[dpdk-dev] [PATCH v6 4/4] lib/lpm: use atomic store to avoid partial update

Ruifeng Wang ruifeng.wang at arm.com
Thu Jul 18 08:22:30 CEST 2019


Compiler could generate non-atomic stores for whole table entry
updating. This may cause incorrect nexthop to be returned, if
the byte with valid flag is updated prior to the byte with nexthop
is updated.
Besides, field by field updating of table entries follow
read-modify-write sequences. The operations are not atomic,
nor efficient. And could cause entries out of synchronization.

Changed to use atomic store to update whole table entry.

Suggested-by: Medvedkin Vladimir <vladimir.medvedkin at intel.com>
Suggested-by: Gavin Hu <gavin.hu at arm.com>
Signed-off-by: Ruifeng Wang <ruifeng.wang at arm.com>
Reviewed-by: Gavin Hu <gavin.hu at arm.com>
---
 lib/librte_lpm/rte_lpm.c | 138 +++++++++++++++++++++++++++++----------
 1 file changed, 104 insertions(+), 34 deletions(-)

diff --git a/lib/librte_lpm/rte_lpm.c b/lib/librte_lpm/rte_lpm.c
index 0a94630db..95e2f75aa 100644
--- a/lib/librte_lpm/rte_lpm.c
+++ b/lib/librte_lpm/rte_lpm.c
@@ -654,11 +654,19 @@ tbl8_alloc_v20(struct rte_lpm_tbl_entry_v20 *tbl8)
 		tbl8_entry = &tbl8[group_idx * RTE_LPM_TBL8_GROUP_NUM_ENTRIES];
 		/* If a free tbl8 group is found clean it and set as VALID. */
 		if (!tbl8_entry->valid_group) {
+			struct rte_lpm_tbl_entry_v20 new_tbl8_entry = {
+				.valid = INVALID,
+				.depth = 0,
+				.valid_group = VALID,
+			};
+			new_tbl8_entry.next_hop = 0;
+
 			memset(&tbl8_entry[0], 0,
 					RTE_LPM_TBL8_GROUP_NUM_ENTRIES *
 					sizeof(tbl8_entry[0]));
 
-			tbl8_entry->valid_group = VALID;
+			__atomic_store(tbl8_entry, &new_tbl8_entry,
+					__ATOMIC_RELAXED);
 
 			/* Return group index for allocated tbl8 group. */
 			return group_idx;
@@ -680,11 +688,19 @@ tbl8_alloc_v1604(struct rte_lpm_tbl_entry *tbl8, uint32_t number_tbl8s)
 		tbl8_entry = &tbl8[group_idx * RTE_LPM_TBL8_GROUP_NUM_ENTRIES];
 		/* If a free tbl8 group is found clean it and set as VALID. */
 		if (!tbl8_entry->valid_group) {
+			struct rte_lpm_tbl_entry new_tbl8_entry = {
+				.next_hop = 0,
+				.valid = INVALID,
+				.depth = 0,
+				.valid_group = VALID,
+			};
+
 			memset(&tbl8_entry[0], 0,
 					RTE_LPM_TBL8_GROUP_NUM_ENTRIES *
 					sizeof(tbl8_entry[0]));
 
-			tbl8_entry->valid_group = VALID;
+			__atomic_store(tbl8_entry, &new_tbl8_entry,
+					__ATOMIC_RELAXED);
 
 			/* Return group index for allocated tbl8 group. */
 			return group_idx;
@@ -699,14 +715,25 @@ static void
 tbl8_free_v20(struct rte_lpm_tbl_entry_v20 *tbl8, uint32_t tbl8_group_start)
 {
 	/* Set tbl8 group invalid*/
-	tbl8[tbl8_group_start].valid_group = INVALID;
+	struct rte_lpm_tbl_entry_v20 zero_tbl8_entry = {
+		.valid = INVALID,
+		.depth = 0,
+		.valid_group = INVALID,
+	};
+	zero_tbl8_entry.next_hop = 0;
+
+	__atomic_store(&tbl8[tbl8_group_start], &zero_tbl8_entry,
+			__ATOMIC_RELAXED);
 }
 
 static void
 tbl8_free_v1604(struct rte_lpm_tbl_entry *tbl8, uint32_t tbl8_group_start)
 {
 	/* Set tbl8 group invalid*/
-	tbl8[tbl8_group_start].valid_group = INVALID;
+	struct rte_lpm_tbl_entry zero_tbl8_entry = {0};
+
+	__atomic_store(&tbl8[tbl8_group_start], &zero_tbl8_entry,
+			__ATOMIC_RELAXED);
 }
 
 static __rte_noinline int32_t
@@ -767,7 +794,9 @@ add_depth_small_v20(struct rte_lpm_v20 *lpm, uint32_t ip, uint8_t depth,
 					 * Setting tbl8 entry in one go to avoid
 					 * race conditions
 					 */
-					lpm->tbl8[j] = new_tbl8_entry;
+					__atomic_store(&lpm->tbl8[j],
+						&new_tbl8_entry,
+						__ATOMIC_RELAXED);
 
 					continue;
 				}
@@ -837,7 +866,9 @@ add_depth_small_v1604(struct rte_lpm *lpm, uint32_t ip, uint8_t depth,
 					 * Setting tbl8 entry in one go to avoid
 					 * race conditions
 					 */
-					lpm->tbl8[j] = new_tbl8_entry;
+					__atomic_store(&lpm->tbl8[j],
+						&new_tbl8_entry,
+						__ATOMIC_RELAXED);
 
 					continue;
 				}
@@ -875,9 +906,14 @@ add_depth_big_v20(struct rte_lpm_v20 *lpm, uint32_t ip_masked, uint8_t depth,
 
 		/* Set tbl8 entry. */
 		for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
-			lpm->tbl8[i].depth = depth;
-			lpm->tbl8[i].next_hop = next_hop;
-			lpm->tbl8[i].valid = VALID;
+			struct rte_lpm_tbl_entry_v20 new_tbl8_entry = {
+				.valid = VALID,
+				.depth = depth,
+				.valid_group = lpm->tbl8[i].valid_group,
+			};
+			new_tbl8_entry.next_hop = next_hop;
+			__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+					__ATOMIC_RELAXED);
 		}
 
 		/*
@@ -912,19 +948,29 @@ add_depth_big_v20(struct rte_lpm_v20 *lpm, uint32_t ip_masked, uint8_t depth,
 
 		/* Populate new tbl8 with tbl24 value. */
 		for (i = tbl8_group_start; i < tbl8_group_end; i++) {
-			lpm->tbl8[i].valid = VALID;
-			lpm->tbl8[i].depth = lpm->tbl24[tbl24_index].depth;
-			lpm->tbl8[i].next_hop =
-					lpm->tbl24[tbl24_index].next_hop;
+			struct rte_lpm_tbl_entry_v20 new_tbl8_entry = {
+				.valid = VALID,
+				.depth = lpm->tbl24[tbl24_index].depth,
+				.valid_group = lpm->tbl8[i].valid_group,
+			};
+			new_tbl8_entry.next_hop =
+				lpm->tbl24[tbl24_index].next_hop;
+			__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+					__ATOMIC_RELAXED);
 		}
 
 		tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
 
 		/* Insert new rule into the tbl8 entry. */
 		for (i = tbl8_index; i < tbl8_index + tbl8_range; i++) {
-			lpm->tbl8[i].valid = VALID;
-			lpm->tbl8[i].depth = depth;
-			lpm->tbl8[i].next_hop = next_hop;
+			struct rte_lpm_tbl_entry_v20 new_tbl8_entry = {
+				.valid = VALID,
+				.depth = depth,
+				.valid_group = lpm->tbl8[i].valid_group,
+			};
+			new_tbl8_entry.next_hop = next_hop;
+			__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+					__ATOMIC_RELAXED);
 		}
 
 		/*
@@ -965,7 +1011,8 @@ add_depth_big_v20(struct rte_lpm_v20 *lpm, uint32_t ip_masked, uint8_t depth,
 				 * Setting tbl8 entry in one go to avoid race
 				 * condition
 				 */
-				lpm->tbl8[i] = new_tbl8_entry;
+				__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+						__ATOMIC_RELAXED);
 
 				continue;
 			}
@@ -1003,9 +1050,14 @@ add_depth_big_v1604(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
 
 		/* Set tbl8 entry. */
 		for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
-			lpm->tbl8[i].depth = depth;
-			lpm->tbl8[i].next_hop = next_hop;
-			lpm->tbl8[i].valid = VALID;
+			struct rte_lpm_tbl_entry new_tbl8_entry = {
+				.valid = VALID,
+				.depth = depth,
+				.valid_group = lpm->tbl8[i].valid_group,
+				.next_hop = next_hop,
+			};
+			__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+					__ATOMIC_RELAXED);
 		}
 
 		/*
@@ -1043,19 +1095,28 @@ add_depth_big_v1604(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
 
 		/* Populate new tbl8 with tbl24 value. */
 		for (i = tbl8_group_start; i < tbl8_group_end; i++) {
-			lpm->tbl8[i].valid = VALID;
-			lpm->tbl8[i].depth = lpm->tbl24[tbl24_index].depth;
-			lpm->tbl8[i].next_hop =
-					lpm->tbl24[tbl24_index].next_hop;
+			struct rte_lpm_tbl_entry new_tbl8_entry = {
+				.valid = VALID,
+				.depth = lpm->tbl24[tbl24_index].depth,
+				.valid_group = lpm->tbl8[i].valid_group,
+				.next_hop = lpm->tbl24[tbl24_index].next_hop,
+			};
+			__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+					__ATOMIC_RELAXED);
 		}
 
 		tbl8_index = tbl8_group_start + (ip_masked & 0xFF);
 
 		/* Insert new rule into the tbl8 entry. */
 		for (i = tbl8_index; i < tbl8_index + tbl8_range; i++) {
-			lpm->tbl8[i].valid = VALID;
-			lpm->tbl8[i].depth = depth;
-			lpm->tbl8[i].next_hop = next_hop;
+			struct rte_lpm_tbl_entry new_tbl8_entry = {
+				.valid = VALID,
+				.depth = depth,
+				.valid_group = lpm->tbl8[i].valid_group,
+				.next_hop = next_hop,
+			};
+			__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+					__ATOMIC_RELAXED);
 		}
 
 		/*
@@ -1100,7 +1161,8 @@ add_depth_big_v1604(struct rte_lpm *lpm, uint32_t ip_masked, uint8_t depth,
 				 * Setting tbl8 entry in one go to avoid race
 				 * condition
 				 */
-				lpm->tbl8[i] = new_tbl8_entry;
+				__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+						__ATOMIC_RELAXED);
 
 				continue;
 			}
@@ -1393,7 +1455,9 @@ delete_depth_small_v20(struct rte_lpm_v20 *lpm, uint32_t ip_masked,
 					RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
 
 					if (lpm->tbl8[j].depth <= depth)
-						lpm->tbl8[j] = new_tbl8_entry;
+						__atomic_store(&lpm->tbl8[j],
+							&new_tbl8_entry,
+							__ATOMIC_RELAXED);
 				}
 			}
 		}
@@ -1490,7 +1554,9 @@ delete_depth_small_v1604(struct rte_lpm *lpm, uint32_t ip_masked,
 					RTE_LPM_TBL8_GROUP_NUM_ENTRIES); j++) {
 
 					if (lpm->tbl8[j].depth <= depth)
-						lpm->tbl8[j] = new_tbl8_entry;
+						__atomic_store(&lpm->tbl8[j],
+							&new_tbl8_entry,
+							__ATOMIC_RELAXED);
 				}
 			}
 		}
@@ -1646,7 +1712,8 @@ delete_depth_big_v20(struct rte_lpm_v20 *lpm, uint32_t ip_masked,
 		 */
 		for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 			if (lpm->tbl8[i].depth <= depth)
-				lpm->tbl8[i] = new_tbl8_entry;
+				__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+						__ATOMIC_RELAXED);
 		}
 	}
 
@@ -1677,7 +1744,8 @@ delete_depth_big_v20(struct rte_lpm_v20 *lpm, uint32_t ip_masked,
 		/* Set tbl24 before freeing tbl8 to avoid race condition.
 		 * Prevent the free of the tbl8 group from hoisting.
 		 */
-		lpm->tbl24[tbl24_index] = new_tbl24_entry;
+		__atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
+				__ATOMIC_RELAXED);
 		__atomic_thread_fence(__ATOMIC_RELEASE);
 		tbl8_free_v20(lpm->tbl8, tbl8_group_start);
 	}
@@ -1730,7 +1798,8 @@ delete_depth_big_v1604(struct rte_lpm *lpm, uint32_t ip_masked,
 		 */
 		for (i = tbl8_index; i < (tbl8_index + tbl8_range); i++) {
 			if (lpm->tbl8[i].depth <= depth)
-				lpm->tbl8[i] = new_tbl8_entry;
+				__atomic_store(&lpm->tbl8[i], &new_tbl8_entry,
+						__ATOMIC_RELAXED);
 		}
 	}
 
@@ -1761,7 +1830,8 @@ delete_depth_big_v1604(struct rte_lpm *lpm, uint32_t ip_masked,
 		/* Set tbl24 before freeing tbl8 to avoid race condition.
 		 * Prevent the free of the tbl8 group from hoisting.
 		 */
-		lpm->tbl24[tbl24_index] = new_tbl24_entry;
+		__atomic_store(&lpm->tbl24[tbl24_index], &new_tbl24_entry,
+				__ATOMIC_RELAXED);
 		__atomic_thread_fence(__ATOMIC_RELEASE);
 		tbl8_free_v1604(lpm->tbl8, tbl8_group_start);
 	}
-- 
2.17.1



More information about the dev mailing list