X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=net%2Fcore%2Fdst.c;h=57bc4d5b8d084c053cded6ebddfe66bf6255d9f5;hb=56baa667973e53d6d38af2ad3731d558566d818b;hp=836ec66069254752c7ca2b9db0a597419dd758a2;hpb=54e6ecb23951b195d02433a741c7f7cb0b796c78;p=safe%2Fjmp%2Flinux-2.6 diff --git a/net/core/dst.c b/net/core/dst.c index 836ec66..57bc4d5 100644 --- a/net/core/dst.c +++ b/net/core/dst.c @@ -9,60 +9,83 @@ #include #include #include +#include #include #include #include -#include #include #include #include +#include #include -/* Locking strategy: - * 1) Garbage collection state of dead destination cache - * entries is protected by dst_lock. - * 2) GC is run only from BH context, and is the only remover - * of entries. - * 3) Entries are added to the garbage list from both BH - * and non-BH context, so local BH disabling is needed. - * 4) All operations modify state, so a spinlock is used. +/* + * Theory of operations: + * 1) We use a list, protected by a spinlock, to add + * new entries from both BH and non-BH context. + * 2) In order to keep spinlock held for a small delay, + * we use a second list where are stored long lived + * entries, that are handled by the garbage collect thread + * fired by a workqueue. + * 3) This list is guarded by a mutex, + * so that the gc_task and dst_dev_event() can be synchronized. */ -static struct dst_entry *dst_garbage_list; -#if RT_CACHE_DEBUG >= 2 +#if RT_CACHE_DEBUG >= 2 static atomic_t dst_total = ATOMIC_INIT(0); #endif -static DEFINE_SPINLOCK(dst_lock); -static unsigned long dst_gc_timer_expires; -static unsigned long dst_gc_timer_inc = DST_GC_MAX; -static void dst_run_gc(unsigned long); +/* + * We want to keep lock & list close together + * to dirty as few cache lines as possible in __dst_free(). + * As this is not a very strong hint, we dont force an alignment on SMP. + */ +static struct { + spinlock_t lock; + struct dst_entry *list; + unsigned long timer_inc; + unsigned long timer_expires; +} dst_garbage = { + .lock = __SPIN_LOCK_UNLOCKED(dst_garbage.lock), + .timer_inc = DST_GC_MAX, +}; +static void dst_gc_task(struct work_struct *work); static void ___dst_free(struct dst_entry * dst); -static DEFINE_TIMER(dst_gc_timer, dst_run_gc, DST_GC_MIN, 0); +static DECLARE_DELAYED_WORK(dst_gc_work, dst_gc_task); -static void dst_run_gc(unsigned long dummy) +static DEFINE_MUTEX(dst_gc_mutex); +/* + * long lived entries are maintained in this list, guarded by dst_gc_mutex + */ +static struct dst_entry *dst_busy_list; + +static void dst_gc_task(struct work_struct *work) { int delayed = 0; - int work_performed; - struct dst_entry * dst, **dstp; + int work_performed = 0; + unsigned long expires = ~0L; + struct dst_entry *dst, *next, head; + struct dst_entry *last = &head; +#if RT_CACHE_DEBUG >= 2 + ktime_t time_start = ktime_get(); + struct timespec elapsed; +#endif - if (!spin_trylock(&dst_lock)) { - mod_timer(&dst_gc_timer, jiffies + HZ/10); - return; - } + mutex_lock(&dst_gc_mutex); + next = dst_busy_list; - del_timer(&dst_gc_timer); - dstp = &dst_garbage_list; - work_performed = 0; - while ((dst = *dstp) != NULL) { - if (atomic_read(&dst->__refcnt)) { - dstp = &dst->next; +loop: + while ((dst = next) != NULL) { + next = dst->next; + prefetch(&next->next); + if (likely(atomic_read(&dst->__refcnt))) { + last->next = dst; + last = dst; delayed++; continue; } - *dstp = dst->next; - work_performed = 1; + work_performed++; dst = dst_destroy(dst); if (dst) { @@ -78,64 +101,82 @@ static void dst_run_gc(unsigned long dummy) continue; ___dst_free(dst); - dst->next = *dstp; - *dstp = dst; - dstp = &dst->next; + dst->next = next; + next = dst; } } - if (!dst_garbage_list) { - dst_gc_timer_inc = DST_GC_MAX; - goto out; + + spin_lock_bh(&dst_garbage.lock); + next = dst_garbage.list; + if (next) { + dst_garbage.list = NULL; + spin_unlock_bh(&dst_garbage.lock); + goto loop; } - if (!work_performed) { - if ((dst_gc_timer_expires += dst_gc_timer_inc) > DST_GC_MAX) - dst_gc_timer_expires = DST_GC_MAX; - dst_gc_timer_inc += DST_GC_INC; - } else { - dst_gc_timer_inc = DST_GC_INC; - dst_gc_timer_expires = DST_GC_MIN; + last->next = NULL; + dst_busy_list = head.next; + if (!dst_busy_list) + dst_garbage.timer_inc = DST_GC_MAX; + else { + /* + * if we freed less than 1/10 of delayed entries, + * we can sleep longer. + */ + if (work_performed <= delayed/10) { + dst_garbage.timer_expires += dst_garbage.timer_inc; + if (dst_garbage.timer_expires > DST_GC_MAX) + dst_garbage.timer_expires = DST_GC_MAX; + dst_garbage.timer_inc += DST_GC_INC; + } else { + dst_garbage.timer_inc = DST_GC_INC; + dst_garbage.timer_expires = DST_GC_MIN; + } + expires = dst_garbage.timer_expires; + /* + * if the next desired timer is more than 4 seconds in the future + * then round the timer to whole seconds + */ + if (expires > 4*HZ) + expires = round_jiffies_relative(expires); + schedule_delayed_work(&dst_gc_work, expires); } + + spin_unlock_bh(&dst_garbage.lock); + mutex_unlock(&dst_gc_mutex); #if RT_CACHE_DEBUG >= 2 - printk("dst_total: %d/%d %ld\n", - atomic_read(&dst_total), delayed, dst_gc_timer_expires); + elapsed = ktime_to_timespec(ktime_sub(ktime_get(), time_start)); + printk(KERN_DEBUG "dst_total: %d delayed: %d work_perf: %d" + " expires: %lu elapsed: %lu us\n", + atomic_read(&dst_total), delayed, work_performed, + expires, + elapsed.tv_sec * USEC_PER_SEC + elapsed.tv_nsec / NSEC_PER_USEC); #endif - mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires); - -out: - spin_unlock(&dst_lock); } -static int dst_discard_in(struct sk_buff *skb) -{ - kfree_skb(skb); - return 0; -} - -static int dst_discard_out(struct sk_buff *skb) +int dst_discard(struct sk_buff *skb) { kfree_skb(skb); return 0; } +EXPORT_SYMBOL(dst_discard); void * dst_alloc(struct dst_ops * ops) { struct dst_entry * dst; if (ops->gc && atomic_read(&ops->entries) > ops->gc_thresh) { - if (ops->gc()) + if (ops->gc(ops)) return NULL; } - dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC); + dst = kmem_cache_zalloc(ops->kmem_cachep, GFP_ATOMIC); if (!dst) return NULL; - memset(dst, 0, ops->entry_size); atomic_set(&dst->__refcnt, 0); dst->ops = ops; dst->lastuse = jiffies; dst->path = dst; - dst->input = dst_discard_in; - dst->output = dst_discard_out; -#if RT_CACHE_DEBUG >= 2 + dst->input = dst->output = dst_discard; +#if RT_CACHE_DEBUG >= 2 atomic_inc(&dst_total); #endif atomic_inc(&ops->entries); @@ -148,24 +189,24 @@ static void ___dst_free(struct dst_entry * dst) protocol module is unloaded. */ if (dst->dev == NULL || !(dst->dev->flags&IFF_UP)) { - dst->input = dst_discard_in; - dst->output = dst_discard_out; + dst->input = dst->output = dst_discard; } dst->obsolete = 2; } void __dst_free(struct dst_entry * dst) { - spin_lock_bh(&dst_lock); + spin_lock_bh(&dst_garbage.lock); ___dst_free(dst); - dst->next = dst_garbage_list; - dst_garbage_list = dst; - if (dst_gc_timer_inc > DST_GC_INC) { - dst_gc_timer_inc = DST_GC_INC; - dst_gc_timer_expires = DST_GC_MIN; - mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires); + dst->next = dst_garbage.list; + dst_garbage.list = dst; + if (dst_garbage.timer_inc > DST_GC_INC) { + dst_garbage.timer_inc = DST_GC_INC; + dst_garbage.timer_expires = DST_GC_MIN; + cancel_delayed_work(&dst_gc_work); + schedule_delayed_work(&dst_gc_work, dst_garbage.timer_expires); } - spin_unlock_bh(&dst_lock); + spin_unlock_bh(&dst_garbage.lock); } struct dst_entry *dst_destroy(struct dst_entry * dst) @@ -196,7 +237,7 @@ again: dst->ops->destroy(dst); if (dst->dev) dev_put(dst->dev); -#if RT_CACHE_DEBUG >= 2 +#if RT_CACHE_DEBUG >= 2 atomic_dec(&dst_total); #endif kmem_cache_free(dst->ops->kmem_cachep, dst); @@ -219,6 +260,18 @@ again: return NULL; } +void dst_release(struct dst_entry *dst) +{ + if (dst) { + int newrefcnt; + + smp_mb__before_atomic_dec(); + newrefcnt = atomic_dec_return(&dst->__refcnt); + WARN_ON(newrefcnt < 0); + } +} +EXPORT_SYMBOL(dst_release); + /* Dirty hack. We did it in 2.2 (in __dst_free), * we have _very_ good reasons not to repeat * this mistake in 2.3, but we have no choice @@ -237,16 +290,15 @@ static inline void dst_ifdown(struct dst_entry *dst, struct net_device *dev, return; if (!unregister) { - dst->input = dst_discard_in; - dst->output = dst_discard_out; + dst->input = dst->output = dst_discard; } else { - dst->dev = &loopback_dev; - dev_hold(&loopback_dev); + dst->dev = dev_net(dst->dev)->loopback_dev; + dev_hold(dst->dev); dev_put(dev); if (dst->neighbour && dst->neighbour->dev == dev) { - dst->neighbour->dev = &loopback_dev; + dst->neighbour->dev = dst->dev; + dev_hold(dst->dev); dev_put(dev); - dev_hold(&loopback_dev); } } } @@ -254,16 +306,30 @@ static inline void dst_ifdown(struct dst_entry *dst, struct net_device *dev, static int dst_dev_event(struct notifier_block *this, unsigned long event, void *ptr) { struct net_device *dev = ptr; - struct dst_entry *dst; + struct dst_entry *dst, *last = NULL; switch (event) { case NETDEV_UNREGISTER: case NETDEV_DOWN: - spin_lock_bh(&dst_lock); - for (dst = dst_garbage_list; dst; dst = dst->next) { + mutex_lock(&dst_gc_mutex); + for (dst = dst_busy_list; dst; dst = dst->next) { + last = dst; + dst_ifdown(dst, dev, event != NETDEV_DOWN); + } + + spin_lock_bh(&dst_garbage.lock); + dst = dst_garbage.list; + dst_garbage.list = NULL; + spin_unlock_bh(&dst_garbage.lock); + + if (last) + last->next = dst; + else + dst_busy_list = dst; + for (; dst; dst = dst->next) { dst_ifdown(dst, dev, event != NETDEV_DOWN); } - spin_unlock_bh(&dst_lock); + mutex_unlock(&dst_gc_mutex); break; } return NOTIFY_DONE;