[Qemu-devel] [PATCH v4 0/3] target/arm: Reduce overhead of cpu_get_tb_cpu_state

[Qemu-devel] [PATCH v4 0/3] target/arm: Reduce overhead of cpu_get_tb_cpu_state

[Richard Henderson]

Version 3 was back in February:
https://lists.gnu.org/archive/html/qemu-devel/2019-02/msg06002.html

Changes since v3:
  * Rebase.
  * Do not cache XSCALE_CPAR now that it overlaps VECSTRIDE.
  * Leave the new v7m bits as uncached.  I haven't figured
    out all of the ways fpccr is modified.

Changes since v2:
  * Do not cache VECLEN, VECSTRIDE, VFPEN.
    These variables come from VFP_FPSCR and VFP_FPEXC, not from
    system control registers.
  * Move HANDLER and STACKCHECK to rebuild_hflags_a32,
    instead of building them in rebuild_hflags_common.

Changes since v1:
  * Apparently I had started a last-minute API change, and failed to
    covert all of the users, and also failed to re-test afterward.
  * Retain assertions for --enable-debug-tcg.


r~


Richard Henderson (3):
  target/arm: Split out recompute_hflags et al
  target/arm: Rebuild hflags at EL changes and MSR writes
  target/arm: Rely on hflags correct in cpu_get_tb_cpu_state

 target/arm/cpu.h           |  35 ++--
 target/arm/helper.h        |   3 +
 target/arm/internals.h     |   3 +
 linux-user/syscall.c       |   1 +
 target/arm/cpu.c           |   1 +
 target/arm/helper-a64.c    |   3 +
 target/arm/helper.c        | 334 ++++++++++++++++++++++---------------
 target/arm/machine.c       |   1 +
 target/arm/op_helper.c     |   1 +
 target/arm/translate-a64.c |   6 +-
 target/arm/translate.c     |  14 +-
 11 files changed, 254 insertions(+), 148 deletions(-)

-- 
2.17.1

[Qemu-devel] [PATCH v4 1/3] target/arm: Split out recompute_hflags et al

[Richard Henderson]

Create functions to compute the values of the a64 and a32 hflags,
as well as one to compute the values that are shared between them.
For now, the env->hflags variable is not used, and the results are
fed back to cpu_get_tb_cpu_state.

Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
v3: Do not cache VECLEN, VECSTRIDE, VFPEN.
    Move HANDLER and STACKCHECK to rebuild_hflags_a32.
v4: Do not cache XSCALE_CPAR now that it overlaps VECSTRIDE.
---
 target/arm/cpu.h       |  35 +++--
 target/arm/helper.h    |   3 +
 target/arm/internals.h |   3 +
 target/arm/helper.c    | 322 ++++++++++++++++++++++++-----------------
 4 files changed, 218 insertions(+), 145 deletions(-)

diff --git a/target/arm/cpu.h b/target/arm/cpu.h
index 94c990cddb..c13633e6a0 100644
--- a/target/arm/cpu.h
+++ b/target/arm/cpu.h
@@ -231,6 +231,9 @@ typedef struct CPUARMState {
     uint32_t pstate;
     uint32_t aarch64; /* 1 if CPU is in aarch64 state; inverse of PSTATE.nRW */
 
+    /* Cached TBFLAGS state.  See below for which bits are included.  */
+    uint32_t hflags;
+
     /* Frequently accessed CPSR bits are stored separately for efficiency.
        This contains all the other bits.  Use cpsr_{read,write} to access
        the whole CPSR.  */
@@ -3130,27 +3133,31 @@ typedef ARMCPU ArchCPU;
 
 #include "exec/cpu-all.h"
 
-/* Bit usage in the TB flags field: bit 31 indicates whether we are
+/*
+ * Bit usage in the TB flags field: bit 31 indicates whether we are
  * in 32 or 64 bit mode. The meaning of the other bits depends on that.
  * We put flags which are shared between 32 and 64 bit mode at the top
  * of the word, and flags which apply to only one mode at the bottom.
+ *
+ * Unless otherwise noted, these bits are cached in env->hflags.
  */
 FIELD(TBFLAG_ANY, AARCH64_STATE, 31, 1)
 FIELD(TBFLAG_ANY, MMUIDX, 28, 3)
 FIELD(TBFLAG_ANY, SS_ACTIVE, 27, 1)
-FIELD(TBFLAG_ANY, PSTATE_SS, 26, 1)
+FIELD(TBFLAG_ANY, PSTATE_SS, 26, 1)     /* Not cached. */
 /* Target EL if we take a floating-point-disabled exception */
 FIELD(TBFLAG_ANY, FPEXC_EL, 24, 2)
 FIELD(TBFLAG_ANY, BE_DATA, 23, 1)
 
 /* Bit usage when in AArch32 state: */
-FIELD(TBFLAG_A32, THUMB, 0, 1)
-FIELD(TBFLAG_A32, VECLEN, 1, 3)
-FIELD(TBFLAG_A32, VECSTRIDE, 4, 2)
+FIELD(TBFLAG_A32, THUMB, 0, 1)          /* Not cached. */
+FIELD(TBFLAG_A32, VECLEN, 1, 3)         /* Not cached. */
+FIELD(TBFLAG_A32, VECSTRIDE, 4, 2)      /* Not cached. */
 /*
  * We store the bottom two bits of the CPAR as TB flags and handle
  * checks on the other bits at runtime. This shares the same bits as
  * VECSTRIDE, which is OK as no XScale CPU has VFP.
+ * Not cached, because VECLEN+VECSTRIDE are not cached.
  */
 FIELD(TBFLAG_A32, XSCALE_CPAR, 4, 2)
 /*
@@ -3159,15 +3166,15 @@ FIELD(TBFLAG_A32, XSCALE_CPAR, 4, 2)
  * the same thing as the current security state of the processor!
  */
 FIELD(TBFLAG_A32, NS, 6, 1)
-FIELD(TBFLAG_A32, VFPEN, 7, 1)
-FIELD(TBFLAG_A32, CONDEXEC, 8, 8)
+FIELD(TBFLAG_A32, VFPEN, 7, 1)          /* Not cached. */
+FIELD(TBFLAG_A32, CONDEXEC, 8, 8)       /* Not cached. */
 FIELD(TBFLAG_A32, SCTLR_B, 16, 1)
 /* For M profile only, set if FPCCR.LSPACT is set */
-FIELD(TBFLAG_A32, LSPACT, 18, 1)
+FIELD(TBFLAG_A32, LSPACT, 18, 1)        /* Not cached. */
 /* For M profile only, set if we must create a new FP context */
-FIELD(TBFLAG_A32, NEW_FP_CTXT_NEEDED, 19, 1)
+FIELD(TBFLAG_A32, NEW_FP_CTXT_NEEDED, 19, 1) /* Not cached. */
 /* For M profile only, set if FPCCR.S does not match current security state */
-FIELD(TBFLAG_A32, FPCCR_S_WRONG, 20, 1)
+FIELD(TBFLAG_A32, FPCCR_S_WRONG, 20, 1) /* Not cached. */
 /* For M profile only, Handler (ie not Thread) mode */
 FIELD(TBFLAG_A32, HANDLER, 21, 1)
 /* For M profile only, whether we should generate stack-limit checks */
@@ -3179,7 +3186,7 @@ FIELD(TBFLAG_A64, SVEEXC_EL, 2, 2)
 FIELD(TBFLAG_A64, ZCR_LEN, 4, 4)
 FIELD(TBFLAG_A64, PAUTH_ACTIVE, 8, 1)
 FIELD(TBFLAG_A64, BT, 9, 1)
-FIELD(TBFLAG_A64, BTYPE, 10, 2)
+FIELD(TBFLAG_A64, BTYPE, 10, 2)         /* Not cached. */
 FIELD(TBFLAG_A64, TBID, 12, 2)
 
 static inline bool bswap_code(bool sctlr_b)
@@ -3264,6 +3271,12 @@ void arm_register_pre_el_change_hook(ARMCPU *cpu, ARMELChangeHookFn *hook,
 void arm_register_el_change_hook(ARMCPU *cpu, ARMELChangeHookFn *hook, void
         *opaque);
 
+/**
+ * arm_rebuild_hflags:
+ * Rebuild the cached TBFLAGS for arbitrary changed processor state.
+ */
+void arm_rebuild_hflags(CPUARMState *env);
+
 /**
  * aa32_vfp_dreg:
  * Return a pointer to the Dn register within env in 32-bit mode.
diff --git a/target/arm/helper.h b/target/arm/helper.h
index 132aa1682e..3919acbe63 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -91,6 +91,9 @@ DEF_HELPER_4(msr_banked, void, env, i32, i32, i32)
 DEF_HELPER_2(get_user_reg, i32, env, i32)
 DEF_HELPER_3(set_user_reg, void, env, i32, i32)
 
+DEF_HELPER_FLAGS_2(rebuild_hflags_a32, TCG_CALL_NO_RWG, void, env, i32)
+DEF_HELPER_FLAGS_2(rebuild_hflags_a64, TCG_CALL_NO_RWG, void, env, i32)
+
 DEF_HELPER_1(vfp_get_fpscr, i32, env)
 DEF_HELPER_2(vfp_set_fpscr, void, env, i32)
 
diff --git a/target/arm/internals.h b/target/arm/internals.h
index 232d963875..db6f010798 100644
--- a/target/arm/internals.h
+++ b/target/arm/internals.h
@@ -992,6 +992,9 @@ ARMVAParameters aa64_va_parameters_both(CPUARMState *env, uint64_t va,
 ARMVAParameters aa64_va_parameters(CPUARMState *env, uint64_t va,
                                    ARMMMUIdx mmu_idx, bool data);
 
+uint32_t rebuild_hflags_a32(CPUARMState *env, int el);
+uint32_t rebuild_hflags_a64(CPUARMState *env, int el);
+
 static inline int exception_target_el(CPUARMState *env)
 {
     int target_el = MAX(1, arm_current_el(env));
diff --git a/target/arm/helper.c b/target/arm/helper.c
index b74c23a9bc..43b7c41f11 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -11013,165 +11013,219 @@ ARMMMUIdx arm_stage1_mmu_idx(CPUARMState *env)
 }
 #endif
 
-void cpu_get_tb_cpu_state(CPUARMState *env, target_ulong *pc,
-                          target_ulong *cs_base, uint32_t *pflags)
+static uint32_t common_hflags(CPUARMState *env, int el, ARMMMUIdx mmu_idx,
+                              int fp_el, uint32_t flags)
 {
-    ARMMMUIdx mmu_idx = arm_mmu_idx(env);
-    int current_el = arm_current_el(env);
-    int fp_el = fp_exception_el(env, current_el);
-    uint32_t flags = 0;
-
-    if (is_a64(env)) {
-        ARMCPU *cpu = env_archcpu(env);
-        uint64_t sctlr;
-
-        *pc = env->pc;
-        flags = FIELD_DP32(flags, TBFLAG_ANY, AARCH64_STATE, 1);
-
-        /* Get control bits for tagged addresses.  */
-        {
-            ARMMMUIdx stage1 = stage_1_mmu_idx(mmu_idx);
-            ARMVAParameters p0 = aa64_va_parameters_both(env, 0, stage1);
-            int tbii, tbid;
-
-            /* FIXME: ARMv8.1-VHE S2 translation regime.  */
-            if (regime_el(env, stage1) < 2) {
-                ARMVAParameters p1 = aa64_va_parameters_both(env, -1, stage1);
-                tbid = (p1.tbi << 1) | p0.tbi;
-                tbii = tbid & ~((p1.tbid << 1) | p0.tbid);
-            } else {
-                tbid = p0.tbi;
-                tbii = tbid & !p0.tbid;
-            }
-
-            flags = FIELD_DP32(flags, TBFLAG_A64, TBII, tbii);
-            flags = FIELD_DP32(flags, TBFLAG_A64, TBID, tbid);
-        }
-
-        if (cpu_isar_feature(aa64_sve, cpu)) {
-            int sve_el = sve_exception_el(env, current_el);
-            uint32_t zcr_len;
-
-            /* If SVE is disabled, but FP is enabled,
-             * then the effective len is 0.
-             */
-            if (sve_el != 0 && fp_el == 0) {
-                zcr_len = 0;
-            } else {
-                zcr_len = sve_zcr_len_for_el(env, current_el);
-            }
-            flags = FIELD_DP32(flags, TBFLAG_A64, SVEEXC_EL, sve_el);
-            flags = FIELD_DP32(flags, TBFLAG_A64, ZCR_LEN, zcr_len);
-        }
-
-        sctlr = arm_sctlr(env, current_el);
-
-        if (cpu_isar_feature(aa64_pauth, cpu)) {
-            /*
-             * In order to save space in flags, we record only whether
-             * pauth is "inactive", meaning all insns are implemented as
-             * a nop, or "active" when some action must be performed.
-             * The decision of which action to take is left to a helper.
-             */
-            if (sctlr & (SCTLR_EnIA | SCTLR_EnIB | SCTLR_EnDA | SCTLR_EnDB)) {
-                flags = FIELD_DP32(flags, TBFLAG_A64, PAUTH_ACTIVE, 1);
-            }
-        }
-
-        if (cpu_isar_feature(aa64_bti, cpu)) {
-            /* Note that SCTLR_EL[23].BT == SCTLR_BT1.  */
-            if (sctlr & (current_el == 0 ? SCTLR_BT0 : SCTLR_BT1)) {
-                flags = FIELD_DP32(flags, TBFLAG_A64, BT, 1);
-            }
-            flags = FIELD_DP32(flags, TBFLAG_A64, BTYPE, env->btype);
-        }
-    } else {
-        *pc = env->regs[15];
-        flags = FIELD_DP32(flags, TBFLAG_A32, THUMB, env->thumb);
-        flags = FIELD_DP32(flags, TBFLAG_A32, VECLEN, env->vfp.vec_len);
-        flags = FIELD_DP32(flags, TBFLAG_A32, VECSTRIDE, env->vfp.vec_stride);
-        flags = FIELD_DP32(flags, TBFLAG_A32, CONDEXEC, env->condexec_bits);
-        flags = FIELD_DP32(flags, TBFLAG_A32, SCTLR_B, arm_sctlr_b(env));
-        flags = FIELD_DP32(flags, TBFLAG_A32, NS, !access_secure_reg(env));
-        if (env->vfp.xregs[ARM_VFP_FPEXC] & (1 << 30)
-            || arm_el_is_aa64(env, 1) || arm_feature(env, ARM_FEATURE_M)) {
-            flags = FIELD_DP32(flags, TBFLAG_A32, VFPEN, 1);
-        }
-        /* Note that XSCALE_CPAR shares bits with VECSTRIDE */
-        if (arm_feature(env, ARM_FEATURE_XSCALE)) {
-            flags = FIELD_DP32(flags, TBFLAG_A32,
-                               XSCALE_CPAR, env->cp15.c15_cpar);
-        }
-    }
-
-    flags = FIELD_DP32(flags, TBFLAG_ANY, MMUIDX, arm_to_core_mmu_idx(mmu_idx));
-
-    /* The SS_ACTIVE and PSTATE_SS bits correspond to the state machine
-     * states defined in the ARM ARM for software singlestep:
-     *  SS_ACTIVE   PSTATE.SS   State
-     *     0            x       Inactive (the TB flag for SS is always 0)
-     *     1            0       Active-pending
-     *     1            1       Active-not-pending
-     */
-    if (arm_singlestep_active(env)) {
-        flags = FIELD_DP32(flags, TBFLAG_ANY, SS_ACTIVE, 1);
-        if (is_a64(env)) {
-            if (env->pstate & PSTATE_SS) {
-                flags = FIELD_DP32(flags, TBFLAG_ANY, PSTATE_SS, 1);
-            }
-        } else {
-            if (env->uncached_cpsr & PSTATE_SS) {
-                flags = FIELD_DP32(flags, TBFLAG_ANY, PSTATE_SS, 1);
-            }
-        }
-    }
+    flags = FIELD_DP32(flags, TBFLAG_ANY, FPEXC_EL, fp_el);
+    flags = FIELD_DP32(flags, TBFLAG_ANY, MMUIDX,
+                       arm_to_core_mmu_idx(mmu_idx));
     if (arm_cpu_data_is_big_endian(env)) {
         flags = FIELD_DP32(flags, TBFLAG_ANY, BE_DATA, 1);
     }
-    flags = FIELD_DP32(flags, TBFLAG_ANY, FPEXC_EL, fp_el);
+    if (arm_singlestep_active(env)) {
+        flags = FIELD_DP32(flags, TBFLAG_ANY, SS_ACTIVE, 1);
+    }
+    return flags;
+}
+
+uint32_t rebuild_hflags_a32(CPUARMState *env, int el)
+{
+    uint32_t flags = 0;
+    ARMMMUIdx mmu_idx;
+    int fp_el;
+
+    flags = FIELD_DP32(flags, TBFLAG_A32, SCTLR_B, arm_sctlr_b(env));
+    flags = FIELD_DP32(flags, TBFLAG_A32, NS, !access_secure_reg(env));
 
     if (arm_v7m_is_handler_mode(env)) {
         flags = FIELD_DP32(flags, TBFLAG_A32, HANDLER, 1);
     }
 
-    /* v8M always applies stack limit checks unless CCR.STKOFHFNMIGN is
-     * suppressing them because the requested execution priority is less than 0.
+    mmu_idx = arm_mmu_idx(env);
+
+    /*
+     * v8M always applies stack limit checks unless CCR.STKOFHFNMIGN
+     * is suppressing them because the requested execution priority
+     * is less than 0.
      */
     if (arm_feature(env, ARM_FEATURE_V8) &&
         arm_feature(env, ARM_FEATURE_M) &&
-        !((mmu_idx  & ARM_MMU_IDX_M_NEGPRI) &&
+        !((mmu_idx & ARM_MMU_IDX_M_NEGPRI) &&
           (env->v7m.ccr[env->v7m.secure] & R_V7M_CCR_STKOFHFNMIGN_MASK))) {
         flags = FIELD_DP32(flags, TBFLAG_A32, STACKCHECK, 1);
     }
 
-    if (arm_feature(env, ARM_FEATURE_M_SECURITY) &&
-        FIELD_EX32(env->v7m.fpccr[M_REG_S], V7M_FPCCR, S) != env->v7m.secure) {
-        flags = FIELD_DP32(flags, TBFLAG_A32, FPCCR_S_WRONG, 1);
+    fp_el = fp_exception_el(env, el);
+    return common_hflags(env, el, mmu_idx, fp_el, flags);
+}
+
+uint32_t rebuild_hflags_a64(CPUARMState *env, int el)
+{
+    ARMCPU *cpu = env_archcpu(env);
+    ARMMMUIdx mmu_idx = arm_mmu_idx(env);
+    ARMMMUIdx stage1 = stage_1_mmu_idx(mmu_idx);
+    ARMVAParameters p0 = aa64_va_parameters_both(env, 0, stage1);
+    int fp_el = fp_exception_el(env, el);
+    uint32_t flags = 0;
+    uint64_t sctlr;
+    int tbii, tbid;
+
+    flags = FIELD_DP32(flags, TBFLAG_ANY, AARCH64_STATE, 1);
+
+    /* Get control bits for tagged addresses.  */
+    /* FIXME: ARMv8.1-VHE S2 translation regime.  */
+    if (regime_el(env, stage1) < 2) {
+        ARMVAParameters p1 = aa64_va_parameters_both(env, -1, stage1);
+        tbid = (p1.tbi << 1) | p0.tbi;
+        tbii = tbid & ~((p1.tbid << 1) | p0.tbid);
+    } else {
+        tbid = p0.tbi;
+        tbii = tbid & !p0.tbid;
     }
 
-    if (arm_feature(env, ARM_FEATURE_M) &&
-        (env->v7m.fpccr[env->v7m.secure] & R_V7M_FPCCR_ASPEN_MASK) &&
-        (!(env->v7m.control[M_REG_S] & R_V7M_CONTROL_FPCA_MASK) ||
-         (env->v7m.secure &&
-          !(env->v7m.control[M_REG_S] & R_V7M_CONTROL_SFPA_MASK)))) {
-        /*
-         * ASPEN is set, but FPCA/SFPA indicate that there is no active
-         * FP context; we must create a new FP context before executing
-         * any FP insn.
+    flags = FIELD_DP32(flags, TBFLAG_A64, TBII, tbii);
+    flags = FIELD_DP32(flags, TBFLAG_A64, TBID, tbid);
+
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        int sve_el = sve_exception_el(env, el);
+        uint32_t zcr_len;
+
+        /* If SVE is disabled, but FP is enabled,
+         * then the effective len is 0.
          */
-        flags = FIELD_DP32(flags, TBFLAG_A32, NEW_FP_CTXT_NEEDED, 1);
+        if (sve_el != 0 && fp_el == 0) {
+            zcr_len = 0;
+        } else {
+            zcr_len = sve_zcr_len_for_el(env, el);
+        }
+        flags = FIELD_DP32(flags, TBFLAG_A64, SVEEXC_EL, sve_el);
+        flags = FIELD_DP32(flags, TBFLAG_A64, ZCR_LEN, zcr_len);
     }
 
-    if (arm_feature(env, ARM_FEATURE_M)) {
-        bool is_secure = env->v7m.fpccr[M_REG_S] & R_V7M_FPCCR_S_MASK;
-
-        if (env->v7m.fpccr[is_secure] & R_V7M_FPCCR_LSPACT_MASK) {
-            flags = FIELD_DP32(flags, TBFLAG_A32, LSPACT, 1);
+    if (el == 0) {
+        /* FIXME: ARMv8.1-VHE S2 translation regime.  */
+        sctlr = env->cp15.sctlr_el[1];
+    } else {
+        sctlr = env->cp15.sctlr_el[el];
+    }
+    if (cpu_isar_feature(aa64_pauth, cpu)) {
+        /*
+         * In order to save space in flags, we record only whether
+         * pauth is "inactive", meaning all insns are implemented as
+         * a nop, or "active" when some action must be performed.
+         * The decision of which action to take is left to a helper.
+         */
+        if (sctlr & (SCTLR_EnIA | SCTLR_EnIB | SCTLR_EnDA | SCTLR_EnDB)) {
+            flags = FIELD_DP32(flags, TBFLAG_A64, PAUTH_ACTIVE, 1);
         }
     }
 
-    *pflags = flags;
+    if (cpu_isar_feature(aa64_bti, cpu)) {
+        /* Note that SCTLR_EL[23].BT == SCTLR_BT1.  */
+        if (sctlr & (el == 0 ? SCTLR_BT0 : SCTLR_BT1)) {
+            flags = FIELD_DP32(flags, TBFLAG_A64, BT, 1);
+        }
+    }
+
+    return common_hflags(env, el, mmu_idx, fp_el, flags);
+}
+
+void arm_rebuild_hflags(CPUARMState *env)
+{
+    int el = arm_current_el(env);
+    env->hflags = (is_a64(env)
+                   ? rebuild_hflags_a64(env, el)
+                   : rebuild_hflags_a32(env, el));
+}
+
+void HELPER(rebuild_hflags_a32)(CPUARMState *env, uint32_t el)
+{
+    tcg_debug_assert(!is_a64(env));
+    env->hflags = rebuild_hflags_a32(env, el);
+}
+
+void HELPER(rebuild_hflags_a64)(CPUARMState *env, uint32_t el)
+{
+    tcg_debug_assert(is_a64(env));
+    env->hflags = rebuild_hflags_a64(env, el);
+}
+
+void cpu_get_tb_cpu_state(CPUARMState *env, target_ulong *pc,
+                          target_ulong *cs_base, uint32_t *pflags)
+{
+    int current_el = arm_current_el(env);
+    uint32_t flags;
+    uint32_t pstate_for_ss;
+
     *cs_base = 0;
+    if (is_a64(env)) {
+        *pc = env->pc;
+        flags = rebuild_hflags_a64(env, current_el);
+        flags = FIELD_DP32(flags, TBFLAG_A64, BTYPE, env->btype);
+        pstate_for_ss = env->pstate;
+    } else {
+        *pc = env->regs[15];
+        flags = rebuild_hflags_a32(env, current_el);
+        flags = FIELD_DP32(flags, TBFLAG_A32, THUMB, env->thumb);
+        flags = FIELD_DP32(flags, TBFLAG_A32, CONDEXEC, env->condexec_bits);
+        /* Note that XSCALE_CPAR shares bits with VECSTRIDE */
+        if (arm_feature(env, ARM_FEATURE_XSCALE)) {
+            flags = FIELD_DP32(flags, TBFLAG_A32, XSCALE_CPAR,
+                               env->cp15.c15_cpar);
+        } else {
+            flags = FIELD_DP32(flags, TBFLAG_A32, VECLEN, env->vfp.vec_len);
+            flags = FIELD_DP32(flags, TBFLAG_A32, VECSTRIDE,
+                               env->vfp.vec_stride);
+        }
+        if (env->vfp.xregs[ARM_VFP_FPEXC] & (1 << 30)
+            || arm_el_is_aa64(env, 1) || arm_feature(env, ARM_FEATURE_M)) {
+            flags = FIELD_DP32(flags, TBFLAG_A32, VFPEN, 1);
+        }
+
+        /* TODO: Perhaps cache these bits too?  */
+        if (arm_feature(env, ARM_FEATURE_M)) {
+            if (arm_feature(env, ARM_FEATURE_M_SECURITY) &&
+                FIELD_EX32(env->v7m.fpccr[M_REG_S], V7M_FPCCR, S)
+                != env->v7m.secure) {
+                flags = FIELD_DP32(flags, TBFLAG_A32, FPCCR_S_WRONG, 1);
+            }
+
+            if ((env->v7m.fpccr[env->v7m.secure] & R_V7M_FPCCR_ASPEN_MASK) &&
+                (!(env->v7m.control[M_REG_S] & R_V7M_CONTROL_FPCA_MASK) ||
+                 (env->v7m.secure &&
+                  !(env->v7m.control[M_REG_S] & R_V7M_CONTROL_SFPA_MASK)))) {
+                /*
+                 * ASPEN is set, but FPCA/SFPA indicate that there is no
+                 * active FP context; we must create a new FP context
+                 * before executing any FP insn.
+                 */
+                flags = FIELD_DP32(flags, TBFLAG_A32, NEW_FP_CTXT_NEEDED, 1);
+            }
+
+            bool is_secure = env->v7m.fpccr[M_REG_S] & R_V7M_FPCCR_S_MASK;
+            if (env->v7m.fpccr[is_secure] & R_V7M_FPCCR_LSPACT_MASK) {
+                flags = FIELD_DP32(flags, TBFLAG_A32, LSPACT, 1);
+            }
+        }
+
+        pstate_for_ss = env->uncached_cpsr;
+    }
+
+    /*
+     * The SS_ACTIVE and PSTATE_SS bits correspond to the state machine
+     * states defined in the ARM ARM for software singlestep:
+     *  SS_ACTIVE   PSTATE.SS   State
+     *     0            x       Inactive (the TB flag for SS is always 0)
+     *     1            0       Active-pending
+     *     1            1       Active-not-pending
+     * SS_ACTIVE is set in hflags; PSTATE_SS is computed every TB.
+     */
+    if (FIELD_EX32(flags, TBFLAG_ANY, SS_ACTIVE)
+        && (pstate_for_ss & PSTATE_SS)) {
+        flags = FIELD_DP32(flags, TBFLAG_ANY, PSTATE_SS, 1);
+    }
+
+    *pflags = flags;
 }
 
 #ifdef TARGET_AARCH64
-- 
2.17.1

[Qemu-devel] [PATCH v4 2/3] target/arm: Rebuild hflags at EL changes and MSR writes

[Richard Henderson]

Now setting, but not relying upon, env->hflags.

Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
v2: Fixed partial conversion to assignment to env->hflags.
---
 linux-user/syscall.c       |  1 +
 target/arm/cpu.c           |  1 +
 target/arm/helper-a64.c    |  3 +++
 target/arm/helper.c        |  2 ++
 target/arm/machine.c       |  1 +
 target/arm/op_helper.c     |  1 +
 target/arm/translate-a64.c |  6 +++++-
 target/arm/translate.c     | 14 ++++++++++++--
 8 files changed, 26 insertions(+), 3 deletions(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 8367cb138d..55d5fdadf3 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -9979,6 +9979,7 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
                     aarch64_sve_narrow_vq(env, vq);
                 }
                 env->vfp.zcr_el[1] = vq - 1;
+                arm_rebuild_hflags(env);
                 ret = vq * 16;
             }
             return ret;
diff --git a/target/arm/cpu.c b/target/arm/cpu.c
index ec2ab95dbe..995f4ea355 100644
--- a/target/arm/cpu.c
+++ b/target/arm/cpu.c
@@ -406,6 +406,7 @@ static void arm_cpu_reset(CPUState *s)
 
     hw_breakpoint_update_all(cpu);
     hw_watchpoint_update_all(cpu);
+    arm_rebuild_hflags(env);
 }
 
 bool arm_cpu_exec_interrupt(CPUState *cs, int interrupt_request)
diff --git a/target/arm/helper-a64.c b/target/arm/helper-a64.c
index 060699b901..3bc364ebb7 100644
--- a/target/arm/helper-a64.c
+++ b/target/arm/helper-a64.c
@@ -1025,6 +1025,7 @@ void HELPER(exception_return)(CPUARMState *env, uint64_t new_pc)
         } else {
             env->regs[15] = new_pc & ~0x3;
         }
+        env->hflags = rebuild_hflags_a32(env, new_el);
         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
                       "AArch32 EL%d PC 0x%" PRIx32 "\n",
                       cur_el, new_el, env->regs[15]);
@@ -1036,10 +1037,12 @@ void HELPER(exception_return)(CPUARMState *env, uint64_t new_pc)
         }
         aarch64_restore_sp(env, new_el);
         env->pc = new_pc;
+        env->hflags = rebuild_hflags_a64(env, new_el);
         qemu_log_mask(CPU_LOG_INT, "Exception return from AArch64 EL%d to "
                       "AArch64 EL%d PC 0x%" PRIx64 "\n",
                       cur_el, new_el, env->pc);
     }
+
     /*
      * Note that cur_el can never be 0.  If new_el is 0, then
      * el0_a64 is return_to_aa64, else el0_a64 is ignored.
diff --git a/target/arm/helper.c b/target/arm/helper.c
index 43b7c41f11..9b07350cfe 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -7905,6 +7905,7 @@ static void take_aarch32_exception(CPUARMState *env, int new_mode,
         env->regs[14] = env->regs[15] + offset;
     }
     env->regs[15] = newpc;
+    env->hflags = rebuild_hflags_a32(env, arm_current_el(env));
 }
 
 static void arm_cpu_do_interrupt_aarch32_hyp(CPUState *cs)
@@ -8251,6 +8252,7 @@ static void arm_cpu_do_interrupt_aarch64(CPUState *cs)
 
     pstate_write(env, PSTATE_DAIF | new_mode);
     env->aarch64 = 1;
+    env->hflags = rebuild_hflags_a64(env, new_el);
     aarch64_restore_sp(env, new_el);
 
     env->pc = addr;
diff --git a/target/arm/machine.c b/target/arm/machine.c
index 3fd319a309..838d154a3c 100644
--- a/target/arm/machine.c
+++ b/target/arm/machine.c
@@ -758,6 +758,7 @@ static int cpu_post_load(void *opaque, int version_id)
     if (!kvm_enabled()) {
         pmu_op_finish(&cpu->env);
     }
+    arm_rebuild_hflags(&cpu->env);
 
     return 0;
 }
diff --git a/target/arm/op_helper.c b/target/arm/op_helper.c
index 5e1625a1c8..6e6613b8a8 100644
--- a/target/arm/op_helper.c
+++ b/target/arm/op_helper.c
@@ -420,6 +420,7 @@ void HELPER(cpsr_write_eret)(CPUARMState *env, uint32_t val)
      */
     env->regs[15] &= (env->thumb ? ~1 : ~3);
 
+    env->hflags = rebuild_hflags_a32(env, arm_current_el(env));
     qemu_mutex_lock_iothread();
     arm_call_el_change_hook(env_archcpu(env));
     qemu_mutex_unlock_iothread();
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index d3231477a2..f8b5debf82 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -1799,11 +1799,15 @@ static void handle_sys(DisasContext *s, uint32_t insn, bool isread,
         /* I/O operations must end the TB here (whether read or write) */
         gen_io_end();
         s->base.is_jmp = DISAS_UPDATE;
-    } else if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
+    }
+    if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
         /* We default to ending the TB on a coprocessor register write,
          * but allow this to be suppressed by the register definition
          * (usually only necessary to work around guest bugs).
          */
+        TCGv_i32 tcg_el = tcg_const_i32(s->current_el);
+        gen_helper_rebuild_hflags_a64(cpu_env, tcg_el);
+        tcg_temp_free_i32(tcg_el);
         s->base.is_jmp = DISAS_UPDATE;
     }
 }
diff --git a/target/arm/translate.c b/target/arm/translate.c
index 7853462b21..cd14bbebf1 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -7164,6 +7164,8 @@ static int disas_coproc_insn(DisasContext *s, uint32_t insn)
     ri = get_arm_cp_reginfo(s->cp_regs,
             ENCODE_CP_REG(cpnum, is64, s->ns, crn, crm, opc1, opc2));
     if (ri) {
+        bool need_exit_tb;
+
         /* Check access permissions */
         if (!cp_access_ok(s->current_el, ri, isread)) {
             return 1;
@@ -7336,15 +7338,23 @@ static int disas_coproc_insn(DisasContext *s, uint32_t insn)
             }
         }
 
+        need_exit_tb = false;
         if ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
             /* I/O operations must end the TB here (whether read or write) */
             gen_io_end();
-            gen_lookup_tb(s);
-        } else if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
+            need_exit_tb = true;
+        }
+        if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
             /* We default to ending the TB on a coprocessor register write,
              * but allow this to be suppressed by the register definition
              * (usually only necessary to work around guest bugs).
              */
+            TCGv_i32 tcg_el = tcg_const_i32(s->current_el);
+            gen_helper_rebuild_hflags_a32(cpu_env, tcg_el);
+            tcg_temp_free_i32(tcg_el);
+            need_exit_tb = true;
+        }
+        if (need_exit_tb) {
             gen_lookup_tb(s);
         }
 
-- 
2.17.1

[Qemu-devel] [PATCH v4 3/3] target/arm: Rely on hflags correct in cpu_get_tb_cpu_state

[Richard Henderson]

This is the payoff.

From perf record -g data of ubuntu 18 boot and shutdown:

BEFORE:

-   23.02%     2.82%  qemu-system-aar  [.] helper_lookup_tb_ptr
   - 20.22% helper_lookup_tb_ptr
      + 10.05% tb_htable_lookup
      - 9.13% cpu_get_tb_cpu_state
           3.20% aa64_va_parameters_both
           0.55% fp_exception_el

-   11.66%     4.74%  qemu-system-aar  [.] cpu_get_tb_cpu_state
   - 6.96% cpu_get_tb_cpu_state
        3.63% aa64_va_parameters_both
        0.60% fp_exception_el
        0.53% sve_exception_el

AFTER:

-   16.40%     3.40%  qemu-system-aar  [.] helper_lookup_tb_ptr
   - 13.03% helper_lookup_tb_ptr
      + 11.19% tb_htable_lookup
        0.55% cpu_get_tb_cpu_state

     0.98%     0.71%  qemu-system-aar  [.] cpu_get_tb_cpu_state

     0.87%     0.24%  qemu-system-aar  [.] rebuild_hflags_a64

Before, helper_lookup_tb_ptr is the second hottest function in the
application, consuming almost a quarter of the runtime.  Within the
entire execution, cpu_get_tb_cpu_state consumes about 12%.

After, helper_lookup_tb_ptr has dropped to the fourth hottest function,
with consumption dropping to a sixth of the runtime.  Within the
entire execution, cpu_get_tb_cpu_state has dropped below 1%, and the
supporting function to rebuild hflags also consumes about 1%.

Assertions are retained for --enable-debug-tcg.

Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
v2: Retain asserts for future debugging.
---
 target/arm/helper.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/target/arm/helper.c b/target/arm/helper.c
index 9b07350cfe..abb8cc52dd 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -11155,19 +11155,29 @@ void HELPER(rebuild_hflags_a64)(CPUARMState *env, uint32_t el)
 void cpu_get_tb_cpu_state(CPUARMState *env, target_ulong *pc,
                           target_ulong *cs_base, uint32_t *pflags)
 {
-    int current_el = arm_current_el(env);
-    uint32_t flags;
+    uint32_t flags = env->hflags;
     uint32_t pstate_for_ss;
 
+#ifdef CONFIG_DEBUG_TCG
+    {
+        int el = arm_current_el(env);
+        uint32_t check_flags;
+        if (is_a64(env)) {
+            check_flags = rebuild_hflags_a64(env, el);
+        } else {
+            check_flags = rebuild_hflags_a32(env, el);
+        }
+        assert(flags == check_flags);
+    }
+#endif
+
     *cs_base = 0;
-    if (is_a64(env)) {
+    if (FIELD_EX32(flags, TBFLAG_ANY, AARCH64_STATE)) {
         *pc = env->pc;
-        flags = rebuild_hflags_a64(env, current_el);
         flags = FIELD_DP32(flags, TBFLAG_A64, BTYPE, env->btype);
         pstate_for_ss = env->pstate;
     } else {
         *pc = env->regs[15];
-        flags = rebuild_hflags_a32(env, current_el);
         flags = FIELD_DP32(flags, TBFLAG_A32, THUMB, env->thumb);
         flags = FIELD_DP32(flags, TBFLAG_A32, CONDEXEC, env->condexec_bits);
         /* Note that XSCALE_CPAR shares bits with VECSTRIDE */
-- 
2.17.1