[PATCH v2] eal: change alloc_sz calculation which may cause unnecessarily allocation

[Dmitry Kozlyuk]

2022-09-21 15:35 (UTC+0200), David Marchand:
> On Thu, Jul 28, 2022 at 8:03 PM Dmitry Kozlyuk <dmitry.kozliuk at gmail.com> wrote:
> >
> > 2022-07-28 14:41 (UTC+0500), Fidaullah Noonari:  
> > > The amount of memory to allocate from the system for heap expansion
> > > was calculated in a way that may yield one page more than needed.
> > > This could hit the allocation limit from the system or EAL.
> > > The allocation would fail despite enough memory being available.
> > >
> > > In response to mail:
> > > http://inbox.dpdk.org/dev/CAEYuUWCnRZNwxiOHEeTHw0Gy9aFJRLZtvAG9g=smuUvUEMcFXg@mail.gmail.com/
> > >
> > > Signed-off-by: Fidaullah Noonari <fidaullah.noonari at emumba.com>  
> >
> > Acked-by: Dmitry Kozlyuk <dmitry.kozliuk at gmail.com>
> >  
> 
> Do you have a reproducer?
> This sounds like a fix, should we backport it?

We should, thanks for the catch.

Reproducer:

#include <rte_eal.h>
#include <rte_malloc.h>
#include <stdio.h>

int
main(int argc, char **argv)
{
	static const size_t hugepage_sz = 2 * (1 << 20);
	static const size_t malloc_elem_overhead = 128;
	static const size_t size = hugepage_sz - malloc_elem_overhead;
	struct rte_malloc_socket_stats stats_before, stats_after;
	size_t alloc_delta;
	int ret;

	ret = rte_eal_init(argc, argv);
	if (ret < 0)
		rte_panic("Cannot init EAL\n");
	rte_malloc_get_socket_stats(0, &stats_before);
	(void)rte_malloc(NULL, size, 0);
	rte_malloc_get_socket_stats(0, &stats_after);
	alloc_delta = stats_after.heap_totalsz_bytes - stats_before.heap_totalsz_bytes;
	puts(alloc_delta == hugepage_sz ? "fixed" : "affected");
	rte_eal_cleanup();
	return 0;
}