Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20171012' into staging

target-arm queue:
 * v8M: SG, BLXNS, secure-return
 * v8M: fixes for coverity issues in previous patches
 * arm: fix armv7m_init() declaration to match definition
 * watchdog/aspeed: fix variable type to store reload value

# gpg: Signature made Thu 12 Oct 2017 17:02:49 BST
# gpg:                using RSA key 0x3C2525ED14360CDE
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>"
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>"
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>"
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* remotes/pmaydell/tags/pull-target-arm-20171012:
  nvic: Fix miscalculation of offsets into ITNS array
  nvic: Add missing 'break'
  target/arm: Implement SG instruction corner cases
  target/arm: Support some Thumb insns being always unconditional
  target-arm: Simplify insn_crosses_page()
  target/arm: Pull Thumb insn word loads up to top level
  target-arm: Don't check for "Thumb2 or M profile" for not-Thumb1
  target/arm: Implement secure function return
  target/arm: Implement BLXNS
  target/arm: Implement SG instruction
  target/arm: Add M profile secure MMU index values to get_a32_user_mem_index()
  arm: fix armv7m_init() declaration to match definition
  watchdog/aspeed: fix variable type to store reload value

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

7 files changed, 521 insertions, 115 deletions

diff --git a/hw/intc/armv7m_nvic.c b/hw/intc/armv7m_nvic.c
index 22d5e6e6af..be46639b63 100644
--- a/hw/intc/armv7m_nvic.c
+++ b/hw/intc/armv7m_nvic.c
@@ -698,7 +698,7 @@ static uint32_t nvic_readl(NVICState *s, uint32_t offset, MemTxAttrs attrs)
         return ((s->num_irq - NVIC_FIRST_IRQ) / 32) - 1;
     case 0x380 ... 0x3bf: /* NVIC_ITNS<n> */
     {
-        int startvec = 32 * (offset - 0x380) + NVIC_FIRST_IRQ;
+        int startvec = 8 * (offset - 0x380) + NVIC_FIRST_IRQ;
         int i;
 
         if (!arm_feature(&cpu->env, ARM_FEATURE_V8)) {
@@ -1102,7 +1102,7 @@ static void nvic_writel(NVICState *s, uint32_t offset, uint32_t value,
     switch (offset) {
     case 0x380 ... 0x3bf: /* NVIC_ITNS<n> */
     {
-        int startvec = 32 * (offset - 0x380) + NVIC_FIRST_IRQ;
+        int startvec = 8 * (offset - 0x380) + NVIC_FIRST_IRQ;
         int i;
 
         if (!arm_feature(&cpu->env, ARM_FEATURE_V8)) {
@@ -1447,6 +1447,7 @@ static void nvic_writel(NVICState *s, uint32_t offset, uint32_t value,
             return;
         }
         cpu->env.sau.ctrl = value & 3;
+        break;
     case 0xdd4: /* SAU_TYPE */
         if (!arm_feature(&cpu->env, ARM_FEATURE_V8)) {
             goto bad_offset;
diff --git a/hw/watchdog/wdt_aspeed.c b/hw/watchdog/wdt_aspeed.c
index 22bce364d7..95f6ad186d 100644
--- a/hw/watchdog/wdt_aspeed.c
+++ b/hw/watchdog/wdt_aspeed.c
@@ -100,13 +100,13 @@ static uint64_t aspeed_wdt_read(void *opaque, hwaddr offset, unsigned size)
 
 static void aspeed_wdt_reload(AspeedWDTState *s, bool pclk)
 {
-    uint32_t reload;
+    uint64_t reload;
 
     if (pclk) {
         reload = muldiv64(s->regs[WDT_RELOAD_VALUE], NANOSECONDS_PER_SECOND,
                           s->pclk_freq);
     } else {
-        reload = s->regs[WDT_RELOAD_VALUE] * 1000;
+        reload = s->regs[WDT_RELOAD_VALUE] * 1000ULL;
     }
 
     if (aspeed_wdt_is_enabled(s)) {
diff --git a/include/hw/arm/arm.h b/include/hw/arm/arm.h
index a3f79d3379..ce769bde6a 100644
--- a/include/hw/arm/arm.h
+++ b/include/hw/arm/arm.h
@@ -25,7 +25,7 @@ typedef enum {
 
 /* armv7m.c */
 DeviceState *armv7m_init(MemoryRegion *system_memory, int mem_size, int num_irq,
-                      const char *kernel_filename, const char *cpu_model);
+                         const char *kernel_filename, const char *cpu_type);
 /**
  * armv7m_load_kernel:
  * @cpu: CPU
diff --git a/target/arm/helper.c b/target/arm/helper.c
index 1d689f00b3..96113fe989 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -41,6 +41,10 @@ typedef struct V8M_SAttributes {
     bool irvalid;
 } V8M_SAttributes;
 
+static void v8m_security_lookup(CPUARMState *env, uint32_t address,
+                                MMUAccessType access_type, ARMMMUIdx mmu_idx,
+                                V8M_SAttributes *sattrs);
+
 /* Definitions for the PMCCNTR and PMCR registers */
 #define PMCRD   0x8
 #define PMCRC   0x4
@@ -5893,6 +5897,12 @@ void HELPER(v7m_bxns)(CPUARMState *env, uint32_t dest)
     g_assert_not_reached();
 }
 
+void HELPER(v7m_blxns)(CPUARMState *env, uint32_t dest)
+{
+    /* translate.c should never generate calls here in user-only mode */
+    g_assert_not_reached();
+}
+
 void switch_mode(CPUARMState *env, int mode)
 {
     ARMCPU *cpu = arm_env_get_cpu(env);
@@ -6164,7 +6174,17 @@ void HELPER(v7m_bxns)(CPUARMState *env, uint32_t dest)
      *  - if the return value is a magic value, do exception return (like BX)
      *  - otherwise bit 0 of the return value is the target security state
      */
-    if (dest >= 0xff000000) {
+    uint32_t min_magic;
+
+    if (arm_feature(env, ARM_FEATURE_M_SECURITY)) {
+        /* Covers FNC_RETURN and EXC_RETURN magic */
+        min_magic = FNC_RETURN_MIN_MAGIC;
+    } else {
+        /* EXC_RETURN magic only */
+        min_magic = EXC_RETURN_MIN_MAGIC;
+    }
+
+    if (dest >= min_magic) {
         /* This is an exception return magic value; put it where
          * do_v7m_exception_exit() expects and raise EXCEPTION_EXIT.
          * Note that if we ever add gen_ss_advance() singlestep support to
@@ -6185,6 +6205,59 @@ void HELPER(v7m_bxns)(CPUARMState *env, uint32_t dest)
     env->regs[15] = dest & ~1;
 }
 
+void HELPER(v7m_blxns)(CPUARMState *env, uint32_t dest)
+{
+    /* Handle v7M BLXNS:
+     *  - bit 0 of the destination address is the target security state
+     */
+
+    /* At this point regs[15] is the address just after the BLXNS */
+    uint32_t nextinst = env->regs[15] | 1;
+    uint32_t sp = env->regs[13] - 8;
+    uint32_t saved_psr;
+
+    /* translate.c will have made BLXNS UNDEF unless we're secure */
+    assert(env->v7m.secure);
+
+    if (dest & 1) {
+        /* target is Secure, so this is just a normal BLX,
+         * except that the low bit doesn't indicate Thumb/not.
+         */
+        env->regs[14] = nextinst;
+        env->thumb = 1;
+        env->regs[15] = dest & ~1;
+        return;
+    }
+
+    /* Target is non-secure: first push a stack frame */
+    if (!QEMU_IS_ALIGNED(sp, 8)) {
+        qemu_log_mask(LOG_GUEST_ERROR,
+                      "BLXNS with misaligned SP is UNPREDICTABLE\n");
+    }
+
+    saved_psr = env->v7m.exception;
+    if (env->v7m.control[M_REG_S] & R_V7M_CONTROL_SFPA_MASK) {
+        saved_psr |= XPSR_SFPA;
+    }
+
+    /* Note that these stores can throw exceptions on MPU faults */
+    cpu_stl_data(env, sp, nextinst);
+    cpu_stl_data(env, sp + 4, saved_psr);
+
+    env->regs[13] = sp;
+    env->regs[14] = 0xfeffffff;
+    if (arm_v7m_is_handler_mode(env)) {
+        /* Write a dummy value to IPSR, to avoid leaking the current secure
+         * exception number to non-secure code. This is guaranteed not
+         * to cause write_v7m_exception() to actually change stacks.
+         */
+        write_v7m_exception(env, 1);
+    }
+    switch_v7m_security_state(env, 0);
+    env->thumb = 1;
+    env->regs[15] = dest;
+}
+
 static uint32_t *get_v7m_sp_ptr(CPUARMState *env, bool secure, bool threadmode,
                                 bool spsel)
 {
@@ -6407,12 +6480,19 @@ static void do_v7m_exception_exit(ARMCPU *cpu)
     bool exc_secure = false;
     bool return_to_secure;
 
-    /* We can only get here from an EXCP_EXCEPTION_EXIT, and
-     * gen_bx_excret() enforces the architectural rule
-     * that jumps to magic addresses don't have magic behaviour unless
-     * we're in Handler mode (compare pseudocode BXWritePC()).
+    /* If we're not in Handler mode then jumps to magic exception-exit
+     * addresses don't have magic behaviour. However for the v8M
+     * security extensions the magic secure-function-return has to
+     * work in thread mode too, so to avoid doing an extra check in
+     * the generated code we allow exception-exit magic to also cause the
+     * internal exception and bring us here in thread mode. Correct code
+     * will never try to do this (the following insn fetch will always
+     * fault) so we the overhead of having taken an unnecessary exception
+     * doesn't matter.
      */
-    assert(arm_v7m_is_handler_mode(env));
+    if (!arm_v7m_is_handler_mode(env)) {
+        return;
+    }
 
     /* In the spec pseudocode ExceptionReturn() is called directly
      * from BXWritePC() and gets the full target PC value including
@@ -6702,6 +6782,78 @@ static void do_v7m_exception_exit(ARMCPU *cpu)
     qemu_log_mask(CPU_LOG_INT, "...successful exception return\n");
 }
 
+static bool do_v7m_function_return(ARMCPU *cpu)
+{
+    /* v8M security extensions magic function return.
+     * We may either:
+     *  (1) throw an exception (longjump)
+     *  (2) return true if we successfully handled the function return
+     *  (3) return false if we failed a consistency check and have
+     *      pended a UsageFault that needs to be taken now
+     *
+     * At this point the magic return value is split between env->regs[15]
+     * and env->thumb. We don't bother to reconstitute it because we don't
+     * need it (all values are handled the same way).
+     */
+    CPUARMState *env = &cpu->env;
+    uint32_t newpc, newpsr, newpsr_exc;
+
+    qemu_log_mask(CPU_LOG_INT, "...really v7M secure function return\n");
+
+    {
+        bool threadmode, spsel;
+        TCGMemOpIdx oi;
+        ARMMMUIdx mmu_idx;
+        uint32_t *frame_sp_p;
+        uint32_t frameptr;
+
+        /* Pull the return address and IPSR from the Secure stack */
+        threadmode = !arm_v7m_is_handler_mode(env);
+        spsel = env->v7m.control[M_REG_S] & R_V7M_CONTROL_SPSEL_MASK;
+
+        frame_sp_p = get_v7m_sp_ptr(env, true, threadmode, spsel);
+        frameptr = *frame_sp_p;
+
+        /* These loads may throw an exception (for MPU faults). We want to
+         * do them as secure, so work out what MMU index that is.
+         */
+        mmu_idx = arm_v7m_mmu_idx_for_secstate(env, true);
+        oi = make_memop_idx(MO_LE, arm_to_core_mmu_idx(mmu_idx));
+        newpc = helper_le_ldul_mmu(env, frameptr, oi, 0);
+        newpsr = helper_le_ldul_mmu(env, frameptr + 4, oi, 0);
+
+        /* Consistency checks on new IPSR */
+        newpsr_exc = newpsr & XPSR_EXCP;
+        if (!((env->v7m.exception == 0 && newpsr_exc == 0) ||
+              (env->v7m.exception == 1 && newpsr_exc != 0))) {
+            /* Pend the fault and tell our caller to take it */
+            env->v7m.cfsr[env->v7m.secure] |= R_V7M_CFSR_INVPC_MASK;
+            armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_USAGE,
+                                    env->v7m.secure);
+            qemu_log_mask(CPU_LOG_INT,
+                          "...taking INVPC UsageFault: "
+                          "IPSR consistency check failed\n");
+            return false;
+        }
+
+        *frame_sp_p = frameptr + 8;
+    }
+
+    /* This invalidates frame_sp_p */
+    switch_v7m_security_state(env, true);
+    env->v7m.exception = newpsr_exc;
+    env->v7m.control[M_REG_S] &= ~R_V7M_CONTROL_SFPA_MASK;
+    if (newpsr & XPSR_SFPA) {
+        env->v7m.control[M_REG_S] |= R_V7M_CONTROL_SFPA_MASK;
+    }
+    xpsr_write(env, 0, XPSR_IT);
+    env->thumb = newpc & 1;
+    env->regs[15] = newpc & ~1;
+
+    qemu_log_mask(CPU_LOG_INT, "...function return successful\n");
+    return true;
+}
+
 static void arm_log_exception(int idx)
 {
     if (qemu_loglevel_mask(CPU_LOG_INT)) {
@@ -6736,6 +6888,126 @@ static void arm_log_exception(int idx)
     }
 }
 
+static bool v7m_read_half_insn(ARMCPU *cpu, ARMMMUIdx mmu_idx,
+                               uint32_t addr, uint16_t *insn)
+{
+    /* Load a 16-bit portion of a v7M instruction, returning true on success,
+     * or false on failure (in which case we will have pended the appropriate
+     * exception).
+     * We need to do the instruction fetch's MPU and SAU checks
+     * like this because there is no MMU index that would allow
+     * doing the load with a single function call. Instead we must
+     * first check that the security attributes permit the load
+     * and that they don't mismatch on the two halves of the instruction,
+     * and then we do the load as a secure load (ie using the security
+     * attributes of the address, not the CPU, as architecturally required).
+     */
+    CPUState *cs = CPU(cpu);
+    CPUARMState *env = &cpu->env;
+    V8M_SAttributes sattrs = {};
+    MemTxAttrs attrs = {};
+    ARMMMUFaultInfo fi = {};
+    MemTxResult txres;
+    target_ulong page_size;
+    hwaddr physaddr;
+    int prot;
+    uint32_t fsr;
+
+    v8m_security_lookup(env, addr, MMU_INST_FETCH, mmu_idx, &sattrs);
+    if (!sattrs.nsc || sattrs.ns) {
+        /* This must be the second half of the insn, and it straddles a
+         * region boundary with the second half not being S&NSC.
+         */
+        env->v7m.sfsr |= R_V7M_SFSR_INVEP_MASK;
+        armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_SECURE, false);
+        qemu_log_mask(CPU_LOG_INT,
+                      "...really SecureFault with SFSR.INVEP\n");
+        return false;
+    }
+    if (get_phys_addr(env, addr, MMU_INST_FETCH, mmu_idx,
+                      &physaddr, &attrs, &prot, &page_size, &fsr, &fi)) {
+        /* the MPU lookup failed */
+        env->v7m.cfsr[env->v7m.secure] |= R_V7M_CFSR_IACCVIOL_MASK;
+        armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_MEM, env->v7m.secure);
+        qemu_log_mask(CPU_LOG_INT, "...really MemManage with CFSR.IACCVIOL\n");
+        return false;
+    }
+    *insn = address_space_lduw_le(arm_addressspace(cs, attrs), physaddr,
+                                 attrs, &txres);
+    if (txres != MEMTX_OK) {
+        env->v7m.cfsr[M_REG_NS] |= R_V7M_CFSR_IBUSERR_MASK;
+        armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_BUS, false);
+        qemu_log_mask(CPU_LOG_INT, "...really BusFault with CFSR.IBUSERR\n");
+        return false;
+    }
+    return true;
+}
+
+static bool v7m_handle_execute_nsc(ARMCPU *cpu)
+{
+    /* Check whether this attempt to execute code in a Secure & NS-Callable
+     * memory region is for an SG instruction; if so, then emulate the
+     * effect of the SG instruction and return true. Otherwise pend
+     * the correct kind of exception and return false.
+     */
+    CPUARMState *env = &cpu->env;
+    ARMMMUIdx mmu_idx;
+    uint16_t insn;
+
+    /* We should never get here unless get_phys_addr_pmsav8() caused
+     * an exception for NS executing in S&NSC memory.
+     */
+    assert(!env->v7m.secure);
+    assert(arm_feature(env, ARM_FEATURE_M_SECURITY));
+
+    /* We want to do the MPU lookup as secure; work out what mmu_idx that is */
+    mmu_idx = arm_v7m_mmu_idx_for_secstate(env, true);
+
+    if (!v7m_read_half_insn(cpu, mmu_idx, env->regs[15], &insn)) {
+        return false;
+    }
+
+    if (!env->thumb) {
+        goto gen_invep;
+    }
+
+    if (insn != 0xe97f) {
+        /* Not an SG instruction first half (we choose the IMPDEF
+         * early-SG-check option).
+         */
+        goto gen_invep;
+    }
+
+    if (!v7m_read_half_insn(cpu, mmu_idx, env->regs[15] + 2, &insn)) {
+        return false;
+    }
+
+    if (insn != 0xe97f) {
+        /* Not an SG instruction second half (yes, both halves of the SG
+         * insn have the same hex value)
+         */
+        goto gen_invep;
+    }
+
+    /* OK, we have confirmed that we really have an SG instruction.
+     * We know we're NS in S memory so don't need to repeat those checks.
+     */
+    qemu_log_mask(CPU_LOG_INT, "...really an SG instruction at 0x%08" PRIx32
+                  ", executing it\n", env->regs[15]);
+    env->regs[14] &= ~1;
+    switch_v7m_security_state(env, true);
+    xpsr_write(env, 0, XPSR_IT);
+    env->regs[15] += 4;
+    return true;
+
+gen_invep:
+    env->v7m.sfsr |= R_V7M_SFSR_INVEP_MASK;
+    armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_SECURE, false);
+    qemu_log_mask(CPU_LOG_INT,
+                  "...really SecureFault with SFSR.INVEP\n");
+    return false;
+}
+
 void arm_v7m_cpu_do_interrupt(CPUState *cs)
 {
     ARMCPU *cpu = ARM_CPU(cs);
@@ -6778,12 +7050,10 @@ void arm_v7m_cpu_do_interrupt(CPUState *cs)
              * the SG instruction have the same security attributes.)
              * Everything else must generate an INVEP SecureFault, so we
              * emulate the SG instruction here.
-             * TODO: actually emulate SG.
              */
-            env->v7m.sfsr |= R_V7M_SFSR_INVEP_MASK;
-            armv7m_nvic_set_pending(env->nvic, ARMV7M_EXCP_SECURE, false);
-            qemu_log_mask(CPU_LOG_INT,
-                          "...really SecureFault with SFSR.INVEP\n");
+            if (v7m_handle_execute_nsc(cpu)) {
+                return;
+            }
             break;
         case M_FAKE_FSR_SFAULT:
             /* Various flavours of SecureFault for attempts to execute or
@@ -6868,8 +7138,18 @@ void arm_v7m_cpu_do_interrupt(CPUState *cs)
     case EXCP_IRQ:
         break;
     case EXCP_EXCEPTION_EXIT:
-        do_v7m_exception_exit(cpu);
-        return;
+        if (env->regs[15] < EXC_RETURN_MIN_MAGIC) {
+            /* Must be v8M security extension function return */
+            assert(env->regs[15] >= FNC_RETURN_MIN_MAGIC);
+            assert(arm_feature(env, ARM_FEATURE_M_SECURITY));
+            if (do_v7m_function_return(cpu)) {
+                return;
+            }
+        } else {
+            do_v7m_exception_exit(cpu);
+            return;
+        }
+        break;
     default:
         cpu_abort(cs, "Unhandled exception 0x%x\n", cs->exception_index);
         return; /* Never happens.  Keep compiler happy.  */
diff --git a/target/arm/helper.h b/target/arm/helper.h
index 64afbac59f..2cf6f74152 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -64,6 +64,7 @@ DEF_HELPER_3(v7m_msr, void, env, i32, i32)
 DEF_HELPER_2(v7m_mrs, i32, env, i32)
 
 DEF_HELPER_2(v7m_bxns, void, env, i32)
+DEF_HELPER_2(v7m_blxns, void, env, i32)
 
 DEF_HELPER_4(access_check_cp_reg, void, env, ptr, i32, i32)
 DEF_HELPER_3(set_cp_reg, void, env, ptr, i32)
diff --git a/target/arm/internals.h b/target/arm/internals.h
index fd9a7e8181..43106a2d6c 100644
--- a/target/arm/internals.h
+++ b/target/arm/internals.h
@@ -60,6 +60,7 @@ static inline bool excp_is_internal(int excp)
 FIELD(V7M_CONTROL, NPRIV, 0, 1)
 FIELD(V7M_CONTROL, SPSEL, 1, 1)
 FIELD(V7M_CONTROL, FPCA, 2, 1)
+FIELD(V7M_CONTROL, SFPA, 3, 1)
 
 /* Bit definitions for v7M exception return payload */
 FIELD(V7M_EXCRET, ES, 0, 1)
@@ -71,6 +72,13 @@ FIELD(V7M_EXCRET, DCRS, 5, 1)
 FIELD(V7M_EXCRET, S, 6, 1)
 FIELD(V7M_EXCRET, RES1, 7, 25) /* including the must-be-1 prefix */
 
+/* Minimum value which is a magic number for exception return */
+#define EXC_RETURN_MIN_MAGIC 0xff000000
+/* Minimum number which is a magic number for function or exception return
+ * when using v8M security extension
+ */
+#define FNC_RETURN_MIN_MAGIC 0xfefffffe
+
 /* We use a few fake FSR values for internal purposes in M profile.
  * M profile cores don't have A/R format FSRs, but currently our
  * get_phys_addr() code assumes A/R profile and reports failures via
diff --git a/target/arm/translate.c b/target/arm/translate.c
index fdc46cc525..4da1a4cbc6 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -165,6 +165,10 @@ static inline int get_a32_user_mem_index(DisasContext *s)
     case ARMMMUIdx_MPriv:
     case ARMMMUIdx_MNegPri:
         return arm_to_core_mmu_idx(ARMMMUIdx_MUser);
+    case ARMMMUIdx_MSUser:
+    case ARMMMUIdx_MSPriv:
+    case ARMMMUIdx_MSNegPri:
+        return arm_to_core_mmu_idx(ARMMMUIdx_MSUser);
     case ARMMMUIdx_S2NS:
     default:
         g_assert_not_reached();
@@ -960,7 +964,8 @@ static inline void gen_bx_excret(DisasContext *s, TCGv_i32 var)
      * s->base.is_jmp that we need to do the rest of the work later.
      */
     gen_bx(s, var);
-    if (s->v7m_handler_mode && arm_dc_feature(s, ARM_FEATURE_M)) {
+    if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY) ||
+        (s->v7m_handler_mode && arm_dc_feature(s, ARM_FEATURE_M))) {
         s->base.is_jmp = DISAS_BX_EXCRET;
     }
 }
@@ -969,9 +974,18 @@ static inline void gen_bx_excret_final_code(DisasContext *s)
 {
     /* Generate the code to finish possible exception return and end the TB */
     TCGLabel *excret_label = gen_new_label();
+    uint32_t min_magic;
+
+    if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY)) {
+        /* Covers FNC_RETURN and EXC_RETURN magic */
+        min_magic = FNC_RETURN_MIN_MAGIC;
+    } else {
+        /* EXC_RETURN magic only */
+        min_magic = EXC_RETURN_MIN_MAGIC;
+    }
 
     /* Is the new PC value in the magic range indicating exception return? */
-    tcg_gen_brcondi_i32(TCG_COND_GEU, cpu_R[15], 0xff000000, excret_label);
+    tcg_gen_brcondi_i32(TCG_COND_GEU, cpu_R[15], min_magic, excret_label);
     /* No: end the TB as we would for a DISAS_JMP */
     if (is_singlestepping(s)) {
         gen_singlestep_exception(s);
@@ -1013,6 +1027,20 @@ static inline void gen_bxns(DisasContext *s, int rm)
     s->base.is_jmp = DISAS_EXIT;
 }
 
+static inline void gen_blxns(DisasContext *s, int rm)
+{
+    TCGv_i32 var = load_reg(s, rm);
+
+    /* We don't need to sync condexec state, for the same reason as bxns.
+     * We do however need to set the PC, because the blxns helper reads it.
+     * The blxns helper may throw an exception.
+     */
+    gen_set_pc_im(s, s->pc);
+    gen_helper_v7m_blxns(cpu_env, var);
+    tcg_temp_free_i32(var);
+    s->base.is_jmp = DISAS_EXIT;
+}
+
 /* Variant of store_reg which uses branch&exchange logic when storing
    to r15 in ARM architecture v7 and above. The source must be a temporary
    and will be marked as dead. */
@@ -9592,6 +9620,44 @@ static void disas_arm_insn(DisasContext *s, unsigned int insn)
     }
 }
 
+static bool thumb_insn_is_16bit(DisasContext *s, uint32_t insn)
+{
+    /* Return true if this is a 16 bit instruction. We must be precise
+     * about this (matching the decode).  We assume that s->pc still
+     * points to the first 16 bits of the insn.
+     */
+    if ((insn >> 11) < 0x1d) {
+        /* Definitely a 16-bit instruction */
+        return true;
+    }
+
+    /* Top five bits 0b11101 / 0b11110 / 0b11111 : this is the
+     * first half of a 32-bit Thumb insn. Thumb-1 cores might
+     * end up actually treating this as two 16-bit insns, though,
+     * if it's half of a bl/blx pair that might span a page boundary.
+     */
+    if (arm_dc_feature(s, ARM_FEATURE_THUMB2)) {
+        /* Thumb2 cores (including all M profile ones) always treat
+         * 32-bit insns as 32-bit.
+         */
+        return false;
+    }
+
+    if ((insn >> 11) == 0x1e && (s->pc < s->next_page_start - 3)) {
+        /* 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix, and the suffix
+         * is not on the next page; we merge this into a 32-bit
+         * insn.
+         */
+        return false;
+    }
+    /* 0b1110_1xxx_xxxx_xxxx : BLX suffix (or UNDEF);
+     * 0b1111_1xxx_xxxx_xxxx : BL suffix;
+     * 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix on the end of a page
+     *  -- handle as single 16 bit insn
+     */
+    return true;
+}
+
 /* Return true if this is a Thumb-2 logical op.  */
 static int
 thumb2_logic_op(int op)
@@ -9677,9 +9743,9 @@ gen_thumb2_data_op(DisasContext *s, int op, int conds, uint32_t shifter_out,
 
 /* Translate a 32-bit thumb instruction.  Returns nonzero if the instruction
    is not legal.  */
-static int disas_thumb2_insn(CPUARMState *env, DisasContext *s, uint16_t insn_hw1)
+static int disas_thumb2_insn(DisasContext *s, uint32_t insn)
 {
-    uint32_t insn, imm, shift, offset;
+    uint32_t imm, shift, offset;
     uint32_t rd, rn, rm, rs;
     TCGv_i32 tmp;
     TCGv_i32 tmp2;
@@ -9691,52 +9757,9 @@ static int disas_thumb2_insn(CPUARMState *env, DisasContext *s, uint16_t insn_hw
     int conds;
     int logic_cc;
 
-    if (!(arm_dc_feature(s, ARM_FEATURE_THUMB2)
-          || arm_dc_feature(s, ARM_FEATURE_M))) {
-        /* Thumb-1 cores may need to treat bl and blx as a pair of
-           16-bit instructions to get correct prefetch abort behavior.  */
-        insn = insn_hw1;
-        if ((insn & (1 << 12)) == 0) {
-            ARCH(5);
-            /* Second half of blx.  */
-            offset = ((insn & 0x7ff) << 1);
-            tmp = load_reg(s, 14);
-            tcg_gen_addi_i32(tmp, tmp, offset);
-            tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
-
-            tmp2 = tcg_temp_new_i32();
-            tcg_gen_movi_i32(tmp2, s->pc | 1);
-            store_reg(s, 14, tmp2);
-            gen_bx(s, tmp);
-            return 0;
-        }
-        if (insn & (1 << 11)) {
-            /* Second half of bl.  */
-            offset = ((insn & 0x7ff) << 1) | 1;
-            tmp = load_reg(s, 14);
-            tcg_gen_addi_i32(tmp, tmp, offset);
-
-            tmp2 = tcg_temp_new_i32();
-            tcg_gen_movi_i32(tmp2, s->pc | 1);
-            store_reg(s, 14, tmp2);
-            gen_bx(s, tmp);
-            return 0;
-        }
-        if ((s->pc & ~TARGET_PAGE_MASK) == 0) {
-            /* Instruction spans a page boundary.  Implement it as two
-               16-bit instructions in case the second half causes an
-               prefetch abort.  */
-            offset = ((int32_t)insn << 21) >> 9;
-            tcg_gen_movi_i32(cpu_R[14], s->pc + 2 + offset);
-            return 0;
-        }
-        /* Fall through to 32-bit decode.  */
-    }
-
-    insn = arm_lduw_code(env, s->pc, s->sctlr_b);
-    s->pc += 2;
-    insn |= (uint32_t)insn_hw1 << 16;
-
+    /* The only 32 bit insn that's allowed for Thumb1 is the combined
+     * BL/BLX prefix and suffix.
+     */
     if ((insn & 0xf800e800) != 0xf000e800) {
         ARCH(6T2);
     }
@@ -9755,7 +9778,28 @@ static int disas_thumb2_insn(CPUARMState *env, DisasContext *s, uint16_t insn_hw
              * - load/store doubleword, load/store exclusive, ldacq/strel,
              *   table branch.
              */
-            if (insn & 0x01200000) {
+            if (insn == 0xe97fe97f && arm_dc_feature(s, ARM_FEATURE_M) &&
+                arm_dc_feature(s, ARM_FEATURE_V8)) {
+                /* 0b1110_1001_0111_1111_1110_1001_0111_111
+                 *  - SG (v8M only)
+                 * The bulk of the behaviour for this instruction is implemented
+                 * in v7m_handle_execute_nsc(), which deals with the insn when
+                 * it is executed by a CPU in non-secure state from memory
+                 * which is Secure & NonSecure-Callable.
+                 * Here we only need to handle the remaining cases:
+                 *  * in NS memory (including the "security extension not
+                 *    implemented" case) : NOP
+                 *  * in S memory but CPU already secure (clear IT bits)
+                 * We know that the attribute for the memory this insn is
+                 * in must match the current CPU state, because otherwise
+                 * get_phys_addr_pmsav8 would have generated an exception.
+                 */
+                if (s->v8m_secure) {
+                    /* Like the IT insn, we don't need to generate any code */
+                    s->condexec_cond = 0;
+                    s->condexec_mask = 0;
+                }
+            } else if (insn & 0x01200000) {
                 /* 0b1110_1000_x11x_xxxx_xxxx_xxxx_xxxx_xxxx
                  *  - load/store dual (post-indexed)
                  * 0b1111_1001_x10x_xxxx_xxxx_xxxx_xxxx_xxxx
@@ -11051,27 +11095,15 @@ illegal_op:
     return 1;
 }
 
-static void disas_thumb_insn(CPUARMState *env, DisasContext *s)
+static void disas_thumb_insn(DisasContext *s, uint32_t insn)
 {
-    uint32_t val, insn, op, rm, rn, rd, shift, cond;
+    uint32_t val, op, rm, rn, rd, shift, cond;
     int32_t offset;
     int i;
     TCGv_i32 tmp;
     TCGv_i32 tmp2;
     TCGv_i32 addr;
 
-    if (s->condexec_mask) {
-        cond = s->condexec_cond;
-        if (cond != 0x0e) {     /* Skip conditional when condition is AL. */
-          s->condlabel = gen_new_label();
-          arm_gen_test_cc(cond ^ 1, s->condlabel);
-          s->condjmp = 1;
-        }
-    }
-
-    insn = arm_lduw_code(env, s->pc, s->sctlr_b);
-    s->pc += 2;
-
     switch (insn >> 12) {
     case 0: case 1:
 
@@ -11218,8 +11250,7 @@ static void disas_thumb_insn(CPUARMState *env, DisasContext *s)
                         goto undef;
                     }
                     if (link) {
-                        /* BLXNS: not yet implemented */
-                        goto undef;
+                        gen_blxns(s, rm);
                     } else {
                         gen_bxns(s, rm);
                     }
@@ -11803,8 +11834,21 @@ static void disas_thumb_insn(CPUARMState *env, DisasContext *s)
 
     case 14:
         if (insn & (1 << 11)) {
-            if (disas_thumb2_insn(env, s, insn))
-              goto undef32;
+            /* thumb_insn_is_16bit() ensures we can't get here for
+             * a Thumb2 CPU, so this must be a thumb1 split BL/BLX:
+             * 0b1110_1xxx_xxxx_xxxx : BLX suffix (or UNDEF)
+             */
+            assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
+            ARCH(5);
+            offset = ((insn & 0x7ff) << 1);
+            tmp = load_reg(s, 14);
+            tcg_gen_addi_i32(tmp, tmp, offset);
+            tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
+
+            tmp2 = tcg_temp_new_i32();
+            tcg_gen_movi_i32(tmp2, s->pc | 1);
+            store_reg(s, 14, tmp2);
+            gen_bx(s, tmp);
             break;
         }
         /* unconditional branch */
@@ -11815,15 +11859,30 @@ static void disas_thumb_insn(CPUARMState *env, DisasContext *s)
         break;
 
     case 15:
-        if (disas_thumb2_insn(env, s, insn))
-            goto undef32;
+        /* thumb_insn_is_16bit() ensures we can't get here for
+         * a Thumb2 CPU, so this must be a thumb1 split BL/BLX.
+         */
+        assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
+
+        if (insn & (1 << 11)) {
+            /* 0b1111_1xxx_xxxx_xxxx : BL suffix */
+            offset = ((insn & 0x7ff) << 1) | 1;
+            tmp = load_reg(s, 14);
+            tcg_gen_addi_i32(tmp, tmp, offset);
+
+            tmp2 = tcg_temp_new_i32();
+            tcg_gen_movi_i32(tmp2, s->pc | 1);
+            store_reg(s, 14, tmp2);
+            gen_bx(s, tmp);
+        } else {
+            /* 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix */
+            uint32_t uoffset = ((int32_t)insn << 21) >> 9;
+
+            tcg_gen_movi_i32(cpu_R[14], s->pc + 2 + uoffset);
+        }
         break;
     }
     return;
-undef32:
-    gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
-                       default_exception_el(s));
-    return;
 illegal_op:
 undef:
     gen_exception_insn(s, 2, EXCP_UDEF, syn_uncategorized(),
@@ -11834,29 +11893,14 @@ static bool insn_crosses_page(CPUARMState *env, DisasContext *s)
 {
     /* Return true if the insn at dc->pc might cross a page boundary.
      * (False positives are OK, false negatives are not.)
+     * We know this is a Thumb insn, and our caller ensures we are
+     * only called if dc->pc is less than 4 bytes from the page
+     * boundary, so we cross the page if the first 16 bits indicate
+     * that this is a 32 bit insn.
      */
-    uint16_t insn;
-
-    if ((s->pc & 3) == 0) {
-        /* At a 4-aligned address we can't be crossing a page */
-        return false;
-    }
-
-    /* This must be a Thumb insn */
-    insn = arm_lduw_code(env, s->pc, s->sctlr_b);
+    uint16_t insn = arm_lduw_code(env, s->pc, s->sctlr_b);
 
-    if ((insn >> 11) >= 0x1d) {
-        /* Top five bits 0b11101 / 0b11110 / 0b11111 : this is the
-         * First half of a 32-bit Thumb insn. Thumb-1 cores might
-         * end up actually treating this as two 16-bit insns (see the
-         * code at the start of disas_thumb2_insn()) but we don't bother
-         * to check for that as it is unlikely, and false positives here
-         * are harmless.
-         */
-        return true;
-    }
-    /* Definitely a 16-bit insn, can't be crossing a page. */
-    return false;
+    return !thumb_insn_is_16bit(s, insn);
 }
 
 static int arm_tr_init_disas_context(DisasContextBase *dcbase,
@@ -12089,16 +12133,88 @@ static void arm_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
        in init_disas_context by adjusting max_insns.  */
 }
 
+static bool thumb_insn_is_unconditional(DisasContext *s, uint32_t insn)
+{
+    /* Return true if this Thumb insn is always unconditional,
+     * even inside an IT block. This is true of only a very few
+     * instructions: BKPT, HLT, and SG.
+     *
+     * A larger class of instructions are UNPREDICTABLE if used
+     * inside an IT block; we do not need to detect those here, because
+     * what we do by default (perform the cc check and update the IT
+     * bits state machine) is a permitted CONSTRAINED UNPREDICTABLE
+     * choice for those situations.
+     *
+     * insn is either a 16-bit or a 32-bit instruction; the two are
+     * distinguishable because for the 16-bit case the top 16 bits
+     * are zeroes, and that isn't a valid 32-bit encoding.
+     */
+    if ((insn & 0xffffff00) == 0xbe00) {
+        /* BKPT */
+        return true;
+    }
+
+    if ((insn & 0xffffffc0) == 0xba80 && arm_dc_feature(s, ARM_FEATURE_V8) &&
+        !arm_dc_feature(s, ARM_FEATURE_M)) {
+        /* HLT: v8A only. This is unconditional even when it is going to
+         * UNDEF; see the v8A ARM ARM DDI0487B.a H3.3.
+         * For v7 cores this was a plain old undefined encoding and so
+         * honours its cc check. (We might be using the encoding as
+         * a semihosting trap, but we don't change the cc check behaviour
+         * on that account, because a debugger connected to a real v7A
+         * core and emulating semihosting traps by catching the UNDEF
+         * exception would also only see cases where the cc check passed.
+         * No guest code should be trying to do a HLT semihosting trap
+         * in an IT block anyway.
+         */
+        return true;
+    }
+
+    if (insn == 0xe97fe97f && arm_dc_feature(s, ARM_FEATURE_V8) &&
+        arm_dc_feature(s, ARM_FEATURE_M)) {
+        /* SG: v8M only */
+        return true;
+    }
+
+    return false;
+}
+
 static void thumb_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
 {
     DisasContext *dc = container_of(dcbase, DisasContext, base);
     CPUARMState *env = cpu->env_ptr;
+    uint32_t insn;
+    bool is_16bit;
 
     if (arm_pre_translate_insn(dc)) {
         return;
     }
 
-    disas_thumb_insn(env, dc);
+    insn = arm_lduw_code(env, dc->pc, dc->sctlr_b);
+    is_16bit = thumb_insn_is_16bit(dc, insn);
+    dc->pc += 2;
+    if (!is_16bit) {
+        uint32_t insn2 = arm_lduw_code(env, dc->pc, dc->sctlr_b);
+
+        insn = insn << 16 | insn2;
+        dc->pc += 2;
+    }
+
+    if (dc->condexec_mask && !thumb_insn_is_unconditional(dc, insn)) {
+        uint32_t cond = dc->condexec_cond;
+
+        if (cond != 0x0e) {     /* Skip conditional when condition is AL. */
+            dc->condlabel = gen_new_label();
+            arm_gen_test_cc(cond ^ 1, dc->condlabel);
+            dc->condjmp = 1;
+        }
+    }
+
+    if (is_16bit) {
+        disas_thumb_insn(dc, insn);
+    } else {
+        disas_thumb2_insn(dc, insn);
+    }
 
     /* Advance the Thumb condexec condition.  */
     if (dc->condexec_mask) {