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

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

On 12/09/18 10:46, Sergey Dyasli wrote:
> On Wed, 2018-09-12 at 10:12 +0100, Andrew Cooper wrote:
>> 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.
> Actually, the current VMX/SVM (but not PV) code does:
>
>         switch ( wrmsr_hypervisor_regs(msr, msr_content) )
>         {
>         case -ERESTART:
>             return X86EMUL_RETRY;
>
> This code is removed in 1/3 patch but I wasn't CCed.

Ah right, in which case I need to temporarily transplant this switch
into patch 1.  Given its only the PV side which is then broken, I can
probably see about doing a bugfix for that.

~Andrew

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