Merge branch 'generic-ipi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
authorLinus Torvalds <torvalds@linux-foundation.org>
Tue, 15 Jul 2008 21:02:33 +0000 (14:02 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 15 Jul 2008 21:12:03 +0000 (14:12 -0700)
* 'generic-ipi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (22 commits)
  generic-ipi: more merge fallout
  generic-ipi: merge fix
  x86, visws: use mach-default/entry_arch.h
  x86, visws: fix generic-ipi build
  generic-ipi: fixlet
  generic-ipi: fix s390 build bug
  generic-ipi: fix linux-next tree build failure
  fix: "smp_call_function: get rid of the unused nonatomic/retry argument"
  fix: "smp_call_function: get rid of the unused nonatomic/retry argument"
  fix "smp_call_function: get rid of the unused nonatomic/retry argument"
  on_each_cpu(): kill unused 'retry' parameter
  smp_call_function: get rid of the unused nonatomic/retry argument
  sh: convert to generic helpers for IPI function calls
  parisc: convert to generic helpers for IPI function calls
  mips: convert to generic helpers for IPI function calls
  m32r: convert to generic helpers for IPI function calls
  arm: convert to generic helpers for IPI function calls
  alpha: convert to generic helpers for IPI function calls
  ia64: convert to generic helpers for IPI function calls
  powerpc: convert to generic helpers for IPI function calls
  ...

Fix trivial conflicts due to rcu updates in kernel/rcupdate.c manually

1  2 
init/main.c
kernel/rcupdate.c
kernel/smp.c

diff --cc init/main.c
Simple merge
@@@ -128,7 -127,7 +128,7 @@@ static void _rcu_barrier(enum rcu_barri
         * until all the callbacks are queued.
         */
        rcu_read_lock();
-       on_each_cpu(rcu_barrier_func, (void *)type, 0, 1);
 -      on_each_cpu(rcu_barrier_func, NULL, 1);
++      on_each_cpu(rcu_barrier_func, (void *)type, 1);
        rcu_read_unlock();
        wait_for_completion(&rcu_barrier_completion);
        mutex_unlock(&rcu_barrier_mutex);
diff --cc kernel/smp.c
index 0000000,4f582b2..ab10793
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,381 +1,382 @@@
+ /*
+  * Generic helpers for smp ipi calls
+  *
+  * (C) Jens Axboe <jens.axboe@oracle.com> 2008
+  *
+  */
+ #include <linux/init.h>
+ #include <linux/module.h>
+ #include <linux/percpu.h>
+ #include <linux/rcupdate.h>
++#include <linux/rculist.h>
+ #include <linux/smp.h>
+ static DEFINE_PER_CPU(struct call_single_queue, call_single_queue);
+ static LIST_HEAD(call_function_queue);
+ __cacheline_aligned_in_smp DEFINE_SPINLOCK(call_function_lock);
+ enum {
+       CSD_FLAG_WAIT           = 0x01,
+       CSD_FLAG_ALLOC          = 0x02,
+ };
+ struct call_function_data {
+       struct call_single_data csd;
+       spinlock_t lock;
+       unsigned int refs;
+       cpumask_t cpumask;
+       struct rcu_head rcu_head;
+ };
+ struct call_single_queue {
+       struct list_head list;
+       spinlock_t lock;
+ };
+ void __cpuinit init_call_single_data(void)
+ {
+       int i;
+       for_each_possible_cpu(i) {
+               struct call_single_queue *q = &per_cpu(call_single_queue, i);
+               spin_lock_init(&q->lock);
+               INIT_LIST_HEAD(&q->list);
+       }
+ }
+ static void csd_flag_wait(struct call_single_data *data)
+ {
+       /* Wait for response */
+       do {
+               /*
+                * We need to see the flags store in the IPI handler
+                */
+               smp_mb();
+               if (!(data->flags & CSD_FLAG_WAIT))
+                       break;
+               cpu_relax();
+       } while (1);
+ }
+ /*
+  * Insert a previously allocated call_single_data element for execution
+  * on the given CPU. data must already have ->func, ->info, and ->flags set.
+  */
+ static void generic_exec_single(int cpu, struct call_single_data *data)
+ {
+       struct call_single_queue *dst = &per_cpu(call_single_queue, cpu);
+       int wait = data->flags & CSD_FLAG_WAIT, ipi;
+       unsigned long flags;
+       spin_lock_irqsave(&dst->lock, flags);
+       ipi = list_empty(&dst->list);
+       list_add_tail(&data->list, &dst->list);
+       spin_unlock_irqrestore(&dst->lock, flags);
+       if (ipi)
+               arch_send_call_function_single_ipi(cpu);
+       if (wait)
+               csd_flag_wait(data);
+ }
+ static void rcu_free_call_data(struct rcu_head *head)
+ {
+       struct call_function_data *data;
+       data = container_of(head, struct call_function_data, rcu_head);
+       kfree(data);
+ }
+ /*
+  * Invoked by arch to handle an IPI for call function. Must be called with
+  * interrupts disabled.
+  */
+ void generic_smp_call_function_interrupt(void)
+ {
+       struct call_function_data *data;
+       int cpu = get_cpu();
+       /*
+        * It's ok to use list_for_each_rcu() here even though we may delete
+        * 'pos', since list_del_rcu() doesn't clear ->next
+        */
+       rcu_read_lock();
+       list_for_each_entry_rcu(data, &call_function_queue, csd.list) {
+               int refs;
+               if (!cpu_isset(cpu, data->cpumask))
+                       continue;
+               data->csd.func(data->csd.info);
+               spin_lock(&data->lock);
+               cpu_clear(cpu, data->cpumask);
+               WARN_ON(data->refs == 0);
+               data->refs--;
+               refs = data->refs;
+               spin_unlock(&data->lock);
+               if (refs)
+                       continue;
+               spin_lock(&call_function_lock);
+               list_del_rcu(&data->csd.list);
+               spin_unlock(&call_function_lock);
+               if (data->csd.flags & CSD_FLAG_WAIT) {
+                       /*
+                        * serialize stores to data with the flag clear
+                        * and wakeup
+                        */
+                       smp_wmb();
+                       data->csd.flags &= ~CSD_FLAG_WAIT;
+               } else
+                       call_rcu(&data->rcu_head, rcu_free_call_data);
+       }
+       rcu_read_unlock();
+       put_cpu();
+ }
+ /*
+  * Invoked by arch to handle an IPI for call function single. Must be called
+  * from the arch with interrupts disabled.
+  */
+ void generic_smp_call_function_single_interrupt(void)
+ {
+       struct call_single_queue *q = &__get_cpu_var(call_single_queue);
+       LIST_HEAD(list);
+       /*
+        * Need to see other stores to list head for checking whether
+        * list is empty without holding q->lock
+        */
+       smp_mb();
+       while (!list_empty(&q->list)) {
+               unsigned int data_flags;
+               spin_lock(&q->lock);
+               list_replace_init(&q->list, &list);
+               spin_unlock(&q->lock);
+               while (!list_empty(&list)) {
+                       struct call_single_data *data;
+                       data = list_entry(list.next, struct call_single_data,
+                                               list);
+                       list_del(&data->list);
+                       /*
+                        * 'data' can be invalid after this call if
+                        * flags == 0 (when called through
+                        * generic_exec_single(), so save them away before
+                        * making the call.
+                        */
+                       data_flags = data->flags;
+                       data->func(data->info);
+                       if (data_flags & CSD_FLAG_WAIT) {
+                               smp_wmb();
+                               data->flags &= ~CSD_FLAG_WAIT;
+                       } else if (data_flags & CSD_FLAG_ALLOC)
+                               kfree(data);
+               }
+               /*
+                * See comment on outer loop
+                */
+               smp_mb();
+       }
+ }
+ /*
+  * smp_call_function_single - Run a function on a specific CPU
+  * @func: The function to run. This must be fast and non-blocking.
+  * @info: An arbitrary pointer to pass to the function.
+  * @wait: If true, wait until function has completed on other CPUs.
+  *
+  * Returns 0 on success, else a negative status code. Note that @wait
+  * will be implicitly turned on in case of allocation failures, since
+  * we fall back to on-stack allocation.
+  */
+ int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
+                            int wait)
+ {
+       struct call_single_data d;
+       unsigned long flags;
+       /* prevent preemption and reschedule on another processor */
+       int me = get_cpu();
+       /* Can deadlock when called with interrupts disabled */
+       WARN_ON(irqs_disabled());
+       if (cpu == me) {
+               local_irq_save(flags);
+               func(info);
+               local_irq_restore(flags);
+       } else {
+               struct call_single_data *data = NULL;
+               if (!wait) {
+                       data = kmalloc(sizeof(*data), GFP_ATOMIC);
+                       if (data)
+                               data->flags = CSD_FLAG_ALLOC;
+               }
+               if (!data) {
+                       data = &d;
+                       data->flags = CSD_FLAG_WAIT;
+               }
+               data->func = func;
+               data->info = info;
+               generic_exec_single(cpu, data);
+       }
+       put_cpu();
+       return 0;
+ }
+ EXPORT_SYMBOL(smp_call_function_single);
+ /**
+  * __smp_call_function_single(): Run a function on another CPU
+  * @cpu: The CPU to run on.
+  * @data: Pre-allocated and setup data structure
+  *
+  * Like smp_call_function_single(), but allow caller to pass in a pre-allocated
+  * data structure. Useful for embedding @data inside other structures, for
+  * instance.
+  *
+  */
+ void __smp_call_function_single(int cpu, struct call_single_data *data)
+ {
+       /* Can deadlock when called with interrupts disabled */
+       WARN_ON((data->flags & CSD_FLAG_WAIT) && irqs_disabled());
+       generic_exec_single(cpu, data);
+ }
+ /**
+  * smp_call_function_mask(): Run a function on a set of other CPUs.
+  * @mask: The set of cpus to run on.
+  * @func: The function to run. This must be fast and non-blocking.
+  * @info: An arbitrary pointer to pass to the function.
+  * @wait: If true, wait (atomically) until function has completed on other CPUs.
+  *
+  * Returns 0 on success, else a negative status code.
+  *
+  * If @wait is true, then returns once @func has returned. Note that @wait
+  * will be implicitly turned on in case of allocation failures, since
+  * we fall back to on-stack allocation.
+  *
+  * You must not call this function with disabled interrupts or from a
+  * hardware interrupt handler or from a bottom half handler. Preemption
+  * must be disabled when calling this function.
+  */
+ int smp_call_function_mask(cpumask_t mask, void (*func)(void *), void *info,
+                          int wait)
+ {
+       struct call_function_data d;
+       struct call_function_data *data = NULL;
+       cpumask_t allbutself;
+       unsigned long flags;
+       int cpu, num_cpus;
+       /* Can deadlock when called with interrupts disabled */
+       WARN_ON(irqs_disabled());
+       cpu = smp_processor_id();
+       allbutself = cpu_online_map;
+       cpu_clear(cpu, allbutself);
+       cpus_and(mask, mask, allbutself);
+       num_cpus = cpus_weight(mask);
+       /*
+        * If zero CPUs, return. If just a single CPU, turn this request
+        * into a targetted single call instead since it's faster.
+        */
+       if (!num_cpus)
+               return 0;
+       else if (num_cpus == 1) {
+               cpu = first_cpu(mask);
+               return smp_call_function_single(cpu, func, info, wait);
+       }
+       if (!wait) {
+               data = kmalloc(sizeof(*data), GFP_ATOMIC);
+               if (data)
+                       data->csd.flags = CSD_FLAG_ALLOC;
+       }
+       if (!data) {
+               data = &d;
+               data->csd.flags = CSD_FLAG_WAIT;
+       }
+       spin_lock_init(&data->lock);
+       data->csd.func = func;
+       data->csd.info = info;
+       data->refs = num_cpus;
+       data->cpumask = mask;
+       spin_lock_irqsave(&call_function_lock, flags);
+       list_add_tail_rcu(&data->csd.list, &call_function_queue);
+       spin_unlock_irqrestore(&call_function_lock, flags);
+       /* Send a message to all CPUs in the map */
+       arch_send_call_function_ipi(mask);
+       /* optionally wait for the CPUs to complete */
+       if (wait)
+               csd_flag_wait(&data->csd);
+       return 0;
+ }
+ EXPORT_SYMBOL(smp_call_function_mask);
+ /**
+  * smp_call_function(): Run a function on all other CPUs.
+  * @func: The function to run. This must be fast and non-blocking.
+  * @info: An arbitrary pointer to pass to the function.
+  * @wait: If true, wait (atomically) until function has completed on other CPUs.
+  *
+  * Returns 0 on success, else a negative status code.
+  *
+  * If @wait is true, then returns once @func has returned; otherwise
+  * it returns just before the target cpu calls @func. In case of allocation
+  * failure, @wait will be implicitly turned on.
+  *
+  * You must not call this function with disabled interrupts or from a
+  * hardware interrupt handler or from a bottom half handler.
+  */
+ int smp_call_function(void (*func)(void *), void *info, int wait)
+ {
+       int ret;
+       preempt_disable();
+       ret = smp_call_function_mask(cpu_online_map, func, info, wait);
+       preempt_enable();
+       return ret;
+ }
+ EXPORT_SYMBOL(smp_call_function);
+ void ipi_call_lock(void)
+ {
+       spin_lock(&call_function_lock);
+ }
+ void ipi_call_unlock(void)
+ {
+       spin_unlock(&call_function_lock);
+ }
+ void ipi_call_lock_irq(void)
+ {
+       spin_lock_irq(&call_function_lock);
+ }
+ void ipi_call_unlock_irq(void)
+ {
+       spin_unlock_irq(&call_function_lock);
+ }