Re: [PATCH v3 8/8] RISC-V: KVM: Implement guest external interrupt line management

[Anup Patel]

On Tue, Apr 4, 2023 at 6:15 PM Andrew Jones <ajones@xxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Apr 03, 2023 at 03:03:10PM +0530, Anup Patel wrote:
> > The RISC-V host will have one guest external interrupt line for each
> > VS-level IMSICs associated with a HART. The guest external interrupt
> > lines are per-HART resources and hypervisor can use HGEIE, HGEIP, and
> > HIE CSRs to manage these guest external interrupt lines.
> >
> > Signed-off-by: Anup Patel <apatel@xxxxxxxxxxxxxxxx>
> > ---
> >  arch/riscv/include/asm/kvm_aia.h |  10 ++
> >  arch/riscv/kvm/aia.c             | 241 +++++++++++++++++++++++++++++++
> >  arch/riscv/kvm/main.c            |   3 +-
> >  arch/riscv/kvm/vcpu.c            |   2 +
> >  4 files changed, 255 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/riscv/include/asm/kvm_aia.h b/arch/riscv/include/asm/kvm_aia.h
> > index 1de0717112e5..0938e0cadf80 100644
> > --- a/arch/riscv/include/asm/kvm_aia.h
> > +++ b/arch/riscv/include/asm/kvm_aia.h
> > @@ -44,10 +44,15 @@ struct kvm_vcpu_aia {
> >
> >  #define irqchip_in_kernel(k)         ((k)->arch.aia.in_kernel)
> >
> > +extern unsigned int kvm_riscv_aia_nr_hgei;
> >  DECLARE_STATIC_KEY_FALSE(kvm_riscv_aia_available);
> >  #define kvm_riscv_aia_available() \
> >       static_branch_unlikely(&kvm_riscv_aia_available)
> >
> > +static inline void kvm_riscv_vcpu_aia_imsic_release(struct kvm_vcpu *vcpu)
> > +{
> > +}
> > +
> >  #define KVM_RISCV_AIA_IMSIC_TOPEI    (ISELECT_MASK + 1)
> >  static inline int kvm_riscv_vcpu_aia_imsic_rmw(struct kvm_vcpu *vcpu,
> >                                              unsigned long isel,
> > @@ -119,6 +124,11 @@ static inline void kvm_riscv_aia_destroy_vm(struct kvm *kvm)
> >  {
> >  }
> >
> > +int kvm_riscv_aia_alloc_hgei(int cpu, struct kvm_vcpu *owner,
> > +                          void __iomem **hgei_va, phys_addr_t *hgei_pa);
> > +void kvm_riscv_aia_free_hgei(int cpu, int hgei);
> > +void kvm_riscv_aia_wakeon_hgei(struct kvm_vcpu *owner, bool enable);
> > +
> >  void kvm_riscv_aia_enable(void);
> >  void kvm_riscv_aia_disable(void);
> >  int kvm_riscv_aia_init(void);
> > diff --git a/arch/riscv/kvm/aia.c b/arch/riscv/kvm/aia.c
> > index d530912f28bc..1264783e7c4d 100644
> > --- a/arch/riscv/kvm/aia.c
> > +++ b/arch/riscv/kvm/aia.c
> > @@ -7,11 +7,46 @@
> >   *   Anup Patel <apatel@xxxxxxxxxxxxxxxx>
> >   */
> >
> > +#include <linux/bitops.h>
> > +#include <linux/irq.h>
> > +#include <linux/irqdomain.h>
> >  #include <linux/kvm_host.h>
> > +#include <linux/percpu.h>
> > +#include <linux/spinlock.h>
> >  #include <asm/hwcap.h>
> >
> > +struct aia_hgei_control {
> > +     raw_spinlock_t lock;
> > +     unsigned long free_bitmap;
> > +     struct kvm_vcpu *owners[BITS_PER_LONG];
> > +};
> > +static DEFINE_PER_CPU(struct aia_hgei_control, aia_hgei);
> > +static int hgei_parent_irq;
> > +
> > +unsigned int kvm_riscv_aia_nr_hgei;
> >  DEFINE_STATIC_KEY_FALSE(kvm_riscv_aia_available);
> >
> > +static int aia_find_hgei(struct kvm_vcpu *owner)
> > +{
> > +     int i, hgei;
> > +     unsigned long flags;
> > +     struct aia_hgei_control *hgctrl = this_cpu_ptr(&aia_hgei);
> > +
> > +     raw_spin_lock_irqsave(&hgctrl->lock, flags);
> > +
> > +     hgei = -1;
> > +     for (i = 1; i <= kvm_riscv_aia_nr_hgei; i++) {
> > +             if (hgctrl->owners[i] == owner) {
> > +                     hgei = i;
> > +                     break;
> > +             }
> > +     }
> > +
> > +     raw_spin_unlock_irqrestore(&hgctrl->lock, flags);
> > +
> > +     return hgei;
> > +}
> > +
> >  static void aia_set_hvictl(bool ext_irq_pending)
> >  {
> >       unsigned long hvictl;
> > @@ -55,6 +90,7 @@ void kvm_riscv_vcpu_aia_sync_interrupts(struct kvm_vcpu *vcpu)
> >
> >  bool kvm_riscv_vcpu_aia_has_interrupts(struct kvm_vcpu *vcpu, u64 mask)
> >  {
> > +     int hgei;
> >       unsigned long seip;
> >
> >       if (!kvm_riscv_aia_available())
> > @@ -72,6 +108,10 @@ bool kvm_riscv_vcpu_aia_has_interrupts(struct kvm_vcpu *vcpu, u64 mask)
> >       if (!kvm_riscv_aia_initialized(vcpu->kvm) || !seip)
> >               return false;
> >
> > +     hgei = aia_find_hgei(vcpu);
> > +     if (hgei > 0)
> > +             return (csr_read(CSR_HGEIP) & BIT(hgei)) ? true : false;
>
> nit: return !!(csr_read(CSR_HGEIP) & BIT(hgei)) is a bit less verbose.

Okay, I will update.

>
> > +
> >       return false;
> >  }
> >
> > @@ -343,6 +383,144 @@ int kvm_riscv_vcpu_aia_rmw_ireg(struct kvm_vcpu *vcpu, unsigned int csr_num,
> >       return KVM_INSN_EXIT_TO_USER_SPACE;
> >  }
> >
> > +int kvm_riscv_aia_alloc_hgei(int cpu, struct kvm_vcpu *owner,
> > +                          void __iomem **hgei_va, phys_addr_t *hgei_pa)
> > +{
> > +     int ret = -ENOENT;
> > +     unsigned long flags;
> > +     struct aia_hgei_control *hgctrl = per_cpu_ptr(&aia_hgei, cpu);
> > +
> > +     if (!kvm_riscv_aia_available())
> > +             return -ENODEV;
> > +     if (!hgctrl)
> > +             return -ENODEV;
>
> nit:
>
> if (!kvm_riscv_aia_available() || !hgctrl)
>    return -ENODEV;

Okay, I will update.

>
> > +
> > +     raw_spin_lock_irqsave(&hgctrl->lock, flags);
> > +
> > +     if (hgctrl->free_bitmap) {
> > +             ret = __ffs(hgctrl->free_bitmap);
> > +             hgctrl->free_bitmap &= ~BIT(ret);
> > +             hgctrl->owners[ret] = owner;
> > +     }
> > +
> > +     raw_spin_unlock_irqrestore(&hgctrl->lock, flags);
> > +
> > +     /* TODO: To be updated later by AIA in-kernel irqchip support */
> > +     if (hgei_va)
> > +             *hgei_va = NULL;
> > +     if (hgei_pa)
> > +             *hgei_pa = 0;
> > +
> > +     return ret;
> > +}
> > +
> > +void kvm_riscv_aia_free_hgei(int cpu, int hgei)
> > +{
> > +     unsigned long flags;
> > +     struct aia_hgei_control *hgctrl = per_cpu_ptr(&aia_hgei, cpu);
> > +
> > +     if (!kvm_riscv_aia_available() || !hgctrl)
> > +             return;
> > +
> > +     raw_spin_lock_irqsave(&hgctrl->lock, flags);
> > +
> > +     if (hgei > 0 && hgei <= kvm_riscv_aia_nr_hgei) {
> > +             if (!(hgctrl->free_bitmap & BIT(hgei))) {
> > +                     hgctrl->free_bitmap |= BIT(hgei);
> > +                     hgctrl->owners[hgei] = NULL;
> > +             }
> > +     }
> > +
> > +     raw_spin_unlock_irqrestore(&hgctrl->lock, flags);
> > +}
> > +
> > +void kvm_riscv_aia_wakeon_hgei(struct kvm_vcpu *owner, bool enable)
> > +{
> > +     int hgei;
> > +
> > +     if (!kvm_riscv_aia_available())
> > +             return;
> > +
> > +     hgei = aia_find_hgei(owner);
> > +     if (hgei > 0) {
> > +             if (enable)
> > +                     csr_set(CSR_HGEIE, BIT(hgei));
> > +             else
> > +                     csr_clear(CSR_HGEIE, BIT(hgei));
> > +     }
> > +}
> > +
> > +static irqreturn_t hgei_interrupt(int irq, void *dev_id)
> > +{
> > +     int i;
> > +     unsigned long hgei_mask, flags;
> > +     struct aia_hgei_control *hgctrl = this_cpu_ptr(&aia_hgei);
> > +
> > +     hgei_mask = csr_read(CSR_HGEIP) & csr_read(CSR_HGEIE);
> > +     csr_clear(CSR_HGEIE, hgei_mask);
> > +
> > +     raw_spin_lock_irqsave(&hgctrl->lock, flags);
> > +
> > +     for_each_set_bit(i, &hgei_mask, BITS_PER_LONG) {
> > +             if (hgctrl->owners[i])
> > +                     kvm_vcpu_kick(hgctrl->owners[i]);
> > +     }
> > +
> > +     raw_spin_unlock_irqrestore(&hgctrl->lock, flags);
> > +
> > +     return IRQ_HANDLED;
> > +}
> > +
> > +static int aia_hgei_init(void)
> > +{
> > +     int cpu, rc;
> > +     struct irq_domain *domain;
> > +     struct aia_hgei_control *hgctrl;
> > +
> > +     /* Initialize per-CPU guest external interrupt line management */
> > +     for_each_possible_cpu(cpu) {
> > +             hgctrl = per_cpu_ptr(&aia_hgei, cpu);
> > +             raw_spin_lock_init(&hgctrl->lock);
> > +             if (kvm_riscv_aia_nr_hgei) {
> > +                     hgctrl->free_bitmap =
> > +                             BIT(kvm_riscv_aia_nr_hgei + 1) - 1;
> > +                     hgctrl->free_bitmap &= ~BIT(0);
> > +             } else
> > +                     hgctrl->free_bitmap = 0;
> > +     }
> > +
> > +     /* Find INTC irq domain */
> > +     domain = irq_find_matching_fwnode(riscv_get_intc_hwnode(),
> > +                                       DOMAIN_BUS_ANY);
> > +     if (!domain) {
> > +             kvm_err("unable to find INTC domain\n");
> > +             return -ENOENT;
> > +     }
> > +
> > +     /* Map per-CPU SGEI interrupt from INTC domain */
> > +     hgei_parent_irq = irq_create_mapping(domain, IRQ_S_GEXT);
> > +     if (!hgei_parent_irq) {
> > +             kvm_err("unable to map SGEI IRQ\n");
> > +             return -ENOMEM;
> > +     }
> > +
> > +     /* Request per-CPU SGEI interrupt */
> > +     rc = request_percpu_irq(hgei_parent_irq, hgei_interrupt,
> > +                             "riscv-kvm", &aia_hgei);
> > +     if (rc) {
> > +             kvm_err("failed to request SGEI IRQ\n");
> > +             return rc;
> > +     }
> > +
> > +     return 0;
> > +}
> > +
> > +static void aia_hgei_exit(void)
> > +{
> > +     /* Free per-CPU SGEI interrupt */
> > +     free_percpu_irq(hgei_parent_irq, &aia_hgei);
> > +}
> > +
> >  void kvm_riscv_aia_enable(void)
> >  {
> >       if (!kvm_riscv_aia_available())
> > @@ -357,21 +535,79 @@ void kvm_riscv_aia_enable(void)
> >       csr_write(CSR_HVIPRIO1H, 0x0);
> >       csr_write(CSR_HVIPRIO2H, 0x0);
> >  #endif
> > +
> > +     /* Enable per-CPU SGEI interrupt */
> > +     enable_percpu_irq(hgei_parent_irq,
> > +                       irq_get_trigger_type(hgei_parent_irq));
> > +     csr_set(CSR_HIE, BIT(IRQ_S_GEXT));
> >  }
> >
> >  void kvm_riscv_aia_disable(void)
> >  {
> > +     int i;
> > +     unsigned long flags;
> > +     struct kvm_vcpu *vcpu;
> > +     struct aia_hgei_control *hgctrl = this_cpu_ptr(&aia_hgei);
> > +
> >       if (!kvm_riscv_aia_available())
> >               return;
> >
> > +     /* Disable per-CPU SGEI interrupt */
> > +     csr_clear(CSR_HIE, BIT(IRQ_S_GEXT));
> > +     disable_percpu_irq(hgei_parent_irq);
> > +
> >       aia_set_hvictl(false);
> > +
> > +     raw_spin_lock_irqsave(&hgctrl->lock, flags);
> > +
> > +     for (i = 0; i <= kvm_riscv_aia_nr_hgei; i++) {
>
> I guess this should start at i = 1, but in this case it doesn't
> matter since hgctrl->owners[0] should always be NULL.

Yes, it doesn't matter in this case.

>
> > +             vcpu = hgctrl->owners[i];
> > +             if (!vcpu)
> > +                     continue;
> > +
> > +             /*
> > +              * We release hgctrl->lock before notifying IMSIC
> > +              * so that we don't have lock ordering issues.
> > +              */
> > +             raw_spin_unlock_irqrestore(&hgctrl->lock, flags);
> > +
> > +             /* Notify IMSIC */
> > +             kvm_riscv_vcpu_aia_imsic_release(vcpu);
> > +
> > +             /*
> > +              * Wakeup VCPU if it was blocked so that it can
> > +              * run on other HARTs
> > +              */
> > +             if (csr_read(CSR_HGEIE) & BIT(i)) {
> > +                     csr_clear(CSR_HGEIE, BIT(i));
> > +                     kvm_vcpu_kick(vcpu);
>
> Doing all this outside the lock makes me wonder what happens when 'vcpu'
> is no longer the owner at the time of kvm_riscv_vcpu_aia_imsic_release()
> or kvm_vcpu_kick(). Even if the calls on the wrong vcpu are just noise,
> then don't we still need to confirm that we release/kick the real owner
> before we return from this function?
>
> It appears safe to call kvm_vcpu_kick() while holding the lock and
> hgei_interrupt() does that. So, since there's currently no
> implementation of kvm_riscv_vcpu_aia_imsic_release(), I'm not sure what
> lock ordering issues we need to avoid.

Next batch of AIA patches has kvm_riscv_vcpu_aia_imsic_release() where
kvm_riscv_vcpu_aia_imsic_release() function will release the HGEI interrupt
line.

This is not to protect the kvm_vcpu_kick() call.

Regards,
Anup

>
> > +             }
> > +
> > +             raw_spin_lock_irqsave(&hgctrl->lock, flags);
> > +     }
> > +
> > +     raw_spin_unlock_irqrestore(&hgctrl->lock, flags);
> >  }
> >
> >  int kvm_riscv_aia_init(void)
> >  {
> > +     int rc;
> > +
> >       if (!riscv_isa_extension_available(NULL, SxAIA))
> >               return -ENODEV;
> >
> > +     /* Figure-out number of bits in HGEIE */
> > +     csr_write(CSR_HGEIE, -1UL);
> > +     kvm_riscv_aia_nr_hgei = fls_long(csr_read(CSR_HGEIE));
> > +     csr_write(CSR_HGEIE, 0);
> > +     if (kvm_riscv_aia_nr_hgei)
> > +             kvm_riscv_aia_nr_hgei--;
> > +
> > +     /* Initialize guest external interrupt line management */
> > +     rc = aia_hgei_init();
> > +     if (rc)
> > +             return rc;
> > +
> >       /* Enable KVM AIA support */
> >       static_branch_enable(&kvm_riscv_aia_available);
> >
> > @@ -380,4 +616,9 @@ int kvm_riscv_aia_init(void)
> >
> >  void kvm_riscv_aia_exit(void)
> >  {
> > +     if (!kvm_riscv_aia_available())
> > +             return;
> > +
> > +     /* Cleanup the HGEI state */
> > +     aia_hgei_exit();
> >  }
> > diff --git a/arch/riscv/kvm/main.c b/arch/riscv/kvm/main.c
> > index 6396352b4e4d..b0b46f48f31e 100644
> > --- a/arch/riscv/kvm/main.c
> > +++ b/arch/riscv/kvm/main.c
> > @@ -116,7 +116,8 @@ static int __init riscv_kvm_init(void)
> >       kvm_info("VMID %ld bits available\n", kvm_riscv_gstage_vmid_bits());
> >
> >       if (kvm_riscv_aia_available())
> > -             kvm_info("AIA available\n");
> > +             kvm_info("AIA available with %d guest external interrupts\n",
> > +                      kvm_riscv_aia_nr_hgei);
> >
> >       rc = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
> >       if (rc) {
> > diff --git a/arch/riscv/kvm/vcpu.c b/arch/riscv/kvm/vcpu.c
> > index 30acf3ebdc3d..eace51dd896f 100644
> > --- a/arch/riscv/kvm/vcpu.c
> > +++ b/arch/riscv/kvm/vcpu.c
> > @@ -249,10 +249,12 @@ int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
> >
> >  void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
> >  {
> > +     kvm_riscv_aia_wakeon_hgei(vcpu, true);
> >  }
> >
> >  void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
> >  {
> > +     kvm_riscv_aia_wakeon_hgei(vcpu, false);
> >  }
> >
> >  int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
> > --
> > 2.34.1
> >
>
> Thanks,
> drew