Re: [PATCH v6 1/3] eal: add rte thread create control API

[Tyler Retzlaff <roretzla@linux.microsoft.com>]

On Fri, Feb 24, 2023 at 09:13:56AM +0100, David Marchand wrote:
> On Wed, Feb 8, 2023 at 10:26 PM Tyler Retzlaff
> <roretzla@linux.microsoft.com> wrote:
> >
> > Add rte_thread_create_control API as a replacement for
> > rte_ctrl_thread_create to allow deprecation of the use of platform
> > specific types in DPDK public API.
> >
> > Add test from David Marchand to exercise the new API.
> >
> > Signed-off-by: Tyler Retzlaff <roretzla@linux.microsoft.com>
> > Acked-by: Morten Brørup <mb@smartsharesystems.com>
> > Reviewed-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
> > ---
> >  app/test/test_threads.c            | 26 ++++++++++++
> >  lib/eal/common/eal_common_thread.c | 85 ++++++++++++++++++++++++++++++++++----
> >  lib/eal/include/rte_thread.h       | 33 +++++++++++++++
> >  lib/eal/version.map                |  1 +
> >  4 files changed, 137 insertions(+), 8 deletions(-)
> >
> > diff --git a/app/test/test_threads.c b/app/test/test_threads.c
> > index e0f18e4..657ecad 100644
> > --- a/app/test/test_threads.c
> > +++ b/app/test/test_threads.c
> > @@ -232,6 +232,31 @@
> >         return 0;
> >  }
> >
> > +static int
> > +test_thread_create_control_join(void)
> > +{
> > +       rte_thread_t thread_id;
> > +       rte_thread_t thread_main_id;
> > +
> > +       thread_id_ready = 0;
> > +       RTE_TEST_ASSERT(rte_thread_create_control(&thread_id, "test_control_threads",
> > +               NULL, thread_main, &thread_main_id) == 0,
> > +               "Failed to create thread.");
> > +
> > +       while (__atomic_load_n(&thread_id_ready, __ATOMIC_ACQUIRE) == 0)
> > +               ;
> > +
> > +       RTE_TEST_ASSERT(rte_thread_equal(thread_id, thread_main_id) != 0,
> > +               "Unexpected thread id.");
> > +
> > +       __atomic_store_n(&thread_id_ready, 2, __ATOMIC_RELEASE);
> > +
> > +       RTE_TEST_ASSERT(rte_thread_join(thread_id, NULL) == 0,
> > +               "Failed to join thread.");
> > +
> > +       return 0;
> > +}
> > +
> >  static struct unit_test_suite threads_test_suite = {
> >         .suite_name = "threads autotest",
> >         .setup = NULL,
> > @@ -243,6 +268,7 @@
> >                 TEST_CASE(test_thread_priority),
> >                 TEST_CASE(test_thread_attributes_affinity),
> >                 TEST_CASE(test_thread_attributes_priority),
> > +               TEST_CASE(test_thread_create_control_join),
> >                 TEST_CASES_END()
> >         }
> >  };
> > diff --git a/lib/eal/common/eal_common_thread.c b/lib/eal/common/eal_common_thread.c
> > index 3181515..4f83c97 100644
> > --- a/lib/eal/common/eal_common_thread.c
> > +++ b/lib/eal/common/eal_common_thread.c
> > @@ -232,7 +232,10 @@ enum __rte_ctrl_thread_status {
> >  };
> >
> >  struct rte_thread_ctrl_params {
> > -       void *(*start_routine)(void *);
> > +       union {
> > +               void *(*ctrl_start_routine)(void *arg);
> > +               rte_thread_func control_start_routine;
> > +       } u;
> >         void *arg;
> >         int ret;
> >         /* Control thread status.
> > @@ -241,27 +244,47 @@ struct rte_thread_ctrl_params {
> >         enum __rte_ctrl_thread_status ctrl_thread_status;
> >  };
> >
> > -static void *ctrl_thread_init(void *arg)
> > +static int ctrl_thread_init(void *arg)
> >  {
> >         struct internal_config *internal_conf =
> >                 eal_get_internal_configuration();
> >         rte_cpuset_t *cpuset = &internal_conf->ctrl_cpuset;
> >         struct rte_thread_ctrl_params *params = arg;
> > -       void *(*start_routine)(void *) = params->start_routine;
> > -       void *routine_arg = params->arg;
> >
> >         __rte_thread_init(rte_lcore_id(), cpuset);
> >         params->ret = rte_thread_set_affinity_by_id(rte_thread_self(), cpuset);
> >         if (params->ret != 0) {
> >                 __atomic_store_n(&params->ctrl_thread_status,
> >                         CTRL_THREAD_ERROR, __ATOMIC_RELEASE);
> > -               return NULL;
> > +               return params->ret;

here too, though a narrow window.

> >         }
> >
> >         __atomic_store_n(&params->ctrl_thread_status,
> >                 CTRL_THREAD_RUNNING, __ATOMIC_RELEASE);
> >
> > -       return start_routine(routine_arg);
> > +       return 0;
> > +}
> > +
> > +static void *ctrl_thread_start(void *arg)
> > +{
> > +       struct rte_thread_ctrl_params *params = arg;
> > +       void *(*start_routine)(void *) = params->u.ctrl_start_routine;
> > +
> > +       if (ctrl_thread_init(arg) != 0)
> > +               return NULL;
> > +
> > +       return start_routine(params->arg);
> 
> Does this patch open a race on params->arg between the newly created
> thread, and the one that asked for creation?
> Reading params->arg is not under the ctrl_thread_status
> syncrhonisation point anymore (which was protecting against
> free(params)).
> 
> WDYT?

yeah, this is definitely wrong. i'm preparing a patch now. both
params->ret and params->arg are loaded unsynchronized in multiple
places.

sorry about the mistake. fix coming soon.

> 
> (Note that I doubt this patch has anything to do with the recent bz
> opened by Intel QE)
> 
> 
> > +}
> > +
> > +static uint32_t control_thread_start(void *arg)
> > +{
> > +       struct rte_thread_ctrl_params *params = arg;
> > +       rte_thread_func start_routine = params->u.control_start_routine;
> > +
> > +       if (ctrl_thread_init(arg) != 0)
> > +               return params->ret;
> 
> Idem here for params->ret.

> 
> 
> > +
> > +       return start_routine(params->arg);
> >  }
> >
> >  int
> > @@ -277,12 +300,12 @@ static void *ctrl_thread_init(void *arg)
> >         if (!params)
> >                 return -ENOMEM;
> >
> > -       params->start_routine = start_routine;
> > +       params->u.ctrl_start_routine = start_routine;
> >         params->arg = arg;
> >         params->ret = 0;
> >         params->ctrl_thread_status = CTRL_THREAD_LAUNCHING;
> >
> > -       ret = pthread_create(thread, attr, ctrl_thread_init, (void *)params);
> > +       ret = pthread_create(thread, attr, ctrl_thread_start, (void *)params);
> >         if (ret != 0) {
> >                 free(params);
> >                 return -ret;
> > @@ -315,6 +338,52 @@ static void *ctrl_thread_init(void *arg)
> >  }
> >
> >  int
> > +rte_thread_create_control(rte_thread_t *thread, const char *name,
> > +       const rte_thread_attr_t *attr,
> > +       rte_thread_func start_routine, void *arg)
> > +{
> > +       struct rte_thread_ctrl_params *params;
> > +       enum __rte_ctrl_thread_status ctrl_thread_status;
> > +       int ret;
> > +
> > +       params = malloc(sizeof(*params));
> > +       if (params == NULL)
> > +               return -ENOMEM;
> > +
> > +       params->u.control_start_routine = start_routine;
> > +       params->arg = arg;
> > +       params->ret = 0;
> > +       params->ctrl_thread_status = CTRL_THREAD_LAUNCHING;
> > +
> > +       ret = rte_thread_create(thread, attr, control_thread_start, params);
> > +       if (ret != 0) {
> > +               free(params);
> > +               return -ret;
> > +       }
> > +
> > +       if (name != NULL)
> > +               rte_thread_set_name(*thread, name);
> > +
> > +       /* Wait for the control thread to initialize successfully */
> > +       while ((ctrl_thread_status =
> > +                       __atomic_load_n(&params->ctrl_thread_status,
> > +                       __ATOMIC_ACQUIRE)) == CTRL_THREAD_LAUNCHING) {
> > +               rte_delay_us_sleep(1);
> > +       }
> > +
> > +       /* Check if the control thread encountered an error */
> > +       if (ctrl_thread_status == CTRL_THREAD_ERROR) {
> > +               /* ctrl thread is exiting */
> > +               rte_thread_join(*thread, NULL);
> > +       }
> > +
> > +       ret = params->ret;
> > +       free(params);
> > +
> > +       return ret;
> > +}
> > +
> > +int
> >  rte_thread_register(void)
> >  {
> >         unsigned int lcore_id;
> > diff --git a/lib/eal/include/rte_thread.h b/lib/eal/include/rte_thread.h
> > index d247930..fae26a7 100644
> > --- a/lib/eal/include/rte_thread.h
> > +++ b/lib/eal/include/rte_thread.h
> > @@ -98,6 +98,39 @@ int rte_thread_create(rte_thread_t *thread_id,
> >   * @warning
> >   * @b EXPERIMENTAL: this API may change without prior notice.
> >   *
> > + * Create a control thread.
> > + *
> > + * Creates a control thread with the given name and attributes. The
> > + * affinity of the new thread is based on the CPU affinity retrieved
> > + * at the time rte_eal_init() was called, the EAL threads are then
> > + * excluded. If setting the name of the thread fails, the error is
> > + * ignored and a debug message is logged.
> > + *
> > + * @param thread
> > + *   Filled with the thread id of the new created thread.
> > + * @param name
> > + *   The name of the control thread
> > + *   (max RTE_MAX_THREAD_NAME_LEN characters including '\0').
> > + * @param thread_attr
> > + *   Attributes for the new thread.
> > + * @param thread_func
> > + *   Function to be executed by the new thread.
> > + * @param arg
> > + *   Argument passed to start_routine.
> > + * @return
> > + *   On success, returns 0; on error, it returns a negative value
> > + *   corresponding to the error number.
> > + */
> > +__rte_experimental
> > +int
> > +rte_thread_create_control(rte_thread_t *thread, const char *name,
> > +               const rte_thread_attr_t *thread_attr,
> > +               rte_thread_func thread_func, void *arg);
> > +
> > +/**
> > + * @warning
> > + * @b EXPERIMENTAL: this API may change without prior notice.
> > + *
> >   * Waits for the thread identified by 'thread_id' to terminate
> >   *
> >   * @param thread_id
> > diff --git a/lib/eal/version.map b/lib/eal/version.map
> > index 6523102..285a0bd 100644
> > --- a/lib/eal/version.map
> > +++ b/lib/eal/version.map
> > @@ -441,6 +441,7 @@ EXPERIMENTAL {
> >         rte_thread_join;
> >
> >         # added in 23.03
> > +       rte_thread_create_control;
> >         rte_thread_set_name;
> >  };
> >
> > --
> > 1.8.3.1
> >
> 
> 
> -- 
> David Marchand