summaryrefslogtreecommitdiffstats
path: root/lib/libstand/zalloc.c
diff options
context:
space:
mode:
authormsmith <msmith@FreeBSD.org>1998-09-26 01:42:40 +0000
committermsmith <msmith@FreeBSD.org>1998-09-26 01:42:40 +0000
commitea783268e73e47b0c3012339d1e278e7045e2305 (patch)
tree622472325886767e4a3905d55b5180ea188a7e99 /lib/libstand/zalloc.c
parent73ca5cb35b01711ebbbdbdc1d00d3f320d0a852c (diff)
downloadFreeBSD-src-ea783268e73e47b0c3012339d1e278e7045e2305.zip
FreeBSD-src-ea783268e73e47b0c3012339d1e278e7045e2305.tar.gz
Replace the old and extremely icky Mach/NetBSD allocator with a similarly
compact and much better one donated by Matt Dillon. Implement a simple sbrk() which uses the existing setheap() api. Remove the custom allocator from the UFS code. It wasn't working quite right, and it shouldn't be needed with the new allocator. Fix a serious problem with changing the value of already-existent environment variables. Don't attempt to modify the supposedly-const argument to putenv() Fix an off-by-one sizing error in the zipfs code detected by the new allocator. Submitted by: zmalloc from Matt Dillon <dillon@backplane.com>
Diffstat (limited to 'lib/libstand/zalloc.c')
-rw-r--r--lib/libstand/zalloc.c591
1 files changed, 591 insertions, 0 deletions
diff --git a/lib/libstand/zalloc.c b/lib/libstand/zalloc.c
new file mode 100644
index 0000000..b441702
--- /dev/null
+++ b/lib/libstand/zalloc.c
@@ -0,0 +1,591 @@
+/*
+ * This module derived from code donated to the FreeBSD Project by
+ * Matthew Dillon <dillon@backplane.com>
+ *
+ * Copyright (c) 1998 The FreeBSD Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * $Id$
+ */
+
+/*
+ * LIB/MEMORY/ZALLOC.C - self contained low-overhead memory pool/allocation
+ * subsystem
+ *
+ * This subsystem implements memory pools and memory allocation
+ * routines.
+ *
+ * Pools are managed via a linked list of 'free' areas. Allocating
+ * memory creates holes in the freelist, freeing memory fills them.
+ * Since the freelist consists only of free memory areas, it is possible
+ * to allocate the entire pool without incuring any structural overhead.
+ *
+ * The system works best when allocating similarly-sized chunks of
+ * memory. Care must be taken to avoid fragmentation when
+ * allocating/deallocating dissimilar chunks.
+ *
+ * When a memory pool is first allocated, the entire pool is marked as
+ * allocated. This is done mainly because we do not want to modify any
+ * portion of a pool's data area until we are given permission. The
+ * caller must explicitly deallocate portions of the pool to make them
+ * available.
+ *
+ * z[n]xalloc() works like z[n]alloc() but the allocation is made from
+ * within the specified address range. If the segment could not be
+ * allocated, NULL is returned. WARNING! The address range will be
+ * aligned to an 8 or 16 byte boundry depending on the cpu so if you
+ * give an unaligned address range, unexpected results may occur.
+ *
+ * If a standard allocation fails, the reclaim function will be called
+ * to recover some space. This usually causes other portions of the
+ * same pool to be released. Memory allocations at this low level
+ * should not block but you can do that too in your reclaim function
+ * if you want. Reclaim does not function when z[n]xalloc() is used,
+ * only for z[n]alloc().
+ *
+ * Allocation and frees of 0 bytes are valid operations.
+ */
+
+#include "zalloc_defs.h"
+
+Prototype struct MemPool *DummyStructMemPool;
+Library void *znalloc(struct MemPool *mpool, iaddr_t bytes);
+Library void *zalloc(struct MemPool *mpool, iaddr_t bytes);
+Library void *zallocAlign(struct MemPool *mpool, iaddr_t bytes, iaddr_t align);
+Library void *zxalloc(struct MemPool *mp, void *addr1, void *addr2, iaddr_t bytes);
+Library void *znxalloc(struct MemPool *mp, void *addr1, void *addr2, iaddr_t bytes);
+Library char *zallocStr(struct MemPool *mpool, const char *s, int slen);
+Library void zfree(struct MemPool *mpool, void *ptr, iaddr_t bytes);
+Library void zfreeStr(struct MemPool *mpool, char *s);
+Library void zinitPool(struct MemPool *mp, const char *id, void (*fpanic)(const char *ctl, ...), int (*freclaim)(struct MemPool *memPool, iaddr_t bytes), void *pBase, iaddr_t pSize);
+Library void zclearPool(struct MemPool *mp);
+Library void znop(const char *ctl, ...);
+Library int znot(struct MemPool *memPool, iaddr_t bytes);
+Library void zallocstats(struct MemPool *mp);
+
+/*
+ * znop() - panic function if none supplied.
+ */
+
+void
+znop(const char *ctl, ...)
+{
+}
+
+/*
+ * znot() - reclaim function if none supplied
+ */
+
+int
+znot(struct MemPool *memPool, iaddr_t bytes)
+{
+ return(-1);
+}
+
+#ifndef MALLOCLIB
+
+/*
+ * zalloc() - allocate and zero memory from pool. Call reclaim
+ * and retry if appropriate, return NULL if unable to allocate
+ * memory.
+ */
+
+void *
+zalloc(MemPool *mp, iaddr_t bytes)
+{
+ void *ptr;
+
+ if ((ptr = znalloc(mp, bytes)) != NULL)
+ bzero(ptr, bytes);
+ return(ptr);
+}
+
+/*
+ * zallocAlign() - allocate and zero memory from pool, enforce specified
+ * alignment (must be power of 2) on allocated memory.
+ */
+
+void *
+zallocAlign(struct MemPool *mp, iaddr_t bytes, iaddr_t align)
+{
+ void *ptr;
+
+ --align;
+ bytes = (bytes + align) & ~align;
+
+ if ((ptr = znalloc(mp, bytes)) != NULL) {
+ bzero(ptr, bytes);
+ }
+ return(ptr);
+}
+
+#endif
+
+/*
+ * znalloc() - allocate memory (without zeroing) from pool. Call reclaim
+ * and retry if appropriate, return NULL if unable to allocate
+ * memory.
+ */
+
+void *
+znalloc(MemPool *mp, iaddr_t bytes)
+{
+ /*
+ * align according to pool object size (can be 0). This is
+ * inclusive of the MEMNODE_SIZE_MASK minimum alignment.
+ *
+ */
+ bytes = (bytes + MEMNODE_SIZE_MASK) & ~MEMNODE_SIZE_MASK;
+
+ if (bytes == 0)
+ return((void *)-1);
+
+ do {
+ /*
+ * locate freelist entry big enough to hold the object. If all objects
+ * are the same size, this is a constant-time function.
+ */
+
+ if (bytes <= mp->mp_Size - mp->mp_Used) {
+ MemNode **pmn;
+ MemNode *mn;
+
+ for (pmn = &mp->mp_First; (mn=*pmn) != NULL; pmn = &mn->mr_Next) {
+ if (bytes > mn->mr_Bytes)
+ continue;
+
+ /*
+ * Cut a chunk of memory out of the beginning of this
+ * block and fixup the link appropriately.
+ */
+
+ {
+ char *ptr = (char *)mn;
+
+ if (mn->mr_Bytes == bytes) {
+ *pmn = mn->mr_Next;
+ } else {
+ mn = (MemNode *)((char *)mn + bytes);
+ mn->mr_Next = ((MemNode *)ptr)->mr_Next;
+ mn->mr_Bytes = ((MemNode *)ptr)->mr_Bytes - bytes;
+ *pmn = mn;
+ }
+ mp->mp_Used += bytes;
+ return(ptr);
+ }
+ }
+ }
+ } while (mp->mp_Reclaim(mp, bytes) == 0);
+
+ /*
+ * Memory pool is full, return NULL.
+ */
+
+ return(NULL);
+}
+
+#ifndef MALLOCLIB
+
+/*
+ * z[n]xalloc() - allocate memory from within a specific address region.
+ * If allocating AT a specific address, then addr2 must be
+ * set to addr1 + bytes (and this only works if addr1 is
+ * already aligned). addr1 and addr2 are aligned by
+ * MEMNODE_SIZE_MASK + 1 (i.e. they wlill be 8 or 16 byte
+ * aligned depending on the machine core).
+ */
+
+void *
+zxalloc(MemPool *mp, void *addr1, void *addr2, iaddr_t bytes)
+{
+ void *ptr;
+
+ if ((ptr = znxalloc(mp, addr1, addr2, bytes)) != NULL)
+ bzero(ptr, bytes);
+ return(ptr);
+}
+
+void *
+znxalloc(MemPool *mp, void *addr1, void *addr2, iaddr_t bytes)
+{
+ /*
+ * align according to pool object size (can be 0). This is
+ * inclusive of the MEMNODE_SIZE_MASK minimum alignment.
+ */
+ bytes = (bytes + MEMNODE_SIZE_MASK) & ~MEMNODE_SIZE_MASK;
+ addr1= (void *)(((iaddr_t)addr1 + MEMNODE_SIZE_MASK) & ~MEMNODE_SIZE_MASK);
+ addr2= (void *)(((iaddr_t)addr2 + MEMNODE_SIZE_MASK) & ~MEMNODE_SIZE_MASK);
+
+ if (bytes == 0)
+ return((void *)addr1);
+
+ /*
+ * Locate freelist entry big enough to hold the object that is within
+ * the allowed address range.
+ */
+
+ if (bytes <= mp->mp_Size - mp->mp_Used) {
+ MemNode **pmn;
+ MemNode *mn;
+
+ for (pmn = &mp->mp_First; (mn = *pmn) != NULL; pmn = &mn->mr_Next) {
+ int mrbytes = mn->mr_Bytes;
+ int offset = 0;
+
+ /*
+ * offset from base of mn to satisfy addr1. 0 or positive
+ */
+
+ if ((char *)mn < (char *)addr1)
+ offset = (char *)addr1 - (char *)mn;
+
+ /*
+ * truncate mrbytes to satisfy addr2. mrbytes may go negative
+ * if the mn is beyond the last acceptable address.
+ */
+
+ if ((char *)mn + mrbytes > (char *)addr2)
+ mrbytes = (saddr_t)((iaddr_t)addr2 - (iaddr_t)mn); /* signed */
+
+ /*
+ * beyond last acceptable address.
+ *
+ * before first acceptable address (if offset > mrbytes, the
+ * second conditional will always succeed).
+ *
+ * area overlapping acceptable address range is not big enough.
+ */
+
+ if (mrbytes < 0)
+ break;
+
+ if (mrbytes - offset < bytes)
+ continue;
+
+ /*
+ * Cut a chunk of memory out of the block and fixup the link
+ * appropriately.
+ *
+ * If offset != 0, we have to cut a chunk out from the middle of
+ * the block.
+ */
+
+ if (offset) {
+ MemNode *mnew = (MemNode *)((char *)mn + offset);
+
+ mnew->mr_Bytes = mn->mr_Bytes - offset;
+ mnew->mr_Next = mn->mr_Next;
+ mn->mr_Bytes = offset;
+ mn->mr_Next = mnew;
+ pmn = &mn->mr_Next;
+ mn = mnew;
+ }
+ {
+ char *ptr = (char *)mn;
+ if (mn->mr_Bytes == bytes) {
+ *pmn = mn->mr_Next;
+ } else {
+ mn = (MemNode *)((char *)mn + bytes);
+ mn->mr_Next = ((MemNode *)ptr)->mr_Next;
+ mn->mr_Bytes = ((MemNode *)ptr)->mr_Bytes - bytes;
+ *pmn = mn;
+ }
+ mp->mp_Used += bytes;
+ return(ptr);
+ }
+ }
+ }
+ return(NULL);
+}
+
+#endif
+
+/*
+ * zfree() - free previously allocated memory
+ */
+
+void
+zfree(MemPool *mp, void *ptr, iaddr_t bytes)
+{
+ /*
+ * align according to pool object size (can be 0). This is
+ * inclusive of the MEMNODE_SIZE_MASK minimum alignment.
+ */
+ bytes = (bytes + MEMNODE_SIZE_MASK) & ~MEMNODE_SIZE_MASK;
+
+ if (bytes == 0)
+ return;
+
+ /*
+ * panic if illegal pointer
+ */
+
+ if ((char *)ptr < (char *)mp->mp_Base ||
+ (char *)ptr + bytes > (char *)mp->mp_End ||
+ ((iaddr_t)ptr & MEMNODE_SIZE_MASK) != 0
+ ) {
+ mp->mp_Panic(
+ "zfree(%s,0x%08lx,%d): wild pointer",
+ mp->mp_Ident,
+ (long)ptr,
+ bytes
+ );
+ }
+
+ /*
+ * free the segment
+ */
+
+ {
+ MemNode **pmn;
+ MemNode *mn;
+
+ mp->mp_Used -= bytes;
+
+ for (pmn = &mp->mp_First; (mn = *pmn) != NULL; pmn = &mn->mr_Next) {
+ /*
+ * If area between last node and current node
+ * - check range
+ * - check merge with next area
+ * - check merge with previous area
+ */
+ if ((char *)ptr <= (char *)mn) {
+ /*
+ * range check
+ */
+ if ((char *)ptr + bytes > (char *)mn) {
+ mp->mp_Panic("zfree(%s,0x%08lx,%d): corrupt memlist1",
+ mp->mp_Ident,
+ (long)ptr,
+ bytes
+ );
+ }
+
+ /*
+ * merge against next area or create independant area
+ */
+
+ if ((char *)ptr + bytes == (char *)mn) {
+ ((MemNode *)ptr)->mr_Next = mn->mr_Next;
+ ((MemNode *)ptr)->mr_Bytes= bytes + mn->mr_Bytes;
+ } else {
+ ((MemNode *)ptr)->mr_Next = mn;
+ ((MemNode *)ptr)->mr_Bytes= bytes;
+ }
+ *pmn = mn = (MemNode *)ptr;
+
+ /*
+ * merge against previous area (if there is a previous
+ * area).
+ */
+
+ if (pmn != &mp->mp_First) {
+ if ((char*)pmn + ((MemNode*)pmn)->mr_Bytes == (char*)ptr) {
+ ((MemNode *)pmn)->mr_Next = mn->mr_Next;
+ ((MemNode *)pmn)->mr_Bytes += mn->mr_Bytes;
+ mn = (MemNode *)pmn;
+ }
+ }
+ return;
+ /* NOT REACHED */
+ }
+ if ((char *)ptr < (char *)mn + mn->mr_Bytes) {
+ mp->mp_Panic("zfree(%s,0x%08lx,%d): corrupt memlist2",
+ mp->mp_Ident,
+ (long)ptr,
+ bytes
+ );
+ }
+ }
+ /*
+ * We are beyond the last MemNode, append new MemNode. Merge against
+ * previous area if possible.
+ */
+ if (pmn == &mp->mp_First ||
+ (char *)pmn + ((MemNode *)pmn)->mr_Bytes != (char *)ptr
+ ) {
+ ((MemNode *)ptr)->mr_Next = NULL;
+ ((MemNode *)ptr)->mr_Bytes = bytes;
+ *pmn = (MemNode *)ptr;
+ mn = (MemNode *)ptr;
+ } else {
+ ((MemNode *)pmn)->mr_Bytes += bytes;
+ mn = (MemNode *)pmn;
+ }
+ }
+}
+
+#ifndef MALLOCLIB
+
+/*
+ * zallocStr() - allocate memory and copy string.
+ */
+
+char *
+zallocStr(MemPool *mp, const char *s, int slen)
+{
+ char *ptr;
+
+ if (slen < 0)
+ slen = strlen(s);
+ if ((ptr = znalloc(mp, slen + 1)) != NULL) {
+ bcopy(s, ptr, slen);
+ ptr[slen] = 0;
+ }
+ return(ptr);
+}
+
+/*
+ * zfreeStr() - free memory associated with an allocated string.
+ */
+
+void
+zfreeStr(MemPool *mp, char *s)
+{
+ zfree(mp, s, strlen(s) + 1);
+}
+
+#endif
+
+/*
+ * zinitpool() - initialize a memory pool
+ */
+
+void
+zinitPool(
+ MemPool *mp,
+ const char *id,
+ void (*fpanic)(const char *ctl, ...),
+ int (*freclaim)(MemPool *memPool, iaddr_t bytes),
+ void *pBase,
+ iaddr_t pSize
+) {
+ if (fpanic == NULL)
+ fpanic = znop;
+ if (freclaim == NULL)
+ freclaim = znot;
+
+ if (id != (const char *)-1)
+ mp->mp_Ident = id;
+ mp->mp_Base = pBase;
+ mp->mp_End = (char *)pBase + pSize;
+ mp->mp_First = NULL;
+ mp->mp_Size = pSize;
+ mp->mp_Used = pSize;
+ mp->mp_Panic = fpanic;
+ mp->mp_Reclaim = freclaim;
+}
+
+/*
+ * zextendPool() - extend memory pool to cover additional space.
+ *
+ * Note: the added memory starts out as allocated, you
+ * must free it to make it available to the memory subsystem.
+ *
+ * Note: mp_Size may not reflect (mp_End - mp_Base) range
+ * due to other parts of the system doing their own sbrk()
+ * calls.
+ */
+
+void
+zextendPool(MemPool *mp, void *base, iaddr_t bytes)
+{
+ if (mp->mp_Size == 0) {
+ mp->mp_Base = base;
+ mp->mp_Used = bytes;
+ } else {
+ void *pend = (char *)mp->mp_Base + mp->mp_Size;
+
+ if (base < mp->mp_Base) {
+ /* mp->mp_Size += (char *)mp->mp_Base - (char *)base; */
+ mp->mp_Used += (char *)mp->mp_Base - (char *)base;
+ mp->mp_Base = base;
+ }
+ base = (char *)base + bytes;
+ if (base > pend) {
+ /* mp->mp_Size += (char *)base - (char *)pend; */
+ mp->mp_Used += (char *)base - (char *)pend;
+ }
+ mp->mp_End = (char *)mp->mp_Base + mp->mp_Size;
+ }
+ mp->mp_Size += bytes;
+}
+
+#ifndef MALLOCLIB
+
+/*
+ * zclearpool() - Free all memory associated with a memory pool,
+ * destroying any previous allocations. Commonly
+ * called afte zinitPool() to make a pool available
+ * for use.
+ */
+
+void
+zclearPool(MemPool *mp)
+{
+ MemNode *mn = mp->mp_Base;
+
+ mn->mr_Next = NULL;
+ mn->mr_Bytes = mp->mp_Size;
+ mp->mp_First = mn;
+}
+
+#endif
+
+#ifdef ZALLOCDEBUG
+
+void
+zallocstats(MemPool *mp)
+{
+ int abytes = 0;
+ int hbytes = 0;
+ int fcount = 0;
+ MemNode *mn;
+
+ printf("Pool %s, %d bytes reserved", mp->mp_Ident, mp->mp_Size);
+
+ mn = mp->mp_First;
+
+ if ((void *)mn != (void *)mp->mp_Base) {
+ abytes += (char *)mn - (char *)mp->mp_Base;
+ }
+
+ while (mn) {
+ if ((char *)mn + mn->mr_Bytes != mp->mp_End) {
+ hbytes += mn->mr_Bytes;
+ ++fcount;
+ }
+ if (mn->mr_Next)
+ abytes += (char *)mn->mr_Next - ((char *)mn + mn->mr_Bytes);
+ mn = mn->mr_Next;
+ }
+ printf(" %d bytes allocated\n%d fragments (%d bytes fragmented)\n",
+ abytes,
+ fcount,
+ hbytes
+ );
+}
+
+#endif
+
OpenPOWER on IntegriCloud