sched: add thread management syscalls

arch/at91sam3x8e/entry.c

@@ -16,7 +16,7 @@
 extern uint16_t __syscall_return_point;
 #endif
 
-void arch_enter(struct exc_context *context)
+void enter_syscall(struct exc_context *context)
 {
 	enum syscall number = sc_num(context);
 	long (*handler)(sysarg_t arg1, sysarg_t arg2, sysarg_t arg3,
@@ -47,6 +47,8 @@ void arch_enter(struct exc_context *context)
 		return;
 	}
 
+	current->tcb.exc_context = context;
+
 	/* TODO: not every syscall uses the max amount of parameters (duh) */
 	sc_ret = handler(sc_arg1(context), sc_arg2(context), sc_arg3(context),
 			 sc_arg4(context), sc_arg5(context), sc_arg6(context));
@@ -54,6 +56,12 @@ void arch_enter(struct exc_context *context)
 	sc_set_rval(context, sc_ret);
 }
 
+void enter_sched(struct exc_context *context)
+{
+	current->tcb.exc_context = context;
+	schedule();
+}
+
 /*
  * This file is part of Ardix.
  * Copyright (c) 2020, 2021 Felix Kopp <owo@fef.moe>.

arch/at91sam3x8e/handle_fault.c

@@ -106,8 +106,6 @@ static void print_regs(struct exc_context *context)
 	print_reg("xPSR", context->sp->psr);
 }
 
-#include <arch/debug.h>
-
 __naked __noreturn void handle_fault(struct exc_context *context, enum irqno irqno)
 {
 	uart_emergency_setup();
@@ -120,7 +118,6 @@ __naked __noreturn void handle_fault(struct exc_context *context, enum irqno irq
 
 	uart_write_sync("\nSystem halted, goodbye\n\n");
 
-	__breakpoint;
 	while (1);
 }
 

arch/at91sam3x8e/handle_pend_sv.S

@@ -4,14 +4,15 @@
 
 .text
 
-/* void schedule(void); */
-.extern schedule
+/* void enter_sched(struct exc_context *context); */
+.extern enter_sched
 
 /* void handle_pend_sv(void); */
 func_begin handle_pend_sv
 
 	prepare_entry
-	bl	schedule
+	mov	r0,	sp
+	bl	enter_sched
 	prepare_leave
 
 	bx	lr

arch/at91sam3x8e/handle_svc.S

@@ -4,15 +4,15 @@
 
 .text
 
-/* void arch_enter(struct exc_context *context); */
-.extern arch_enter
+/* void enter_syscall(struct exc_context *context); */
+.extern enter_syscall
 
 /* void handle_svc(void); */
 func_begin handle_svc
 
 	prepare_entry
 	mov	r0,	sp
-	bl	arch_enter		/* arch_enter(sp); */
+	bl	enter_syscall		/* enter_syscall(sp); */
 	prepare_leave
 	bx	lr
 

arch/at91sam3x8e/include/arch/hardware.h

@@ -69,11 +69,17 @@ struct context {
 /**
  * @brief Task Control Block.
  * This is a low level structure used by `do_switch()` to do the actual context
- * switching,
+ * switching, and embedded into `struct task`.  We do this nesting because it
+ * makes it easier to access the TCB's fields from assembly, and it also makes
+ * us less dependent on a specific architecture.
  */
 struct tcb {
 	struct context context;
-	struct hw_context *hw_context;
+	/*
+	 * Needed for exec() because the child task leaves kernel space over a
+	 * different route than the parent one.
+	 */
+	struct exc_context *exc_context;
 };
 
 __always_inline sysarg_t sc_num(const struct exc_context *ctx)

arch/at91sam3x8e/sched.c

@@ -14,6 +14,7 @@
 
 #include <errno.h>
 #include <string.h>
+#include <unistd.h>
 
 volatile unsigned long int tick = 0;
 
@@ -51,13 +52,23 @@ int arch_sched_init(unsigned int freq)
 	return 0;
 }
 
-void arch_task_init(struct task *task, void (*entry)(void))
+void task_init(struct task *task, int (*entry)(void))
 {
+	task->bottom = task->stack + CONFIG_STACK_SIZE;
 	/* TODO: Use separate stacks for kernel and program */
 	struct hw_context *hw_context = task->bottom - sizeof(*hw_context);
 	struct exc_context *exc_context = (void *)hw_context - sizeof(*exc_context);
 
 	memset(hw_context, 0, task->bottom - (void *)hw_context);
+	/*
+	 * The return value of entry(), which is the exit code, will be stored
+	 * in r0 as per the AAPCS.  Conveniently, this happens to be the same
+	 * register that is also used for passing the first argument to a
+	 * function, so by setting the initial link register to exit() we
+	 * effectively inject a call to that function after the task's main
+	 * routine returns.
+	 */
+	hw_context->lr = exit;
 	hw_context->pc = entry;
 	hw_context->psr = 0x01000000; /* Thumb = 1, unprivileged */
 
@@ -67,29 +78,15 @@ void arch_task_init(struct task *task, void (*entry)(void))
 	memset(&task->tcb, 0, sizeof(task->tcb));
 	task->tcb.context.sp = exc_context;
 	task->tcb.context.pc = _leave;
+	task->tcb.exc_context = exc_context;
 }
 
-__naked __noreturn void _idle(void)
+__naked int _idle(void)
 {
 	/* TODO: put the CPU to sleep */
 	while (1);
 }
 
-int arch_idle_task_init(struct task *task)
-{
-	void *stack = malloc(CONFIG_STACK_SIZE);
-	if (stack == NULL)
-		return -ENOMEM;
-
-	task->bottom = stack + CONFIG_STACK_SIZE; /* full-descending stack */
-	arch_task_init(task, _idle);
-	task->sleep = 0;
-	task->last_tick = 0;
-	task->state = TASK_READY;
-	task->pid = -1;
-	return 0;
-}
-
 unsigned long int ms_to_ticks(unsigned long int ms)
 {
 	return ( ms * (unsigned long int)tick_freq ) / 1000lu /* 1 s = 1000 ms */;

include/arch-generic/sched.h

@@ -16,16 +16,16 @@ struct task; /* see include/ardix/sched.h */
 int arch_sched_init(unsigned int freq);
 
 /**
- * Initialize a new process.
- * This requires the process' `stack_base` field to be initialized as the
- * initial register values are written to the stack.
+ * @brief Initialize a new task.
  *
- * @param process: The process.
- * @param entry: The process entry point.
+ *
+ * @param task Task to initialize
+ * @param entry Task entry point
  */
-void arch_task_init(struct task *task, void (*entry)(void));
+void task_init(struct task *task, int (*entry)(void));
 
-int arch_idle_task_init(struct task *task);
+/** @brief Idle task entry point. */
+__naked int _idle(void);
 
 /**
  * @brief Convert milliseconds to system ticks, rounding to zero.

include/arch-generic/syscall.h

@@ -7,6 +7,9 @@
 #define ARCH_SYS_sleep		2
 #define ARCH_SYS_malloc		3
 #define ARCH_SYS_free		4
+#define ARCH_SYS_exec		5
+#define ARCH_SYS_exit		6
+#define ARCH_SYS_waitpid	7
 
 /*
  * This file is part of Ardix.

include/ardix/kevent.h

@@ -25,6 +25,8 @@ enum kevent_kind {
 	KEVENT_DEVICE,
 	/** @brief File has changed */
 	KEVENT_FILE,
+	/** @brief Task has exited */
+	KEVENT_TASK,
 
 	KEVENT_KIND_COUNT,
 };

include/ardix/sched.h

@@ -6,6 +6,8 @@
 
 #include <ardix/kent.h>
 #include <ardix/list.h>
+#include <ardix/mutex.h>
+#include <ardix/task.h>
 #include <ardix/types.h>
 
 #include <config.h>
@@ -14,40 +16,6 @@
 #warning "CONFIG_SCHED_MAXTASK is > 64, this could have a significant performance impact"
 #endif
 
-enum task_state {
-	/** Task is dead / doesn't exist */
-	TASK_DEAD,
-	/** Task is ready for execution or currently running. */
-	TASK_READY,
-	/** Task is waiting for its next time share. */
-	TASK_QUEUE,
-	/** Task is sleeping, `task::sleep` specifies for how many ticks. */
-	TASK_SLEEP,
-	/** Task is waiting for I/O to flush buffers. */
-	TASK_IOWAIT,
-	/** Task is waiting for a mutex to be unlocked. */
-	TASK_LOCKWAIT,
-};
-
-/** @brief Core structure holding information about a task. */
-struct task {
-	struct tcb tcb;
-
-	struct kent kent;
-	/**
-	 * @brief Points to the bottom of the stack.
-	 * In a full-descending stack, this is one word after the highest stack address.
-	 */
-	void *bottom;
-	/** @brief If state is `TASK_SLEEP`, the total amount of ticks to sleep */
-	unsigned long int sleep;
-	/** @brief Last execution in ticks */
-	unsigned long int last_tick;
-
-	enum task_state state;
-	pid_t pid;
-};
-
 /** @brief Current task (access from syscall context only) */
 extern struct task *volatile current;
 
@@ -64,7 +32,7 @@ int sched_init(void);
  * @brief Main scheduler routine.
  * This will iterate over the process table and choose a new task to be run,
  * which `current` is then updated to.  If the old task was in state
- * `TASK_READY`, it is set to `TASK_QUEUE`.
+ * `TASK_RUNNING`, it is set to `TASK_QUEUE`.
  */
 void schedule(void);
 
@@ -77,9 +45,10 @@ void schedule(void);
  * setup work.
  *
  * @param task Task to make a copy of
+ * @param err Where to store the error code (will be written 0 on success)
  * @returns The new (child) task copy, or `NULL` on failure
  */
-struct task *task_clone(struct task *task);
+struct task *task_clone(struct task *task, int *trr);
 
 /**
  * @brief Sleep for an approximate amount of milliseconds.

include/ardix/syscall.h

@@ -15,6 +15,9 @@ enum syscall {
 	SYS_sleep		= ARCH_SYS_sleep,
 	SYS_malloc		= ARCH_SYS_malloc,
 	SYS_free		= ARCH_SYS_free,
+	SYS_exec		= ARCH_SYS_exec,
+	SYS_exit		= ARCH_SYS_exit,
+	SYS_waitpid		= ARCH_SYS_waitpid,
 	NSYSCALLS
 };
 
@@ -31,6 +34,11 @@ long sys_stub(void);
 long sys_read(int fd, void *buf, size_t len);
 long sys_write(int fd, const void *buf, size_t len);
 long sys_sleep(unsigned long millis);
+long sys_malloc(size_t size);
+void sys_free(void *ptr);
+long sys_exec(int (*entry)(void));
+void sys_exit(int code);
+long sys_waitpid(pid_t pid, int *stat_loc, int options);
 
 /*
  * This file is part of Ardix.

include/ardix/task.h

+/* See the end of this file for copyright, license, and warranty information. */
+
+#pragma once
+
+#include <arch/hardware.h>
+
+#include <ardix/kent.h>
+#include <ardix/kevent.h>
+#include <ardix/malloc.h>
+#include <ardix/sched.h>
+#include <ardix/util.h>
+
+enum task_state {
+	/** Task is dead / doesn't exist */
+	TASK_DEAD,
+	/** Task is currently running. */
+	TASK_RUNNING,
+	/** Task is waiting for its next time share. */
+	TASK_QUEUE,
+	/** Task is sleeping, `task::sleep` specifies for how many ticks. */
+	TASK_SLEEP,
+	/** Task is waiting for I/O to flush buffers. */
+	TASK_IOWAIT,
+	/** Task is waiting for a mutex to be unlocked. */
+	TASK_LOCKWAIT,
+	/** Task is waiting for child to  */
+	TASK_WAITPID,
+};
+
+/** @brief Core structure holding information about a task. */
+struct task {
+	struct tcb tcb;
+
+	struct kent kent;
+	/**
+	 * @brief Points to the bottom of the stack.
+	 * In a full-descending stack, this is one word after the highest stack address.
+	 */
+	void *bottom;
+	/** @brief Lowest address in the stack, as returned by malloc. */
+	void *stack;
+	/** @brief If state is `TASK_SLEEP`, the total amount of ticks to sleep */
+	unsigned long int sleep;
+	/** @brief Last execution in ticks */
+	unsigned long int last_tick;
+
+	/*
+	 * if a child process exited before its parent called waitpid(),
+	 * this is where the children are stored temporarily
+	 */
+	struct list_head pending_sigchld;
+	struct mutex pending_sigchld_lock;
+
+	enum task_state state;
+	pid_t pid;
+};
+
+__always_inline void task_get(struct task *task)
+{
+	kent_get(&task->kent);
+}
+
+__always_inline void task_put(struct task *task)
+{
+	kent_put(&task->kent);
+}
+
+__always_inline struct task *task_parent(struct task *task)
+{
+	if (task->pid == 0)
+		return NULL;
+	else
+		return container_of(task->kent.parent, struct task, kent);
+}
+
+struct task_kevent {
+	struct kevent kevent;
+	struct task *task;
+	int status;
+};
+
+void task_kevent_create_and_dispatch(struct task *task, int status);
+
+/*
+ * This file is part of Ardix.
+ * Copyright (c) 2020, 2021 Felix Kopp <owo@fef.moe>.
+ *
+ * Ardix is non-violent software: you may only use, redistribute,
+ * and/or modify it under the terms of the CNPLv6+ as found in
+ * the LICENSE file in the source code root directory or at
+ * <https://git.pixie.town/thufie/CNPL>.
+ *
+ * Ardix comes with ABSOLUTELY NO WARRANTY, to the extent
+ * permitted by applicable law.  See the CNPLv6+ for details.
+ */

include/sys/wait.h

+/* See the end of this file for copyright, license, and warranty information. */
+
+#pragma once
+
+#include <stdint.h>
+#include <toolchain.h>
+
+__shared pid_t waitpid(pid_t pid, int *stat_loc, int options);
+
+/*
+ * This file is part of Ardix.
+ * Copyright (c) 2020, 2021 Felix Kopp <owo@fef.moe>.
+ *
+ * Ardix is non-violent software: you may only use, redistribute,
+ * and/or modify it under the terms of the CNPLv6+ as found in
+ * the LICENSE file in the source code root directory or at
+ * <https://git.pixie.town/thufie/CNPL>.
+ *
+ * Ardix comes with ABSOLUTELY NO WARRANTY, to the extent
+ * permitted by applicable law.  See the CNPLv6+ for details.
+ */

include/unistd.h

@@ -3,10 +3,21 @@
 #pragma once
 
 #include <stdint.h>
+#include <toolchain.h>
 
 __shared ssize_t read(int fildes, void *buf, size_t nbyte);
 __shared ssize_t write(int fildes, const void *buf, size_t nbyte);
 __shared ssize_t sleep(unsigned long int millis);
+/**
+ * @brief Create a new thread.
+ *
+ * Embedded systems typically don't have a MMU and thus no virtual memory,
+ * meaning it is impossible to implement a proper fork.  So, the `fork()` and
+ * `execve()` system calls have to be combined into one.
+ */
+__shared pid_t exec(int (*entry)(void));
+__shared __noreturn void exit(int status);
+__shared pid_t waitpid(pid_t pid, int *stat_loc, int options);
 
 /*
  * This file is part of Ardix.

kernel/CMakeLists.txt

@@ -21,6 +21,7 @@ target_sources(ardix_kernel PRIVATE
 	sched.c
 	serial.c
 	syscall.c
+	task.c
 	userspace.c
 )
 

kernel/mutex.c

@@ -44,9 +44,7 @@ void mutex_unlock(struct mutex *mutex)
 	spin_unlock(&mutex->wait_queue_lock);
 
 	if (waiter != NULL) {
-		struct task *task = waiter->task;
-		current->state = TASK_QUEUE;
-		do_switch(current, task);
+		waiter->task->state = TASK_QUEUE;
 	} else {
 		_mutex_unlock(&mutex->lock);
 	}

kernel/sched.c

@@ -38,6 +38,7 @@
 #include <ardix/kevent.h>
 #include <ardix/malloc.h>
 #include <ardix/sched.h>
+#include <ardix/task.h>
 #include <ardix/types.h>
 
 #include <errno.h>
@@ -48,6 +49,7 @@ extern uint32_t _sstack;
 extern uint32_t _estack;
 
 static struct task *tasks[CONFIG_SCHED_MAXTASK];
+static MUTEX(tasks_lock);
 struct task *volatile current;
 
 static struct task kernel_task;
@@ -56,7 +58,12 @@ static struct task idle_task;
 static void task_destroy(struct kent *kent)
 {
 	struct task *task = container_of(kent, struct task, kent);
+
+	mutex_lock(&tasks_lock);
 	tasks[task->pid] = NULL;
+	mutex_unlock(&tasks_lock);
+
+	kfree(task->stack);
 	kfree(task);
 }
 
@@ -72,8 +79,12 @@ int sched_init(void)
 
 	memset(&kernel_task.tcb, 0, sizeof(kernel_task.tcb));
 	kernel_task.bottom = &_estack;
+	kernel_task.stack = kernel_task.bottom - CONFIG_STACK_SIZE;
 	kernel_task.pid = 0;
-	kernel_task.state = TASK_READY;
+	kernel_task.state = TASK_RUNNING;
+
+	list_init(&kernel_task.pending_sigchld);
+	mutex_init(&kernel_task.pending_sigchld_lock);
 
 	tasks[0] = &kernel_task;
 	current = &kernel_task;
@@ -85,11 +96,17 @@ int sched_init(void)
 	if (err != 0)
 		goto out;
 
-	err = arch_sched_init(CONFIG_SCHED_FREQ);
-	if (err != 0)
+	idle_task.stack = kmalloc(CONFIG_STACK_SIZE);
+	if (idle_task.stack == NULL)
 		goto out;
+	idle_task.bottom = idle_task.stack + CONFIG_STACK_SIZE;
+	idle_task.pid = -1;
+	idle_task.state = TASK_QUEUE;
+	list_init(&idle_task.pending_sigchld);
+	mutex_init(&idle_task.pending_sigchld_lock);
+	task_init(&idle_task, _idle);
 
-	err = arch_idle_task_init(&idle_task);
+	err = arch_sched_init(CONFIG_SCHED_FREQ);
 	if (err != 0)
 		goto out;
 
@@ -113,11 +130,12 @@ static inline bool can_run(const struct task *task)
 	case TASK_SLEEP:
 		return tick - task->last_tick >= task->sleep;
 	case TASK_QUEUE:
-	case TASK_READY:
+	case TASK_RUNNING:
 		return true;
 	case TASK_DEAD:
 	case TASK_IOWAIT:
 	case TASK_LOCKWAIT:
+	case TASK_WAITPID:
 		return false;
 	}
 
@@ -135,7 +153,7 @@ void schedule(void)
 
 	kevents_process();
 
-	if (old->state == TASK_READY)
+	if (old->state == TASK_RUNNING)
 		old->state = TASK_QUEUE;
 
 	for (unsigned int i = 0; i < ARRAY_SIZE(tasks); i++) {
@@ -156,7 +174,7 @@ void schedule(void)
 	if (new == NULL)
 		new = &idle_task;
 
-	new->state = TASK_READY;
+	new->state = TASK_RUNNING;
 	new->last_tick = tick;
 	current = new;
 
@@ -177,9 +195,67 @@ long sys_sleep(unsigned long int millis)
 	current->sleep = ms_to_ticks(millis);
 	yield(TASK_SLEEP);
 	/* TODO: return actual milliseconds */
+	/*
+	 * TODO: actually, use fucking hardware timers which were specifically
+	 *       invented for this exact kind of feature because (1) the tick
+	 *       resolution is often less than 1 ms and (2) ticks aren't really
+	 *       supposed to be guaranteed to happen at regular intervals and
+	 *       (3) the scheduler doesn't even check whether there is a task
+	 *       whose sleep period just expired
+	 */
 	return 0;
 }
 
+long sys_exec(int (*entry)(void))
+{
+	pid_t pid;
+	struct task *child = NULL;
+
+	mutex_lock(&tasks_lock);
+
+	for (pid = 1; pid < CONFIG_SCHED_MAXTASK; pid++) {
+		if (tasks[pid] == NULL)
+			break;
+	}
+	if (pid == CONFIG_SCHED_MAXTASK) {
+		pid = -EAGAIN;
+		goto out;
+	}
+
+	child = kmalloc(sizeof(*child));
+	if (child == NULL) {
+		pid = -ENOMEM;
+		goto out;
+	}
+
+	child->pid = pid;
+	child->stack = kmalloc(CONFIG_STACK_SIZE);
+	if (child->stack == NULL) {
+		pid = -ENOMEM;
+		goto err_stack_malloc;
+	}
+
+	child->kent.parent = &current->kent;
+	child->kent.destroy = task_destroy;
+	kent_init(&child->kent);
+
+	child->bottom = child->stack + CONFIG_STACK_SIZE;
+	task_init(child, entry);
+
+	list_init(&child->pending_sigchld);
+	mutex_init(&child->pending_sigchld_lock);
+
+	child->state = TASK_QUEUE;
+	tasks[pid] = child;
+	goto out;
+
+err_stack_malloc:
+	kfree(child);
+out:
+	mutex_unlock(&tasks_lock);
+	return pid;
+}
+
 /*
  * This file is part of Ardix.
  * Copyright (c) 2020, 2021 Felix Kopp <owo@fef.moe>.

kernel/syscall.c

@@ -15,6 +15,9 @@ long (*const sys_table[NSYSCALLS])(sysarg_t arg1, sysarg_t arg2, sysarg_t arg3,
 	sys_table_entry(SYS_sleep,		sys_sleep),
 	sys_table_entry(SYS_malloc,		sys_malloc),