On Tue, Dec 09, 2014 at 09:53:18AM +0100, Olivier MATZ wrote:
> Hi Neil,
>
> On 12/08/2014 04:04 PM, Neil Horman wrote:
> >On Fri, Nov 07, 2014 at 09:28:09AM -0800, Jia Yu wrote:
> >>Include rte_memory.h for lib files that use __rte_cache_aligned
> >>attribute.
> >>
> >>Signed-off-by: Jia Yu <jyu at vmware.com>
> >>
> >Why? I presume there was a build break or something. Please repost with a
> >changelog that details what this patch is for.
> >Neil
>
> I don't know if Yu's issue was the same, but I had a very "fun" issue
> with __rte_cache_aligned in my application. Consider the following code:
>
> struct per_core_foo {
> ...
> } __rte_cache_aligned;
>
> struct global_foo {
> struct per_core_foo foo[RTE_MAX_CORE];
> };
>
> If __rte_cache_aligned is not defined (rte_memory.h is not included),
> the code compiles but the structure is not aligned... it defines the
> structure and creates a global variable called __rte_cache_aligned.
> And this can lead to really bad things if this code is in a .h that
> is included by files that may or may not include rte_memory.h
>
> I have no idea about how we could prevent this issue, except using
> __attribute__((aligned(CACHE_LINE))) instead of __rte_cache_aligned.
>
> Anyway this could probably explain the willing to include rte_memory.h
> everywhere.
>
> Regards,
> Olivier
>
>
So, that is a great explination, and would be good to have in the changelog.
Also, to avoid the problem that you describe, while its preferred to have it at
the end of a struct, you can also put the alignment attribute right after the
struct keyword in gcc:
https://gcc.gnu.org/onlinedocs/gcc/Attribute-Syntax.html#Attribute-Syntax
That seems like it would solve the problem going forward.
Neil
