freebsd-skq/sys/kern/subr_busdma_bufalloc.c
jeff f375b4dd66 Implement NUMA support in uma(9) and malloc(9). Allocations from specific
domains can be done by the _domain() API variants.  UMA also supports a
first-touch policy via the NUMA zone flag.

The slab layer is now segregated by VM domains and is precise.  It handles
iteration for round-robin directly.  The per-cpu cache layer remains
a mix of domains according to where memory is allocated and freed.  Well
behaved clients can achieve perfect locality with no performance penalty.

The direct domain allocation functions have to visit the slab layer and
so require per-zone locks which come at some expense.

Reviewed by:	Attilio (a slightly older version)
Tested by:	pho
Sponsored by:	Netflix, Dell/EMC Isilon
2018-01-12 23:25:05 +00:00

177 lines
5.2 KiB
C

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2012 Ian Lepore
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
/*
* Buffer allocation support routines for bus_dmamem_alloc implementations.
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/busdma_bufalloc.h>
#include <sys/malloc.h>
#include <vm/vm.h>
#include <vm/vm_extern.h>
#include <vm/vm_kern.h>
#include <vm/uma.h>
/*
* We manage buffer zones up to a page in size. Buffers larger than a page can
* be managed by one of the kernel's page-oriented memory allocation routines as
* efficiently as what we can do here. Also, a page is the largest size for
* which we can g'tee contiguity when using uma, and contiguity is one of the
* requirements we have to fulfill.
*/
#define MIN_ZONE_BUFSIZE 32
#define MAX_ZONE_BUFSIZE PAGE_SIZE
/*
* The static array of 12 bufzones is big enough to handle all the zones for the
* smallest supported allocation size of 32 through the largest supported page
* size of 64K. If you up the biggest page size number, up the array size too.
* Basically the size of the array needs to be log2(maxsize)-log2(minsize)+1,
* but I don't know of an easy way to express that as a compile-time constant.
*/
#if PAGE_SIZE > 65536
#error Unsupported page size
#endif
struct busdma_bufalloc {
bus_size_t min_size;
size_t num_zones;
struct busdma_bufzone buf_zones[12];
};
busdma_bufalloc_t
busdma_bufalloc_create(const char *name, bus_size_t minimum_alignment,
uma_alloc alloc_func, uma_free free_func, u_int32_t zcreate_flags)
{
struct busdma_bufalloc *ba;
struct busdma_bufzone *bz;
int i;
bus_size_t cursize;
ba = malloc(sizeof(struct busdma_bufalloc), M_DEVBUF,
M_ZERO | M_WAITOK);
ba->min_size = MAX(MIN_ZONE_BUFSIZE, minimum_alignment);
/*
* Each uma zone is created with an alignment of size-1, meaning that
* the alignment is equal to the size (I.E., 64 byte buffers are aligned
* to 64 byte boundaries, etc). This allows for a fast efficient test
* when deciding whether a pool buffer meets the constraints of a given
* tag used for allocation: the buffer is usable if tag->alignment <=
* bufzone->size.
*/
for (i = 0, bz = ba->buf_zones, cursize = ba->min_size;
i < nitems(ba->buf_zones) && cursize <= MAX_ZONE_BUFSIZE;
++i, ++bz, cursize <<= 1) {
snprintf(bz->name, sizeof(bz->name), "dma %.10s %ju",
name, (uintmax_t)cursize);
bz->size = cursize;
bz->umazone = uma_zcreate(bz->name, bz->size,
NULL, NULL, NULL, NULL, bz->size - 1, zcreate_flags);
if (bz->umazone == NULL) {
busdma_bufalloc_destroy(ba);
return (NULL);
}
if (alloc_func != NULL)
uma_zone_set_allocf(bz->umazone, alloc_func);
if (free_func != NULL)
uma_zone_set_freef(bz->umazone, free_func);
++ba->num_zones;
}
return (ba);
}
void
busdma_bufalloc_destroy(busdma_bufalloc_t ba)
{
struct busdma_bufzone *bz;
int i;
if (ba == NULL)
return;
for (i = 0, bz = ba->buf_zones; i < ba->num_zones; ++i, ++bz) {
uma_zdestroy(bz->umazone);
}
free(ba, M_DEVBUF);
}
struct busdma_bufzone *
busdma_bufalloc_findzone(busdma_bufalloc_t ba, bus_size_t size)
{
struct busdma_bufzone *bz;
int i;
if (size > MAX_ZONE_BUFSIZE)
return (NULL);
for (i = 0, bz = ba->buf_zones; i < ba->num_zones; ++i, ++bz) {
if (bz->size >= size)
return (bz);
}
panic("Didn't find a buffer zone of the right size");
}
void *
busdma_bufalloc_alloc_uncacheable(uma_zone_t zone, vm_size_t size, int domain,
uint8_t *pflag, int wait)
{
#ifdef VM_MEMATTR_UNCACHEABLE
/* Inform UMA that this allocator uses kernel_arena/object. */
*pflag = UMA_SLAB_KERNEL;
return ((void *)kmem_alloc_attr_domain(domain, size, wait, 0,
BUS_SPACE_MAXADDR, VM_MEMATTR_UNCACHEABLE));
#else
panic("VM_MEMATTR_UNCACHEABLE unavailable");
#endif /* VM_MEMATTR_UNCACHEABLE */
}
void
busdma_bufalloc_free_uncacheable(void *item, vm_size_t size, uint8_t pflag)
{
kmem_free(kernel_arena, (vm_offset_t)item, size);
}