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

84865c8b · Frans Kaashoek · af23ec9e · dbf9bbef · 84865c8b · 84865c8b
--- a/include/filetable.hh
+++ b/include/filetable.hh
@@ -42,6 +42,10 @@ public:
    // XXX(sbw) if f->ref_ > 1 the kernel will not actually close 
    // the file when this function returns (i.e. sys_close can return 
    // while the file/pipe/socket is still open).
+    if (fd >= NOFILE) {
+      cprintf("filetable::close: bad fd %u\n", fd);
+      return;
+    }

    file* f = ofile_[fd].exchange(nullptr);
    if (f != nullptr)

--- a/include/markptr.hh
+++ b/include/markptr.hh
@@ -25,10 +25,16 @@ class markptr {
  markptr_ptr<T>& ptr() {
    return *(markptr_ptr<T>*) this;
  }
+  const markptr_ptr<T>& ptr() const {
+    return *(const markptr_ptr<T>*) this;
+  }

  markptr_mark<T>& mark() {
    return *(markptr_mark<T>*) this;
  }
+  const markptr_mark<T>& mark() const {
+    return *(const markptr_mark<T>*) this;
+  }

  // Convenience operator to avoid having to write out xx.ptr()->...
  T* operator->() { return ptr(); }

--- a/include/radix.hh
+++ b/include/radix.hh
@@ -78,6 +78,39 @@ struct radix_iterator {
    return r_ != other.r_ || k_ != other.k_; }
 };

+struct radix_iterator2 {
+  const radix* r_;
+  u64 k_;
+  // path_[i] is the node at level i. Note that the leaf is at zero
+  // and is radix_elem. The rest are radix_node. For now we assume all
+  // leaves are at level 0. Later we'll steal a bit for them. The root
+  // is path_[radix_levels].
+  void *path_[radix_levels+1];
+
+  radix_iterator2(const radix* r, u64 k);
+
+  radix_iterator2 &operator++() {
+    if (!advance(radix_levels-1)) k_ = ~0ULL;
+    return *this;
+  }
+  radix_elem* operator*() {
+    return (radix_elem*)path_[0];
+  }
+  radix_node* node(u32 level) { return (radix_node*)path_[level]; }
+
+  // Compare equality on just the key.
+  bool operator==(const radix_iterator2 &other) {
+    return r_ == other.r_ && k_ == other.k_; }
+  bool operator!=(const radix_iterator2 &other) {
+    return r_ != other.r_ || k_ != other.k_; }
+
+private:
+  bool find_first_leaf(u32 level);
+  bool advance(u32 level);
+};
+
+#define radix_iterator radix_iterator2
+
 static inline radix_iterator
 begin(const radix &r) { return radix_iterator(&r, 0); }

@@ -91,3 +124,4 @@ begin(const radix_range &rr) { return radix_iterator(rr.r_, rr.start_); }
 static inline radix_iterator
 end(const radix_range &rr) { return radix_iterator(rr.r_, rr.start_ + rr.size_); }

+#undef radix_iterator
--- a/kernel/idle.cc
+++ b/kernel/idle.cc
@@ -93,12 +93,10 @@ idleloop(void)
  mtstart(idleloop, myproc());

  sti();
-  sampidle(true);
  for (;;) {
    acquire(&myproc()->lock);
    myproc()->set_state(RUNNABLE);
    sched();
-    sampidle(true);

    finishzombies();


--- a/kernel/radix.cc
+++ b/kernel/radix.cc
 #include "crange_arch.hh"
 #include "radix.hh"

+enum { crange_debug = 0 };
+
+#define dprintf(...) do { if (crange_debug) cprintf(__VA_ARGS__); } while(0)
+
+// Returns the index needed to reach |level| from |level+1|.
+static u32
+index(u64 key, u32 level)
+{
+  u64 idx = key >> (bits_per_level * level);
+  idx &= (1 << bits_per_level) - 1;
+  return idx;
+}
+
 // Returns the level we stopped at.
 template<class CB>
 u32
@@ -25,9 +38,7 @@ descend(u64 key, markptr<void> *n, u32 level, CB cb, bool create)

  radix_node *rn = (radix_node*) v;

-  u64 idx = key >> (bits_per_level * level);
-  idx &= (1<<bits_per_level)-1;
-  markptr<void> *vptr = &rn->ptr[idx];
+  markptr<void> *vptr = &rn->ptr[index(key, level)];
  if (level == 0) {
    cb(vptr);
    return level;
@@ -43,6 +54,7 @@ radix::search(u64 key)
  descend(key >> shift_, &root_, radix_levels-1, [&result](markptr<void> *v) {
      result = (radix_elem*) v->ptr().load();
    }, false);
+  dprintf("%p: search(%lu) -> %p\n", this, key >> shift_, result);
  return result;
 }

@@ -104,6 +116,7 @@ radix_range::replace(u64 start, u64 size, radix_elem *val)
 {
  start = start >> r_->shift_;
  size = size >> r_->shift_;
+  dprintf("%p: replace: [%lu, %lu) with %p\n", r_, start, start + size, val);

  assert(start >= start_);
  assert(start + size <= start_ + size_);
@@ -132,3 +145,59 @@ radix_iterator::operator*()
    }, false);
  return result;
 }
+
+radix_iterator2::radix_iterator2(const radix* r, u64 k)
+  : r_(r), k_(k) {
+  dprintf("%p: Made iterator with k = %lu\n", r_, k_);
+  if (k_ != ~0ULL) {
+    path_[radix_levels] = r_->root_.ptr().load();
+    if (!find_first_leaf(radix_levels - 1))
+      k_ = ~0ULL;
+  }
+  dprintf("%p: Adjusted: k = %lu\n", r_, k_);
+}
+
+bool
+radix_iterator2::advance(u32 level)
+{
+  // First, see if we can advance a lower level.
+  if (level > 0 && advance(level-1)) {
+    // Nothing more to do.
+    return true;
+  }
+
+  // Try to advance this level, if we can.
+  u32 start_idx = index(k_, level)+1;
+  if (start_idx < (1<<bits_per_level)) {
+    // Find the first leaf starting at our sibling node.
+    k_ &= ~((1ULL<<((level+1) * bits_per_level)) - 1);
+    k_ |= (u64(start_idx) << (level * bits_per_level));
+    return find_first_leaf(level);
+  } else {
+    return false;
+  }
+}
+
+bool
+radix_iterator2::find_first_leaf(u32 level)
+{
+  // Find the first non-empty node after k_ on this level.
+  for (u32 idx = index(k_, level); idx < (1<<bits_per_level); idx++) {
+    void *next = node(level+1)->ptr[idx].ptr().load();
+    if (next != nullptr) {
+      if (index(k_, level) != idx) {
+        // We had to advance; clear everything this level and under
+        // and set this one.
+        k_ &= ~((1ULL<<((level+1) * bits_per_level)) - 1);
+        k_ |= (u64(idx) << (level * bits_per_level));
+      }
+      path_[level] = next;
+
+      if (level == 0 || find_first_leaf(level-1))
+        return true;
+    }
+  }
+
+  // Failed to find a leaf. Abort.
+  return false;
+}
--- a/kernel/sampler.cc
+++ b/kernel/sampler.cc
@@ -26,7 +26,6 @@ struct pmu pmu;
 struct pmulog {
  u64 count;
  u64 capacity;
-  u8 idle:1;                    // Currently idle?
  struct pmuevent *event;
  __padout__;
 } __mpalign__;
@@ -72,12 +71,6 @@ sampdump(void)
 }

 void
-sampidle(bool b)
-{
-  pmulog[myid()].idle = b;
-}
-
-void
 sampconf(void)
 {
  pushcli();
@@ -112,7 +105,7 @@ samplog(struct trapframe *tf)

  e = &l->event[l->count];

-  e->idle = l->idle;
+  e->idle = (myproc() == idleproc());
  e->rip = tf->rip;
  getcallerpcs((void*)tf->rbp, e->trace, NELEM(e->trace));
  l->count++;

--- a/kernel/sched.cc
+++ b/kernel/sched.cc
@@ -74,8 +74,6 @@ sched(void)
      release(&myproc()->lock);
      return;
    }
-  } else {
-    sampidle(false);
  }

  if (next->get_state() != RUNNABLE)

--- a/kernel/uwq.cc
+++ b/kernel/uwq.cc
@@ -123,7 +123,6 @@ uwq::alloc(vmap* vmap, filetable *ftable)
  }

  if (mapkva(vmap->pml4, (char*)ipc, USERWQ, USERWQSIZE)) {
-    ksfree(slab_userwq, ipc);
    u->dec();
    return nullptr;
  }

--- a/kernel/vm.cc
+++ b/kernel/vm.cc
@@ -685,12 +685,12 @@ vmap::sbrk(ssize_t n, uptr *addr)
  auto curbrk = brk_;
  *addr = curbrk;

-  if(n < 0 && 0 - n <= curbrk){
+  if(n <= 0 && 0 - n <= curbrk){
    brk_ += n;
    return 0;
  }

-  if(n < 0 || n > USERTOP || curbrk + n > USERTOP)
+  if(n <= 0 || n > USERTOP || curbrk + n > USERTOP)
    return -1;

  // look one page ahead, to check if the newly allocated region would