On Fri, May 22, 2015 at 2:04 PM, Maxim Uvarov <[email protected]> wrote:
> On 05/22/15 13:25, Ciprian Barbu wrote:
>>
>> On Thu, May 21, 2015 at 6:32 PM, Maxim Uvarov <[email protected]>
>> wrote:
>>>
>>> On init odp creates odp_sched_pool. Because of we can not
>>> modify API to add new parameter to odp_pool_param_t this
>>> pool should not be shared between different processes. To
>>> make that add special value to shm provided to pool saying that
>>> this pool should be in local memory only.
>>
>> This description is not very clear.
>>
>>> Signed-off-by: Maxim Uvarov <[email protected]>
>>> ---
>>> platform/linux-generic/include/odp/plat/shared_memory_types.h | 4 ++++
>>> platform/linux-generic/odp_pool.c | 9
>>> +++++++--
>>> platform/linux-generic/odp_schedule.c | 2 +-
>>> 3 files changed, 12 insertions(+), 3 deletions(-)
>>>
>>> diff --git
>>> a/platform/linux-generic/include/odp/plat/shared_memory_types.h
>>> b/platform/linux-generic/include/odp/plat/shared_memory_types.h
>>> index 4be7356..908bb2e 100644
>>> --- a/platform/linux-generic/include/odp/plat/shared_memory_types.h
>>> +++ b/platform/linux-generic/include/odp/plat/shared_memory_types.h
>>> @@ -30,6 +30,10 @@ typedef ODP_HANDLE_T(odp_shm_t);
>>>
>>> #define ODP_SHM_INVALID _odp_cast_scalar(odp_shm_t, 0)
>>> #define ODP_SHM_NULL ODP_SHM_INVALID
>>> +/** NULL shared memory but do not create IPC object for it.
>>> + * Platfrom specific flag.
>>> + */
>>> +#define _ODP_SHM_NULL_LOCAL _odp_cast_scalar(odp_shm_t, 0xffffffffULL -
>>> 1)
>>
>> I don't think we should have this visible to ODP applications, because
>> it's a workaround.
>>
>>> /** Get printable format of odp_shm_t */
>>> static inline uint64_t odp_shm_to_u64(odp_shm_t hdl)
>>> diff --git a/platform/linux-generic/odp_pool.c
>>> b/platform/linux-generic/odp_pool.c
>>> index cd2c449..b2f30dc 100644
>>> --- a/platform/linux-generic/odp_pool.c
>>> +++ b/platform/linux-generic/odp_pool.c
>>> @@ -151,10 +151,14 @@ odp_pool_t odp_pool_create(const char *name,
>>> odp_pool_t pool_hdl = ODP_POOL_INVALID;
>>> pool_entry_t *pool;
>>> uint32_t i, headroom = 0, tailroom = 0;
>>> + uint32_t shm_flags = 0;
>>>
>>> if (params == NULL)
>>> return ODP_POOL_INVALID;
>>>
>>> + if (shm == ODP_SHM_NULL)
>>> + shm_flags = ODP_SHM_PROC;
>>> +
>>> /* Default size and align for timeouts */
>>> if (params->type == ODP_POOL_TIMEOUT) {
>>> params->buf.size = 0; /* tmo.__res1 */
>>> @@ -289,10 +293,11 @@ odp_pool_t odp_pool_create(const char *name,
>>> mdata_size +
>>> udata_size);
>>>
>>> - if (shm == ODP_SHM_NULL) {
>>> + if (shm == ODP_SHM_NULL || shm == _ODP_SHM_NULL_LOCAL) {
>>> shm = odp_shm_reserve(pool->s.name,
>>> pool->s.pool_size,
>>> - ODP_PAGE_SIZE, 0);
>>> + ODP_PAGE_SIZE,
>>> + shm_flags);
>>> if (shm == ODP_SHM_INVALID) {
>>> POOL_UNLOCK(&pool->s.lock);
>>> return ODP_POOL_INVALID;
>>> diff --git a/platform/linux-generic/odp_schedule.c
>>> b/platform/linux-generic/odp_schedule.c
>>> index a63f97a..07422bd 100644
>>> --- a/platform/linux-generic/odp_schedule.c
>>> +++ b/platform/linux-generic/odp_schedule.c
>>> @@ -129,7 +129,7 @@ int odp_schedule_init_global(void)
>>> params.buf.num = NUM_SCHED_CMD;
>>> params.type = ODP_POOL_BUFFER;
>>>
>>> - pool = odp_pool_create("odp_sched_pool", ODP_SHM_NULL, ¶ms);
>>> + pool = odp_pool_create("odp_sched_pool", _ODP_SHM_NULL_LOCAL,
>>> ¶ms);
>>
>> I think you're going about this the wrong way. ODP_SHM_NULL is a
>> convenient way to let the implementation chose the memory where it
>> will create buffers / packets / tmo. It was introduced because most
>> hardware platforms don't use shared memory, since they have their own
>> buffer management. But linux-generic can only use shared memory, and
>> it's ok to allocate an shm here because it's inside the
>> implementation, there is no visibility towards the application. So I
>> think you should allocated the shm here with the right flags and
>> remove the _ODP_SHM_NULL_LOCAL hack.
>
>
> we were planned to remove shm argument from pool_create() in future. For not
> it will work, but later
> I'm not sure what is the good solution for that.
So why try to workaround it when we haven't even decided when to
remove the shm param from odp_pktio_open?
>
> Maxim.
>
>
>>> if (pool == ODP_POOL_INVALID) {
>>> ODP_ERR("Schedule init: Pool create failed.\n");
>>> --
>>> 1.9.1
>>>
>>> _______________________________________________
>>> lng-odp mailing list
>>> [email protected]
>>> https://lists.linaro.org/mailman/listinfo/lng-odp
>
>
_______________________________________________
lng-odp mailing list
[email protected]
https://lists.linaro.org/mailman/listinfo/lng-odp
