prov/efa: Use long CTS protocol if runting read protocol fails because of memory registration limits #9493
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
sunkuamzn
force-pushed
the
runt-read-fallback-both
branch
2 times, most recently
from
1c4e49a to
3774491
Compare
sunkuamzn
force-pushed
the
runt-read-fallback-both
branch
from
3774491 to
22e50bc
Compare
shijin-aws
reviewed
Nov 1, 2023
| /* The data_offset will be non-zero when the long CTS RTM packet | ||
| * is sent to continue a runting read transfer after the | ||
| * receiver has run out of memory registrations */ | ||
| assert((data_offset == 0 || ope->internal_flags & EFA_RDM_OPE_READ_NACK) && data_size == -1); |
There was a problem hiding this comment.
So for runting read, receiver already gets some data before try to register the memory. If we fallback to long cts, we restart from the data_offset. Is that correct?
There was a problem hiding this comment.
Yes, that's correct
| send any data with the RTM packet. This is because the runting read RTM | ||
| packets have already delivered some of the data and the long CTS RTM | ||
| packet does not have a seg_offset field */ | ||
| if (txe->internal_flags & EFA_RDM_OPE_READ_NACK) { |
There was a problem hiding this comment.
so it is because long cts rtm currently doesn't support sending from a non-zero offset. data pkts supports offset so it is not impacted. Do I understand correctly?
sunkuamzn
force-pushed
the
runt-read-fallback-both
branch
from
22e50bc to
248ed7f
Compare
shijin-aws
approved these changes
Nov 1, 2023
|
@wzamazon could you have a look at it? |
FI_ENOMR is returned when the hardware memory registration limit is reached Signed-off-by: Sai Sunku <sunkusa@amazon.com>
The READ_NACK feature is checked before sending a EFA_RDM_READ_NACK_PKT packet. The EFA_RDM_READ_NACK_PKT packet is sent by a receiver when it fails to register a buffer to receive the RDMA read data in a long read or runting read protocol Signed-off-by: Sai Sunku <sunkusa@amazon.com>
…ENOMR Long read protocol could fail with ENOMR if the EFA provider is unable to register the buffer with the NIC. In that case, we should fall back to long CTS instead This commit is for the changes when the sender fails to register the source buffer. The sender will switch to the long CTS protocol. Signed-off-by: Sai Sunku <sunkusa@amazon.com>
This change is required for the long read nack protocol where we get the msg_id from ope instead of from the pke Signed-off-by: Sai Sunku <sunkusa@amazon.com>
Long read protocol could fail with ENOMR if the EFA provider is unable to register the buffer with the NIC. In that case, we should fall back to long CTS protocol. This commit is for the changes when the receiver fails to register the destination memory. Receiver sends a NACK packet (packet type EFA_RDM_READ_NACK_PKT) to the sender. The sender switches to the long CTS protocol. Signed-off-by: Sai Sunku <sunkusa@amazon.com>
…ENOMR Runting read protocol could fail with ENOMR if the EFA provider is unable to register the buffer with the NIC. In that case, we should fall back to long CTS instead This commit is for the changes when the sender fails to register the source buffer. The sender will switch to the long CTS protocol. Signed-off-by: Sai Sunku <sunkusa@amazon.com>
Runting read protocol could fail with ENOMR if the EFA provider is unable to register the buffer with the NIC. In that case, we should fall back to long CTS protocol. This commit is for the changes when the receiver fails to register the destination memory. Receiver sends a NACK packet (packet type EFA_RDM_READ_NACK_PKT) to the sender. The sender switches to the long CTS protocol. Signed-off-by: Sai Sunku <sunkusa@amazon.com>
Signed-off-by: Sai Sunku <sunkusa@amazon.com>
sunkuamzn
force-pushed
the
runt-read-fallback-both
branch
from
248ed7f to
592dd68
Compare
This was referenced Nov 7, 2023
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.
The runting read protocol can fail because of MR registration limits on the hardware. This PR has changes to switch to long CTS protocol when that happens.
The first four commits are in #9432