uring enablement #8
Open
+35,157
−15,579
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
9c8b7eb to
9ebf395
Compare
Signed-off-by: Thomas Huber <thomas.huber@cornelisnetworks.com>
Replace running of on-merge workflow with a nightly workflow instead. Signed-off-by: Jack Morrison <jack.morrison@cornelisnetworks.com>
Do not use PR closed events as workflow triggers. Allow triggering PR events when targeting any branch, not just main. Change cron schedule to account for UTC. Improve conditional execution of reusable workflows. Signed-off-by: Jack Morrison <jack.morrison@cornelisnetworks.com>
Signed-off-by: Bob Cernohous <bob.cernohous@cornelisnetworks.com>
This commit adds support for link up/down events in OPX for WFR platforms. Signed-off-by: Archana Venkatesha <archana.venkatesha@cornelisnetworks.com>
Signed-off-by: Mike Wilkins <michael.wilkins@cornelisnetworks.com>
Signed-off-by: Bob Cernohous <bob.cernohous@cornelisnetworks.com>
Signed-off-by: Bob Cernohous <bob.cernohous@cornelisnetworks.com>
Fix 16B PBC/payload lengths Signed-off-by: Bob Cernohous <bob.cernohous@cornelisnetworks.com>
Signed-off-by: Mike Wilkins <michael.wilkins@cornelisnetworks.com>
Signed-off-by: Bob Cernohous <bob.cernohous@cornelisnetworks.com>
Signed-off-by: Mike Wilkins <michael.wilkins@cornelisnetworks.com> Signed-off-by: Ben Lynam <Ben.Lynam@cornelisnetworks.com>
Signed-off-by: Ben Lynam <Ben.Lynam@cornelisnetworks.com>
Signed-off-by: Lindsay Reiser <lindsay.reiser@cornelisnetworks.com>
… rendezvous Signed-off-by: Ben Lynam <Ben.Lynam@cornelisnetworks.com>
Add ability to independently tune the minimum threshold to use expected receive (TID) when sending. Signed-off-by: Ben Lynam <Ben.Lynam@cornelisnetworks.com>
Shorten the name field of opx-ci. Remove schedule-triggered Nightly job. Signed-off-by: Jack Morrison <jack.morrison@cornelisnetworks.com>
Signed-off-by: Bob Cernohous <bob.cernohous@cornelisnetworks.com>
Also store pad value not buffer data Signed-off-by: Bob Cernohous <bob.cernohous@cornelisnetworks.com>
Signed-off-by: Thomas Huber <thomas.huber@cornelisnetworks.com>
Signed-off-by: Elias Kozah <elias.elkozah@cornelisnetworks.com>
…resulting from ignored context creation error Signed-off-by: Elias Kozah <Elias.Elkozah@cornelisnetworks.com>
…vous performance Signed-off-by: Ben Lynam <Ben.Lynam@cornelisnetworks.com>
The OPX provider was explicitly setting FI_REMOTE_CQ_DATA on all receive operations; however, it should only set the flag to indicate that the data field contains the completion data provided by the peer as part of their transmit request. Signed-off-by: Lindsay Reiser <lindsay.reiser@cornelisnetworks.com>
Signed-off-by: Ben Lynam <Ben.Lynam@cornelisnetworks.com>
Signed-off-by: Jack Morrison <jack.morrison@cornelisnetworks.com>
Currently, efa_base_ep's default rnr_retry is 3 which only does a few retry in the firmware level for RNR. This is due to the efa_rdm_ep supports libfabric level RNR retry. However, the efa-direct ep doesn't support libfabric level RNR retry. Then we should make it do infinite RNR retry (7), which is also the default behavior of SRD QP. Signed-off-by: Shi Jin <sjina@amazon.com>
This commit removes the x86-64 architecture check from the static_assert conditional compilation directive. The static_assert feature is not architecture-dependent and should be checked on all platforms that support it. Signed-off-by: Jessie Yang <jiaxiyan@amazon.com>
Store the completion flags and peer address in FI_CONTEXT2 and retrieve later when writing cq. Signed-off-by: Jessie Yang <jiaxiyan@amazon.com>
Signed-off-by: Sai Sunku <sunkusa@amazon.com>
Other memory monitors, such as CUDA, ROCR, and ZE, have a .c file for the implementation. This change cleans up the util_mem_monitor.c code by defining a uffd and import .c file, thus aligning to other memory monitor implementations. Signed-off-by: Mike Uttormark <mike.uttormark@hpe.com> Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
Some memory monitors, such as kdreg2, have a subscription context per MR cache entry. These memory monitors require unsubscribe to be called for each freed MR cache entry. To support this, call unsubscribe when an entry is remove from the MR cache RB tree. If a memory monitor does not support a subscription context per MR, unsubscribe must be implemented as a noop. Update uffd and rocr memory monitors accordingly. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
ROCR deallocation CB will call rocr_unsubscribe with mm_lock held. If memhooks is used, since rocr_unsubscribe may call free, this can result in memhooks intercepting the free and leading to deadlock. To avoid this, freeing is deferred until locks are released. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
Subscribe, unsubscribe, and valid are callbacks which are dynamically setup. Change this to be statically set. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
An MR cache utilizing kdreg2 will have incorrect MR cache count stats if unsubscribe is not called. Signed-off-by: Mike Uttormark <mike.uttormark@hpe.com> Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
prior to this patch, when efa_hmem_info_check_p2p_support_cuda elected to attempt dmabuf for p2p, we previously leaked the file descriptor returned by cuMemGetHandleForAddressRange in all cases. This ultimately meant the dmabuf stuck around for the lifetime of the process, even after dereg and after releasing the memory back to the device mempool. All calls to cuda_get_dmabuf_fd need a corresponding close call. Signed-off-by: Nicholas Sielicki <nslick@amazon.com>
For some HMEM ifaces, ofi_hmem_get_dmabuf_fd() may result in a new FD being allocated. Define ofi_hmem_put_dmabuf_fd() to close FD. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
With ROCR, callers of ofi_hmem_get_dmabuf_fd() should call ofi_hmem_put_dmabuf_fd() once the DMA buf region is no longer used. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
With CUDA, callers of ofi_hmem_get_dmabuf_fd() should call ofi_hmem_put_dmabuf_fd() once the DMA buf region is no longer used. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
Performing multiple HSA allocations appears to result in a DMA buf offset. Verify that the CXI provider can register a DMA buf offset memory region. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
When a MR is freed, the CXI provider should free the DMA buf FD used for the ROCR region. Failing to do this will result in FDs being exhausted. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
When a MR is freed, the CXI provider should free the DMA buf FD used for the CUDA region. Failing to do this will result in FDs being exhausted. Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
This allows testing FI_CONTEXT2 in providers that require this mode bit. Signed-off-by: Jessie Yang <jiaxiyan@amazon.com>
Use cuda_put_dmabuf_fd to close fd Signed-off-by: Shi Jin <sjina@amazon.com>
Signed-off-by: Zach Dworkin <zachary.dworkin@intel.com>
Signed-off-by: Jessie Yang <jiaxiyan@amazon.com>
When multiple multi-recv buffers are posted, FI_MULTI_RECV would only be set on error if an mrecv entry was already created, meaning the buffer would have already been in-use. If the buffer has not been used yet and a cancelation for this buffer has been processed, correctly set FI_MULTI_RECV when reporting the error, indicating that the buffer is no longer in use. Signed-off-by: Jerome Soumagne <jerome.soumagne@hpe.com>
This ensures that the libcurl dlopen path is correct If the user passes '--with-curl=<path>' to configure, then the dlopen of libcurl should honor that selection and use the file path passed in Signed-off-by: John Biddiscombe <biddisco@cscs.ch>
Signed-off-by: John Biddiscombe <biddisco@cscs.ch>
Bumps [actions/stale](https://github.com/actions/stale) from 9.0.0 to 9.1.0. - [Release notes](https://github.com/actions/stale/releases) - [Changelog](https://github.com/actions/stale/blob/main/CHANGELOG.md) - [Commits](actions/stale@28ca103...5bef64f) --- updated-dependencies: - dependency-name: actions/stale dependency-type: direct:production update-type: version-update:semver-minor ... Signed-off-by: dependabot[bot] <support@github.com>
Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.28.1 to 3.28.5. - [Release notes](https://github.com/github/codeql-action/releases) - [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md) - [Commits](github/codeql-action@b6a472f...f6091c0) --- updated-dependencies: - dependency-name: github/codeql-action dependency-type: direct:production update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com>
we may receive uring events before we're fully connected so don't try to progress rx until that connection is established Signed-off-by: Stephen Oost <stephen.oost@intel.com>
the previously used io_uring_prep_readv function does not support flags, instead flags were being passed as an offset, triggering an illegal seek error Signed-off-by: Stephen Oost <stephen.oost@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.
No description provided.