prov/efa: Cleanup ep and peer structures #10382
Conversation
These dlists are now migrated to srx context. Signed-off-by: Shi Jin <[email protected]>
Currently, extra_info is a 256*8 bytes array in efa_rdm_ep and efa_rdm_peer, this is way too much as it can cover 256*64 bits of extra features/requests that RDM endpoint is likely to have. This patch reduces it to 4*8 bytes array that can cover 256 bits which should be enough for the foreseeable future. Signed-off-by: Shi Jin <[email protected]>
shijin-aws
changed the title
| * and efa_rdm_peer | ||
| * 4 means 64*4=256 bits of extra features or requests | ||
| */ | ||
| #define EFA_RDM_MAX_NUM_EXINFO (4) |
There was a problem hiding this comment.
Hmmm did this bug cause any issue?
There was a problem hiding this comment.
The code was actually correct: https://github.com/ofiwg/libfabric/blob/main/prov/efa/src/rdm/efa_rdm_pke_nonreq.c#L40 that it divided by 64 when calculating the number of extra_info. But I have no idea why we make it 256 as the array length. Likely an overlook
There was a problem hiding this comment.
The arrays were originally of length 1 (
There was a problem hiding this comment.
It was planned that RDM v4 will support only 64 extra features, v5 will support 128, and so on. So it was reasonable to have it as length 1 at that time. I think this PR aimed to make it support 256/64 = 4 so it can support 256 features.
|
bot:aws:retest |
1 similar comment
|
bot:aws:retest |
Signed-off-by: Shi Jin <[email protected]>
This PR contains 2 commits that cleans up efa_rdm_ep and efa_rdm_peer structs.