[PATCH] rcu: join rcu_ctrlblk and rcu_state
authorOleg Nesterov <oleg@tv-sign.ru>
Tue, 10 Jan 2006 13:48:02 +0000 (16:48 +0300)
committerLinus Torvalds <torvalds@g5.osdl.org>
Tue, 10 Jan 2006 16:42:50 +0000 (08:42 -0800)
This patch moves rcu_state into the rcu_ctrlblk. I think there
are no reasons why we should have 2 different variables to control
rcu state. Every user of rcu_state has also "rcu_ctrlblk *rcp" in
the parameter list.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
include/linux/rcupdate.h
kernel/rcupdate.c

index a1d26cb28925dea07283800ff4a3888f79face2b..981f9aa433538a8ec2e8b0142da769b2e5a4dd87 100644 (file)
@@ -65,6 +65,10 @@ struct rcu_ctrlblk {
        long    cur;            /* Current batch number.                      */
        long    completed;      /* Number of the last completed batch         */
        int     next_pending;   /* Is the next batch already waiting?         */
+
+       spinlock_t      lock    ____cacheline_internodealigned_in_smp;
+       cpumask_t       cpumask; /* CPUs that need to switch in order    */
+                                /* for current batch to proceed.        */
 } ____cacheline_internodealigned_in_smp;
 
 /* Is batch a before batch b ? */
index 05ee48316f70097f386e9edefe9962a5d66ec51c..e18f9190eafa112a9a1b08b25a608ab4dcb733cd 100644 (file)
 #include <linux/cpu.h>
 
 /* Definition for rcupdate control block. */
-struct rcu_ctrlblk rcu_ctrlblk = 
-       { .cur = -300, .completed = -300 };
-struct rcu_ctrlblk rcu_bh_ctrlblk =
-       { .cur = -300, .completed = -300 };
-
-/* Bookkeeping of the progress of the grace period */
-struct rcu_state {
-       spinlock_t      lock; /* Guard this struct and writes to rcu_ctrlblk */
-       cpumask_t       cpumask; /* CPUs that need to switch in order    */
-                                     /* for current batch to proceed.        */
+struct rcu_ctrlblk rcu_ctrlblk = {
+       .cur = -300,
+       .completed = -300,
+       .lock = SPIN_LOCK_UNLOCKED,
+       .cpumask = CPU_MASK_NONE,
+};
+struct rcu_ctrlblk rcu_bh_ctrlblk = {
+       .cur = -300,
+       .completed = -300,
+       .lock = SPIN_LOCK_UNLOCKED,
+       .cpumask = CPU_MASK_NONE,
 };
-
-static struct rcu_state rcu_state ____cacheline_internodealigned_in_smp =
-         {.lock = SPIN_LOCK_UNLOCKED, .cpumask = CPU_MASK_NONE };
-static struct rcu_state rcu_bh_state ____cacheline_internodealigned_in_smp =
-         {.lock = SPIN_LOCK_UNLOCKED, .cpumask = CPU_MASK_NONE };
 
 DEFINE_PER_CPU(struct rcu_data, rcu_data) = { 0L };
 DEFINE_PER_CPU(struct rcu_data, rcu_bh_data) = { 0L };
@@ -220,13 +216,13 @@ static void rcu_do_batch(struct rcu_data *rdp)
  *   This is done by rcu_start_batch. The start is not broadcasted to
  *   all cpus, they must pick this up by comparing rcp->cur with
  *   rdp->quiescbatch. All cpus are recorded  in the
- *   rcu_state.cpumask bitmap.
+ *   rcu_ctrlblk.cpumask bitmap.
  * - All cpus must go through a quiescent state.
  *   Since the start of the grace period is not broadcasted, at least two
  *   calls to rcu_check_quiescent_state are required:
  *   The first call just notices that a new grace period is running. The
  *   following calls check if there was a quiescent state since the beginning
- *   of the grace period. If so, it updates rcu_state.cpumask. If
+ *   of the grace period. If so, it updates rcu_ctrlblk.cpumask. If
  *   the bitmap is empty, then the grace period is completed.
  *   rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace
  *   period (if necessary).
@@ -234,9 +230,9 @@ static void rcu_do_batch(struct rcu_data *rdp)
 /*
  * Register a new batch of callbacks, and start it up if there is currently no
  * active batch and the batch to be registered has not already occurred.
- * Caller must hold rcu_state.lock.
+ * Caller must hold rcu_ctrlblk.lock.
  */
-static void rcu_start_batch(struct rcu_ctrlblk *rcp, struct rcu_state *rsp)
+static void rcu_start_batch(struct rcu_ctrlblk *rcp)
 {
        if (rcp->next_pending &&
                        rcp->completed == rcp->cur) {
@@ -251,11 +247,11 @@ static void rcu_start_batch(struct rcu_ctrlblk *rcp, struct rcu_state *rsp)
                /*
                 * Accessing nohz_cpu_mask before incrementing rcp->cur needs a
                 * Barrier  Otherwise it can cause tickless idle CPUs to be
-                * included in rsp->cpumask, which will extend graceperiods
+                * included in rcp->cpumask, which will extend graceperiods
                 * unnecessarily.
                 */
                smp_mb();
-               cpus_andnot(rsp->cpumask, cpu_online_map, nohz_cpu_mask);
+               cpus_andnot(rcp->cpumask, cpu_online_map, nohz_cpu_mask);
 
        }
 }
@@ -265,13 +261,13 @@ static void rcu_start_batch(struct rcu_ctrlblk *rcp, struct rcu_state *rsp)
  * Clear it from the cpu mask and complete the grace period if it was the last
  * cpu. Start another grace period if someone has further entries pending
  */
-static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp, struct rcu_state *rsp)
+static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp)
 {
-       cpu_clear(cpu, rsp->cpumask);
-       if (cpus_empty(rsp->cpumask)) {
+       cpu_clear(cpu, rcp->cpumask);
+       if (cpus_empty(rcp->cpumask)) {
                /* batch completed ! */
                rcp->completed = rcp->cur;
-               rcu_start_batch(rcp, rsp);
+               rcu_start_batch(rcp);
        }
 }
 
@@ -281,7 +277,7 @@ static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp, struct rcu_state *rsp)
  * quiescent cycle, then indicate that it has done so.
  */
 static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp,
-                       struct rcu_state *rsp, struct rcu_data *rdp)
+                                       struct rcu_data *rdp)
 {
        if (rdp->quiescbatch != rcp->cur) {
                /* start new grace period: */
@@ -306,15 +302,15 @@ static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp,
                return;
        rdp->qs_pending = 0;
 
-       spin_lock(&rsp->lock);
+       spin_lock(&rcp->lock);
        /*
         * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync
         * during cpu startup. Ignore the quiescent state.
         */
        if (likely(rdp->quiescbatch == rcp->cur))
-               cpu_quiet(rdp->cpu, rcp, rsp);
+               cpu_quiet(rdp->cpu, rcp);
 
-       spin_unlock(&rsp->lock);
+       spin_unlock(&rcp->lock);
 }
 
 
@@ -335,16 +331,16 @@ static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list,
 }
 
 static void __rcu_offline_cpu(struct rcu_data *this_rdp,
-       struct rcu_ctrlblk *rcp, struct rcu_state *rsp, struct rcu_data *rdp)
+                               struct rcu_ctrlblk *rcp, struct rcu_data *rdp)
 {
        /* if the cpu going offline owns the grace period
         * we can block indefinitely waiting for it, so flush
         * it here
         */
-       spin_lock_bh(&rsp->lock);
+       spin_lock_bh(&rcp->lock);
        if (rcp->cur != rcp->completed)
-               cpu_quiet(rdp->cpu, rcp, rsp);
-       spin_unlock_bh(&rsp->lock);
+               cpu_quiet(rdp->cpu, rcp);
+       spin_unlock_bh(&rcp->lock);
        rcu_move_batch(this_rdp, rdp->curlist, rdp->curtail);
        rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail);
 
@@ -354,9 +350,9 @@ static void rcu_offline_cpu(int cpu)
        struct rcu_data *this_rdp = &get_cpu_var(rcu_data);
        struct rcu_data *this_bh_rdp = &get_cpu_var(rcu_bh_data);
 
-       __rcu_offline_cpu(this_rdp, &rcu_ctrlblk, &rcu_state,
+       __rcu_offline_cpu(this_rdp, &rcu_ctrlblk,
                                        &per_cpu(rcu_data, cpu));
-       __rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk, &rcu_bh_state,
+       __rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk,
                                        &per_cpu(rcu_bh_data, cpu));
        put_cpu_var(rcu_data);
        put_cpu_var(rcu_bh_data);
@@ -375,7 +371,7 @@ static void rcu_offline_cpu(int cpu)
  * This does the RCU processing work from tasklet context. 
  */
 static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp,
-                       struct rcu_state *rsp, struct rcu_data *rdp)
+                                       struct rcu_data *rdp)
 {
        if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) {
                *rdp->donetail = rdp->curlist;
@@ -405,25 +401,23 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp,
 
                if (!rcp->next_pending) {
                        /* and start it/schedule start if it's a new batch */
-                       spin_lock(&rsp->lock);
+                       spin_lock(&rcp->lock);
                        rcp->next_pending = 1;
-                       rcu_start_batch(rcp, rsp);
-                       spin_unlock(&rsp->lock);
+                       rcu_start_batch(rcp);
+                       spin_unlock(&rcp->lock);
                }
        } else {
                local_irq_enable();
        }
-       rcu_check_quiescent_state(rcp, rsp, rdp);
+       rcu_check_quiescent_state(rcp, rdp);
        if (rdp->donelist)
                rcu_do_batch(rdp);
 }
 
 static void rcu_process_callbacks(unsigned long unused)
 {
-       __rcu_process_callbacks(&rcu_ctrlblk, &rcu_state,
-                               &__get_cpu_var(rcu_data));
-       __rcu_process_callbacks(&rcu_bh_ctrlblk, &rcu_bh_state,
-                               &__get_cpu_var(rcu_bh_data));
+       __rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data));
+       __rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data));
 }
 
 static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp)