#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: subr_extent.c,v 1.91 2026/01/04 03:17:47 riastradh Exp $");
#ifdef _KERNEL
#ifdef _KERNEL_OPT
#include "opt_lockdebug.h"
#endif
#include <sys/param.h>
#include <sys/types.h>
#include <sys/extent.h>
#include <sys/kmem.h>
#include <sys/pool.h>
#include <sys/proc.h>
#include <sys/sdt.h>
#include <sys/systm.h>
#include <sys/time.h>
#include <uvm/uvm_extern.h>
#elif defined(_EXTENT_TESTING)
#include <sys/param.h>
#include <sys/pool.h>
#include <sys/extent.h>
#include <errno.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
static inline void no_op(void) { return; }
#define \
kmem_alloc(s, flags) malloc(s)
#define \
kmem_free(p, s) free(p)
#define \
cv_wait_sig(cv, lock) (EWOULDBLOCK)
#define \
pool_get(pool, flags) kmem_alloc((pool)->pr_size,0)
#define \
pool_put(pool, rp) kmem_free(rp,0)
#define \
panic(a ...) printf(a)
#define mutex_init(a, b, c) no_op()
#define mutex_destroy(a) no_op()
#define mutex_enter(l) no_op()
#define mutex_exit(l) no_op()
#define cv_wait(cv, lock) no_op()
#define cv_broadcast(cv) no_op()
#define cv_init(a, b) no_op()
#define cv_destroy(a) no_op()
#define KMEM_IS_RUNNING (1)
#define IPL_VM (0)
#define MUTEX_DEFAULT (0)
#define KASSERT(exp)
#define SET_ERROR(error) (error)
#endif
static struct pool expool;
#define EXTENT_ALIGN(_start, _align, _skew) \
(((((_start) - (_skew)) + ((_align) - 1)) & (-(_align))) + (_skew))
void
extent_init(void)
{
#if defined(_KERNEL)
pool_init(&expool, sizeof(struct extent_region), 0, 0, 0,
"extent", NULL, IPL_VM);
#else
expool.pr_size = sizeof(struct extent_region);
#endif
}
static struct extent_region *
extent_alloc_region_descriptor(struct extent *ex, int flags)
{
struct extent_region *rp;
int error;
if (ex->ex_flags & EXF_FIXED) {
struct extent_fixed *fex = (struct extent_fixed *)ex;
if (!(ex->ex_flags & EXF_EARLY))
mutex_enter(&ex->ex_lock);
for (;;) {
if ((rp = LIST_FIRST(&fex->fex_freelist)) != NULL) {
LIST_REMOVE(rp, er_link);
if (!(ex->ex_flags & EXF_EARLY))
mutex_exit(&ex->ex_lock);
return (rp);
}
if (flags & EX_MALLOCOK) {
if (!(ex->ex_flags & EXF_EARLY))
mutex_exit(&ex->ex_lock);
goto alloc;
}
if ((flags & EX_WAITOK) == 0) {
if (!(ex->ex_flags & EXF_EARLY))
mutex_exit(&ex->ex_lock);
return (NULL);
}
KASSERT(mutex_owned(&ex->ex_lock));
ex->ex_flwanted = true;
if ((flags & EX_CATCH) != 0)
error = cv_wait_sig(&ex->ex_cv, &ex->ex_lock);
else {
cv_wait(&ex->ex_cv, &ex->ex_lock);
error = 0;
}
if (error != 0) {
mutex_exit(&ex->ex_lock);
return (NULL);
}
}
}
alloc:
rp = pool_get(&expool, (flags & EX_WAITOK) ? PR_WAITOK : PR_NOWAIT);
if (rp != NULL)
rp->er_flags = ER_ALLOC;
return (rp);
}
static void
extent_free_region_descriptor(struct extent *ex, struct extent_region *rp)
{
if (ex->ex_flags & EXF_FIXED) {
struct extent_fixed *fex = (struct extent_fixed *)ex;
if (rp->er_flags & ER_ALLOC) {
if (ex->ex_flwanted) {
rp->er_flags = ER_ALLOC;
LIST_INSERT_HEAD(&fex->fex_freelist, rp,
er_link);
goto wake_em_up;
} else
pool_put(&expool, rp);
} else {
rp->er_flags = 0;
LIST_INSERT_HEAD(&fex->fex_freelist, rp, er_link);
}
wake_em_up:
if (!(ex->ex_flags & EXF_EARLY)) {
ex->ex_flwanted = false;
cv_broadcast(&ex->ex_cv);
}
return;
}
pool_put(&expool, rp);
}
struct extent *
extent_create(const char *name, u_long start, u_long end,
void *storage, size_t storagesize, int flags)
{
struct extent *ex;
char *cp = storage;
size_t sz = storagesize;
struct extent_region *rp;
int fixed_extent = (storage != NULL);
#ifndef _KERNEL
extent_init();
#endif
#ifdef DIAGNOSTIC
if (name == NULL)
panic("extent_create: name == NULL");
if (end < start) {
printf("extent_create: extent `%s', start 0x%lx, end 0x%lx\n",
name, start, end);
panic("extent_create: end < start");
}
if (fixed_extent && (storagesize < sizeof(struct extent_fixed)))
panic("extent_create: fixed extent, bad storagesize 0x%lx",
(u_long)storagesize);
if (fixed_extent == 0 && (storagesize != 0 || storage != NULL))
panic("extent_create: storage provided for non-fixed");
#endif
if (fixed_extent) {
struct extent_fixed *fex;
memset(storage, 0, storagesize);
fex = (struct extent_fixed *)cp;
ex = (struct extent *)fex;
cp += ALIGN(sizeof(struct extent_fixed));
sz -= ALIGN(sizeof(struct extent_fixed));
fex->fex_storage = storage;
fex->fex_storagesize = storagesize;
LIST_INIT(&fex->fex_freelist);
while (sz >= ALIGN(sizeof(struct extent_region))) {
rp = (struct extent_region *)cp;
cp += ALIGN(sizeof(struct extent_region));
sz -= ALIGN(sizeof(struct extent_region));
LIST_INSERT_HEAD(&fex->fex_freelist, rp, er_link);
}
} else {
ex = kmem_alloc(sizeof(*ex),
(flags & EX_WAITOK) ? KM_SLEEP : KM_NOSLEEP);
if (ex == NULL)
return (NULL);
}
if ((flags & EX_EARLY) == 0) {
mutex_init(&ex->ex_lock, MUTEX_DEFAULT, IPL_VM);
cv_init(&ex->ex_cv, "extent");
}
LIST_INIT(&ex->ex_regions);
ex->ex_name = name;
ex->ex_start = start;
ex->ex_end = end;
ex->ex_flags = 0;
ex->ex_flwanted = false;
if (fixed_extent)
ex->ex_flags |= EXF_FIXED;
if (flags & EX_NOCOALESCE)
ex->ex_flags |= EXF_NOCOALESCE;
if (flags & EX_EARLY)
ex->ex_flags |= EXF_EARLY;
return (ex);
}
void
extent_destroy(struct extent *ex)
{
struct extent_region *rp, *orp;
#ifdef DIAGNOSTIC
if (ex == NULL)
panic("extent_destroy: NULL extent");
#endif
for (rp = LIST_FIRST(&ex->ex_regions); rp != NULL; ) {
orp = rp;
rp = LIST_NEXT(rp, er_link);
LIST_REMOVE(orp, er_link);
extent_free_region_descriptor(ex, orp);
}
cv_destroy(&ex->ex_cv);
mutex_destroy(&ex->ex_lock);
if ((ex->ex_flags & EXF_FIXED) == 0)
kmem_free(ex, sizeof(*ex));
}
static void
extent_insert_and_optimize(struct extent *ex, u_long start, u_long size,
int flags, struct extent_region *after, struct extent_region *rp)
{
struct extent_region *nextr;
int appended = 0;
if (after == NULL) {
if (((ex->ex_flags & EXF_NOCOALESCE) == 0) &&
(LIST_FIRST(&ex->ex_regions) != NULL) &&
((start + size) == LIST_FIRST(&ex->ex_regions)->er_start)) {
LIST_FIRST(&ex->ex_regions)->er_start = start;
extent_free_region_descriptor(ex, rp);
return;
}
rp->er_start = start;
rp->er_end = start + (size - 1);
LIST_INSERT_HEAD(&ex->ex_regions, rp, er_link);
return;
}
if (ex->ex_flags & EXF_NOCOALESCE)
goto cant_coalesce;
if ((after->er_end + 1) == start) {
after->er_end = start + (size - 1);
appended = 1;
}
if ((LIST_NEXT(after, er_link) != NULL) &&
((start + size) == LIST_NEXT(after, er_link)->er_start)) {
if (appended) {
after->er_end = LIST_NEXT(after, er_link)->er_end;
nextr = LIST_NEXT(after, er_link);
LIST_REMOVE(nextr, er_link);
extent_free_region_descriptor(ex, nextr);
} else {
LIST_NEXT(after, er_link)->er_start = start;
}
extent_free_region_descriptor(ex, rp);
return;
}
if (appended) {
extent_free_region_descriptor(ex, rp);
return;
}
cant_coalesce:
rp->er_start = start;
rp->er_end = start + (size - 1);
LIST_INSERT_AFTER(after, rp, er_link);
}
int
extent_alloc_region(struct extent *ex, u_long start, u_long size, int flags)
{
struct extent_region *rp, *last, *myrp;
u_long end = start + (size - 1);
int error;
#ifdef DIAGNOSTIC
if (ex == NULL)
panic("extent_alloc_region: NULL extent");
if (size < 1) {
printf("extent_alloc_region: extent `%s', size 0x%lx\n",
ex->ex_name, size);
panic("extent_alloc_region: bad size");
}
if (end < start) {
printf(
"extent_alloc_region: extent `%s', start 0x%lx, size 0x%lx\n",
ex->ex_name, start, size);
panic("extent_alloc_region: overflow");
}
#endif
#ifdef LOCKDEBUG
if (flags & EX_WAITSPACE) {
ASSERT_SLEEPABLE();
}
#endif
if ((start < ex->ex_start) || (end > ex->ex_end)) {
#ifdef DIAGNOSTIC
printf("extent_alloc_region: extent `%s' (0x%lx - 0x%lx)\n",
ex->ex_name, ex->ex_start, ex->ex_end);
printf("extent_alloc_region: start 0x%lx, end 0x%lx\n",
start, end);
panic("extent_alloc_region: region lies outside extent");
#else
return SET_ERROR(EINVAL);
#endif
}
myrp = extent_alloc_region_descriptor(ex, flags);
if (myrp == NULL) {
#ifdef DIAGNOSTIC
printf(
"extent_alloc_region: can't allocate region descriptor\n");
#endif
return SET_ERROR(ENOMEM);
}
if (!(ex->ex_flags & EXF_EARLY))
mutex_enter(&ex->ex_lock);
alloc_start:
last = NULL;
LIST_FOREACH(rp, &ex->ex_regions, er_link) {
if (rp->er_start > end) {
break;
}
if (rp->er_end >= start) {
if (flags & EX_WAITSPACE) {
KASSERT(!(ex->ex_flags & EXF_EARLY));
if ((flags & EX_CATCH) != 0)
error = cv_wait_sig(&ex->ex_cv,
&ex->ex_lock);
else {
cv_wait(&ex->ex_cv, &ex->ex_lock);
error = 0;
}
if (error == 0)
goto alloc_start;
mutex_exit(&ex->ex_lock);
} else {
if (!(ex->ex_flags & EXF_EARLY))
mutex_exit(&ex->ex_lock);
error = SET_ERROR(EAGAIN);
}
extent_free_region_descriptor(ex, myrp);
return error;
}
last = rp;
}
extent_insert_and_optimize(ex, start, size, flags, last, myrp);
if (!(ex->ex_flags & EXF_EARLY))
mutex_exit(&ex->ex_lock);
return (0);
}
#define LE_OV(x, y, z) ((((x) + (y)) >= (x)) && (((x) + (y)) <= (z)))
int
extent_alloc_subregion1(struct extent *ex, u_long substart, u_long subend,
u_long size, u_long alignment, u_long skew, u_long boundary,
int flags, u_long *result)
{
struct extent_region *rp, *myrp, *last, *bestlast;
u_long newstart, newend, exend, beststart, bestovh, ovh;
u_long dontcross;
int error;
#ifdef DIAGNOSTIC
if (ex == NULL)
panic("extent_alloc_subregion: NULL extent");
if (result == NULL)
panic("extent_alloc_subregion: NULL result pointer");
if ((substart < ex->ex_start) || (substart > ex->ex_end) ||
(subend > ex->ex_end) || (subend < ex->ex_start)) {
printf("extent_alloc_subregion: extent `%s', ex_start 0x%lx, "
"ex_end 0x%lx\n", ex->ex_name, ex->ex_start, ex->ex_end);
printf("extent_alloc_subregion: substart 0x%lx, subend 0x%lx\n",
substart, subend);
panic("extent_alloc_subregion: bad subregion");
}
if ((size < 1) || ((size - 1) > (subend - substart))) {
printf("extent_alloc_subregion: extent `%s', size 0x%lx\n",
ex->ex_name, size);
printf("extent_alloc_subregion: substart 0x%lx, subend 0x%lx\n",
substart, subend);
panic("extent_alloc_subregion: bad size");
}
if (alignment == 0)
panic("extent_alloc_subregion: bad alignment");
if (boundary && (boundary < size)) {
printf(
"extent_alloc_subregion: extent `%s', size 0x%lx, "
"boundary 0x%lx\n", ex->ex_name, size, boundary);
panic("extent_alloc_subregion: bad boundary");
}
#endif
#ifdef LOCKDEBUG
if (flags & EX_WAITSPACE) {
ASSERT_SLEEPABLE();
}
#endif
myrp = extent_alloc_region_descriptor(ex, flags);
if (myrp == NULL) {
#ifdef DIAGNOSTIC
printf(
"extent_alloc_subregion: can't allocate region descriptor\n");
#endif
return SET_ERROR(ENOMEM);
}
alloc_start:
mutex_enter(&ex->ex_lock);
last = NULL;
bestovh = 0;
beststart = 0;
bestlast = NULL;
exend = ex->ex_end;
newstart = EXTENT_ALIGN(substart, alignment, skew);
if (newstart < ex->ex_start) {
#ifdef DIAGNOSTIC
printf(
"extent_alloc_subregion: extent `%s' (0x%lx - 0x%lx), alignment 0x%lx\n",
ex->ex_name, ex->ex_start, ex->ex_end, alignment);
mutex_exit(&ex->ex_lock);
panic("extent_alloc_subregion: overflow after alignment");
#else
extent_free_region_descriptor(ex, myrp);
mutex_exit(&ex->ex_lock);
return SET_ERROR(EINVAL);
#endif
}
LIST_FOREACH(rp, &ex->ex_regions, er_link) {
if (rp->er_start >= newstart)
break;
last = rp;
}
if (last != NULL && last->er_end >= newstart)
newstart = EXTENT_ALIGN((last->er_end + 1), alignment, skew);
for (; rp != NULL; rp = LIST_NEXT(rp, er_link)) {
if (rp->er_start > subend) {
exend = rp->er_start;
break;
}
if (LE_OV(newstart, size, rp->er_start)) {
if (boundary) {
newend = newstart + (size - 1);
dontcross = EXTENT_ALIGN(newstart+1, boundary,
(flags & EX_BOUNDZERO) ? 0 : ex->ex_start)
- 1;
#if 0
printf("newstart=%lx newend=%lx ex_start=%lx ex_end=%lx boundary=%lx dontcross=%lx\n",
newstart, newend, ex->ex_start, ex->ex_end,
boundary, dontcross);
#endif
if (dontcross < ex->ex_start)
dontcross = ex->ex_end;
else if (newend > dontcross) {
newstart = dontcross + 1;
newend = newstart + (size - 1);
dontcross += boundary;
if (!LE_OV(newstart, size, rp->er_start))
goto skip;
}
if (newstart + size - 1 > ex->ex_end ||
dontcross < newstart)
goto fail;
}
ovh = rp->er_start - newstart - size;
if ((flags & EX_FAST) || (ovh == 0))
goto found;
if ((bestovh == 0) || (ovh < bestovh)) {
bestovh = ovh;
beststart = newstart;
bestlast = last;
}
}
skip:
newstart = EXTENT_ALIGN((rp->er_end + 1), alignment, skew);
if (newstart < rp->er_end) {
goto fail;
}
last = rp;
}
if (LE_OV(newstart, (size - 1), subend)) {
if (boundary) {
newend = newstart + (size - 1);
dontcross = EXTENT_ALIGN(newstart+1, boundary,
(flags & EX_BOUNDZERO) ? 0 : ex->ex_start)
- 1;
#if 0
printf("newstart=%lx newend=%lx ex_start=%lx ex_end=%lx boundary=%lx dontcross=%lx\n",
newstart, newend, ex->ex_start, ex->ex_end,
boundary, dontcross);
#endif
if (dontcross < ex->ex_start)
dontcross = ex->ex_end;
else if (newend > dontcross) {
newstart = dontcross + 1;
newend = newstart + (size - 1);
dontcross += boundary;
if (!LE_OV(newstart, (size - 1), subend))
goto fail;
}
if (newstart + size - 1 > ex->ex_end ||
dontcross < newstart)
goto fail;
}
ovh = exend - newstart - (size - 1);
if ((flags & EX_FAST) || (ovh == 0))
goto found;
if ((bestovh == 0) || (ovh < bestovh)) {
bestovh = ovh;
beststart = newstart;
bestlast = last;
}
}
fail:
if (((flags & EX_FAST) == 0) && (bestovh != 0)) {
newstart = beststart;
last = bestlast;
goto found;
}
if (flags & EX_WAITSPACE) {
if ((flags & EX_CATCH) != 0) {
error = cv_wait_sig(&ex->ex_cv, &ex->ex_lock);
} else {
cv_wait(&ex->ex_cv, &ex->ex_lock);
error = 0;
}
if (error == 0)
goto alloc_start;
mutex_exit(&ex->ex_lock);
} else {
mutex_exit(&ex->ex_lock);
error = SET_ERROR(EAGAIN);
}
extent_free_region_descriptor(ex, myrp);
return error;
found:
extent_insert_and_optimize(ex, newstart, size, flags, last, myrp);
mutex_exit(&ex->ex_lock);
*result = newstart;
return (0);
}
int
extent_alloc_subregion(struct extent *ex, u_long start, u_long end, u_long size,
u_long alignment, u_long boundary, int flags, u_long *result)
{
return (extent_alloc_subregion1(ex, start, end, size, alignment,
0, boundary, flags, result));
}
int
extent_alloc(struct extent *ex, u_long size, u_long alignment, u_long boundary,
int flags, u_long *result)
{
return (extent_alloc_subregion1(ex, ex->ex_start, ex->ex_end,
size, alignment, 0, boundary,
flags, result));
}
int
extent_alloc1(struct extent *ex, u_long size, u_long alignment, u_long skew,
u_long boundary, int flags, u_long *result)
{
return (extent_alloc_subregion1(ex, ex->ex_start, ex->ex_end,
size, alignment, skew, boundary,
flags, result));
}
int
extent_free(struct extent *ex, u_long start, u_long size, int flags)
{
struct extent_region *rp, *nrp = NULL;
u_long end = start + (size - 1);
#ifdef DIAGNOSTIC
if (ex == NULL)
panic("extent_free: NULL extent");
if ((start < ex->ex_start) || (end > ex->ex_end)) {
extent_print(ex);
printf("extent_free: extent `%s', start 0x%lx, size 0x%lx\n",
ex->ex_name, start, size);
panic("extent_free: extent `%s', region not within extent",
ex->ex_name);
}
if (end < start) {
extent_print(ex);
printf("extent_free: extent `%s', start 0x%lx, size 0x%lx\n",
ex->ex_name, start, size);
panic("extent_free: overflow");
}
#endif
const bool coalesce = (ex->ex_flags & EXF_NOCOALESCE) == 0;
if (coalesce) {
nrp = extent_alloc_region_descriptor(ex, flags);
if (nrp == NULL)
return SET_ERROR(ENOMEM);
}
if (!(ex->ex_flags & EXF_EARLY))
mutex_enter(&ex->ex_lock);
LIST_FOREACH(rp, &ex->ex_regions, er_link) {
if (rp->er_end < start)
continue;
if (rp->er_start > end)
break;
if ((start == rp->er_start) && (end == rp->er_end)) {
LIST_REMOVE(rp, er_link);
extent_free_region_descriptor(ex, rp);
goto done;
}
if (!coalesce)
continue;
if ((start == rp->er_start) && (end < rp->er_end)) {
rp->er_start = (end + 1);
goto done;
}
if ((start > rp->er_start) && (end == rp->er_end)) {
rp->er_end = (start - 1);
goto done;
}
if ((start > rp->er_start) && (end < rp->er_end)) {
nrp->er_start = end + 1;
nrp->er_end = rp->er_end;
rp->er_end = start - 1;
LIST_INSERT_AFTER(rp, nrp, er_link);
nrp = NULL;
goto done;
}
}
if (!(ex->ex_flags & EXF_EARLY))
mutex_exit(&ex->ex_lock);
extent_print(ex);
printf("extent_free: start 0x%lx, end 0x%lx\n", start, end);
panic("extent_free: region not found");
done:
if (nrp != NULL)
extent_free_region_descriptor(ex, nrp);
if (!(ex->ex_flags & EXF_EARLY)) {
cv_broadcast(&ex->ex_cv);
mutex_exit(&ex->ex_lock);
}
return (0);
}
void
extent_print(struct extent *ex)
{
struct extent_region *rp;
if (ex == NULL)
panic("extent_print: NULL extent");
if (!(ex->ex_flags & EXF_EARLY))
mutex_enter(&ex->ex_lock);
printf("extent `%s' (0x%lx - 0x%lx), flags = 0x%x\n", ex->ex_name,
ex->ex_start, ex->ex_end, ex->ex_flags);
LIST_FOREACH(rp, &ex->ex_regions, er_link)
printf(" 0x%lx - 0x%lx\n", rp->er_start, rp->er_end);
if (!(ex->ex_flags & EXF_EARLY))
mutex_exit(&ex->ex_lock);
}