[v2,2/2] target/riscv: Legalize MPP value in write_mstatus

Commit Message

Weiwei Li April 6, 2023, 7:25 a.m. UTC

mstatus.MPP field is a WARL field since priv version 1.11, so we
remain it unchanged if an invalid value is written into it. And
after this, RVH shouldn't be passed to riscv_cpu_set_mode().

Signed-off-by: Weiwei Li <[e-mail]>
Signed-off-by: Junqiang Wang <[e-mail]>
---
 target/riscv/cpu_helper.c |  5 +----
 target/riscv/csr.c        | 18 ++++++++++++++++++
 2 files changed, 19 insertions(+), 4 deletions(-)

Comments

Richard Henderson April 6, 2023, 7:28 p.m. UTC | #1

On 4/6/23 00:25, Weiwei Li wrote:
>   void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv)
>   {
> -    if (newpriv > PRV_M) {
> +    if (newpriv > PRV_M || newpriv == PRV_H) {
>           g_assert_not_reached();
>       }

Nit: if (test) { assert_not_reached } -> assert(!test).

which emits a more helpful error message before abort.


r~

Richard Henderson April 6, 2023, 7:33 p.m. UTC | #2

On 4/6/23 00:25, Weiwei Li wrote:
> +static target_ulong legalize_mpp(CPURISCVState *env, target_ulong old_mpp,
> +                                 target_ulong val)
> +{
> +    target_ulong new_mpp = get_field(val, MSTATUS_MPP);
> +    bool mpp_invalid = (new_mpp == PRV_S && !riscv_has_ext(env, RVS)) ||
> +                       (new_mpp == PRV_U && !riscv_has_ext(env, RVU)) ||
> +                       (new_mpp == PRV_H);
> +
> +    /* Remain field unchanged if new_mpp value is invalid */
> +    return mpp_invalid ? set_field(val, MSTATUS_MPP, old_mpp) : val;
> +}

Does anyone find PRV_H confusing, since that's not what it is?
I think it would be nice to remove it entirely.

This function might be better as

     bool valid = false;

     switch (new_mpp) {
     case PRV_M:
         valid = true;
         break;
     case PRV_S:
         valid = riscv_has_ext(env, RVS);
         break;
     case PRV_U:
         valid = riscv_has_ext(env, RVU);
         break;
     }
     if (!valid) {
         val = set_field(...);
     }
     return val;


r~

Weiwei Li April 7, 2023, 1:16 a.m. UTC | #3

On 2023/4/7 03:28, Richard Henderson wrote:
> On 4/6/23 00:25, Weiwei Li wrote:
>>   void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv)
>>   {
>> -    if (newpriv > PRV_M) {
>> +    if (newpriv > PRV_M || newpriv == PRV_H) {
>>           g_assert_not_reached();
>>       }
>
> Nit: if (test) { assert_not_reached } -> assert(!test).
>
> which emits a more helpful error message before abort.
>
OK. I'll update this.

Regards,

Weiwei Li

>
> r~

Weiwei Li April 7, 2023, 1:24 a.m. UTC | #4

On 2023/4/7 03:33, Richard Henderson wrote:
> On 4/6/23 00:25, Weiwei Li wrote:
>> +static target_ulong legalize_mpp(CPURISCVState *env, target_ulong 
>> old_mpp,
>> +                                 target_ulong val)
>> +{
>> +    target_ulong new_mpp = get_field(val, MSTATUS_MPP);
>> +    bool mpp_invalid = (new_mpp == PRV_S && !riscv_has_ext(env, 
>> RVS)) ||
>> +                       (new_mpp == PRV_U && !riscv_has_ext(env, 
>> RVU)) ||
>> +                       (new_mpp == PRV_H);
>> +
>> +    /* Remain field unchanged if new_mpp value is invalid */
>> +    return mpp_invalid ? set_field(val, MSTATUS_MPP, old_mpp) : val;
>> +}
>
> Does anyone find PRV_H confusing, since that's not what it is?
> I think it would be nice to remove it entirely.
I agree. Maybe PRV_RESERVED is more readable in some cases.
>
> This function might be better as
>
>     bool valid = false;
>
>     switch (new_mpp) {
>     case PRV_M:
>         valid = true;
>         break;
>     case PRV_S:
>         valid = riscv_has_ext(env, RVS);
>         break;
>     case PRV_U:
>         valid = riscv_has_ext(env, RVU);
>         break;
>     }
>     if (!valid) {
>         val = set_field(...);
>     }
>     return val;
>
>
OK. This is more readable. I'll update this.

Regards,

Weiwei Li

> r~

Patch

diff --git a/target/riscv/cpu_helper.c b/target/riscv/cpu_helper.c
index 2310c7905f..6148f221c3 100644
--- a/target/riscv/cpu_helper.c
+++ b/target/riscv/cpu_helper.c
@@ -647,12 +647,9 @@  void riscv_cpu_set_aia_ireg_rmw_fn(CPURISCVState *env, uint32_t priv,
 
 void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv)
 {
-    if (newpriv > PRV_M) {
+    if (newpriv > PRV_M || newpriv == PRV_H) {
         g_assert_not_reached();
     }
-    if (newpriv == PRV_H) {
-        newpriv = PRV_U;
-    }
     if (icount_enabled() && newpriv != env->priv) {
         riscv_itrigger_update_priv(env);
     }
diff --git a/target/riscv/csr.c b/target/riscv/csr.c
index e0b871f6dc..f3ae726853 100644
--- a/target/riscv/csr.c
+++ b/target/riscv/csr.c
@@ -1230,6 +1230,18 @@  static bool validate_vm(CPURISCVState *env, target_ulong vm)
            satp_mode_max_from_map(riscv_cpu_cfg(env)->satp_mode.map);
 }
 
+static target_ulong legalize_mpp(CPURISCVState *env, target_ulong old_mpp,
+                                 target_ulong val)
+{
+    target_ulong new_mpp = get_field(val, MSTATUS_MPP);
+    bool mpp_invalid = (new_mpp == PRV_S && !riscv_has_ext(env, RVS)) ||
+                       (new_mpp == PRV_U && !riscv_has_ext(env, RVU)) ||
+                       (new_mpp == PRV_H);
+
+    /* Remain field unchanged if new_mpp value is invalid */
+    return mpp_invalid ? set_field(val, MSTATUS_MPP, old_mpp) : val;
+}
+
 static RISCVException write_mstatus(CPURISCVState *env, int csrno,
                                     target_ulong val)
 {
@@ -1237,6 +1249,12 @@  static RISCVException write_mstatus(CPURISCVState *env, int csrno,
     uint64_t mask = 0;
     RISCVMXL xl = riscv_cpu_mxl(env);
 
+    /*
+     * MPP field have been made WARL since priv version 1.11. However,
+     * legalization for it will not break any software running on 1.10.
+     */
+    val = legalize_mpp(env, get_field(mstatus, MSTATUS_MPP), val);
+
     /* flush tlb on mstatus fields that affect VM */
     if ((val ^ mstatus) & (MSTATUS_MXR | MSTATUS_MPP | MSTATUS_MPV |
             MSTATUS_MPRV | MSTATUS_SUM)) {