On Tue, 11 Aug 2020 20:13:24 +0000
"Eads, Gage" <gage.e...@intel.com> wrote:
> Hi Steven,
>
> > -----Original Message-----
> > From: Steven Lariau <steven.lar...@arm.com>
> > Sent: Wednesday, August 5, 2020 10:57 AM
> > To: Eads, Gage <gage.e...@intel.com>; Olivier Matz
> > <olivier.m...@6wind.com>
> > Cc: dev@dpdk.org; honnappa.nagaraha...@arm.com;
> > dharmik.thak...@arm.com; n...@arm.com; Steven Lariau
> > <steven.lar...@arm.com>
> > Subject: [PATCH 1/4] test/stack: avoid trivial memory allocations
> >
> > Replace the arguments array by one argument.
> > All objects in the args array have the same values, so there is no need
> > to use an array, only one struct is enough.
> > The args object is a lot smaller, and the allocation can be replaced
> > with a stack variable.
> >
> > The allocation of obj_table isn't needed either, because MAX_BULK is
> > small. The allocation can instead be replaced with a static array.
> >
> > Signed-off-by: Steven Lariau <steven.lar...@arm.com>
> > Reviewed-by: Dharmik Thakkar <dharmik.thak...@arm.com>
> > Reviewed-by: Phil Yang <phil.y...@arm.com>
> > Reviewed-by: Ruifeng Wang <ruifeng.w...@arm.com>
> > ---
> > app/test/test_stack.c | 39 ++++++---------------------------------
> > 1 file changed, 6 insertions(+), 33 deletions(-)
> >
> > diff --git a/app/test/test_stack.c b/app/test/test_stack.c
> > index c8dac1f55..5a7273a7d 100644
> > --- a/app/test/test_stack.c
> > +++ b/app/test/test_stack.c
> > @@ -280,16 +280,9 @@ static int
> > stack_thread_push_pop(void *args)
> > {
> > struct test_args *t = args;
> > - void **obj_table;
> > + void *obj_table[MAX_BULK];
> > int i;
> >
> > - obj_table = rte_calloc(NULL, STACK_SIZE, sizeof(void *), 0);
> > - if (obj_table == NULL) {
> > - printf("[%s():%u] failed to calloc %zu bytes\n",
> > - __func__, __LINE__, STACK_SIZE * sizeof(void *));
> > - return -1;
> > - }
> > -
> > for (i = 0; i < NUM_ITERS_PER_THREAD; i++) {
> > unsigned int success, num;
> >
> > @@ -310,28 +303,25 @@ stack_thread_push_pop(void *args)
> > if (rte_stack_push(t->s, obj_table, num) != num) {
> > printf("[%s():%u] Failed to push %u pointers\n",
> > __func__, __LINE__, num);
> > - rte_free(obj_table);
> > return -1;
> > }
> >
> > if (rte_stack_pop(t->s, obj_table, num) != num) {
> > printf("[%s():%u] Failed to pop %u pointers\n",
> > __func__, __LINE__, num);
> > - rte_free(obj_table);
> > return -1;
> > }
> >
> > rte_atomic64_sub(t->sz, num);
> > }
> >
> > - rte_free(obj_table);
> > return 0;
> > }
>
> Agreed, the dynamic allocation is unnecessary.
>
> >
> > static int
> > test_stack_multithreaded(uint32_t flags)
> > {
> > - struct test_args *args;
> > + struct test_args args;
> > unsigned int lcore_id;
> > struct rte_stack *s;
> > rte_atomic64_t size;
> > @@ -344,45 +334,28 @@ test_stack_multithreaded(uint32_t flags)
> > printf("[%s():%u] Running with %u lcores\n",
> > __func__, __LINE__, rte_lcore_count());
> >
> > - args = rte_malloc(NULL, sizeof(struct test_args) * RTE_MAX_LCORE,
> > 0);
> > - if (args == NULL) {
> > - printf("[%s():%u] failed to malloc %zu bytes\n",
> > - __func__, __LINE__,
> > - sizeof(struct test_args) * RTE_MAX_LCORE);
> > - return -1;
> > - }
> > -
> > s = rte_stack_create("test", STACK_SIZE, rte_socket_id(), flags);
> > if (s == NULL) {
> > printf("[%s():%u] Failed to create a stack\n",
> > __func__, __LINE__);
> > - rte_free(args);
> > return -1;
> > }
> >
> > rte_atomic64_init(&size);
> > + args.s = s;
> > + args.sz = &size;
> >
> > RTE_LCORE_FOREACH_SLAVE(lcore_id) {
> > - args[lcore_id].s = s;
> > - args[lcore_id].sz = &size;
> > -
> > if (rte_eal_remote_launch(stack_thread_push_pop,
> > - &args[lcore_id], lcore_id))
> > + &args, lcore_id))
> > rte_panic("Failed to launch lcore %d\n", lcore_id);
> > }
>
>
> In general we shouldn't pass a stack variable to other threads. Though your
> code here looks fine, I'd rather err on the safe side in case this is ever
> used
> as a template/basis for some other code...particularly since there's no
> performance/correctness/etc. penalty to using dynamically allocated memory.
>
> To support patch 2/4, you can instead convert the rte_malloc to allocate a
> single shared test_args structure. Or perhaps move patch 4 earlier in the
> series,
> and simply pass the stack pointer instead.
>
> Thanks,
> Gage
There is no gain to using rte_malloc unless you are doing primary/secondary
process
or trying to test rte_malloc. Why not use regular malloc which has good tools
and library support.
