diff --git a/src/runtime/dynamic_arm64.go b/src/runtime/dynamic_arm64.go
index 6fa6ef3c..645c7975 100644
--- a/src/runtime/dynamic_arm64.go
+++ b/src/runtime/dynamic_arm64.go
@@ -59,8 +59,15 @@ func dynamicLoader(base uintptr, dyn *dyn64) {
 	for relasz > 0 && rela != nil {
 		switch rela.Info {
 		case rAARCH64_RELATIVE:
+			if debugLoader {
+				println("relocating ", uintptr(rela.Addend), " to ", base+uintptr(rela.Addend))
+			}
 			ptr := (*uint64)(unsafe.Pointer(base + uintptr(rela.Off)))
 			*ptr = uint64(base + uintptr(rela.Addend))
+		default:
+			if debugLoader {
+				println("unknown section to load:", rela.Info)
+			}
 		}
 
 		rptr := uintptr(unsafe.Pointer(rela))
diff --git a/src/runtime/runtime_nintendoswitch.go b/src/runtime/runtime_nintendoswitch.go
index f5107fa8..6930e371 100644
--- a/src/runtime/runtime_nintendoswitch.go
+++ b/src/runtime/runtime_nintendoswitch.go
@@ -8,35 +8,67 @@ type timeUnit int64
 
 const asyncScheduler = false
 
+const (
+	// Handles
+	infoTypeTotalMemorySize = 6          // Total amount of memory available for process.
+	infoTypeUsedMemorySize  = 7          // Amount of memory currently used by process.
+	currentProcessHandle    = 0xFFFF8001 // Pseudo handle for the current process.
+
+	// Types of config Entry
+	envEntryTypeEndOfList        = 0 // Entry list terminator.
+	envEntryTypeMainThreadHandle = 1 // Provides the handle to the main thread.
+	envEntryTypeOverrideHeap     = 3 // Provides heap override information.
+
+	// Default heap size allocated by libnx
+	defaultHeapSize = 0x2000000 * 16
+
+	debugInit = false
+)
+
+//go:extern _saved_return_address
+var savedReturnAddress uintptr
+
+//export __stack_top
 var stackTop uintptr
 
+//go:extern _context
+var context uintptr
+
+//go:extern _main_thread
+var mainThread uintptr
+
+var (
+	heapStart = uintptr(0)
+	heapEnd   = uintptr(0)
+	usedRam   = uint64(0)
+	totalRam  = uint64(0)
+	totalHeap = uint64(0)
+)
+
 func postinit() {}
 
+func preinit() {
+	// Unsafe to use heap here
+	setupEnv()
+	setupHeap()
+}
+
 // Entry point for Go. Initialize all packages and call main.main().
 //export main
-func main() int {
+func main() {
 	preinit()
-
-	// Obtain the initial stack pointer right before calling the run() function.
-	// The run function has been moved to a separate (non-inlined) function so
-	// that the correct stack pointer is read.
-	stackTop = getCurrentStackPointer()
-	runMain()
+	run()
 
 	// Call exit to correctly finish the program
 	// Without this, the application crashes at start, not sure why
-	return exit(0)
-}
-
-// Must be a separate function to get the correct stack pointer.
-//go:noinline
-func runMain() {
-	run()
+	for {
+		exit(0)
+	}
 }
 
 // sleepTicks sleeps for the specified system ticks
 func sleepTicks(d timeUnit) {
-	sleepThread(uint64(ticksToNanoseconds(d)))
+	svcSleepThread(uint64(ticksToNanoseconds(d)))
 }
 
 // armTicksToNs converts cpu ticks to nanoseconds
@@ -60,7 +92,7 @@ var position = 0
 
 func putchar(c byte) {
 	if c == '\n' || position >= len(stdoutBuffer) {
-		nxOutputString(&stdoutBuffer[0], uint64(position))
+		svcOutputDebugString(&stdoutBuffer[0], uint64(position))
 		position = 0
 		return
 	}
@@ -85,11 +117,159 @@ func write(fd int32, buf *byte, count int) int {
 	return count
 }
 
-//export sleepThread
-func sleepThread(nanos uint64)
+// exit checks if a savedReturnAddress were provided by the launcher
+// if so, calls the nxExit which restores the stack and returns to launcher
+// otherwise just calls systemcall exit
+func exit(code int) {
+	if savedReturnAddress == 0 {
+		svcExitProcess(code)
+		return
+	}
 
-//export exit
-func exit(code int) int
+	nxExit(code, stackTop, savedReturnAddress)
+}
+
+type configEntry struct {
+	Key   uint32
+	Flags uint32
+	Value [2]uint64
+}
+
+func setupEnv() {
+	if debugInit {
+		println("Saved Return Address:", savedReturnAddress)
+		println("Context:", context)
+		println("Main Thread Handle:", mainThread)
+	}
+
+	// See https://switchbrew.org/w/index.php?title=Homebrew_ABI
+	// Here we parse only the required configs for initializing
+	if context != 0 {
+		ptr := context
+		entry := (*configEntry)(unsafe.Pointer(ptr))
+		for entry.Key != envEntryTypeEndOfList {
+			switch entry.Key {
+			case envEntryTypeOverrideHeap:
+				if debugInit {
+					println("Got heap override")
+				}
+				heapStart = uintptr(entry.Value[0])
+				heapEnd = heapStart + uintptr(entry.Value[1])
+			case envEntryTypeMainThreadHandle:
+				mainThread = uintptr(entry.Value[0])
+			default:
+				if entry.Flags&1 > 0 {
+					// Mandatory but not parsed
+					runtimePanic("mandatory config entry not parsed")
+				}
+			}
+			ptr += unsafe.Sizeof(configEntry{})
+			entry = (*configEntry)(unsafe.Pointer(ptr))
+		}
+	}
+	// Fetch used / total RAM for allocating HEAP
+	svcGetInfo(&totalRam, infoTypeTotalMemorySize, currentProcessHandle, 0)
+	svcGetInfo(&usedRam, infoTypeUsedMemorySize, currentProcessHandle, 0)
+}
+
+func setupHeap() {
+	if heapStart != 0 {
+		if debugInit {
+			print("Heap already overrided by hblauncher")
+		}
+		// Already overrided
+		return
+	}
+
+	if debugInit {
+		print("No heap override. Using normal initialization")
+	}
+
+	size := uint32(defaultHeapSize)
+
+	if totalRam > usedRam+0x200000 {
+		// Get maximum possible heap
+		size = uint32(totalRam-usedRam-0x200000) & ^uint32(0x1FFFFF)
+	}
+
+	if size < defaultHeapSize {
+		size = defaultHeapSize
+	}
+
+	if debugInit {
+		println("Trying to allocate", size, "bytes of heap")
+	}
+
+	svcSetHeapSize(&heapStart, uint64(size))
+
+	if heapStart == 0 {
+		runtimePanic("failed to allocate heap")
+	}
+
+	totalHeap = uint64(size)
+
+	heapEnd = heapStart + uintptr(size)
+
+	if debugInit {
+		println("Heap Start", heapStart)
+		println("Heap End  ", heapEnd)
+		println("Total Heap", totalHeap)
+	}
+}
+
+// getHeapBase returns the start address of the heap
+// this is externally linked by gonx
+func getHeapBase() uintptr {
+	return heapStart
+}
+
+// getHeapEnd returns the end address of the heap
+// this is externally linked by gonx
+func getHeapEnd() uintptr {
+	return heapEnd
+}
+
+// getContextPtr returns the hblauncher context
+// this is externally linked by gonx
+func getContextPtr() uintptr {
+	return context
+}
+
+// getMainThreadHandle returns the main thread handler if any
+// this is externally linked by gonx
+func getMainThreadHandle() uintptr {
+	return mainThread
+}
 
 //export armGetSystemTick
 func getArmSystemTick() int64
+
+// nxExit exits the program to homebrew launcher
+//export __nx_exit
+func nxExit(code int, stackTop uintptr, exitFunction uintptr)
+
+// Horizon System Calls
+// svcSetHeapSize Set the process heap to a given size. It can both extend and shrink the heap.
+// svc 0x01
+//export svcSetHeapSize
+func svcSetHeapSize(addr *uintptr, length uint64) uint64
+
+// svcExitProcess Exits the current process.
+// svc 0x07
+//export svcExitProcess
+func svcExitProcess(code int)
+
+// svcSleepThread Sleeps the current thread for the specified amount of time.
+// svc 0x0B
+//export svcSleepThread
+func svcSleepThread(nanos uint64)
+
+// svcOutputDebugString Outputs debug text, if used during debugging.
+// svc 0x27
+//export svcOutputDebugString
+func svcOutputDebugString(str *uint8, size uint64) uint64
+
+// svcGetInfo Retrieves information about the system, or a certain kernel object.
+// svc 0x29
+//export svcGetInfo
+func svcGetInfo(output *uint64, id0 uint32, handle uint32, id1 uint64) uint64
diff --git a/src/runtime/runtime_nintendoswitch.s b/src/runtime/runtime_nintendoswitch.s
index 58ffe273..a4cac23e 100644
--- a/src/runtime/runtime_nintendoswitch.s
+++ b/src/runtime/runtime_nintendoswitch.s
@@ -1,45 +1,40 @@
-.section .text.armGetSystemTick, "ax", %progbits
-.global  armGetSystemTick
-.type    armGetSystemTick, %function
-.align 2
-armGetSystemTick:
+// Macro for writing less code
+.macro FUNC name
+    .section .text.\name, "ax", %progbits
+    .global \name
+    .type \name, %function
+    .align 2
+\name:
+.endm
+
+FUNC armGetSystemTick
     mrs x0, cntpct_el0
     ret
 
-.section .text.nxOutputString, "ax", %progbits
-.global nxOutputString
-.type nxOutputString, %function
-.align 2
-.cfi_startproc
-nxOutputString:
-	svc 0x27
-	ret
-.cfi_endproc
+// Horizon System Calls
+// https://switchbrew.org/wiki/SVC
+FUNC svcSetHeapSize
+    str x0, [sp, #-16]!
+    svc 0x1
+    ldr x2, [sp], #16
+    str x1, [x2]
+    ret
 
-.section .text.exit, "ax", %progbits
-.global exit
-.type exit, %function
-.align 2
-exit:
-	svc 0x7
-	ret
+FUNC svcExitProcess
+    svc 0x7
+    ret
 
-.section .text.setHeapSize, "ax", %progbits
-.global setHeapSize
-.type setHeapSize, %function
-.align 2
-setHeapSize:
-	str x0, [sp, #-16]!
-	svc 0x1
-	ldr x2, [sp], #16
-	str x1, [x2]
-	ret
+FUNC svcSleepThread
+    svc 0xB
+    ret
 
+FUNC svcOutputDebugString
+    svc 0x27
+    ret
 
-.section .text.sleepThread, "ax", %progbits
-.global sleepThread
-.type sleepThread, %function
-.align 2
-sleepThread:
-	svc 0xB
-	ret
+FUNC svcGetInfo
+    str x0, [sp, #-16]!
+    svc 0x29
+    ldr x2, [sp], #16
+    str x1, [x2]
+    ret
diff --git a/src/runtime/runtime_nintendoswitch_heap.go b/src/runtime/runtime_nintendoswitch_heap.go
deleted file mode 100644
index e8df7e4b..00000000
--- a/src/runtime/runtime_nintendoswitch_heap.go
+++ /dev/null
@@ -1,25 +0,0 @@
-// +build nintendoswitch
-
-// +build gc.conservative gc.leaking
-
-package runtime
-
-const heapSize = 0x2000000 * 16 // Default by libnx
-
-var (
-	heapStart = uintptr(0)
-	heapEnd   = uintptr(0)
-)
-
-//export setHeapSize
-func setHeapSize(addr *uintptr, length uint64) uint64
-
-func preinit() {
-	setHeapSize(&heapStart, heapSize)
-
-	if heapStart == 0 {
-		runtimePanic("failed to allocate heap")
-	}
-
-	heapEnd = heapStart + heapSize
-}
diff --git a/src/runtime/runtime_nintendoswitch_noheap.go b/src/runtime/runtime_nintendoswitch_noheap.go
deleted file mode 100644
index 41e0b368..00000000
--- a/src/runtime/runtime_nintendoswitch_noheap.go
+++ /dev/null
@@ -1,7 +0,0 @@
-// +build nintendoswitch
-
-// +build gc.none gc.extalloc
-
-package runtime
-
-func preinit() {}
diff --git a/src/runtime/runtime_nintendoswitch_svc.go b/src/runtime/runtime_nintendoswitch_svc.go
deleted file mode 100644
index 2bc2003f..00000000
--- a/src/runtime/runtime_nintendoswitch_svc.go
+++ /dev/null
@@ -1,22 +0,0 @@
-// +build nintendoswitch
-
-package runtime
-
-import (
-	"unsafe"
-)
-
-// Result nxOutputString(const char *str, u64 size)
-//export nxOutputString
-func nxOutputString(str *uint8, size uint64) uint64
-
-func NxOutputString(str string) uint64 {
-	strData := (*_string)(unsafe.Pointer(&str))
-	return nxOutputString((*uint8)(unsafe.Pointer(strData.ptr)), uint64(strData.length))
-}
-
-//export malloc
-func extalloc(size uintptr) unsafe.Pointer
-
-//export free
-func extfree(ptr unsafe.Pointer)
diff --git a/targets/nintendoswitch.json b/targets/nintendoswitch.json
index aa8567b0..b7bc5bcd 100644
--- a/targets/nintendoswitch.json
+++ b/targets/nintendoswitch.json
@@ -9,9 +9,10 @@
   "libc": "picolibc",
   "gc": "conservative",
   "relocation-model": "pic",
+  "cpu": "cortex-a57",
   "cflags": [
-    "-target", "aarch64-none-linux-gnu",
-    "-mtune=cortex-a57",
+    "-target", "aarch64-unknown-none",
+    "-mcpu=cortex-a57",
     "-fPIE",
     "-Werror",
     "-Qunused-arguments",
diff --git a/targets/nintendoswitch.ld b/targets/nintendoswitch.ld
index 40179866..b8e9f5d9 100644
--- a/targets/nintendoswitch.ld
+++ b/targets/nintendoswitch.ld
@@ -17,6 +17,7 @@ SECTIONS
     KEEP(*(.text.jmp))
 
     . = 0x80;
+    KEEP(*(.text.start))
 
     *(.text .text.*)
     *(.plt .plt.*)
diff --git a/targets/nintendoswitch.s b/targets/nintendoswitch.s
index 38e0b39e..48edc857 100644
--- a/targets/nintendoswitch.s
+++ b/targets/nintendoswitch.s
@@ -14,11 +14,11 @@ _start:
     .word 0                    // flags (unused)
 
     // segment headers
-    .word 0 // __text_start
+    .word __text_start - _start
     .word __text_size
-    .word 0 //__rodata_start
+    .word __rodata_start - _start
     .word __rodata_size
-    .word 0 //__data_start
+    .word __data_start - _start
     .word __data_size
     .word __bss_size
     .word 0
@@ -41,19 +41,37 @@ _mod_header:
     .word 0, 0 // eh_frame_hdr start/end
     .word 0 // runtime-generated module object offset
 
-.section .text, "x"
+.section .text.start, "x"
 .global start
 start:
-    // Get ASLR Base
-    adrp x6, _start
+    // save lr
+    mov  x7, x30
+
+    // get aslr base
+    bl   +4
+    sub  x6, x30, #0x88
 
     // context ptr and main thread handle
-    mov  x5, x0
-    mov  x4, x1
+    mov  x25, x0
+    mov  x26, x1
 
     // Save ASLR Base to use later
     mov x0, x6
 
+    adrp x4, _saved_return_address
+    str  x7, [x4, #:lo12:_saved_return_address]
+
+    adrp x4, _context
+    str x25, [x4, #:lo12:_context]
+
+    adrp x4, _main_thread
+    str x26, [x4, #:lo12:_main_thread]
+
+    // store stack pointer
+    mov  x26, sp
+    adrp x4, _stack_top
+    str  x26, [x4, #:lo12:_stack_top]
+
     // clear .bss
     adrp x5, __bss_start
     add x5, x5, #:lo12:__bss_start
@@ -76,3 +94,33 @@ run:
 
     // call entrypoint
     b    main
+
+.global __nx_exit
+.type   __nx_exit, %function
+__nx_exit:
+    // Exit code in x0
+
+    // restore stack pointer
+    mov sp, x1
+
+    // jump back to loader
+    br   x2
+
+.section .data.horizon
+.align 8
+.global _saved_return_address // Saved return address.
+                              // This might be different than null when coming from launcher
+_saved_return_address:
+    .dword 0
+.global _context // Homebrew Launcher Context
+                 // This might be different than null when not coming from launcher
+_context:
+    .dword 0
+
+.global _main_thread
+_main_thread:
+    .dword 0
+
+.global _stack_top
+_stack_top:
+    .dword 0