lockmem);
+		if (munlock(td->pinned_mem, td->o.lockmem) < 0)
 			perror("munlock");
-		munmap(pinned_mem, mlock_size);
-		pinned_mem = NULL;
+		munmap(td->pinned_mem, td->o.lockmem);
+		td->pinned_mem = NULL;
 	}
 }
 
-int fio_pin_memory(void)
+int fio_pin_memory(struct thread_data *td)
 {
 	unsigned long long phys_mem;
 
-	if (!mlock_size)
+	if (!td->o.lockmem)
 		return 0;
 
-	dprint(FD_MEM, "pinning %llu bytes\n", mlock_size);
+	dprint(FD_MEM, "pinning %llu bytes\n", td->o.lockmem);
 
 	/*
 	 * Don't allow mlock of more than real_mem-128MB
 	 */
 	phys_mem = os_phys_mem();
 	if (phys_mem) {
-		if ((mlock_size + 128 * 1024 * 1024) > phys_mem) {
-			mlock_size = phys_mem - 128 * 1024 * 1024;
+		if ((td->o.lockmem + 128 * 1024 * 1024) > phys_mem) {
+			td->o.lockmem = phys_mem - 128 * 1024 * 1024;
 			log_info("fio: limiting mlocked memory to %lluMB\n",
-							mlock_size >> 20);
+							td->o.lockmem >> 20);
 		}
 	}
 
-	pinned_mem = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE,
+	td->pinned_mem = mmap(NULL, td->o.lockmem, PROT_READ | PROT_WRITE,
 				MAP_PRIVATE | OS_MAP_ANON, -1, 0);
-	if (pinned_mem == MAP_FAILED) {
+	if (td->pinned_mem == MAP_FAILED) {
 		perror("malloc locked mem");
-		pinned_mem = NULL;
+		td->pinned_mem = NULL;
 		return 1;
 	}
-	if (mlock(pinned_mem, mlock_size) < 0) {
+	if (mlock(td->pinned_mem, td->o.lockmem) < 0) {
 		perror("mlock");
-		munmap(pinned_mem, mlock_size);
-		pinned_mem = NULL;
+		munmap(td->pinned_mem, td->o.lockmem);
+		td->pinned_mem = NULL;
 		return 1;
 	}