On 21/7/23 15:34, Daniel Henrique Barboza wrote:
The 'host' CPU is available in a CONFIG_KVM build and it's currently
available for all accels, but is a KVM only CPU. This means that in a
RISC-V KVM capable host we can do things like this:
$ ./build/qemu-system-riscv64 -M virt,accel=tcg -cpu host --nographic
qemu-system-riscv64: H extension requires priv spec 1.12.0
This CPU does not have a priv spec because we don't filter its extensions
via priv spec. We shouldn't be reaching riscv_cpu_realize_tcg() at all
with the 'host' CPU.
We don't have a way to filter the 'host' CPU out of the available CPU
options (-cpu help) if the build includes both KVM and TCG. What we can
do is to error out during riscv_cpu_realize_tcg() if the user chooses
the 'host' CPU with accel=tcg:
$ ./build/qemu-system-riscv64 -M virt,accel=tcg -cpu host --nographic
qemu-system-riscv64: 'host' CPU is not compatible with TCG acceleration
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
---
target/riscv/cpu.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/target/riscv/cpu.c b/target/riscv/cpu.c
index 6b93b04453..08db3d613f 100644
--- a/target/riscv/cpu.c
+++ b/target/riscv/cpu.c
@@ -1395,6 +1395,11 @@ static void riscv_cpu_realize_tcg(DeviceState *dev,
Error **errp)
/me wonders why this method isn't used as AccelCPUClass:cpu_realizefn().
CPURISCVState *env = &cpu->env;
Error *local_err = NULL;
+ if (object_dynamic_cast(OBJECT(dev), TYPE_RISCV_CPU_HOST)) {
+ error_setg(errp, "'host' CPU is not compatible with TCG acceleration");
+ return;
+ }
We should manage that generically, likely in CPUClass.
We could define riscv CPU with a DEFINE_CPU_WITH_CLASS() macro to pass
the class_init() handler. Anyhow meanwhile your patch is sufficient, so
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>