ref: a73b2d8351853bdfde84b2792b8b231472271284
dir: /lib/std/alloc.myr/
use "die"
use "extremum"
use "types"
use "units"
use "syswrap"
use "memops"
/*
The allocator implementation here is based on Bonwick's slab allocator.
For small allocations (up to Bktmax), it works by requesting large,
power of two aligned chunks from the operating system, and breaking
them into a linked list of equal sized chunks. Allocations are then
satisfied by taking the head of the list of chunks. Empty slabs
are removed from the freelist.
The data structure looks something like this:
Bkts:
[16 byte] -> [slab hdr | chunk -> chunk -> chunk] -> [slab hdr | chunk -> chunk -> chunk]
[32 byte] -> Zslab
[64 byte] -> [slab hdr | chunk -> chunk]
...
[32k byte] -> ...
Large allocations are simply satisfied by mmap().
*/
pkg std =
generic alloc : ( -> @a#)
generic zalloc : ( -> @a#)
generic free : (v:@a# -> void)
generic slalloc : (len : size -> @a[:])
generic slzalloc : (len : size -> @a[:])
generic slgrow : (sl : @a[:]#, len : size -> @a[:])
generic slzgrow : (sl : @a[:]#, len : size -> @a[:])
generic slfree : (sl : @a[:] -> void)
const bytealloc : (sz:size -> byte#)
const zbytealloc : (sz:size -> byte#)
const bytefree : (m:byte#, sz:size -> void)
;;
/* null pointers. only used internally. */
const Zslab = 0 castto(slab#)
const Zchunk = 0 castto(chunk#)
const Zsliceptr = 0 castto(byte#)
const Slabsz = 1*MiB /* 1 meg slabs */
const Cachemax = 16 /* maximum number of slabs in the cache */
const Bktmax = 32*KiB /* Slabsz / 8; a balance. */
const Pagesz = 4*KiB
const Align = 16 /* minimum allocation alignment */
var buckets : bucket[32] /* excessive */
extern const put : (str : byte[:], args : ... -> void)
type slheader = struct
cap : size /* capacity in bytes */
magic : size /* magic check value */
;;
type bucket = struct
sz : size /* aligned size */
nper : size /* max number of elements per slab */
slabs : slab# /* partially filled or free slabs */
cache : slab# /* cache of empty slabs, to prevent thrashing */
ncache : size /* size of cache */
;;
type slab = struct
head : byte# /* head of virtual addresses, so we don't leak address space */
next : slab# /* the next slab on the chain */
freehd : chunk# /* the nodes we're allocating */
nfree : size /* the number of free nodes */
;;
type chunk = struct /* NB: must be smaller than sizeof(slab) */
next : chunk# /* the next chunk in the free list */
;;
const __init__ = {
for var i = 0; i < buckets.len && (Align << i) <= Bktmax; i++
bktinit(&buckets[i], Align << i)
;;
}
/* Allocates an object of type @a, returning a pointer to it. */
generic alloc = {-> @a#
-> bytealloc(sizeof(@a)) castto(@a#)
}
generic zalloc = {-> @a#
-> zbytealloc(sizeof(@a)) castto(@a#)
}
/* Frees a value of type @a */
generic free = {v:@a# -> void
bytefree(v castto(byte#), sizeof(@a))
}
/* allocates a slice of 'len' elements. */
generic slalloc = {len
var p, sz
if len == 0
-> [][:]
;;
sz = len*sizeof(@a) + align(sizeof(slheader), Align)
p = bytealloc(sz)
p = inithdr(p, sz)
-> (p castto(@a#))[0:len]
}
generic slzalloc = {len
var p, sz
if len == 0
-> [][:]
;;
sz = len*sizeof(@a) + align(sizeof(slheader), Align)
p = zbytealloc(sz)
p = inithdr(p, sz)
-> (p castto(@a#))[0:len]
}
const inithdr = {p, sz
var phdr, prest
phdr = p castto(slheader#)
phdr.cap = allocsz(sz) - align(sizeof(slheader), Align)
phdr.magic = (0xdeadbeefbadf00d castto(size))
prest = (p castto(size)) + align(sizeof(slheader), Align)
-> prest castto(byte#)
}
const checkhdr = {p
var phdr, addr
addr = p castto(size)
addr -= align(sizeof(slheader), Align)
phdr = addr castto(slheader#)
iassert(phdr.magic == (0xdeadbeefbadf00d castto(size)), "corrupt memory\n")
}
/* Frees a slice */
generic slfree = {sl
var head
if sl castto(byte#) == Zsliceptr
-> void
;;
checkhdr(sl castto(byte#))
head = (sl castto(byte#)) castto(size)
head -= align(sizeof(slheader), Align)
bytefree(head castto(byte#), slcap(sl castto(byte#)))
}
/* Grows a slice */
generic slgrow = {sl : @a[:]#, len
var cap
var new
var n
/* if the slice doesn't need a bigger bucket, we don't need to realloc. */
cap = 0
if sl# castto(byte#) != Zsliceptr
cap = slcap(sl# castto(byte#))
;;
if cap >= allocsz(len*sizeof(@a))
sl# = (sl# castto(@a#))[:len]
-> sl#
;;
/* grow in factors of 1.5 */
cap = max(Align, cap)
while cap < len
cap += (cap >> 1)
;;
new = slalloc(cap)
n = min(len, sl#.len)
for var i = 0; i < n; i++
new[i] = sl#[i]
;;
if sl#.len > 0
slfree(sl#)
;;
sl# = new[:len]
-> sl#
}
/* Grows a slice, filling new entries with zero bytes */
generic slzgrow = {sl : @a[:]#, len
var oldlen
var base
oldlen = sl#.len
slgrow(sl, len)
base = sl# castto(byte#) castto(intptr)
if oldlen < len
memfill(sl#[oldlen:] castto(byte#), 0, (len - oldlen)*sizeof(@a))
;;
-> sl#
}
const slcap = {p
var phdr
phdr = (p castto(size)) - align(sizeof(slheader), Align) castto(slheader#)
-> phdr.cap
}
const zbytealloc = {sz
var p
p = bytealloc(sz)
memfill(p, 0, sz)
-> p
}
/* Allocates a blob that is 'sz' bytes long. Dies if the allocation fails */
const bytealloc = {sz
var bkt, p
if (sz <= Bktmax)
bkt = &buckets[bktnum(sz)]
p = bktalloc(bkt)
else
p = getmem(sz)
if p == Failmem
die("could not get memory\n")
;;
;;
-> p
}
/* frees a blob that is 'sz' bytes long. */
const bytefree = {p, sz
var bkt
memfill(p, 0xa8, sz)
if (sz < Bktmax)
bkt = &buckets[bktnum(sz)]
bktfree(bkt, p)
else
freemem(p, sz)
;;
}
/* Sets up a single empty bucket */
const bktinit = {b, sz
b.sz = align(sz, Align)
b.nper = (Slabsz - sizeof(slab))/b.sz
b.slabs = Zslab
b.cache = Zslab
b.ncache = 0
}
/* Creates a slab for bucket 'bkt', and fills the chunk list */
const mkslab = {bkt
var p, s
var b, bnext
var off /* offset of chunk head */
if bkt.ncache > 0
s = bkt.cache
bkt.cache = s.next
bkt.ncache--
;;
/*
tricky: we need power of two alignment, so we allocate double the
needed size, chop off the unaligned ends, and waste the address
space. Since the OS is "smart enough", this shouldn't actually
cost us memory, and 64 bits of address space means that we're not
going to have issues with running out of address space for a
while. On a 32 bit system this would be a bad idea.
*/
p = getmem(Slabsz*2)
if p == Failmem
die("Unable to mmap")
;;
s = align(p castto(size), Slabsz) castto(slab#)
s.head = p
s.nfree = bkt.nper
/* skip past the slab header */
off = align(sizeof(slab), Align)
bnext = nextchunk(s castto(chunk#), off)
s.freehd = bnext
for var i = 0; i < bkt.nper; i++
b = bnext
bnext = nextchunk(b, bkt.sz)
b.next = bnext
;;
b.next = Zchunk
-> s
}
/*
Allocates a node from bucket 'bkt', crashing if the
allocation cannot be satisfied. Will create a new slab
if there are no slabs on the freelist.
*/
const bktalloc = {bkt
var s
var b
/* find a slab */
s = bkt.slabs
if s == Zslab
s = mkslab(bkt)
if s == Zslab
die("No memory left")
;;
bkt.slabs = s
;;
/* grab the first chunk on the slab */
b = s.freehd
s.freehd = b.next
s.nfree--
if s.nfree == 0
bkt.slabs = s.next
s.next = Zslab
;;
-> b castto(byte#)
}
/*
Frees a chunk of memory 'm' into bucket 'bkt'.
Assumes that the memory already came from a slab
that was created for bucket 'bkt'. Will crash
if this is not the case.
*/
const bktfree = {bkt, m
var s, b
s = mtrunc(m, Slabsz) castto(slab#)
b = m castto(chunk#)
if s.nfree == 0
s.next = bkt.slabs
bkt.slabs = s
elif s.nfree == bkt.nper
/*
HACK HACK HACK: if we can't unmap, keep an infinite cache per slab size.
We should solve this better somehow.
*/
if bkt.ncache < Cachemax || !Canunmap
s.next = bkt.cache
bkt.cache = s
else
/* we mapped 2*Slabsz so we could align it,
so we need to unmap the same */
freemem(s.head, Slabsz*2)
;;
;;
s.nfree++
b.next = s.freehd
s.freehd = b
}
/*
Finds the correct bucket index to allocate from
for allocations of size 'sz'
*/
const bktnum = {sz
var bktsz
bktsz = Align
for var i = 0; bktsz <= Bktmax; i++
if bktsz >= sz
-> i
;;
bktsz *= 2
;;
die("Size does not match any buckets")
}
/*
returns the actual size we allocated for a given
size request
*/
const allocsz = {sz
var bktsz
if sz <= Bktmax
bktsz = Align
for var i = 0; bktsz <= Bktmax; i++
if bktsz >= sz
-> bktsz
;;
bktsz *= 2
;;
else
-> align(sz, Pagesz)
;;
die("Size does not match any buckets")
}
/*
aligns a size to a requested alignment.
'align' must be a power of two
*/
const align = {v, align
-> (v + align - 1) & ~(align - 1)
}
/*
chunks are variable sizes, so we can't just
index to get to the next one
*/
const nextchunk = {b, sz : size
-> ((b castto(intptr)) + (sz castto(intptr))) castto(chunk#)
}
/*
truncates a pointer to 'align'. 'align' must
be a power of two.
*/
const mtrunc = {m, align
-> ((m castto(intptr)) & ~((align castto(intptr)) - 1)) castto(byte#)
}