rust/src/rt/rust_task.cpp

#include "rust_internal.h"
#include "rust_cc.h"

#include "vg/valgrind.h"
#include "vg/memcheck.h"

#ifndef __WIN32__
#include <execinfo.h>
#endif
#include <iostream>
#include <cassert>
#include <cstring>
#include <algorithm>

#include "globals.h"

// Each stack gets some guard bytes that valgrind will verify we don't touch
#ifndef NVALGRIND
#define STACK_NOACCESS_SIZE 16
#else
#define STACK_NOACCESS_SIZE 0
#endif

// The amount of extra space at the end of each stack segment, available
// to the rt, compiler and dynamic linker for running small functions
// FIXME: We want this to be 128 but need to slim the red zone calls down
#ifdef __i386__
#define RED_ZONE_SIZE (65536 + STACK_NOACCESS_SIZE)
#endif

#ifdef __x86_64__
#define RED_ZONE_SIZE (65536 + STACK_NOACCESS_SIZE)
#endif

// Stack size
size_t g_custom_min_stack_size = 0;

static size_t
get_min_stk_size(size_t default_size) {
    if (g_custom_min_stack_size != 0) {
        return g_custom_min_stack_size;
    } else {
        return default_size;
    }
}

static size_t
get_next_stk_size(rust_scheduler *sched, rust_task *task,
                  size_t min, size_t current, size_t requested) {
    LOG(task, mem, "calculating new stack size for 0x%" PRIxPTR, task);
    LOG(task, mem,
        "min: %" PRIdPTR " current: %" PRIdPTR " requested: %" PRIdPTR,
        min, current, requested);

    // Allocate at least enough to accomodate the next frame
    size_t sz = std::max(min, requested);

    // And double the stack size each allocation
    const size_t max = 1024 * 1024;
    size_t next = std::min(max, current * 2);

    sz = std::max(sz, next);

    LOG(task, mem, "next stack size: %" PRIdPTR, sz);
    I(sched, requested <= sz);
    return sz;
}

// Task stack segments. Heap allocated and chained together.

static void
config_valgrind_stack(stk_seg *stk) {
    stk->valgrind_id =
        VALGRIND_STACK_REGISTER(&stk->data[0],
                                stk->end);
#ifndef NVALGRIND
    // Establish that the stack is accessible.  This must be done when reusing
    // old stack segments, since the act of popping the stack previously
    // caused valgrind to consider the whole thing inaccessible.
    size_t sz = stk->end - (uintptr_t)&stk->data[0];
    VALGRIND_MAKE_MEM_UNDEFINED(stk->data, sz);

    // Establish some guard bytes so valgrind will tell
    // us if we run off the end of the stack
    VALGRIND_MAKE_MEM_NOACCESS(stk->data, STACK_NOACCESS_SIZE);
#endif
}

static void
unconfig_valgrind_stack(stk_seg *stk) {
#ifndef NVALGRIND
    // Make the guard bytes accessible again, but undefined
    VALGRIND_MAKE_MEM_UNDEFINED(stk->data, STACK_NOACCESS_SIZE);
#endif
VALGRIND_STACK_DEREGISTER(stk->valgrind_id);
}

static void
free_stk(rust_task *task, stk_seg *stk) {
    LOGPTR(task->sched, "freeing stk segment", (uintptr_t)stk);
    task->free(stk);
}

static stk_seg*
new_stk(rust_scheduler *sched, rust_task *task, size_t requested_sz)
{
    LOG(task, mem, "creating new stack for task %" PRIxPTR, task);

    // The minimum stack size, in bytes, of a Rust stack, excluding red zone
    size_t min_sz = get_min_stk_size(sched->min_stack_size);

    // Try to reuse an existing stack segment
    if (task->stk != NULL && task->stk->prev != NULL) {
        size_t prev_sz = (size_t)(task->stk->prev->end
                                  - (uintptr_t)&task->stk->prev->data[0]
                                  - RED_ZONE_SIZE);
        if (min_sz <= prev_sz) {
            LOG(task, mem, "reusing existing stack");
            task->stk = task->stk->prev;
            A(sched, task->stk->prev == NULL, "Bogus stack ptr");
            config_valgrind_stack(task->stk);
            return task->stk;
        } else {
            LOG(task, mem, "existing stack is not big enough");
            free_stk(task, task->stk->prev);
            task->stk->prev = NULL;
        }
    }

    // The size of the current stack segment, excluding red zone
    size_t current_sz = 0;
    if (task->stk != NULL) {
        current_sz = (size_t)(task->stk->end
                              - (uintptr_t)&task->stk->data[0]
                              - RED_ZONE_SIZE);
    }
    // The calculated size of the new stack, excluding red zone
    size_t rust_stk_sz = get_next_stk_size(sched, task, min_sz,
                                           current_sz, requested_sz);

    size_t sz = sizeof(stk_seg) + rust_stk_sz + RED_ZONE_SIZE;
    stk_seg *stk = (stk_seg *)task->malloc(sz, "stack");
    LOGPTR(task->sched, "new stk", (uintptr_t)stk);
    memset(stk, 0, sizeof(stk_seg));
    stk->prev = NULL;
    stk->next = task->stk;
    stk->end = (uintptr_t) &stk->data[rust_stk_sz + RED_ZONE_SIZE];
    LOGPTR(task->sched, "stk end", stk->end);

    task->stk = stk;
    config_valgrind_stack(task->stk);
    return stk;
}

static void
del_stk(rust_task *task, stk_seg *stk)
{
    assert(stk == task->stk && "Freeing stack segments out of order!");

    task->stk = stk->next;

    bool delete_stack = false;
    if (task->stk != NULL) {
        // Don't actually delete this stack. Save it to reuse later,
        // preventing the pathological case where we repeatedly reallocate
        // the stack for the next frame.
        task->stk->prev = stk;
    } else {
        // This is the last stack, delete it.
        delete_stack = true;
    }

    // Delete the previous previous stack
    if (stk->prev != NULL) {
        free_stk(task, stk->prev);
        stk->prev = NULL;
    }

    unconfig_valgrind_stack(stk);
    if (delete_stack) {
        free_stk(task, stk);
    }
}

// Tasks
rust_task::rust_task(rust_scheduler *sched, rust_task_list *state,
                     rust_task *spawner, const char *name) :
    ref_count(1),
    stk(NULL),
    runtime_sp(0),
    sched(sched),
    cache(NULL),
    kernel(sched->kernel),
    name(name),
    state(state),
    cond(NULL),
    cond_name("none"),
    supervisor(spawner),
    list_index(-1),
    next_port_id(0),
    rendezvous_ptr(0),
    running_on(-1),
    pinned_on(-1),
    local_region(&sched->srv->local_region),
    failed(false),
    killed(false),
    propagate_failure(true),
    dynastack(this),
    cc_counter(0)
{
    LOGPTR(sched, "new task", (uintptr_t)this);
    DLOG(sched, task, "sizeof(task) = %d (0x%x)", sizeof *this, sizeof *this);

    assert((void*)this == (void*)&user);

    user.notify_enabled = 0;

    stk = new_stk(sched, this, 0);
    user.rust_sp = stk->end;
    if (supervisor) {
        supervisor->ref();
    }
}

rust_task::~rust_task()
{
    I(sched, !sched->lock.lock_held_by_current_thread());
    I(sched, port_table.is_empty());
    DLOG(sched, task, "~rust_task %s @0x%" PRIxPTR ", refcnt=%d",
         name, (uintptr_t)this, ref_count);

    if (supervisor) {
        supervisor->deref();
    }

    kernel->release_task_id(user.id);

    /* FIXME: tighten this up, there are some more
       assertions that hold at task-lifecycle events. */
    I(sched, ref_count == 0); // ||
    //   (ref_count == 1 && this == sched->root_task));

    // Delete all the stacks. There may be more than one if the task failed
    // FIXME: This is not correct. During unwinding we need to delete
    // the stacks and record the stack limit, otherwise the stack
    // stack is corrupted when destructors are running.
    while (stk != NULL) {
        del_stk(this, stk);
    }
}

struct spawn_args {
    rust_task *task;
    uintptr_t a3;
    uintptr_t a4;
    void (*CDECL f)(int *, uintptr_t, uintptr_t);
};

struct rust_closure_env {
    intptr_t ref_count;
    type_desc *td;
};

// This runs on the Rust stack
extern "C" CDECL
void task_start_wrapper(spawn_args *a)
{
    rust_task *task = a->task;
    int rval = 42;

    bool failed = false;
    try {
        a->f(&rval, a->a3, a->a4);
    } catch (rust_task *ex) {
        A(task->sched, ex == task,
          "Expected this task to be thrown for unwinding");
        failed = true;
    }

    cc::do_cc(task);

    rust_closure_env* env = (rust_closure_env*)a->a3;
    if(env) {
        // free the environment.
        I(task->sched, 1 == env->ref_count); // the ref count better be 1
        //env->td->drop_glue(NULL, task, NULL, env->td->first_param, env);
        //env->td->free_glue(NULL, task, NULL, env->td->first_param, env);
        task->free(env);
    }

    task->die();

    if (task->killed && !failed) {
        LOG(task, task, "Task killed during termination");
        failed = true;
    }

    task->notify(!failed);

    if (failed) {
#ifndef __WIN32__
        task->conclude_failure();
#else
        A(task->sched, false, "Shouldn't happen");
#endif
    }
    task->ctx.next->swap(task->ctx);
}

void
rust_task::start(uintptr_t spawnee_fn,
                 uintptr_t args,
                 uintptr_t env)
{
    LOG(this, task, "starting task from fn 0x%" PRIxPTR
        " with args 0x%" PRIxPTR, spawnee_fn, args);

    I(sched, stk->data != NULL);

    char *sp = (char *)user.rust_sp;

    sp -= sizeof(spawn_args);

    spawn_args *a = (spawn_args *)sp;

    a->task = this;
    a->a3 = env;
    a->a4 = args;
    void **f = (void **)&a->f;
    *f = (void *)spawnee_fn;

    ctx.call((void *)task_start_wrapper, a, sp);

    this->start();
}

void
rust_task::start(uintptr_t spawnee_fn,
                 uintptr_t args)
{
    start(spawnee_fn, args, 0);
}

void rust_task::start()
{
    yield_timer.reset_us(0);
    transition(&sched->newborn_tasks, &sched->running_tasks);
    sched->lock.signal();
}

void
rust_task::grow(size_t n_frame_bytes)
{
    // FIXME (issue #151): Just fail rather than almost certainly crashing
    // mysteriously later. The commented-out logic below won't work at all in
    // the presence of non-word-aligned pointers.
    abort();

}

// Only run this on the rust stack
void
rust_task::yield(size_t time_in_us, bool *killed) {
    if (this->killed) {
        *killed = true;
    }

    yield_timer.reset_us(time_in_us);

    // Return to the scheduler.
    ctx.next->swap(ctx);

    if (this->killed) {
        *killed = true;
    }
}

void
rust_task::kill() {
    if (dead()) {
        // Task is already dead, can't kill what's already dead.
        fail_parent();
        return;
    }

    // Note the distinction here: kill() is when you're in an upcall
    // from task A and want to force-fail task B, you do B->kill().
    // If you want to fail yourself you do self->fail().
    LOG(this, task, "killing task %s @0x%" PRIxPTR, name, this);
    // When the task next goes to yield or resume it will fail
    killed = true;
    // Unblock the task so it can unwind.
    unblock();

    sched->lock.signal();

    LOG(this, task, "preparing to unwind task: 0x%" PRIxPTR, this);
    // run_on_resume(rust_unwind_glue);
}

void
rust_task::fail() {
    // See note in ::kill() regarding who should call this.
    DLOG(sched, task, "task %s @0x%" PRIxPTR " failing", name, this);
    backtrace();
#ifndef __WIN32__
    throw this;
#else
    die();
    conclude_failure();
#endif
}

void
rust_task::conclude_failure() {
    fail_parent();
    failed = true;
}

void
rust_task::fail_parent() {
    if (supervisor) {
        DLOG(sched, task,
             "task %s @0x%" PRIxPTR
             " propagating failure to supervisor %s @0x%" PRIxPTR,
             name, this, supervisor->name, supervisor);
        supervisor->kill();
    }
    // FIXME: implement unwinding again.
    if (NULL == supervisor && propagate_failure)
        sched->fail();
}

void
rust_task::unsupervise()
{
    DLOG(sched, task,
             "task %s @0x%" PRIxPTR
             " disconnecting from supervisor %s @0x%" PRIxPTR,
             name, this, supervisor->name, supervisor);
    if (supervisor) {
        supervisor->deref();
    }
    supervisor = NULL;
    propagate_failure = false;
}

frame_glue_fns*
rust_task::get_frame_glue_fns(uintptr_t fp) {
    fp -= sizeof(uintptr_t);
    return *((frame_glue_fns**) fp);
}

bool
rust_task::running()
{
    return state == &sched->running_tasks;
}

bool
rust_task::blocked()
{
    return state == &sched->blocked_tasks;
}

bool
rust_task::blocked_on(rust_cond *on)
{
    return blocked() && cond == on;
}

bool
rust_task::dead()
{
    return state == &sched->dead_tasks;
}

void *
rust_task::malloc(size_t sz, const char *tag, type_desc *td)
{
    return local_region.malloc(sz, tag);
}

void *
rust_task::realloc(void *data, size_t sz, bool is_gc)
{
    return local_region.realloc(data, sz);
}

void
rust_task::free(void *p, bool is_gc)
{
    local_region.free(p);
}

void
rust_task::transition(rust_task_list *src, rust_task_list *dst) {
    bool unlock = false;
    if(!sched->lock.lock_held_by_current_thread()) {
        unlock = true;
        sched->lock.lock();
    }
    DLOG(sched, task,
         "task %s " PTR " state change '%s' -> '%s' while in '%s'",
         name, (uintptr_t)this, src->name, dst->name, state->name);
    I(sched, state == src);
    src->remove(this);
    dst->append(this);
    state = dst;
    if(unlock)
        sched->lock.unlock();
}

void
rust_task::block(rust_cond *on, const char* name) {
    I(sched, !lock.lock_held_by_current_thread());
    scoped_lock with(lock);
    LOG(this, task, "Blocking on 0x%" PRIxPTR ", cond: 0x%" PRIxPTR,
                         (uintptr_t) on, (uintptr_t) cond);
    A(sched, cond == NULL, "Cannot block an already blocked task.");
    A(sched, on != NULL, "Cannot block on a NULL object.");

    transition(&sched->running_tasks, &sched->blocked_tasks);
    cond = on;
    cond_name = name;
}

void
rust_task::wakeup(rust_cond *from) {
    I(sched, !lock.lock_held_by_current_thread());
    scoped_lock with(lock);
    A(sched, cond != NULL, "Cannot wake up unblocked task.");
    LOG(this, task, "Blocked on 0x%" PRIxPTR " woken up on 0x%" PRIxPTR,
                        (uintptr_t) cond, (uintptr_t) from);
    A(sched, cond == from, "Cannot wake up blocked task on wrong condition.");

    transition(&sched->blocked_tasks, &sched->running_tasks);
    I(sched, cond == from);
    cond = NULL;
    cond_name = "none";

    sched->lock.signal();
}

void
rust_task::die() {
    I(sched, !lock.lock_held_by_current_thread());
    scoped_lock with(lock);
    transition(&sched->running_tasks, &sched->dead_tasks);
    sched->lock.signal();
}

void
rust_task::unblock() {
    if (blocked()) {
        // FIXME: What if another thread unblocks the task between when
        // we checked and here?
        wakeup(cond);
    }
}

rust_crate_cache *
rust_task::get_crate_cache()
{
    if (!cache) {
        DLOG(sched, task, "fetching cache for current crate");
        cache = sched->get_cache();
    }
    return cache;
}

void
rust_task::backtrace() {
    if (!log_rt_backtrace) return;
#ifndef __WIN32__
    void *call_stack[256];
    int nframes = ::backtrace(call_stack, 256);
    backtrace_symbols_fd(call_stack + 1, nframes - 1, 2);
#endif
}

bool rust_task::can_schedule(int id)
{
    return yield_timer.has_timed_out() &&
        running_on == -1 &&
        (pinned_on == -1 || pinned_on == id);
}

void *
rust_task::calloc(size_t size, const char *tag) {
    return local_region.calloc(size, tag);
}

void rust_task::pin() {
    I(this->sched, running_on != -1);
    pinned_on = running_on;
}

void rust_task::pin(int id) {
    I(this->sched, running_on == -1);
    pinned_on = id;
}

void rust_task::unpin() {
    pinned_on = -1;
}

rust_port_id rust_task::register_port(rust_port *port) {
    I(sched, !lock.lock_held_by_current_thread());
    scoped_lock with(lock);

    rust_port_id id = next_port_id++;
    port_table.put(id, port);
    return id;
}

void rust_task::release_port(rust_port_id id) {
    I(sched, lock.lock_held_by_current_thread());
    port_table.remove(id);
}

rust_port *rust_task::get_port_by_id(rust_port_id id) {
    I(sched, !lock.lock_held_by_current_thread());
    scoped_lock with(lock);
    rust_port *port = NULL;
    port_table.get(id, &port);
    if (port) {
        port->ref();
    }
    return port;
}


// Temporary routine to allow boxes on one task's shared heap to be reparented
// to another.
const type_desc *
rust_task::release_alloc(void *alloc) {
    I(sched, !lock.lock_held_by_current_thread());
    lock.lock();

    assert(local_allocs.find(alloc) != local_allocs.end());
    const type_desc *tydesc = local_allocs[alloc];
    local_allocs.erase(alloc);

    local_region.release_alloc(alloc);

    lock.unlock();
    return tydesc;
}

// Temporary routine to allow boxes from one task's shared heap to be
// reparented to this one.
void
rust_task::claim_alloc(void *alloc, const type_desc *tydesc) {
    I(sched, !lock.lock_held_by_current_thread());
    lock.lock();

    assert(local_allocs.find(alloc) == local_allocs.end());
    local_allocs[alloc] = tydesc;
    local_region.claim_alloc(alloc);

    lock.unlock();
}

void
rust_task::notify(bool success) {
    // FIXME (1078) Do this in rust code
    if(user.notify_enabled) {
        rust_task *target_task = kernel->get_task_by_id(user.notify_chan.task);
        if (target_task) {
            rust_port *target_port =
                target_task->get_port_by_id(user.notify_chan.port);
            if(target_port) {
                task_notification msg;
                msg.id = user.id;
                msg.result = !success ? tr_failure : tr_success;

                target_port->send(&msg);
                scoped_lock with(target_task->lock);
                target_port->deref();
            }
            target_task->deref();
        }
    }
}

extern "C" CDECL void
record_sp(void *limit);

void *
rust_task::new_stack(size_t stk_sz, void *args_addr, size_t args_sz) {

    stk_seg *stk_seg = new_stk(sched, this, stk_sz + args_sz);

    uint8_t *new_sp = (uint8_t*)stk_seg->end;
    // Push the function arguments to the new stack
    new_sp = align_down(new_sp - args_sz);
    memcpy(new_sp, args_addr, args_sz);
    record_stack_limit();
    return new_sp;
}

void
rust_task::del_stack() {
    del_stk(this, stk);
    record_stack_limit();
}

void
rust_task::record_stack_limit() {
    // The function prolog compares the amount of stack needed to the end of
    // the stack. As an optimization, when the frame size is less than 256
    // bytes, it will simply compare %esp to to the stack limit instead of
    // subtracting the frame size. As a result we need our stack limit to
    // account for those 256 bytes.
    const unsigned LIMIT_OFFSET = 256;
    A(sched,
      (uintptr_t)stk->end - RED_ZONE_SIZE
      - (uintptr_t)stk->data >= LIMIT_OFFSET,
      "Stack size must be greater than LIMIT_OFFSET");
    record_sp(stk->data + LIMIT_OFFSET + RED_ZONE_SIZE);
}

extern "C" uintptr_t get_sp();

/*
Called by landing pads during unwinding to figure out which
stack segment we are currently running on, delete the others,
and record the stack limit (which was not restored when unwinding
through __morestack).
 */
void
rust_task::reset_stack_limit() {
    uintptr_t sp = get_sp();
    // Not positive these bounds for sp are correct.
    // I think that the first possible value for esp on a new
    // stack is stk->end, which points one word in front of
    // the first work to be pushed onto a new stack.
    while (sp <= (uintptr_t)stk->data || stk->end < sp) {
        del_stk(this, stk);
        A(sched, stk != NULL, "Failed to find the current stack");
    }
    record_stack_limit();
}

//
// Local Variables:
// mode: C++
// fill-column: 78;
// indent-tabs-mode: nil
// c-basic-offset: 4
// buffer-file-coding-system: utf-8-unix
// End:
//