diff --git a/.gitlab-ci.d/static_checks.yml b/.gitlab-ci.d/static_checks.yml
index 902843f8b3..5e955540d3 100644
--- a/.gitlab-ci.d/static_checks.yml
+++ b/.gitlab-ci.d/static_checks.yml
@@ -46,4 +46,6 @@ check-python-tox:
     QEMU_TOX_EXTRA_ARGS: --skip-missing-interpreters=false
   needs:
     job: python-container
+  rules:
+    - when: manual
   allow_failure: true
diff --git a/configure b/configure
index eb977e5b6f..030728d11e 100755
--- a/configure
+++ b/configure
@@ -3821,7 +3821,7 @@ done
 (for i in $cross_cc_vars; do
   export $i
 done
-export target_list source_path use_containers ARCH
+export target_list source_path use_containers cpu
 $source_path/tests/tcg/configure.sh)
 
 # temporary config to build submodules
diff --git a/meson.build b/meson.build
index 886f0a9343..53065e96ec 100644
--- a/meson.build
+++ b/meson.build
@@ -74,6 +74,8 @@ if cpu not in supported_cpus
   host_arch = 'unknown'
 elif cpu == 'x86'
   host_arch = 'i386'
+elif cpu == 'mips64'
+  host_arch = 'mips'
 else
   host_arch = cpu
 endif
diff --git a/target/hppa/op_helper.c b/target/hppa/op_helper.c
index 96d9391c39..1b86557d5d 100644
--- a/target/hppa/op_helper.c
+++ b/target/hppa/op_helper.c
@@ -57,26 +57,29 @@ void HELPER(tcond)(CPUHPPAState *env, target_ureg cond)
     }
 }
 
-static void atomic_store_3(CPUHPPAState *env, target_ulong addr, uint32_t val,
-                           uint32_t mask, uintptr_t ra)
+static void atomic_store_3(CPUHPPAState *env, target_ulong addr,
+                           uint32_t val, uintptr_t ra)
 {
-#ifdef CONFIG_USER_ONLY
-    uint32_t old, new, cmp;
+    int mmu_idx = cpu_mmu_index(env, 0);
+    uint32_t old, new, cmp, mask, *haddr;
+    void *vaddr;
+
+    vaddr = probe_access(env, addr, 3, MMU_DATA_STORE, mmu_idx, ra);
+    if (vaddr == NULL) {
+        cpu_loop_exit_atomic(env_cpu(env), ra);
+    }
+    haddr = (uint32_t *)((uintptr_t)vaddr & -4);
+    mask = addr & 1 ? 0x00ffffffu : 0xffffff00u;
 
-    uint32_t *haddr = g2h(env_cpu(env), addr - 1);
     old = *haddr;
     while (1) {
-        new = (old & ~mask) | (val & mask);
+        new = be32_to_cpu((cpu_to_be32(old) & ~mask) | (val & mask));
         cmp = qatomic_cmpxchg(haddr, old, new);
         if (cmp == old) {
             return;
         }
         old = cmp;
     }
-#else
-    /* FIXME -- we can do better.  */
-    cpu_loop_exit_atomic(env_cpu(env), ra);
-#endif
 }
 
 static void do_stby_b(CPUHPPAState *env, target_ulong addr, target_ureg val,
@@ -92,7 +95,7 @@ static void do_stby_b(CPUHPPAState *env, target_ulong addr, target_ureg val,
     case 1:
         /* The 3 byte store must appear atomic.  */
         if (parallel) {
-            atomic_store_3(env, addr, val, 0x00ffffffu, ra);
+            atomic_store_3(env, addr, val, ra);
         } else {
             cpu_stb_data_ra(env, addr, val >> 16, ra);
             cpu_stw_data_ra(env, addr + 1, val, ra);
@@ -122,7 +125,7 @@ static void do_stby_e(CPUHPPAState *env, target_ulong addr, target_ureg val,
     case 3:
         /* The 3 byte store must appear atomic.  */
         if (parallel) {
-            atomic_store_3(env, addr - 3, val, 0xffffff00u, ra);
+            atomic_store_3(env, addr - 3, val, ra);
         } else {
             cpu_stw_data_ra(env, addr - 3, val >> 16, ra);
             cpu_stb_data_ra(env, addr - 1, val >> 8, ra);
diff --git a/tests/tcg/Makefile.target b/tests/tcg/Makefile.target
index 63cf1b2573..0f8645f782 100644
--- a/tests/tcg/Makefile.target
+++ b/tests/tcg/Makefile.target
@@ -82,13 +82,12 @@ QEMU_OPTS=
 
 
 # If TCG debugging, or TCI is enabled things are a lot slower
-ifneq ($(CONFIG_TCG_INTERPRETER),)
+# ??? Makefile no longer has any indication that TCI is enabled,
+# but for the record:
+#   15s    original default
+#   60s    with --enable-debug
+#   90s    with --enable-tcg-interpreter
 TIMEOUT=90
-else ifneq ($(CONFIG_DEBUG_TCG),)
-TIMEOUT=60
-else
-TIMEOUT=15
-endif
 
 ifdef CONFIG_USER_ONLY
 # The order we include is important. We include multiarch first and
@@ -144,7 +143,6 @@ PLUGINS=$(patsubst %.c, lib%.so, $(notdir $(wildcard $(PLUGIN_SRC)/*.c)))
 $(foreach p,$(PLUGINS), \
 	$(foreach t,$(TESTS),\
 		$(eval run-plugin-$(t)-with-$(p): $t $p) \
-		$(eval run-plugin-$(t)-with-$(p): TIMEOUT=60) \
 		$(eval RUN_TESTS+=run-plugin-$(t)-with-$(p))))
 endif
 
diff --git a/tests/tcg/configure.sh b/tests/tcg/configure.sh
index 9ef913df5b..8eb4287c84 100755
--- a/tests/tcg/configure.sh
+++ b/tests/tcg/configure.sh
@@ -326,7 +326,7 @@ for target in $target_list; do
   elif test $got_cross_cc = no && test "$container" != no && \
           test -n "$container_image"; then
       for host in $container_hosts; do
-          if test "$host" = "$ARCH"; then
+          if test "$host" = "$cpu"; then
               echo "DOCKER_IMAGE=$container_image" >> $config_target_mak
               echo "DOCKER_CROSS_CC_GUEST=$container_cross_cc" >> \
                    $config_target_mak
diff --git a/tests/tcg/hppa/Makefile.target b/tests/tcg/hppa/Makefile.target
index d0d5e0e257..b78e6b4849 100644
--- a/tests/tcg/hppa/Makefile.target
+++ b/tests/tcg/hppa/Makefile.target
@@ -12,3 +12,8 @@ run-signals: signals
 	$(call skip-test, $<, "BROKEN awaiting vdso support")
 run-plugin-signals-with-%:
 	$(call skip-test, $<, "BROKEN awaiting vdso support")
+
+VPATH += $(SRC_PATH)/tests/tcg/hppa
+TESTS += stby
+
+stby: CFLAGS += -pthread
diff --git a/tests/tcg/hppa/stby.c b/tests/tcg/hppa/stby.c
new file mode 100644
index 0000000000..36bd5f723c
--- /dev/null
+++ b/tests/tcg/hppa/stby.c
@@ -0,0 +1,87 @@
+/* Test STBY */
+
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+
+struct S {
+    unsigned a;
+    unsigned b;
+    unsigned c;
+};
+
+static void check(const struct S *s, unsigned e,
+                  const char *which, const char *insn, int ofs)
+{
+    int err = 0;
+
+    if (s->a != 0) {
+        fprintf(stderr, "%s %s %d: garbage before word 0x%08x\n",
+                which, insn, ofs, s->a);
+        err = 1;
+    }
+    if (s->c != 0) {
+        fprintf(stderr, "%s %s %d: garbage after word 0x%08x\n",
+                which, insn, ofs, s->c);
+        err = 1;
+    }
+    if (s->b != e) {
+        fprintf(stderr, "%s %s %d: 0x%08x != 0x%08x\n",
+                which, insn, ofs, s->b, e);
+        err = 1;
+    }
+
+    if (err) {
+        exit(1);
+    }
+}
+
+#define TEST(INSN, OFS, E)                                         \
+    do {                                                           \
+        s.b = 0;                                                   \
+        asm volatile(INSN " %1, " #OFS "(%0)"                      \
+                     : : "r"(&s.b), "r" (0x11223344) : "memory");  \
+        check(&s, E, which, INSN, OFS);                            \
+    } while (0)
+
+static void test(const char *which)
+{
+    struct S s = { };
+
+    TEST("stby,b", 0, 0x11223344);
+    TEST("stby,b", 1, 0x00223344);
+    TEST("stby,b", 2, 0x00003344);
+    TEST("stby,b", 3, 0x00000044);
+
+    TEST("stby,e", 0, 0x00000000);
+    TEST("stby,e", 1, 0x11000000);
+    TEST("stby,e", 2, 0x11220000);
+    TEST("stby,e", 3, 0x11223300);
+}
+
+static void *child(void *x)
+{
+    return NULL;
+}
+
+int main()
+{
+    int err;
+    pthread_t thr;
+
+    /* Run test in serial mode */
+    test("serial");
+
+    /* Create a dummy thread to start parallel mode. */
+    err = pthread_create(&thr, NULL, child, NULL);
+    if (err != 0) {
+        fprintf(stderr, "pthread_create: %s\n", strerror(err));
+        return 2;
+    }
+
+    /* Run test in parallel mode */
+    test("parallel");
+    return 0;
+}