Hi Fidaullah,

Thanks for the fix,

Acked-by: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com>

Anatoly, I noticed a couple of other things while testing this.

1. Consider:

elt_size = pg_sz - MALLOC_ELEM_OVERHEAD

rte_malloc(align=0) which is converted to align = 1.

Obviously, such an element fits into one page, however:

alloc_sz = RTE_ALIGN_CEIL(1 + pg_sz +

                (MALLOC_ELEM_OVERHEAD - MALLOC_ELEM_OVERHEAD),

                pg_sz) == 2 * pg_sz.

This can unnecessarily hit an allocation limit from the system or EAL.

I suggest, in both places:

alloc_sz = RTE_ALIGN_CEIL(RTE_ALIGN_CEIL(elt_size, align) +
                MALLOC_ELEM_OVERHEAD, pg_sz);

This would be symmetric with malloc_elem_can_hold().

2. Alignment calculation depends on whether we allocated new pages or not:

malloc_heap_alloc_on_heap_id(align = 0) ->

heap_alloc(align = 1) ->

find_suitable_element(align = RTE_CACHE_LINE_ROUNDUP(align))

malloc_heap_alloc_on_heap_id(align = 0) ->
alloc_more_mem_on_socket(align = 1) ->
try_expand_heap() -> ... ->
alloc_pages_on_heap(align = 1) ->
find_suitable_element(align = 1)