diff mbox series

[v3,56/60] target/arm: Enable FEAT_CSV2_2 for -cpu max

Message ID 20220417174426.711829-57-richard.henderson@linaro.org
State Superseded
Headers show
Series target/arm: Cleanups, new features, new cpus | expand

Commit Message

Richard Henderson April 17, 2022, 5:44 p.m. UTC
There is no branch prediction in TCG, therefore there is no
need to actually include the context number into the predictor.
Therefore all we need to do is add the state for SCXTNUM_ELx.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
v2: Update emulation.rst; clear CSV2_FRAC; use decimal; tidy access_scxtnum.
v3: Rely on EL3-no-EL2 squashing during registration.
---
 docs/system/arm/emulation.rst |  3 ++
 target/arm/cpu.h              | 16 +++++++++
 target/arm/cpu64.c            |  3 +-
 target/arm/helper.c           | 66 ++++++++++++++++++++++++++++++++++-
 4 files changed, 86 insertions(+), 2 deletions(-)

Comments

Damien Hedde April 29, 2022, 9:52 a.m. UTC | #1
On 4/17/22 19:44, Richard Henderson wrote:
> There is no branch prediction in TCG, therefore there is no
> need to actually include the context number into the predictor.
> Therefore all we need to do is add the state for SCXTNUM_ELx. >
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
> v2: Update emulation.rst; clear CSV2_FRAC; use decimal; tidy access_scxtnum.
> v3: Rely on EL3-no-EL2 squashing during registration.
> ---
>   docs/system/arm/emulation.rst |  3 ++
>   target/arm/cpu.h              | 16 +++++++++
>   target/arm/cpu64.c            |  3 +-
>   target/arm/helper.c           | 66 ++++++++++++++++++++++++++++++++++-
>   4 files changed, 86 insertions(+), 2 deletions(-)
> 
> diff --git a/target/arm/helper.c b/target/arm/helper.c
> @@ -7233,7 +7243,57 @@ static const ARMCPRegInfo mte_el0_cacheop_reginfo[] = {
>       },
>   };

Hi Richard,

I tried to compare with the pseudocode from arm doc and I've a few 
interrogations. It seems to me there are missing cases in the access 
checks, but I lack the background to know if these are not checked 
somewhere else or guaranteed to never happen.

>   
> -#endif > +static CPAccessResult access_scxtnum(CPUARMState *env, const 
ARMCPRegInfo *ri,
> +                                     bool isread)
> +{
The following checks are missing:
    + for HFG[W/R]TR_EL2.SCXTNUM_EL0/1
    + HCR_EL2.<NV2,NV1,NV> when accessing SCXTNUM_EL1, but maybe these 
are always guaranteed to fail because we don't support the features ?
    + HCR_EL2.NV when accessing SCXTNUM_EL2
> +    int el = arm_current_el(env);
> +
> +    if (el == 0) {
> +        uint64_t hcr = arm_hcr_el2_eff(env);
> +        if ((hcr & (HCR_TGE | HCR_E2H)) != (HCR_TGE | HCR_E2H)) {
> +            if (env->cp15.sctlr_el[1] & SCTLR_TSCXT) {
> +                if (hcr & HCR_TGE) {
> +                    return CP_ACCESS_TRAP_EL2;
> +                }
> +                return CP_ACCESS_TRAP;
> +            }
> +            if (arm_is_el2_enabled(env) && !(hcr & HCR_ENSCXT)) {
This case is also present when accessing SCXTNUM_EL0 from el1 (but 
without "(hcr & (HCR_TGE | HCR_E2H)) != (HCR_TGE | HCR_E2H)" precondition)
> +                return CP_ACCESS_TRAP_EL2;
> +            }
> +            goto no_sctlr_el2;
> +        }
> +    }
> +    if (el < 2 && (env->cp15.sctlr_el[2] & SCTLR_TSCXT)) {
> +        return CP_ACCESS_TRAP_EL2;
> +    }
> + no_sctlr_el2:
> +    if (el < 3
> +        && arm_feature(env, ARM_FEATURE_EL3)
> +        && !(env->cp15.scr_el3 & SCR_ENSCXT)) {
> +        return CP_ACCESS_TRAP_EL3;
> +    }
> +    return CP_ACCESS_OK;
> +}

Regards,
--
Damien
Richard Henderson April 29, 2022, 6:06 p.m. UTC | #2
On 4/29/22 02:52, Damien Hedde wrote:
> The following checks are missing:
>     + for HFG[W/R]TR_EL2.SCXTNUM_EL0/1

We do not yet support FEAT_FGT.

>     + HCR_EL2.<NV2,NV1,NV> when accessing SCXTNUM_EL1, but maybe these are always 
> guaranteed to fail because we don't support the features ?
>     + HCR_EL2.NV when accessing SCXTNUM_EL2

We do not yet support FEAT_NV or FEAT_NV2.

>> +            if (arm_is_el2_enabled(env) && !(hcr & HCR_ENSCXT)) {
> This case is also present when accessing SCXTNUM_EL0 from el1 (but without "(hcr & 
> (HCR_TGE | HCR_E2H)) != (HCR_TGE | HCR_E2H)" precondition)

Quite right.  Will fix.


r~
diff mbox series

Patch

diff --git a/docs/system/arm/emulation.rst b/docs/system/arm/emulation.rst
index 91fb06c579..aebe3be1ba 100644
--- a/docs/system/arm/emulation.rst
+++ b/docs/system/arm/emulation.rst
@@ -12,6 +12,9 @@  the following architecture extensions:
 - FEAT_BF16 (AArch64 BFloat16 instructions)
 - FEAT_BTI (Branch Target Identification)
 - FEAT_CSV2 (Cache speculation variant 2)
+- FEAT_CSV2_1p1 (Cache speculation variant 2, version 1.1)
+- FEAT_CSV2_1p2 (Cache speculation variant 2, version 1.2)
+- FEAT_CSV2_2 (Cache speculation variant 2, version 2)
 - FEAT_DIT (Data Independent Timing instructions)
 - FEAT_DPB (DC CVAP instruction)
 - FEAT_Debugv8p2 (Debug changes for v8.2)
diff --git a/target/arm/cpu.h b/target/arm/cpu.h
index b90b6d91bd..a7582da7c2 100644
--- a/target/arm/cpu.h
+++ b/target/arm/cpu.h
@@ -687,6 +687,8 @@  typedef struct CPUArchState {
         ARMPACKey apdb;
         ARMPACKey apga;
     } keys;
+
+    uint64_t scxtnum_el[4];
 #endif
 
 #if defined(CONFIG_USER_ONLY)
@@ -1210,6 +1212,7 @@  void pmu_init(ARMCPU *cpu);
 #define SCTLR_WXN     (1U << 19)
 #define SCTLR_ST      (1U << 20) /* up to ??, RAZ in v6 */
 #define SCTLR_UWXN    (1U << 20) /* v7 onward, AArch32 only */
+#define SCTLR_TSCXT   (1U << 20) /* FEAT_CSV2_1p2, AArch64 only */
 #define SCTLR_FI      (1U << 21) /* up to v7, v8 RES0 */
 #define SCTLR_IESB    (1U << 21) /* v8.2-IESB, AArch64 only */
 #define SCTLR_U       (1U << 22) /* up to v6, RAO in v7 */
@@ -4021,6 +4024,19 @@  static inline bool isar_feature_aa64_dit(const ARMISARegisters *id)
     return FIELD_EX64(id->id_aa64pfr0, ID_AA64PFR0, DIT) != 0;
 }
 
+static inline bool isar_feature_aa64_scxtnum(const ARMISARegisters *id)
+{
+    int key = FIELD_EX64(id->id_aa64pfr0, ID_AA64PFR0, CSV2);
+    if (key >= 2) {
+        return true;      /* FEAT_CSV2_2 */
+    }
+    if (key == 1) {
+        key = FIELD_EX64(id->id_aa64pfr1, ID_AA64PFR1, CSV2_FRAC);
+        return key >= 2;  /* FEAT_CSV2_1p2 */
+    }
+    return false;
+}
+
 static inline bool isar_feature_aa64_ssbs(const ARMISARegisters *id)
 {
     return FIELD_EX64(id->id_aa64pfr1, ID_AA64PFR1, SSBS) != 0;
diff --git a/target/arm/cpu64.c b/target/arm/cpu64.c
index b62656f3c3..6ccbcb857d 100644
--- a/target/arm/cpu64.c
+++ b/target/arm/cpu64.c
@@ -748,7 +748,7 @@  static void aarch64_max_initfn(Object *obj)
     t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
     t = FIELD_DP64(t, ID_AA64PFR0, SEL2, 1);      /* FEAT_SEL2 */
     t = FIELD_DP64(t, ID_AA64PFR0, DIT, 1);       /* FEAT_DIT */
-    t = FIELD_DP64(t, ID_AA64PFR0, CSV2, 1);      /* FEAT_CSV2 */
+    t = FIELD_DP64(t, ID_AA64PFR0, CSV2, 2);      /* FEAT_CSV2_2 */
     cpu->isar.id_aa64pfr0 = t;
 
     t = cpu->isar.id_aa64pfr1;
@@ -760,6 +760,7 @@  static void aarch64_max_initfn(Object *obj)
      * we do for EL2 with the virtualization=on property.
      */
     t = FIELD_DP64(t, ID_AA64PFR1, MTE, 3);       /* FEAT_MTE3 */
+    t = FIELD_DP64(t, ID_AA64PFR1, CSV2_FRAC, 0); /* FEAT_CSV2_2 */
     cpu->isar.id_aa64pfr1 = t;
 
     t = cpu->isar.id_aa64mmfr0;
diff --git a/target/arm/helper.c b/target/arm/helper.c
index 7e4178c594..22728b9fbf 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -1771,6 +1771,9 @@  static void scr_write(CPUARMState *env, const ARMCPRegInfo *ri, uint64_t value)
         if (cpu_isar_feature(aa64_mte, cpu)) {
             valid_mask |= SCR_ATA;
         }
+        if (cpu_isar_feature(aa64_scxtnum, cpu)) {
+            valid_mask |= SCR_ENSCXT;
+        }
     } else {
         valid_mask &= ~(SCR_RW | SCR_ST);
         if (cpu_isar_feature(aa32_ras, cpu)) {
@@ -5149,6 +5152,9 @@  static void do_hcr_write(CPUARMState *env, uint64_t value, uint64_t valid_mask)
         if (cpu_isar_feature(aa64_mte, cpu)) {
             valid_mask |= HCR_ATA | HCR_DCT | HCR_TID5;
         }
+        if (cpu_isar_feature(aa64_scxtnum, cpu)) {
+            valid_mask |= HCR_ENSCXT;
+        }
     }
 
     /* Clear RES0 bits.  */
@@ -5798,6 +5804,10 @@  static void define_arm_vh_e2h_redirects_aliases(ARMCPU *cpu)
         { K(3, 0,  5, 6, 0), K(3, 4,  5, 6, 0), K(3, 5, 5, 6, 0),
           "TFSR_EL1", "TFSR_EL2", "TFSR_EL12", isar_feature_aa64_mte },
 
+        { K(3, 0, 13, 0, 7), K(3, 4, 13, 0, 7), K(3, 5, 13, 0, 7),
+          "SCXTNUM_EL1", "SCXTNUM_EL2", "SCXTNUM_EL12",
+          isar_feature_aa64_scxtnum },
+
         /* TODO: ARMv8.2-SPE -- PMSCR_EL2 */
         /* TODO: ARMv8.4-Trace -- TRFCR_EL2 */
     };
@@ -7233,7 +7243,57 @@  static const ARMCPRegInfo mte_el0_cacheop_reginfo[] = {
     },
 };
 
-#endif
+static CPAccessResult access_scxtnum(CPUARMState *env, const ARMCPRegInfo *ri,
+                                     bool isread)
+{
+    int el = arm_current_el(env);
+
+    if (el == 0) {
+        uint64_t hcr = arm_hcr_el2_eff(env);
+        if ((hcr & (HCR_TGE | HCR_E2H)) != (HCR_TGE | HCR_E2H)) {
+            if (env->cp15.sctlr_el[1] & SCTLR_TSCXT) {
+                if (hcr & HCR_TGE) {
+                    return CP_ACCESS_TRAP_EL2;
+                }
+                return CP_ACCESS_TRAP;
+            }
+            if (arm_is_el2_enabled(env) && !(hcr & HCR_ENSCXT)) {
+                return CP_ACCESS_TRAP_EL2;
+            }
+            goto no_sctlr_el2;
+        }
+    }
+    if (el < 2 && (env->cp15.sctlr_el[2] & SCTLR_TSCXT)) {
+        return CP_ACCESS_TRAP_EL2;
+    }
+ no_sctlr_el2:
+    if (el < 3
+        && arm_feature(env, ARM_FEATURE_EL3)
+        && !(env->cp15.scr_el3 & SCR_ENSCXT)) {
+        return CP_ACCESS_TRAP_EL3;
+    }
+    return CP_ACCESS_OK;
+}
+
+static const ARMCPRegInfo scxtnum_reginfo[] = {
+    { .name = "SCXTNUM_EL0", .state = ARM_CP_STATE_AA64,
+      .opc0 = 3, .opc1 = 3, .crn = 13, .crm = 0, .opc2 = 7,
+      .access = PL0_RW, .accessfn = access_scxtnum,
+      .fieldoffset = offsetof(CPUARMState, scxtnum_el[0]) },
+    { .name = "SCXTNUM_EL1", .state = ARM_CP_STATE_AA64,
+      .opc0 = 3, .opc1 = 0, .crn = 13, .crm = 0, .opc2 = 7,
+      .access = PL1_RW, .accessfn = access_scxtnum,
+      .fieldoffset = offsetof(CPUARMState, scxtnum_el[1]) },
+    { .name = "SCXTNUM_EL2", .state = ARM_CP_STATE_AA64,
+      .opc0 = 3, .opc1 = 4, .crn = 13, .crm = 0, .opc2 = 7,
+      .access = PL2_RW, .accessfn = access_scxtnum,
+      .fieldoffset = offsetof(CPUARMState, scxtnum_el[2]) },
+    { .name = "SCXTNUM_EL3", .state = ARM_CP_STATE_AA64,
+      .opc0 = 3, .opc1 = 6, .crn = 13, .crm = 0, .opc2 = 7,
+      .access = PL3_RW,
+      .fieldoffset = offsetof(CPUARMState, scxtnum_el[3]) },
+};
+#endif /* TARGET_AARCH64 */
 
 static CPAccessResult access_predinv(CPUARMState *env, const ARMCPRegInfo *ri,
                                      bool isread)
@@ -8372,6 +8432,10 @@  void register_cp_regs_for_features(ARMCPU *cpu)
         define_arm_cp_regs(cpu, mte_tco_ro_reginfo);
         define_arm_cp_regs(cpu, mte_el0_cacheop_reginfo);
     }
+
+    if (cpu_isar_feature(aa64_scxtnum, cpu)) {
+        define_arm_cp_regs(cpu, scxtnum_reginfo);
+    }
 #endif
 
     if (cpu_isar_feature(any_predinv, cpu)) {