Merge branch 'master' of git.cs.vt.edu:gback/pintos-2017

doc/threads.texi

@@ -608,7 +608,7 @@ that no CPU is idle while threads are ready to run (but are not currently
 running). However, using a global queue has two main weaknesses.
 
 The first weakness is lack of
-scability. The global queue must be locked while choosing the next job to run.
+scalability. The global queue must be locked while choosing the next job to run.
 Locking greatly reduces performance as the number of CPUs grows. Each CPU will
 spend more and more time contenting for the global queue lock and less time
 actually running threads.

src/devices/usb_uhci.c

@@ -724,7 +724,7 @@ uhci_add_td_to_qh (struct queue_head *qh, struct tx_descriptor *td)
     {
       /* queue is empty */
       td->flp.terminate = 1;
-      barrier ();
+      smp_barrier ();
       td->flp.flp = 0;
       qh->qelp.flp = ptr_to_flp (vtop (td));
       qh->qelp.terminate = 0;
@@ -745,7 +745,7 @@ uhci_add_td_to_qh (struct queue_head *qh, struct tx_descriptor *td)
       fp->qh_select = 0;
       fp->depth_select = 0;
       fp->flp = ptr_to_flp (vtop (td));
-      barrier ();
+      smp_barrier ();
       fp->terminate = 0;
     }
 }
@@ -928,7 +928,7 @@ uhci_create_chan (host_info hi, int dev_addr, int ver)
   /* queue data */
   memset (ud->qh, 0, sizeof (*ud->qh));
   ud->qh->qelp.terminate = 1;
-  barrier ();
+  smp_barrier ();
   ud->qh->qelp.flp = 0;
   ud->qh->qelp.qh_select = 0;
   ud->qh->qhlp.qh_select = 1;

src/threads/synch.h

@@ -52,4 +52,12 @@ void cond_broadcast (struct condition *, struct lock *);
    reference guide for more information.*/
 #define barrier() asm volatile ("" : : : "memory")
 
+/* Memory barrier.
+
+   The hardware will not complete instructions out of 
+   order around a memory barrier. The compiler will 
+   similarly refrain from reordering operations across
+   a memory barrier */
+#define smp_barrier() asm volatile("mfence" : : : "memory")
+
 #endif /* threads/synch.h */

src/userprog/pagedir.c

@@ -297,7 +297,7 @@ invalidate_pagedir_others (uint32_t *pd)
   lock_acquire (&tlb_flush_state.lock);
   tlb_flush_state.remaining = ncpu - 1;
   tlb_flush_state.pd = pd;
-  barrier ();
+  smp_barrier ();
   lapic_send_ipi_to_all_but_self (IPI_TLB);
 
   /* We busy-wait here rather than blocking the calling thread