diff --git a/.github/workflows/nightly.yml b/.github/workflows/nightly.yml
index 60ccbd84..176bfa41 100644
--- a/.github/workflows/nightly.yml
+++ b/.github/workflows/nightly.yml
@@ -272,6 +272,24 @@ jobs:
         test/health.awk < log > json
         scp -o "StrictHostKeyChecking=no" json "root@176.9.217.245:/data/downloads/health/bsd_${{ matrix.cc }}_$(date +"%Y%m%d").json"
 
+  bsdbe:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        target: [bsd_s390]
+    name: bsd be ${{ matrix.target }}
+    steps:
+    - uses: actions/checkout@v5
+      with: { fetch-depth: 2 }
+    - run: |
+        sudo apt -y update ; sudo apt -y install binfmt-support qemu-user-static && docker run --rm --privileged multiarch/qemu-user-static --reset -p yes
+        source ./test/setup_ga_network.sh
+        make -C test port_tap_bridge
+        test/port_tap_bridge &
+        sleep 2
+        make -C test/bsd ${{ matrix.target }}
+
   s390:
     runs-on: ubuntu-latest
     strategy:
diff --git a/test/bsd/Makefile b/test/bsd/Makefile
index 5c1cf693..7be18f2a 100644
--- a/test/bsd/Makefile
+++ b/test/bsd/Makefile
@@ -13,6 +13,7 @@ else
 C_WARN ?= -Wno-deprecated
 endif
 
+.PHONY: all bsd_test bsd_client shm_tap_bridge shm_port_bridge bsd_s390 clean
 
 SOURCES = bsd_shm_test.c ../../mongoose.c
 
@@ -23,26 +24,42 @@ SOURCES += FreeRTOS-Kernel/portable/ThirdParty/GCC/Posix/utils/wait_for_event.c
 CFLAGS += -IFreeRTOS-Kernel/include 
 CFLAGS += -IFreeRTOS-Kernel/portable/ThirdParty/GCC/Posix -Wno-conversion -Wno-undef
 
+DOCKER_BIN ?= docker
+CWD ?= $(realpath $(CURDIR))
+ROOT_DIR = $(realpath $(CWD)/../..)
+ENV ?= -e Tmp=. -e WINEDEBUG=-all
+DOCKER ?= $(DOCKER_BIN) run --platform linux/amd64 --net=host --rm $(ENV) -v $(ROOT_DIR):$(ROOT_DIR) -w $(CWD)
+
 # Mongoose options are defined in mongoose_config.h, this forces loading and skips OS detection
 CFLAGS_MONGOOSE += -DMG_ARCH=MG_ARCH_CUSTOM
 
 
-all:  shm_tap_bridge bsd_client FreeRTOS-Kernel bsd_test
+all bsd: shm_tap_bridge bsd_client FreeRTOS-Kernel bsd_test
 	ASAN_OPTIONS=$(ASAN_OPTIONS) $(RUN) ./bsd_test $(ARGS)
 
 # requires shm_tap_bridge to be running
-bsd_test: $(SOURCES) bsd_x_test.c ../../mongoose.h
+bsd_test: $(SOURCES) bsd_x_test.c ../../mongoose.h shm_queue.h
 	$(CC) $(SOURCES) $(wildcard FreeRTOS-Kernel/*.c) $(CFLAGS) $(CFLAGS_MONGOOSE) $(CFLAGS_EXTRA) -o $@
 
 bsd_client: bsd_client.c
 	$(CC) bsd_client.c -pthread -o $@
 
 # bridge shm queue <--> TAP interface
-shm_tap_bridge: shm_tap_bridge.c
+shm_tap_bridge: shm_tap_bridge.c shm_queue.h
 	$(CC) shm_tap_bridge.c -pthread -o $@
 
+# requires test/port_tap_bridge to be running
+bsd_s390: ASAN=
+bsd_s390: CC = $(DOCKER) mdashnet/s390 cc
+bsd_s390: RUN = $(DOCKER) --init mdashnet/s390
+bsd_s390: shm_port_bridge bsd_client FreeRTOS-Kernel bsd_test
+	$(RUN) sh -c './shm_port_bridge & pid=$$!; sleep 1; ./bsd_test; rc=$$?; kill $$pid >/dev/null 2>&1; wait $$pid >/dev/null 2>&1; exit $$rc'
+
+shm_port_bridge: shm_port_bridge.c shm_queue.h
+	$(CC) shm_port_bridge.c $(CFLAGS) -pthread -o $@
+
 clean:                    # Cleanup. Delete built program and all build artifacts
-	rm -rf bsd_test bsd_client *.o *.obj FreeRTOS-Kernel
+	rm -rf bsd_test bsd_client shm_port_bridge *.o *.obj FreeRTOS-Kernel
 
 FreeRTOS-Kernel:  # FreeRTOS sources
 	git clone -q -c advice.detachedHead=false --depth 1 -b V11.2.0 https://github.com/FreeRTOS/FreeRTOS-Kernel $@
diff --git a/test/bsd/bsd_shm_test.c b/test/bsd/bsd_shm_test.c
index 474f1bd5..e3214466 100644
--- a/test/bsd/bsd_shm_test.c
+++ b/test/bsd/bsd_shm_test.c
@@ -1,36 +1,15 @@
+#include <errno.h>
 #include <stdio.h>
-#include <semaphore.h>
 #include <sys/mman.h>
 #include <sys/stat.h>
 
 #include "mongoose.h"
+#include "shm_queue.h"
 
 #ifndef O_RDWR
 #define O_RDWR 02
 #endif
 
-#define FRAME_SIZE 1540
-#define FRAME_QUEUE_DEPTH 8
-#define SHM_NAME "/mongoose-shm"
-
-struct shm_frame {
-  size_t len;
-  uint8_t buf[FRAME_SIZE];
-};
-
-struct shm_queue {
-  sem_t full;
-  sem_t empty;
-  size_t head;
-  size_t tail;
-  struct shm_frame frames[FRAME_QUEUE_DEPTH];
-};
-
-struct shm_link {
-  struct shm_queue to_tap;
-  struct shm_queue from_tap;
-};
-
 struct shm_driver_data {
   const char *name;
   FILE *fp;
@@ -48,14 +27,21 @@ static bool shm_init(struct mg_tcpip_if *ifp) {
   int fd;
 
   fd = shm_open(d->name, O_RDWR, 0600);
-  if (fd < 0) return false;
+  if (fd < 0) {
+    printf("shm_open failed: %d\n", errno);
+    return false;
+  }
   d->fp = fdopen(fd, "r+");
-  if (d->fp == NULL) return false;
+  if (d->fp == NULL) {
+    printf("fdopen failed: %d\n", errno);
+    return false;
+  }
 
   d->shm = (struct shm_link *) mmap(NULL, sizeof(*d->shm),
                                     PROT_READ | PROT_WRITE, MAP_SHARED,
                                     fileno(d->fp), 0);
   if (d->shm == MAP_FAILED) {
+    printf("mmap failed: %d\n", errno);
     fclose(d->fp);
     d->fp = NULL;
     d->shm = NULL;
@@ -67,39 +53,17 @@ static bool shm_init(struct mg_tcpip_if *ifp) {
 
 static size_t shm_tx(const void *buf, size_t len, struct mg_tcpip_if *ifp) {
   struct shm_driver_data *d = (struct shm_driver_data *) ifp->driver_data;
-  struct shm_queue *q = &d->shm->to_tap;
-  struct shm_frame *f;
 
-  if (len > sizeof(q->frames[0].buf)) return 0;
-  if (sem_trywait(&q->empty) != 0) {
+  if (!shm_queue_push(&d->shm->to_tap, buf, len)) {
     printf("\nFRAME LOST\n");
     return 0;
   }
-  f = &q->frames[q->tail];
-  memcpy(f->buf, buf, len);
-  f->len = len;
-  q->tail = (q->tail + 1) % FRAME_QUEUE_DEPTH;
-  sem_post(&q->full);
   return len;
 }
 
 static size_t shm_rx(void *buf, size_t len, struct mg_tcpip_if *ifp) {
   struct shm_driver_data *d = (struct shm_driver_data *) ifp->driver_data;
-  struct shm_queue *q = &d->shm->from_tap;
-  struct shm_frame *f;
-  size_t n;
-
-  if (sem_trywait(&q->full) != 0) return 0;
-  f = &q->frames[q->head];
-  n = f->len;
-  q->head = (q->head + 1) % FRAME_QUEUE_DEPTH;
-  if (n > len) {
-    sem_post(&q->empty);
-    return 0;
-  }
-  memcpy(buf, f->buf, n);
-  sem_post(&q->empty);
-  return n;
+  return shm_queue_pop(&d->shm->from_tap, buf, len);
 }
 
 static bool shm_poll(struct mg_tcpip_if *ifp, bool s1) {
@@ -140,7 +104,7 @@ int main(void) {
   // Setup interface
   const char *mac = "02:00:01:02:03:78";  // MAC address
 
-  printf("Opened SHM interface\n");
+  printf("Opening SHM interface\n");
   usleep(200000);  // 200 ms
 
   result = bsd_x_test(&shm_driver, &shm_data, mac, debug_level);
diff --git a/test/bsd/shm_port_bridge.c b/test/bsd/shm_port_bridge.c
new file mode 100644
index 00000000..c7757390
--- /dev/null
+++ b/test/bsd/shm_port_bridge.c
@@ -0,0 +1,138 @@
+#define _POSIX_C_SOURCE 200809L
+
+#include <arpa/inet.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <netinet/in.h>
+#include <signal.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/mman.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <poll.h>
+#include <unistd.h>
+
+#include "shm_queue.h"
+
+static volatile sig_atomic_t s_signo;
+
+static void signal_handler(int signo) {
+  s_signo = signo;
+}
+
+static int open_udp_socket(void) {
+  const uint16_t port = 0xAA55;      // Local UDP port
+  const uint16_t tap_port = 0x55AA;  // Host port_tap_bridge UDP port
+  struct sockaddr_in port_addr, tap_addr;
+  int fd = socket(AF_INET, SOCK_DGRAM, 0);
+  if (fd < 0) {
+    perror("socket");
+    return -1;
+  }
+
+  memset(&port_addr, 0, sizeof(port_addr));
+  port_addr.sin_family = AF_INET;
+  port_addr.sin_addr.s_addr = htonl(INADDR_ANY);
+  port_addr.sin_port = htons(port);
+  if (bind(fd, (struct sockaddr *) &port_addr, sizeof(port_addr)) < 0) {
+    perror("bind");
+    close(fd);
+    return -1;
+  }
+  printf("Opened UDP socket: 127.0.0.1:%u (0x%04x)\n", (unsigned) port,
+         (unsigned) port);
+
+  memset(&tap_addr, 0, sizeof(tap_addr));
+  tap_addr.sin_family = AF_INET;
+  tap_addr.sin_addr.s_addr = htonl(0x7f000001U);  // 127.0.0.1
+  tap_addr.sin_port = htons(tap_port);
+  if (connect(fd, (struct sockaddr *) &tap_addr, sizeof(tap_addr)) < 0) {
+    perror("connect");
+    close(fd);
+    return -1;
+  }
+  printf("Connected to TAP UDP socket: 127.0.0.1:%u (0x%04x)\n",
+         (unsigned) tap_port, (unsigned) tap_port);
+
+  return fd;
+}
+
+static struct shm_link *open_shm(int *shmfd) {
+  struct shm_link *shm;
+
+  shm_unlink(SHM_NAME);
+  *shmfd = shm_open(SHM_NAME, O_CREAT | O_RDWR, 0600);
+  if (*shmfd < 0) {
+    perror("shm_open");
+    return NULL;
+  }
+  if (ftruncate(*shmfd, (off_t) sizeof(*shm)) != 0) {
+    perror("ftruncate");
+    close(*shmfd);
+    shm_unlink(SHM_NAME);
+    return NULL;
+  }
+  shm = (struct shm_link *) mmap(NULL, sizeof(*shm), PROT_READ | PROT_WRITE,
+                                 MAP_SHARED, *shmfd, 0);
+  if (shm == MAP_FAILED) {
+    perror("mmap");
+    close(*shmfd);
+    shm_unlink(SHM_NAME);
+    return NULL;
+  }
+
+  memset(shm, 0, sizeof(*shm));
+  printf("Opened shared memory: %s\n", SHM_NAME);
+
+  return shm;
+}
+
+int main(void) {
+  int sockfd, shmfd = -1;
+  struct shm_link *shm;
+
+  sockfd = open_udp_socket();
+  if (sockfd < 0) return EXIT_FAILURE;
+  shm = open_shm(&shmfd);
+  if (shm == NULL) {
+    close(sockfd);
+    return EXIT_FAILURE;
+  }
+
+  signal(SIGINT, signal_handler);
+  signal(SIGTERM, signal_handler);
+
+  while (s_signo == 0) {
+    struct pollfd pfd = {sockfd, (short) (POLLIN | POLLPRI), 0};
+    int result = poll(&pfd, 1, 1);
+    if (result < 0 && errno != EINTR) {
+      perror("poll");
+      break;
+    }
+    if (result > 0 && (pfd.revents & (POLLIN | POLLPRI))) {
+      uint8_t buf[FRAME_SIZE];
+      ssize_t nread = recv(sockfd, buf, sizeof(buf), 0);
+      if (nread > 0 &&
+          !shm_queue_push(&shm->from_tap, buf, (size_t) nread)) {
+        printf("FRAME LOST: port -> SHM\n");
+      }
+    }
+    {
+      uint8_t buf[FRAME_SIZE];
+      size_t n = shm_queue_pop(&shm->to_tap, buf, sizeof(buf));
+      if (n > 0) {
+        if (send(sockfd, buf, n, 0) < 0) perror("send");
+      }
+    }
+  }
+
+  munmap(shm, sizeof(*shm));
+  close(shmfd);
+  shm_unlink(SHM_NAME);
+  close(sockfd);
+  return EXIT_SUCCESS;
+}
diff --git a/test/bsd/shm_queue.h b/test/bsd/shm_queue.h
new file mode 100644
index 00000000..c5925f2d
--- /dev/null
+++ b/test/bsd/shm_queue.h
@@ -0,0 +1,60 @@
+#pragma once
+
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+#define FRAME_SIZE 1540
+#define FRAME_QUEUE_DEPTH 32
+#define SHM_NAME "/mongoose-shm"
+
+// Keep this SHM ABI semaphore-free. Some S390/Linux headers pull fcntl/socket
+// declarations through <semaphore.h>, colliding with Mongoose BSD prototypes.
+// Each direction is single-producer/single-consumer and polled, so a simple
+// ring is enough for this test path.
+struct shm_frame {
+  size_t len;
+  uint8_t buf[FRAME_SIZE];
+};
+
+struct shm_queue {
+  volatile size_t head;
+  volatile size_t tail;
+  struct shm_frame frames[FRAME_QUEUE_DEPTH];
+};
+
+struct shm_link {
+  struct shm_queue to_tap;
+  struct shm_queue from_tap;
+};
+
+static inline bool shm_queue_push(struct shm_queue *q, const void *buf,
+                                  size_t len) {
+  size_t tail = q->tail, next = (tail + 1) % FRAME_QUEUE_DEPTH;
+  struct shm_frame *f;
+  if (len > sizeof(q->frames[0].buf) || next == q->head) return false;
+  f = &q->frames[tail];
+  memcpy(f->buf, buf, len);
+  f->len = len;
+  __sync_synchronize();
+  q->tail = next;
+  return true;
+}
+
+static inline size_t shm_queue_pop(struct shm_queue *q, void *buf, size_t len) {
+  size_t head = q->head, n;
+  struct shm_frame *f;
+  if (head == q->tail) return 0;
+  __sync_synchronize();
+  f = &q->frames[head];
+  n = f->len;
+  if (n > len) {
+    q->head = (head + 1) % FRAME_QUEUE_DEPTH;
+    return 0;
+  }
+  memcpy(buf, f->buf, n);
+  __sync_synchronize();
+  q->head = (head + 1) % FRAME_QUEUE_DEPTH;
+  return n;
+}
diff --git a/test/bsd/shm_tap_bridge.c b/test/bsd/shm_tap_bridge.c
index 138e2a68..1a20ed20 100644
--- a/test/bsd/shm_tap_bridge.c
+++ b/test/bsd/shm_tap_bridge.c
@@ -4,39 +4,18 @@
 #include <string.h>
 #include <time.h>
 #include <unistd.h>
+#include <errno.h>
 #include <sys/types.h>
 #include <fcntl.h>
 #include <sys/ioctl.h>
 #include <sys/mman.h>
 #include <sys/poll.h>
 #include <sys/stat.h>
-#include <semaphore.h>
 #include <signal.h>
 
 #include <linux/if.h>
 #include <linux/if_tun.h>
-
-#define FRAME_SIZE 1540
-#define FRAME_QUEUE_DEPTH 8
-#define SHM_NAME "/mongoose-shm"
-
-struct shm_frame {
-  size_t len;
-  uint8_t buf[FRAME_SIZE];
-};
-
-struct shm_queue {
-  sem_t full;
-  sem_t empty;
-  size_t head;
-  size_t tail;
-  struct shm_frame frames[FRAME_QUEUE_DEPTH];
-};
-
-struct shm_link {
-  struct shm_queue to_tap;
-  struct shm_queue from_tap;
-};
+#include "shm_queue.h"
 
 static volatile sig_atomic_t s_signo;
 void signal_handler(int signo) {
@@ -63,17 +42,22 @@ int main(void) {
 
   shm_unlink(SHM_NAME);
   shmfd = shm_open(SHM_NAME, O_CREAT | O_RDWR, 0600);
-  if (shmfd < 0) return EXIT_FAILURE;
-  if (ftruncate(shmfd, sizeof(*shm)) != 0) return EXIT_FAILURE;
+  if (shmfd < 0) {
+    printf("shm_open failed: %d\n", errno);
+    return EXIT_FAILURE;
+  }
+  if (ftruncate(shmfd, sizeof(*shm)) != 0) {
+    printf("ftruncate failed: %d\n", errno);
+    return EXIT_FAILURE;
+  }
   shm = (struct shm_link *) mmap(NULL, sizeof(*shm), PROT_READ | PROT_WRITE,
                                  MAP_SHARED, shmfd, 0);
-  if (shm == MAP_FAILED) return EXIT_FAILURE;
+  if (shm == MAP_FAILED) {
+    printf("mmap failed: %d\n", errno);
+    return EXIT_FAILURE;
+  }
 
   memset(shm, 0, sizeof(*shm));
-  sem_init(&shm->to_tap.full, 1, 0);
-  sem_init(&shm->to_tap.empty, 1, FRAME_QUEUE_DEPTH);
-  sem_init(&shm->from_tap.full, 1, 0);
-  sem_init(&shm->from_tap.empty, 1, FRAME_QUEUE_DEPTH);
   printf("Opened shared memory: %s\n", SHM_NAME);
   signal(SIGINT, signal_handler);
   signal(SIGTERM, signal_handler);
@@ -94,41 +78,31 @@ int main(void) {
 				break;
 			default: 	// the number of fd/sockets with data available (see pfd.revents for each one);
         if (pfd[0].revents & (POLLIN | POLLPRI)) {
-          struct shm_queue *q = &shm->from_tap;
-          ssize_t nread;
-          if (sem_trywait(&q->empty) == 0) {
-            struct shm_frame *f = &q->frames[q->tail];
-            nread = read(fd, f->buf, sizeof(f->buf));
+          uint8_t buf[FRAME_SIZE];
+          ssize_t nread = read(fd, buf, sizeof(buf));
 //          printf("Got at TAP: %d bytes\n", (int) nread);
-            if (nread > 0) {
-              f->len = (size_t) nread;
-              q->tail = (q->tail + 1) % FRAME_QUEUE_DEPTH;
-              sem_post(&q->full);
-//            printf("Sent to SHM\n");
-            } else {
-              sem_post(&q->empty);
+          if (nread > 0) {
+            if (!shm_queue_push(&shm->from_tap, buf, (size_t) nread)) {
+              printf("FRAME LOST: TAP -> SHM\n");
             }
+//            printf("Sent to SHM\n");
           }
         }
 				break;
 			}
-      if (sem_trywait(&shm->to_tap.full) == 0) {
-        struct shm_queue *q = &shm->to_tap;
-        struct shm_frame *f = &q->frames[q->head];
+      {
+        uint8_t buf[FRAME_SIZE];
+        size_t n = shm_queue_pop(&shm->to_tap, buf, sizeof(buf));
+        if (n > 0) {
 //      printf("Got at SHM: %d bytes\n", (int) f->len);
-        write(fd, f->buf, f->len);
-        q->head = (q->head + 1) % FRAME_QUEUE_DEPTH;
-        sem_post(&q->empty);
+          write(fd, buf, n);
 //      printf("Sent to TAP\n");
+        }
       }
 		} while (result >= 0 && s_signo == 0);
     if (result < 0)
       printf("looks like we got an error here... should I quit ? Waiting for CTRL-C\n");
 	}
-  sem_destroy(&shm->from_tap.empty);
-  sem_destroy(&shm->from_tap.full);
-  sem_destroy(&shm->to_tap.empty);
-  sem_destroy(&shm->to_tap.full);
   munmap(shm, sizeof(*shm));
   close(shmfd);
   shm_unlink(SHM_NAME);