Josep Puigdemont(joseppc) replied on github web page:
platform/linux-generic/odp_packet.c
line 8
@@ -816,10 +816,7 @@ odp_packet_t odp_packet_alloc(odp_pool_t pool_hdl,
uint32_t len)
odp_packet_t pkt;
int num, num_seg;
- if (odp_unlikely(pool->params.type != ODP_POOL_PACKET)) {
- __odp_errno = EINVAL;
- return ODP_PACKET_INVALID;
- }
+ ODP_ASSERT(pool->params.type == ODP_POOL_PACKET);
Comment:
I think this is not correct. Before if we detected an error we interrupted the
function returning an error code, now we will only interrupt this function (by
calling abort_fn() global function) if and only if ODP was compiled with
ODP_DEBUG enabled.
> Bill Fischofer(Bill-Fischofer-Linaro) wrote:
> My point is that we want two things: Cache alignment and no wasted space.
> Kevin's PR addressed this issue. It seems all we need is document the
> implications for those who wish to set `CONFIG_POOL_CACHE_SIZE` since
> otherwise we'll just be inserting a lot of padding that once again will waste
> space.
>> nagarahalli wrote
>> Kevin's PR is to adjust the default value so that by default we are using 1
>> cache line. However, CONFIG_POOL_CACHE_SIZE is configurable. If the user is
>> not aware of the internal details of the pool implementation, parts of 2
>> elements of the pool cache array will be in the same cache line. This change
>> will guard against that. If the CONFIG_POOL_CACHE_SIZE is configured
>> correctly, this will not add any additional space.
>>> Bill Fischofer(Bill-Fischofer-Linaro) wrote:
>>> With Kevin's latest change (PR #326) this is already an integral multiple
>>> of the cache size. Not sure what value we're gaining here. Is this to guard
>>> against random changes to `CONFIG_POOL_CACHE_SIZE`? That still could leave
>>> a lot of wasted space, which was the idea behind Kevin's PR.
https://github.com/Linaro/odp/pull/346#discussion_r157737389
updated_at 2017-12-19 12:16:42
