On Tue, Jun 25, 2019 at 6:12 PM Anatoly Burakov <anatoly.bura...@intel.com>
wrote:
> Currently, whenever timer library is initialized, the memory is leaked
> because there is no telling when primary or secondary processes get
> to use the state, and there is no way to initialize/deinitialize
> timer library state without race conditions because the data itself
> must live in shared memory.
>
> However, there is now a timer library lock in the shared memory
> config, which can be used to synchronize access to the timer
> library shared memory. Use it to initialize/deinitialize timer
> library shared data in a safe way. There is still a way to leak
> the memory (by killing one of the processes), but we can't do
> anything about that.
>
> Also, update the API doc. Note that the behavior of the API
> itself did not change - the requirement to call init in every
> process was simply not documented explicitly.
>
> Signed-off-by: Erik Gabriel Carrillo <erik.g.carri...@intel.com>
> Signed-off-by: Anatoly Burakov <anatoly.bura...@intel.com>
> ---
> lib/librte_timer/rte_timer.c | 41 ++++++++++++++++++++++++------------
> lib/librte_timer/rte_timer.h | 5 +++--
> 2 files changed, 31 insertions(+), 15 deletions(-)
>
> diff --git a/lib/librte_timer/rte_timer.c b/lib/librte_timer/rte_timer.c
> index dd7953922..08517c120 100644
> --- a/lib/librte_timer/rte_timer.c
> +++ b/lib/librte_timer/rte_timer.c
> @@ -13,6 +13,7 @@
> #include <rte_atomic.h>
> #include <rte_common.h>
> #include <rte_cycles.h>
> +#include <rte_eal_memconfig.h>
> #include <rte_per_lcore.h>
> #include <rte_memory.h>
> #include <rte_launch.h>
> @@ -61,6 +62,8 @@ struct rte_timer_data {
> };
>
> #define RTE_MAX_DATA_ELS 64
> +static const struct rte_memzone *rte_timer_data_mz;
> +static int *volatile rte_timer_mz_refcnt;
> static struct rte_timer_data *rte_timer_data_arr;
> static const uint32_t default_data_id;
> static uint32_t rte_timer_subsystem_initialized;
> @@ -157,28 +160,30 @@ rte_timer_subsystem_init_v1905(void)
> int i, lcore_id;
> static const char *mz_name = "rte_timer_mz";
> const size_t data_arr_size =
> - RTE_MAX_DATA_ELS *
> sizeof(*rte_timer_data_arr);
> + RTE_MAX_DATA_ELS * sizeof(*rte_timer_data_arr);
> + const size_t mem_size = data_arr_size +
> sizeof(*rte_timer_mz_refcnt);
> bool do_full_init = true;
>
> if (rte_timer_subsystem_initialized)
> return -EALREADY;
>
> -reserve:
> - rte_errno = 0;
> - mz = rte_memzone_reserve_aligned(mz_name, data_arr_size,
> SOCKET_ID_ANY,
> - 0, RTE_CACHE_LINE_SIZE);
> + rte_mcfg_timer_lock();
> +
> + mz = rte_memzone_lookup(mz_name);
> if (mz == NULL) {
> - if (rte_errno == EEXIST) {
> - mz = rte_memzone_lookup(mz_name);
> - if (mz == NULL)
> - goto reserve;
> -
> - do_full_init = false;
> - } else
> + mz = rte_memzone_reserve_aligned(mz_name, mem_size,
> + SOCKET_ID_ANY, 0, RTE_CACHE_LINE_SIZE);
> + if (mz == NULL) {
> + rte_mcfg_timer_unlock();
> return -ENOMEM;
> - }
> + }
> + do_full_init = true;
> + } else
> + do_full_init = false;
>
> + rte_timer_data_mz = mz;
> rte_timer_data_arr = mz->addr;
> + rte_timer_mz_refcnt = (void *)((char *)mz->addr + data_arr_size);
>
> if (do_full_init) {
> for (i = 0; i < RTE_MAX_DATA_ELS; i++) {
> @@ -195,6 +200,9 @@ rte_timer_subsystem_init_v1905(void)
> }
>
> rte_timer_data_arr[default_data_id].internal_flags |= FL_ALLOCATED;
> + (*rte_timer_mz_refcnt)++;
> +
> + rte_mcfg_timer_unlock();
>
> rte_timer_subsystem_initialized = 1;
>
> @@ -210,6 +218,13 @@ rte_timer_subsystem_finalize(void)
> if (!rte_timer_subsystem_initialized)
> return;
>
> + rte_mcfg_timer_lock();
> +
> + if (--(*rte_timer_mz_refcnt) == 0)
> + rte_memzone_free(rte_timer_data_mz);
> +
> + rte_mcfg_timer_unlock();
> +
> rte_timer_subsystem_initialized = 0;
> }
>
> diff --git a/lib/librte_timer/rte_timer.h b/lib/librte_timer/rte_timer.h
> index 2196934b2..a7af10176 100644
> --- a/lib/librte_timer/rte_timer.h
> +++ b/lib/librte_timer/rte_timer.h
> @@ -170,10 +170,11 @@ int __rte_experimental
> rte_timer_data_dealloc(uint32_t id);
> * Initializes internal variables (list, locks and so on) for the RTE
> * timer library.
> *
> + * @note
> + * This function must be called in every process before using the
> library.
> + *
> * @return
> * - 0: Success
> - * - -EEXIST: Returned in secondary process when primary process has not
> - * yet initialized the timer subsystem
> * - -ENOMEM: Unable to allocate memory needed to initialize timer
> * subsystem
> */
> --
> 2.17.1
>
Can be squashed with the first patch.
--
David Marchand
