freebsd-dev/sys/kern/kern_malloc.c

891 lines
23 KiB
C
Raw Normal View History

/*-
1994-05-24 10:09:53 +00:00
* Copyright (c) 1987, 1991, 1993
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
* The Regents of the University of California.
* Copyright (c) 2005-2006 Robert N. M. Watson
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
* All rights reserved.
1994-05-24 10:09:53 +00:00
*
* 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.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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.
*
* @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94
*/
/*
* Kernel malloc(9) implementation -- general purpose kernel memory allocator
* based on memory types. Back end is implemented using the UMA(9) zone
* allocator. A set of fixed-size buckets are used for smaller allocations,
* and a special UMA allocation interface is used for larger allocations.
* Callers declare memory types, and statistics are maintained independently
* for each memory type. Statistics are maintained per-CPU for performance
* reasons. See malloc(9) and comments in malloc.h for a detailed
* description.
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_ddb.h"
#include "opt_vm.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kdb.h>
1994-05-24 10:09:53 +00:00
#include <sys/kernel.h>
#include <sys/lock.h>
1994-05-24 10:09:53 +00:00
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/mutex.h>
#include <sys/vmmeter.h>
#include <sys/proc.h>
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
#include <sys/sbuf.h>
#include <sys/sysctl.h>
#include <sys/time.h>
1994-05-24 10:09:53 +00:00
#include <vm/vm.h>
#include <vm/pmap.h>
#include <vm/vm_param.h>
1994-05-24 10:09:53 +00:00
#include <vm/vm_kern.h>
#include <vm/vm_extern.h>
#include <vm/vm_map.h>
#include <vm/vm_page.h>
#include <vm/uma.h>
#include <vm/uma_int.h>
#include <vm/uma_dbg.h>
1994-05-24 10:09:53 +00:00
#ifdef DEBUG_MEMGUARD
#include <vm/memguard.h>
#endif
#ifdef DEBUG_REDZONE
#include <vm/redzone.h>
#endif
#if defined(INVARIANTS) && defined(__i386__)
#include <machine/cpu.h>
#endif
#include <ddb/ddb.h>
/*
* When realloc() is called, if the new size is sufficiently smaller than
* the old size, realloc() will allocate a new, smaller block to avoid
* wasting memory. 'Sufficiently smaller' is defined as: newsize <=
* oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
*/
#ifndef REALLOC_FRACTION
#define REALLOC_FRACTION 1 /* new block if <= half the size */
#endif
/*
* Centrally define some common malloc types.
*/
MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
2002-03-19 21:25:46 +00:00
static void kmeminit(void *);
SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL)
static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
static struct malloc_type *kmemstatistics;
static vm_offset_t kmembase;
static vm_offset_t kmemlimit;
Introduce a new sysctl, kern.malloc_stats, which exports kernel malloc statistics via a binary structure stream: - Add structure 'malloc_type_stream_header', which defines a stream version, definition of MAXCPUS used in the stream, and a number of malloc_type records in the stream. - Add structure 'malloc_type_header', which defines the name of the malloc type being reported on. - When the sysctl is queried, return a stream header, followed by a series of type descriptions, each consisting of a type header followed by a series of MAXCPUS malloc_type_stats structures holding per-CPU allocation information. Typical values of MAXCPUS will be 1 (UP compiled kernel) and 16 (SMP compiled kernel). This query mechanism allows user space monitoring tools to extract memory allocation statistics in a machine-readable form, and to do so at a per-CPU granularity, allowing monitoring of allocation patterns across CPUs in order to better understand the distribution of work and memory flow over multiple CPUs. While here: - Bump statistics width to uint64_t, and hard code using fixed-width type in order to be more sure about structure layout in the stream. We allocate and free a lot of memory. - Add kmemcount, a counter of the number of registered malloc types, in order to avoid excessive manual counting of types. Export via a new sysctl to allow user-space code to better size buffers. - De-XXX comment on no longer maintaining the high watermark in old sysctl monitoring code. A follow-up commit of libmemstat(3), a library to monitor kernel memory allocation, will occur in the next few days. Likewise, similar changes to UMA.
2005-07-14 11:52:06 +00:00
static int kmemcount;
#define KMEM_ZSHIFT 4
#define KMEM_ZBASE 16
#define KMEM_ZMASK (KMEM_ZBASE - 1)
#define KMEM_ZMAX PAGE_SIZE
#define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT)
static u_int8_t kmemsize[KMEM_ZSIZE + 1];
/*
* Small malloc(9) memory allocations are allocated from a set of UMA buckets
* of various sizes.
*
* XXX: The comment here used to read "These won't be powers of two for
* long." It's possible that a significant amount of wasted memory could be
* recovered by tuning the sizes of these buckets.
*/
struct {
int kz_size;
char *kz_name;
uma_zone_t kz_zone;
} kmemzones[] = {
{16, "16", NULL},
{32, "32", NULL},
{64, "64", NULL},
{128, "128", NULL},
{256, "256", NULL},
{512, "512", NULL},
{1024, "1024", NULL},
{2048, "2048", NULL},
{4096, "4096", NULL},
#if PAGE_SIZE > 4096
{8192, "8192", NULL},
#if PAGE_SIZE > 8192
{16384, "16384", NULL},
#if PAGE_SIZE > 16384
{32768, "32768", NULL},
#if PAGE_SIZE > 32768
{65536, "65536", NULL},
#if PAGE_SIZE > 65536
#error "Unsupported PAGE_SIZE"
#endif /* 65536 */
#endif /* 32768 */
#endif /* 16384 */
#endif /* 8192 */
#endif /* 4096 */
{0, NULL},
1994-05-24 10:09:53 +00:00
};
/*
* Zone to allocate malloc type descriptions from. For ABI reasons, memory
* types are described by a data structure passed by the declaring code, but
* the malloc(9) implementation has its own data structure describing the
* type and statistics. This permits the malloc(9)-internal data structures
* to be modified without breaking binary-compiled kernel modules that
* declare malloc types.
*/
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
static uma_zone_t mt_zone;
u_int vm_kmem_size;
SYSCTL_UINT(_vm, OID_AUTO, kmem_size, CTLFLAG_RD, &vm_kmem_size, 0,
"Size of kernel memory");
u_int vm_kmem_size_min;
SYSCTL_UINT(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RD, &vm_kmem_size_min, 0,
"Minimum size of kernel memory");
u_int vm_kmem_size_max;
SYSCTL_UINT(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RD, &vm_kmem_size_max, 0,
"Maximum size of kernel memory");
u_int vm_kmem_size_scale;
SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RD, &vm_kmem_size_scale, 0,
"Scale factor for kernel memory size");
/*
* The malloc_mtx protects the kmemstatistics linked list.
*/
struct mtx malloc_mtx;
1994-05-24 10:09:53 +00:00
#ifdef MALLOC_PROFILE
uint64_t krequests[KMEM_ZSIZE + 1];
static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
#endif
Introduce a new sysctl, kern.malloc_stats, which exports kernel malloc statistics via a binary structure stream: - Add structure 'malloc_type_stream_header', which defines a stream version, definition of MAXCPUS used in the stream, and a number of malloc_type records in the stream. - Add structure 'malloc_type_header', which defines the name of the malloc type being reported on. - When the sysctl is queried, return a stream header, followed by a series of type descriptions, each consisting of a type header followed by a series of MAXCPUS malloc_type_stats structures holding per-CPU allocation information. Typical values of MAXCPUS will be 1 (UP compiled kernel) and 16 (SMP compiled kernel). This query mechanism allows user space monitoring tools to extract memory allocation statistics in a machine-readable form, and to do so at a per-CPU granularity, allowing monitoring of allocation patterns across CPUs in order to better understand the distribution of work and memory flow over multiple CPUs. While here: - Bump statistics width to uint64_t, and hard code using fixed-width type in order to be more sure about structure layout in the stream. We allocate and free a lot of memory. - Add kmemcount, a counter of the number of registered malloc types, in order to avoid excessive manual counting of types. Export via a new sysctl to allow user-space code to better size buffers. - De-XXX comment on no longer maintaining the high watermark in old sysctl monitoring code. A follow-up commit of libmemstat(3), a library to monitor kernel memory allocation, will occur in the next few days. Likewise, similar changes to UMA.
2005-07-14 11:52:06 +00:00
static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
1994-05-24 10:09:53 +00:00
/*
* time_uptime of the last malloc(9) failure (induced or real).
*/
static time_t t_malloc_fail;
/*
* malloc(9) fault injection -- cause malloc failures every (n) mallocs when
* the caller specifies M_NOWAIT. If set to 0, no failures are caused.
*/
#ifdef MALLOC_MAKE_FAILURES
SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
"Kernel malloc debugging options");
static int malloc_failure_rate;
static int malloc_nowait_count;
static int malloc_failure_count;
SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
&malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
&malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
#endif
int
malloc_last_fail(void)
{
return (time_uptime - t_malloc_fail);
}
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
/*
* An allocation has succeeded -- update malloc type statistics for the
* amount of bucket size. Occurs within a critical section so that the
* thread isn't preempted and doesn't migrate while updating per-PCU
* statistics.
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
*/
static void
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
int zindx)
{
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
struct malloc_type_internal *mtip;
struct malloc_type_stats *mtsp;
critical_enter();
mtip = mtp->ks_handle;
mtsp = &mtip->mti_stats[curcpu];
if (size > 0) {
mtsp->mts_memalloced += size;
mtsp->mts_numallocs++;
}
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
if (zindx != -1)
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
mtsp->mts_size |= 1 << zindx;
critical_exit();
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
}
void
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
{
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
if (size > 0)
malloc_type_zone_allocated(mtp, size, -1);
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
}
/*
* A free operation has occurred -- update malloc type statistics for the
* amount of the bucket size. Occurs within a critical section so that the
* thread isn't preempted and doesn't migrate while updating per-CPU
* statistics.
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
*/
void
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
malloc_type_freed(struct malloc_type *mtp, unsigned long size)
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
{
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
struct malloc_type_internal *mtip;
struct malloc_type_stats *mtsp;
critical_enter();
mtip = mtp->ks_handle;
mtsp = &mtip->mti_stats[curcpu];
mtsp->mts_memfreed += size;
mtsp->mts_numfrees++;
critical_exit();
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
}
1994-05-24 10:09:53 +00:00
/*
* malloc:
*
* Allocate a block of memory.
*
* If M_NOWAIT is set, this routine will not block and return NULL if
* the allocation fails.
1994-05-24 10:09:53 +00:00
*/
void *
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
malloc(unsigned long size, struct malloc_type *mtp, int flags)
1994-05-24 10:09:53 +00:00
{
int indx;
caddr_t va;
uma_zone_t zone;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
uma_keg_t keg;
#if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
unsigned long osize = size;
#endif
#ifdef INVARIANTS
/*
* Check that exactly one of M_WAITOK or M_NOWAIT is specified.
*/
indx = flags & (M_WAITOK | M_NOWAIT);
if (indx != M_NOWAIT && indx != M_WAITOK) {
static struct timeval lasterr;
static int curerr, once;
if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
printf("Bad malloc flags: %x\n", indx);
kdb_backtrace();
flags |= M_WAITOK;
once++;
}
}
#endif
#ifdef MALLOC_MAKE_FAILURES
if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
atomic_add_int(&malloc_nowait_count, 1);
if ((malloc_nowait_count % malloc_failure_rate) == 0) {
atomic_add_int(&malloc_failure_count, 1);
t_malloc_fail = time_uptime;
return (NULL);
}
}
#endif
if (flags & M_WAITOK)
KASSERT(curthread->td_intr_nesting_level == 0,
("malloc(M_WAITOK) in interrupt context"));
#ifdef DEBUG_MEMGUARD
if (memguard_cmp(mtp))
return memguard_alloc(size, flags);
#endif
#ifdef DEBUG_REDZONE
size = redzone_size_ntor(size);
#endif
if (size <= KMEM_ZMAX) {
if (size & KMEM_ZMASK)
size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
indx = kmemsize[size >> KMEM_ZSHIFT];
zone = kmemzones[indx].kz_zone;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
keg = zone->uz_keg;
#ifdef MALLOC_PROFILE
krequests[size >> KMEM_ZSHIFT]++;
#endif
va = uma_zalloc(zone, flags);
Reimplement contigmalloc(9) with an algorithm which stands a greatly- improved chance of working despite pressure from running programs. Instead of trying to throw a bunch of pages out to swap and hope for the best, only a range that can potentially fulfill contigmalloc(9)'s request will have its contents paged out (potentially, not forcibly) at a time. The new contigmalloc operation still operates in three passes, but it could potentially be tuned to more or less. The first pass only looks at pages in the cache and free pages, so they would be thrown out without having to block. If this is not enough, the subsequent passes page out any unwired memory. To combat memory pressure refragmenting the section of memory being laundered, each page is removed from the systems' free memory queue once it has been freed so that blocking later doesn't cause the memory laundered so far to get reallocated. The page-out operations are now blocking, as it would make little sense to try to push out a page, then get its status immediately afterward to remove it from the available free pages queue, if it's unlikely to have been freed. Another change is that if KVA allocation fails, the allocated memory segment will be freed and not leaked. There is a sysctl/tunable, defaulting to on, which causes the old contigmalloc() algorithm to be used. Nonetheless, I have been using vm.old_contigmalloc=0 for over a month. It is safe to switch at run-time to see the difference it makes. A new interface has been used which does not require mapping the allocated pages into KVA: vm_page.h functions vm_page_alloc_contig() and vm_page_release_contig(). These are what vm.old_contigmalloc=0 uses internally, so the sysctl/tunable does not affect their operation. When using the contigmalloc(9) and contigfree(9) interfaces, memory is now tracked with malloc(9) stats. Several functions have been exported from kern_malloc.c to allow other subsystems to use these statistics, as well. This invalidates the BUGS section of the contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
if (va != NULL)
size = keg->uk_size;
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
} else {
size = roundup(size, PAGE_SIZE);
zone = NULL;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
keg = NULL;
va = uma_large_malloc(size, flags);
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
malloc_type_allocated(mtp, va == NULL ? 0 : size);
1994-05-24 10:09:53 +00:00
}
if (flags & M_WAITOK)
KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
else if (va == NULL)
t_malloc_fail = time_uptime;
#ifdef DIAGNOSTIC
if (va != NULL && !(flags & M_ZERO)) {
memset(va, 0x70, osize);
}
#endif
#ifdef DEBUG_REDZONE
if (va != NULL)
va = redzone_setup(va, osize);
#endif
1994-05-24 10:09:53 +00:00
return ((void *) va);
}
/*
* free:
*
* Free a block of memory allocated by malloc.
*
* This routine may not block.
1994-05-24 10:09:53 +00:00
*/
void
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
free(void *addr, struct malloc_type *mtp)
1994-05-24 10:09:53 +00:00
{
uma_slab_t slab;
u_long size;
/* free(NULL, ...) does nothing */
if (addr == NULL)
return;
#ifdef DEBUG_MEMGUARD
if (memguard_cmp(mtp)) {
memguard_free(addr);
return;
}
#endif
#ifdef DEBUG_REDZONE
redzone_check(addr);
addr = redzone_addr_ntor(addr);
#endif
size = 0;
slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
if (slab == NULL)
panic("free: address %p(%p) has not been allocated.\n",
addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
#ifdef INVARIANTS
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
struct malloc_type **mtpp = addr;
#endif
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
size = slab->us_keg->uk_size;
#ifdef INVARIANTS
/*
* Cache a pointer to the malloc_type that most recently freed
* this memory here. This way we know who is most likely to
* have stepped on it later.
*
* This code assumes that size is a multiple of 8 bytes for
* 64 bit machines
*/
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
mtpp = (struct malloc_type **)
((unsigned long)mtpp & ~UMA_ALIGN_PTR);
mtpp += (size - sizeof(struct malloc_type *)) /
sizeof(struct malloc_type *);
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
*mtpp = mtp;
#endif
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
} else {
size = slab->us_size;
uma_large_free(slab);
}
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
malloc_type_freed(mtp, size);
1994-05-24 10:09:53 +00:00
}
/*
* realloc: change the size of a memory block
*/
void *
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
{
uma_slab_t slab;
unsigned long alloc;
void *newaddr;
/* realloc(NULL, ...) is equivalent to malloc(...) */
if (addr == NULL)
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
return (malloc(size, mtp, flags));
/*
* XXX: Should report free of old memory and alloc of new memory to
* per-CPU stats.
*/
#ifdef DEBUG_MEMGUARD
if (memguard_cmp(mtp)) {
slab = NULL;
alloc = size;
} else {
#endif
#ifdef DEBUG_REDZONE
slab = NULL;
alloc = redzone_get_size(addr);
#else
slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
/* Sanity check */
KASSERT(slab != NULL,
("realloc: address %p out of range", (void *)addr));
/* Get the size of the original block */
if (!(slab->us_flags & UMA_SLAB_MALLOC))
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
alloc = slab->us_keg->uk_size;
else
alloc = slab->us_size;
/* Reuse the original block if appropriate */
if (size <= alloc
&& (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
return (addr);
#endif /* !DEBUG_REDZONE */
#ifdef DEBUG_MEMGUARD
}
#endif
/* Allocate a new, bigger (or smaller) block */
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
if ((newaddr = malloc(size, mtp, flags)) == NULL)
return (NULL);
/* Copy over original contents */
bcopy(addr, newaddr, min(size, alloc));
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
free(addr, mtp);
return (newaddr);
}
/*
* reallocf: same as realloc() but free memory on failure.
*/
void *
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
{
void *mem;
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
if ((mem = realloc(addr, size, mtp, flags)) == NULL)
free(addr, mtp);
return (mem);
}
1994-05-24 10:09:53 +00:00
/*
* Initialize the kernel memory allocator
*/
/* ARGSUSED*/
static void
kmeminit(void *dummy)
1994-05-24 10:09:53 +00:00
{
u_int8_t indx;
u_long mem_size;
int i;
mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
/*
* Try to auto-tune the kernel memory size, so that it is
* more applicable for a wider range of machine sizes.
* On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while
* a VM_KMEM_SIZE of 12MB is a fair compromise. The
* VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
* available, and on an X86 with a total KVA space of 256MB,
* try to keep VM_KMEM_SIZE_MAX at 80MB or below.
*
* Note that the kmem_map is also used by the zone allocator,
* so make sure that there is enough space.
*/
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
mem_size = cnt.v_page_count;
#if defined(VM_KMEM_SIZE_SCALE)
vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
#endif
TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
if (vm_kmem_size_scale > 0 &&
(mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
#if defined(VM_KMEM_SIZE_MIN)
vm_kmem_size_min = VM_KMEM_SIZE_MIN;
#endif
TUNABLE_INT_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) {
vm_kmem_size = vm_kmem_size_min;
}
#if defined(VM_KMEM_SIZE_MAX)
vm_kmem_size_max = VM_KMEM_SIZE_MAX;
#endif
TUNABLE_INT_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
vm_kmem_size = vm_kmem_size_max;
/* Allow final override from the kernel environment */
#ifndef BURN_BRIDGES
if (TUNABLE_INT_FETCH("kern.vm.kmem.size", &vm_kmem_size) != 0)
printf("kern.vm.kmem.size is now called vm.kmem_size!\n");
#endif
TUNABLE_INT_FETCH("vm.kmem_size", &vm_kmem_size);
/*
* Limit kmem virtual size to twice the physical memory.
* This allows for kmem map sparseness, but limits the size
* to something sane. Be careful to not overflow the 32bit
* ints while doing the check.
*/
if (((vm_kmem_size / 2) / PAGE_SIZE) > cnt.v_page_count)
vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE;
/*
* Tune settings based on the kernel map's size at this time.
*/
init_param3(vm_kmem_size / PAGE_SIZE);
kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit,
vm_kmem_size);
kmem_map->system_map = 1;
#ifdef DEBUG_MEMGUARD
/*
* Initialize MemGuard if support compiled in. MemGuard is a
* replacement allocator used for detecting tamper-after-free
* scenarios as they occur. It is only used for debugging.
*/
vm_memguard_divisor = 10;
TUNABLE_INT_FETCH("vm.memguard.divisor", &vm_memguard_divisor);
/* Pick a conservative value if provided value sucks. */
if ((vm_memguard_divisor <= 0) ||
((vm_kmem_size / vm_memguard_divisor) == 0))
vm_memguard_divisor = 10;
memguard_init(kmem_map, vm_kmem_size / vm_memguard_divisor);
#endif
uma_startup2();
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
#ifdef INVARIANTS
mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
#else
NULL, NULL, NULL, NULL,
#endif
UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
int size = kmemzones[indx].kz_size;
char *name = kmemzones[indx].kz_name;
kmemzones[indx].kz_zone = uma_zcreate(name, size,
#ifdef INVARIANTS
mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
#else
NULL, NULL, NULL, NULL,
#endif
UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
for (;i <= size; i+= KMEM_ZBASE)
kmemsize[i >> KMEM_ZSHIFT] = indx;
1994-05-24 10:09:53 +00:00
}
}
void
malloc_init(void *data)
{
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
struct malloc_type_internal *mtip;
struct malloc_type *mtp;
KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
mtp = data;
mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
mtp->ks_handle = mtip;
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
mtx_lock(&malloc_mtx);
mtp->ks_next = kmemstatistics;
kmemstatistics = mtp;
Introduce a new sysctl, kern.malloc_stats, which exports kernel malloc statistics via a binary structure stream: - Add structure 'malloc_type_stream_header', which defines a stream version, definition of MAXCPUS used in the stream, and a number of malloc_type records in the stream. - Add structure 'malloc_type_header', which defines the name of the malloc type being reported on. - When the sysctl is queried, return a stream header, followed by a series of type descriptions, each consisting of a type header followed by a series of MAXCPUS malloc_type_stats structures holding per-CPU allocation information. Typical values of MAXCPUS will be 1 (UP compiled kernel) and 16 (SMP compiled kernel). This query mechanism allows user space monitoring tools to extract memory allocation statistics in a machine-readable form, and to do so at a per-CPU granularity, allowing monitoring of allocation patterns across CPUs in order to better understand the distribution of work and memory flow over multiple CPUs. While here: - Bump statistics width to uint64_t, and hard code using fixed-width type in order to be more sure about structure layout in the stream. We allocate and free a lot of memory. - Add kmemcount, a counter of the number of registered malloc types, in order to avoid excessive manual counting of types. Export via a new sysctl to allow user-space code to better size buffers. - De-XXX comment on no longer maintaining the high watermark in old sysctl monitoring code. A follow-up commit of libmemstat(3), a library to monitor kernel memory allocation, will occur in the next few days. Likewise, similar changes to UMA.
2005-07-14 11:52:06 +00:00
kmemcount++;
mtx_unlock(&malloc_mtx);
1994-05-24 10:09:53 +00:00
}
void
malloc_uninit(void *data)
{
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
struct malloc_type_internal *mtip;
struct malloc_type_stats *mtsp;
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
struct malloc_type *mtp, *temp;
uma_slab_t slab;
long temp_allocs, temp_bytes;
int i;
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
mtp = data;
KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
mtx_lock(&malloc_mtx);
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
mtip = mtp->ks_handle;
mtp->ks_handle = NULL;
if (mtp != kmemstatistics) {
for (temp = kmemstatistics; temp != NULL;
temp = temp->ks_next) {
if (temp->ks_next == mtp)
temp->ks_next = mtp->ks_next;
}
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
} else
kmemstatistics = mtp->ks_next;
Introduce a new sysctl, kern.malloc_stats, which exports kernel malloc statistics via a binary structure stream: - Add structure 'malloc_type_stream_header', which defines a stream version, definition of MAXCPUS used in the stream, and a number of malloc_type records in the stream. - Add structure 'malloc_type_header', which defines the name of the malloc type being reported on. - When the sysctl is queried, return a stream header, followed by a series of type descriptions, each consisting of a type header followed by a series of MAXCPUS malloc_type_stats structures holding per-CPU allocation information. Typical values of MAXCPUS will be 1 (UP compiled kernel) and 16 (SMP compiled kernel). This query mechanism allows user space monitoring tools to extract memory allocation statistics in a machine-readable form, and to do so at a per-CPU granularity, allowing monitoring of allocation patterns across CPUs in order to better understand the distribution of work and memory flow over multiple CPUs. While here: - Bump statistics width to uint64_t, and hard code using fixed-width type in order to be more sure about structure layout in the stream. We allocate and free a lot of memory. - Add kmemcount, a counter of the number of registered malloc types, in order to avoid excessive manual counting of types. Export via a new sysctl to allow user-space code to better size buffers. - De-XXX comment on no longer maintaining the high watermark in old sysctl monitoring code. A follow-up commit of libmemstat(3), a library to monitor kernel memory allocation, will occur in the next few days. Likewise, similar changes to UMA.
2005-07-14 11:52:06 +00:00
kmemcount--;
mtx_unlock(&malloc_mtx);
/*
* Look for memory leaks.
*/
temp_allocs = temp_bytes = 0;
for (i = 0; i < MAXCPU; i++) {
mtsp = &mtip->mti_stats[i];
temp_allocs += mtsp->mts_numallocs;
temp_allocs -= mtsp->mts_numfrees;
temp_bytes += mtsp->mts_memalloced;
temp_bytes -= mtsp->mts_memfreed;
}
if (temp_allocs > 0 || temp_bytes > 0) {
printf("Warning: memory type %s leaked memory on destroy "
"(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
temp_allocs, temp_bytes);
}
slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
uma_zfree_arg(mt_zone, mtip, slab);
}
struct malloc_type *
malloc_desc2type(const char *desc)
{
struct malloc_type *mtp;
mtx_assert(&malloc_mtx, MA_OWNED);
for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
if (strcmp(mtp->ks_shortdesc, desc) == 0)
return (mtp);
}
return (NULL);
}
Introduce a new sysctl, kern.malloc_stats, which exports kernel malloc statistics via a binary structure stream: - Add structure 'malloc_type_stream_header', which defines a stream version, definition of MAXCPUS used in the stream, and a number of malloc_type records in the stream. - Add structure 'malloc_type_header', which defines the name of the malloc type being reported on. - When the sysctl is queried, return a stream header, followed by a series of type descriptions, each consisting of a type header followed by a series of MAXCPUS malloc_type_stats structures holding per-CPU allocation information. Typical values of MAXCPUS will be 1 (UP compiled kernel) and 16 (SMP compiled kernel). This query mechanism allows user space monitoring tools to extract memory allocation statistics in a machine-readable form, and to do so at a per-CPU granularity, allowing monitoring of allocation patterns across CPUs in order to better understand the distribution of work and memory flow over multiple CPUs. While here: - Bump statistics width to uint64_t, and hard code using fixed-width type in order to be more sure about structure layout in the stream. We allocate and free a lot of memory. - Add kmemcount, a counter of the number of registered malloc types, in order to avoid excessive manual counting of types. Export via a new sysctl to allow user-space code to better size buffers. - De-XXX comment on no longer maintaining the high watermark in old sysctl monitoring code. A follow-up commit of libmemstat(3), a library to monitor kernel memory allocation, will occur in the next few days. Likewise, similar changes to UMA.
2005-07-14 11:52:06 +00:00
static int
sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
{
struct malloc_type_stream_header mtsh;
struct malloc_type_internal *mtip;
struct malloc_type_header mth;
struct malloc_type *mtp;
int buflen, count, error, i;
struct sbuf sbuf;
char *buffer;
mtx_lock(&malloc_mtx);
restart:
mtx_assert(&malloc_mtx, MA_OWNED);
count = kmemcount;
mtx_unlock(&malloc_mtx);
buflen = sizeof(mtsh) + count * (sizeof(mth) +
sizeof(struct malloc_type_stats) * MAXCPU) + 1;
buffer = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
mtx_lock(&malloc_mtx);
if (count < kmemcount) {
free(buffer, M_TEMP);
goto restart;
}
sbuf_new(&sbuf, buffer, buflen, SBUF_FIXEDLEN);
/*
* Insert stream header.
*/
bzero(&mtsh, sizeof(mtsh));
mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
mtsh.mtsh_maxcpus = MAXCPU;
mtsh.mtsh_count = kmemcount;
if (sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)) < 0) {
mtx_unlock(&malloc_mtx);
error = ENOMEM;
goto out;
}
/*
* Insert alternating sequence of type headers and type statistics.
*/
for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
mtip = (struct malloc_type_internal *)mtp->ks_handle;
/*
* Insert type header.
*/
bzero(&mth, sizeof(mth));
strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
if (sbuf_bcat(&sbuf, &mth, sizeof(mth)) < 0) {
mtx_unlock(&malloc_mtx);
error = ENOMEM;
goto out;
}
/*
* Insert type statistics for each CPU.
*/
for (i = 0; i < MAXCPU; i++) {
if (sbuf_bcat(&sbuf, &mtip->mti_stats[i],
sizeof(mtip->mti_stats[i])) < 0) {
mtx_unlock(&malloc_mtx);
error = ENOMEM;
goto out;
}
}
}
mtx_unlock(&malloc_mtx);
sbuf_finish(&sbuf);
error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
out:
sbuf_delete(&sbuf);
free(buffer, M_TEMP);
return (error);
}
SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
"Return malloc types");
SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
"Count of kernel malloc types");
#ifdef DDB
DB_SHOW_COMMAND(malloc, db_show_malloc)
{
struct malloc_type_internal *mtip;
struct malloc_type *mtp;
u_int64_t allocs, frees;
u_int64_t alloced, freed;
int i;
db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse",
"Requests");
for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
mtip = (struct malloc_type_internal *)mtp->ks_handle;
allocs = 0;
frees = 0;
alloced = 0;
freed = 0;
for (i = 0; i < MAXCPU; i++) {
allocs += mtip->mti_stats[i].mts_numallocs;
frees += mtip->mti_stats[i].mts_numfrees;
alloced += mtip->mti_stats[i].mts_memalloced;
freed += mtip->mti_stats[i].mts_memfreed;
}
db_printf("%18s %12ju %12juK %12ju\n",
mtp->ks_shortdesc, allocs - frees,
(alloced - freed + 1023) / 1024, allocs);
}
}
#endif
#ifdef MALLOC_PROFILE
static int
sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
{
int linesize = 64;
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
struct sbuf sbuf;
uint64_t count;
uint64_t waste;
uint64_t mem;
int bufsize;
int error;
char *buf;
int rsize;
int size;
int i;
bufsize = linesize * (KMEM_ZSIZE + 1);
bufsize += 128; /* For the stats line */
bufsize += 128; /* For the banner line */
waste = 0;
mem = 0;
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
buf = malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
sbuf_new(&sbuf, buf, bufsize, SBUF_FIXEDLEN);
sbuf_printf(&sbuf,
"\n Size Requests Real Size\n");
for (i = 0; i < KMEM_ZSIZE; i++) {
size = i << KMEM_ZSHIFT;
rsize = kmemzones[kmemsize[i]].kz_size;
count = (long long unsigned)krequests[i];
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
(unsigned long long)count, rsize);
if ((rsize * count) > (size * count))
waste += (rsize * count) - (size * count);
mem += (rsize * count);
}
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
sbuf_printf(&sbuf,
"\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
(unsigned long long)mem, (unsigned long long)waste);
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
sbuf_finish(&sbuf);
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
Kernel malloc layers malloc_type allocation over one of two underlying allocators: a set of power-of-two UMA zones for small allocations, and the VM page allocator for large allocations. In order to maintain unified statistics for specific malloc types, kernel malloc maintains a separate per-type statistics pool, which can be monitored using vmstat -m. Prior to this commit, each pool of per-type statistics was protected using a per-type mutex associated with the malloc type. This change modifies kernel malloc to maintain per-CPU statistics pools for each malloc type, and protects writing those statistics using critical sections. It also moves to unsynchronized reads of per-CPU statistics when generating coalesced statistics. To do this, several changes are implemented: - In the previous world order, the statistics memory was allocated by the owner of the malloc type structure, allocated statically using MALLOC_DEFINE(). This embedded the definition of the malloc_type structure into all kernel modules. Move to a model in which a pointer within struct malloc_type points at a UMA-allocated malloc_type_internal data structure owned and maintained by kern_malloc.c, and not part of the exported ABI/API to the rest of the kernel. For the purposes of easing a possible MFC, re-use an existing pointer in 'struct malloc_type', and maintain the current malloc_type structure size, as well as layout with respect to the fields reused outside of the malloc subsystem (such as ks_shortdesc). There are several unused fields as a result of no longer requiring the mutex in malloc_type. - Struct malloc_type_internal contains an array of malloc_type_stats, of size MAXCPU. The structure defined above avoids hard-coding a kernel compile-time value of MAXCPU into kernel modules that interact with malloc. - When accessing per-cpu statistics for a malloc type, surround read - modify - update requests with critical_enter()/critical_exit() in order to avoid races during write. The per-CPU fields are written only from the CPU that owns them. - Per-CPU stats now maintained "allocated" and "freed" counters for number of allocations/frees and bytes allocated/freed, since there is no longer a coherent global notion of the totals. When coalescing malloc stats, accept a slight race between reading stats across CPUs, and avoid showing the user a negative allocation count for the type in the event of a race. The global high watermark is no longer maintained for a malloc type, as there is no global notion of the number of allocations. - While tearing up the sysctl() path, also switch to using sbufs. The current "export as text" sysctl format is retained with the same syntax. We may want to change this in the future to export more per-CPU information, such as how allocations and frees are balanced across CPUs. This change results in a substantial speedup of kernel malloc and free paths on SMP, as critical sections (where usable) out-perform mutexes due to avoiding atomic/bus-locked operations. There is also a minor improvement on UP due to the slightly lower cost of critical sections there. The cost of the change to this approach is the loss of a continuous notion of total allocations that can be exploited to track per-type high watermarks, as well as increased complexity when monitoring statistics. Due to carefully avoiding changing the ABI, as well as hardening the ABI against future changes, it is not necessary to recompile kernel modules for this change. However, MFC'ing this change to RELENG_5 will require also MFC'ing optimizations for soft critical sections, which may modify exposed kernel ABIs. The internal malloc API is changed, and modifications to vmstat in order to restore "vmstat -m" on core dumps will follow shortly. Several improvements from: bde Statistics approach discussed with: ups Tested by: scottl, others
2005-05-29 13:38:07 +00:00
sbuf_delete(&sbuf);
free(buf, M_TEMP);
return (error);
}
SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
#endif /* MALLOC_PROFILE */