diff --git a/py/nlr.h b/py/nlr.h
index 4cc66d8c2..b94ca7d3f 100644
--- a/py/nlr.h
+++ b/py/nlr.h
@@ -11,7 +11,11 @@ struct _nlr_buf_t {
 #if defined(__i386__)
     void *regs[6];
 #elif defined(__x86_64__)
+  #if defined(__CYGWIN__)
+    void *regs[12];
+  #else
     void *regs[8];
+  #endif
 #elif defined(__thumb2__)
     void *regs[10];
 #else
diff --git a/py/nlrx64.S b/py/nlrx64.S
index c5711001e..a4073981a 100644
--- a/py/nlrx64.S
+++ b/py/nlrx64.S
@@ -4,6 +4,8 @@
     .file   "nlr.s"
     .text
 
+#if !defined(__CYGWIN__)
+
 /* uint nlr_push(rdi=nlr_buf_t *nlr) */
 #if !(defined(__APPLE__) && defined(__MACH__))
     .globl  nlr_push
@@ -82,4 +84,63 @@ nlr_jump:
     .local  nlr_top
 #endif
     .comm   nlr_top,8,8
-#endif
+
+#else // !defined(__CYGWIN__)
+
+/* uint nlr_push(rcx=nlr_buf_t *nlr) */
+    .globl  nlr_push
+nlr_push:
+    movq    (%rsp), %rax            # load return %rip
+    movq    %rax, 16(%rcx)          # store %rip into nlr_buf
+    movq    %rbp, 24(%rcx)          # store %rbp into nlr_buf
+    movq    %rsp, 32(%rcx)          # store %rsp into nlr_buf
+    movq    %rbx, 40(%rcx)          # store %rbx into nlr_buf
+    movq    %r12, 48(%rcx)          # store %r12 into nlr_buf
+    movq    %r13, 56(%rcx)          # store %r13 into nlr_buf
+    movq    %r14, 64(%rcx)          # store %r14 into nlr_buf
+    movq    %r15, 72(%rcx)          # store %r15 into
+    movq    %rdi, 80(%rcx)          # store %rdr into
+    movq    %rsi, 88(%rcx)          # store %rsi into
+    movq    nlr_top(%rip), %rax     # get last nlr_buf
+    movq    %rax, (%rcx)            # store it
+    movq    %rcx, nlr_top(%rip)     # stor new nlr_buf (to make linked list)
+    xorq    %rax, %rax              # return 0, normal return
+    ret                             # return
+
+/* void nlr_jump(rcx=uint val) */
+
+    .globl  nlr_jump
+nlr_jump:
+    movq    %rcx, %rax              # put return value in %rax
+    movq    nlr_top(%rip), %rcx     # get nlr_top into %rsi
+    movq    %rax, 8(%rcx)           # store return value
+    movq    (%rcx), %rax            # load prev nlr_buf
+    movq    %rax, nlr_top(%rip)     # store prev nlr_buf (to unlink list)
+    movq    72(%rcx), %r15          # load saved %r15
+    movq    64(%rcx), %r14          # load saved %r14
+    movq    56(%rcx), %r13          # load saved %r13
+    movq    48(%rcx), %r12          # load saved %r12
+    movq    40(%rcx), %rbx          # load saved %rbx
+    movq    32(%rcx), %rsp          # load saved %rsp
+    movq    24(%rcx), %rbp          # load saved %rbp
+    movq    16(%rcx), %rax          # load saved %rip
+    movq    80(%rcx), %rdi          # store %rdr into
+    movq    88(%rcx), %rsi          # store %rsi into
+    movq    %rax, (%rsp)            # store saved %rip to stack
+    xorq    %rax, %rax              # clear return register
+    inc     %al                     # increase to make 1, non-local return
+    ret                             # return
+
+    .comm   nlr_top,8,8
+
+    /* void nlr_pop() */
+    .globl  nlr_pop
+nlr_pop:
+    movq    nlr_top(%rip), %rax     # get nlr_top into %rax
+    movq    (%rax), %rax            # load prev nlr_buf
+    movq    %rax, nlr_top(%rip)     # store prev nlr_buf (to unlink list)
+    ret                             # return
+
+#endif // !defined(__CYGWIN__)
+
+#endif // __x86_64__
diff --git a/tests/run-tests b/tests/run-tests
index 1d99830bc..5be819297 100755
--- a/tests/run-tests
+++ b/tests/run-tests
@@ -36,7 +36,13 @@ if test_on_pyboard:
 for test_file in tests:
     test_name = os.path.splitext(os.path.basename(test_file))[0]
 
-    output_expected = subprocess.check_output([CPYTHON3, '-B', test_file])
+    # run CPython
+    try:
+        output_expected = subprocess.check_output([CPYTHON3, '-B', test_file])
+    except subprocess.CalledProcessError:
+        output_expected = b'CPYTHON3 CRASH'
+
+    # run Micro Python
     try:
         if test_on_pyboard:
             pyb.enter_raw_repl()
diff --git a/unix/gccollect.c b/unix/gccollect.c
index 450a69778..33748d240 100644
--- a/unix/gccollect.c
+++ b/unix/gccollect.c
@@ -78,9 +78,14 @@ void gc_collect(void) {
 
     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, ((machine_uint_t)&_end - (machine_uint_t)&__bss_start) / sizeof(machine_uint_t));
+#ifdef __CYGWIN__
+#define BSS_START __bss_start__
+#else
+#define BSS_START __bss_start
+#endif
+    extern char BSS_START, _end;
+    //printf(".bss: %p-%p\n", &BSS_START, &_end);
+    gc_collect_root((void**)&BSS_START, ((machine_uint_t)&_end - (machine_uint_t)&BSS_START) / sizeof(machine_uint_t));
     regs_t regs;
     gc_helper_get_regs(regs);
     // GC stack (and regs because we captured them)