runtime: don't touch pages of memory unnecessarily.

cuts working size for hello world from 6 MB to 1.2 MB.
still some work to be done, but diminishing returns.

R=r
https://golang.org/cl/165080

8 files changed, 15 insertions, 45 deletions

diff --git a/src/pkg/runtime/malloc.h b/src/pkg/runtime/malloc.h
index 2e0f1143dd..b3fa8e0dfb 100644
--- a/src/pkg/runtime/malloc.h
+++ b/src/pkg/runtime/malloc.h
@@ -113,8 +113,9 @@ struct MLink
 	MLink *next;
 };
 
-// SysAlloc obtains a large chunk of memory from the operating system,
-// typically on the order of a hundred kilobytes or a megabyte.
+// SysAlloc obtains a large chunk of zeroed memory from the
+// operating system, typically on the order of a hundred kilobytes
+// or a megabyte.
 //
 // SysUnused notifies the operating system that the contents
 // of the memory region are no longer needed and can be reused
@@ -312,4 +313,3 @@ enum
 	RefSome,		// some references
 	RefNoPointers = 0x80000000U,	// flag - no pointers here     
 };
-
diff --git a/src/pkg/runtime/mcentral.c b/src/pkg/runtime/mcentral.c
index 5c9f720c09..9881812e32 100644
--- a/src/pkg/runtime/mcentral.c
+++ b/src/pkg/runtime/mcentral.c
@@ -40,7 +40,6 @@ MCentral_AllocList(MCentral *c, int32 n, MLink **pfirst)
 	MLink *first, *last, *v;
 	int32 i;
 
-
 	lock(c);
 	// Replenish central list if empty.
 	if(MSpanList_IsEmpty(&c->nonempty)) {
diff --git a/src/pkg/runtime/mgc0.c b/src/pkg/runtime/mgc0.c
index d01429f349..f0eafe3fd6 100644
--- a/src/pkg/runtime/mgc0.c
+++ b/src/pkg/runtime/mgc0.c
@@ -91,8 +91,11 @@ mark(void)
 {
 	G *gp;
 
-	// mark data+bss
-	scanblock(0, data, end - data);
+	// mark data+bss.
+	// skip mheap itself, which has no interesting pointers
+	// and is mostly zeroed and would not otherwise be paged in.
+	scanblock(0, data, (byte*)&mheap - data);
+	scanblock(0, (byte*)(&mheap+1), end - (byte*)(&mheap+1));
 
 	// mark stacks
 	for(gp=allg; gp!=nil; gp=gp->alllink) {
diff --git a/src/pkg/runtime/mheap.c b/src/pkg/runtime/mheap.c
index 8661bd2a1b..e78c860c31 100644
--- a/src/pkg/runtime/mheap.c
+++ b/src/pkg/runtime/mheap.c
@@ -194,7 +194,6 @@ MHeap_Grow(MHeap *h, uintptr npage)
 	// NOTE(rsc): In tcmalloc, if we've accumulated enough
 	// system allocations, the heap map gets entirely allocated
 	// in 32-bit mode.  (In 64-bit mode that's not practical.)
-
 	if(!MHeapMap_Preallocate(&h->map, ((uintptr)v>>PageShift) - 1, (ask>>PageShift) + 2)) {
 		SysFree(v, ask);
 		return false;
diff --git a/src/pkg/runtime/mheapmap32.c b/src/pkg/runtime/mheapmap32.c
index 8cea825c33..13491595da 100644
--- a/src/pkg/runtime/mheapmap32.c
+++ b/src/pkg/runtime/mheapmap32.c
@@ -84,7 +84,6 @@ MHeapMap_Preallocate(MHeapMap *m, PageID k, uintptr len)
 			p2 = m->allocator(sizeof *p2);
 			if(p2 == nil)
 				return false;
-			runtime_memclr((byte*)p2, sizeof *p2);
 			m->p[i1] = p2;
 		}
 
diff --git a/src/pkg/runtime/mheapmap64.c b/src/pkg/runtime/mheapmap64.c
index 7930291786..97e20b6d20 100644
--- a/src/pkg/runtime/mheapmap64.c
+++ b/src/pkg/runtime/mheapmap64.c
@@ -96,7 +96,6 @@ MHeapMap_Preallocate(MHeapMap *m, PageID k, uintptr len)
 			p2 = m->allocator(sizeof *p2);
 			if(p2 == nil)
 				return false;
-			runtime_memclr((byte*)p2, sizeof *p2);
 			m->p[i1] = p2;
 		}
 
@@ -105,7 +104,6 @@ MHeapMap_Preallocate(MHeapMap *m, PageID k, uintptr len)
 			p3 = m->allocator(sizeof *p3);
 			if(p3 == nil)
 				return false;
-			runtime_memclr((byte*)p3, sizeof *p3);
 			p2->p[i2] = p3;
 		}
 
diff --git a/src/pkg/runtime/proc.c b/src/pkg/runtime/proc.c
index 2f36868f5f..f04cb66928 100644
--- a/src/pkg/runtime/proc.c
+++ b/src/pkg/runtime/proc.c
@@ -102,9 +102,10 @@ schedinit(void)
 	mallocinit();
 	goargs();
 
+	// For debugging:
 	// Allocate internal symbol table representation now,
 	// so that we don't need to call malloc when we crash.
-	findfunc(0);
+	// findfunc(0);
 
 	sched.gomaxprocs = 1;
 	p = getenv("GOMAXPROCS");
diff --git a/src/pkg/runtime/symtab.c b/src/pkg/runtime/symtab.c
index 18ce07a1cf..0b5499474f 100644
--- a/src/pkg/runtime/symtab.c
+++ b/src/pkg/runtime/symtab.c
@@ -24,40 +24,6 @@
 #define SYMDATA ((byte*)(0x99LL<<24) + 8)
 #endif
 
-
-// Return a pointer to a byte array containing the symbol table segment.
-void
-runtime·symdat(Slice *symtab, Slice *pclntab)
-{
-	Slice *a;
-	int32 *v;
-
-	// TODO(rsc): Remove once TODO at top of file is done.
-	if(goos != nil && strcmp((uint8*)goos, (uint8*)"nacl") == 0) {
-		symtab = mal(sizeof *a);
-		pclntab = mal(sizeof *a);
-		FLUSH(&symtab);
-		FLUSH(&pclntab);
-		return;
-	}
-
-	v = SYMCOUNTS;
-
-	a = mal(sizeof *a);
-	a->len = v[0];
-	a->cap = a->len;
-	a->array = SYMDATA;
-	symtab = a;
-	FLUSH(&symtab);
-
-	a = mal(sizeof *a);
-	a->len = v[1];
-	a->cap = a->len;
-	a->array = SYMDATA + v[0];
-	pclntab = a;
-	FLUSH(&pclntab);
-}
-
 typedef struct Sym Sym;
 struct Sym
 {
@@ -122,6 +88,8 @@ static int32 nfunc;
 static byte **fname;
 static int32 nfname;
 
+static Lock funclock;
+
 static void
 dofunc(Sym *sym)
 {
@@ -379,8 +347,11 @@ findfunc(uintptr addr)
 	Func *f;
 	int32 nf, n;
 
+	lock(&funclock);
 	if(func == nil)
 		buildfuncs();
+	unlock(&funclock);
+
 	if(nfunc == 0)
 		return nil;
 	if(addr < func[0].entry || addr >= func[nfunc].entry)