2007-05-09 09:00:38 +08:00
|
|
|
/*
|
|
|
|
* Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
|
2008-07-26 01:32:52 +08:00
|
|
|
* Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
|
2007-05-09 09:00:38 +08:00
|
|
|
*
|
|
|
|
* This software is available to you under a choice of one of two
|
|
|
|
* licenses. You may choose to be licensed under the terms of the GNU
|
|
|
|
* General Public License (GPL) Version 2, available from the file
|
|
|
|
* COPYING in the main directory of this source tree, or the
|
|
|
|
* OpenIB.org BSD license below:
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or
|
|
|
|
* without modification, are permitted provided that the following
|
|
|
|
* conditions are met:
|
|
|
|
*
|
|
|
|
* - Redistributions of source code must retain the above
|
|
|
|
* copyright notice, this list of conditions and the following
|
|
|
|
* disclaimer.
|
|
|
|
*
|
|
|
|
* - Redistributions in binary form must reproduce the above
|
|
|
|
* copyright notice, this list of conditions and the following
|
|
|
|
* disclaimer in the documentation and/or other materials
|
|
|
|
* provided with the distribution.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
|
|
|
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
|
|
|
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
|
|
|
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
|
|
|
|
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
|
|
|
|
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
|
|
|
|
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
|
|
* SOFTWARE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/errno.h>
|
|
|
|
#include <linux/slab.h>
|
2008-09-06 05:04:07 +08:00
|
|
|
#include <linux/mm.h>
|
2011-05-28 04:14:23 +08:00
|
|
|
#include <linux/export.h>
|
2007-05-09 09:00:38 +08:00
|
|
|
#include <linux/bitmap.h>
|
2007-05-16 03:36:30 +08:00
|
|
|
#include <linux/dma-mapping.h>
|
2008-02-11 10:22:57 +08:00
|
|
|
#include <linux/vmalloc.h>
|
2007-05-09 09:00:38 +08:00
|
|
|
|
|
|
|
#include "mlx4.h"
|
|
|
|
|
|
|
|
u32 mlx4_bitmap_alloc(struct mlx4_bitmap *bitmap)
|
|
|
|
{
|
|
|
|
u32 obj;
|
|
|
|
|
|
|
|
spin_lock(&bitmap->lock);
|
|
|
|
|
|
|
|
obj = find_next_zero_bit(bitmap->table, bitmap->max, bitmap->last);
|
|
|
|
if (obj >= bitmap->max) {
|
2008-10-23 01:25:29 +08:00
|
|
|
bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
|
|
|
|
& bitmap->mask;
|
2007-05-09 09:00:38 +08:00
|
|
|
obj = find_first_zero_bit(bitmap->table, bitmap->max);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (obj < bitmap->max) {
|
|
|
|
set_bit(obj, bitmap->table);
|
2008-10-23 01:25:29 +08:00
|
|
|
bitmap->last = (obj + 1);
|
|
|
|
if (bitmap->last == bitmap->max)
|
|
|
|
bitmap->last = 0;
|
2007-05-09 09:00:38 +08:00
|
|
|
obj |= bitmap->top;
|
|
|
|
} else
|
|
|
|
obj = -1;
|
|
|
|
|
2011-03-23 06:38:45 +08:00
|
|
|
if (obj != -1)
|
|
|
|
--bitmap->avail;
|
|
|
|
|
2007-05-09 09:00:38 +08:00
|
|
|
spin_unlock(&bitmap->lock);
|
|
|
|
|
|
|
|
return obj;
|
|
|
|
}
|
|
|
|
|
mlx4_core: Roll back round robin bitmap allocation commit for CQs, SRQs, and MPTs
Commit f4ec9e9 "mlx4_core: Change bitmap allocator to work in round-robin fashion"
introduced round-robin allocation (via bitmap) for all resources which allocate
via a bitmap.
Round robin allocation is desirable for mcgs, counters, pd's, UARs, and xrcds.
These are simply numbers, with no involvement of ICM memory mapping.
Round robin is required for QPs, since we had a problem with immediate
reuse of a 24-bit QP number (commit f4ec9e9).
However, for other resources which use the bitmap allocator and involve
mapping ICM memory -- MPTs, CQs, SRQs -- round-robin is not desirable.
What happens in these cases is the following:
ICM memory is allocated and mapped in chunks of 256K.
Since the resource allocation index goes up monotonically, the allocator
will eventually require mapping a new chunk. Now, chunks are also unmapped
when their reference count goes back to zero. Thus, if a single app is
running and starts/exits frequently we will have the following situation:
When the app starts, a new chunk must be allocated and mapped.
When the app exits, the chunk reference count goes back to zero, and the
chunk is unmapped and freed. Therefore, the app must pay the cost of allocation
and mapping of ICM memory each time it runs (although the price is paid only when
allocating the initial entry in the new chunk).
For apps which allocate MPTs/SRQs/CQs and which operate as described above,
this presented a performance problem.
We therefore roll back the round-robin allocator modification for MPTs, CQs, SRQs.
Reported-by: Matthew Finlay <matt@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-08 22:50:17 +08:00
|
|
|
void mlx4_bitmap_free(struct mlx4_bitmap *bitmap, u32 obj, int use_rr)
|
2007-05-09 09:00:38 +08:00
|
|
|
{
|
mlx4_core: Roll back round robin bitmap allocation commit for CQs, SRQs, and MPTs
Commit f4ec9e9 "mlx4_core: Change bitmap allocator to work in round-robin fashion"
introduced round-robin allocation (via bitmap) for all resources which allocate
via a bitmap.
Round robin allocation is desirable for mcgs, counters, pd's, UARs, and xrcds.
These are simply numbers, with no involvement of ICM memory mapping.
Round robin is required for QPs, since we had a problem with immediate
reuse of a 24-bit QP number (commit f4ec9e9).
However, for other resources which use the bitmap allocator and involve
mapping ICM memory -- MPTs, CQs, SRQs -- round-robin is not desirable.
What happens in these cases is the following:
ICM memory is allocated and mapped in chunks of 256K.
Since the resource allocation index goes up monotonically, the allocator
will eventually require mapping a new chunk. Now, chunks are also unmapped
when their reference count goes back to zero. Thus, if a single app is
running and starts/exits frequently we will have the following situation:
When the app starts, a new chunk must be allocated and mapped.
When the app exits, the chunk reference count goes back to zero, and the
chunk is unmapped and freed. Therefore, the app must pay the cost of allocation
and mapping of ICM memory each time it runs (although the price is paid only when
allocating the initial entry in the new chunk).
For apps which allocate MPTs/SRQs/CQs and which operate as described above,
this presented a performance problem.
We therefore roll back the round-robin allocator modification for MPTs, CQs, SRQs.
Reported-by: Matthew Finlay <matt@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-08 22:50:17 +08:00
|
|
|
mlx4_bitmap_free_range(bitmap, obj, 1, use_rr);
|
2008-10-11 03:01:37 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
u32 mlx4_bitmap_alloc_range(struct mlx4_bitmap *bitmap, int cnt, int align)
|
|
|
|
{
|
2010-08-28 03:08:13 +08:00
|
|
|
u32 obj;
|
2008-10-11 03:01:37 +08:00
|
|
|
|
|
|
|
if (likely(cnt == 1 && align == 1))
|
|
|
|
return mlx4_bitmap_alloc(bitmap);
|
|
|
|
|
|
|
|
spin_lock(&bitmap->lock);
|
|
|
|
|
2009-12-16 08:48:29 +08:00
|
|
|
obj = bitmap_find_next_zero_area(bitmap->table, bitmap->max,
|
|
|
|
bitmap->last, cnt, align - 1);
|
2008-10-11 03:01:37 +08:00
|
|
|
if (obj >= bitmap->max) {
|
2008-10-23 01:25:29 +08:00
|
|
|
bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
|
|
|
|
& bitmap->mask;
|
2009-12-16 08:48:29 +08:00
|
|
|
obj = bitmap_find_next_zero_area(bitmap->table, bitmap->max,
|
|
|
|
0, cnt, align - 1);
|
2008-10-11 03:01:37 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (obj < bitmap->max) {
|
2010-08-28 03:08:13 +08:00
|
|
|
bitmap_set(bitmap->table, obj, cnt);
|
2008-10-11 03:01:37 +08:00
|
|
|
if (obj == bitmap->last) {
|
|
|
|
bitmap->last = (obj + cnt);
|
|
|
|
if (bitmap->last >= bitmap->max)
|
|
|
|
bitmap->last = 0;
|
|
|
|
}
|
|
|
|
obj |= bitmap->top;
|
|
|
|
} else
|
|
|
|
obj = -1;
|
|
|
|
|
2011-03-23 06:38:45 +08:00
|
|
|
if (obj != -1)
|
|
|
|
bitmap->avail -= cnt;
|
|
|
|
|
2008-10-11 03:01:37 +08:00
|
|
|
spin_unlock(&bitmap->lock);
|
|
|
|
|
|
|
|
return obj;
|
|
|
|
}
|
|
|
|
|
2011-03-23 06:38:45 +08:00
|
|
|
u32 mlx4_bitmap_avail(struct mlx4_bitmap *bitmap)
|
|
|
|
{
|
|
|
|
return bitmap->avail;
|
|
|
|
}
|
|
|
|
|
mlx4_core: Roll back round robin bitmap allocation commit for CQs, SRQs, and MPTs
Commit f4ec9e9 "mlx4_core: Change bitmap allocator to work in round-robin fashion"
introduced round-robin allocation (via bitmap) for all resources which allocate
via a bitmap.
Round robin allocation is desirable for mcgs, counters, pd's, UARs, and xrcds.
These are simply numbers, with no involvement of ICM memory mapping.
Round robin is required for QPs, since we had a problem with immediate
reuse of a 24-bit QP number (commit f4ec9e9).
However, for other resources which use the bitmap allocator and involve
mapping ICM memory -- MPTs, CQs, SRQs -- round-robin is not desirable.
What happens in these cases is the following:
ICM memory is allocated and mapped in chunks of 256K.
Since the resource allocation index goes up monotonically, the allocator
will eventually require mapping a new chunk. Now, chunks are also unmapped
when their reference count goes back to zero. Thus, if a single app is
running and starts/exits frequently we will have the following situation:
When the app starts, a new chunk must be allocated and mapped.
When the app exits, the chunk reference count goes back to zero, and the
chunk is unmapped and freed. Therefore, the app must pay the cost of allocation
and mapping of ICM memory each time it runs (although the price is paid only when
allocating the initial entry in the new chunk).
For apps which allocate MPTs/SRQs/CQs and which operate as described above,
this presented a performance problem.
We therefore roll back the round-robin allocator modification for MPTs, CQs, SRQs.
Reported-by: Matthew Finlay <matt@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-08 22:50:17 +08:00
|
|
|
void mlx4_bitmap_free_range(struct mlx4_bitmap *bitmap, u32 obj, int cnt,
|
|
|
|
int use_rr)
|
2008-10-11 03:01:37 +08:00
|
|
|
{
|
2008-10-23 01:25:29 +08:00
|
|
|
obj &= bitmap->max + bitmap->reserved_top - 1;
|
2007-05-09 09:00:38 +08:00
|
|
|
|
|
|
|
spin_lock(&bitmap->lock);
|
mlx4_core: Roll back round robin bitmap allocation commit for CQs, SRQs, and MPTs
Commit f4ec9e9 "mlx4_core: Change bitmap allocator to work in round-robin fashion"
introduced round-robin allocation (via bitmap) for all resources which allocate
via a bitmap.
Round robin allocation is desirable for mcgs, counters, pd's, UARs, and xrcds.
These are simply numbers, with no involvement of ICM memory mapping.
Round robin is required for QPs, since we had a problem with immediate
reuse of a 24-bit QP number (commit f4ec9e9).
However, for other resources which use the bitmap allocator and involve
mapping ICM memory -- MPTs, CQs, SRQs -- round-robin is not desirable.
What happens in these cases is the following:
ICM memory is allocated and mapped in chunks of 256K.
Since the resource allocation index goes up monotonically, the allocator
will eventually require mapping a new chunk. Now, chunks are also unmapped
when their reference count goes back to zero. Thus, if a single app is
running and starts/exits frequently we will have the following situation:
When the app starts, a new chunk must be allocated and mapped.
When the app exits, the chunk reference count goes back to zero, and the
chunk is unmapped and freed. Therefore, the app must pay the cost of allocation
and mapping of ICM memory each time it runs (although the price is paid only when
allocating the initial entry in the new chunk).
For apps which allocate MPTs/SRQs/CQs and which operate as described above,
this presented a performance problem.
We therefore roll back the round-robin allocator modification for MPTs, CQs, SRQs.
Reported-by: Matthew Finlay <matt@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-08 22:50:17 +08:00
|
|
|
if (!use_rr) {
|
|
|
|
bitmap->last = min(bitmap->last, obj);
|
|
|
|
bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
|
|
|
|
& bitmap->mask;
|
|
|
|
}
|
2010-08-28 03:08:13 +08:00
|
|
|
bitmap_clear(bitmap->table, obj, cnt);
|
2011-03-23 06:38:45 +08:00
|
|
|
bitmap->avail += cnt;
|
2007-05-09 09:00:38 +08:00
|
|
|
spin_unlock(&bitmap->lock);
|
|
|
|
}
|
|
|
|
|
2008-10-23 01:25:29 +08:00
|
|
|
int mlx4_bitmap_init(struct mlx4_bitmap *bitmap, u32 num, u32 mask,
|
|
|
|
u32 reserved_bot, u32 reserved_top)
|
2007-05-09 09:00:38 +08:00
|
|
|
{
|
|
|
|
/* num must be a power of 2 */
|
|
|
|
if (num != roundup_pow_of_two(num))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
bitmap->last = 0;
|
|
|
|
bitmap->top = 0;
|
2008-10-23 01:25:29 +08:00
|
|
|
bitmap->max = num - reserved_top;
|
2007-05-09 09:00:38 +08:00
|
|
|
bitmap->mask = mask;
|
2008-10-23 01:25:29 +08:00
|
|
|
bitmap->reserved_top = reserved_top;
|
2011-03-23 06:38:45 +08:00
|
|
|
bitmap->avail = num - reserved_top - reserved_bot;
|
2007-05-09 09:00:38 +08:00
|
|
|
spin_lock_init(&bitmap->lock);
|
2008-10-23 01:25:29 +08:00
|
|
|
bitmap->table = kzalloc(BITS_TO_LONGS(bitmap->max) *
|
|
|
|
sizeof (long), GFP_KERNEL);
|
2007-05-09 09:00:38 +08:00
|
|
|
if (!bitmap->table)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2010-08-28 03:08:13 +08:00
|
|
|
bitmap_set(bitmap->table, 0, reserved_bot);
|
2007-05-09 09:00:38 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
void mlx4_bitmap_cleanup(struct mlx4_bitmap *bitmap)
|
|
|
|
{
|
|
|
|
kfree(bitmap->table);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Handling for queue buffers -- we allocate a bunch of memory and
|
|
|
|
* register it in a memory region at HCA virtual address 0. If the
|
|
|
|
* requested size is > max_direct, we split the allocation into
|
|
|
|
* multiple pages, so we don't require too much contiguous memory.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
|
2014-05-11 20:15:12 +08:00
|
|
|
struct mlx4_buf *buf, gfp_t gfp)
|
2007-05-09 09:00:38 +08:00
|
|
|
{
|
|
|
|
dma_addr_t t;
|
|
|
|
|
|
|
|
if (size <= max_direct) {
|
|
|
|
buf->nbufs = 1;
|
|
|
|
buf->npages = 1;
|
|
|
|
buf->page_shift = get_order(size) + PAGE_SHIFT;
|
2008-02-07 13:17:59 +08:00
|
|
|
buf->direct.buf = dma_alloc_coherent(&dev->pdev->dev,
|
2014-05-11 20:15:12 +08:00
|
|
|
size, &t, gfp);
|
2008-02-07 13:17:59 +08:00
|
|
|
if (!buf->direct.buf)
|
2007-05-09 09:00:38 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
|
2008-02-07 13:17:59 +08:00
|
|
|
buf->direct.map = t;
|
2007-05-09 09:00:38 +08:00
|
|
|
|
|
|
|
while (t & ((1 << buf->page_shift) - 1)) {
|
|
|
|
--buf->page_shift;
|
|
|
|
buf->npages *= 2;
|
|
|
|
}
|
|
|
|
|
2008-02-07 13:17:59 +08:00
|
|
|
memset(buf->direct.buf, 0, size);
|
2007-05-09 09:00:38 +08:00
|
|
|
} else {
|
|
|
|
int i;
|
|
|
|
|
2011-01-11 09:42:06 +08:00
|
|
|
buf->direct.buf = NULL;
|
2007-05-09 09:00:38 +08:00
|
|
|
buf->nbufs = (size + PAGE_SIZE - 1) / PAGE_SIZE;
|
|
|
|
buf->npages = buf->nbufs;
|
|
|
|
buf->page_shift = PAGE_SHIFT;
|
2010-08-11 15:02:48 +08:00
|
|
|
buf->page_list = kcalloc(buf->nbufs, sizeof(*buf->page_list),
|
2014-05-11 20:15:12 +08:00
|
|
|
gfp);
|
2008-02-07 13:17:59 +08:00
|
|
|
if (!buf->page_list)
|
2007-05-09 09:00:38 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
for (i = 0; i < buf->nbufs; ++i) {
|
2008-02-07 13:17:59 +08:00
|
|
|
buf->page_list[i].buf =
|
2007-05-09 09:00:38 +08:00
|
|
|
dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
|
2014-05-11 20:15:12 +08:00
|
|
|
&t, gfp);
|
2008-02-07 13:17:59 +08:00
|
|
|
if (!buf->page_list[i].buf)
|
2007-05-09 09:00:38 +08:00
|
|
|
goto err_free;
|
|
|
|
|
2008-02-07 13:17:59 +08:00
|
|
|
buf->page_list[i].map = t;
|
2007-05-09 09:00:38 +08:00
|
|
|
|
2008-02-07 13:17:59 +08:00
|
|
|
memset(buf->page_list[i].buf, 0, PAGE_SIZE);
|
2007-05-09 09:00:38 +08:00
|
|
|
}
|
2008-01-28 16:40:51 +08:00
|
|
|
|
|
|
|
if (BITS_PER_LONG == 64) {
|
|
|
|
struct page **pages;
|
2014-05-11 20:15:12 +08:00
|
|
|
pages = kmalloc(sizeof *pages * buf->nbufs, gfp);
|
2008-01-28 16:40:51 +08:00
|
|
|
if (!pages)
|
|
|
|
goto err_free;
|
|
|
|
for (i = 0; i < buf->nbufs; ++i)
|
2008-02-07 13:17:59 +08:00
|
|
|
pages[i] = virt_to_page(buf->page_list[i].buf);
|
|
|
|
buf->direct.buf = vmap(pages, buf->nbufs, VM_MAP, PAGE_KERNEL);
|
2008-01-28 16:40:51 +08:00
|
|
|
kfree(pages);
|
2008-02-07 13:17:59 +08:00
|
|
|
if (!buf->direct.buf)
|
2008-01-28 16:40:51 +08:00
|
|
|
goto err_free;
|
|
|
|
}
|
2007-05-09 09:00:38 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
err_free:
|
|
|
|
mlx4_buf_free(dev, size, buf);
|
|
|
|
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(mlx4_buf_alloc);
|
|
|
|
|
|
|
|
void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
if (buf->nbufs == 1)
|
2008-02-07 13:17:59 +08:00
|
|
|
dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
|
|
|
|
buf->direct.map);
|
2007-05-09 09:00:38 +08:00
|
|
|
else {
|
2011-01-11 09:42:06 +08:00
|
|
|
if (BITS_PER_LONG == 64 && buf->direct.buf)
|
2008-02-07 13:17:59 +08:00
|
|
|
vunmap(buf->direct.buf);
|
2008-01-28 16:40:51 +08:00
|
|
|
|
2007-05-09 09:00:38 +08:00
|
|
|
for (i = 0; i < buf->nbufs; ++i)
|
2008-02-07 13:17:59 +08:00
|
|
|
if (buf->page_list[i].buf)
|
2007-11-14 07:26:57 +08:00
|
|
|
dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
|
2008-02-07 13:17:59 +08:00
|
|
|
buf->page_list[i].buf,
|
|
|
|
buf->page_list[i].map);
|
|
|
|
kfree(buf->page_list);
|
2007-05-09 09:00:38 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(mlx4_buf_free);
|
2008-04-24 02:55:45 +08:00
|
|
|
|
2014-05-11 20:15:12 +08:00
|
|
|
static struct mlx4_db_pgdir *mlx4_alloc_db_pgdir(struct device *dma_device,
|
|
|
|
gfp_t gfp)
|
2008-04-24 02:55:45 +08:00
|
|
|
{
|
|
|
|
struct mlx4_db_pgdir *pgdir;
|
|
|
|
|
2014-05-11 20:15:12 +08:00
|
|
|
pgdir = kzalloc(sizeof *pgdir, gfp);
|
2008-04-24 02:55:45 +08:00
|
|
|
if (!pgdir)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
bitmap_fill(pgdir->order1, MLX4_DB_PER_PAGE / 2);
|
|
|
|
pgdir->bits[0] = pgdir->order0;
|
|
|
|
pgdir->bits[1] = pgdir->order1;
|
|
|
|
pgdir->db_page = dma_alloc_coherent(dma_device, PAGE_SIZE,
|
2014-05-11 20:15:12 +08:00
|
|
|
&pgdir->db_dma, gfp);
|
2008-04-24 02:55:45 +08:00
|
|
|
if (!pgdir->db_page) {
|
|
|
|
kfree(pgdir);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return pgdir;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir *pgdir,
|
|
|
|
struct mlx4_db *db, int order)
|
|
|
|
{
|
|
|
|
int o;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (o = order; o <= 1; ++o) {
|
|
|
|
i = find_first_bit(pgdir->bits[o], MLX4_DB_PER_PAGE >> o);
|
|
|
|
if (i < MLX4_DB_PER_PAGE >> o)
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
found:
|
|
|
|
clear_bit(i, pgdir->bits[o]);
|
|
|
|
|
|
|
|
i <<= o;
|
|
|
|
|
|
|
|
if (o > order)
|
|
|
|
set_bit(i ^ 1, pgdir->bits[order]);
|
|
|
|
|
|
|
|
db->u.pgdir = pgdir;
|
|
|
|
db->index = i;
|
|
|
|
db->db = pgdir->db_page + db->index;
|
|
|
|
db->dma = pgdir->db_dma + db->index * 4;
|
|
|
|
db->order = order;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2014-05-11 20:15:12 +08:00
|
|
|
int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order, gfp_t gfp)
|
2008-04-24 02:55:45 +08:00
|
|
|
{
|
|
|
|
struct mlx4_priv *priv = mlx4_priv(dev);
|
|
|
|
struct mlx4_db_pgdir *pgdir;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
mutex_lock(&priv->pgdir_mutex);
|
|
|
|
|
|
|
|
list_for_each_entry(pgdir, &priv->pgdir_list, list)
|
|
|
|
if (!mlx4_alloc_db_from_pgdir(pgdir, db, order))
|
|
|
|
goto out;
|
|
|
|
|
2014-05-11 20:15:12 +08:00
|
|
|
pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev), gfp);
|
2008-04-24 02:55:45 +08:00
|
|
|
if (!pgdir) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
list_add(&pgdir->list, &priv->pgdir_list);
|
|
|
|
|
|
|
|
/* This should never fail -- we just allocated an empty page: */
|
|
|
|
WARN_ON(mlx4_alloc_db_from_pgdir(pgdir, db, order));
|
|
|
|
|
|
|
|
out:
|
|
|
|
mutex_unlock(&priv->pgdir_mutex);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(mlx4_db_alloc);
|
|
|
|
|
|
|
|
void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db)
|
|
|
|
{
|
|
|
|
struct mlx4_priv *priv = mlx4_priv(dev);
|
|
|
|
int o;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
mutex_lock(&priv->pgdir_mutex);
|
|
|
|
|
|
|
|
o = db->order;
|
|
|
|
i = db->index;
|
|
|
|
|
|
|
|
if (db->order == 0 && test_bit(i ^ 1, db->u.pgdir->order0)) {
|
|
|
|
clear_bit(i ^ 1, db->u.pgdir->order0);
|
|
|
|
++o;
|
|
|
|
}
|
|
|
|
i >>= o;
|
|
|
|
set_bit(i, db->u.pgdir->bits[o]);
|
|
|
|
|
|
|
|
if (bitmap_full(db->u.pgdir->order1, MLX4_DB_PER_PAGE / 2)) {
|
|
|
|
dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
|
|
|
|
db->u.pgdir->db_page, db->u.pgdir->db_dma);
|
|
|
|
list_del(&db->u.pgdir->list);
|
|
|
|
kfree(db->u.pgdir);
|
|
|
|
}
|
|
|
|
|
|
|
|
mutex_unlock(&priv->pgdir_mutex);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(mlx4_db_free);
|
2008-04-26 05:27:08 +08:00
|
|
|
|
|
|
|
int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
|
|
|
|
int size, int max_direct)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
|
2014-05-11 20:15:12 +08:00
|
|
|
err = mlx4_db_alloc(dev, &wqres->db, 1, GFP_KERNEL);
|
2008-04-26 05:27:08 +08:00
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
*wqres->db.db = 0;
|
|
|
|
|
2014-05-11 20:15:12 +08:00
|
|
|
err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf, GFP_KERNEL);
|
2008-04-26 05:27:08 +08:00
|
|
|
if (err)
|
|
|
|
goto err_db;
|
|
|
|
|
|
|
|
err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift,
|
|
|
|
&wqres->mtt);
|
|
|
|
if (err)
|
|
|
|
goto err_buf;
|
|
|
|
|
2014-05-11 20:15:12 +08:00
|
|
|
err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf, GFP_KERNEL);
|
2008-04-26 05:27:08 +08:00
|
|
|
if (err)
|
|
|
|
goto err_mtt;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
err_mtt:
|
|
|
|
mlx4_mtt_cleanup(dev, &wqres->mtt);
|
|
|
|
err_buf:
|
|
|
|
mlx4_buf_free(dev, size, &wqres->buf);
|
|
|
|
err_db:
|
|
|
|
mlx4_db_free(dev, &wqres->db);
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res);
|
|
|
|
|
|
|
|
void mlx4_free_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
|
|
|
|
int size)
|
|
|
|
{
|
|
|
|
mlx4_mtt_cleanup(dev, &wqres->mtt);
|
|
|
|
mlx4_buf_free(dev, size, &wqres->buf);
|
|
|
|
mlx4_db_free(dev, &wqres->db);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(mlx4_free_hwq_res);
|