diff --git a/py/gc.c b/py/gc.c
index c8256fb03b..9c8d203d28 100644
--- a/py/gc.c
+++ b/py/gc.c
@@ -240,7 +240,7 @@ void gc_info(gc_info_t *info) {
 
 void *gc_alloc(machine_uint_t n_bytes) {
     machine_uint_t n_blocks = ((n_bytes + BYTES_PER_BLOCK - 1) & (~(BYTES_PER_BLOCK - 1))) / BYTES_PER_BLOCK;
-    //printf("gc_alloc(%u bytes -> %u blocks)\n", n_bytes, n_blocks);
+    DEBUG_printf("gc_alloc(%u bytes -> %u blocks)\n", n_bytes, n_blocks);
 
     // check for 0 allocation
     if (n_blocks == 0) {
@@ -267,6 +267,7 @@ void *gc_alloc(machine_uint_t n_bytes) {
         if (collected) {
             return NULL;
         }
+        DEBUG_printf("gc_alloc(" UINT_FMT "): no free mem, triggering GC\n", n_bytes);
         gc_collect();
         collected = 1;
     }
@@ -341,6 +342,14 @@ void *gc_realloc(void *ptr, machine_uint_t n_bytes) {
     }
 }
 
+void gc_dump_info() {
+    gc_info_t info;
+    gc_info(&info);
+    printf("GC: total: " UINT_FMT ", used: " UINT_FMT ", free: " UINT_FMT "\n", info.total, info.used, info.free);
+    printf(" No. of 1-blocks: " UINT_FMT ", 2-blocks: " UINT_FMT ", max blk sz: " UINT_FMT "\n",
+           info.num_1block, info.num_2block, info.max_block);
+}
+
 #if DEBUG_PRINT
 static void gc_dump_at(void) {
     for (machine_uint_t bl = 0; bl < gc_alloc_table_byte_len * BLOCKS_PER_ATB; bl++) {
diff --git a/py/malloc.c b/py/malloc.c
index c87d91c0a0..7f55fa7c8e 100644
--- a/py/malloc.c
+++ b/py/malloc.c
@@ -19,6 +19,22 @@ static int peak_bytes_allocated = 0;
 #define UPDATE_PEAK() { if (current_bytes_allocated > peak_bytes_allocated) peak_bytes_allocated = current_bytes_allocated; }
 #endif
 
+#if MICROPY_ENABLE_GC
+#include "gc.h"
+
+// We redirect standard alloc functions to GC heap - just for the rest of
+// this module. In the rest of micropython source, system malloc can be
+// freely accessed - for interfacing with system and 3rd-party libs for
+// example. On the other hand, some (e.g. bare-metal) ports may use GC
+// heap as system heap, so, to avoid warnings, we do undef's first.
+#undef malloc
+#undef free
+#undef realloc
+#define malloc gc_alloc
+#define free gc_free
+#define realloc gc_realloc
+#endif // MICROPY_ENABLE_GC
+
 void *m_malloc(int num_bytes) {
     if (num_bytes == 0) {
         return NULL;
diff --git a/stm/malloc0.c b/stm/malloc0.c
index eaa436f4fb..85a643f72d 100644
--- a/stm/malloc0.c
+++ b/stm/malloc0.c
@@ -29,18 +29,6 @@ void *realloc(void *ptr, size_t n) {
 
 #endif
 
-void *malloc(size_t n) {
-    return gc_alloc(n);
-}
-
-void free(void *ptr) {
-    gc_free(ptr);
-}
-
-void *realloc(void *ptr, size_t n) {
-    return gc_realloc(ptr, n);
-}
-
 void __assert_func(void) {
     printf("\nASSERT FAIL!");
     for (;;) {
diff --git a/stm/mpconfigport.h b/stm/mpconfigport.h
index fd7af29d27..1d79243f9b 100644
--- a/stm/mpconfigport.h
+++ b/stm/mpconfigport.h
@@ -21,6 +21,14 @@ typedef float machine_float_t;
 
 machine_float_t machine_sqrt(machine_float_t x);
 
+// There is no classical C heap in bare-metal ports, only Python
+// garbage-collected heap. For completeness, emulate C heap via
+// GC heap. Note that MicroPython core never uses malloc() and friends,
+// so these defines are mostly to help extension module writers.
+#define malloc gc_alloc
+#define free gc_free
+#define realloc gc_realloc
+
 // board specific definitions
 
 // choose 1 of these boards
diff --git a/stm/std.h b/stm/std.h
index a2bf706176..843ddd8270 100644
--- a/stm/std.h
+++ b/stm/std.h
@@ -2,10 +2,6 @@ typedef unsigned int size_t;
 
 void __assert_func(void);
 
-void *malloc(size_t n);
-void free(void *ptr);
-void *realloc(void *ptr, size_t n);
-
 void *memcpy(void *dest, const void *src, size_t n);
 void *memmove(void *dest, const void *src, size_t n);
 void *memset(void *s, int c, size_t n);
diff --git a/unix/Makefile b/unix/Makefile
index 317bcb5ebf..1ce00eb28f 100644
--- a/unix/Makefile
+++ b/unix/Makefile
@@ -35,6 +35,7 @@ endif
 # source files
 SRC_C = \
 	main.c \
+	gccollect.c \
 	file.c \
 	socket.c \
 	$(SRC_MOD)
diff --git a/unix/gccollect.c b/unix/gccollect.c
new file mode 100644
index 0000000000..60bc99323e
--- /dev/null
+++ b/unix/gccollect.c
@@ -0,0 +1,68 @@
+#include <stdint.h>
+#include <stdio.h>
+
+#include "misc.h"
+#include "mpconfig.h"
+#include "gc.h"
+
+#if MICROPY_ENABLE_GC
+
+extern void *stack_top;
+
+// We capture here callee-save registers, i.e. ones which may contain
+// interesting values held there by our callers. It doesn't make sense
+// to capture caller-saved registers, because they, well, put on the
+// stack already by the caller.
+#ifdef __x86_64__
+typedef machine_uint_t regs_t[6];
+
+void gc_helper_get_regs(regs_t arr) {
+    register long rbx asm ("rbx");
+    register long rbp asm ("rbp");
+    register long r12 asm ("r12");
+    register long r13 asm ("r13");
+    register long r14 asm ("r14");
+    register long r15 asm ("r15");
+    arr[0] = rbx;
+    arr[1] = rbp;
+    arr[2] = r12;
+    arr[3] = r13;
+    arr[4] = r14;
+    arr[5] = r15;
+}
+#endif
+
+#ifdef __i386__
+typedef machine_uint_t regs_t[4];
+
+void gc_helper_get_regs(regs_t arr) {
+    register long ebx asm ("ebx");
+    register long esi asm ("esi");
+    register long edi asm ("edi");
+    register long ebp asm ("ebp");
+    arr[0] = ebx;
+    arr[1] = esi;
+    arr[2] = edi;
+    arr[3] = ebp;
+}
+#endif
+
+void gc_collect(void) {
+    //gc_dump_info();
+
+    gc_collect_start();
+    // this traces .data and .bss sections
+    extern char __bss_start, _end;
+    //printf(".bss: %p-%p\n", &__bss_start, &_end);
+    gc_collect_root((void**)&__bss_start, ((uint32_t)&_end - (uint32_t)&__bss_start) / sizeof(uint32_t));
+    regs_t regs;
+    gc_helper_get_regs(regs);
+    // GC stack (and regs because we captured them)
+    gc_collect_root((void**)&regs, ((uint32_t)stack_top - (uint32_t)&regs) / sizeof(uint32_t));
+    gc_collect_end();
+
+    //printf("-----\n");
+    //gc_dump_info();
+}
+
+#endif //MICROPY_ENABLE_GC
diff --git a/unix/main.c b/unix/main.c
index 9bdcc53a7c..192a0e6e8b 100644
--- a/unix/main.c
+++ b/unix/main.c
@@ -15,12 +15,20 @@
 #include "runtime0.h"
 #include "runtime.h"
 #include "repl.h"
+#include "gc.h"
 
 #if MICROPY_USE_READLINE
 #include <readline/readline.h>
 #include <readline/history.h>
 #endif
 
+// Heap size of GC heap (if enabled)
+// TODO: allow to specify on command line
+#define HEAP_SIZE 128*1024
+
+// Stack top at the start of program
+void *stack_top;
+
 extern const mp_obj_fun_native_t mp_builtin_open_obj;
 void file_init();
 void microsocket_init();
@@ -217,7 +225,24 @@ mp_obj_t qstr_info(void) {
     return mp_const_none;
 }
 
+#if MICROPY_ENABLE_GC
+// TODO: this doesn't belong here
+static mp_obj_t pyb_gc(void) {
+    gc_collect();
+    return mp_const_none;
+}
+MP_DEFINE_CONST_FUN_OBJ_0(pyb_gc_obj, pyb_gc);
+#endif
+
 int main(int argc, char **argv) {
+    volatile int stack_dummy;
+    stack_top = (void*)&stack_dummy;
+
+#if MICROPY_ENABLE_GC
+    char *heap = malloc(HEAP_SIZE);
+    gc_init(heap, heap + HEAP_SIZE);
+#endif
+
     qstr_init();
     rt_init();
 
@@ -263,6 +288,9 @@ int main(int argc, char **argv) {
     rt_store_name(qstr_from_str("test"), test_obj_new(42));
     rt_store_name(qstr_from_str("mem_info"), rt_make_function_n(0, mem_info));
     rt_store_name(qstr_from_str("qstr_info"), rt_make_function_n(0, qstr_info));
+#if MICROPY_ENABLE_GC
+    rt_store_name(qstr_from_str("gc"), (mp_obj_t)&pyb_gc_obj);
+#endif
 
     file_init();
     microsocket_init();