PM: Add user-space wake lock api.
[linux-2.6.git] / kernel / rcutree.h
index 5e872bb..01b2ccd 100644 (file)
+/*
+ * Read-Copy Update mechanism for mutual exclusion (tree-based version)
+ * Internal non-public definitions.
+ *
+ * This program 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.
+ *
+ * This program 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 this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright IBM Corporation, 2008
+ *
+ * Author: Ingo Molnar <mingo@elte.hu>
+ *        Paul E. McKenney <paulmck@linux.vnet.ibm.com>
+ */
+
+#include <linux/cache.h>
+#include <linux/spinlock.h>
+#include <linux/threads.h>
+#include <linux/cpumask.h>
+#include <linux/seqlock.h>
+
+/*
+ * Define shape of hierarchy based on NR_CPUS and CONFIG_RCU_FANOUT.
+ * In theory, it should be possible to add more levels straightforwardly.
+ * In practice, this did work well going from three levels to four.
+ * Of course, your mileage may vary.
+ */
+#define MAX_RCU_LVLS 4
+#if CONFIG_RCU_FANOUT > 16
+#define RCU_FANOUT_LEAF       16
+#else /* #if CONFIG_RCU_FANOUT > 16 */
+#define RCU_FANOUT_LEAF       (CONFIG_RCU_FANOUT)
+#endif /* #else #if CONFIG_RCU_FANOUT > 16 */
+#define RCU_FANOUT_1         (RCU_FANOUT_LEAF)
+#define RCU_FANOUT_2         (RCU_FANOUT_1 * CONFIG_RCU_FANOUT)
+#define RCU_FANOUT_3         (RCU_FANOUT_2 * CONFIG_RCU_FANOUT)
+#define RCU_FANOUT_4         (RCU_FANOUT_3 * CONFIG_RCU_FANOUT)
+
+#if NR_CPUS <= RCU_FANOUT_1
+#  define NUM_RCU_LVLS       1
+#  define NUM_RCU_LVL_0              1
+#  define NUM_RCU_LVL_1              (NR_CPUS)
+#  define NUM_RCU_LVL_2              0
+#  define NUM_RCU_LVL_3              0
+#  define NUM_RCU_LVL_4              0
+#elif NR_CPUS <= RCU_FANOUT_2
+#  define NUM_RCU_LVLS       2
+#  define NUM_RCU_LVL_0              1
+#  define NUM_RCU_LVL_1              DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_LVL_2              (NR_CPUS)
+#  define NUM_RCU_LVL_3              0
+#  define NUM_RCU_LVL_4              0
+#elif NR_CPUS <= RCU_FANOUT_3
+#  define NUM_RCU_LVLS       3
+#  define NUM_RCU_LVL_0              1
+#  define NUM_RCU_LVL_1              DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
+#  define NUM_RCU_LVL_2              DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_LVL_3              (NR_CPUS)
+#  define NUM_RCU_LVL_4              0
+#elif NR_CPUS <= RCU_FANOUT_4
+#  define NUM_RCU_LVLS       4
+#  define NUM_RCU_LVL_0              1
+#  define NUM_RCU_LVL_1              DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
+#  define NUM_RCU_LVL_2              DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
+#  define NUM_RCU_LVL_3              DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
+#  define NUM_RCU_LVL_4              (NR_CPUS)
+#else
+# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
+#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
+
+#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
+#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
+
+/*
+ * Dynticks per-CPU state.
+ */
+struct rcu_dynticks {
+       int dynticks_nesting;   /* Track irq/process nesting level. */
+       int dynticks_nmi_nesting; /* Track NMI nesting level. */
+       atomic_t dynticks;      /* Even value for dynticks-idle, else odd. */
+};
+
+/* RCU's kthread states for tracing. */
+#define RCU_KTHREAD_STOPPED  0
+#define RCU_KTHREAD_RUNNING  1
+#define RCU_KTHREAD_WAITING  2
+#define RCU_KTHREAD_OFFCPU   3
+#define RCU_KTHREAD_YIELDING 4
+#define RCU_KTHREAD_MAX      4
+
+/*
+ * Definition for node within the RCU grace-period-detection hierarchy.
+ */
+struct rcu_node {
+       raw_spinlock_t lock;    /* Root rcu_node's lock protects some */
+                               /*  rcu_state fields as well as following. */
+       unsigned long gpnum;    /* Current grace period for this node. */
+                               /*  This will either be equal to or one */
+                               /*  behind the root rcu_node's gpnum. */
+       unsigned long completed; /* Last GP completed for this node. */
+                               /*  This will either be equal to or one */
+                               /*  behind the root rcu_node's gpnum. */
+       unsigned long qsmask;   /* CPUs or groups that need to switch in */
+                               /*  order for current grace period to proceed.*/
+                               /*  In leaf rcu_node, each bit corresponds to */
+                               /*  an rcu_data structure, otherwise, each */
+                               /*  bit corresponds to a child rcu_node */
+                               /*  structure. */
+       unsigned long expmask;  /* Groups that have ->blkd_tasks */
+                               /*  elements that need to drain to allow the */
+                               /*  current expedited grace period to */
+                               /*  complete (only for TREE_PREEMPT_RCU). */
+       atomic_t wakemask;      /* CPUs whose kthread needs to be awakened. */
+                               /*  Since this has meaning only for leaf */
+                               /*  rcu_node structures, 32 bits suffices. */
+       unsigned long qsmaskinit;
+                               /* Per-GP initial value for qsmask & expmask. */
+       unsigned long grpmask;  /* Mask to apply to parent qsmask. */
+                               /*  Only one bit will be set in this mask. */
+       int     grplo;          /* lowest-numbered CPU or group here. */
+       int     grphi;          /* highest-numbered CPU or group here. */
+       u8      grpnum;         /* CPU/group number for next level up. */
+       u8      level;          /* root is at level 0. */
+       struct rcu_node *parent;
+       struct list_head blkd_tasks;
+                               /* Tasks blocked in RCU read-side critical */
+                               /*  section.  Tasks are placed at the head */
+                               /*  of this list and age towards the tail. */
+       struct list_head *gp_tasks;
+                               /* Pointer to the first task blocking the */
+                               /*  current grace period, or NULL if there */
+                               /*  is no such task. */
+       struct list_head *exp_tasks;
+                               /* Pointer to the first task blocking the */
+                               /*  current expedited grace period, or NULL */
+                               /*  if there is no such task.  If there */
+                               /*  is no current expedited grace period, */
+                               /*  then there can cannot be any such task. */
+#ifdef CONFIG_RCU_BOOST
+       struct list_head *boost_tasks;
+                               /* Pointer to first task that needs to be */
+                               /*  priority boosted, or NULL if no priority */
+                               /*  boosting is needed for this rcu_node */
+                               /*  structure.  If there are no tasks */
+                               /*  queued on this rcu_node structure that */
+                               /*  are blocking the current grace period, */
+                               /*  there can be no such task. */
+       unsigned long boost_time;
+                               /* When to start boosting (jiffies). */
+       struct task_struct *boost_kthread_task;
+                               /* kthread that takes care of priority */
+                               /*  boosting for this rcu_node structure. */
+       unsigned int boost_kthread_status;
+                               /* State of boost_kthread_task for tracing. */
+       unsigned long n_tasks_boosted;
+                               /* Total number of tasks boosted. */
+       unsigned long n_exp_boosts;
+                               /* Number of tasks boosted for expedited GP. */
+       unsigned long n_normal_boosts;
+                               /* Number of tasks boosted for normal GP. */
+       unsigned long n_balk_blkd_tasks;
+                               /* Refused to boost: no blocked tasks. */
+       unsigned long n_balk_exp_gp_tasks;
+                               /* Refused to boost: nothing blocking GP. */
+       unsigned long n_balk_boost_tasks;
+                               /* Refused to boost: already boosting. */
+       unsigned long n_balk_notblocked;
+                               /* Refused to boost: RCU RS CS still running. */
+       unsigned long n_balk_notyet;
+                               /* Refused to boost: not yet time. */
+       unsigned long n_balk_nos;
+                               /* Refused to boost: not sure why, though. */
+                               /*  This can happen due to race conditions. */
+#endif /* #ifdef CONFIG_RCU_BOOST */
+       struct task_struct *node_kthread_task;
+                               /* kthread that takes care of this rcu_node */
+                               /*  structure, for example, awakening the */
+                               /*  per-CPU kthreads as needed. */
+       unsigned int node_kthread_status;
+                               /* State of node_kthread_task for tracing. */
+} ____cacheline_internodealigned_in_smp;
+
+/*
+ * Do a full breadth-first scan of the rcu_node structures for the
+ * specified rcu_state structure.
+ */
+#define rcu_for_each_node_breadth_first(rsp, rnp) \
+       for ((rnp) = &(rsp)->node[0]; \
+            (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
+
+/*
+ * Do a breadth-first scan of the non-leaf rcu_node structures for the
+ * specified rcu_state structure.  Note that if there is a singleton
+ * rcu_node tree with but one rcu_node structure, this loop is a no-op.
+ */
+#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
+       for ((rnp) = &(rsp)->node[0]; \
+            (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
+
+/*
+ * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
+ * structure.  Note that if there is a singleton rcu_node tree with but
+ * one rcu_node structure, this loop -will- visit the rcu_node structure.
+ * It is still a leaf node, even if it is also the root node.
+ */
+#define rcu_for_each_leaf_node(rsp, rnp) \
+       for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
+            (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
+
+/* Index values for nxttail array in struct rcu_data. */
+#define RCU_DONE_TAIL          0       /* Also RCU_WAIT head. */
+#define RCU_WAIT_TAIL          1       /* Also RCU_NEXT_READY head. */
+#define RCU_NEXT_READY_TAIL    2       /* Also RCU_NEXT head. */
+#define RCU_NEXT_TAIL          3
+#define RCU_NEXT_SIZE          4
+
+/* Per-CPU data for read-copy update. */
+struct rcu_data {
+       /* 1) quiescent-state and grace-period handling : */
+       unsigned long   completed;      /* Track rsp->completed gp number */
+                                       /*  in order to detect GP end. */
+       unsigned long   gpnum;          /* Highest gp number that this CPU */
+                                       /*  is aware of having started. */
+       unsigned long   passed_quiesc_completed;
+                                       /* Value of completed at time of qs. */
+       bool            passed_quiesc;  /* User-mode/idle loop etc. */
+       bool            qs_pending;     /* Core waits for quiesc state. */
+       bool            beenonline;     /* CPU online at least once. */
+       bool            preemptible;    /* Preemptible RCU? */
+       struct rcu_node *mynode;        /* This CPU's leaf of hierarchy */
+       unsigned long grpmask;          /* Mask to apply to leaf qsmask. */
+
+       /* 2) batch handling */
+       /*
+        * If nxtlist is not NULL, it is partitioned as follows.
+        * Any of the partitions might be empty, in which case the
+        * pointer to that partition will be equal to the pointer for
+        * the following partition.  When the list is empty, all of
+        * the nxttail elements point to the ->nxtlist pointer itself,
+        * which in that case is NULL.
+        *
+        * [nxtlist, *nxttail[RCU_DONE_TAIL]):
+        *      Entries that batch # <= ->completed
+        *      The grace period for these entries has completed, and
+        *      the other grace-period-completed entries may be moved
+        *      here temporarily in rcu_process_callbacks().
+        * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
+        *      Entries that batch # <= ->completed - 1: waiting for current GP
+        * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
+        *      Entries known to have arrived before current GP ended
+        * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
+        *      Entries that might have arrived after current GP ended
+        *      Note that the value of *nxttail[RCU_NEXT_TAIL] will
+        *      always be NULL, as this is the end of the list.
+        */
+       struct rcu_head *nxtlist;
+       struct rcu_head **nxttail[RCU_NEXT_SIZE];
+       long            qlen;           /* # of queued callbacks */
+       long            qlen_last_fqs_check;
+                                       /* qlen at last check for QS forcing */
+       unsigned long   n_cbs_invoked;  /* count of RCU cbs invoked. */
+       unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
+       unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
+       unsigned long   n_force_qs_snap;
+                                       /* did other CPU force QS recently? */
+       long            blimit;         /* Upper limit on a processed batch */
+
+#ifdef CONFIG_NO_HZ
+       /* 3) dynticks interface. */
+       struct rcu_dynticks *dynticks;  /* Shared per-CPU dynticks state. */
+       int dynticks_snap;              /* Per-GP tracking for dynticks. */
+#endif /* #ifdef CONFIG_NO_HZ */
+
+       /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
+#ifdef CONFIG_NO_HZ
+       unsigned long dynticks_fqs;     /* Kicked due to dynticks idle. */
+#endif /* #ifdef CONFIG_NO_HZ */
+       unsigned long offline_fqs;      /* Kicked due to being offline. */
+       unsigned long resched_ipi;      /* Sent a resched IPI. */
+
+       /* 5) __rcu_pending() statistics. */
+       unsigned long n_rcu_pending;    /* rcu_pending() calls since boot. */
+       unsigned long n_rp_qs_pending;
+       unsigned long n_rp_report_qs;
+       unsigned long n_rp_cb_ready;
+       unsigned long n_rp_cpu_needs_gp;
+       unsigned long n_rp_gp_completed;
+       unsigned long n_rp_gp_started;
+       unsigned long n_rp_need_fqs;
+       unsigned long n_rp_need_nothing;
+
+       int cpu;
+};
+
+/* Values for signaled field in struct rcu_state. */
+#define RCU_GP_IDLE            0       /* No grace period in progress. */
+#define RCU_GP_INIT            1       /* Grace period being initialized. */
+#define RCU_SAVE_DYNTICK       2       /* Need to scan dyntick state. */
+#define RCU_FORCE_QS           3       /* Need to force quiescent state. */
+#ifdef CONFIG_NO_HZ
+#define RCU_SIGNAL_INIT                RCU_SAVE_DYNTICK
+#else /* #ifdef CONFIG_NO_HZ */
+#define RCU_SIGNAL_INIT                RCU_FORCE_QS
+#endif /* #else #ifdef CONFIG_NO_HZ */
+
+#define RCU_JIFFIES_TILL_FORCE_QS       3      /* for rsp->jiffies_force_qs */
+
+#ifdef CONFIG_PROVE_RCU
+#define RCU_STALL_DELAY_DELTA         (5 * HZ)
+#else
+#define RCU_STALL_DELAY_DELTA         0
+#endif
+
+#define RCU_SECONDS_TILL_STALL_CHECK   (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
+                                       RCU_STALL_DELAY_DELTA)
+                                               /* for rsp->jiffies_stall */
+#define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
+                                               /* for rsp->jiffies_stall */
+#define RCU_STALL_RAT_DELAY            2       /* Allow other CPUs time */
+                                               /*  to take at least one */
+                                               /*  scheduling clock irq */
+                                               /*  before ratting on them. */
+
+#define rcu_wait(cond)                                                 \
+do {                                                                   \
+       for (;;) {                                                      \
+               set_current_state(TASK_INTERRUPTIBLE);                  \
+               if (cond)                                               \
+                       break;                                          \
+               schedule();                                             \
+       }                                                               \
+       __set_current_state(TASK_RUNNING);                              \
+} while (0)
+
+/*
+ * RCU global state, including node hierarchy.  This hierarchy is
+ * represented in "heap" form in a dense array.  The root (first level)
+ * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
+ * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
+ * and the third level in ->node[m+1] and following (->node[m+1] referenced
+ * by ->level[2]).  The number of levels is determined by the number of
+ * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
+ * consisting of a single rcu_node.
+ */
+struct rcu_state {
+       struct rcu_node node[NUM_RCU_NODES];    /* Hierarchy. */
+       struct rcu_node *level[NUM_RCU_LVLS];   /* Hierarchy levels. */
+       u32 levelcnt[MAX_RCU_LVLS + 1];         /* # nodes in each level. */
+       u8 levelspread[NUM_RCU_LVLS];           /* kids/node in each level. */
+       struct rcu_data __percpu *rda;          /* pointer of percu rcu_data. */
+
+       /* The following fields are guarded by the root rcu_node's lock. */
+
+       u8      signaled ____cacheline_internodealigned_in_smp;
+                                               /* Force QS state. */
+       u8      fqs_active;                     /* force_quiescent_state() */
+                                               /*  is running. */
+       u8      fqs_need_gp;                    /* A CPU was prevented from */
+                                               /*  starting a new grace */
+                                               /*  period because */
+                                               /*  force_quiescent_state() */
+                                               /*  was running. */
+       u8      boost;                          /* Subject to priority boost. */
+       unsigned long gpnum;                    /* Current gp number. */
+       unsigned long completed;                /* # of last completed gp. */
+
+       /* End of fields guarded by root rcu_node's lock. */
+
+       raw_spinlock_t onofflock;               /* exclude on/offline and */
+                                               /*  starting new GP. */
+       raw_spinlock_t fqslock;                 /* Only one task forcing */
+                                               /*  quiescent states. */
+       unsigned long jiffies_force_qs;         /* Time at which to invoke */
+                                               /*  force_quiescent_state(). */
+       unsigned long n_force_qs;               /* Number of calls to */
+                                               /*  force_quiescent_state(). */
+       unsigned long n_force_qs_lh;            /* ~Number of calls leaving */
+                                               /*  due to lock unavailable. */
+       unsigned long n_force_qs_ngp;           /* Number of calls leaving */
+                                               /*  due to no GP active. */
+       unsigned long gp_start;                 /* Time at which GP started, */
+                                               /*  but in jiffies. */
+       unsigned long jiffies_stall;            /* Time at which to check */
+                                               /*  for CPU stalls. */
+       unsigned long gp_max;                   /* Maximum GP duration in */
+                                               /*  jiffies. */
+       char *name;                             /* Name of structure. */
+};
+
+/* Return values for rcu_preempt_offline_tasks(). */
+
+#define RCU_OFL_TASKS_NORM_GP  0x1             /* Tasks blocking normal */
+                                               /*  GP were moved to root. */
+#define RCU_OFL_TASKS_EXP_GP   0x2             /* Tasks blocking expedited */
+                                               /*  GP were moved to root. */
 
 /*
  * RCU implementation internal declarations:
  */
-extern struct rcu_state rcu_state;
-DECLARE_PER_CPU(struct rcu_data, rcu_data);
+extern struct rcu_state rcu_sched_state;
+DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
 
 extern struct rcu_state rcu_bh_state;
 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
 
+#ifdef CONFIG_TREE_PREEMPT_RCU
+extern struct rcu_state rcu_preempt_state;
+DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
+#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
+
+#ifndef RCU_TREE_NONCORE
+
+/* Forward declarations for rcutree_plugin.h */
+static void rcu_bootup_announce(void);
+long rcu_batches_completed(void);
+static void rcu_preempt_note_context_switch(int cpu);
+static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
+#ifdef CONFIG_HOTPLUG_CPU
+static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
+                                     unsigned long flags);
+static void rcu_stop_cpu_kthread(int cpu);
+#endif /* #ifdef CONFIG_HOTPLUG_CPU */
+static void rcu_print_detail_task_stall(struct rcu_state *rsp);
+static void rcu_print_task_stall(struct rcu_node *rnp);
+static void rcu_preempt_stall_reset(void);
+static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
+#ifdef CONFIG_HOTPLUG_CPU
+static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
+                                    struct rcu_node *rnp,
+                                    struct rcu_data *rdp);
+static void rcu_preempt_offline_cpu(int cpu);
+#endif /* #ifdef CONFIG_HOTPLUG_CPU */
+static void rcu_preempt_check_callbacks(int cpu);
+static void rcu_preempt_process_callbacks(void);
+void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
+#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
+static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp);
+#endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
+static int rcu_preempt_pending(int cpu);
+static int rcu_preempt_needs_cpu(int cpu);
+static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
+static void rcu_preempt_send_cbs_to_online(void);
+static void __init __rcu_init_preempt(void);
+static void rcu_needs_cpu_flush(void);
+static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
+static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
+static void invoke_rcu_callbacks_kthread(void);
+#ifdef CONFIG_RCU_BOOST
+static void rcu_preempt_do_callbacks(void);
+static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
+                                         cpumask_var_t cm);
+static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
+                                                struct rcu_node *rnp,
+                                                int rnp_index);
+static void invoke_rcu_node_kthread(struct rcu_node *rnp);
+static void rcu_yield(void (*f)(unsigned long), unsigned long arg);
+#endif /* #ifdef CONFIG_RCU_BOOST */
+static void rcu_cpu_kthread_setrt(int cpu, int to_rt);
+static void __cpuinit rcu_prepare_kthreads(int cpu);
+
+#endif /* #ifndef RCU_TREE_NONCORE */