Fix BCC CI: shallow-clone versioning, clang/tools_smoke test regressions #5428

HoyeonRhee · 2025-11-25T19:56:51Z

bcc CI has been red from shallow-clone builds emitting PEP 440–invalid
"128-NOTFOUND-…" versions, and from py_test_clang/py_test_tools_smoke
breakage on newer kernels. This series supersedes the prior PR #5414.

By first commit, cmake/version.cmake now falls back to a compliant
'0.0.0+<hash>' when tags are missing, satisfying newer python setuptools
in shallow-clone workflows.

The clang tests are fixed to use blk-mq (rq->q->disk in place of rq_disk)
and the percpu-counter layout for mm->rss_stat[MM_ANONPAGES].count,
and kmalloc kprobe are replaced with kmalloc_noprof for recent kernels.
[1] [2] [3]

Additionally, smoke tests now only runs btrfs/f2fs tools when those
modules are loaded, and softirqslower.py is fixed with missing exec bit to
avoid error.

Together these changes clear the py_test_clang and py_test_tools_smoke
failures and restore CI.

1: commit torvalds/linux@f3fa33acca9f ("block: remove the ->rq_disk field in struct request")
2: commit torvalds/linux@f1a7941243c1 ("mm: convert mm's rss stats into percpu_counter")
3: commit torvalds/linux@7bd230a26648 ("mm/slab: enable slab allocation tagging for kmalloc and friends")

cmake/version.cmake

HoyeonRhee · 2025-11-28T14:25:34Z

Updated GIT_TAG_LAST tag match with -NOTFOUND placeholder to cover other edge cases defined in
iovisor/bcc/blob/master/cmake/GitRevisionDescription.cmake.

ekyooo · 2025-12-04T11:06:23Z

tests/python/test_clang.py


-    if (!rq->rq_disk || rq->rq_disk->major != 5 ||
-        rq->rq_disk->first_minor != 6)
+    struct gendisk *disk = rq->q->disk;


Is a NULL check needed for rq->q?

This code is used as a standalone test, so it isn’t attached to any events.
In normal blk-mq paths blk_mq_start_request, rq->q should always be initialized, so a NULL here would indicate a kernel issue.

Still, if you think adding a NULL check improves clarity, I’m happy to include it.

ekyooo · 2025-12-04T11:12:20Z

tests/python/test_clang.py


-    if (!rq->rq_disk || rq->rq_disk->major != 5 ||
-        rq->rq_disk->first_minor != 6)
+    struct gendisk *disk = rq->q->disk;


Please follow declaration-first style.

Got it. But to follow the rq->q NULL check, I’ll drop the separate declaration.

ekyooo · 2025-12-05T02:10:34Z

tests/python/test_clang.py

        text = b"""
 #include <linux/blk_types.h>
-#include <linux/blkdev.h>
+#include <linux/blk-mq.h>


Does this patch require updating the @skipUnless kernel version requirement at line 1297?

Got it. I’ll update the @skipUnless accordingly.

ekyooo · 2025-12-05T02:27:10Z

tests/python/test_clang.py

 #include <linux/mm_types.h>
 int test(struct pt_regs *ctx, struct mm_struct *mm) {
-    return mm->rss_stat.count[MM_ANONPAGES].counter;
+    return mm->rss_stat[MM_ANONPAGES].count;


Does the rss_stat test need @skipUnless?

I’ll adjust as suggested.

ekyooo · 2025-12-05T02:28:43Z

tests/python/test_clang.py

 struct leaf { size_t size; };
 BPF_HASH(simple_map, u32, struct leaf);
-int kprobe____kmalloc(struct pt_regs *ctx, size_t size) {
+int kprobe____kmalloc_noprof(struct pt_regs *ctx, size_t size) {


Does the kmalloc_noprof test need @skipUnless as well?

Got it. I’ll update the @skipUnless accordingly.

ekyooo · 2025-12-05T02:34:27Z

cmake/version.cmake

-    set(GIT_TAG_EXACT "")
+  if(GIT_TAG_LAST MATCHES "-NOTFOUND")
+    set(REVISION "0.0.0+${GIT_SHA1_SHORT}")
+    message(STATUS "Using PEP 440 compliant version for shallow clone")


How about making the message more informative? e.g.,
message(STATUS "No valid tag reachable for shallow clone, using fallback 0.0.0+${GIT_SHA1_SHORT}")

Good point. Since a tarball isn’t a shallow clone, I’ll adjust it to:

"No valid Git tag found, using fallback 0.0.0+${GIT_SHA1_SHORT}"

Thanks!

ekyooo · 2025-12-05T02:51:22Z

Updated GIT_TAG_LAST tag match with -NOTFOUND placeholder to cover other edge cases defined in iovisor/bcc/blob/master/cmake/GitRevisionDescription.cmake.

It's good.

bcc's GitHub workflow has been failing for a while when building from a shallow clone (git clone --depth 1). In this case, git_describe() from CMake returns an invalid version string (128-NOTFOUND-deadbeef). Older setuptools versions used in Fedora 38 tolerated this format, but with Ubuntu 24.04's newer setuptools enforces PEP 440 compliance and treats it as an error, breaking the CI builds. This updates cmake/version.cmake to detect this case and fall back to a compliant 0.0.0+<hash> version when tags are missing. Commit hash is kept for traceability and behavior with full clone hasn't changed. The '-NOTFOUND' placeholder is used to cover other edge cases defined in cmake/GitRevisionDescription.cmake. Signed-off-by: Hoyeon Lee <[email protected]>

Linux kernel commit f3fa33acca9f ("block: remove the ->rq_disk field in struct request") dropped rq_disk in favor of rq->q->disk. The clang tests still include <linux/blkdev.h> and read rq->rq_disk, which fails on recent kernels. 16: include/linux/blkdev.h:34:8: note: forward declaration of 'struct request' 16: struct request; 16: ^ 16: /virtual/main.c:15:12: error: incomplete definition of type 'struct request' 16: if (!rq->rq_disk || rq->rq_disk->major != 5 || 16: ~~^ Since blk-mq became default in linux 5.x, this commit includes <linux/blk-mq.h> and fetch the disk via rq->q->disk so the tests stay valid on current kernels. Signed-off-by: Hoyeon Lee <[email protected]>

Linux Kernel commit f1a7941243c1 ("mm: convert mm's rss stats into percpu_counter") replaced the old rss_stat.count[] layout. The clang test still dereferences mm->rss_stat.count[MM_ANONPAGES].counter, which is old approach. 16: E../virtual/main.c:4:24: error: member reference base type 'struct percpu_counter[4]' is not a structure or union 16: return mm->rss_stat.count[MM_ANONPAGES].counter; 16: ~~~~~~~~~~~~^~~~~~ 16: 1 error generated. This commit change access with mm->rss_stat[MM_ANONPAGES].count instead so the test stay valid on modern kernels. Signed-off-by: Hoyeon Lee <[email protected]>

As part of the Memory Allocation Profiling patchset, Linux commit 7bd230a26648 ("mm/slab: enable slab allocation tagging for kmalloc and friends") converted __kmalloc into a wrapper and moved the actual implementation to __kmalloc_noprof, breaking kmalloc kprobes. This commit update the clang/tools_smoke python test to attach to __kmalloc_noprof instead, restoring the kmalloc kprobe test behavior stay valid on modern kernels. Signed-off-by: Hoyeon Lee <[email protected]>

btrfsdist/btrfsslower/f2fsslower fails on systems without the specific modules. Check for a loaded btrfs or f2fs module before invoking the scripts so the smoke test only runs when specific module is loaded. Signed-off-by: Hoyeon Lee <[email protected]>

The smoke tool test runs tool directly, and especially softirqslower.py lacked the exec bit and failed with "Permission denied". Set the mode to 0755 so the script can be executed by the tests. 29: .... timeout: failed to run command ‘/bcc/tools/softirqslower.py’: Permission denied Signed-off-by: Hoyeon Lee <[email protected]>

HoyeonRhee requested review from brendangregg, chenhengqi, ekyooo and yonghong-song as code owners November 25, 2025 19:56

HoyeonRhee force-pushed the bcc-ci-fix branch 2 times, most recently from a0c8bfc to 833a594 Compare November 25, 2025 20:33

ekyooo reviewed Nov 27, 2025

View reviewed changes

cmake/version.cmake Outdated Show resolved Hide resolved

HoyeonRhee force-pushed the bcc-ci-fix branch from 833a594 to d2b204e Compare November 28, 2025 14:21

ekyooo reviewed Dec 5, 2025

View reviewed changes

HoyeonRhee force-pushed the bcc-ci-fix branch 2 times, most recently from 38550b3 to d545980 Compare December 5, 2025 11:04

HoyeonRhee added 6 commits December 5, 2025 20:06

HoyeonRhee force-pushed the bcc-ci-fix branch from d545980 to 12c2beb Compare December 5, 2025 11:07

Fix BCC CI: shallow-clone versioning, clang/tools_smoke test regressions #5428

Are you sure you want to change the base?

Fix BCC CI: shallow-clone versioning, clang/tools_smoke test regressions #5428

Uh oh!

Conversation

HoyeonRhee commented Nov 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

HoyeonRhee commented Nov 28, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HoyeonRhee Dec 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ekyooo commented Dec 5, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

HoyeonRhee commented Nov 25, 2025 •

edited

Loading

HoyeonRhee Dec 5, 2025 •

edited

Loading