Skip to contentSkip to conversation

target/riscv: Mask the implicitly enabled extensions in isa_string based on priv version

PATCH In review Weiwei Li 2 participants 4 messages Apr 7, 2023 – Apr 12, 2023 RISC-V

lore
Share & tools
lore

Weiwei Li

author
#0 lore raw

PATCH 1/1 target/riscv: Mask the implicitly enabled extensions in isa_string based on priv version

Using implicitly enabled extensions such as Zca/Zcf/Zcd instead of their super extensions can simplify the extension related check. However, they may have higher priv version than their super extensions. So we should mask them in the isa_string based on priv version to make them invisible to user if the specified priv version is lower than their minimal priv version.

---
 target/riscv/cpu.c | 1 +
 1 file changed, 1 insertion(+)
target/riscv/cpu.c +1 -0
index cb68916fce..1a5099382c 100644
@@ -1709,6 +1709,7 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str,
 
     for (i = 0; i < ARRAY_SIZE(isa_edata_arr); i++) {
         if (isa_edata_arr[i].multi_letter &&
+            (cpu->env.priv_ver >= isa_edata_arr[i].min_version) &&
             isa_ext_is_enabled(cpu, &isa_edata_arr[i])) {
             new = g_strconcat(old, "_", isa_edata_arr[i].name, NULL);
             g_free(old);
Signature · 1 line
2.25.1
#1 lore raw
Quoted 17 lines · Weiwei Li

Using implicitly enabled extensions such as Zca/Zcf/Zcd instead of their super extensions can simplify the extension related check. However, they may have higher priv version than their super extensions. So we should mask them in the isa_string based on priv version to make them invisible to user if the specified priv version is lower than their minimal priv version.

---
  target/riscv/cpu.c | 1 +
  1 file changed, 1 insertion(+)
target/riscv/cpu.c +0 -0
index cb68916fce..1a5099382c 100644

target/riscv/cpu.c static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, · line 1712

Quoted 4 lines
target/riscv/cpu.c +1 -0
@@ -1709,6 +1709,7 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str,
 
      for (i = 0; i < ARRAY_SIZE(isa_edata_arr); i++) {
          if (isa_edata_arr[i].multi_letter &&
+            (cpu->env.priv_ver >= isa_edata_arr[i].min_version) &&

We don't have a way of telling whether an extension was enabled by us or by the user. This will end up filtering user extensions from the isa_string.

IMHO, the way the logic is working today, we can't enable Z extensions based on enabled MISA bits alone and disable extensions based on priv_spec at the same time. We would need to check for priv_version when enabling these extensions implicitly during realize(). Another alternative is to not disable any extensions at all based on priv spec - we send a warning about the priv mismatch and that's it.

Thanks,

Daniel

Quoted 3 lines
target/riscv/cpu.c +0 -0
              isa_ext_is_enabled(cpu, &isa_edata_arr[i])) {
              new = g_strconcat(old, "_", isa_edata_arr[i].name, NULL);
              g_free(old);
author
#2 lore raw
Quoted 15 lines · Daniel Henrique Barboza
Quoted 12 lines · Weiwei Li

Using implicitly enabled extensions such as Zca/Zcf/Zcd instead of their super extensions can simplify the extension related check. However, they may have higher priv version than their super extensions. So we should mask them in the isa_string based on priv version to make them invisible to user if the specified priv version is lower than their minimal priv version.

---
  target/riscv/cpu.c | 1 +

  1 file changed, 1 insertion(+)
Quoted 12 lines
Quoted 9 lines
target/riscv/cpu.c +0 -0
index cb68916fce..1a5099382c 100644
@@ -1709,6 +1709,7 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, 
 char **isa_str,
         for (i = 0; i < ARRAY_SIZE(isa_edata_arr); i++) {

          if (isa_edata_arr[i].multi_letter &&
+            (cpu->env.priv_ver >= isa_edata_arr[i].min_version) &&

We don't have a way of telling whether an extension was enabled by us or by the user. This will end up filtering user extensions from the isa_string.

As the method described in the https://lists.gnu.org/archive/html/qemu-riscv/2023-04/msg00200.html:

The user specified extension have been disabled in realize(), So what needs masked is the implied extensions here.

Quoted 4 lines

IMHO, the way the logic is working today, we can't enable Z extensions based on enabled MISA bits alone and disable extensions based on priv_spec at the same time. We would need to check for priv_version when enabling these extensions implicitly during realize().

If we check this in realize(), we also need check them in support of the extension. So this is unnecessary.

Regards,

Weiwei Li

Quoted 13 lines

Another alternative is to not disable any extensions at all based on priv spec - we send a warning about the priv mismatch and that's it.

Thanks,

Daniel

Quoted 3 lines

              isa_ext_is_enabled(cpu, &isa_edata_arr[i])) { new = g_strconcat(old, "_", isa_edata_arr[i].name, NULL); g_free(old);

Reviewed-by on v1 1/1
#3 lore raw
In reply to Weiwei Li
Quoted 9 lines · Weiwei Li

Using implicitly enabled extensions such as Zca/Zcf/Zcd instead of their super extensions can simplify the extension related check. However, they may have higher priv version than their super extensions. So we should mask them in the isa_string based on priv version to make them invisible to user if the specified priv version is lower than their minimal priv version.

---

And I'll fold it into the next version of "[PATCH v6 0/9] target/riscv: rework CPU extensions validation​" to fix the sifive break I'm experiencing there.

Thanks,

Daniel

Quoted 15 lines
  target/riscv/cpu.c | 1 +
  1 file changed, 1 insertion(+)
target/riscv/cpu.c +1 -0
index cb68916fce..1a5099382c 100644
@@ -1709,6 +1709,7 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str,
 
      for (i = 0; i < ARRAY_SIZE(isa_edata_arr); i++) {
          if (isa_edata_arr[i].multi_letter &&
+            (cpu->env.priv_ver >= isa_edata_arr[i].min_version) &&
              isa_ext_is_enabled(cpu, &isa_edata_arr[i])) {
              new = g_strconcat(old, "_", isa_edata_arr[i].name, NULL);
              g_free(old);