prov/net: Updates from v1.17.x branch back into main #8283
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Allow dumping message details for sent and received data transfers and completions for debug purposes. Signed-off-by: Sean Hefty <sean.hefty@intel.com>
This is a simple rename of the structures to indicate that these messages are actives. A separate change can reuse these structures to track other active messages (i.e. unexpected receives). Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Move common functionality from xnet_match_claim and xnet_match_rx_tag into a common function. This will be reused in a later patch as well. Signed-off-by: Sean Hefty <sean.hefty@intel.com>
There's only 1 unexpected list associated with the progress structure. Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Handle 0-byte messages with claim/discard. Avoid trying to allocate 0-bytes. Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Rename function, as it will be re-used later, and remove the match handler. That tool will be needed to simplify matching in a future patch. Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
If we're waiting for an ACK or an RMA read response, it may be
blocked behind a unexpected receive message. Unfortunately, this
is a common MPI flow:
for each rank
RMA read()
barrier()
The barrier messages may begin to show up from peers before
the local rank has finished its RMA read operations. And the
local rank won't post the receive message for the barrier until
the RMA read operations have completed.
There seems to be some sort of assumption that RMA read responses
have a separate from from messages, which is true when using an
RDMA device. However, it's not true for tcp streams. To handle
this case, we need to look ahead in the tcp stream for the response,
queuing the barrier message.
To minimize the changes to the provider and avoid re-introducing
unexpected message buffering at the receiver, we limit peeking
ahead in the tcp stream to a single 0-byte tagged message (what MPI
uses for barrier), at least for now.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
We need the saved_entry not unexp_entry. Signed-off-by: Sean Hefty <sean.hefty@intel.com>
If we have a saved rx entry, remove the ep from the progress' saved list. Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Look-ahead patch series 'front'-ported to main, with conflicts from io_uring changes resolved.