[CRIU] [PATCH 1/6] Replace PAGE_SIZE in vaddr_to_pfn

[Pavel Emelyanov]

On 10/08/2014 06:58 PM, Cyrill Gorcunov wrote:
> On Wed, Oct 08, 2014 at 10:42:29AM -0400, Christopher Covington wrote:
>>>
>>> or something like that. SO we don't have to change PAGE_SIZE all over
>>> the criu code at all, but in one place only. Hmm? What do you think?
>>
>> I did try that and I'd like to end up with something mostly equivalent.
>> However, some existing code assumes PAGE_SIZE is constant, using it for
>> compile-time buffer sizes. My instinct is to not turn all of these into
>> run-time allocations.
>>
>> (As a brief aside, I it's a reasonable assumption to think PAGE_SIZE or
>> ANY_MACRO_REALLY is a compile-time constant. In my opinion, using a function
>> such as sysconf(_SC_PAGESIZE) or lowercase variable name such as page_size
>> best communicates to those reading and writing the common code that the value
>> cannot be relied upon to be a compile time constant across all systems.)
>>
>> Here are the different possible uses of PAGE_SIZE and corresponding
>> replacements that I came up with:
>>
>> PAGE_OR_LESS -- (local) compile-time constant minimum page size, guaranteed to
>> be less than or equal to one page but not necessarily page aligned
>>
>> EXEC_PAGESIZE -- (libc) compile-time constant maximum page size, guaranteed to
>> be greater than or equal to one page and page aligned
>>
>> sysconf(_SC_PAGESIZE) or page_size -- (libc or local) run-time determined
>> value of the current page size
>>
>> (And maybe also something like BUFFER_SIZE, if I run into uses that really
>> have no functional or performance connection to the page size:
>> http://lists.busybox.net/pipermail/busybox/2011-June/075864.html)
> 
> I see. Indeed we've a number of places where PAGE_SIZE is assumed to be
> a constant expanded at compile time. I like the idea of introducing
> these variables. Pavel?

I've grep-ed around all the '\[PAGE_SIZE' lines. It looks like there's only
one place that requires a page -- in memory restorer. All other places just
declare a buffer big enough to fit what it needs (log message, proc file line,
etc.).  That said, yes I agree that we might need another set of constants
for such cases, but in all but one of them has nothing to do with a page :)

Thanks,
Pavel