diff --git a/compileopts/config.go b/compileopts/config.go
index 944322eb69..fe3a189fa3 100644
--- a/compileopts/config.go
+++ b/compileopts/config.go
@@ -549,6 +549,9 @@ func (c *Config) LDFlags() []string {
 	if c.Target.LinkerScript != "" {
 		ldflags = append(ldflags, "-T", c.Target.LinkerScript)
 	}
+	if c.Target.Linker == "wasm-ld" && c.Scheduler() == "none" && c.Options.StackSize != 0 {
+		ldflags = append(ldflags, "-z", fmt.Sprintf("stack-size=%d", c.Options.StackSize))
+	}
 	ldflags = append(ldflags, c.Options.ExtLDFlags...)
 
 	return ldflags
diff --git a/src/internal/poll/fd_wasip1.go b/src/internal/poll/fd_wasip1.go
index 5dcddbc26d..8ccf4020f9 100644
--- a/src/internal/poll/fd_wasip1.go
+++ b/src/internal/poll/fd_wasip1.go
@@ -519,6 +519,10 @@ const (
 // package syscall (intentionally duplicated rather than linknamed
 // across the package boundary — see project memory on shim avoidance).
 func wait(fd int, mode uint8) {
+	if task.OnSystemStack() {
+		task.Pause()
+		return
+	}
 	pd := runtime_netpoll_addwait(uint32(fd), mode)
 	task.Pause()
 	runtime_netpoll_done(pd)
@@ -538,6 +542,10 @@ func (fd *FD) parkUntil(mode uint8, deadline time.Time) error {
 	if d <= 0 {
 		return ErrDeadlineExceeded
 	}
+	if task.OnSystemStack() {
+		task.Pause()
+		return nil
+	}
 	pd := runtime_netpoll_addwait(uint32(fd.Sysfd), mode)
 	timer := time.AfterFunc(d, func() {
 		runtime_netpoll_wake(pd)
@@ -548,6 +556,15 @@ func (fd *FD) parkUntil(mode uint8, deadline time.Time) error {
 	return nil
 }
 
+// WaitWrite waits until data can be written to fd.
+func (fd *FD) WaitWrite() error {
+	if fd.wDeadline.IsZero() {
+		wait(fd.Sysfd, pollModeWrite)
+		return nil
+	}
+	return fd.parkUntil(pollModeWrite, fd.wDeadline)
+}
+
 // byteCount turns the -1 a failed read or write returns into 0, as Go's
 // internal/poll does in fd_unix.go.
 func byteCount(n int, err error) (int, error) {
diff --git a/src/runtime/arch_tinygowasm.go b/src/runtime/arch_tinygowasm.go
index cfb8dfef64..6089118303 100644
--- a/src/runtime/arch_tinygowasm.go
+++ b/src/runtime/arch_tinygowasm.go
@@ -91,16 +91,17 @@ func saveStackPointer() {
 // growHeap tries to grow the heap size. It returns true if it succeeds, false
 // otherwise.
 func growHeap() bool {
-	// Grow memory by the available size, which means the heap size is doubled.
+	// Grow memory by up to 64 MiB at a time and back off if the host caps memory.
+	// Cap total pages at 65535 so uintptr(pages) * wasmPageSize does not wrap to 0.
+	const maxPages = 65535
 	memorySize := wasm_memory_size(wasmMemoryIndex)
-	result := wasm_memory_grow(wasmMemoryIndex, memorySize)
-	if result == -1 {
-		// Grow failed.
-		return false
+	growPages := min(memorySize, 1024, maxPages-memorySize)
+	for growPages > 0 {
+		if wasm_memory_grow(wasmMemoryIndex, growPages) != -1 {
+			setHeapEnd(uintptr(wasm_memory_size(wasmMemoryIndex)) * wasmPageSize)
+			return true
+		}
+		growPages /= 2
 	}
-
-	setHeapEnd(uintptr(wasm_memory_size(wasmMemoryIndex) * wasmPageSize))
-
-	// Heap has grown successfully.
-	return true
+	return false
 }
diff --git a/src/runtime/hashmap.go b/src/runtime/hashmap.go
index 631308a63e..53754ea1c0 100644
--- a/src/runtime/hashmap.go
+++ b/src/runtime/hashmap.go
@@ -176,7 +176,7 @@ func hashmapMake(keySize, valueSize uintptr, sizeHint uintptr, typeInfo unsafe.P
 // it. This is used for the clear builtin, and can be used to reuse a map (to
 // avoid extra heap allocations).
 func hashmapClear(m *hashmap) {
-	if m == nil {
+	if m == nil || m.count == 0 {
 		// Nothing to do. According to the spec:
 		// > If the map or slice is nil, clear is a no-op.
 		return
@@ -415,15 +415,19 @@ func hashmapCopy(m *hashmap, sizeBits uint8) hashmap {
 	bucketBufSize := hashmapBucketSize(m)
 	n.buckets = alloc(bucketBufSize*numBuckets, hashmapType(m).bucketLayout)
 
-	// use a hashmap iterator to go through the old map
-	var it hashmapIterator
-
-	var key = alloc(m.keySize, hashmapType(m).keyLayout)
-	var value = alloc(m.valueSize, hashmapType(m).valueLayout)
-
-	for hashmapNext(m, &it, key, value) {
-		h := n.keyHash(key, uintptr(n.keySize), n.seed)
-		hashmapSet(&n, key, value, h)
+	oldNumBuckets := uintptr(1) << m.bucketBits
+	for i := uintptr(0); i < oldNumBuckets; i++ {
+		for b := hashmapBucketAddr(m, m.buckets, i); b != nil; b = b.next {
+			for j := uint8(0); j < 8; j++ {
+				if b.tophash[j] == 0 {
+					continue
+				}
+				key := hashmapSlotKeyData(m, hashmapSlotKey(m, b, j))
+				value := hashmapSlotValueData(m, hashmapSlotValue(m, b, j))
+				h := n.keyHash(key, uintptr(n.keySize), n.seed)
+				hashmapSet(&n, key, value, h)
+			}
+		}
 	}
 
 	return n
diff --git a/src/runtime/slice.go b/src/runtime/slice.go
index 5206634fd2..f5d954940e 100644
--- a/src/runtime/slice.go
+++ b/src/runtime/slice.go
@@ -29,12 +29,20 @@ func sliceGrow(oldBuf unsafe.Pointer, oldLen, oldCap, newCap, elemSize uintptr,
 		// No need to grow, return the input slice.
 		return oldBuf, oldLen, oldCap
 	}
-
-	// This can be made more memory-efficient by multiplying by some other constant, such as 1.5,
-	// which seems to be allowed by the Go language specification (but this can be observed by
-	// programs); however, due to memory fragmentation and the current state of the TinyGo
-	// memory allocators, this causes some difficult to debug issues.
-	newCap = 1 << bits.Len(uint(newCap))
+	if elemSize == 0 {
+		return oldBuf, oldLen, newCap
+	}
+	if oldCap < 256 {
+		newCap = 1 << bits.Len(uint(newCap))
+	} else {
+		growCap := oldCap
+		for 0 < growCap && growCap < newCap {
+			growCap += (growCap + 3*256) / 4
+		}
+		if growCap > 0 {
+			newCap = growCap
+		}
+	}
 
 	buf := alloc(newCap*elemSize, layout)
 	if oldLen > 0 {