Commit e4044cfc authored by Roland Dreier's avatar Roland Dreier
Browse files

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: default avatarRoland Dreier <rolandd@cisco.com>
parent 899698da

drivers/net/mlx4/mlx4.h

+1 −0
Original line number Diff line number Diff line
@@ -118,6 +118,7 @@ struct mlx4_bitmap { 


struct mlx4_buddy {

	unsigned long	      **bits;

	unsigned int	       *num_free;

	int			max_order;

	spinlock_t		lock;

};

drivers/net/mlx4/mr.c

+18 −8
Original line number Diff line number Diff line
@@ -79,7 +79,8 @@ static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order) 


	spin_lock(&buddy->lock);



	for (o = order; o <= buddy->max_order; ++o) {

	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)
@@ -91,11 +92,13 @@ static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order) 


 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)

CRA Git | Maintained and supported by SUSTech CRA and CCSE