Skip to content

X
xv6-public
提交 ff1eef00 创建 作者: Nickolai Zeldovich's avatar Nickolai Zeldovich
浏览文件
操作

Merge branch 'scale-amd64' of git+ssh://pdos.csail.mit.edu/home/am0/6.828/xv6 into scale-amd64

上级 bfdbe916 275a8c9f
隐藏空白字符变更
内嵌 并排
正在显示 包含 203 行增加 和 156 行删除
include/kernel.hh
浏览文件 @ ff1eef00
......@@ -126,7 +126,8 @@ void ideintr(void);
void iderw(struct buf*);
// idle.cc
extern struct proc *idlep[NCPU];
struct proc * idleproc(void);
void idlezombie(struct proc*);
// ioapic.c
void ioapicenable(int irq, int cpu);
......@@ -179,6 +180,7 @@ int pipewrite(struct pipe*, char*, int);
// proc.c
struct proc* allocproc(void);
struct proc* copyproc(struct proc*);
void finishproc(struct proc*);
void exit(void);
int fork(int);
int growproc(int);
......@@ -190,7 +192,7 @@ void userinit(void);
int wait(void);
void yield(void);
struct proc* threadalloc(void (*fn)(void*), void *arg);
void threadpin(void (*fn)(void*), void *arg, const char *name, int cpu);
struct proc* threadpin(void (*fn)(void*), void *arg, const char *name, int cpu);
// prof.c
extern int profenable;
......@@ -262,7 +264,7 @@ void freework(struct work *w);
// cilk.c
#if CILKENABLE
void cilk_push(void *rip, u64 arg0, u64 arg1);
void cilk_push(void (*fn)(uptr, uptr), u64 arg0, u64 arg1);
void cilk_start(void);
void cilk_end(void);
void cilk_dump(void);
......
include/lockstat.h
浏览文件 @ ff1eef00
......@@ -42,11 +42,13 @@ struct klockstat;
// Debug knobs
#define LOCKSTAT_BIO 1
#define LOCKSTAT_CILK 1
#define LOCKSTAT_CONDVAR 0
#define LOCKSTAT_CONSOLE 1
#define LOCKSTAT_CRANGE 1
#define LOCKSTAT_FS 1
#define LOCKSTAT_GC 1
#define LOCKSTAT_IDLE 1
#define LOCKSTAT_KALLOC 1
#define LOCKSTAT_KMALLOC 1
#define LOCKSTAT_NET 1
......
include/percpu.hh 0 → 100644
浏览文件 @ ff1eef00
template <typename T>
struct percpu {
int myid() {
return mycpu()->id;
}
const T* operator->() const {
return cpu(myid());
}
T* operator->() {
return cpu(myid());
}
T& operator*() {
return *cpu(myid());
}
T& operator[](int id) {
return *cpu(id);
}
const T& operator[](int id) const {
return *cpu(id);
}
T* cpu(int id) {
return &pad_[id].v_;
}
struct {
T v_ __mpalign__;
__padout__;
} pad_[NCPU];
};
kernel/async.cc
浏览文件 @ ff1eef00
......@@ -92,3 +92,4 @@ sys_async(int fd, size_t count, off_t off,
msg->submitted = 1;
return 0;
}
kernel/cilk.cc
浏览文件 @ ff1eef00
......@@ -36,6 +36,7 @@
#include "queue.h"
#include "proc.hh"
#include "mtrace.h"
#include "wq.hh"
#define NSLOTS (1 << CILKSHIFT)
......@@ -59,7 +60,6 @@ struct cilkthread {
struct cilkstat {
u64 push;
u64 full;
u64 pop;
u64 steal;
__padout__;
} __mpalign__;
......@@ -67,12 +67,6 @@ struct cilkstat {
static struct cilkqueue queue[NCPU] __mpalign__;
static struct cilkstat stat[NCPU] __mpalign__;
static struct cilkqueue *
cilk_cur(void)
{
return &queue[mycpu()->id];
}
static struct cilkframe *
cilk_frame(void)
{
......@@ -85,73 +79,20 @@ cilk_stat(void)
return &stat[mycpu()->id];
}
static int
__cilk_push(struct cilkqueue *q, struct cilkthread *t)
{
int i;
i = q->head;
if ((i - q->tail) == NSLOTS) {
cilk_stat()->full++;
return -1;
}
i = i & (NSLOTS-1);
q->thread[i] = t;
q->head++;
cilk_stat()->push++;
return 0;
}
static struct cilkthread *
__cilk_pop(struct cilkqueue *q)
{
int i;
acquire(&q->lock);
i = q->head;
if ((i - q->tail) == 0) {
release(&q->lock);
return 0;
}
i = (i-1) & (NSLOTS-1);
q->head--;
release(&q->lock);
cilk_stat()->pop++;
return q->thread[i];
}
static struct cilkthread *
__cilk_steal(struct cilkqueue *q)
{
int i;
acquire(&q->lock);
i = q->tail;
if ((i - q->head) == 0) {
release(&q->lock);
return 0;
}
i = i & (NSLOTS-1);
q->tail++;
release(&q->lock);
cilk_stat()->steal++;
return q->thread[i];
}
static void
__cilk_run(struct cilkthread *th)
__cilk_run(struct work *w, void *xfn, void *arg0, void *arg1, void *xframe)
{
void (*fn)(uptr arg0, uptr arg1) = (void(*)(uptr,uptr))th->rip;
void (*fn)(uptr arg0, uptr arg1) = (void(*)(uptr,uptr))xfn;
struct cilkframe *frame = (struct cilkframe *)xframe;
struct cilkframe *old = mycpu()->cilkframe;
mycpu()->cilkframe = th->frame;
fn(th->arg0, th->arg1);
if (old != frame)
cilk_stat()->steal++;
mycpu()->cilkframe = frame;
fn((uptr)arg0, (uptr)arg1);
mycpu()->cilkframe = old;
th->frame->ref--;
kfree(th);
frame->ref--;
}
// Add the (rip, arg0, arg1) work to the local work queue.
......@@ -160,56 +101,27 @@ __cilk_run(struct cilkthread *th)
void
cilk_push(void (*fn)(uptr, uptr), u64 arg0, u64 arg1)
{
struct cilkthread *th;
struct work *w;
th = (struct cilkthread *) kalloc();
if (th == nullptr) {
w = allocwork();
if (w == nullptr) {
fn(arg0, arg1);
return;
}
th->rip = (uptr) fn;
th->arg0 = arg0;
th->arg1 = arg1;
th->frame = cilk_frame();
if (__cilk_push(cilk_cur(), th)) {
kfree(th);
w->rip = (void*)__cilk_run;
w->arg0 = (void*)fn;
w->arg1 = (void*)arg0;
w->arg2 = (void*)arg1;
w->arg3 = (void*)cilk_frame();
if (wq_push(w)) {
freework(w);
fn(arg0, arg1);
} else
cilk_stat()->full++;
} else {
cilk_frame()->ref++;
}
// Try to execute one cilkthread.
// Check local queue then steal from other queues.
int
cilk_trywork(void)
{
struct cilkthread *th;
int i;
pushcli();
th = __cilk_pop(cilk_cur());
if (th != nullptr) {
__cilk_run(th);
popcli();
return 1;
}
// XXX(sbw) should be random
for (i = 0; i < NCPU; i++) {
if (i == mycpu()->id)
continue;
th = __cilk_steal(&queue[i]);
if (th != nullptr) {
__cilk_run(th);
popcli();
return 1;
}
cilk_stat()->push++;
}
popcli();
return 0;
}
// Start a new work queue frame.
......@@ -229,21 +141,9 @@ cilk_start(void)
void
cilk_end(void)
{
while (cilk_frame()->ref != 0) {
struct cilkthread *th;
int i;
while (cilk_frame()->ref != 0)
wq_trywork();
while ((th = __cilk_pop(cilk_cur())) != nullptr)
__cilk_run(th);
for (i = 0; i < NCPU; i++) {
th = __cilk_steal(&queue[i]);
if (th != nullptr) {
__cilk_run(th);
break;
}
}
}
mycpu()->cilkframe = 0;
popcli();
}
......@@ -253,8 +153,8 @@ cilk_dump(void)
{
int i;
for (i = 0; i < NCPU; i++)
cprintf("push %lu full %lu pop %lu steal %lu\n",
stat[i].push, stat[i].full, stat[i].pop, stat[i].steal);
cprintf("push %lu full %lu steal %lu\n",
stat[i].push, stat[i].full, stat[i].steal);
}
static void
......@@ -285,7 +185,7 @@ testcilk(void)
running = 0;
} else {
while (running)
cilk_trywork();
wq_trywork();
}
popcli();
}
......
kernel/idle.cc
浏览文件 @ ff1eef00
......@@ -6,8 +6,75 @@
#include "proc.hh"
#include "cpu.hh"
#include "sched.hh"
#include "percpu.hh"
struct proc *idlep[NCPU] __mpalign__;
struct idle {
struct proc *cur;
struct proc *heir;
SLIST_HEAD(zombies, proc) zombies;
struct spinlock lock;
};
static percpu<idle> idlem;
void idleloop(void);
struct proc *
idleproc(void)
{
assert(mycpu()->ncli > 0);
return idlem->cur;
}
void
idlezombie(struct proc *p)
{
acquire(&idlem[mycpu()->id].lock);
SLIST_INSERT_HEAD(&idlem[mycpu()->id].zombies, p, child_next);
release(&idlem[mycpu()->id].lock);
}
void
idlebequeath(void)
{
// Only the current idle thread may call this function
assert(mycpu()->ncli > 0);
assert(myproc() == idlem->cur);
assert(idlem->heir != nullptr);
idlem->cur = idlem->heir;
acquire(&idlem->heir->lock);
idlem->heir->set_state(RUNNABLE);
release(&idlem->heir->lock);
}
static void
idleheir(void *x)
{
post_swtch();
idlem->heir = nullptr;
idleloop();
}
static inline void
finishzombies(void)
{
struct idle *i = &idlem[mycpu()->id];
if (!SLIST_EMPTY(&i->zombies)) {
struct proc *p, *np;
acquire(&i->lock);
SLIST_FOREACH_SAFE(p, &i->zombies, child_next, np) {
SLIST_REMOVE(&i->zombies, p, proc, child_next);
finishproc(p);
}
release(&i->lock);
}
}
void
idleloop(void)
......@@ -29,11 +96,31 @@ idleloop(void)
myproc()->set_state(RUNNABLE);
sched();
finishzombies();
if (steal() == 0) {
int worked;
do {
assert(mycpu()->ncli == 0);
// If we don't have an heir, try to allocate one
if (idlem->heir == nullptr) {
struct proc *p;
p = allocproc();
if (p == nullptr)
break;
snprintf(p->name, sizeof(p->name), "idleh_%u", mycpu()->id);
p->cpuid = mycpu()->id;
p->cpu_pin = 1;
p->context->rip = (u64)idleheir;
p->cwd = nullptr;
idlem->heir = p;
}
worked = wq_trywork();
// If we are no longer the idle thread, exit
if (worked && idlem->cur != myproc())
exit();
} while(worked);
sti();
}
......@@ -43,13 +130,16 @@ idleloop(void)
void
initidle(void)
{
// allocate a fake PID for each scheduler thread
struct proc *p = allocproc();
if (!p)
panic("initidle allocproc");
SLIST_INIT(&idlem[cpunum()].zombies);
initlock(&idlem[cpunum()].lock, "idle_lock", LOCKSTAT_IDLE);
snprintf(p->name, sizeof(p->name), "idle_%u", cpunum());
mycpu()->proc = p;
myproc()->cpuid = cpunum();
myproc()->cpu_pin = 1;
idlep[cpunum()] = p;
idlem->cur = p;
}
kernel/main.cc
浏览文件 @ ff1eef00
......@@ -83,6 +83,7 @@ cmain(u64 mbmagic, u64 mbaddr)
initkalloc(mbaddr);
initproc(); // process table
initsched(); // scheduler run queues
initidle();
initgc(); // gc epochs and threads
initbio(); // buffer cache
initinode(); // inode cache
......@@ -95,7 +96,6 @@ cmain(u64 mbmagic, u64 mbaddr)
initlockstat();
initpci();
initnet();
initidle();
if (VERBOSE)
cprintf("ncpu %d %lu MHz\n", ncpu, cpuhz / 1000000);
......
kernel/proc.cc
浏览文件 @ ff1eef00
......@@ -152,9 +152,13 @@ exit(void)
// Parent might be sleeping in wait().
acquire(&(myproc()->lock));
// Kernel threads might not have a parent
if (myproc()->parent != nullptr)
cv_wakeup(&(myproc()->parent->cv));
cv_wakeup(&(myproc()->parent->cv));
else
idlezombie(myproc());
if (wakeupinit)
cv_wakeup(&bootproc->cv);
......@@ -458,6 +462,21 @@ fork(int flags)
return pid;
}
void
finishproc(struct proc *p)
{
ksfree(slab_stack, p->kstack);
p->kstack = 0;
p->vmap->decref();
if (!xnspid->remove(p->pid, &p))
panic("wait: ns_remove");
p->pid = 0;
p->parent = 0;
p->name[0] = 0;
p->killed = 0;
freeproc(p);
}
// Wait for a child process to exit and return its pid.
// Return -1 if this process has no children.
int
......@@ -478,16 +497,7 @@ wait(void)
pid = p->pid;
SLIST_REMOVE(&myproc()->childq, p, proc, child_next);
release(&myproc()->lock);
ksfree(slab_stack, p->kstack);
p->kstack = 0;
p->vmap->decref();
if (!xnspid->remove(p->pid, &p))
panic("wait: ns_remove");
p->pid = 0;
p->parent = 0;
p->name[0] = 0;
p->killed = 0;
freeproc(p);
finishproc(p);
return pid;
}
release(&p->lock);
......@@ -533,12 +543,12 @@ threadalloc(void (*fn)(void *), void *arg)
p->context->rip = (u64)threadstub;
p->context->r12 = (u64)fn;
p->context->r13 = (u64)arg;
p->parent = myproc();
p->cwd = 0;
p->parent = nullptr;
p->cwd = nullptr;
return p;
}
void
struct proc*
threadpin(void (*fn)(void*), void *arg, const char *name, int cpu)
{
struct proc *p;
......@@ -553,4 +563,5 @@ threadpin(void (*fn)(void*), void *arg, const char *name, int cpu)
acquire(&p->lock);
addrun(p);
release(&p->lock);
return p;
}
kernel/sched.cc
浏览文件 @ ff1eef00
......@@ -26,7 +26,7 @@ void
post_swtch(void)
{
if (mycpu()->prev->get_state() == RUNNABLE &&
mycpu()->prev != idlep[mycpu()->id])
mycpu()->prev != idleproc())
addrun(mycpu()->prev);
release(&mycpu()->prev->lock);
}
......@@ -42,6 +42,11 @@ sched(void)
if(!holding(&myproc()->lock))
panic("sched proc->lock");
#endif
if (myproc() == idleproc() &&
myproc()->get_state() != RUNNABLE) {
extern void idlebequeath(void);
idlebequeath();
}
if(mycpu()->ncli != 1)
panic("sched locks");
if(myproc()->get_state() == RUNNING)
......@@ -59,7 +64,7 @@ sched(void)
struct proc *next = schednext();
if (next == nullptr) {
if (myproc()->get_state() != RUNNABLE) {
next = idlep[mycpu()->id];
next = idleproc();
} else {
myproc()->set_state(RUNNING);
mycpu()->intena = intena;
......
kernel/wq.cc
浏览文件 @ ff1eef00
......@@ -175,11 +175,11 @@ wq_trywork(void)
struct work *w;
int i;
pushcli();
//pushcli();
w = __wq_pop(mycpu()->id);
if (w != nullptr) {
__wq_run(w);
popcli();
//popcli();
return 1;
}
// XXX(sbw) should be random
......@@ -190,12 +190,12 @@ wq_trywork(void)
w = __wq_steal(i);
if (w != nullptr) {
__wq_run(w);
popcli();
//popcli();
return 1;
}
}
popcli();
//popcli();
return 0;
}
......
param.h
浏览文件 @ ff1eef00
......@@ -24,6 +24,7 @@
#define ALLOC_MEMSET DEBUG
#define KSHAREDSIZE (32 << 10)
#define WQSHIFT 7
#define CILKENABLE 0
#if defined(HW_josmp)
#define NCPU 16 // maximum number of CPUs
#define MTRACE 0
......
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论