[RFC PATCH 6/7] riscv: kvm: add SBI FWFT support for SBI_FWFT_DOUBLE_TRAP_ENABLE

[Clément Léger]

On 27/04/2024 03:17, Deepak Gupta wrote:
> On Thu, Apr 18, 2024 at 04:26:45PM +0200, Clément Léger wrote:
>> Add support in KVM SBI FWFT extension to allow VS-mode to request double
>> trap enabling. Double traps can then be generated by VS-mode, allowing
>> M-mode to redirect them to S-mode.
>>
>> Signed-off-by: Clément Léger <cleger at rivosinc.com>
>> ---
>> arch/riscv/include/asm/csr.h               |  1 +
>> arch/riscv/include/asm/kvm_vcpu_sbi_fwft.h |  2 +-
>> arch/riscv/kvm/vcpu_sbi_fwft.c             | 41 ++++++++++++++++++++++
>> 3 files changed, 43 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/riscv/include/asm/csr.h b/arch/riscv/include/asm/csr.h
>> index 905cdf894a57..ee1b73655bec 100644
>> --- a/arch/riscv/include/asm/csr.h
>> +++ b/arch/riscv/include/asm/csr.h
>> @@ -196,6 +196,7 @@
>> /* xENVCFG flags */
>> #define ENVCFG_STCE            (_AC(1, ULL) << 63)
>> #define ENVCFG_PBMTE            (_AC(1, ULL) << 62)
>> +#define ENVCFG_DTE            (_AC(1, ULL) << 59)
>> #define ENVCFG_CBZE            (_AC(1, UL) << 7)
>> #define ENVCFG_CBCFE            (_AC(1, UL) << 6)
>> #define ENVCFG_CBIE_SHIFT        4
>> diff --git a/arch/riscv/include/asm/kvm_vcpu_sbi_fwft.h
>> b/arch/riscv/include/asm/kvm_vcpu_sbi_fwft.h
>> index 7dc1b80c7e6c..a9e20d655126 100644
>> --- a/arch/riscv/include/asm/kvm_vcpu_sbi_fwft.h
>> +++ b/arch/riscv/include/asm/kvm_vcpu_sbi_fwft.h
>> @@ -11,7 +11,7 @@
>>
>> #include <asm/sbi.h>
>>
>> -#define KVM_SBI_FWFT_FEATURE_COUNT    1
>> +#define KVM_SBI_FWFT_FEATURE_COUNT    2
>>
>> struct kvm_sbi_fwft_config;
>> struct kvm_vcpu;
>> diff --git a/arch/riscv/kvm/vcpu_sbi_fwft.c
>> b/arch/riscv/kvm/vcpu_sbi_fwft.c
>> index b9b7f8fa6d22..9e8e397eb02f 100644
>> --- a/arch/riscv/kvm/vcpu_sbi_fwft.c
>> +++ b/arch/riscv/kvm/vcpu_sbi_fwft.c
>> @@ -9,10 +9,19 @@
>> #include <linux/errno.h>
>> #include <linux/err.h>
>> #include <linux/kvm_host.h>
>> +#include <linux/riscv_dbltrp.h>
>> #include <asm/sbi.h>
>> #include <asm/kvm_vcpu_sbi.h>
>> #include <asm/kvm_vcpu_sbi_fwft.h>
>>
>> +#ifdef CONFIG_32BIT
>> +# define CSR_HENVCFG_DBLTRP    CSR_HENVCFGH
>> +# define DBLTRP_DTE    (ENVCFG_DTE >> 32)
>> +#else
>> +# define CSR_HENVCFG_DBLTRP    CSR_HENVCFG
>> +# define DBLTRP_DTE    ENVCFG_DTE
>> +#endif
>> +
>> #define MIS_DELEG (1UL << EXC_LOAD_MISALIGNED | 1UL <<
>> EXC_STORE_MISALIGNED)
>>
>> static int kvm_sbi_fwft_set_misaligned_delegation(struct kvm_vcpu *vcpu,
>> @@ -36,6 +45,33 @@ static int
>> kvm_sbi_fwft_get_misaligned_delegation(struct kvm_vcpu *vcpu,
>>     return SBI_SUCCESS;
>> }
>>
>> +static int kvm_sbi_fwft_set_double_trap(struct kvm_vcpu *vcpu,
>> +                    struct kvm_sbi_fwft_config *conf,
>> +                    unsigned long value)
>> +{
>> +    if (!riscv_double_trap_enabled())
>> +        return SBI_ERR_NOT_SUPPORTED;
> 
> Why its required to check whether host has enabled double trap for itself ?
> It's orthogonal to guest asking hypervisor to enable double trap.

Hi Deepak,

Indeed, as you saw, henvcfg.DTE needs menvcfg.DTE to be enabled in order
to be usable.

> 
> Probably you need a check here whether underlying FW supports handling
> double
> trap.
> 
> Am I missing something here?
> 
>> +
>> +    if (value)
>> +        csr_set(CSR_HENVCFG_DBLTRP, DBLTRP_DTE);
>> +    else
>> +        csr_clear(CSR_HENVCFG_DBLTRP, DBLTRP_DTE);
> 
> I think vcpu->arch.cfg has `henvcfg` field. Can we reflect it there as
> well so that current
> `henvcfg` copy in vcpu arch specifci config is consistent? Otherwise
> it'll be lost when vCPU
> is scheduled out and later scheduled back in (on vcpu load)

henvcfg is restored when loading the vpcu (kvm_arch_vcpu_load()) and
saved when the CPU is put (kvm_arch_vcpu_put()). But I just saw that
this change is included in the next patch. Should have been this one ,
I'll fix that.


> 
> Furthermore, lets not do feature specific alias names for CSR.
> 
> Instead let's keep consistent 64bit image of henvcfg in vcpu->arch.cfg.
> 
> And whenever it's time to pick up the setting, pick up logic either
> perform the writes in
> henvcfg. And if required it'll perform henvcfgh too (as
> `kvm_arch_vcpu_load` already does)

I don't have a strong opinion on that point so if you think it really is
better, I'll switch to that.

Thanks,

Clément

> 
>> +
>> +    return SBI_SUCCESS;
>> +}
>> +
>> +static int kvm_sbi_fwft_get_double_trap(struct kvm_vcpu *vcpu,
>> +                    struct kvm_sbi_fwft_config *conf,
>> +                    unsigned long *value)
>> +{
>> +    if (!riscv_double_trap_enabled())
>> +        return SBI_ERR_NOT_SUPPORTED;
>> +
>> +    *value = (csr_read(CSR_HENVCFG_DBLTRP) & DBLTRP_DTE) != 0;
>> +
>> +    return SBI_SUCCESS;
>> +}
>> +
>> static struct kvm_sbi_fwft_config *
>> kvm_sbi_fwft_get_config(struct kvm_vcpu *vcpu, enum sbi_fwft_feature_t
>> feature)
>> {
>> @@ -111,6 +147,11 @@ static const struct kvm_sbi_fwft_feature
>> features[] = {
>>         .id = SBI_FWFT_MISALIGNED_DELEG,
>>         .set = kvm_sbi_fwft_set_misaligned_delegation,
>>         .get = kvm_sbi_fwft_get_misaligned_delegation,
>> +    },
>> +    {
>> +        .id = SBI_FWFT_DOUBLE_TRAP_ENABLE,
>> +        .set = kvm_sbi_fwft_set_double_trap,
>> +        .get = kvm_sbi_fwft_get_double_trap,
>>     }
>> };
>>
>> -- 
>> 2.43.0
>>
>>