Re: vMCE vs migration

[George Dunlap <george.dunlap@citrix.com>]

On Mon, 2012-01-30 at 13:47 +0000, Jan Beulich wrote:
> >>> On 26.01.12 at 17:54, George Dunlap <george.dunlap@citrix.com> wrote:
> > On Tue, 2012-01-24 at 11:08 +0000, Jan Beulich wrote:
> >> >>> On 24.01.12 at 11:29, George Dunlap <George.Dunlap@eu.citrix.com> wrote:
> >> > On Mon, Jan 23, 2012 at 11:08 AM, Jan Beulich <JBeulich@suse.com> wrote:
> >> >> x86's vMCE implementation lets a guest know of as many MCE reporting
> >> >> banks as there are in the host. While a PV guest could be expected to
> >> >> deal with this number changing (particularly decreasing) during migration
> >> >> (not currently handled anywhere afaict), for HVM guests this is certainly
> >> >> wrong.
> >> >>
> >> >> At least to me it isn't, however, clear how to properly handle this. The
> >> >> easiest would appear to be to save and restore the number of banks
> >> >> the guest was made believe it can access, making vmce_{rd,wr}msr()
> >> >> silently tolerate accesses between the host and guest values.
> >> >
> >> > We ran into this in the XS 6.0 release as well.  I think that the
> >> > ideal thing to do would be to have a parameter that can be set at
> >> > boot, to say how many vMCE banks a guest has, defaulting to the number
> >> > of MCE banks on the host.  This parameter would be preserved across
> >> > migration.  Ideally, a pool-aware toolstack like xapi would then set
> >> > this value to be the value of the host in the pool with the largest
> >> > number of banks, allowing a guest to access all the banks on any host
> >> > to which it migrates.
> >> >
> >> > What do you think?
> >>
> >> That sounds like the way to go.
> >
> > So should we put this on IanC's to-do-be-done list?  Are you going to
> > put it on your to-do list? :-)
> 
> Below/attached a draft patch (compile tested only), handling save/
> restore of the bank count, but not allowing for a config setting to
> specify its initial value (yet).

Looks pretty good for a first blush.  Just one question: Why is the vmce
count made on a per-vcpu basis, rather than on a per-domain basis, like
the actual banks are?  Is the host MCE stuff per-vcpu?

 -George

> +static int bank_mce_rdmsr(const struct vcpu *v, uint32_t msr, uint64_t *val)
> +{
> +    int ret = 1;
> +    unsigned int bank = (msr - MSR_IA32_MC0_CTL) / 4;
> +    struct domain_mca_msrs *vmce = dom_vmce(v->domain);
>      struct bank_entry *entry;
> 
> -    bank = (msr - MSR_IA32_MC0_CTL) / 4;
> -    if ( bank >= nr_mce_banks )
> -        return -1;
> +    *val = 0;
> 
>      switch ( msr & (MSR_IA32_MC0_CTL | 3) )
>      {
>      case MSR_IA32_MC0_CTL:
> -        *val = vmce->mci_ctl[bank] &
> -            (h_mci_ctrl ? h_mci_ctrl[bank] : ~0UL);
> +        if ( bank < nr_mce_banks )
> +            *val = vmce->mci_ctl[bank] &
> +                   (h_mci_ctrl ? h_mci_ctrl[bank] : ~0UL);
>          mce_printk(MCE_VERBOSE, "MCE: rdmsr MC%u_CTL 0x%"PRIx64"\n",
>                     bank, *val);
>          break;
> @@ -126,7 +132,7 @@ static int bank_mce_rdmsr(struct domain
>          switch ( boot_cpu_data.x86_vendor )
>          {
>          case X86_VENDOR_INTEL:
> -            ret = intel_mce_rdmsr(msr, val);
> +            ret = intel_mce_rdmsr(v, msr, val);
>              break;
>          default:
>              ret = 0;
> @@ -145,13 +151,13 @@ static int bank_mce_rdmsr(struct domain
>   */
>  int vmce_rdmsr(uint32_t msr, uint64_t *val)
>  {
> -    struct domain *d = current->domain;
> -    struct domain_mca_msrs *vmce = dom_vmce(d);
> +    const struct vcpu *cur = current;
> +    struct domain_mca_msrs *vmce = dom_vmce(cur->domain);
>      int ret = 1;
> 
>      *val = 0;
> 
> -    spin_lock(&dom_vmce(d)->lock);
> +    spin_lock(&vmce->lock);
> 
>      switch ( msr )
>      {
> @@ -173,28 +179,26 @@ int vmce_rdmsr(uint32_t msr, uint64_t *v
>                     *val);
>          break;
>      default:
> -        ret = mce_bank_msr(msr) ? bank_mce_rdmsr(d, msr, val) : 0;
> +        ret = mce_bank_msr(cur, msr) ? bank_mce_rdmsr(cur, msr, val) : 0;
>          break;
>      }
> 
> -    spin_unlock(&dom_vmce(d)->lock);
> +    spin_unlock(&vmce->lock);
>      return ret;
>  }
> 
> -static int bank_mce_wrmsr(struct domain *d, u32 msr, u64 val)
> +static int bank_mce_wrmsr(struct vcpu *v, u32 msr, u64 val)
>  {
> -    int bank, ret = 1;
> -    struct domain_mca_msrs *vmce = dom_vmce(d);
> +    int ret = 1;
> +    unsigned int bank = (msr - MSR_IA32_MC0_CTL) / 4;
> +    struct domain_mca_msrs *vmce = dom_vmce(v->domain);
>      struct bank_entry *entry = NULL;
> 
> -    bank = (msr - MSR_IA32_MC0_CTL) / 4;
> -    if ( bank >= nr_mce_banks )
> -        return -EINVAL;
> -
>      switch ( msr & (MSR_IA32_MC0_CTL | 3) )
>      {
>      case MSR_IA32_MC0_CTL:
> -        vmce->mci_ctl[bank] = val;
> +        if ( bank < nr_mce_banks )
> +            vmce->mci_ctl[bank] = val;
>          break;
>      case MSR_IA32_MC0_STATUS:
>          /* Give the first entry of the list, it corresponds to current
> @@ -202,9 +206,9 @@ static int bank_mce_wrmsr(struct domain
>           * the guest, this node will be deleted.
>           * Only error bank is written. Non-error banks simply return.
>           */
> -        if ( !list_empty(&dom_vmce(d)->impact_header) )
> +        if ( !list_empty(&vmce->impact_header) )
>          {
> -            entry = list_entry(dom_vmce(d)->impact_header.next,
> +            entry = list_entry(vmce->impact_header.next,
>                                 struct bank_entry, list);
>              if ( entry->bank == bank )
>                  entry->mci_status = val;
> @@ -228,7 +232,7 @@ static int bank_mce_wrmsr(struct domain
>          switch ( boot_cpu_data.x86_vendor )
>          {
>          case X86_VENDOR_INTEL:
> -            ret = intel_mce_wrmsr(msr, val);
> +            ret = intel_mce_wrmsr(v, msr, val);
>              break;
>          default:
>              ret = 0;
> @@ -247,9 +251,9 @@ static int bank_mce_wrmsr(struct domain
>   */
>  int vmce_wrmsr(u32 msr, u64 val)
>  {
> -    struct domain *d = current->domain;
> +    struct vcpu *cur = current;
>      struct bank_entry *entry = NULL;
> -    struct domain_mca_msrs *vmce = dom_vmce(d);
> +    struct domain_mca_msrs *vmce = dom_vmce(cur->domain);
>      int ret = 1;
> 
>      if ( !g_mcg_cap )
> @@ -266,7 +270,7 @@ int vmce_wrmsr(u32 msr, u64 val)
>          vmce->mcg_status = val;
>          mce_printk(MCE_VERBOSE, "MCE: wrmsr MCG_STATUS %"PRIx64"\n", val);
>          /* For HVM guest, this is the point for deleting vMCE injection node */
> -        if ( d->is_hvm && (vmce->nr_injection > 0) )
> +        if ( is_hvm_vcpu(cur) && (vmce->nr_injection > 0) )
>          {
>              vmce->nr_injection--; /* Should be 0 */
>              if ( !list_empty(&vmce->impact_header) )
> @@ -293,7 +297,7 @@ int vmce_wrmsr(u32 msr, u64 val)
>          ret = -1;
>          break;
>      default:
> -        ret = mce_bank_msr(msr) ? bank_mce_wrmsr(d, msr, val) : 0;
> +        ret = mce_bank_msr(cur, msr) ? bank_mce_wrmsr(cur, msr, val) : 0;
>          break;
>      }
> 
> @@ -301,6 +305,47 @@ int vmce_wrmsr(u32 msr, u64 val)
>      return ret;
>  }
> 
> +static int vmce_save_vcpu_ctxt(struct domain *d, hvm_domain_context_t *h)
> +{
> +    struct vcpu *v;
> +    int err = 0;
> +
> +    for_each_vcpu( d, v ) {
> +        struct hvm_vmce_vcpu ctxt = {
> +            .nr_banks = v->arch.nr_vmce_banks
> +        };
> +
> +        err = hvm_save_entry(VMCE_VCPU, v->vcpu_id, h, &ctxt);
> +        if ( err )
> +            break;
> +    }
> +
> +    return err;
> +}
> +
> +static int vmce_load_vcpu_ctxt(struct domain *d, hvm_domain_context_t *h)
> +{
> +    unsigned int vcpuid = hvm_load_instance(h);
> +    struct vcpu *v;
> +    struct hvm_vmce_vcpu ctxt;
> +    int err;
> +
> +    if ( vcpuid >= d->max_vcpus || (v = d->vcpu[vcpuid]) == NULL )
> +    {
> +        gdprintk(XENLOG_ERR, "HVM restore: domain has no vcpu %u\n", vcpuid);
> +        return -EINVAL;
> +    }
> +
> +    err = hvm_load_entry(VMCE_VCPU, h, &ctxt);
> +    if ( !err )
> +        v->arch.nr_vmce_banks = ctxt.nr_banks;
> +
> +    return err;
> +}
> +
> +HVM_REGISTER_SAVE_RESTORE(VMCE_VCPU, vmce_save_vcpu_ctxt,
> +                          vmce_load_vcpu_ctxt, 1, HVMSR_PER_VCPU);
> +
>  int inject_vmce(struct domain *d)
>  {
>      int cpu = smp_processor_id();
> --- a/xen/arch/x86/domain.c
> +++ b/xen/arch/x86/domain.c
> @@ -468,6 +468,8 @@ int vcpu_initialise(struct vcpu *v)
> 
>      v->arch.pv_vcpu.ctrlreg[4] = real_cr4_to_pv_guest_cr4(mmu_cr4_features);
> 
> +    vmce_init_vcpu(v);
> +
>      rc = is_pv_32on64_vcpu(v) ? setup_compat_l4(v) : 0;
>   done:
>      if ( rc )
> --- a/xen/arch/x86/domctl.c
> +++ b/xen/arch/x86/domctl.c
> @@ -1027,11 +1027,12 @@ long arch_do_domctl(
>                  evc->syscall32_callback_eip    = 0;
>                  evc->syscall32_disables_events = 0;
>              }
> +            evc->nr_mce_banks = v->arch.nr_vmce_banks;
>          }
>          else
>          {
>              ret = -EINVAL;
> -            if ( evc->size != sizeof(*evc) )
> +            if ( evc->size < offsetof(typeof(*evc), nr_mce_banks) )
>                  goto ext_vcpucontext_out;
>  #ifdef __x86_64__
>              if ( !is_hvm_domain(d) )
> @@ -1059,6 +1060,16 @@ long arch_do_domctl(
>                   (evc->syscall32_callback_cs & ~3) ||
>                   evc->syscall32_callback_eip )
>                  goto ext_vcpucontext_out;
> +
> +            if ( evc->size >= offsetof(typeof(*evc), mbz[ARRAY_SIZE(evc->mbz)]) )
> +            {
> +                unsigned int i;
> +
> +                for ( i = 0; i < ARRAY_SIZE(evc->mbz); ++i )
> +                    if ( evc->mbz[i] )
> +                        goto ext_vcpucontext_out;
> +                v->arch.nr_vmce_banks = evc->nr_mce_banks;
> +            }
>          }
> 
>          ret = 0;
> --- a/xen/include/asm-x86/domain.h
> +++ b/xen/include/asm-x86/domain.h
> @@ -488,6 +488,8 @@ struct arch_vcpu
>      /* This variable determines whether nonlazy extended state has been used,
>       * and thus should be saved/restored. */
>      bool_t nonlazy_xstate_used;
> +
> +    uint8_t nr_vmce_banks;
> 
>      struct paging_vcpu paging;
> 
> --- a/xen/include/asm-x86/mce.h
> +++ b/xen/include/asm-x86/mce.h
> @@ -28,6 +28,7 @@ struct domain_mca_msrs
>  /* Guest vMCE MSRs virtualization */
>  extern int vmce_init_msr(struct domain *d);
>  extern void vmce_destroy_msr(struct domain *d);
> +extern void vmce_init_vcpu(struct vcpu *);
>  extern int vmce_wrmsr(uint32_t msr, uint64_t val);
>  extern int vmce_rdmsr(uint32_t msr, uint64_t *val);
> 
> --- a/xen/include/public/arch-x86/hvm/save.h
> +++ b/xen/include/public/arch-x86/hvm/save.h
> @@ -585,9 +585,15 @@ struct hvm_viridian_vcpu_context {
> 
>  DECLARE_HVM_SAVE_TYPE(VIRIDIAN_VCPU, 17, struct hvm_viridian_vcpu_context);
> 
> +struct hvm_vmce_vcpu {
> +    uint8_t nr_banks;
> +};
> +
> +DECLARE_HVM_SAVE_TYPE(VMCE_VCPU, 18, struct hvm_vmce_vcpu);
> +
>  /*
>   * Largest type-code in use
>   */
> -#define HVM_SAVE_CODE_MAX 17
> +#define HVM_SAVE_CODE_MAX 18
> 
>  #endif /* __XEN_PUBLIC_HVM_SAVE_X86_H__ */
> --- a/xen/include/public/domctl.h
> +++ b/xen/include/public/domctl.h
> @@ -559,7 +559,7 @@ struct xen_domctl_ext_vcpucontext {
>      uint32_t         vcpu;
>      /*
>       * SET: Size of struct (IN)
> -     * GET: Size of struct (OUT)
> +     * GET: Size of struct (OUT, up to 128 bytes)
>       */
>      uint32_t         size;
>  #if defined(__i386__) || defined(__x86_64__)
> @@ -571,6 +571,10 @@ struct xen_domctl_ext_vcpucontext {
>      uint16_t         sysenter_callback_cs;
>      uint8_t          syscall32_disables_events;
>      uint8_t          sysenter_disables_events;
> +    uint8_t          nr_mce_banks;
> +    /* This array can be split and used for future extension. */
> +    /* It is there just to grow the structure beyond 4.1's size. */
> +    uint8_t          mbz[9];
>  #endif
>  };
>  typedef struct xen_domctl_ext_vcpucontext xen_domctl_ext_vcpucontext_t;
>