Skip to contentSkip to conversation

target/riscv: add zicsr/zifencei to isa_string

PATCH Applied Hongren (Zenithal) Zheng 3 participants 6 messages May 18, 2022 – May 23, 2022 RISC-V

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 target/riscv: add zicsr/zifencei to isa_string

Zicsr/Zifencei is not in 'I' since ISA version 20190608, thus to fully express the capability of the CPU, they should be exposed in isa_string.

---
 target/riscv/cpu.c | 2 ++
 1 file changed, 2 insertions(+)
target/riscv/cpu.c +2 -0
index 6d01569cad..61fa9b97a4 100644
@@ -1027,6 +1027,8 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, int max_str_len)
      *    extensions by an underscore.
      */
     struct isa_ext_data isa_edata_arr[] = {
+        ISA_EDATA_ENTRY(zicsr, ext_icsr),
+        ISA_EDATA_ENTRY(zifencei, ext_ifencei),
         ISA_EDATA_ENTRY(zfh, ext_zfh),
         ISA_EDATA_ENTRY(zfhmin, ext_zfhmin),
         ISA_EDATA_ENTRY(zfinx, ext_zfinx),
Signature · 1 line
2.35.1
Tested-by on v1 1/1
#1 lore raw

This patch could work successfully in qemu. With command "cat /proc/device-tree/cpus/cpu@0/riscv\,isa", string "zicsr" and "zifencei" could be found in linux device tree.

#2 lore raw
In reply to Hongren (Zenithal) Zheng
Quoted 15 lines · Hongren (Zenithal) Zheng

Zicsr/Zifencei is not in 'I' since ISA version 20190608, thus to fully express the capability of the CPU, they should be exposed in isa_string.

---
 target/riscv/cpu.c | 2 ++
 1 file changed, 2 insertions(+)
target/riscv/cpu.c +0 -0
index 6d01569cad..61fa9b97a4 100644

target/riscv/cpu.c static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, int max_str_len) · line 1031

Quoted 5 lines
target/riscv/cpu.c +2 -0
@@ -1027,6 +1027,8 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, int max_str_len)
      *    extensions by an underscore.
      */
     struct isa_ext_data isa_edata_arr[] = {
+        ISA_EDATA_ENTRY(zicsr, ext_icsr),
+        ISA_EDATA_ENTRY(zifencei, ext_ifencei),

Shouldn't we have a spec version check here?

Alistair

Quoted 7 lines
target/riscv/cpu.c +0 -1
         ISA_EDATA_ENTRY(zfh, ext_zfh),
         ISA_EDATA_ENTRY(zfhmin, ext_zfhmin),
         ISA_EDATA_ENTRY(zfinx, ext_zfinx),
--

2.35.1

#3 lore raw
Quoted 11 lines · Alistair Francis
Quoted 10 lines · Hongren (Zenithal) Zheng

Zicsr/Zifencei is not in 'I' since ISA version 20190608, thus to fully express the capability of the CPU, they should be exposed in isa_string.

---
 target/riscv/cpu.c | 2 ++
 1 file changed, 2 insertions(+)
Quoted 12 lines

target/riscv/cpu.c static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, int max_str_len) · line 1031

Quoted 10 lines
target/riscv/cpu.c +2 -0
index 6d01569cad..61fa9b97a4 100644
@@ -1027,6 +1027,8 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, int max_str_len)
      *    extensions by an underscore.
      */
     struct isa_ext_data isa_edata_arr[] = {
+        ISA_EDATA_ENTRY(zicsr, ext_icsr),
+        ISA_EDATA_ENTRY(zifencei, ext_ifencei),

Shouldn't we have a spec version check here?

I think that can be done, but most of the time it is not necessary. For old specs, the "short-isa-string" can be a workaround.

This patch is actually a follow-up of "target/riscv: Change "G" expansion"
https://github.com/alistair23/qemu/commit/72bd25b7b88d0536bfb5666990e296587d4057a5
where "G" is expanded with "zicsr"/"zifencei" without checking unpriv spec version.

From the summary from Kito from gnu toolchain
https://lkml.org/lkml/2022/1/24/537
we know that there are at least 3 formally released unpriv specs:
2.2, 20190608 and 20191213, and frequent informal release from
https://github.com/riscv/riscv-isa-manual/releases

If we add a spec check, we need to add a PROP_STRING and and parse the version. We then need an enum like PRIV_VERSION_1_12_0 and we needs to g_strcmp0 in riscv_cpu_realize, just as what priv spec had done.

And we need to define a default unpriv spec version, which should be 20191213.

I can add a separate patch for it if you do think it is necessary.

If the guest does want old spec version, I think the recently added "short-isa-string" option from Tsukasa OI is suitable. Instead of -cpu rv64,unpriv_spec=2.2 they can just use -cpu rv64,short-isa-string=true to solve the issue.

Cc Tsukasa OI

Quoted 10 lines

Alistair

Quoted 7 lines
target/riscv/cpu.c +0 -1
         ISA_EDATA_ENTRY(zfh, ext_zfh),
         ISA_EDATA_ENTRY(zfhmin, ext_zfhmin),
         ISA_EDATA_ENTRY(zfinx, ext_zfinx),
--

2.35.1

Reviewed-by
#4 lore raw
Quoted 48 lines · Hongren (Zenithal) Zheng
Quoted 11 lines · Alistair Francis
Quoted 10 lines · Hongren (Zenithal) Zheng

Zicsr/Zifencei is not in 'I' since ISA version 20190608, thus to fully express the capability of the CPU, they should be exposed in isa_string.

---
 target/riscv/cpu.c | 2 ++
 1 file changed, 2 insertions(+)
Quoted 12 lines

target/riscv/cpu.c static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, int max_str_len) · line 1031

Quoted 10 lines
target/riscv/cpu.c +2 -0
index 6d01569cad..61fa9b97a4 100644
@@ -1027,6 +1027,8 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, int max_str_len)
      *    extensions by an underscore.
      */
     struct isa_ext_data isa_edata_arr[] = {
+        ISA_EDATA_ENTRY(zicsr, ext_icsr),
+        ISA_EDATA_ENTRY(zifencei, ext_ifencei),

Shouldn't we have a spec version check here?

I think that can be done, but most of the time it is not necessary. For old specs, the "short-isa-string" can be a workaround.

This patch is actually a follow-up of "target/riscv: Change "G" expansion"
https://github.com/alistair23/qemu/commit/72bd25b7b88d0536bfb5666990e296587d4057a5
where "G" is expanded with "zicsr"/"zifencei" without checking unpriv spec version.

From the summary from Kito from gnu toolchain
https://lkml.org/lkml/2022/1/24/537
we know that there are at least 3 formally released unpriv specs:
2.2, 20190608 and 20191213, and frequent informal release from
https://github.com/riscv/riscv-isa-manual/releases

If we add a spec check, we need to add a PROP_STRING and and parse the version. We then need an enum like PRIV_VERSION_1_12_0 and we needs to g_strcmp0 in riscv_cpu_realize, just as what priv spec had done.

And we need to define a default unpriv spec version, which should be 20191213.

Quoted 6 lines

I can add a separate patch for it if you do think it is necessary.

If the guest does want old spec version, I think the recently added "short-isa-string" option from Tsukasa OI is suitable. Instead of -cpu rv64,unpriv_spec=2.2 they can just use -cpu rv64,short-isa-string=true to solve the issue.

Ok, fair point

Alistair

Quoted 13 lines

Cc Tsukasa OI

Quoted 10 lines

Alistair

Quoted 7 lines
target/riscv/cpu.c +0 -1
         ISA_EDATA_ENTRY(zfh, ext_zfh),
         ISA_EDATA_ENTRY(zfhmin, ext_zfhmin),
         ISA_EDATA_ENTRY(zfinx, ext_zfinx),
--

2.35.1

#5 lore raw
In reply to Hongren (Zenithal) Zheng
Quoted 6 lines · Hongren (Zenithal) Zheng

Zicsr/Zifencei is not in 'I' since ISA version 20190608, thus to fully express the capability of the CPU, they should be exposed in isa_string.

Thanks!

Applied to riscv-to-apply.next

Alistair

Quoted 21 lines
---
 target/riscv/cpu.c | 2 ++
 1 file changed, 2 insertions(+)
target/riscv/cpu.c +2 -1
index 6d01569cad..61fa9b97a4 100644
@@ -1027,6 +1027,8 @@ static void riscv_isa_string_ext(RISCVCPU *cpu, char **isa_str, int max_str_len)
      *    extensions by an underscore.
      */
     struct isa_ext_data isa_edata_arr[] = {
+        ISA_EDATA_ENTRY(zicsr, ext_icsr),
+        ISA_EDATA_ENTRY(zifencei, ext_ifencei),
         ISA_EDATA_ENTRY(zfh, ext_zfh),
         ISA_EDATA_ENTRY(zfhmin, ext_zfhmin),
         ISA_EDATA_ENTRY(zfinx, ext_zfinx),
--

2.35.1