exit has to be assert(), it's actually bug if buffer type do not match one
of these:
if (buf_type == ODP_BUFFER_TYPE_RAW) {
hdr_size = sizeof(odp_raw_buffer_hdr_t);
} else if (buf_type == ODP_BUFFER_TYPE_PACKET) {
hdr_size = sizeof(odp_packet_hdr_t);
} else if (buf_type == ODP_BUFFER_TYPE_TIMEOUT) {
hdr_size = sizeof(odp_timeout_hdr_t);
} else if (buf_type == ODP_BUFFER_TYPE_ANY) {
hdr_size = sizeof(odp_any_buffer_hdr_t);
} else {
ODP_ERR("odp_buffer_pool_create: Bad type %i\n", buf_type);
exit(0);
}
Btw, for ks2 we do no have exit(0) there.
On 27 August 2014 05:08, Bill Fischofer <[email protected]> wrote:
> Many of the linux-generic routines that were written early in the project
> took various shortcuts in this area. They need to be standardized as part
> of the ODP v1.0 cleanup/packaging/validation/documentation effort post-LCU.
> We should start compiling a list of these and get BPs created for these
> now and at LCU.
>
> ODP routines, should not call exit() or similar global "give up" routines
> but should provide appropriate error returns in the case of most problems.
> In the specific case of odp_buffer_pool_create(), it should never abort.
> The parameter list also does not reflect the current design so that too
> will need to be harmonized.
>
> Bill
>
>
> On Tue, Aug 26, 2014 at 7:30 PM, Wiles, Roger Keith <
> [email protected]> wrote:
>
>>
>> On Aug 26, 2014, at 4:25 PM, Mike Holmes <[email protected]> wrote:
>>
>> The odp_buffer_pool_create function has a return code that defaults
>> to ODP_BUFFER_POOL_INVALID, although the API header file doxygen comment
>> does not indicate that there is any error return code possible.
>> I was going to add that information however there truly is no error
>> returned because link_bufs actually calls exit(0) on error.
>>
>> My question is do we update odp_buffer_pool_create docs to say this
>> function may never return, or should link_bufs be rewritten ?
>>
>> I would normally say we need to remove the exit(0) but presumably this
>> was done as as an optimization ? In which case we just need to document it
>> properly ?
>>
>> Thoughts ?
>>
>>
>> odp_buffer_pool_t odp_buffer_pool_create(const char *name,
>> void *base_addr, uint64_t size,
>> size_t buf_size, size_t
>> buf_align,
>> int buf_type)
>> {
>> odp_buffer_pool_t i;
>> pool_entry_t *pool;
>> odp_buffer_pool_t pool_id = ODP_BUFFER_POOL_INVALID;
>>
>> for (i = 0; i < ODP_CONFIG_BUFFER_POOLS; i++) {
>> pool = get_pool_entry(i);
>>
>> LOCK(&pool->s.lock);
>>
>> if (pool->s.buf_base == 0) {
>> /* found free pool */
>>
>> strncpy(pool->s.name, name,
>> ODP_BUFFER_POOL_NAME_LEN - 1);
>> pool->s.name[ODP_BUFFER_POOL_NAME_LEN - 1] = 0;
>> pool->s.pool_base_addr = base_addr;
>> pool->s.pool_size = size;
>> pool->s.user_size = buf_size;
>> pool->s.user_align = buf_align;
>> pool->s.buf_type = buf_type;
>>
>> link_bufs(pool); NEVER returns on some errors
>>
>>
>> I would want this function to not do an exit() call without allowing the
>> application to caught the error. It would mean the link_bufs() routine
>> would need to return and error/error code instead of exiting. Having
>> something exit is not a very friendly action IMO.
>>
>>
>> UNLOCK(&pool->s.lock);
>>
>> pool_id = i;
>> break;
>> }
>>
>> UNLOCK(&pool->s.lock);
>> }
>>
>> return pool_id;
>> }
>>
>>
>>
>> --
>> *Mike Holmes*
>> Linaro Technical Manager / Lead
>> LNG - ODP
>> _______________________________________________
>> lng-odp mailing list
>> [email protected]
>> http://lists.linaro.org/mailman/listinfo/lng-odp
>>
>>
>> *Keith **Wiles*, Principal Technologist with CTO office, *Wind River *mobile
>> 972-213-5533
>>
>>
>> _______________________________________________
>> lng-odp mailing list
>> [email protected]
>> http://lists.linaro.org/mailman/listinfo/lng-odp
>>
>>
>
> _______________________________________________
> lng-odp mailing list
> [email protected]
> http://lists.linaro.org/mailman/listinfo/lng-odp
>
>
_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp
