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

c132eade · Nickolai Zeldovich · b002deab · a0943975 · c132eade · c132eade
--- a/bin/pls.cc
+++ b/bin/pls.cc
+#include "types.h"
+#include "stat.h"
+#include "user.h"
+#include "fs.h"
+const char*
+fmtname(const char *path)
+{
+  static char buf[DIRSIZ+1];
+  const char *p;
+  // Find first character after last slash.
+  for(p=path+strlen(path); p >= path && *p != '/'; p--)
+    ;
+  p++;
+  // Return blank-padded name.
+  if(strlen(p) >= DIRSIZ)
+    return p;
+  memmove(buf, p, strlen(p));
+  memset(buf+strlen(p), ' ', DIRSIZ-strlen(p));
+  return buf;
+}
+void
+ls(const char *path)
+{
+  char buf[512], *p;
+  int fd;
+  struct dirent de;
+  struct stat st;
+  if((fd = open(path, 0)) < 0){
+    fprintf(2, "ls: cannot open %s\n", path);
+    return;
+  }
+  if(fstat(fd, &st) < 0){
+    fprintf(2, "ls: cannot stat %s\n", path);
+    close(fd);
+    return;
+  }
+  switch(st.type){
+  case T_FILE:
+    fprintf(1, "%s %d %d %d\n", fmtname(path), st.type, st.ino, st.size);
+    break;
+  case T_DIR:
+    if(strlen(path) + 1 + DIRSIZ + 1 > sizeof buf){
+      fprintf(1, "ls: path too long\n");
+      break;
+    }
+    strcpy(buf, path);
+    p = buf+strlen(buf);
+    *p++ = '/';
+    while(read(fd, &de, sizeof(de)) == sizeof(de)){
+      if(de.inum == 0)
+        continue;
+      memmove(p, de.name, DIRSIZ);
+      p[DIRSIZ] = 0;
+      if(stat(buf, &st) < 0){
+        fprintf(1, "ls: cannot stat %s\n", buf);
+        continue;
+      }
+      fprintf(1, "%s %d %d %d\n", fmtname(buf), st.type, st.ino, st.size);
+    }
+    break;
+  }
+  close(fd);
+}
+void
+work(void *arg)
+{
+  u64 tid = (u64)arg;
+  // grab and push work  (may divide into blocks? and call ls on a block)?
+  // maybe implement getdirent sys call that gives you some unread dir entry
+}
+int
+main(int argc, char *argv[])
+{
+  int i;
+  int nthread = 4;
+  for(int i = 0; i < nthread; i++) {
+    sbrk(8192);
+    void *tstack = sbrk(0);
+    // fprintf(1, "tstack %lx\n", tstack);
+    int tid = forkt(tstack, (void*) thr, (void *)(u64)i);
+    if (0) fprintf(1, "pls[%d]: child %d\n", getpid(), tid);
+  }
+  // push work wq.cc
+  if(argc < 2){
+    ls(".");
+    exit();
+  }
+  for(i=1; i<argc; i++)
+    ls(argv[i]);
+  for(int i = 0; i < nthread; i++)
+    wait();
+  exit();
+}
--- a/include/kernel.hh
+++ b/include/kernel.hh
@@ -263,54 +263,27 @@ struct work *   allocwork(void);
 void            freework(struct work *w);
 // cilk.c
+void            initcilkframe(struct cilkframe*);
 #if CILKENABLE
 void            cilk_push(void (*fn)(uptr, uptr), u64 arg0, u64 arg1);
 void            cilk_start(void);
-void            cilk_end(void);
+u64             cilk_end(void);
 void            cilk_dump(void);
-int             cilk_trywork(void);
+void            cilk_abort(u64 val);
-void            initcilkframe(struct cilkframe *wq);
 #else
 #define cilk_push(rip, arg0, arg1) do { \
  void (*fn)(uptr, uptr) = rip; \
  fn(arg0, arg1); \
 } while(0)
 #define cilk_start() do { } while(0)
-#define cilk_end() do { } while(0)
+#define cilk_end() (myproc()->cilkframe.abort)
 #define cilk_dump() do { } while(0)
-#define cilk_trywork() 0
+#define cilk_abort(val) do { \
-#define initcilkframe(x) do { } while (0)
+  cmpxch(&myproc()->cilkframe.abort, (u64)0, (u64)val);   \
+} while (0)
 #endif
-// various init functions
-void initpic(void);
-void initioapic(void);
-void inituart(void);
-void initcga(void);
-void initconsole(void);
-void initpg(void);
-void initmp(void);
-void initlapic(void);
-void inittls(void);
-void initnmi(void);
-void inittrap(void);
-void initseg(void);
-void initkalloc(u64 mbaddr);
-void initrcu(void);
-void initproc(void);
-void initbio(void);
-void initinode(void);
-void initdisk(void);
-void inituser(void);
-void initcilk(void);
-void initsamp(void);
-void initpci(void);
-void initnet(void);
-void initsched(void);
-void initlockstat(void);
-void initwq(void);
-void initsperf(void);
 // other exported/imported functions
 void cmain(u64 mbmagic, u64 mbaddr);
 void mpboot(void);

--- a/include/mmu.h
+++ b/include/mmu.h
@@ -24,6 +24,7 @@
 #define PGROUNDUP(sz)  (((sz)+PGSIZE-1) & ~(PGSIZE-1))
 #define PGROUNDDOWN(a) ((__typeof__(a))((((uptr)(a)) & ~(PGSIZE-1))))
+#define PGOFFSET(a) ((a) & ((1<<PGSHIFT)-1))
 // Address in page table or page directory entry
 #define PTE_ADDR(pte)	((uptr)(pte) & ~0xFFF)

--- a/include/proc.hh
+++ b/include/proc.hh
@@ -19,6 +19,7 @@ struct context {
 // Work queue frame
 struct cilkframe {
  volatile std::atomic<u64> ref;
+  volatile std::atomic<u64> abort;
 };
 // Per-process, per-stack meta data for mtrace

--- a/kernel/cilk.cc
+++ b/kernel/cilk.cc
-// cilk style run queue
+// cilk style run queue built on wq.cc:
-// A work queue is built from NCPU per-core wqueues.
+//   A core pushes work to the head of its per-core wq.
-// A core pushes work to the head of its per-core wqueue.
+//   A core pops work from the head of its per-core wq.
-// A core pops work from the head of its per-core wqueue.
 //   A core pops work from the tail of another core's per-core wqueue.
 //
 // Usage:
@@ -25,7 +24,6 @@
 //     cprintf("%c %c\n", arg[0], arg[1]);
 //   }
-#if CILKENABLE
 #include "types.h"
 #include "kernel.hh"
 #include "amd64.h"
@@ -35,37 +33,18 @@
 #include "condvar.h"
 #include "queue.h"
 #include "proc.hh"
+#if CILKENABLE
 #include "mtrace.h"
 #include "wq.hh"
+#include "percpu.hh"
-#define NSLOTS (1 << CILKSHIFT)
-struct cilkqueue {
-  struct cilkthread *thread[NSLOTS];
-  volatile int head __mpalign__;
-  struct spinlock lock;
-  volatile int tail;
-  __padout__;
-} __mpalign__;
-struct cilkthread {
-  u64 rip;
-  u64 arg0;
-  u64 arg1;
-  struct cilkframe *frame;  // parent cilkframe
-  __padout__;
-} __mpalign__;
 struct cilkstat {
  u64 push;
  u64 full;
  u64 steal;
-  __padout__;
+};
-} __mpalign__;
+static percpu<cilkstat> stat;
-static struct cilkqueue queue[NCPU] __mpalign__;
-static struct cilkstat stat[NCPU] __mpalign__;
 static struct cilkframe *
 cilk_frame(void)
@@ -73,12 +52,6 @@ cilk_frame(void)
  return mycpu()->cilkframe;
 }
-static struct cilkstat *
-cilk_stat(void)
-{
-  return &stat[mycpu()->id];
-}
 static void
 __cilk_run(struct work *w, void *xfn, void *arg0, void *arg1, void *xframe)
 {
@@ -87,9 +60,10 @@ __cilk_run(struct work *w, void *xfn, void *arg0, void *arg1, void *xframe)
  struct cilkframe *old = mycpu()->cilkframe;
  if (old != frame)
-    cilk_stat()->steal++;
+    stat->steal++;
  mycpu()->cilkframe = frame;
+  if (frame->abort == 0)
    fn((uptr)arg0, (uptr)arg1);
  mycpu()->cilkframe = old;
  frame->ref--;
@@ -117,10 +91,10 @@ cilk_push(void (*fn)(uptr, uptr), u64 arg0, u64 arg1)
  if (wq_push(w)) {
    freework(w);
    fn(arg0, arg1);
-    cilk_stat()->full++;
+    stat->full++;
  } else {
    cilk_frame()->ref++;
-    cilk_stat()->push++;
+    stat->push++;
  }
 }
@@ -138,14 +112,26 @@ cilk_start(void)
 // End of the current work queue frame.
 // The core works while the reference count of the current
 // work queue frame is not 0.
-void
+u64
 cilk_end(void)
 {
+  u64 r;
  while (cilk_frame()->ref != 0)
    wq_trywork();
+  r = cilk_frame()->abort;
  mycpu()->cilkframe = 0;
  popcli();
+  return r;
+}
+void
+cilk_abort(u64 val)
+{
+  cmpxch(&cilk_frame()->abort, (u64)0, val);
 }
 void
@@ -189,19 +175,11 @@ testcilk(void)
  }
  popcli();
 }
+#endif // CILKENABLE
 void
 initcilkframe(struct cilkframe *cilk)
 {
-  memset(cilk, 0, sizeof(*cilk));
+  cilk->ref = 0;
-}
+  cilk->abort = 0;
-void
-initcilk(void)
-{
-  int i;
-  for (i = 0; i < NCPU; i++)
-    initlock(&queue[i].lock, "queue lock", LOCKSTAT_CILK);
 }
-#endif // CILKENABLE
--- a/kernel/exec.cc
+++ b/kernel/exec.cc
@@ -34,6 +34,10 @@ dosegment(uptr a0, u64 a1)
  u64 off = a1;
  struct vmnode *vmn = nullptr;
  struct proghdr ph;
+  uptr va_start, va_end;
+  uptr in_off;
+  uptr in_sz;
+  int npg;
  prof_start(dosegment_prof);
  if(readi(args->ip, (char*)&ph, off, sizeof(ph)) != sizeof(ph))
@@ -42,33 +46,30 @@ dosegment(uptr a0, u64 a1)
    goto bad;
  if(ph.memsz < ph.filesz)
    goto bad;
-  // XXX(sbw) vaddr doesn't have to be page aligned..
+  if (ph.offset < PGOFFSET(ph.vaddr))
-  if(ph.vaddr % PGSIZE) {
-    cprintf("unaligned ph.va\n");
    goto bad;
-  }
-  {
+  va_start = PGROUNDDOWN(ph.vaddr);
-    uptr va_start = PGROUNDDOWN(ph.vaddr);
+  va_end = PGROUNDUP(ph.vaddr + ph.memsz);
-    uptr va_end = PGROUNDUP(ph.vaddr + ph.memsz);
+  in_off = ph.offset - PGOFFSET(ph.vaddr);
+  in_sz = ph.filesz + PGOFFSET(ph.vaddr);
-    int npg = (va_end - va_start) / PGSIZE;
+  npg = (va_end - va_start) / PGSIZE;
  if ((vmn = new vmnode(npg, odp ? ONDEMAND : EAGER,
-                          args->ip, ph.offset, ph.filesz)) == 0)
+                        args->ip, in_off, in_sz)) == 0)
    goto bad;
-    if(args->vmap->insert(vmn, ph.vaddr, 1) < 0)
+  if(args->vmap->insert(vmn, va_start, 1) < 0)
    goto bad;
  prof_end(dosegment_prof);
  return;
-  }
 bad:
-  panic("dosegment: Oops");
+  cilk_abort(-1);
 }
-static void dostack(uptr a0, u64 a1)
+static void  __attribute__((unused)) dostack(uptr a0, u64 a1)
 {
  struct vmnode *vmn = nullptr;
  struct eargs *args = (eargs*) a0;  
@@ -110,18 +111,19 @@ static void dostack(uptr a0, u64 a1)
  for(last=s=args->path; *s; s++)
    if(*s == '/')
      last = s+1;
-  safestrcpy(args->proc->name, last, sizeof(args->proc->name));
+  // XXX(sbw) Oops, don't want to do this, unless we have abort
+  safestrcpy(args->proc->name, last, sizeof(args->proc->name));
  args->proc->tf->rsp = sp;
  prof_end(dostack_prof);
  return;
 bad:
-  panic("dostack: Oops");
+  cilk_abort(-1);
 }
-static void doheap(uptr a0, u64 a1)
+static void __attribute__((unused)) doheap(uptr a0, u64 a1)
 {
  struct vmnode *vmn = nullptr;
  struct eargs *args = (eargs*) a0;
@@ -138,7 +140,7 @@ static void doheap(uptr a0, u64 a1)
  return;
 bad:
-  panic("doheap: Oops");
+  cilk_abort(-1);
 }
 int
@@ -204,7 +206,8 @@ exec(const char *path, char **argv)
  //cilk_push(dostack, (uptr)&args, (uptr)0);
  dostack((uptr)&args, (uptr)0);
-  cilk_end();
+  if (cilk_end())
+    goto bad;
  // Commit to the user image.
  oldvmap = myproc()->vmap;

--- a/kernel/main.cc
+++ b/kernel/main.cc
@@ -9,8 +9,34 @@
 #include "condvar.h"
 #include "proc.hh"
-extern void initidle(void);
+void initpic(void);
-extern void idleloop(void);
+void initioapic(void);
+void inituart(void);
+void initcga(void);
+void initconsole(void);
+void initpg(void);
+void initmp(void);
+void initlapic(void);
+void inittls(void);
+void initnmi(void);
+void inittrap(void);
+void initseg(void);
+void initkalloc(u64 mbaddr);
+void initrcu(void);
+void initproc(void);
+void initbio(void);
+void initinode(void);
+void initdisk(void);
+void inituser(void);
+void initsamp(void);
+void initpci(void);
+void initnet(void);
+void initsched(void);
+void initlockstat(void);
+void initwq(void);
+void initsperf(void);
+void initidle(void);
+void idleloop(void);
 static volatile int bstate;
@@ -89,9 +115,6 @@ cmain(u64 mbmagic, u64 mbaddr)
  initinode();     // inode cache
  initdisk();      // disk
  initwq();
-#if CILKENABLE
-  initcilk();
-#endif
  initsamp();
  initlockstat();
  initpci();

--- a/kernel/wq.cc
+++ b/kernel/wq.cc
@@ -109,10 +109,18 @@ wq_push2(void (*fn)(struct work*, void*, void*), void *a0, void *a1)
  return 0;
 }
-static struct work *
+static void
+__wq_run(struct work *w)
+{
+  void (*fn)(struct work*, void*, void*, void*, void*, void*) = 
+    (void(*)(work*,void*,void*,void*,void*,void*))w->rip;
+  fn(w, w->arg0, w->arg1, w->arg2, w->arg3, w->arg4);
+  freework(w);
+}
+static inline struct work *
 __wq_pop(int c)
 {
-  // Called with cli
  struct wqueue *wq = &queue[c];
  struct work *w;
  int i;
@@ -136,10 +144,9 @@ __wq_pop(int c)
  return w;
 }
-static struct work *
+static inline struct work *
 __wq_steal(int c)
 {
-  // Called with cli
  struct wqueue *wq = &queue[c];
  struct work *w;
  int i;
@@ -160,42 +167,34 @@ __wq_steal(int c)
  return w;
 }
-static void
-__wq_run(struct work *w)
-{
-  void (*fn)(struct work*, void*, void*, void*, void*, void*) = 
-    (void(*)(work*,void*,void*,void*,void*,void*))w->rip;
-  fn(w, w->arg0, w->arg1, w->arg2, w->arg3, w->arg4);
-  freework(w);
-}
 int
 wq_trywork(void)
 {
  struct work *w;
-  int i;
+  int i, k;
+  // A "random" victim CPU
+  k = rdtsc();
-  //pushcli();
  w = __wq_pop(mycpu()->id);
  if (w != nullptr) {
    __wq_run(w);
-    //popcli();
    return 1;
  }
-  // XXX(sbw) should be random
  for (i = 0; i < NCPU; i++) {
-    if (i == mycpu()->id)
+    int j = (i+k) % NCPU;
+    if (j == mycpu()->id)
        continue;
-    w = __wq_steal(i);
+    w = __wq_steal(j);
    if (w != nullptr) {
      __wq_run(w);
-      //popcli();
      return 1;
    }
  }
-  //popcli();
  return 0;
 }