Re: [PATCH v3 3/3] x86: Clean up the Xen MSR infrastructure

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 12/09/18 09:29, Sergey Dyasli wrote:
> On Tue, 2018-09-11 at 19:56 +0100, Andrew Cooper wrote:
>> Rename them to guest_{rd,wr}msr_xen() for consistency, and because the _regs
>> suffix isn't very appropriate.
>>
>> Update them to take a vcpu pointer rather than presuming that they act on
>> current, and switch to using X86EMUL_* return values.
>>
>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
>> ---
>> CC: Jan Beulich <JBeulich@suse.com>
>> CC: Wei Liu <wei.liu2@citrix.com>
>> CC: Roger Pau Monné <roger.pau@citrix.com>
>> CC: Sergey Dyasli <sergey.dyasli@citrix.com>
>>
>> v3:
>>  * Clean up after splitting the series.
>> ---
>>  xen/arch/x86/msr.c              |  6 ++----
>>  xen/arch/x86/traps.c            | 29 +++++++++++++----------------
>>  xen/include/asm-x86/processor.h |  4 ++--
>>  3 files changed, 17 insertions(+), 22 deletions(-)
>>
>> diff --git a/xen/arch/x86/msr.c b/xen/arch/x86/msr.c
>> index cf0dc27..8f02a89 100644
>> --- a/xen/arch/x86/msr.c
>> +++ b/xen/arch/x86/msr.c
>> @@ -156,8 +156,7 @@ int guest_rdmsr(const struct vcpu *v, uint32_t msr, uint64_t *val)
>>  
>>          /* Fallthrough. */
>>      case 0x40000200 ... 0x400002ff:
>> -        ret = (rdmsr_hypervisor_regs(msr, val)
>> -               ? X86EMUL_OKAY : X86EMUL_EXCEPTION);
>> +        ret = guest_rdmsr_xen(v, msr, val);
>>          break;
>>  
>>      default:
>> @@ -277,8 +276,7 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
>>  
>>          /* Fallthrough. */
>>      case 0x40000200 ... 0x400002ff:
>> -        ret = (wrmsr_hypervisor_regs(msr, val) == 1
>> -               ? X86EMUL_OKAY : X86EMUL_EXCEPTION);
>> +        ret = guest_wrmsr_xen(v, msr, val);
>>          break;
>>  
>>      default:
>> diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
>> index 7c17806..3988753 100644
>> --- a/xen/arch/x86/traps.c
>> +++ b/xen/arch/x86/traps.c
>> @@ -768,29 +768,25 @@ static void do_trap(struct cpu_user_regs *regs)
>>            trapnr, trapstr(trapnr), regs->error_code);
>>  }
>>  
>> -/* Returns 0 if not handled, and non-0 for success. */
>> -int rdmsr_hypervisor_regs(uint32_t idx, uint64_t *val)
>> +int guest_rdmsr_xen(const struct vcpu *v, uint32_t idx, uint64_t *val)
>>  {
>> -    struct domain *d = current->domain;
>> +    const struct domain *d = v->domain;
>>      /* Optionally shift out of the way of Viridian architectural MSRs. */
>>      uint32_t base = is_viridian_domain(d) ? 0x40000200 : 0x40000000;
>>  
>>      switch ( idx - base )
>>      {
>>      case 0: /* Write hypercall page MSR.  Read as zero. */
>> -    {
>>          *val = 0;
>> -        return 1;
>> -    }
>> +        return X86EMUL_OKAY;
>>      }
>>  
>> -    return 0;
>> +    return X86EMUL_EXCEPTION;
>>  }
>>  
>> -/* Returns 1 if handled, 0 if not and -Exx for error. */
>> -int wrmsr_hypervisor_regs(uint32_t idx, uint64_t val)
>> +int guest_wrmsr_xen(struct vcpu *v, uint32_t idx, uint64_t val)
>>  {
>> -    struct domain *d = current->domain;
>> +    struct domain *d = v->domain;
>>      /* Optionally shift out of the way of Viridian architectural MSRs. */
>>      uint32_t base = is_viridian_domain(d) ? 0x40000200 : 0x40000000;
>>  
>> @@ -809,7 +805,7 @@ int wrmsr_hypervisor_regs(uint32_t idx, uint64_t val)
>>              gdprintk(XENLOG_WARNING,
>>                       "wrmsr hypercall page index %#x unsupported\n",
>>                       page_index);
>> -            return 0;
>> +            return X86EMUL_EXCEPTION;
>>          }
>>  
>>          page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
>> @@ -822,13 +818,13 @@ int wrmsr_hypervisor_regs(uint32_t idx, uint64_t val)
>>              if ( p2m_is_paging(t) )
>>              {
>>                  p2m_mem_paging_populate(d, gmfn);
>> -                return -ERESTART;
>> +                return X86EMUL_RETRY;
> Previously -ERESTART would've been converted to X86EMUL_EXCEPTION. But
> with this patch, X86EMUL_RETRY will actually be returned. I don't think
> that callers can handle this situation.
>
> E.g. the code from vmx_vmexit_handler():
>
>     case EXIT_REASON_MSR_WRITE:
>         switch ( hvm_msr_write_intercept(regs->ecx, msr_fold(regs), 1) )
>         {
>         case X86EMUL_OKAY:
>             update_guest_eip(); /* Safe: WRMSR */
>             break;
>
>         case X86EMUL_EXCEPTION:
>             hvm_inject_hw_exception(TRAP_gp_fault, 0);
>             break;
>         }
>         break;

Hmm lovely, so it was broken before, but should be correct now.

RETRY has caused an entry to go onto the paging ring, which will pause
the vcpu until a reply occurs, after which we will re-enter the guest
without having moved RIP forwards, re-execute the wrmsr instruction, and
this time succeed because the frame has been paged in.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel