Hfi2 - #1710
Open
ddalessa wants to merge 3 commits into
Open
Conversation
ddalessa
force-pushed
the
hfi2
branch
4 times, most recently
from
March 13, 2026 17:33
819425e to
4b9a8b2
Compare
There was a problem hiding this comment.
Pull request overview
This PR replaces the legacy ipath/qib userspace provider with a new hfi2 userspace provider, and removes truescale-related packaging bits while updating RDMA kernel header UAPI to match new ioctl/ABI requirements.
Changes:
- Remove the
providers/ipathverbsprovider (and associated truescale modprobe/dracut artifacts) and add a newproviders/hfi2verbsprovider. - Update distro packaging specs/installs to stop shipping
truescale.confandtruescale-serdes.cmds. - Refresh/install new RDMA kernel header definitions (driver IDs, buffer descriptors, mlx5/bnxt uapi additions) and update udev rule classification for
hfi2.
Reviewed changes
Copilot reviewed 31 out of 31 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| suse/rdma-core.spec | Stops packaging truescale modprobe config and serdes script. |
| redhat/rdma-core.spec | Stops packaging truescale modprobe config and serdes script. |
| providers/ipathverbs/verbs.c | Removes legacy ipathverbs provider implementation. |
| providers/ipathverbs/truescale.conf.in | Removes truescale modprobe install rule. |
| providers/ipathverbs/truescale-serdes.cmds | Removes truescale serdes helper script. |
| providers/ipathverbs/ipathverbs.h | Removes legacy ipathverbs provider header. |
| providers/ipathverbs/ipathverbs.c | Removes legacy ipathverbs provider entry/registration code. |
| providers/ipathverbs/ipath-abi.h | Removes ipathverbs UAPI shim header. |
| providers/ipathverbs/dracut_kmod | Removes dracut module hook for ib_qib. |
| providers/ipathverbs/dracut_install | Removes dracut install hook for truescale assets. |
| providers/ipathverbs/dracut_check | Removes dracut hostonly detection for qib hardware. |
| providers/ipathverbs/COPYING | Removes provider-specific licensing file for ipathverbs. |
| providers/ipathverbs/CMakeLists.txt | Removes provider build/install rules for ipathverbs. |
| providers/hfi2verbs/verbs.c | Adds hfi2 verbs implementation (CQ/QP/SRQ/etc). |
| providers/hfi2verbs/hfiverbs.h | Adds hfi2 provider data structures and function prototypes. |
| providers/hfi2verbs/hfiverbs.c | Adds hfi2 provider registration/match table. |
| providers/hfi2verbs/hfi-abi.h | Adds hfi2 provider UAPI shim header. |
| providers/hfi2verbs/CMakeLists.txt | Adds build rules for the hfi2 provider. |
| MAINTAINERS | Adds a maintainer section for the new hfi2 provider. |
| libibverbs/verbs.h | Adds hfi2 provider extern and includes substantial formatting/macro changes. |
| kernel-headers/rdma/mlx5_user_ioctl_cmds.h | Adds new mlx5 CQ/QP attribute IDs for umem-backed DBR buffers. |
| kernel-headers/rdma/ib_user_verbs.h | Adds a new uverbs device capability bit. |
| kernel-headers/rdma/ib_user_ioctl_verbs.h | Adds RDMA_DRIVER_HFI2 and a new buffer descriptor UAPI type. |
| kernel-headers/rdma/ib_user_ioctl_cmds.h | Adds new uverbs CQ/QP buffer umem attribute IDs. |
| kernel-headers/rdma/hfi2-abi.h | Adds hfi2 kernel ABI header to published kernel headers set. |
| kernel-headers/rdma/bnxt_re-abi.h | Updates bnxt_re QP mask naming/fields and adds create_qp attrs. |
| kernel-headers/CMakeLists.txt | Publishes the new rdma/hfi2-abi.h header. |
| kernel-boot/rdma-description.rules | Marks hfi2 devices as OPA-capable via udev classification. |
| debian/rdma-core.lintian-overrides | Removes obsolete lintian overrides tied to truescale modprobe install usage. |
| debian/rdma-core.install | Stops installing removed truescale config/script into the Debian package. |
| CMakeLists.txt | Switches provider build from ipathverbs to hfi2verbs. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
54
to
58
| #if __GNUC__ >= 3 | ||
| # define __attribute_const __attribute__((const)) | ||
| #define __attribute_const __attribute_const__ | ||
| #else | ||
| # define __attribute_const | ||
| #define __attribute_const | ||
| #endif |
Comment on lines
+100
to
+104
| struct hfi2_cq_wc { | ||
| _Atomic(uint32_t) head; | ||
| _Atomic(uint32_t) tail; | ||
| struct hfi2_wc queue[]; | ||
| }; |
| S: Supported | ||
| F: providers/hfi1verbs/ | ||
|
|
||
| HFI2 USERSPACE PROVIDER (for hf2.ko) |
| HFI2_ATTR_SDMA_FD_RSP = (1U << UVERBS_ID_NS_SHIFT), | ||
| }; | ||
|
|
||
| #endif /* _LINIUX_HFI2_USER_H */ |
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 31 out of 31 changed files in this pull request and generated 1 comment.
Suppressed comments (4)
libibverbs/verbs.h:2304
- This header still declares
verbs_provider_ipathverbs, but the ipathverbs provider has been removed from the build (andproviders/ipathverbs/is gone). Keeping the extern makes the public provider list misleading.
extern const struct verbs_device_ops verbs_provider_erdma;
extern const struct verbs_device_ops verbs_provider_hfi1verbs;
extern const struct verbs_device_ops verbs_provider_hfi2verbs;
extern const struct verbs_device_ops verbs_provider_hns;
extern const struct verbs_device_ops verbs_provider_ipathverbs;
extern const struct verbs_device_ops verbs_provider_irdma;
extern const struct verbs_device_ops verbs_provider_mana;
libibverbs/verbs.h:58
__attribute_constis defined as__attribute_const__, which is not defined here and will typically break compilation when__attribute_constis used in function prototypes.
#if __GNUC__ >= 3
#define __attribute_const __attribute_const__
#else
#define __attribute_const
#endif
MAINTAINERS:74
- Typo in kernel module name: this should reference
hfi2.ko(nothf2.ko).
HFI2 USERSPACE PROVIDER (for hf2.ko)
kernel-headers/rdma/bnxt_re-abi.h:130
- Renaming the enum constant removes the old identifier from the installed UAPI header, which can break source compatibility for out-of-tree consumers. Consider keeping the old name as an alias to the new one.
enum bnxt_re_qp_mask {
BNXT_RE_QP_REQ_MASK_FIXED_QUE_ATTR = 0x1,
};
Comment on lines
+100
to
+104
| struct hfi2_cq_wc { | ||
| _Atomic(uint32_t) head; | ||
| _Atomic(uint32_t) tail; | ||
| struct hfi2_wc queue[]; | ||
| }; |
The ipath driver has been dead for over a decade. Qib has also been removed from the kernel. Toss this old code out of the boat. Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 27 out of 27 changed files in this pull request and generated no new comments.
Suppressed comments (5)
libibverbs/verbs.h:58
__attribute_constis redefined to__attribute_const__, which is not guaranteed to be provided by all libc/toolchain environments (it’s a glibc internal macro). This can break compilation on non-glibc systems. Define it directly using GCC’s attribute syntax, as before.
#if __GNUC__ >= 3
#define __attribute_const __attribute_const__
#else
#define __attribute_const
#endif
providers/hfi2verbs/hfiverbs.h:104
struct hfi2_cq_wcuses a flexible array member (queue[]), but the CQ mmap sizing and wrap logic inverbs.cassume an extra sentinel element (i.e., indices can reachcqe). Withqueue[], the computed mmap size is too small and can lead to out-of-bounds access whentailreachescqe.
struct hfi2_cq_wc {
_Atomic(uint32_t) head;
_Atomic(uint32_t) tail;
struct hfi2_wc queue[];
};
kernel-headers/rdma/hfi2-abi.h:678
- Typo in the header guard comment (
_LINIUX_->_LINUX_).
#endif /* _LINIUX_HFI2_USER_H */
providers/hfi2verbs/hfiverbs.c:69
INT_MAXis used inhfi2_dev_ops.match_max_abi_version, but this file does not include<limits.h>, which can cause build failures depending on include ordering.
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
MAINTAINERS:74
- Typo in kernel module name:
hf2.koshould behfi2.ko. Also, sinceproviders/ipathverbs/is removed in this PR, the existing IPATH provider entry later in this file (currently around line 107) should be removed/updated to avoid stale maintenance metadata.
HFI2 USERSPACE PROVIDER (for hf2.ko)
Add InfiniBand verbs support for the hfi2 driver, which drives the Cornelis Networks CN5000 adapter. Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
To commit: ?? ("RDMA/hfi2: Support ipoib").
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 27 out of 27 changed files in this pull request and generated 1 comment.
Suppressed comments (4)
libibverbs/verbs.h:58
- __attribute_const is redefined to attribute_const, which is not defined by this header and can cause compile failures on libcs/toolchains that don't provide that macro (it turns function declarations like
int __attribute_const foo();into invalid syntax).
#if __GNUC__ >= 3
#define __attribute_const __attribute_const__
#else
#define __attribute_const
#endif
libibverbs/verbs.h:2348
- verbs_provider_ipathverbs is still declared in the public header, but this PR removes the ipathverbs provider from the build (top-level CMakeLists.txt stops adding providers/ipathverbs and the provider sources are deleted). Keeping this declaration can mislead users and can trigger link failures if someone tries to statically reference the symbol.
extern const struct verbs_device_ops verbs_provider_hfi1verbs;
extern const struct verbs_device_ops verbs_provider_hfi2verbs;
extern const struct verbs_device_ops verbs_provider_hns;
extern const struct verbs_device_ops verbs_provider_ipathverbs;
extern const struct verbs_device_ops verbs_provider_irdma;
MAINTAINERS:74
- Typo in kernel module name: this entry says "hf2.ko" but the driver name elsewhere is "hfi2" (e.g., rdma-description.rules uses DRIVERS=="hfi2").
HFI2 USERSPACE PROVIDER (for hf2.ko)
kernel-headers/rdma/hfi2-abi.h:678
- Typo in the closing include-guard comment (LINIUX vs LINUX).
#endif /* _LINIUX_HFI2_USER_H */
Comment on lines
+100
to
+104
| struct hfi2_cq_wc { | ||
| _Atomic(uint32_t) head; | ||
| _Atomic(uint32_t) tail; | ||
| struct hfi2_wc queue[]; | ||
| }; |
Member
|
Strange, i disabled for more than once Copilot and it still reports. |
This file contains hidden or 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
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.
Add support hfi2 and get rid of legacy ipath. The kernel series for this is making it's way to the list now.