On Fri, Mar 1, 2024 at 12:05 PM Chao Du <duchao@xxxxxxxxxxxxxxxxxx> wrote:
>
> On 2024-03-01 13:00, Anup Patel <anup@xxxxxxxxxxxxxx> wrote:
> >
> > On Fri, Mar 1, 2024 at 7:08 AM Chao Du <duchao@xxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is
> > > been checked.
> > >
> > > kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags
> > > from userspace accordingly. Route the breakpoint exceptions to HS mode
> > > if the VCPU is being debugged by userspace, by clearing the
> > > corresponding bit in hedeleg. Write the actual CSR in
> > > kvm_arch_vcpu_load().
> > >
> > > Signed-off-by: Chao Du <duchao@xxxxxxxxxxxxxxxxxx>
> > > ---
> > > arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
> > > arch/riscv/include/uapi/asm/kvm.h | 1 +
> > > arch/riscv/kvm/main.c | 18 ++----------------
> > > arch/riscv/kvm/vcpu.c | 15 +++++++++++++--
> > > arch/riscv/kvm/vm.c | 1 +
> > > 5 files changed, 34 insertions(+), 18 deletions(-)
> > >
> > > diff --git a/arch/riscv/include/asm/kvm_host.h b/arch/riscv/include/asm/kvm_host.h
> > > index 484d04a92fa6..9ee3f03ba5d1 100644
> > > --- a/arch/riscv/include/asm/kvm_host.h
> > > +++ b/arch/riscv/include/asm/kvm_host.h
> > > @@ -43,6 +43,22 @@
> > > KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
> > > #define KVM_REQ_STEAL_UPDATE KVM_ARCH_REQ(6)
> > >
> > > +#define KVM_HEDELEG_DEFAULT ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
> > > + (_AC(1, UL) << EXC_BREAKPOINT) | \
> > > + (_AC(1, UL) << EXC_SYSCALL) | \
> > > + (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
> > > + (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
> > > + (_AC(1, UL) << EXC_STORE_PAGE_FAULT))
> >
> > Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)
>
> Thanks, I will use BIT() instead in next revision.
>
> >
>
> > Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.
>
> I think the bit EXC_BREAKPOINT should be set by default, like what you
> already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak
> and handle it accordingly.
>
> If the guest_debug is enabled, ebreak instructions are inserted by the
> userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU.
> Bit EXC_BREAKPOINT should be cleared in this case.
If EXC_BREAKPOINT is delegated by default then it is not consistent with
vcpu->guest_debug which is not enabled by default.
>
> >
> > > +#define KVM_HEDELEG_GUEST_DEBUG ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
> > > + (_AC(1, UL) << EXC_SYSCALL) | \
> > > + (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
> > > + (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
> > > + (_AC(1, UL) << EXC_STORE_PAGE_FAULT))
> >
> > No need for KVM_HEDELEG_GUEST_DEBUG, see below.
> >
> > > +
> > > +#define KVM_HIDELEG_DEFAULT ((_AC(1, UL) << IRQ_VS_SOFT) | \
> > > + (_AC(1, UL) << IRQ_VS_TIMER) | \
> > > + (_AC(1, UL) << IRQ_VS_EXT))
> > > +
> >
> > Same as above, use BIT(xyz) here.
> >
> > > enum kvm_riscv_hfence_type {
> > > KVM_RISCV_HFENCE_UNKNOWN = 0,
> > > KVM_RISCV_HFENCE_GVMA_VMID_GPA,
> > > @@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
> > > struct kvm_vcpu_config {
> > > u64 henvcfg;
> > > u64 hstateen0;
> > > + unsigned long hedeleg;
> > > };
> > >
> > > struct kvm_vcpu_smstateen_csr {
> > > diff --git a/arch/riscv/include/uapi/asm/kvm.h b/arch/riscv/include/uapi/asm/kvm.h
> > > index 7499e88a947c..39f4f4b9dede 100644
> > > --- a/arch/riscv/include/uapi/asm/kvm.h
> > > +++ b/arch/riscv/include/uapi/asm/kvm.h
> > > @@ -17,6 +17,7 @@
> > >
> > > #define __KVM_HAVE_IRQ_LINE
> > > #define __KVM_HAVE_READONLY_MEM
> > > +#define __KVM_HAVE_GUEST_DEBUG
> > >
> > > #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
> > >
> > > diff --git a/arch/riscv/kvm/main.c b/arch/riscv/kvm/main.c
> > > index 225a435d9c9a..bab2ec34cd87 100644
> > > --- a/arch/riscv/kvm/main.c
> > > +++ b/arch/riscv/kvm/main.c
> > > @@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
> > >
> > > int kvm_arch_hardware_enable(void)
> > > {
> > > - unsigned long hideleg, hedeleg;
> > > -
> > > - hedeleg = 0;
> > > - hedeleg |= (1UL << EXC_INST_MISALIGNED);
> > > - hedeleg |= (1UL << EXC_BREAKPOINT);
> > > - hedeleg |= (1UL << EXC_SYSCALL);
> > > - hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
> > > - hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
> > > - hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
> > > - csr_write(CSR_HEDELEG, hedeleg);
> > > -
> > > - hideleg = 0;
> > > - hideleg |= (1UL << IRQ_VS_SOFT);
> > > - hideleg |= (1UL << IRQ_VS_TIMER);
> > > - hideleg |= (1UL << IRQ_VS_EXT);
> > > - csr_write(CSR_HIDELEG, hideleg);
> > > + csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
> > > + csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
> > >
> > > /* VS should access only the time counter directly. Everything else should trap */
> > > csr_write(CSR_HCOUNTEREN, 0x02);
> > > diff --git a/arch/riscv/kvm/vcpu.c b/arch/riscv/kvm/vcpu.c
> > > index b5ca9f2e98ac..242076c2227f 100644
> > > --- a/arch/riscv/kvm/vcpu.c
> > > +++ b/arch/riscv/kvm/vcpu.c
> > > @@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
> > > int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
> > > struct kvm_guest_debug *dbg)
> > > {
> > > - /* TODO; To be implemented later. */
> > > - return -EINVAL;
> >
> > if (vcpu->arch.ran_atleast_once)
> > return -EBUSY;
>
> If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl
> will come before the first KVM_RUN. This will always cause an ERROR.
The check ensures that KVM user space can only enable/disable
guest debug before the VCPU is run. I don't see why this would
fail for QEMU.
>
> >
> >
> > > + if (dbg->control & KVM_GUESTDBG_ENABLE) {
> > > + vcpu->guest_debug = dbg->control;
> > > + vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
> > > + } else {
> > > + vcpu->guest_debug = 0;
> > > + vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
> > > + }
> >
> > Don't update vcpu->arch.cfg.hedeleg here since it should be only done
> > in kvm_riscv_vcpu_setup_config().
> >
> > > +
> > > + return 0;
> > > }
> > >
> > > static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
> > > @@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
> > > if (riscv_isa_extension_available(isa, SMSTATEEN))
> > > cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
> > > }
> > > +
> > > + if (!vcpu->guest_debug)
> > > + cfg->hedeleg = KVM_HEDELEG_DEFAULT;
> >
> > This should be:
> >
> > cfg->hedeleg = KVM_HEDELEG_DEFAULT;
> > if (vcpu->guest_debug)
> > cfg->hedeleg |= BIT(EXC_BREAKPOINT);
>
> Like above, here the logic should be:
>
> cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT)
> if (vcpu->guest_debug)
> cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);
>
> Another approach is:
> initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create().
> Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().
I disagree. We should handle hedeleg just like we handle henvcfg.
>
> >
> > > }
> > >
> > > void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
> > > @@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
> > > csr_write(CSR_VSEPC, csr->vsepc);
> > > csr_write(CSR_VSCAUSE, csr->vscause);
> > > csr_write(CSR_VSTVAL, csr->vstval);
> > > + csr_write(CSR_HEDELEG, cfg->hedeleg);
> > > csr_write(CSR_HVIP, csr->hvip);
> > > csr_write(CSR_VSATP, csr->vsatp);
> > > csr_write(CSR_HENVCFG, cfg->henvcfg);
> > > diff --git a/arch/riscv/kvm/vm.c b/arch/riscv/kvm/vm.c
> > > index ce58bc48e5b8..7396b8654f45 100644
> > > --- a/arch/riscv/kvm/vm.c
> > > +++ b/arch/riscv/kvm/vm.c
> > > @@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
> > > case KVM_CAP_READONLY_MEM:
> > > case KVM_CAP_MP_STATE:
> > > case KVM_CAP_IMMEDIATE_EXIT:
> > > + case KVM_CAP_SET_GUEST_DEBUG:
> > > r = 1;
> > > break;
> > > case KVM_CAP_NR_VCPUS:
> > > --
> > > 2.17.1
> > >
> >
> > Regards,
> > Anup
>
> Thanks,
> Chao
Regards,
Anup
