diff options
author | Roland Dreier <rolandd@cisco.com> | 2008-07-22 14:19:40 -0700 |
---|---|---|
committer | Roland Dreier <rolandd@cisco.com> | 2008-07-22 14:19:40 -0700 |
commit | e4044cfc493338cd09870bd45dc646336bb66e9f (patch) | |
tree | 2385c18ee6a79393b4e1c5c25ee1f0f980d7da93 /drivers/net/mlx4/mr.c | |
parent | 899698dad72340b562478b8b770317f2f0fe0c09 (diff) | |
download | linux-3.10-e4044cfc493338cd09870bd45dc646336bb66e9f.tar.gz linux-3.10-e4044cfc493338cd09870bd45dc646336bb66e9f.tar.bz2 linux-3.10-e4044cfc493338cd09870bd45dc646336bb66e9f.zip |
mlx4_core: Keep free count for MTT buddy allocator
MTT entries are allocated with a buddy allocator, which just keeps
bitmaps for each level of the buddy table. However, all free space
starts out at the highest order, and small allocations start scanning
from the lowest order. When the lowest order tables have no free
space, this can lead to scanning potentially millions of bits before
finding a free entry at a higher order.
We can avoid this by just keeping a count of how many free entries
each order has, and skipping the bitmap scan when an order is
completely empty. This provides a nice performance boost for a
negligible increase in memory usage.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Diffstat (limited to 'drivers/net/mlx4/mr.c')
-rw-r--r-- | drivers/net/mlx4/mr.c | 26 |
1 files changed, 18 insertions, 8 deletions
diff --git a/drivers/net/mlx4/mr.c b/drivers/net/mlx4/mr.c index 03a9abcce52..b3ea93b9868 100644 --- a/drivers/net/mlx4/mr.c +++ b/drivers/net/mlx4/mr.c @@ -79,23 +79,26 @@ static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order) spin_lock(&buddy->lock); - for (o = order; o <= buddy->max_order; ++o) { - m = 1 << (buddy->max_order - o); - seg = find_first_bit(buddy->bits[o], m); - if (seg < m) - goto found; - } + for (o = order; o <= buddy->max_order; ++o) + if (buddy->num_free[o]) { + m = 1 << (buddy->max_order - o); + seg = find_first_bit(buddy->bits[o], m); + if (seg < m) + goto found; + } spin_unlock(&buddy->lock); return -1; found: clear_bit(seg, buddy->bits[o]); + --buddy->num_free[o]; while (o > order) { --o; seg <<= 1; set_bit(seg ^ 1, buddy->bits[o]); + ++buddy->num_free[o]; } spin_unlock(&buddy->lock); @@ -113,11 +116,13 @@ static void mlx4_buddy_free(struct mlx4_buddy *buddy, u32 seg, int order) while (test_bit(seg ^ 1, buddy->bits[order])) { clear_bit(seg ^ 1, buddy->bits[order]); + --buddy->num_free[order]; seg >>= 1; ++order; } set_bit(seg, buddy->bits[order]); + ++buddy->num_free[order]; spin_unlock(&buddy->lock); } @@ -131,7 +136,9 @@ static int mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order) buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *), GFP_KERNEL); - if (!buddy->bits) + buddy->num_free = kzalloc((buddy->max_order + 1) * sizeof (int *), + GFP_KERNEL); + if (!buddy->bits || !buddy->num_free) goto err_out; for (i = 0; i <= buddy->max_order; ++i) { @@ -143,6 +150,7 @@ static int mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order) } set_bit(0, buddy->bits[buddy->max_order]); + buddy->num_free[buddy->max_order] = 1; return 0; @@ -150,9 +158,10 @@ err_out_free: for (i = 0; i <= buddy->max_order; ++i) kfree(buddy->bits[i]); +err_out: kfree(buddy->bits); + kfree(buddy->num_free); -err_out: return -ENOMEM; } @@ -164,6 +173,7 @@ static void mlx4_buddy_cleanup(struct mlx4_buddy *buddy) kfree(buddy->bits[i]); kfree(buddy->bits); + kfree(buddy->num_free); } static u32 mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order) |