Merge branch 'main' into feat/gui-desktop

include/kernel/futex.h

@@ -0,0 +1,62 @@
+#ifndef _KERNEL_FUTEX_H
+#define _KERNEL_FUTEX_H
+
+#include <stdint.h>
+#include <stddef.h>
+#include <stdbool.h>
+#include <kernel/wait.h>
+
+// Fast Userspace muTEX. A futex is an ordinary 32-bit word that stays entirely
+// in user memory while uncontended; the kernel is only consulted when a thread
+// actually has to block, at which point the word's address is hashed onto one
+// of FUTEX_BUCKETS wait queues.
+
+#define FUTEX_BUCKETS 32
+
+typedef enum {
+    FUTEX_OP_WAIT = 0,
+    FUTEX_OP_WAKE = 1,
+    FUTEX_OP_REQUEUE = 2,
+    FUTEX_OP_WAKE_ALL = 3
+} futex_op_t;
+
+typedef struct {
+    uint64_t waits;
+    uint64_t wakes;
+    uint64_t requeues;
+    uint64_t spurious;
+    uint64_t timeouts;
+    uint32_t active_waiters;
+} futex_stats_t;
+
+void futex_init(void);
+
+// Sleep on `uaddr` only while *uaddr still equals `expected`. Returns 0 when
+// woken, 1 on timeout, -1 if the value changed before parking (the caller
+// should retry its fast path).
+int futex_wait(volatile uint32_t* uaddr, uint32_t expected, uint64_t timeout_ticks);
+
+// Wake up to `count` waiters parked on `uaddr`. Returns how many were woken.
+int futex_wake(volatile uint32_t* uaddr, int count);
+
+// Move waiters from `uaddr1` onto `uaddr2` without waking them (used to avoid
+// the thundering herd when a condvar hands off to a mutex).
+int futex_requeue(volatile uint32_t* uaddr1, volatile uint32_t* uaddr2, int count);
+
+futex_stats_t futex_get_stats(void);
+void          futex_dump(void);
+
+// ---------------------------------------------------------------------------
+// A futex-backed mutex, for kernel code that wants sleeping mutual exclusion.
+// ---------------------------------------------------------------------------
+typedef struct {
+    volatile uint32_t state; // 0 = unlocked, 1 = locked, 2 = locked + waiters
+    char              name[24];
+} fmutex_t;
+
+void fmutex_init(fmutex_t* m, const char* name);
+void fmutex_lock(fmutex_t* m);
+bool fmutex_trylock(fmutex_t* m);
+void fmutex_unlock(fmutex_t* m);
+
+#endif // _KERNEL_FUTEX_H

include/kernel/ktest.h

@@ -0,0 +1,81 @@
+#ifndef _KERNEL_KTEST_H
+#define _KERNEL_KTEST_H
+
+#include <stdint.h>
+#include <stddef.h>
+#include <stdbool.h>
+
+// In-kernel self-test harness (KUnit-style). Suites register a list of cases;
+// `ktest` from the LazyBox shell runs them and reports pass/fail per assertion.
+
+#define KTEST_MAX_SUITES 16
+#define KTEST_MAX_CASES  24
+
+typedef struct ktest_context {
+    const char* suite;
+    const char* test;
+    uint32_t    assertions;
+    uint32_t    failures;
+    bool        aborted;
+} ktest_context_t;
+
+typedef void (*ktest_fn_t)(ktest_context_t* ctx);
+
+typedef struct {
+    const char* name;
+    ktest_fn_t  func;
+} ktest_case_t;
+
+typedef struct {
+    const char*        name;
+    const ktest_case_t* cases;
+    size_t             case_count;
+} ktest_suite_t;
+
+typedef struct {
+    uint32_t suites_run;
+    uint32_t tests_run;
+    uint32_t tests_passed;
+    uint32_t tests_failed;
+    uint32_t assertions;
+    uint32_t assertion_failures;
+    uint64_t elapsed_ticks;
+} ktest_result_t;
+
+void ktest_init(void);
+int  ktest_register_suite(const ktest_suite_t* suite);
+
+// Run every suite, or one by name. Returns the number of failed tests.
+int  ktest_run_all(ktest_result_t* result_out);
+int  ktest_run_suite(const char* name, ktest_result_t* result_out);
+void ktest_list_suites(void);
+
+// Assertion primitives used by the macros below.
+void ktest_report(ktest_context_t* ctx, bool passed, const char* expr,
+                  const char* file, int line, const char* detail);
+
+#define KTEST_ASSERT(ctx, cond) \
+    ktest_report((ctx), (cond), #cond, __FILE__, __LINE__, NULL)
+
+#define KTEST_ASSERT_EQ(ctx, a, b) \
+    do { \
+        long long _a = (long long)(a); \
+        long long _b = (long long)(b); \
+        ktest_expect_eq((ctx), _a, _b, #a " == " #b, __FILE__, __LINE__); \
+    } while (0)
+
+#define KTEST_ASSERT_STREQ(ctx, a, b) \
+    ktest_expect_streq((ctx), (a), (b), #a " == " #b, __FILE__, __LINE__)
+
+#define KTEST_ASSERT_NOT_NULL(ctx, p) \
+    ktest_report((ctx), (p) != NULL, #p " != NULL", __FILE__, __LINE__, NULL)
+
+void ktest_expect_eq(ktest_context_t* ctx, long long a, long long b,
+                     const char* expr, const char* file, int line);
+void ktest_expect_streq(ktest_context_t* ctx, const char* a, const char* b,
+                        const char* expr, const char* file, int line);
+
+// Built-in suites covering the core data structures and subsystems.
+void ktest_register_builtin_suites(void);
+
+#endif // _KERNEL_KTEST_H

include/kernel/rcu.h

@@ -0,0 +1,58 @@
+#ifndef _KERNEL_RCU_H
+#define _KERNEL_RCU_H
+
+#include <stdint.h>
+#include <stddef.h>
+#include <stdbool.h>
+
+// Tiny RCU (Read-Copy-Update).
+//
+// Readers are free: rcu_read_lock() only bumps a per-CPU nesting depth, so a
+// lookup never contends with an update. Writers publish a new version with
+// rcu_assign_pointer() and defer reclamation of the old one until every reader
+// that could still hold a reference has finished — a "grace period".
+
+#define RCU_MAX_CALLBACKS 64
+
+typedef void (*rcu_callback_t)(void* arg);
+
+typedef struct {
+    uint64_t grace_periods;
+    uint64_t callbacks_queued;
+    uint64_t callbacks_invoked;
+    uint64_t read_locks;
+    uint32_t nesting_depth;
+    uint32_t pending;
+} rcu_stats_t;
+
+void rcu_init(void);
+
+void rcu_read_lock(void);
+void rcu_read_unlock(void);
+bool rcu_read_lock_held(void);
+
+// Block until every reader active at call time has left its critical section.
+void rcu_synchronize(void);
+
+// Queue `func(arg)` to run once the current grace period elapses.
+int  call_rcu(rcu_callback_t func, void* arg);
+
+// Free `ptr` after a grace period (the common case for call_rcu).
+int  kfree_rcu(void* ptr);
+
+// Advance the state machine; called from the timer tick and from rcu_synchronize.
+void rcu_process_callbacks(void);
+
+uint64_t   rcu_get_grace_period(void);
+rcu_stats_t rcu_get_stats(void);
+void       rcu_dump(void);
+
+// Publish/consume helpers. The barriers keep the compiler from reordering the
+// initialisation of an object past the store that makes it visible.
+#define rcu_assign_pointer(p, v) \
+    do { __atomic_store_n(&(p), (v), __ATOMIC_RELEASE); } while (0)
+
+#define rcu_dereference(p) \
+    __atomic_load_n(&(p), __ATOMIC_ACQUIRE)
+
+#endif // _KERNEL_RCU_H

include/kernel/wait.h

@@ -0,0 +1,65 @@
+#ifndef _KERNEL_WAIT_H
+#define _KERNEL_WAIT_H
+
+#include <stdint.h>
+#include <stddef.h>
+#include <stdbool.h>
+#include <lib/list.h>
+#include <kernel/sync.h>
+#include <kernel/task.h>
+
+// Wait queues and completions.
+//
+// SUB-OS runs a cooperative round-robin scheduler, so a sleeper parks itself
+// on the queue, marks its task BLOCKED and yields until a waker flips the
+// entry's `woken` flag. Timed waits fall back to the PIT tick counter.
+
+typedef struct wait_entry {
+    task_t*          task;
+    volatile bool    woken;
+    struct list_head node;
+} wait_entry_t;
+
+typedef struct wait_queue {
+    char             name[24];
+    struct list_head waiters;
+    spinlock_t       lock;
+    uint64_t         wakeups;
+    uint64_t         sleeps;
+    uint32_t         nr_waiting;
+} wait_queue_t;
+
+void wait_queue_init(wait_queue_t* wq, const char* name);
+
+// Park the caller until woken. Returns 0 on wake, -1 if the queue is full.
+int  wait_event(wait_queue_t* wq);
+
+// Park the caller for at most `timeout_ticks` PIT ticks (100 Hz).
+// Returns 0 if woken, 1 if the wait timed out.
+int  wait_event_timeout(wait_queue_t* wq, uint64_t timeout_ticks);
+
+// Wake the longest-waiting sleeper, or every sleeper. Returns the count woken.
+int  wake_up(wait_queue_t* wq);
+int  wake_up_all(wait_queue_t* wq);
+
+uint32_t wait_queue_waiters(const wait_queue_t* wq);
+
+// ---------------------------------------------------------------------------
+// Completions: one-shot "this is done" barriers built on a wait queue.
+// ---------------------------------------------------------------------------
+typedef struct completion {
+    wait_queue_t  wq;
+    volatile bool done;
+} completion_t;
+
+void completion_init(completion_t* c, const char* name);
+void wait_for_completion(completion_t* c);
+int  wait_for_completion_timeout(completion_t* c, uint64_t timeout_ticks);
+void complete(completion_t* c);
+void complete_all(completion_t* c);
+bool completion_done(const completion_t* c);
+
+void wait_subsystem_init(void);
+void wait_dump_stats(void);
+
+#endif // _KERNEL_WAIT_H

include/lib/hashtable.h

@@ -0,0 +1,50 @@
+#ifndef _LIB_HASHTABLE_H
+#define _LIB_HASHTABLE_H
+
+#include <stddef.h>
+#include <stdint.h>
+#include <stdbool.h>
+
+// Separately-chained hash table with FNV-1a hashing and automatic growth.
+// Keys are either NUL-terminated strings (owned copies) or raw 64-bit ids.
+
+typedef struct hnode {
+    uint64_t      hash;
+    char*         skey;   // NULL for integer-keyed entries
+    uint64_t      ikey;
+    void*         value;
+    struct hnode* next;
+} hnode_t;
+
+typedef struct hashtable {
+    hnode_t** buckets;
+    size_t    bucket_count;
+    size_t    entries;
+    size_t    collisions;
+    bool      string_keys;
+} hashtable_t;
+
+uint64_t hash_fnv1a(const void* data, size_t len);
+uint64_t hash_string(const char* s);
+uint64_t hash_u64(uint64_t v);
+
+hashtable_t* hashtable_create(size_t initial_buckets, bool string_keys);
+void         hashtable_destroy(hashtable_t* ht);
+
+int   hashtable_put(hashtable_t* ht, const char* key, void* value);
+void* hashtable_get(const hashtable_t* ht, const char* key);
+bool  hashtable_remove(hashtable_t* ht, const char* key);
+
+int   hashtable_put_u64(hashtable_t* ht, uint64_t key, void* value);
+void* hashtable_get_u64(const hashtable_t* ht, uint64_t key);
+bool  hashtable_remove_u64(hashtable_t* ht, uint64_t key);
+
+size_t hashtable_size(const hashtable_t* ht);
+size_t hashtable_buckets(const hashtable_t* ht);
+size_t hashtable_collisions(const hashtable_t* ht);
+size_t hashtable_longest_chain(const hashtable_t* ht);
+
+typedef void (*hashtable_iter_fn)(const hnode_t* node, void* ctx);
+void hashtable_foreach(const hashtable_t* ht, hashtable_iter_fn fn, void* ctx);
+
+#endif // _LIB_HASHTABLE_H

include/lib/kfifo.h

@@ -0,0 +1,40 @@
+#ifndef _LIB_KFIFO_H
+#define _LIB_KFIFO_H
+
+#include <stddef.h>
+#include <stdint.h>
+#include <stdbool.h>
+#include <kernel/sync.h>
+
+// Power-of-two byte ring buffer with an unsigned index-wrap discipline,
+// mirroring the classic Linux kfifo. Producer and consumer indices are
+// free-running; masking happens only at access time.
+
+typedef struct kfifo {
+    uint8_t*  buffer;
+    size_t    size;      // Always a power of two
+    size_t    mask;      // size - 1
+    size_t    in;        // Free-running producer index
+    size_t    out;       // Free-running consumer index
+    bool      owns_buffer;
+    spinlock_t lock;
+} kfifo_t;
+
+int    kfifo_alloc(kfifo_t* fifo, size_t size);
+int    kfifo_init_static(kfifo_t* fifo, void* buffer, size_t size);
+void   kfifo_free(kfifo_t* fifo);
+void   kfifo_reset(kfifo_t* fifo);
+
+size_t kfifo_in(kfifo_t* fifo, const void* src, size_t len);
+size_t kfifo_out(kfifo_t* fifo, void* dst, size_t len);
+size_t kfifo_peek(const kfifo_t* fifo, void* dst, size_t len);
+
+size_t kfifo_len(const kfifo_t* fifo);
+size_t kfifo_avail(const kfifo_t* fifo);
+bool   kfifo_is_empty(const kfifo_t* fifo);
+bool   kfifo_is_full(const kfifo_t* fifo);
+
+int    kfifo_put_byte(kfifo_t* fifo, uint8_t byte);
+int    kfifo_get_byte(kfifo_t* fifo, uint8_t* byte_out);
+
+#endif // _LIB_KFIFO_H

include/lib/rbtree.h

@@ -0,0 +1,63 @@
+#ifndef _LIB_RBTREE_H
+#define _LIB_RBTREE_H
+
+#include <stddef.h>
+#include <stdint.h>
+#include <stdbool.h>
+
+// Linux-style Red-Black Tree: O(log n) ordered container used by the
+// CFS-style scheduler index, VMA lookup and the page cache radix index.
+
+#define RB_RED   0
+#define RB_BLACK 1
+
+struct rb_node {
+    struct rb_node* rb_parent;
+    struct rb_node* rb_left;
+    struct rb_node* rb_right;
+    int             rb_color;
+};
+
+struct rb_root {
+    struct rb_node* rb_node;
+};
+
+#define RB_ROOT_INIT { NULL }
+
+#define rb_entry(ptr, type, member) \
+    ((type*)((char*)(ptr) - offsetof(type, member)))
+
+static inline void rb_root_init(struct rb_root* root) {
+    root->rb_node = NULL;
+}
+
+static inline bool rb_empty(const struct rb_root* root) {
+    return root->rb_node == NULL;
+}
+
+// Attach a freshly allocated node into the tree, then recolour.
+static inline void rb_link_node(struct rb_node* node, struct rb_node* parent,
+                                struct rb_node** rb_link) {
+    node->rb_parent = parent;
+    node->rb_color  = RB_RED;
+    node->rb_left   = NULL;
+    node->rb_right  = NULL;
+    *rb_link = node;
+}
+
+void rb_insert_color(struct rb_node* node, struct rb_root* root);
+void rb_erase(struct rb_node* node, struct rb_root* root);
+
+struct rb_node* rb_first(const struct rb_root* root);
+struct rb_node* rb_last(const struct rb_root* root);
+struct rb_node* rb_next(const struct rb_node* node);
+struct rb_node* rb_prev(const struct rb_node* node);
+
+size_t rb_count(const struct rb_root* root);
+int    rb_black_height(const struct rb_root* root);
+bool   rb_validate(const struct rb_root* root);
+
+#define rb_for_each(pos, root) \
+    for (pos = rb_first(root); pos != NULL; pos = rb_next(pos))
+
+#endif // _LIB_RBTREE_H

include/mm/page_cache.h

@@ -0,0 +1,63 @@
+#ifndef _MM_PAGE_CACHE_H
+#define _MM_PAGE_CACHE_H
+
+#include <stdint.h>
+#include <stddef.h>
+#include <stdbool.h>
+
+// Write-back page cache sitting between the VFS and the block layer.
+//
+// Every entry caches PAGE_CACHE_BLOCK_SIZE bytes of one block device, keyed by
+// (device, page index). Lookups go through a hash table; eviction follows a
+// CLOCK approximation of LRU, and dirty pages are written back before reuse.
+
+#define PAGE_CACHE_BLOCK_SIZE   4096
+#define PAGE_CACHE_SECTORS      (PAGE_CACHE_BLOCK_SIZE / 512)
+#define PAGE_CACHE_MAX_PAGES    64
+
+typedef struct page_cache_entry {
+    char      dev_name[32];
+    uint64_t  page_index;      // Offset in PAGE_CACHE_BLOCK_SIZE units
+    uint8_t*  data;
+    bool      valid;
+    bool      dirty;
+    bool      referenced;      // CLOCK second-chance bit
+    uint32_t  pin_count;
+    uint64_t  last_access;
+    uint64_t  hits;
+} page_cache_entry_t;
+
+typedef struct {
+    uint64_t hits;
+    uint64_t misses;
+    uint64_t evictions;
+    uint64_t writebacks;
+    uint64_t read_ios;
+    uint64_t write_ios;
+    uint32_t resident_pages;
+    uint32_t dirty_pages;
+    size_t   memory_bytes;
+} page_cache_stats_t;
+
+void page_cache_init(void);
+
+// Read/write through the cache. Offsets and lengths are arbitrary; the cache
+// splits them across pages internally. Both return bytes transferred.
+size_t page_cache_read(const char* dev_name, uint64_t offset, void* buf, size_t len);
+size_t page_cache_write(const char* dev_name, uint64_t offset, const void* buf, size_t len);
+
+// Borrow a page directly (pinned until page_cache_release).
+page_cache_entry_t* page_cache_get(const char* dev_name, uint64_t page_index);
+void                page_cache_release(page_cache_entry_t* entry);
+void                page_cache_mark_dirty(page_cache_entry_t* entry);
+
+int  page_cache_sync(void);                       // Flush every dirty page
+int  page_cache_sync_dev(const char* dev_name);   // Flush one device
+int  page_cache_invalidate(const char* dev_name); // Drop clean pages
+void page_cache_drop_all(void);
+
+page_cache_stats_t page_cache_get_stats(void);
+uint32_t           page_cache_hit_percent(void);
+void               page_cache_dump(void);
+
+#endif // _MM_PAGE_CACHE_H

include/userland/coreutils.h

@@ -0,0 +1,22 @@
+#ifndef _USERLAND_COREUTILS_H
+#define _USERLAND_COREUTILS_H
+
+// GNU-style text-processing applets for LazyBox. Each entry point matches
+// applet_fn_t so it can be placed directly in the LazyBox dispatch table.
+
+int coreutils_sort(int argc, char** argv);
+int coreutils_uniq(int argc, char** argv);
+int coreutils_cut(int argc, char** argv);
+int coreutils_tr(int argc, char** argv);
+int coreutils_rev(int argc, char** argv);
+int coreutils_tac(int argc, char** argv);
+int coreutils_nl(int argc, char** argv);
+int coreutils_seq(int argc, char** argv);
+int coreutils_diff(int argc, char** argv);
+int coreutils_xxd(int argc, char** argv);
+int coreutils_du(int argc, char** argv);
+int coreutils_factor(int argc, char** argv);
+int coreutils_sum(int argc, char** argv);
+int coreutils_truncate(int argc, char** argv);
+
+#endif // _USERLAND_COREUTILS_H