Skip to content

Commit

Permalink
net/mlx5: DR, Fix the threshold that defines when pool sync is initiated
Browse files Browse the repository at this point in the history
commit ecd9c5c upstream.

When deciding whether to start syncing and actually free all the "hot"
ICM chunks, we need to consider the type of the ICM chunks that we're
dealing with. For instance, the amount of available ICM for MODIFY_ACTION
is significantly lower than the usual STE ICM, so the threshold should
account for that - otherwise we can deplete MODIFY_ACTION memory just by
creating and deleting the same modify header action in a continuous loop.

This patch replaces the hard-coded threshold with a dynamic value.

Fixes: 1c58651 ("net/mlx5: DR, ICM memory pools sync optimization")
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  • Loading branch information
kliteyn authored and gregkh committed Mar 2, 2022
1 parent 9703a9e commit d4d1884
Showing 1 changed file with 7 additions and 4 deletions.
11 changes: 7 additions & 4 deletions drivers/net/ethernet/mellanox/mlx5/core/steering/dr_icm_pool.c
Expand Up @@ -4,7 +4,6 @@
#include "dr_types.h"

#define DR_ICM_MODIFY_HDR_ALIGN_BASE 64
#define DR_ICM_SYNC_THRESHOLD_POOL (64 * 1024 * 1024)

struct mlx5dr_icm_pool {
enum mlx5dr_icm_type icm_type;
Expand Down Expand Up @@ -324,10 +323,14 @@ dr_icm_chunk_create(struct mlx5dr_icm_pool *pool,

static bool dr_icm_pool_is_sync_required(struct mlx5dr_icm_pool *pool)
{
if (pool->hot_memory_size > DR_ICM_SYNC_THRESHOLD_POOL)
return true;
int allow_hot_size;

return false;
/* sync when hot memory reaches half of the pool size */
allow_hot_size =
mlx5dr_icm_pool_chunk_size_to_byte(pool->max_log_chunk_sz,
pool->icm_type) / 2;

return pool->hot_memory_size > allow_hot_size;
}

static int dr_icm_pool_sync_all_buddy_pools(struct mlx5dr_icm_pool *pool)
Expand Down

0 comments on commit d4d1884

Please sign in to comment.