#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: subr_blist.c,v 1.16 2024/09/08 17:28:36 rillig Exp $");
#if 0
__FBSDID("$FreeBSD: src/sys/kern/subr_blist.c,v 1.17 2004/06/04 04:03:25 alc Exp $");
#endif
#ifdef _KERNEL
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/blist.h>
#include <sys/kmem.h>
#else
#ifndef BLIST_NO_DEBUG
#define BLIST_DEBUG
#endif
#include <sys/types.h>
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <stdarg.h>
#include <inttypes.h>
#define KM_SLEEP 1
#define kmem_zalloc(a,b) calloc(1, (a))
#define kmem_alloc(a,b) malloc(a)
#define kmem_free(a,b) free(a)
#include "../sys/blist.h"
void panic(const char *ctl, ...) __printflike(1, 2);
#endif
typedef struct blmeta {
union {
blist_blkno_t bmu_avail;
blist_bitmap_t bmu_bitmap;
} u;
blist_blkno_t bm_bighint;
} blmeta_t;
struct blist {
blist_blkno_t bl_blocks;
blist_blkno_t bl_radix;
blist_blkno_t bl_skip;
blist_blkno_t bl_free;
blmeta_t *bl_root;
blist_blkno_t bl_rootblks;
};
#define BLIST_META_RADIX 16
static blist_blkno_t blst_leaf_alloc(blmeta_t *scan, blist_blkno_t blk,
int count);
static blist_blkno_t blst_meta_alloc(blmeta_t *scan, blist_blkno_t blk,
blist_blkno_t count, blist_blkno_t radix, blist_blkno_t skip);
static void blst_leaf_free(blmeta_t *scan, blist_blkno_t relblk, int count);
static void blst_meta_free(blmeta_t *scan, blist_blkno_t freeBlk,
blist_blkno_t count, blist_blkno_t radix, blist_blkno_t skip,
blist_blkno_t blk);
static void blst_copy(blmeta_t *scan, blist_blkno_t blk, blist_blkno_t radix,
blist_blkno_t skip, blist_t dest, blist_blkno_t count);
static int blst_leaf_fill(blmeta_t *scan, blist_blkno_t blk, int count);
static blist_blkno_t blst_meta_fill(blmeta_t *scan, blist_blkno_t allocBlk,
blist_blkno_t count, blist_blkno_t radix, blist_blkno_t skip,
blist_blkno_t blk);
static blist_blkno_t blst_radix_init(blmeta_t *scan, blist_blkno_t radix,
blist_blkno_t skip, blist_blkno_t count);
#ifndef _KERNEL
static void blst_radix_print(blmeta_t *scan, blist_blkno_t blk,
blist_blkno_t radix, blist_blkno_t skip, int tab);
#endif
blist_t
blist_create(blist_blkno_t blocks)
{
blist_t bl;
blist_blkno_t radix;
blist_blkno_t skip = 0;
radix = BLIST_BMAP_RADIX;
while (radix < blocks) {
radix *= BLIST_META_RADIX;
skip = (skip + 1) * BLIST_META_RADIX;
}
bl = kmem_zalloc(sizeof(struct blist), KM_SLEEP);
bl->bl_blocks = blocks;
bl->bl_radix = radix;
bl->bl_skip = skip;
bl->bl_rootblks = 1 +
blst_radix_init(NULL, bl->bl_radix, bl->bl_skip, blocks);
bl->bl_root = kmem_alloc(sizeof(blmeta_t) * bl->bl_rootblks, KM_SLEEP);
#if defined(BLIST_DEBUG)
printf(
"BLIST representing %" PRIu64 " blocks (%" PRIu64 " MB of swap)"
", requiring %" PRIu64 "K of ram\n",
(uint64_t)bl->bl_blocks,
(uint64_t)bl->bl_blocks * 4 / 1024,
((uint64_t)bl->bl_rootblks * sizeof(blmeta_t) + 1023) / 1024
);
printf("BLIST raw radix tree contains %" PRIu64 " records\n",
(uint64_t)bl->bl_rootblks);
#endif
blst_radix_init(bl->bl_root, bl->bl_radix, bl->bl_skip, blocks);
return(bl);
}
void
blist_destroy(blist_t bl)
{
kmem_free(bl->bl_root, sizeof(blmeta_t) * bl->bl_rootblks);
kmem_free(bl, sizeof(struct blist));
}
blist_blkno_t
blist_alloc(blist_t bl, blist_blkno_t count)
{
blist_blkno_t blk = BLIST_NONE;
if (bl) {
if (bl->bl_radix == BLIST_BMAP_RADIX)
blk = blst_leaf_alloc(bl->bl_root, 0, count);
else
blk = blst_meta_alloc(bl->bl_root, 0, count, bl->bl_radix, bl->bl_skip);
if (blk != BLIST_NONE)
bl->bl_free -= count;
}
return(blk);
}
void
blist_free(blist_t bl, blist_blkno_t blkno, blist_blkno_t count)
{
if (bl) {
if (bl->bl_radix == BLIST_BMAP_RADIX)
blst_leaf_free(bl->bl_root, blkno, count);
else
blst_meta_free(bl->bl_root, blkno, count, bl->bl_radix, bl->bl_skip, 0);
bl->bl_free += count;
}
}
blist_blkno_t
blist_fill(blist_t bl, blist_blkno_t blkno, blist_blkno_t count)
{
blist_blkno_t filled;
if (bl) {
if (bl->bl_radix == BLIST_BMAP_RADIX)
filled = blst_leaf_fill(bl->bl_root, blkno, count);
else
filled = blst_meta_fill(bl->bl_root, blkno, count,
bl->bl_radix, bl->bl_skip, 0);
bl->bl_free -= filled;
return filled;
} else
return 0;
}
void
blist_resize(blist_t *pbl, blist_blkno_t count, int freenew)
{
blist_t newbl = blist_create(count);
blist_t save = *pbl;
*pbl = newbl;
if (count > save->bl_blocks)
count = save->bl_blocks;
blst_copy(save->bl_root, 0, save->bl_radix, save->bl_skip, newbl, count);
if (freenew && count < newbl->bl_blocks) {
blist_free(newbl, count, newbl->bl_blocks - count);
}
blist_destroy(save);
}
#ifdef BLIST_DEBUG
void
blist_print(blist_t bl)
{
printf("BLIST {\n");
blst_radix_print(bl->bl_root, 0, bl->bl_radix, bl->bl_skip, 4);
printf("}\n");
}
#endif
static blist_blkno_t
blst_leaf_alloc(
blmeta_t *scan,
blist_blkno_t blk,
int count
) {
blist_bitmap_t orig = scan->u.bmu_bitmap;
if (orig == 0) {
scan->bm_bighint = 0;
return(BLIST_NONE);
}
if (count == 1) {
blist_bitmap_t mask;
int j = BLIST_BMAP_RADIX/2;
int r = 0;
mask = (blist_bitmap_t)-1 >> (BLIST_BMAP_RADIX/2);
while (j) {
if ((orig & mask) == 0) {
r += j;
orig >>= j;
}
j >>= 1;
mask >>= j;
}
scan->u.bmu_bitmap &= ~((blist_bitmap_t)1 << r);
return(blk + r);
}
if (count <= BLIST_BMAP_RADIX) {
int j;
int n = BLIST_BMAP_RADIX - count;
blist_bitmap_t mask;
mask = (blist_bitmap_t)-1 >> n;
for (j = 0; j <= n; ++j) {
if ((orig & mask) == mask) {
scan->u.bmu_bitmap &= ~mask;
return(blk + j);
}
mask = (mask << 1);
}
}
scan->bm_bighint = count - 1;
return(BLIST_NONE);
}
static blist_blkno_t
blst_meta_alloc(
blmeta_t *scan,
blist_blkno_t blk,
blist_blkno_t count,
blist_blkno_t radix,
blist_blkno_t skip
) {
blist_blkno_t i;
blist_blkno_t next_skip = (skip / BLIST_META_RADIX);
if (scan->u.bmu_avail == 0) {
scan->bm_bighint = count;
return(BLIST_NONE);
}
if (scan->u.bmu_avail == radix) {
radix /= BLIST_META_RADIX;
for (i = 1; i <= skip; i += next_skip) {
if (scan[i].bm_bighint == (blist_blkno_t)-1)
break;
if (next_skip == 1) {
scan[i].u.bmu_bitmap = (blist_bitmap_t)-1;
scan[i].bm_bighint = BLIST_BMAP_RADIX;
} else {
scan[i].bm_bighint = radix;
scan[i].u.bmu_avail = radix;
}
}
} else {
radix /= BLIST_META_RADIX;
}
for (i = 1; i <= skip; i += next_skip) {
if (scan[i].bm_bighint == (blist_blkno_t)-1) {
break;
} else if (count <= scan[i].bm_bighint) {
blist_blkno_t r;
if (next_skip == 1) {
r = blst_leaf_alloc(&scan[i], blk, count);
} else {
r = blst_meta_alloc(&scan[i], blk, count, radix, next_skip - 1);
}
if (r != BLIST_NONE) {
scan->u.bmu_avail -= count;
if (scan->bm_bighint > scan->u.bmu_avail)
scan->bm_bighint = scan->u.bmu_avail;
return(r);
}
} else if (count > radix) {
panic("blist_meta_alloc: allocation too large");
}
blk += radix;
}
if (scan->bm_bighint >= count)
scan->bm_bighint = count - 1;
return(BLIST_NONE);
}
static void
blst_leaf_free(
blmeta_t *scan,
blist_blkno_t blk,
int count
) {
int n = blk & (BLIST_BMAP_RADIX - 1);
blist_bitmap_t mask;
mask = ((blist_bitmap_t)-1 << n) &
((blist_bitmap_t)-1 >> (BLIST_BMAP_RADIX - count - n));
if (scan->u.bmu_bitmap & mask)
panic("blst_radix_free: freeing free block");
scan->u.bmu_bitmap |= mask;
scan->bm_bighint = BLIST_BMAP_RADIX;
}
static void
blst_meta_free(
blmeta_t *scan,
blist_blkno_t freeBlk,
blist_blkno_t count,
blist_blkno_t radix,
blist_blkno_t skip,
blist_blkno_t blk
) {
blist_blkno_t i;
blist_blkno_t next_skip = (skip / BLIST_META_RADIX);
#if 0
printf("FREE (%" PRIx64 ",%" PRIu64
") FROM (%" PRIx64 ",%" PRIu64 ")\n",
(uint64_t)freeBlk, (uint64_t)count,
(uint64_t)blk, (uint64_t)radix
);
#endif
if (scan->u.bmu_avail == 0) {
scan->u.bmu_avail = count;
scan->bm_bighint = count;
if (count != radix) {
for (i = 1; i <= skip; i += next_skip) {
if (scan[i].bm_bighint == (blist_blkno_t)-1)
break;
scan[i].bm_bighint = 0;
if (next_skip == 1) {
scan[i].u.bmu_bitmap = 0;
} else {
scan[i].u.bmu_avail = 0;
}
}
}
} else {
scan->u.bmu_avail += count;
}
if (scan->u.bmu_avail == radix)
return;
if (scan->u.bmu_avail > radix)
panic("blst_meta_free: freeing already free blocks (%"
PRIu64 ") %" PRIu64 "/%" PRIu64,
(uint64_t)count,
(uint64_t)scan->u.bmu_avail,
(uint64_t)radix);
radix /= BLIST_META_RADIX;
i = (freeBlk - blk) / radix;
blk += i * radix;
i = i * next_skip + 1;
while (i <= skip && blk < freeBlk + count) {
blist_blkno_t v;
v = blk + radix - freeBlk;
if (v > count)
v = count;
if (scan->bm_bighint == (blist_blkno_t)-1)
panic("blst_meta_free: freeing unexpected range");
if (next_skip == 1) {
blst_leaf_free(&scan[i], freeBlk, v);
} else {
blst_meta_free(&scan[i], freeBlk, v, radix, next_skip - 1, blk);
}
if (scan->bm_bighint < scan[i].bm_bighint)
scan->bm_bighint = scan[i].bm_bighint;
count -= v;
freeBlk += v;
blk += radix;
i += next_skip;
}
}
static void blst_copy(
blmeta_t *scan,
blist_blkno_t blk,
blist_blkno_t radix,
blist_blkno_t skip,
blist_t dest,
blist_blkno_t count
) {
blist_blkno_t next_skip;
blist_blkno_t i;
if (radix == BLIST_BMAP_RADIX) {
blist_bitmap_t v = scan->u.bmu_bitmap;
if (v == (blist_bitmap_t)-1) {
blist_free(dest, blk, count);
} else if (v != 0) {
int j;
for (j = 0; j < BLIST_BMAP_RADIX && j < count; ++j) {
if (v & (1 << j))
blist_free(dest, blk + j, 1);
}
}
return;
}
if (scan->u.bmu_avail == 0) {
return;
}
if (scan->u.bmu_avail == radix) {
if (count < radix)
blist_free(dest, blk, count);
else
blist_free(dest, blk, radix);
return;
}
radix /= BLIST_META_RADIX;
next_skip = (skip / BLIST_META_RADIX);
for (i = 1; count && i <= skip; i += next_skip) {
if (scan[i].bm_bighint == (blist_blkno_t)-1)
break;
if (count >= radix) {
blst_copy(
&scan[i],
blk,
radix,
next_skip - 1,
dest,
radix
);
count -= radix;
} else {
if (count) {
blst_copy(
&scan[i],
blk,
radix,
next_skip - 1,
dest,
count
);
}
count = 0;
}
blk += radix;
}
}
static int
blst_leaf_fill(blmeta_t *scan, blist_blkno_t blk, int count)
{
int n = blk & (BLIST_BMAP_RADIX - 1);
int nblks;
blist_bitmap_t mask, bitmap;
mask = ((blist_bitmap_t)-1 << n) &
((blist_bitmap_t)-1 >> (BLIST_BMAP_RADIX - count - n));
bitmap = scan->u.bmu_bitmap & mask;
for (nblks = 0; bitmap != 0; nblks++)
bitmap &= bitmap - 1;
scan->u.bmu_bitmap &= ~mask;
return nblks;
}
static blist_blkno_t
blst_meta_fill(
blmeta_t *scan,
blist_blkno_t allocBlk,
blist_blkno_t count,
blist_blkno_t radix,
blist_blkno_t skip,
blist_blkno_t blk
) {
blist_blkno_t i;
blist_blkno_t next_skip = (skip / BLIST_META_RADIX);
blist_blkno_t nblks = 0;
if (count == radix || scan->u.bmu_avail == 0) {
nblks = scan->u.bmu_avail;
scan->u.bmu_avail = 0;
scan->bm_bighint = count;
return nblks;
}
if (count > radix)
panic("blist_meta_fill: allocation too large");
if (scan->u.bmu_avail == radix) {
radix /= BLIST_META_RADIX;
for (i = 1; i <= skip; i += next_skip) {
if (scan[i].bm_bighint == (blist_blkno_t)-1)
break;
if (next_skip == 1) {
scan[i].u.bmu_bitmap = (blist_bitmap_t)-1;
scan[i].bm_bighint = BLIST_BMAP_RADIX;
} else {
scan[i].bm_bighint = radix;
scan[i].u.bmu_avail = radix;
}
}
} else {
radix /= BLIST_META_RADIX;
}
i = (allocBlk - blk) / radix;
blk += i * radix;
i = i * next_skip + 1;
while (i <= skip && blk < allocBlk + count) {
blist_blkno_t v;
v = blk + radix - allocBlk;
if (v > count)
v = count;
if (scan->bm_bighint == (blist_blkno_t)-1)
panic("blst_meta_fill: filling unexpected range");
if (next_skip == 1) {
nblks += blst_leaf_fill(&scan[i], allocBlk, v);
} else {
nblks += blst_meta_fill(&scan[i], allocBlk, v,
radix, next_skip - 1, blk);
}
count -= v;
allocBlk += v;
blk += radix;
i += next_skip;
}
scan->u.bmu_avail -= nblks;
return nblks;
}
static blist_blkno_t
blst_radix_init(blmeta_t *scan, blist_blkno_t radix, blist_blkno_t skip,
blist_blkno_t count)
{
blist_blkno_t i;
blist_blkno_t next_skip;
blist_blkno_t memindex = 0;
if (radix == BLIST_BMAP_RADIX) {
if (scan) {
scan->bm_bighint = 0;
scan->u.bmu_bitmap = 0;
}
return(memindex);
}
if (scan) {
scan->bm_bighint = 0;
scan->u.bmu_avail = 0;
}
radix /= BLIST_META_RADIX;
next_skip = (skip / BLIST_META_RADIX);
for (i = 1; i <= skip; i += next_skip) {
if (count >= radix) {
memindex = i + blst_radix_init(
((scan) ? &scan[i] : NULL),
radix,
next_skip - 1,
radix
);
count -= radix;
} else if (count > 0) {
memindex = i + blst_radix_init(
((scan) ? &scan[i] : NULL),
radix,
next_skip - 1,
count
);
count = 0;
} else {
if (scan)
scan[i].bm_bighint = (blist_blkno_t)-1;
break;
}
}
if (memindex < i)
memindex = i;
return(memindex);
}
#ifdef BLIST_DEBUG
static void
blst_radix_print(blmeta_t *scan, blist_blkno_t blk, blist_blkno_t radix,
blist_blkno_t skip, int tab)
{
blist_blkno_t i;
blist_blkno_t next_skip;
int lastState = 0;
if (radix == BLIST_BMAP_RADIX) {
printf(
"%*.*s(%0*" PRIx64 ",%" PRIu64
"): bitmap %0*" PRIx64 " big=%" PRIu64 "\n",
tab, tab, "",
sizeof(blk) * 2,
(uint64_t)blk,
(uint64_t)radix,
sizeof(scan->u.bmu_bitmap) * 2,
(uint64_t)scan->u.bmu_bitmap,
(uint64_t)scan->bm_bighint
);
return;
}
if (scan->u.bmu_avail == 0) {
printf(
"%*.*s(%0*" PRIx64 ",%" PRIu64") ALL ALLOCATED\n",
tab, tab, "",
sizeof(blk) * 2,
(uint64_t)blk,
(uint64_t)radix
);
return;
}
if (scan->u.bmu_avail == radix) {
printf(
"%*.*s(%0*" PRIx64 ",%" PRIu64 ") ALL FREE\n",
tab, tab, "",
sizeof(blk) * 2,
(uint64_t)blk,
(uint64_t)radix
);
return;
}
printf(
"%*.*s(%0*" PRIx64 ",%" PRIu64 "): subtree (%" PRIu64 "/%"
PRIu64 ") big=%" PRIu64 " {\n",
tab, tab, "",
sizeof(blk) * 2,
(uint64_t)blk,
(uint64_t)radix,
(uint64_t)scan->u.bmu_avail,
(uint64_t)radix,
(uint64_t)scan->bm_bighint
);
radix /= BLIST_META_RADIX;
next_skip = (skip / BLIST_META_RADIX);
tab += 4;
for (i = 1; i <= skip; i += next_skip) {
if (scan[i].bm_bighint == (blist_blkno_t)-1) {
printf(
"%*.*s(%0*" PRIx64 ",%" PRIu64 "): Terminator\n",
tab, tab, "",
sizeof(blk) * 2,
(uint64_t)blk,
(uint64_t)radix
);
lastState = 0;
break;
}
blst_radix_print(
&scan[i],
blk,
radix,
next_skip - 1,
tab
);
blk += radix;
}
tab -= 4;
printf(
"%*.*s}\n",
tab, tab, ""
);
}
#endif
#ifdef BLIST_DEBUG
int
main(int ac, char **av)
{
blist_blkno_t size = 1024;
int i;
blist_t bl;
for (i = 1; i < ac; ++i) {
const char *ptr = av[i];
if (*ptr != '-') {
size = strtol(ptr, NULL, 0);
continue;
}
ptr += 2;
fprintf(stderr, "Bad option: %s\n", ptr - 2);
exit(1);
}
bl = blist_create(size);
blist_free(bl, 0, size);
for (;;) {
char buf[1024];
uint64_t da = 0;
uint64_t count = 0;
printf("%" PRIu64 "/%" PRIu64 "/%" PRIu64 "> ",
(uint64_t)bl->bl_free,
(uint64_t)size,
(uint64_t)bl->bl_radix);
fflush(stdout);
if (fgets(buf, sizeof(buf), stdin) == NULL)
break;
switch(buf[0]) {
case 'r':
if (sscanf(buf + 1, "%" SCNu64, &count) == 1) {
blist_resize(&bl, count, 1);
} else {
printf("?\n");
}
case 'p':
blist_print(bl);
break;
case 'a':
if (sscanf(buf + 1, "%" SCNu64, &count) == 1) {
blist_blkno_t blk = blist_alloc(bl, count);
printf(" R=%0*" PRIx64 "\n",
sizeof(blk) * 2,
(uint64_t)blk);
} else {
printf("?\n");
}
break;
case 'f':
if (sscanf(buf + 1, "%" SCNx64 " %" SCNu64,
&da, &count) == 2) {
blist_free(bl, da, count);
} else {
printf("?\n");
}
break;
case 'l':
if (sscanf(buf + 1, "%" SCNx64 " %" SCNu64,
&da, &count) == 2) {
printf(" n=%" PRIu64 "\n",
(uint64_t)blist_fill(bl, da, count));
} else {
printf("?\n");
}
break;
case '?':
case 'h':
puts(
"p -print\n"
"a %d -allocate\n"
"f %x %d -free\n"
"l %x %d -fill\n"
"r %d -resize\n"
"h/? -help"
);
break;
default:
printf("?\n");
break;
}
}
return(0);
}
void
panic(const char *ctl, ...)
{
va_list va;
va_start(va, ctl);
vfprintf(stderr, ctl, va);
fprintf(stderr, "\n");
va_end(va);
exit(1);
}
#endif