Re: [PATCH 6/8] xen/x86: Avoid overriding initialisers in arrays

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

On 10/02/16 14:03, Jan Beulich wrote:
>>>> On 10.02.16 at 14:50, <andrew.cooper3@citrix.com> wrote:
>> On 10/02/16 13:22, Jan Beulich wrote:
>>>>>> On 09.02.16 at 21:01, <andrew.cooper3@citrix.com> wrote:
>>>> Clang objects to having multiple initialisers when creating an array.
>>>>
>>>> As this warning is useful for spotting obscure bugs, disabling it is
>>>> unhelpful.  Instead, fix our two deliberate usecases.
>>> Ugly again, but - well ...
>>>
>>>> --- a/xen/arch/x86/mm/p2m-ept.c
>>>> +++ b/xen/arch/x86/mm/p2m-ept.c
>>>> @@ -1201,6 +1201,20 @@ void ept_p2m_uninit(struct p2m_domain *p2m)
>>>>      free_cpumask_var(ept->invalidate);
>>>>  }
>>>>  
>>>> +static const char *memory_type_to_str(unsigned int x)
>>>> +{
>>>> +    static const char memory_types[8][2] = {
>>>> +        [MTRR_TYPE_UNCACHABLE]     = "UC",
>>>> +        [MTRR_TYPE_WRCOMB]         = "WC",
>>>> +        [MTRR_TYPE_WRTHROUGH]      = "WT",
>>>> +        [MTRR_TYPE_WRPROT]         = "WP",
>>>> +        [MTRR_TYPE_WRBACK]         = "WB",
>>>> +        [MTRR_NUM_TYPES]           = "??"
>>>> +    };
>>>> +
>>>> +    return x < ARRAY_SIZE(memory_types) ? (memory_types[x] ?: "?") : "?";
>>> I think this should really ASSERT() the first condition.
>>>
>>>> @@ -1212,15 +1226,6 @@ static void ept_dump_p2m_table(unsigned char key)
>>>>      unsigned long record_counter = 0;
>>>>      struct p2m_domain *p2m;
>>>>      struct ept_data *ept;
>>>> -    static const char memory_types[8][2] = {
>>>> -        [0 ... 7] = "?",
>>>> -        [MTRR_TYPE_UNCACHABLE]     = "UC",
>>>> -        [MTRR_TYPE_WRCOMB]         = "WC",
>>>> -        [MTRR_TYPE_WRTHROUGH]      = "WT",
>>>> -        [MTRR_TYPE_WRPROT]         = "WP",
>>>> -        [MTRR_TYPE_WRBACK]         = "WB",
>>>> -        [MTRR_NUM_TYPES]           = "??"
>>>> -    };
>>>>  
>>>>      for_each_domain(d)
>>>>      {
>>>> @@ -1260,8 +1265,8 @@ static void ept_dump_p2m_table(unsigned char key)
>>>>                             ept_entry->r ? 'r' : ' ',
>>>>                             ept_entry->w ? 'w' : ' ',
>>>>                             ept_entry->x ? 'x' : ' ',
>>>> -                           memory_types[ept_entry->emt][0],
>>>> -                           memory_types[ept_entry->emt][1]
>>>> +                           memory_type_to_str(ept_entry->emt)[0],
>>>> +                           memory_type_to_str(ept_entry->emt)[1]
>>>>                             ?: ept_entry->emt + '0',
>>>>                             c ?: ept_entry->ipat ? '!' : ' ');
>>> There's actually a bug here, which I think is worth fixing at once:
>>> The default initializer was a string of length 1, resulting in a
>>> premature NUL character to get placed into the fully expanded
>>> string, causing - afaict - truncation of the intended message. I
>>> therefore think the default string should be e.g. "? ".
>>
>> The code is very opaque.  However, that appears to be precisely how it
>> is intended to work.  (Having said that - it is your code from c/s
>> 90e9c95f).
> 
> I know.
> 
>> The following line will only format the raw emt value as a number if
>> there is a NUL character returned from memory_type_to_str().  Putting a
>> space in instead would break this.
> 
> Oh, right - this is the operand to a ?:, not by itself passed to
> printk(). Line breaks like this (to aid people with old editors) are
> really undesirable in places like this...

Even more so over-clever undocumented code.  If you're going to do
things like this, you need to leave a comment near the string definition
saying that the second byte being NULL is a flag for the printing
routine to print the number, so that people who come along later (maybe
even yourself, as in this case) know there's a dependency there.

 -George