Re: [PATCH v6 2/2] LongArch: KVM: Add dmsintc inject msi to the dest vcpu

[Huacai Chen]

Hi, Song,

On Fri, Feb 6, 2026 at 9:45 AM Song Gao <gaosong@xxxxxxxxxxx> wrote:
>
> Implement irqfd deliver msi to vcpu and vcpu dmsintc inject irq.
> Add irqfd choice dmsintc to set msi irq by the msg_addr and
> implement dmsintc set msi irq.
>
> Signed-off-by: Song Gao <gaosong@xxxxxxxxxxx>
> ---
>  arch/loongarch/include/asm/kvm_dmsintc.h |  1 +
>  arch/loongarch/include/asm/kvm_host.h    |  5 ++
>  arch/loongarch/kvm/intc/dmsintc.c        |  6 +++
>  arch/loongarch/kvm/interrupt.c           |  1 +
>  arch/loongarch/kvm/irqfd.c               | 42 +++++++++++++++--
>  arch/loongarch/kvm/vcpu.c                | 58 ++++++++++++++++++++++++
>  6 files changed, 109 insertions(+), 4 deletions(-)
>
> diff --git a/arch/loongarch/include/asm/kvm_dmsintc.h b/arch/loongarch/include/asm/kvm_dmsintc.h
> index 1d4f66996f3c..9b5436a2fcbe 100644
> --- a/arch/loongarch/include/asm/kvm_dmsintc.h
> +++ b/arch/loongarch/include/asm/kvm_dmsintc.h
> @@ -11,6 +11,7 @@ struct loongarch_dmsintc  {
>         struct kvm *kvm;
>         uint64_t msg_addr_base;
>         uint64_t msg_addr_size;
> +       uint32_t cpu_mask;
>  };
>
>  struct dmsintc_state {
> diff --git a/arch/loongarch/include/asm/kvm_host.h b/arch/loongarch/include/asm/kvm_host.h
> index 5e9e2af7312f..91e0190aeaec 100644
> --- a/arch/loongarch/include/asm/kvm_host.h
> +++ b/arch/loongarch/include/asm/kvm_host.h
> @@ -258,6 +258,11 @@ struct kvm_vcpu_arch {
>         } st;
>  };
>
> +void loongarch_dmsintc_inject_irq(struct kvm_vcpu *vcpu);
> +int kvm_loongarch_deliver_msi_to_vcpu(struct kvm *kvm,
> +                               struct kvm_vcpu *vcpu,
> +                               u32 vector, int level);
> +
>  static inline unsigned long readl_sw_gcsr(struct loongarch_csrs *csr, int reg)
>  {
>         return csr->csrs[reg];
> diff --git a/arch/loongarch/kvm/intc/dmsintc.c b/arch/loongarch/kvm/intc/dmsintc.c
> index 00e401de0464..1bb61e55d061 100644
> --- a/arch/loongarch/kvm/intc/dmsintc.c
> +++ b/arch/loongarch/kvm/intc/dmsintc.c
> @@ -15,6 +15,7 @@ static int kvm_dmsintc_ctrl_access(struct kvm_device *dev,
>         void __user *data;
>         struct loongarch_dmsintc *s = dev->kvm->arch.dmsintc;
>         u64 tmp;
> +       u32 cpu_bit;
>
>         data = (void __user *)attr->addr;
>         switch (addr) {
> @@ -30,6 +31,11 @@ static int kvm_dmsintc_ctrl_access(struct kvm_device *dev,
>                                 s->msg_addr_base = tmp;
>                         else
>                                 return  -EFAULT;
> +                       s->msg_addr_base = tmp;
> +                       cpu_bit = find_first_bit((unsigned long *)&(s->msg_addr_base), 64)
> +                                               - AVEC_CPU_SHIFT;
> +                       cpu_bit = min(cpu_bit, AVEC_CPU_BIT);
> +                       s->cpu_mask = GENMASK(cpu_bit - 1, 0) & AVEC_CPU_MASK;
>                 }
>                 break;
>         case KVM_DEV_LOONGARCH_DMSINTC_MSG_ADDR_SIZE:
I'm not sure but maybe this part should go to the first patch?

> diff --git a/arch/loongarch/kvm/interrupt.c b/arch/loongarch/kvm/interrupt.c
> index a6d42d399a59..893a81ca1079 100644
> --- a/arch/loongarch/kvm/interrupt.c
> +++ b/arch/loongarch/kvm/interrupt.c
> @@ -33,6 +33,7 @@ static int kvm_irq_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
>                 irq = priority_to_irq[priority];
>
>         if (cpu_has_msgint && (priority == INT_AVEC)) {
> +               loongarch_dmsintc_inject_irq(vcpu);
>                 set_gcsr_estat(irq);
>                 return 1;
>         }
> diff --git a/arch/loongarch/kvm/irqfd.c b/arch/loongarch/kvm/irqfd.c
> index 9a39627aecf0..3bbb26f4e2b7 100644
> --- a/arch/loongarch/kvm/irqfd.c
> +++ b/arch/loongarch/kvm/irqfd.c
> @@ -6,6 +6,7 @@
>  #include <linux/kvm_host.h>
>  #include <trace/events/kvm.h>
>  #include <asm/kvm_pch_pic.h>
> +#include <asm/kvm_vcpu.h>
>
>  static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e,
>                 struct kvm *kvm, int irq_source_id, int level, bool line_status)
> @@ -16,6 +17,38 @@ static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e,
>         return 0;
>  }
>
> +static int kvm_dmsintc_set_msi_irq(struct kvm *kvm, u32 addr, int data, int level)
> +{
> +       unsigned int virq, dest;
> +       struct kvm_vcpu *vcpu;
> +
> +       virq = (addr >> AVEC_IRQ_SHIFT) & AVEC_IRQ_MASK;
> +       dest = (addr >> AVEC_CPU_SHIFT) & kvm->arch.dmsintc->cpu_mask;
> +       if (dest > KVM_MAX_VCPUS)
> +               return -EINVAL;
> +       vcpu = kvm_get_vcpu_by_cpuid(kvm, dest);
> +       if (!vcpu)
> +               return -EINVAL;
> +       return kvm_loongarch_deliver_msi_to_vcpu(kvm, vcpu, virq, level);
> +}
> +
> +static int loongarch_set_msi(struct kvm_kernel_irq_routing_entry *e,
> +                       struct kvm *kvm, int level)
> +{
> +       u64 msg_addr;
> +
> +       msg_addr = (((u64)e->msi.address_hi) << 32) | e->msi.address_lo;
> +       if (cpu_has_msgint && kvm->arch.dmsintc &&
> +               msg_addr >= kvm->arch.dmsintc->msg_addr_base &&
> +               msg_addr < (kvm->arch.dmsintc->msg_addr_base  + kvm->arch.dmsintc->msg_addr_size)) {
> +               return kvm_dmsintc_set_msi_irq(kvm, msg_addr, e->msi.data, level);
> +       } else {
> +               pch_msi_set_irq(kvm, e->msi.data, level);
> +       }
> +
> +       return 0;
> +}
Rename loongarch_set_msi() to loongarch_msi_set_irq(), rename
kvm_dmsintc_set_msi_irq() to dmsintc_msi_set_irq(), this makes the
naming more consistent.

> +
>  /*
>   * kvm_set_msi: inject the MSI corresponding to the
>   * MSI routing entry
> @@ -29,9 +62,7 @@ int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
>         if (!level)
>                 return -1;
>
> -       pch_msi_set_irq(kvm, e->msi.data, level);
> -
> -       return 0;
> +       return loongarch_set_msi(e, kvm, level);
>  }
>
>  /*
> @@ -71,12 +102,15 @@ int kvm_set_routing_entry(struct kvm *kvm,
>  int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
>                 struct kvm *kvm, int irq_source_id, int level, bool line_status)
>  {
> +       if (!level)
> +               return -EWOULDBLOCK;
> +
>         switch (e->type) {
>         case KVM_IRQ_ROUTING_IRQCHIP:
>                 pch_pic_set_irq(kvm->arch.pch_pic, e->irqchip.pin, level);
>                 return 0;
>         case KVM_IRQ_ROUTING_MSI:
> -               pch_msi_set_irq(kvm, e->msi.data, level);
> +               loongarch_set_msi(e, kvm, level);
>                 return 0;
>         default:
>                 return -EWOULDBLOCK;
> diff --git a/arch/loongarch/kvm/vcpu.c b/arch/loongarch/kvm/vcpu.c
> index 656b954c1134..325bb084d704 100644
> --- a/arch/loongarch/kvm/vcpu.c
> +++ b/arch/loongarch/kvm/vcpu.c
> @@ -14,6 +14,64 @@
>  #define CREATE_TRACE_POINTS
>  #include "trace.h"
>
> +void loongarch_dmsintc_inject_irq(struct kvm_vcpu *vcpu)
> +{
> +       struct dmsintc_state *ds = &vcpu->arch.dmsintc_state;
> +       unsigned int i;
> +       unsigned long temp[4], old;
> +
> +       if (!ds)
> +               return;
> +
> +       for (i = 0; i < 4; i++) {
> +               old = atomic64_read(&(ds->vector_map[i]));
> +               if (old)
> +                       temp[i] = atomic64_xchg(&(ds->vector_map[i]), 0);
> +       }
> +
> +       if (temp[0]) {
> +               old = kvm_read_hw_gcsr(LOONGARCH_CSR_ISR0);
> +               kvm_write_hw_gcsr(LOONGARCH_CSR_ISR0, temp[0]|old);
> +       }
> +
> +       if (temp[1]) {
> +               old = kvm_read_hw_gcsr(LOONGARCH_CSR_ISR1);
> +               kvm_write_hw_gcsr(LOONGARCH_CSR_ISR1, temp[1]|old);
> +       }
> +
> +       if (temp[2]) {
> +               old = kvm_read_hw_gcsr(LOONGARCH_CSR_ISR2);
> +               kvm_write_hw_gcsr(LOONGARCH_CSR_ISR2, temp[2]|old);
> +       }
> +
> +       if (temp[3]) {
> +               old = kvm_read_hw_gcsr(LOONGARCH_CSR_ISR3);
> +               kvm_write_hw_gcsr(LOONGARCH_CSR_ISR3, temp[3]|old);
> +       }
> +}
The only caller is in interrupt.c, so rename
loongarch_dmsintc_inject_irq() to msgint_inject_irq() (or
dmsintc_inject_irq() if you prefer), and move it to interrupt.c, then
we don't need to declare it as a extern function.

> +
> +int kvm_loongarch_deliver_msi_to_vcpu(struct kvm *kvm,
> +                               struct kvm_vcpu *vcpu,
> +                               u32 vector, int level)
> +{
> +       struct kvm_interrupt vcpu_irq;
> +       struct dmsintc_state *ds;
> +
> +       if (!level)
> +               return 0;
> +       if (!vcpu || vector >= 256)
> +               return -EINVAL;
> +       ds = &vcpu->arch.dmsintc_state;
> +       if (!ds)
> +               return -ENODEV;
> +       set_bit(vector, (unsigned long *)&ds->vector_map);
> +       vcpu_irq.irq = INT_AVEC;
> +       kvm_vcpu_ioctl_interrupt(vcpu, &vcpu_irq);
> +       kvm_vcpu_kick(vcpu);
> +       return 0;
> +}
The only caller is in irqfd.c, so rename
kvm_loongarch_deliver_msi_to_vcpu() to dmsintc_deliver_msi_to_vcpu(),
and move it to irqfd.c, then we don't need to declare it as a extern
function.

And in addition, from Documentation/arch/loongarch/irq-chip-model.rst,
all msi irq are triggered from "pch_msi_irq", which means it is not
reasonable to dispatch the dmsintc/pch_msi paths in
loongarch_set_msi(). Instead, we should dispatch the dmsintc/eiointc
paths in pch_msi_set_irq(), this needs a rework...


Huacai

> +
> +
>  const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
>         KVM_GENERIC_VCPU_STATS(),
>         STATS_DESC_COUNTER(VCPU, int_exits),
> --
> 2.39.3
>
>