[ubicom32]: move new files out from platform support patch
[openwrt/svn-archive/archive.git] / target / linux / ubicom32 / files / arch / ubicom32 / kernel / semaphore.c
diff --git a/target/linux/ubicom32/files/arch/ubicom32/kernel/semaphore.c b/target/linux/ubicom32/files/arch/ubicom32/kernel/semaphore.c
new file mode 100644 (file)
index 0000000..d996ac2
--- /dev/null
@@ -0,0 +1,159 @@
+/*
+ * arch/ubicom32/kernel/semaphore.c
+ *   Ubicom32 architecture semaphore implementation.
+ *
+ * (C) Copyright 2009, Ubicom, Inc.
+ *
+ * This file is part of the Ubicom32 Linux Kernel Port.
+ *
+ * The Ubicom32 Linux Kernel Port is free software: you can redistribute
+ * it and/or modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation, either version 2 of the
+ * License, or (at your option) any later version.
+ *
+ * The Ubicom32 Linux Kernel Port is distributed in the hope that it
+ * will be useful, but WITHOUT ANY WARRANTY; without even the implied
+ * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
+ * the GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with the Ubicom32 Linux Kernel Port.  If not,
+ * see <http://www.gnu.org/licenses/>.
+ *
+ * Ubicom32 implementation derived from (with many thanks):
+ *   arch/m68knommu
+ *   arch/blackfin
+ *   arch/parisc
+ */
+/*
+ *  Generic semaphore code. Buyer beware. Do your own
+ * specific changes in <asm/semaphore-helper.h>
+ */
+
+#include <linux/sched.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <asm/semaphore-helper.h>
+
+#ifndef CONFIG_RMW_INSNS
+spinlock_t semaphore_wake_lock;
+#endif
+
+/*
+ * Semaphores are implemented using a two-way counter:
+ * The "count" variable is decremented for each process
+ * that tries to sleep, while the "waking" variable is
+ * incremented when the "up()" code goes to wake up waiting
+ * processes.
+ *
+ * Notably, the inline "up()" and "down()" functions can
+ * efficiently test if they need to do any extra work (up
+ * needs to do something only if count was negative before
+ * the increment operation.
+ *
+ * waking_non_zero() (from asm/semaphore.h) must execute
+ * atomically.
+ *
+ * When __up() is called, the count was negative before
+ * incrementing it, and we need to wake up somebody.
+ *
+ * This routine adds one to the count of processes that need to
+ * wake up and exit.  ALL waiting processes actually wake up but
+ * only the one that gets to the "waking" field first will gate
+ * through and acquire the semaphore.  The others will go back
+ * to sleep.
+ *
+ * Note that these functions are only called when there is
+ * contention on the lock, and as such all this is the
+ * "non-critical" part of the whole semaphore business. The
+ * critical part is the inline stuff in <asm/semaphore.h>
+ * where we want to avoid any extra jumps and calls.
+ */
+void __up(struct semaphore *sem)
+{
+       wake_one_more(sem);
+       wake_up(&sem->wait);
+}
+
+/*
+ * Perform the "down" function.  Return zero for semaphore acquired,
+ * return negative for signalled out of the function.
+ *
+ * If called from __down, the return is ignored and the wait loop is
+ * not interruptible.  This means that a task waiting on a semaphore
+ * using "down()" cannot be killed until someone does an "up()" on
+ * the semaphore.
+ *
+ * If called from __down_interruptible, the return value gets checked
+ * upon return.  If the return value is negative then the task continues
+ * with the negative value in the return register (it can be tested by
+ * the caller).
+ *
+ * Either form may be used in conjunction with "up()".
+ *
+ */
+
+
+#define DOWN_HEAD(task_state)                                          \
+                                                                       \
+                                                                       \
+       current->state = (task_state);                                  \
+       add_wait_queue(&sem->wait, &wait);                              \
+                                                                       \
+       /*                                                              \
+        * Ok, we're set up.  sem->count is known to be less than zero  \
+        * so we must wait.                                             \
+        *                                                              \
+        * We can let go the lock for purposes of waiting.              \
+        * We re-acquire it after awaking so as to protect              \
+        * all semaphore operations.                                    \
+        *                                                              \
+        * If "up()" is called before we call waking_non_zero() then    \
+        * we will catch it right away.  If it is called later then     \
+        * we will have to go through a wakeup cycle to catch it.       \
+        *                                                              \
+        * Multiple waiters contend for the semaphore lock to see       \
+        * who gets to gate through and who has to wait some more.      \
+        */                                                             \
+       for (;;) {
+
+#define DOWN_TAIL(task_state)                  \
+               current->state = (task_state);  \
+       }                                       \
+       current->state = TASK_RUNNING;          \
+       remove_wait_queue(&sem->wait, &wait);
+
+void __sched __down(struct semaphore *sem)
+{
+       DECLARE_WAITQUEUE(wait, current);
+
+       DOWN_HEAD(TASK_UNINTERRUPTIBLE)
+       if (waking_non_zero(sem))
+               break;
+       schedule();
+       DOWN_TAIL(TASK_UNINTERRUPTIBLE)
+}
+
+int __sched __down_interruptible(struct semaphore *sem)
+{
+       DECLARE_WAITQUEUE(wait, current);
+       int ret = 0;
+
+       DOWN_HEAD(TASK_INTERRUPTIBLE)
+
+       ret = waking_non_zero_interruptible(sem, current);
+       if (ret) {
+               if (ret == 1)
+                       /* ret != 0 only if we get interrupted -arca */
+                       ret = 0;
+               break;
+       }
+       schedule();
+       DOWN_TAIL(TASK_INTERRUPTIBLE)
+       return ret;
+}
+
+int __down_trylock(struct semaphore *sem)
+{
+       return waking_non_zero_trylock(sem);
+}