On Fri, Dec 07, 2018 at 10:36:12AM -0600, Tom Tucker wrote:
> On 12/05/2018 02:56 AM, Gal Pressman wrote:
>
> On 04-Dec-18 18:03, Jason Gunthorpe wrote:
>
> On Tue, Dec 04, 2018 at 02:04:26PM +0200, Gal Pressman wrote:
>
> Bitmap allocation service is currently used for assigning
> Protection Domain (PD) numbers.
>
> Signed-off-by: Gal Pressman [1]<galpress@xxxxxxxxxx>
> drivers/infiniband/hw/efa/efa_bitmap.c | 76 ++++++++++++++++++++++++++++++++++
> 1 file changed, 76 insertions(+)
> create mode 100644 drivers/infiniband/hw/efa/efa_bitmap.c
>
> diff --git a/drivers/infiniband/hw/efa/efa_bitmap.c b/drivers/infiniband/hw/efa/
> efa_bitmap.c
> new file mode 100644
> index 000000000000..251cc68d25f5
> +++ b/drivers/infiniband/hw/efa/efa_bitmap.c
> @@ -0,0 +1,76 @@
> +// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
> +/*
> + * Copyright 2006, 2007 Cisco Systems, Inc. All rights reserved.
> + * Copyright 2007, 2008 Mellanox Technologies. All rights reserved.
> + * Copyright 2018 Amazon.com, Inc. or its affiliates.
> + */
> +
> +#include <linux/bitmap.h>
> +
> +#include "efa.h"
> +
> +u32 efa_bitmap_alloc(struct efa_bitmap *bitmap)
> +{
> + u32 obj;
> +
> + spin_lock(&bitmap->lock);
> +
> + obj = find_next_zero_bit(bitmap->table, bitmap->max, bitmap->last);
> + if (obj >= bitmap->max)
> + obj = find_first_zero_bit(bitmap->table, bitmap->max);
> +
> + if (obj < bitmap->max) {
> + set_bit(obj, bitmap->table);
> + bitmap->last = obj + 1;
> + if (bitmap->last == bitmap->max)
> + bitmap->last = 0;
> + } else {
> + obj = EFA_BITMAP_INVAL;
> + }
> +
> + if (obj != EFA_BITMAP_INVAL)
> + --bitmap->avail;
> +
> + spin_unlock(&bitmap->lock);
> +
> + return obj;
> +}
>
> Isn't this just an ida?
>
> Jason
>
>
> Will examine.
>
> The idr_alloc() service uses radix trees and is not spin locked. They
> would need to wrap the idr calls anyway and not care about the
> performance implications.
Assiging protection domain numbers is not a performance path - and
these days ida has an internal lock that callers can use.
For the simple case of < 64*64 the performance will be very similar. A
few branches more for the IDA, but the code is also more likely to be
in cache.
Jason
