[PATCH net-next 0/4] mlxsw: spectrum_acl: Get rid of failed_rollback mechanism

[PATCH net-next 0/4] mlxsw: spectrum_acl: Get rid of failed_rollback mechanism

[Ido Schimmel]

Jiri says:

Currently if vregion rehash fails, it rolls back to the original ERP
set. However, in case of unlikely rollback fail, the vregion is in a
zombie state and never gets rehashed again. With the recent changes, it
is possible to try to continue the rollback. Do it from the last failed
ventry.

Jiri Pirko (4):
  mlxsw: spectrum_acl: Remove redundant failed_rollback from
    migrate_start()
  mlxsw: spectrum_acl: Move rehash_dis trace call and err msg to
    vregion_migrate()
  mlxsw: spectrum_acl: Remove failed_rollback dead end
  mlxsw: spectrum_acl: Rename rehash_dis trace

 .../mellanox/mlxsw/spectrum_acl_tcam.c        | 36 +++++++++----------
 include/trace/events/mlxsw.h                  |  2 +-
 2 files changed, 17 insertions(+), 21 deletions(-)

-- 
2.20.1

[PATCH net-next 1/4] mlxsw: spectrum_acl: Remove redundant failed_rollback from migrate_start()

[Ido Schimmel]

From: Jiri Pirko <jiri@mellanox.com>

The flag is set by the caller mlxsw_sp_acl_tcam_vregion_migrate() anyway,
so don't set it here.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c b/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
index 8811f6513e36..a26854f97f0f 100644
--- a/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
+++ b/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
@@ -1256,11 +1256,8 @@ mlxsw_sp_acl_tcam_vchunk_migrate_start(struct mlxsw_sp *mlxsw_sp,
 	struct mlxsw_sp_acl_tcam_chunk *new_chunk;
 
 	new_chunk = mlxsw_sp_acl_tcam_chunk_create(mlxsw_sp, vchunk, region);
-	if (IS_ERR(new_chunk)) {
-		if (ctx->this_is_rollback)
-			vchunk->vregion->failed_rollback = true;
+	if (IS_ERR(new_chunk))
 		return PTR_ERR(new_chunk);
-	}
 	vchunk->chunk2 = vchunk->chunk;
 	vchunk->chunk = new_chunk;
 	ctx->current_vchunk = vchunk;
-- 
2.20.1

[PATCH net-next 2/4] mlxsw: spectrum_acl: Move rehash_dis trace call and err msg to vregion_migrate()

[Ido Schimmel]

From: Jiri Pirko <jiri@mellanox.com>

Move the call of rehash_dis trace and the error message to
vregion_migrate() next to the failed_rollback flag set.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
---
 .../net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c   | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c b/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
index a26854f97f0f..a2af385bcd63 100644
--- a/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
+++ b/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
@@ -1394,8 +1394,12 @@ mlxsw_sp_acl_tcam_vregion_migrate(struct mlxsw_sp *mlxsw_sp,
 		ctx->this_is_rollback = true;
 		err2 = mlxsw_sp_acl_tcam_vchunk_migrate_all(mlxsw_sp, vregion,
 							    ctx, credits);
-		if (err2)
+		if (err2) {
 			vregion->failed_rollback = true;
+			trace_mlxsw_sp_acl_tcam_vregion_rehash_dis(mlxsw_sp,
+								   vregion);
+			dev_err(mlxsw_sp->bus_info->dev, "Failed to rollback during vregion migration fail\n");
+		}
 	}
 	mutex_unlock(&vregion->lock);
 	trace_mlxsw_sp_acl_tcam_vregion_migrate_end(mlxsw_sp, vregion);
@@ -1503,11 +1507,6 @@ mlxsw_sp_acl_tcam_vregion_rehash(struct mlxsw_sp *mlxsw_sp,
 						ctx, credits);
 	if (err) {
 		dev_err(mlxsw_sp->bus_info->dev, "Failed to migrate vregion\n");
-		if (vregion->failed_rollback) {
-			trace_mlxsw_sp_acl_tcam_vregion_rehash_dis(mlxsw_sp,
-								   vregion);
-			dev_err(mlxsw_sp->bus_info->dev, "Failed to rollback during vregion migration fail\n");
-		}
 	}
 
 	if (*credits >= 0)
-- 
2.20.1

[PATCH net-next 3/4] mlxsw: spectrum_acl: Remove failed_rollback dead end

[Ido Schimmel]

From: Jiri Pirko <jiri@mellanox.com>

Currently if a rollback ends with error, the vregion is in a zombie
state until end of the existence. Instead of that, rather try to
continue where rollback ended later on (after rehash interval).

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
---
 .../mellanox/mlxsw/spectrum_acl_tcam.c        | 20 +++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c b/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
index a2af385bcd63..2f0b61b87c99 100644
--- a/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
+++ b/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
@@ -216,7 +216,6 @@ struct mlxsw_sp_acl_tcam_vregion {
 		struct mlxsw_sp_acl_tcam_rehash_ctx ctx;
 	} rehash;
 	struct mlxsw_sp *mlxsw_sp;
-	bool failed_rollback; /* Indicates failed rollback during migration */
 	unsigned int ref_count;
 };
 
@@ -1315,8 +1314,13 @@ mlxsw_sp_acl_tcam_vchunk_migrate_one(struct mlxsw_sp *mlxsw_sp,
 		err = mlxsw_sp_acl_tcam_ventry_migrate(mlxsw_sp, ventry,
 						       vchunk->chunk, credits);
 		if (err) {
-			if (ctx->this_is_rollback)
+			if (ctx->this_is_rollback) {
+				/* Save the ventry which we ended with and try
+				 * to continue later on.
+				 */
+				ctx->start_ventry = ventry;
 				return err;
+			}
 			/* Swap the chunk and chunk2 pointers so the follow-up
 			 * rollback call will see the original chunk pointer
 			 * in vchunk->chunk.
@@ -1395,10 +1399,10 @@ mlxsw_sp_acl_tcam_vregion_migrate(struct mlxsw_sp *mlxsw_sp,
 		err2 = mlxsw_sp_acl_tcam_vchunk_migrate_all(mlxsw_sp, vregion,
 							    ctx, credits);
 		if (err2) {
-			vregion->failed_rollback = true;
 			trace_mlxsw_sp_acl_tcam_vregion_rehash_dis(mlxsw_sp,
 								   vregion);
 			dev_err(mlxsw_sp->bus_info->dev, "Failed to rollback during vregion migration fail\n");
+			/* Let the rollback to be continued later on. */
 		}
 	}
 	mutex_unlock(&vregion->lock);
@@ -1424,8 +1428,6 @@ mlxsw_sp_acl_tcam_vregion_rehash_start(struct mlxsw_sp *mlxsw_sp,
 	int err;
 
 	trace_mlxsw_sp_acl_tcam_vregion_rehash(mlxsw_sp, vregion);
-	if (vregion->failed_rollback)
-		return -EBUSY;
 
 	hints_priv = ops->region_rehash_hints_get(vregion->region->priv);
 	if (IS_ERR(hints_priv))
@@ -1472,11 +1474,9 @@ mlxsw_sp_acl_tcam_vregion_rehash_end(struct mlxsw_sp *mlxsw_sp,
 	struct mlxsw_sp_acl_tcam_region *unused_region = vregion->region2;
 	const struct mlxsw_sp_acl_tcam_ops *ops = mlxsw_sp->acl_tcam_ops;
 
-	if (!vregion->failed_rollback) {
-		vregion->region2 = NULL;
-		mlxsw_sp_acl_tcam_group_region_detach(mlxsw_sp, unused_region);
-		mlxsw_sp_acl_tcam_region_destroy(mlxsw_sp, unused_region);
-	}
+	vregion->region2 = NULL;
+	mlxsw_sp_acl_tcam_group_region_detach(mlxsw_sp, unused_region);
+	mlxsw_sp_acl_tcam_region_destroy(mlxsw_sp, unused_region);
 	ops->region_rehash_hints_put(ctx->hints_priv);
 	ctx->hints_priv = NULL;
 }
-- 
2.20.1

[PATCH net-next 4/4] mlxsw: spectrum_acl: Rename rehash_dis trace

[Ido Schimmel]

From: Jiri Pirko <jiri@mellanox.com>

The name of the trace is no longer correct, since there is no disable of
rehash done. So name it "rehash_rollback_failed".

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c | 4 ++--
 include/trace/events/mlxsw.h                            | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c b/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
index 2f0b61b87c99..e993159e8e4c 100644
--- a/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
+++ b/drivers/net/ethernet/mellanox/mlxsw/spectrum_acl_tcam.c
@@ -1399,8 +1399,8 @@ mlxsw_sp_acl_tcam_vregion_migrate(struct mlxsw_sp *mlxsw_sp,
 		err2 = mlxsw_sp_acl_tcam_vchunk_migrate_all(mlxsw_sp, vregion,
 							    ctx, credits);
 		if (err2) {
-			trace_mlxsw_sp_acl_tcam_vregion_rehash_dis(mlxsw_sp,
-								   vregion);
+			trace_mlxsw_sp_acl_tcam_vregion_rehash_rollback_failed(mlxsw_sp,
+									       vregion);
 			dev_err(mlxsw_sp->bus_info->dev, "Failed to rollback during vregion migration fail\n");
 			/* Let the rollback to be continued later on. */
 		}
diff --git a/include/trace/events/mlxsw.h b/include/trace/events/mlxsw.h
index 6a4cfaef33a2..19a25ed323a5 100644
--- a/include/trace/events/mlxsw.h
+++ b/include/trace/events/mlxsw.h
@@ -93,7 +93,7 @@ TRACE_EVENT(mlxsw_sp_acl_tcam_vregion_migrate_end,
 		  __entry->mlxsw_sp, __entry->vregion)
 );
 
-TRACE_EVENT(mlxsw_sp_acl_tcam_vregion_rehash_dis,
+TRACE_EVENT(mlxsw_sp_acl_tcam_vregion_rehash_rollback_failed,
 	TP_PROTO(const struct mlxsw_sp *mlxsw_sp,
 		 const struct mlxsw_sp_acl_tcam_vregion *vregion),
 
-- 
2.20.1

Re: [PATCH net-next 0/4] mlxsw: spectrum_acl: Get rid of failed_rollback mechanism

[David Miller]

From: Ido Schimmel <idosch@mellanox.com>
Date: Sun, 31 Mar 2019 06:49:37 +0000

> Jiri says:
> 
> Currently if vregion rehash fails, it rolls back to the original ERP
> set. However, in case of unlikely rollback fail, the vregion is in a
> zombie state and never gets rehashed again. With the recent changes, it
> is possible to try to continue the rollback. Do it from the last failed
> ventry.

Series applied.