Re: [PATCH 3/9] x86/vmx: Factor locate_msr_entry() out of vmx_find_msr() and vmx_add_msr()

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

On 24/05/18 11:53, Roger Pau Monné wrote:
> On Wed, May 23, 2018 at 05:55:50PM +0100, Andrew Cooper wrote:
>> On 23/05/18 17:39, Roger Pau Monné wrote:
>>> On Tue, May 22, 2018 at 12:20:40PM +0100, Andrew Cooper wrote:
>>>> Instead of having multiple algorithms searching the MSR lists, implement a
>>>> single one.  It has the semantics required by vmx_add_msr(), to identify the
>>>> position in which an MSR should live, if it isn't already present.
>>>>
>>>> There will be a marginal improvement for vmx_find_msr() by avoiding the
>>>> function pointer calls to vmx_msr_entry_key_cmp(), and a major improvement for
>>>> vmx_add_msr() by using a binary search instead of a linear search.
>>>>
>>>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
>>>> ---
>>>> CC: Jan Beulich <JBeulich@suse.com>
>>>> CC: Jun Nakajima <jun.nakajima@intel.com>
>>>> CC: Kevin Tian <kevin.tian@intel.com>
>>>> CC: Wei Liu <wei.liu2@citrix.com>
>>>> CC: Roger Pau Monné <roger.pau@citrix.com>
>>>> ---
>>>>  xen/arch/x86/hvm/vmx/vmcs.c | 42 ++++++++++++++++++++++++++++--------------
>>>>  1 file changed, 28 insertions(+), 14 deletions(-)
>>>>
>>>> diff --git a/xen/arch/x86/hvm/vmx/vmcs.c b/xen/arch/x86/hvm/vmx/vmcs.c
>>>> index f557857..e4acdc1 100644
>>>> --- a/xen/arch/x86/hvm/vmx/vmcs.c
>>>> +++ b/xen/arch/x86/hvm/vmx/vmcs.c
>>>> @@ -1276,24 +1276,36 @@ static int construct_vmcs(struct vcpu *v)
>>>>      return 0;
>>>>  }
>>>>  
>>>> -static int vmx_msr_entry_key_cmp(const void *key, const void *elt)
>>>> +/*
>>>> + * Search an MSR list looking for an MSR entry, or the slot in which it should
>>>> + * live (to keep the data sorted) if an entry is not found.
>>>> + *
>>>> + * The return pointer is guarenteed to be bounded by start and end.  However,
>>>> + * it may point at end, and may be invalid for the caller to dereference.
>>>> + */
>>>> +static struct vmx_msr_entry *locate_msr_entry(
>>>> +    struct vmx_msr_entry *start, struct vmx_msr_entry *end, uint32_t msr)
>>>>  {
>>>> -    const u32 *msr = key;
>>>> -    const struct vmx_msr_entry *entry = elt;
>>>> +    while ( start < end )
>>>> +    {
>>>> +        struct vmx_msr_entry *mid = start + (end - start) / 2;
>>>>  
>>>> -    if ( *msr > entry->index )
>>>> -        return 1;
>>>> -    if ( *msr < entry->index )
>>>> -        return -1;
>>>> +        if ( msr < mid->index )
>>>> +            end = mid;
>>>> +        else if ( msr > mid->index )
>>>> +            start = mid + 1;
>>>> +        else
>>>> +            return mid;
>>>> +    }
>>> This is basically an open coded version of bsearch, isn't there anyway
>>> to adapt the current bsearch so that it could be used for both
>>> vmx_find_msr and vmx_add_msr?
>>>
>>> I know there will be a performance penalty for using a function
>>> pointer for the comparator function, but this looks like code
>>> duplication to me.
>> A third use appears in a later patch.  bsearch() doesn't have the
>> described property on a miss, which is necessary to maintain the lists.
> I would consider adding a flag to the list of parameters so that
> bsearch returned the position where the item should be added in case
> of a miss. You could then wrap it inside of locate_msr_entry, or get
> rid of this helper altogether.

bsearch() is specified by POSIX, and C89/99, amongst other standards. 
Changing its API is not something I'm going to do.

~Andrew

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