NACK: [SRU][N][PATCH 00/13] Coresight fails to build on 6.8.0-101 due to 'atomic_t' member accessed as 'int'

[Noah Wager]

On Fri, Feb 20, 2026 at 04:49:40PM -0800, Noah Wager wrote:
> BugLink: https://bugs.launchpad.net/bugs/2142336
>
> SRU Justification:
>
> [Impact]
>
> When CONFIG_CORESIGHT is enabled, noble 6.8.0-101 fails to build with the following errors:
>
> ```
> drivers/hwtracing/coresight/coresight-catu.c: In function 'catu_enable':
> drivers/hwtracing/coresight/coresight-catu.c:465:27: error: invalid operands to binary == (have 'atomic_t' and 'int')
>   465 | if (csdev->refcnt == 0) {
>       | ~~~~~~~~~~~~~ ^~
>       | |
>       | atomic_t
> drivers/hwtracing/coresight/coresight-catu.c:471:30: error: wrong type argument to increment
>   471 | csdev->refcnt++;
>       | ^~
> drivers/hwtracing/coresight/coresight-catu.c: In function 'catu_disable':
> drivers/hwtracing/coresight/coresight-catu.c:498:13: error: wrong type argument to decrement
>   498 | if (--csdev->refcnt == 0) {
>       | ^~
> drivers/hwtracing/coresight/coresight-catu.c: In function 'catu_init':
> drivers/hwtracing/coresight/coresight-catu.c:728:15: error: implicit declaration of function 'coresight_init_driver' [-Werror=implicit-function-declaration]
>   728 | ret = coresight_init_driver("catu", &catu_driver, &catu_platform_driver);
>       | ^~~~~~~~~~~~~~~~~~~~~
> ```
>
> These errors are a result of including the following patch in upstream stable
> updates, but missing some dependencies from the original patch set:
>
>   Error description: refcnt accessed as int but defined as atomic_t
>   Break commit: n/linux 0e2198d37507 ("coresight: catu: Introduce refcount and spinlock for enabling/disabling")
>   Missing commit: 4545b38ef004 ("coresight: Remove atomic type from refcnt")
>   Patch series: "coresight: Separate sysfs and Perf usage and some other cleanups"
>                 (https://lists.infradead.org/pipermail/linux-arm-kernel/2024-January/899269.html)
>
> [Fix]
>
> Cherry-pick upstream commit:
> 4545b38ef004 ("coresight: Remove atomic type from refcnt")
> which involves cherry-picking the commits in the series:
> "coresight: Separate sysfs and Perf usage and some other cleanups"
> up to the fix commit.
>
> NOTE: I decided it would be best to backport the entire patch series
> because the rest of the patches, while not *strictly* necessary for a
> successful build, add some clean up and helpers that do not change
> behaviour and may be useful when pulling in future upstream commits,
> as we will be less likely to miss a dependency. That means the full list
> of cherry-picks is:
>
> a0fef3f05cf3 ("coresight: Make language around "activated" sinks consistent")
> a11ebe138b8d ("coresight: Remove ops callback checks")
> 9cae77cf23e3 ("coresight: Move mode to struct coresight_device")
> d5e83f97eb56 ("coresight: Remove the 'enable' field.")
> 1f5149c7751c ("coresight: Move all sysfs code to sysfs file")
> 4545b38ef004 ("coresight: Remove atomic type from refcnt")
> 053ad9ad1d13 ("coresight: Remove unused stubs")
> 812265e26ed3 ("coresight: Add explicit member initializers to coresight_dev_type")
> d724f65218b9 ("coresight: Add helper for atomically taking the device")
> c95c2733e5fe ("coresight: Add a helper for getting csdev->mode")
> bcaabb95f0c9 ("coresight: Add helper for setting csdev->mode")
>
> NOTE 2: To apply the patches cleanly, I had to revert the already-applied
> "coresight-etb10: change etb_drvdata spinlock's type to raw_spinlock_t"
> and then re-cherry-pick it from upstream, where it applies cleanly. I found
> this approach safer than mangling all the new commits to fix context changes.
>
> [Test Plan]
>
> Ensure CONFIG_CORESIGHT and CONFIG_CORESIGHT_CATU are enabled, then compile test.
>
> [Where problems could occur]
>
> There is a low chance of regression since this patchset is purely cleanup and
> does not modify observable behaviour, as stated by the author in the original
> patch set. Additionally, coresight failed to build before this, so this is
> a guaranteed improvement in that respect. These changes do alter some function
> and struct definitions, which could cause build failures in downstream projects
> that have non-upstream coresight patches reliant on the legacy definitions.
>
> James Clark (11):
>   coresight: Make language around "activated" sinks consistent
>   coresight: Remove ops callback checks
>   coresight: Move mode to struct coresight_device
>   coresight: Remove the 'enable' field.
>   coresight: Move all sysfs code to sysfs file
>   coresight: Remove atomic type from refcnt
>   coresight: Remove unused stubs
>   coresight: Add explicit member initializers to coresight_dev_type
>   coresight: Add helper for atomically taking the device
>   coresight: Add a helper for getting csdev->mode
>   coresight: Add helper for setting csdev->mode
>
> Noah Wager (1):
>   Revert "coresight-etb10: change etb_drvdata spinlock's type to
>     raw_spinlock_t"
>
> Yeoreum Yun (1):
>   coresight-etb10: change etb_drvdata spinlock's type to raw_spinlock_t
>
>  drivers/hwtracing/coresight/coresight-core.c  | 490 +-----------------
>  drivers/hwtracing/coresight/coresight-etb10.c |  29 +-
>  drivers/hwtracing/coresight/coresight-etm.h   |   2 -
>  .../coresight/coresight-etm3x-core.c          |  23 +-
>  .../coresight/coresight-etm3x-sysfs.c         |   4 +-
>  .../coresight/coresight-etm4x-core.c          |  26 +-
>  drivers/hwtracing/coresight/coresight-etm4x.h |   1 -
>  drivers/hwtracing/coresight/coresight-priv.h  |   7 +-
>  drivers/hwtracing/coresight/coresight-stm.c   |  30 +-
>  drivers/hwtracing/coresight/coresight-sysfs.c | 391 ++++++++++++++
>  .../hwtracing/coresight/coresight-tmc-core.c  |   2 +-
>  .../hwtracing/coresight/coresight-tmc-etf.c   |  46 +-
>  .../hwtracing/coresight/coresight-tmc-etr.c   |  33 +-
>  drivers/hwtracing/coresight/coresight-tmc.h   |   2 -
>  drivers/hwtracing/coresight/coresight-tpda.c  |  13 +-
>  drivers/hwtracing/coresight/coresight-tpiu.c  |  14 +-
>  drivers/hwtracing/coresight/ultrasoc-smb.c    |  22 +-
>  drivers/hwtracing/coresight/ultrasoc-smb.h    |   2 -
>  include/linux/coresight.h                     | 146 ++----
>  19 files changed, 594 insertions(+), 689 deletions(-)
>
> --
> 2.43.0
>

As per Stefan's comment in this thread, cherry picking the entire
patchset is excessive to fix this issue. This PR needs to be reworked to
a more minimal set of changes, so I will NACK myself and send a V2.