freebsd-dev/sys/vm/uma_int.h

439 lines
15 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2002-2005, 2009 Jeffrey Roberson <jeff@FreeBSD.org>
* Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice unmodified, this list of conditions, and the following
* disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR 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.
*
* $FreeBSD$
*
*/
/*
* This file includes definitions, structures, prototypes, and inlines that
* should not be used outside of the actual implementation of UMA.
*/
/*
* Here's a quick description of the relationship between the objects:
*
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
* Kegs contain lists of slabs which are stored in either the full bin, empty
* bin, or partially allocated bin, to reduce fragmentation. They also contain
* the user supplied value for size, which is adjusted for alignment purposes
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
* and rsize is the result of that. The Keg also stores information for
* managing a hash of page addresses that maps pages to uma_slab_t structures
* for pages that don't have embedded uma_slab_t's.
*
* The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may
* be allocated off the page from a special slab zone. The free list within a
* slab is managed with a linked list of indexes, which are 8 bit values. If
* UMA_SLAB_SIZE is defined to be too large I will have to switch to 16bit
* values. Currently on alpha you can get 250 or so 32 byte items and on x86
* you can get 250 or so 16byte items. For item sizes that would yield more
* than 10% memory waste we potentially allocate a separate uma_slab_t if this
* will improve the number of items per slab that will fit.
*
* Other potential space optimizations are storing the 8bit of linkage in space
* wasted between items due to alignment problems. This may yield a much better
* memory footprint for certain sizes of objects. Another alternative is to
* increase the UMA_SLAB_SIZE, or allow for dynamic slab sizes. I prefer
* dynamic slab sizes because we could stick with 8 bit indexes and only use
* large slab sizes for zones with a lot of waste per slab. This may create
* ineffeciencies in the vm subsystem due to fragmentation in the address space.
*
* The only really gross cases, with regards to memory waste, are for those
* items that are just over half the page size. You can get nearly 50% waste,
* so you fall back to the memory footprint of the power of two allocator. I
* have looked at memory allocation sizes on many of the machines available to
* me, and there does not seem to be an abundance of allocations at this range
* so at this time it may not make sense to optimize for it. This can, of
* course, be solved with dynamic slab sizes.
*
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
* Kegs may serve multiple Zones but by far most of the time they only serve
* one. When a Zone is created, a Keg is allocated and setup for it. While
* the backing Keg stores slabs, the Zone caches Buckets of items allocated
* from the slabs. Each Zone is equipped with an init/fini and ctor/dtor
* pair, as well as with its own set of small per-CPU caches, layered above
* the Zone's general Bucket cache.
*
* The PCPU caches are protected by critical sections, and may be accessed
* safely only from their associated CPU, while the Zones backed by the same
* Keg all share a common Keg lock (to coalesce contention on the backing
* slabs). The backing Keg typically only serves one Zone but in the case of
* multiple Zones, one of the Zones is considered the Master Zone and all
* Zone-related stats from the Keg are done in the Master Zone. For an
* example of a Multi-Zone setup, refer to the Mbuf allocation code.
*/
/*
* This is the representation for normal (Non OFFPAGE slab)
*
* i == item
* s == slab pointer
*
* <---------------- Page (UMA_SLAB_SIZE) ------------------>
* ___________________________________________________________
* | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___________ |
* ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header||
* ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________||
* |___________________________________________________________|
*
*
* This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE.
*
* ___________________________________________________________
* | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ |
* ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |
* ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |
* |___________________________________________________________|
* ___________ ^
* |slab header| |
* |___________|---*
*
*/
#ifndef VM_UMA_INT_H
#define VM_UMA_INT_H
#define UMA_SLAB_SIZE PAGE_SIZE /* How big are our slabs? */
#define UMA_SLAB_MASK (PAGE_SIZE - 1) /* Mask to get back to the page */
#define UMA_SLAB_SHIFT PAGE_SHIFT /* Number of bits PAGE_MASK */
#define UMA_BOOT_PAGES 48 /* Pages allocated for startup */
/* Max waste before going to off page slab management */
#define UMA_MAX_WASTE (UMA_SLAB_SIZE / 10)
/*
* I doubt there will be many cases where this is exceeded. This is the initial
* size of the hash table for uma_slabs that are managed off page. This hash
* does expand by powers of two. Currently it doesn't get smaller.
*/
#define UMA_HASH_SIZE_INIT 32
/*
* I should investigate other hashing algorithms. This should yield a low
* number of collisions if the pages are relatively contiguous.
*
* This is the same algorithm that most processor caches use.
*
* I'm shifting and masking instead of % because it should be faster.
*/
#define UMA_HASH(h, s) ((((unsigned long)s) >> UMA_SLAB_SHIFT) & \
(h)->uh_hashmask)
#define UMA_HASH_INSERT(h, s, mem) \
SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \
(mem))], (s), us_hlink)
#define UMA_HASH_REMOVE(h, s, mem) \
SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h), \
(mem))], (s), uma_slab, us_hlink)
/* Hash table for freed address -> slab translation */
SLIST_HEAD(slabhead, uma_slab);
struct uma_hash {
struct slabhead *uh_slab_hash; /* Hash table for slabs */
int uh_hashsize; /* Current size of the hash table */
int uh_hashmask; /* Mask used during hashing */
};
/*
* Structures for per cpu queues.
*/
struct uma_bucket {
LIST_ENTRY(uma_bucket) ub_link; /* Link into the zone */
int16_t ub_cnt; /* Count of free items. */
int16_t ub_entries; /* Max items. */
void *ub_bucket[]; /* actual allocation storage */
};
typedef struct uma_bucket * uma_bucket_t;
struct uma_cache {
uma_bucket_t uc_freebucket; /* Bucket we're freeing to */
uma_bucket_t uc_allocbucket; /* Bucket to allocate from */
u_int64_t uc_allocs; /* Count of allocations */
u_int64_t uc_frees; /* Count of frees */
};
typedef struct uma_cache * uma_cache_t;
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 management structure
*
* TODO: Optimize for cache line size
*
*/
struct uma_keg {
LIST_ENTRY(uma_keg) uk_link; /* List of all kegs */
struct mtx uk_lock; /* Lock for the keg */
struct uma_hash uk_hash;
char *uk_name; /* Name of creating 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
LIST_HEAD(,uma_zone) uk_zones; /* Keg's zones */
LIST_HEAD(,uma_slab) uk_part_slab; /* partially allocated slabs */
LIST_HEAD(,uma_slab) uk_free_slab; /* empty slab list */
LIST_HEAD(,uma_slab) uk_full_slab; /* full slabs */
u_int32_t uk_recurse; /* Allocation recursion count */
u_int32_t uk_align; /* Alignment mask */
u_int32_t uk_pages; /* Total page count */
u_int32_t uk_free; /* Count of items free in slabs */
u_int32_t uk_size; /* Requested size of each item */
u_int32_t uk_rsize; /* Real size of each item */
u_int32_t uk_maxpages; /* Maximum number of pages to alloc */
uma_init uk_init; /* Keg's init routine */
uma_fini uk_fini; /* Keg's fini routine */
uma_alloc uk_allocf; /* Allocation function */
uma_free uk_freef; /* Free routine */
struct vm_object *uk_obj; /* Zone specific object */
vm_offset_t uk_kva; /* Base kva for zones with objs */
uma_zone_t uk_slabzone; /* Slab zone backing us, if OFFPAGE */
u_int16_t uk_pgoff; /* Offset to uma_slab struct */
u_int16_t uk_ppera; /* pages per allocation from backend */
u_int16_t uk_ipers; /* Items per slab */
u_int32_t uk_flags; /* Internal flags */
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
};
typedef struct uma_keg * uma_keg_t;
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
/* Page management structure */
/* Sorry for the union, but space efficiency is important */
struct uma_slab_head {
uma_keg_t us_keg; /* Keg we live in */
union {
LIST_ENTRY(uma_slab) _us_link; /* slabs in zone */
unsigned long _us_size; /* Size of allocation */
} us_type;
SLIST_ENTRY(uma_slab) us_hlink; /* Link for hash table */
u_int8_t *us_data; /* First item */
u_int8_t us_flags; /* Page flags see uma.h */
u_int8_t us_freecount; /* How many are free? */
u_int8_t us_firstfree; /* First free item index */
};
/* The standard slab structure */
struct uma_slab {
struct uma_slab_head us_head; /* slab header data */
struct {
u_int8_t us_item;
} us_freelist[1]; /* actual number bigger */
};
/*
* The slab structure for UMA_ZONE_REFCNT zones for whose items we
* maintain reference counters in the slab for.
*/
struct uma_slab_refcnt {
struct uma_slab_head us_head; /* slab header data */
struct {
u_int8_t us_item;
u_int32_t us_refcnt;
} us_freelist[1]; /* actual number bigger */
};
#define us_keg us_head.us_keg
#define us_link us_head.us_type._us_link
#define us_size us_head.us_type._us_size
#define us_hlink us_head.us_hlink
#define us_data us_head.us_data
#define us_flags us_head.us_flags
#define us_freecount us_head.us_freecount
#define us_firstfree us_head.us_firstfree
typedef struct uma_slab * uma_slab_t;
typedef struct uma_slab_refcnt * uma_slabrefcnt_t;
typedef uma_slab_t (*uma_slaballoc)(uma_zone_t, uma_keg_t, int);
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
Rework the way slab header storage space is calculated in UMA. - zone_large_init() stays pretty much the same. - zone_small_init() will try to stash the slab header in the slab page being allocated if the amount of calculated wasted space is less than UMA_MAX_WASTE (for both the UMA_ZONE_REFCNT case and regular case). If the amount of wasted space is >= UMA_MAX_WASTE, then UMA_ZONE_OFFPAGE will be set and the slab header will be allocated separately for better use of space. - uma_startup() calculates the maximum ipers required in offpage slabs (so that the offpage slab header zone(s) can be sized accordingly). The algorithm used to calculate this replaces the old calculation (which only happened to work coincidentally). We now iterate over possible object sizes, starting from the smallest one, until we determine that wastedspace calculated in zone_small_init() might end up being greater than UMA_MAX_WASTE, at which point we use the found object size to compute the maximum possible ipers. The reason this works is because: - wastedspace versus objectsize is a see-saw function with local minima all equal to zero and local maxima growing directly proportioned to objectsize. This implies that for objects up to or equal a certain objectsize, the see-saw remains entirely below UMA_MAX_WASTE, so for those objectsizes it is impossible to ever go OFFPAGE for slab headers. - ipers (items-per-slab) versus objectsize is an inversely proportional function which falls off very quickly (very large for small objectsizes). - To determine the maximum ipers we'll ever need from OFFPAGE slab headers we first find the largest objectsize for which we are guaranteed to not go offpage for and use it to compute ipers (as though we were offpage). Since the only objectsizes allowed to go offpage are bigger than the found objectsize, and since ipers vs objectsize is inversely proportional (and monotonically decreasing), then we are guaranteed that the ipers computed is always >= what we will ever need in offpage slab headers. - Define UMA_FRITM_SZ and UMA_FRITMREF_SZ to be the actual (possibly padded) size of each freelist index so that offset calculations are fixed. This might fix weird data corruption problems and certainly allows ARM to now boot to at least single-user (via simulator). Tested on i386 UP by me. Tested on sparc64 SMP by fenner. Tested on ARM simulator to single-user by cognet.
2004-07-29 15:25:40 +00:00
/*
* These give us the size of one free item reference within our corresponding
* uma_slab structures, so that our calculations during zone setup are correct
* regardless of what the compiler decides to do with padding the structure
* arrays within uma_slab.
*/
#define UMA_FRITM_SZ (sizeof(struct uma_slab) - sizeof(struct uma_slab_head))
#define UMA_FRITMREF_SZ (sizeof(struct uma_slab_refcnt) - \
sizeof(struct uma_slab_head))
struct uma_klink {
LIST_ENTRY(uma_klink) kl_link;
uma_keg_t kl_keg;
};
typedef struct uma_klink *uma_klink_t;
/*
* Zone management structure
*
* TODO: Optimize for cache line size
*
*/
struct uma_zone {
char *uz_name; /* Text name of the 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
struct mtx *uz_lock; /* Lock for the zone (keg's lock) */
LIST_ENTRY(uma_zone) uz_link; /* List of all zones in keg */
LIST_HEAD(,uma_bucket) uz_full_bucket; /* full buckets */
LIST_HEAD(,uma_bucket) uz_free_bucket; /* Buckets for frees */
LIST_HEAD(,uma_klink) uz_kegs; /* List of kegs. */
struct uma_klink uz_klink; /* klink for first keg. */
uma_slaballoc uz_slab; /* Allocate a slab from the backend. */
uma_ctor uz_ctor; /* Constructor for each allocation */
uma_dtor uz_dtor; /* Destructor */
uma_init uz_init; /* Initializer for each item */
uma_fini uz_fini; /* Discards memory */
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
u_int64_t uz_allocs; /* Total number of allocations */
Introduce a new sysctl, vm.zone_stats, which exports UMA(9) allocator statistics via a binary structure stream: - Add structure 'uma_stream_header', which defines a stream version, definition of MAXCPUs used in the stream, and the number of zone records in the stream. - Add structure 'uma_type_header', which defines the name, alignment, size, resource allocation limits, current pages allocated, preferred bucket size, and central zone + keg statistics. - Add structure 'uma_percpu_stat', which, for each per-CPU cache, includes the number of allocations and frees, as well as the number of free items in the cache. - 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 uma_percpu_stat structures holding per-CPU allocation information. Typical values of MAXCPU 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, also export the number of UMA zones as a sysctl vm.uma_count, in order to assist in sizing user swpace buffers to receive the stream. A follow-up commit of libmemstat(3), a library to monitor kernel memory allocation, will occur in the next few days. This change directly supports converting netstat(1)'s "-mb" mode to using UMA-sourced stats rather than separately maintained mbuf allocator statistics. MFC after: 1 week
2005-07-14 16:35:13 +00:00
u_int64_t uz_frees; /* Total number of frees */
u_int64_t uz_fails; /* Total number of alloc failures */
u_int32_t uz_flags; /* Flags inherited from kegs */
u_int32_t uz_size; /* Size inherited from kegs */
uint16_t uz_fills; /* Outstanding bucket fills */
uint16_t uz_count; /* Highest value ub_ptr can have */
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
/*
* This HAS to be the last item because we adjust the zone size
* based on NCPU and then allocate the space for the zones.
*/
struct uma_cache uz_cpu[1]; /* Per cpu caches */
};
/*
* These flags must not overlap with the UMA_ZONE flags specified in uma.h.
*/
#define UMA_ZFLAG_BUCKET 0x02000000 /* Bucket zone. */
#define UMA_ZFLAG_MULTI 0x04000000 /* Multiple kegs in the zone. */
#define UMA_ZFLAG_DRAINING 0x08000000 /* Running zone_drain. */
#define UMA_ZFLAG_PRIVALLOC 0x10000000 /* Use uz_allocf. */
#define UMA_ZFLAG_INTERNAL 0x20000000 /* No offpage no PCPU. */
#define UMA_ZFLAG_FULL 0x40000000 /* Reached uz_maxpages */
#define UMA_ZFLAG_CACHEONLY 0x80000000 /* Don't ask VM for buckets. */
#define UMA_ZFLAG_INHERIT (UMA_ZFLAG_INTERNAL | UMA_ZFLAG_CACHEONLY | \
UMA_ZFLAG_BUCKET)
#ifdef _KERNEL
/* Internal prototypes */
static __inline uma_slab_t hash_sfind(struct uma_hash *hash, u_int8_t *data);
void *uma_large_malloc(int size, int wait);
void uma_large_free(uma_slab_t slab);
/* Lock Macros */
#define KEG_LOCK_INIT(k, lc) \
do { \
if ((lc)) \
mtx_init(&(k)->uk_lock, (k)->uk_name, \
(k)->uk_name, MTX_DEF | MTX_DUPOK); \
else \
mtx_init(&(k)->uk_lock, (k)->uk_name, \
"UMA zone", MTX_DEF | MTX_DUPOK); \
} while (0)
#define KEG_LOCK_FINI(k) mtx_destroy(&(k)->uk_lock)
#define KEG_LOCK(k) mtx_lock(&(k)->uk_lock)
#define KEG_UNLOCK(k) mtx_unlock(&(k)->uk_lock)
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
#define ZONE_LOCK(z) mtx_lock((z)->uz_lock)
#define ZONE_UNLOCK(z) mtx_unlock((z)->uz_lock)
/*
* Find a slab within a hash table. This is used for OFFPAGE zones to lookup
* the slab structure.
*
* Arguments:
* hash The hash table to search.
* data The base page of the item.
*
* Returns:
* A pointer to a slab if successful, else NULL.
*/
static __inline uma_slab_t
hash_sfind(struct uma_hash *hash, u_int8_t *data)
{
uma_slab_t slab;
int hval;
hval = UMA_HASH(hash, data);
SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) {
if ((u_int8_t *)slab->us_data == data)
return (slab);
}
return (NULL);
}
static __inline uma_slab_t
vtoslab(vm_offset_t va)
{
vm_page_t p;
uma_slab_t slab;
p = PHYS_TO_VM_PAGE(pmap_kextract(va));
slab = (uma_slab_t )p->object;
if (p->flags & PG_SLAB)
return (slab);
else
return (NULL);
}
static __inline void
vsetslab(vm_offset_t va, uma_slab_t slab)
{
vm_page_t p;
2004-11-26 15:04:26 +00:00
p = PHYS_TO_VM_PAGE(pmap_kextract(va));
p->object = (vm_object_t)slab;
p->flags |= PG_SLAB;
}
static __inline void
vsetobj(vm_offset_t va, vm_object_t obj)
{
vm_page_t p;
2004-11-26 15:04:26 +00:00
p = PHYS_TO_VM_PAGE(pmap_kextract(va));
p->object = obj;
p->flags &= ~PG_SLAB;
}
/*
* The following two functions may be defined by architecture specific code
* if they can provide more effecient allocation functions. This is useful
* for using direct mapped addresses.
*/
void *uma_small_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait);
void uma_small_free(void *mem, int size, u_int8_t flags);
#endif /* _KERNEL */
#endif /* VM_UMA_INT_H */