From fc2219d49ef1606e7fd2c88af2b423b01ff3d319 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Wed, 23 Sep 2009 09:50:41 -0700 Subject: rcu: Clean up code based on review feedback from Josh Triplett These issues identified during an old-fashioned face-to-face code review extended over many hours. o Bury various forms of the "rsp->completed == rsp->gpnum" comparison into an rcu_gp_in_progress() function, which has the beneficial side-effect of forcing consistent use of ACCESS_ONCE(). o Replace hand-coded arithmetic with DIV_ROUND_UP(). o Bury several "!list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x01])" instances into an rcu_preempted_readers() function, as this expression indicates that there are no readers blocked within RCU read-side critical sections blocking the current grace period. (Though there might well be similar readers blocking the next grace period.) o Remove a dangling rcu_restart_cpu() declaration that has been dangling for almost 20 minor releases of the kernel. Signed-off-by: Paul E. McKenney Acked-by: Peter Zijlstra Cc: laijs@cn.fujitsu.com Cc: dipankar@in.ibm.com Cc: akpm@linux-foundation.org Cc: mathieu.desnoyers@polymtl.ca Cc: josh@joshtriplett.org Cc: dvhltc@us.ibm.com Cc: niv@us.ibm.com Cc: rostedt@goodmis.org Cc: Valdis.Kletnieks@vt.edu Cc: dhowells@redhat.com LKML-Reference: <12537246442687-git-send-email-> Signed-off-by: Ingo Molnar --- kernel/rcutree.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel/rcutree.h') diff --git a/kernel/rcutree.h b/kernel/rcutree.h index 8e8287a..9aa8c8a 100644 --- a/kernel/rcutree.h +++ b/kernel/rcutree.h @@ -48,14 +48,14 @@ #elif NR_CPUS <= RCU_FANOUT_SQ # define NUM_RCU_LVLS 2 # define NUM_RCU_LVL_0 1 -# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT - 1) / RCU_FANOUT) +# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT) # define NUM_RCU_LVL_2 (NR_CPUS) # define NUM_RCU_LVL_3 0 #elif NR_CPUS <= RCU_FANOUT_CUBE # define NUM_RCU_LVLS 3 # define NUM_RCU_LVL_0 1 -# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT_SQ - 1) / RCU_FANOUT_SQ) -# define NUM_RCU_LVL_2 (((NR_CPUS) + (RCU_FANOUT) - 1) / (RCU_FANOUT)) +# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_SQ) +# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT) # define NUM_RCU_LVL_3 NR_CPUS #else # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS" -- cgit v1.1 From 1eba8f84380bede3c602bd7758dea96925cead01 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Wed, 23 Sep 2009 09:50:42 -0700 Subject: rcu: Clean up code based on review feedback from Josh Triplett, part 2 These issues identified during an old-fashioned face-to-face code review extending over many hours. o Add comments for tricky parts of code, and correct comments that have passed their sell-by date. o Get rid of the vestiges of rcu_init_sched(), which is no longer needed now that PREEMPT_RCU is gone. o Move the #include of rcutree_plugin.h to the end of rcutree.c, which means that, rather than having a random collection of forward declarations, the new set of forward declarations document the set of plugins. The new home for this #include also allows __rcu_init_preempt() to move into rcutree_plugin.h. o Fix rcu_preempt_check_callbacks() to be static. Suggested-by: Josh Triplett Signed-off-by: Paul E. McKenney Cc: laijs@cn.fujitsu.com Cc: dipankar@in.ibm.com Cc: akpm@linux-foundation.org Cc: mathieu.desnoyers@polymtl.ca Cc: josh@joshtriplett.org Cc: dvhltc@us.ibm.com Cc: niv@us.ibm.com Cc: peterz@infradead.org Cc: rostedt@goodmis.org Cc: Valdis.Kletnieks@vt.edu Cc: dhowells@redhat.com LKML-Reference: <12537246443924-git-send-email-> Signed-off-by: Ingo Molnar Peter Zijlstra --- kernel/rcutree.h | 31 +++++++++++++++++++++++-------- 1 file changed, 23 insertions(+), 8 deletions(-) (limited to 'kernel/rcutree.h') diff --git a/kernel/rcutree.h b/kernel/rcutree.h index 9aa8c8a..a48d11f 100644 --- a/kernel/rcutree.h +++ b/kernel/rcutree.h @@ -79,15 +79,21 @@ struct rcu_dynticks { * Definition for node within the RCU grace-period-detection hierarchy. */ struct rcu_node { - spinlock_t lock; + spinlock_t lock; /* Root rcu_node's lock protects some */ + /* rcu_state fields as well as following. */ 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 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 qsmaskinit; /* Per-GP initialization for qsmask. */ 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. */ @@ -95,6 +101,9 @@ struct rcu_node { struct rcu_node *parent; struct list_head blocked_tasks[2]; /* Tasks blocked in RCU read-side critsect. */ + /* Grace period number (->gpnum) x blocked */ + /* by tasks on the (x & 0x1) element of the */ + /* blocked_tasks[] array. */ } ____cacheline_internodealigned_in_smp; /* Index values for nxttail array in struct rcu_data. */ @@ -126,19 +135,22 @@ struct rcu_data { * 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 nxtlist, which is NULL. + * the nxttail elements point to the ->nxtlist pointer itself, + * which in that case is NULL. * - * [*nxttail[RCU_NEXT_READY_TAIL], NULL = *nxttail[RCU_NEXT_TAIL]): - * Entries that might have arrived after current GP ended - * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]): - * Entries known to have arrived before current GP ended - * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]): - * Entries that batch # <= ->completed - 1: waiting for current GP * [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]; @@ -216,6 +228,9 @@ struct rcu_state { /* Force QS state. */ long gpnum; /* Current gp number. */ long completed; /* # of last completed gp. */ + + /* End of fields guarded by root rcu_node's lock. */ + spinlock_t onofflock; /* exclude on/offline and */ /* starting new GP. */ spinlock_t fqslock; /* Only one task forcing */ -- cgit v1.1 From 9b2619aff0332e95ea5eb7a0d75b0208818d871c Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Wed, 23 Sep 2009 09:50:43 -0700 Subject: rcu: Clean up code to address Ingo's checkpatch feedback Move declarations and update storage classes to make checkpatch happy. Signed-off-by: Paul E. McKenney Cc: laijs@cn.fujitsu.com Cc: dipankar@in.ibm.com Cc: akpm@linux-foundation.org Cc: mathieu.desnoyers@polymtl.ca Cc: josh@joshtriplett.org Cc: dvhltc@us.ibm.com Cc: niv@us.ibm.com Cc: peterz@infradead.org Cc: rostedt@goodmis.org Cc: Valdis.Kletnieks@vt.edu Cc: dhowells@redhat.com LKML-Reference: <12537246441701-git-send-email-> Signed-off-by: Ingo Molnar --- kernel/rcutree.h | 26 +++++++++++++++++++++++++- 1 file changed, 25 insertions(+), 1 deletion(-) (limited to 'kernel/rcutree.h') diff --git a/kernel/rcutree.h b/kernel/rcutree.h index a48d11f..e6ab31c 100644 --- a/kernel/rcutree.h +++ b/kernel/rcutree.h @@ -270,5 +270,29 @@ extern struct rcu_state rcu_preempt_state; DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data); #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ -#endif /* #ifdef RCU_TREE_NONCORE */ +#else /* #ifdef RCU_TREE_NONCORE */ +/* Forward declarations for rcutree_plugin.h */ +static inline void rcu_bootup_announce(void); +long rcu_batches_completed(void); +static void rcu_preempt_note_context_switch(int cpu); +static int rcu_preempted_readers(struct rcu_node *rnp); +#ifdef CONFIG_RCU_CPU_STALL_DETECTOR +static void rcu_print_task_stall(struct rcu_node *rnp); +#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ +static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp); +#ifdef CONFIG_HOTPLUG_CPU +static void 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)); +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 __init __rcu_init_preempt(void); + +#endif /* #else #ifdef RCU_TREE_NONCORE */ -- cgit v1.1 From a0b6c9a78c41dc36732d6e1e90f0f2f57b29816f Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Mon, 28 Sep 2009 07:46:33 -0700 Subject: rcu: Clean up code based on review feedback from Josh Triplett, part 4 These issues identified during an old-fashioned face-to-face code review extending over many hours. This group improves an existing abstraction and introduces two new ones. It also fixes an RCU stall-warning bug found while making the other changes. o Make RCU_INIT_FLAVOR() declare its own variables, removing the need to declare them at each call site. o Create an rcu_for_each_leaf() macro that scans the leaf nodes of the rcu_node tree. o Create an rcu_for_each_node_breadth_first() macro that does a breadth-first traversal of the rcu_node tree, AKA stepping through the array in index-number order. o If all CPUs corresponding to a given leaf rcu_node structure go offline, then any tasks queued on that leaf will be moved to the root rcu_node structure. Therefore, the stall-warning code must dump out tasks queued on the root rcu_node structure as well as those queued on the leaf rcu_node structures. Signed-off-by: Paul E. McKenney Cc: laijs@cn.fujitsu.com Cc: dipankar@in.ibm.com Cc: akpm@linux-foundation.org Cc: mathieu.desnoyers@polymtl.ca Cc: josh@joshtriplett.org Cc: dvhltc@us.ibm.com Cc: niv@us.ibm.com Cc: peterz@infradead.org Cc: rostedt@goodmis.org Cc: Valdis.Kletnieks@vt.edu Cc: dhowells@redhat.com LKML-Reference: <12541491934126-git-send-email-> Signed-off-by: Ingo Molnar --- kernel/rcutree.h | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'kernel/rcutree.h') diff --git a/kernel/rcutree.h b/kernel/rcutree.h index e6ab31c..676eecd 100644 --- a/kernel/rcutree.h +++ b/kernel/rcutree.h @@ -106,6 +106,18 @@ struct rcu_node { /* blocked_tasks[] array. */ } ____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)++) + +#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. */ -- cgit v1.1 From e74f4c4564455c91a3b4075bb1721993c2a95dda Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Tue, 6 Oct 2009 21:48:17 -0700 Subject: rcu: Make hot-unplugged CPU relinquish its own RCU callbacks The current interaction between RCU and CPU hotplug requires that RCU block in CPU notifiers waiting for callbacks to drain. This can be greatly simplified by having each CPU relinquish its own callbacks, and for both _rcu_barrier() and CPU_DEAD notifiers to adopt all callbacks that were previously relinquished. This change also eliminates the possibility of certain types of hangs due to the previous practice of waiting for callbacks to be invoked from within CPU notifiers. If you don't every wait, you cannot hang. Signed-off-by: Paul E. McKenney Cc: laijs@cn.fujitsu.com Cc: dipankar@in.ibm.com Cc: akpm@linux-foundation.org Cc: mathieu.desnoyers@polymtl.ca Cc: josh@joshtriplett.org Cc: dvhltc@us.ibm.com Cc: niv@us.ibm.com Cc: peterz@infradead.org Cc: rostedt@goodmis.org Cc: Valdis.Kletnieks@vt.edu Cc: dhowells@redhat.com LKML-Reference: <1254890898456-git-send-email-> Signed-off-by: Ingo Molnar --- kernel/rcutree.h | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) (limited to 'kernel/rcutree.h') diff --git a/kernel/rcutree.h b/kernel/rcutree.h index 676eecd..b40ac57 100644 --- a/kernel/rcutree.h +++ b/kernel/rcutree.h @@ -244,7 +244,15 @@ struct rcu_state { /* End of fields guarded by root rcu_node's lock. */ spinlock_t onofflock; /* exclude on/offline and */ - /* starting new GP. */ + /* starting new GP. Also */ + /* protects the following */ + /* orphan_cbs fields. */ + struct rcu_head *orphan_cbs_list; /* list of rcu_head structs */ + /* orphaned by all CPUs in */ + /* a given leaf rcu_node */ + /* going offline. */ + struct rcu_head **orphan_cbs_tail; /* And tail pointer. */ + long orphan_qlen; /* Number of orphaned cbs. */ spinlock_t fqslock; /* Only one task forcing */ /* quiescent states. */ unsigned long jiffies_force_qs; /* Time at which to invoke */ @@ -305,6 +313,7 @@ void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *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_orphanage(void); static void __init __rcu_init_preempt(void); #endif /* #else #ifdef RCU_TREE_NONCORE */ -- cgit v1.1