nexoniarz/SchOS
nexoniarz/SchOS
Watch
1
0
Fork
You've already forked SchOS
 0
Code Issues Pull requests Projects Releases 1 Packages Wiki Activity Actions

kalloc: never-used memory stays a range; 256 GB; free is a counter

Browse source 
Every page of RAM went onto the free list at boot, which means writing
each one -- a link, and kfree()'s 4 KB of junk -- before the first
process ran: 64 GB of memset on a 64 GB machine to say it was free.
Memory nobody has used yet is kept as a cursor and an end per region
now, and kalloc() takes from those once the list of given-back pages
is empty. A page is first touched by whoever asked for it.

That makes the cap on RAM cheap to raise, and PHYSTOP_MAX goes from 64
to 256 GB, which still leaves 248 GB of the kernel PDPT for device
windows. MEMREGION_MAX goes from 32 to 64 for the firmware maps that
cut RAM into many pieces.

kalloc_free_bytes() walked the whole free list under the allocator's
lock -- two million links on 8 GB, every second top or fastfetch was
running, with every CPU that wanted a page waiting behind it. It reads
a counter now. And the buffer cache, which keeps what it grows to, is
reported as Buffers in /proc/meminfo and sysinfo(2) rather than as
memory used by nothing.

Checked under QEMU at 1, 2 and 6 GB: MemFree right, zsh splitting
200000 words, fork loops, a 74 MB write.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
This commit is contained in:
Nexoniarz 2026-10-23 14:05:40 +02:00
commit b70e36e0a2
4 changed files with 106 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

12
include/memlayout.h
View file

@ -27,7 +27,7 @@ extern uint64 phystop;
// PHYSTOP still means what it always did -- the top of what kalloc may hand
// out, and the bound kfree() rejects pages above -- so every existing call
// site is unaffected.
#define MEMREGION_MAX 32
#define MEMREGION_MAX 64
#ifndef __ASSEMBLER__
struct memregion {
@ -57,10 +57,14 @@ extern uint64 lowtop;
//
// kvmalloc() now maps as many 1 GB pages as the machine has RAM, so the limit
// is the page table rather than a constant: the kernel PDPT has 512 entries
// and ioremap takes what the direct map does not, so 64 GB of RAM still leaves
// 448 GB of device window. A machine with more than this gets the rest
// and ioremap takes what the direct map does not, so 256 GB of RAM still
// leaves 248 GB of device window. A machine with more than this gets the rest
// ignored, with a line saying so.
#define PHYSTOP_MAX (64ULL << 30)
//
// It was 64 GB while handing memory to the allocator meant writing every page
// of it at boot. kalloc.c keeps never-used memory as ranges now, so a bigger
// machine costs nothing until its memory is used.
#define PHYSTOP_MAX (256ULL << 30)
// What the *loader's* page tables cover, and therefore the ceiling on anything
// that runs before kvmalloc() builds the kernel's own.

103
kernel/core/kalloc.c
View file

@ -19,12 +19,53 @@ struct run {
struct run *next;
};
// Free memory is two things: pages that have been allocated and given back,
// on a list through their first word, and ranges that have never been handed
// out at all, as a cursor and an end. kalloc() takes from the list first.
//
// Everything used to go on the list at boot, which means writing every page
// of RAM -- a pointer and, from kfree(), 4 KB of junk -- before the first
// process ran. On a 64 GB machine that is 64 GB of memset to say "this is
// free". A range costs sixteen bytes however large it is, and a page is first
// touched by whoever asked for it.
//
// `nfree` counts both, so the free total is a read rather than a walk of
// millions of links under the lock every time top refreshes.
#define NFRESH (2 * MEMREGION_MAX)
struct fresh {
addr_t next, end; // physical, page aligned
};
struct {
struct spinlock lock;
int use_lock;
struct run *freelist;
struct fresh fresh[NFRESH];
int nfresh, cur; // ranges, and the first that may have pages
uint64 nfree; // pages, list and ranges together
} kmem;
// Before the allocator has a lock, or under it.
static void
add_fresh(addr_t lo, addr_t hi)
{
lo = PGROUNDUP(lo);
hi &= ~(addr_t)(PGSIZE - 1);
if (lo >= hi)
return;
if (kmem.nfresh == NFRESH) {
// Not reachable with MEMREGION_MAX regions per pass; but if it were, the
// old way still works.
freerange(P2V(lo), P2V(hi));
return;
}
kmem.fresh[kmem.nfresh].next = lo;
kmem.fresh[kmem.nfresh].end = hi;
kmem.nfresh++;
kmem.nfree += (hi - lo) / PGSIZE;
}
// Catch kmem.freelist corruption at the point it's touched, rather than
// letting a bad pointer be silently returned by kalloc() and crash whatever
// code dereferences it minutes later on an unrelated call stack.
@ -60,7 +101,7 @@ kinit1(void *vstart, void *vend)
// synthetic region behind for exactly this reason, so this is belt and
// braces rather than a path anything takes.
if(nmem_usable == 0){
freerange(vstart, vend);
add_fresh(V2P((addr_t)vstart), V2P((addr_t)vend));
return;
}
@ -78,7 +119,7 @@ kinit1(void *vstart, void *vend)
if(lo >= hi)
continue;
freerange(P2V(lo), P2V(hi));
add_fresh(lo, hi);
}
}
@ -110,8 +151,16 @@ mem_dump(struct sbuf *out, const char *arg)
sb_puts(out, " upward; PHYSTOP is 0x");
sb_puthex(out, (addr_t)PHYSTOP, 16);
sb_putc(out, '\n');
sb_kv(out, "free ", n, "pages");
sb_kv(out, " ", n * PGSIZE / 1024, "KB");
sb_kv(out, "listed ", n, "pages");
sb_kv(out, "free ", kmem.nfree, "pages");
sb_kv(out, " ", kmem.nfree * PGSIZE / 1024, "KB");
for (int i = 0; i < kmem.nfresh; i++) {
sb_puts(out, " never used 0x");
sb_puthex(out, kmem.fresh[i].next, 12);
sb_puts(out, "-0x");
sb_puthex(out, kmem.fresh[i].end, 12);
sb_putc(out, '\n');
}
if (looped)
sb_puts(out, "\n the free list did not end -- it has a CYCLE in it\n");
}
@ -154,6 +203,14 @@ mem_whose(struct sbuf *out, struct whose *w)
if (kmem.use_lock)
release(&kmem.lock);
for (int i = 0; !found && i < kmem.nfresh; i++)
if (w->phys >= kmem.fresh[i].next && w->phys < kmem.fresh[i].end) {
w->owners++;
w->free_owners++;
sb_puts(out, " kalloc FREE -- never handed out yet\n");
return;
}
if (found) {
w->owners++;
w->free_owners++;
@ -189,7 +246,11 @@ kinit_high(void)
if (lo >= hi)
continue;
freerange(P2V(lo), P2V(hi));
if (kmem.use_lock)
acquire(&kmem.lock);
add_fresh(lo, hi);
if (kmem.use_lock)
release(&kmem.lock);
added += hi - lo;
}
@ -236,6 +297,7 @@ kfree(char *v)
r = (struct run*)v;
r->next = kmem.freelist;
kmem.freelist = r;
kmem.nfree++;
if(kmem.use_lock)
release(&kmem.lock);
}
@ -252,27 +314,32 @@ kalloc(void)
acquire(&kmem.lock);
check_run(kmem.freelist, "kalloc entry");
r = kmem.freelist;
if(r)
if(r){
kmem.freelist = r->next;
} else {
while(kmem.cur < kmem.nfresh &&
kmem.fresh[kmem.cur].next >= kmem.fresh[kmem.cur].end)
kmem.cur++;
if(kmem.cur < kmem.nfresh){
r = (struct run*)P2V(kmem.fresh[kmem.cur].next);
kmem.fresh[kmem.cur].next += PGSIZE;
}
}
if(r)
kmem.nfree--;
if(kmem.use_lock)
release(&kmem.lock);
return (char*)r;
}
// Total free physical memory, in bytes. Walks the free list; only /proc
// readers call it, so the O(n) scan is not on any hot path.
// Total free physical memory, in bytes.
//
// This walked the free list under the allocator's lock: two million links on
// 8 GB, cache misses all, every second that top or a status bar was running --
// and every CPU that wanted a page in that time waited for it. One word now.
uint64
kalloc_free_bytes(void)
{
struct run *r;
uint64 n = 0;
if(kmem.use_lock)
acquire(&kmem.lock);
for(r = kmem.freelist; r; r = r->next)
n += PGSIZE;
if(kmem.use_lock)
release(&kmem.lock);
return n;
return kmem.nfree * PGSIZE;
}

6
kernel/dev/procfs.c
View file

@ -88,7 +88,11 @@ gen_meminfo(struct sbuf *sb)
sb_kv(sb, "MemTotal: ", total, "kB");
sb_kv(sb, "MemFree: ", freek, "kB");
sb_kv(sb, "MemAvailable: ", freek, "kB");
sb_kv(sb, "Buffers: ", 0, "kB");
// The buffer cache, which grows to 1/32 of RAM and keeps it: without this
// line that showed up in free(1) as memory "used" by nothing.
uint nbuf, maxbuf;
bcache_size(&nbuf, &maxbuf);
sb_kv(sb, "Buffers: ", (uint64)nbuf * (BSIZE / 1024), "kB");
sb_kv(sb, "Cached: ", 0, "kB");
sb_kv(sb, "SwapTotal: ", 0, "kB");
sb_kv(sb, "SwapFree: ", 0, "kB");

8
kernel/sys/sys_proc.c
View file

@ -12,6 +12,10 @@
#include "errno.h"
#include "acpi.h"
#include "fat.h"
#include "spinlock.h"
#include "sleeplock.h"
#include "fs.h"
#include "buf.h"
addr_t sys_fork(void) { return fork(); }
addr_t sys_wait(void) { return wait(); }
@ -522,6 +526,10 @@ addr_t sys_sysinfo(void)
// then PHYSTOP, an address, which counted the PCI hole as memory.
si->totalram = ram_total_bytes();
si->freeram = kalloc_free_bytes();
// The buffer cache, so free(1) does not count it as used by nothing.
uint nbuf, maxbuf;
bcache_size(&nbuf, &maxbuf);
si->bufferram = (uint64)nbuf * BSIZE;
si->mem_unit = 1;
si->procs = (uint16)proc_num();
return 0;