Fix some problems introduced with the last descriptors tables locking

patch:
- Do the correct test for ldt allocation
- Drop dt_lock just before to call kmem_free (since it acquires blocking
  locks inside)
- Solve a deadlock with smp_rendezvous() where other CPU will wait
  undefinitively for dt_lock acquisition.
- Add dt_lock in the WITNESS list of spinlocks

While applying these modifies, change the requirement for user_ldt_free()
making that returning without dt_lock held.

Tested by: marcus, tegge
Reviewed by: tegge
Approved by: jeff (mentor)

4 files changed, 22 insertions, 8 deletions

diff --git a/sys/i386/i386/machdep.c b/sys/i386/i386/machdep.c
index 4b04ab9162a7..cc6984ef15ec 100644
--- a/sys/i386/i386/machdep.c
+++ b/sys/i386/i386/machdep.c
@@ -1174,7 +1174,8 @@ exec_setregs(td, entry, stack, ps_strings)
 	mtx_lock_spin(&dt_lock);
 	if (td->td_proc->p_md.md_ldt)
 		user_ldt_free(td);
-	mtx_unlock_spin(&dt_lock);
+	else
+		mtx_unlock_spin(&dt_lock);
   
 	bzero((char *)regs, sizeof(struct trapframe));
 	regs->tf_eip = entry;
diff --git a/sys/i386/i386/sys_machdep.c b/sys/i386/i386/sys_machdep.c
index 74a7e6c3d18d..31fad625d831 100644
--- a/sys/i386/i386/sys_machdep.c
+++ b/sys/i386/i386/sys_machdep.c
@@ -428,7 +428,7 @@ user_ldt_alloc(struct mdproc *mdp, int len)
 }
 
 /*
- * Must be called with dt_lock held.
+ * Must be called with dt_lock held.  Returns with dt_lock unheld.
  */
 void
 user_ldt_free(struct thread *td)
@@ -446,6 +446,7 @@ user_ldt_free(struct thread *td)
 	}
 
 	mdp->md_ldt = NULL;
+	mtx_unlock_spin(&dt_lock);
 	if (refcount_release(&pldt->ldt_refcnt)) {
 		kmem_free(kernel_map, (vm_offset_t)pldt->ldt_base,
 			pldt->ldt_len * sizeof(union descriptor));
@@ -701,7 +702,7 @@ i386_ldt_grow(struct thread *td, int len)
 		len = NLDT + 1;
 
 	/* Allocate a user ldt. */
-	if ((pldt = mdp->md_ldt) != NULL || len > pldt->ldt_len) {
+	if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) {
 		struct proc_ldt *new_ldt;
 
 		new_ldt = user_ldt_alloc(mdp, len);
@@ -716,26 +717,37 @@ i386_ldt_grow(struct thread *td, int len)
 				pldt->ldt_sd = new_ldt->ldt_sd;
 				pldt->ldt_base = new_ldt->ldt_base;
 				pldt->ldt_len = new_ldt->ldt_len;
+				mtx_unlock_spin(&dt_lock);
 				kmem_free(kernel_map, (vm_offset_t)old_ldt_base,
 					old_ldt_len * sizeof(union descriptor));
 				FREE(new_ldt, M_SUBPROC);
+				mtx_lock_spin(&dt_lock);
 			} else {
 				/*
 				 * If other threads already did the work,
 				 * do nothing.
 				 */
+				mtx_unlock_spin(&dt_lock);
 				kmem_free(kernel_map,
 				   (vm_offset_t)new_ldt->ldt_base,
 				   new_ldt->ldt_len * sizeof(union descriptor));
 				FREE(new_ldt, M_SUBPROC);
+				mtx_lock_spin(&dt_lock);
 				return (0);
 			}
 		} else
 			mdp->md_ldt = pldt = new_ldt;
 #ifdef SMP
-		/* signal other cpus to reload ldt */
+		/*
+		 * Signal other cpus to reload ldt.  We need to unlock dt_lock
+		 * here because other CPU will contest on it since their
+		 * curthreads won't hold the lock and will block when trying
+		 * to acquire it.
+		 */
+		mtx_unlock_spin(&dt_lock);
 		smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv,
 		    NULL, td);
+		mtx_lock_spin(&dt_lock);
 #else
 		set_user_ldt(mdp);
 #endif
diff --git a/sys/i386/i386/vm_machdep.c b/sys/i386/i386/vm_machdep.c
index 116f1dfcccef..73ce4b741d1b 100644
--- a/sys/i386/i386/vm_machdep.c
+++ b/sys/i386/i386/vm_machdep.c
@@ -168,8 +168,8 @@ cpu_fork(td1, p2, td2, flags)
 				mdp1->md_ldt = pldt;
 				set_user_ldt(mdp1);
 				user_ldt_free(td1);
-			}
-			mtx_unlock_spin(&dt_lock);
+			} else
+				mtx_unlock_spin(&dt_lock);
 		}
 		return;
 	}
@@ -312,8 +312,8 @@ cpu_exit(struct thread *td)
 		td->td_pcb->pcb_gs = _udatasel;
 		load_gs(_udatasel);
 		user_ldt_free(td);
-	}
-	mtx_unlock_spin(&dt_lock);
+	} else
+		mtx_unlock_spin(&dt_lock);
 }
 
 void
diff --git a/sys/kern/subr_witness.c b/sys/kern/subr_witness.c
index 1f750e515b2b..606987090cb8 100644
--- a/sys/kern/subr_witness.c
+++ b/sys/kern/subr_witness.c
@@ -394,6 +394,7 @@ static struct witness_order_list_entry order_lists[] = {
 	{ "sio", &lock_class_mtx_spin },
 #ifdef __i386__
 	{ "cy", &lock_class_mtx_spin },
+	{ "descriptor tables", &lock_class_mtx_spin },
 #endif
 	{ "scc_hwmtx", &lock_class_mtx_spin },
 	{ "uart_hwmtx", &lock_class_mtx_spin },