aboutsummaryrefslogtreecommitdiffstats
path: root/src/vnet/util/refcount.h
blob: 873ab6def6997cb034afcd3f7167ad804f1bdf3a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
/*
 * Copyright (c) 2016 Cisco and/or its affiliates.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at:
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

/*
 * vlib provides lock-free counters but those
 * - Have 16bits per-CPU counter, which may overflow.
 * - Would only increment.
 *
 * This is very similar to vlib counters, but may be used to count reference.
 * Such a counter includes an arbitrary number of counters. Each counter
 * is identified by its index. This is used to aggregate per-cpu memory.
 *
 * Warning:
 *   This reference counter is lock-free but is not race-condition free.
 *   The counting result is approximate and another mechanism needs to be used
 *   in order to ensure that an object may be freed.
 *
 */

#include <vnet/vnet.h>

/*
 * Reference counting
 * A specific reference counter is used. The design is quite
 * similar to vlib counters but:
 *   - It is possible to decrease the value
 *   - Summing will not zero the per-thread counters
 *   - Only the thread can reallocate its own counters vector (to avoid concurrency issues)
*/
typedef struct {
  u32 *counters;
  volatile u32 *counter_lock;
  CLIB_CACHE_LINE_ALIGN_MARK(o);
} vlib_refcount_per_cpu_t;

typedef struct {
  vlib_refcount_per_cpu_t *per_cpu;
} vlib_refcount_t;

static_always_inline
void vlib_refcount_lock (volatile u32 *counter_lock)
{
  while (clib_atomic_test_and_set (counter_lock))
    ;
}

static_always_inline
void vlib_refcount_unlock (volatile u32 *counter_lock)
{
  clib_atomic_release(counter_lock);
}

void __vlib_refcount_resize(vlib_refcount_per_cpu_t *per_cpu, u32 size);

static_always_inline
void vlib_refcount_add(vlib_refcount_t *r, u32 thread_index, u32 counter_index, i32 v)
{
  vlib_refcount_per_cpu_t *per_cpu = &r->per_cpu[thread_index];
  if (PREDICT_FALSE(counter_index >= vec_len(per_cpu->counters)))
    __vlib_refcount_resize(per_cpu, clib_max(counter_index + 16,(vec_len(per_cpu->counters)) * 2));

  per_cpu->counters[counter_index] += v;
}

u64 vlib_refcount_get(vlib_refcount_t *r, u32 index);

static_always_inline
void vlib_refcount_init(vlib_refcount_t *r)
{
  vlib_thread_main_t *tm = vlib_get_thread_main ();
  u32 thread_index;
  r->per_cpu = 0;
  vec_validate (r->per_cpu, tm->n_vlib_mains - 1);

  for (thread_index = 0; thread_index < tm->n_vlib_mains; thread_index++)
    {
      r->per_cpu[thread_index].counter_lock =
	  clib_mem_alloc_aligned(CLIB_CACHE_LINE_BYTES,CLIB_CACHE_LINE_BYTES);
      r->per_cpu[thread_index].counter_lock[0] = 0;
    }
}
t * vnm, u32 hw_if_index, u16 queue_id, vnet_hw_interface_rx_mode mode); int vnet_hw_interface_get_rx_mode (vnet_main_t * vnm, u32 hw_if_index, u16 queue_id, vnet_hw_interface_rx_mode * mode); static inline u64 vnet_get_aggregate_rx_packets (void) { vnet_device_main_t *vdm = &vnet_device_main; u64 sum = 0; vnet_device_per_worker_data_t *pwd; vec_foreach (pwd, vdm->workers) sum += pwd->aggregate_rx_packets; return sum; } static inline void vnet_device_increment_rx_packets (u32 thread_index, u64 count) { vnet_device_main_t *vdm = &vnet_device_main; vnet_device_per_worker_data_t *pwd; pwd = vec_elt_at_index (vdm->workers, thread_index); pwd->aggregate_rx_packets += count; } static_always_inline vnet_device_and_queue_t * vnet_get_device_and_queue (vlib_main_t * vm, vlib_node_runtime_t * node) { vnet_device_input_runtime_t *rt = (void *) node->runtime_data; return rt->devices_and_queues; } static_always_inline uword vnet_get_device_input_thread_index (vnet_main_t * vnm, u32 hw_if_index, u16 queue_id) { vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index); ASSERT (queue_id < vec_len (hw->input_node_thread_index_by_queue)); return hw->input_node_thread_index_by_queue[queue_id]; } static_always_inline void vnet_device_input_set_interrupt_pending (vnet_main_t * vnm, u32 hw_if_index, u16 queue_id) { vlib_main_t *vm; vnet_hw_interface_t *hw; vnet_device_input_runtime_t *rt; vnet_device_and_queue_t *dq; uword idx; hw = vnet_get_hw_interface (vnm, hw_if_index); idx = vnet_get_device_input_thread_index (vnm, hw_if_index, queue_id); vm = vlib_mains[idx]; rt = vlib_node_get_runtime_data (vm, hw->input_node_index); idx = hw->dq_runtime_index_by_queue[queue_id]; dq = vec_elt_at_index (rt->devices_and_queues, idx); clib_atomic_store_rel_n (&(dq->interrupt_pending), 1); vlib_node_set_interrupt_pending (vm, hw->input_node_index); } /* * Acquire RMW Access * Paired with Release Store in vnet_device_input_set_interrupt_pending */ #define foreach_device_and_queue(var,vec) \ for (var = (vec); var < vec_end (vec); var++) \ if ((var->mode == VNET_HW_INTERFACE_RX_MODE_POLLING) \ || clib_atomic_swap_acq_n (&((var)->interrupt_pending), 0)) #endif /* included_vnet_vnet_device_h */ /* * fd.io coding-style-patch-verification: ON * * Local Variables: * eval: (c-set-style "gnu") * End: */