Linux Documentation
 help / color / mirror / Atom feed
* [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release
@ 2026-11-19 19:25 Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 1/9] Compiler Attributes: Add __must_init macro Ian Rogers
                   ` (7 more replies)
  0 siblings, 8 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

Clang supports the 'require_explicit_initialization' attribute on struct
fields to enforce that the field is explicitly initialized when an
aggregate object is constructed, emitting -Wuninitialized-explicit-init
(enabled by default) if the field is omitted.

Following discussion at the Linux Plumbers Conference [1], patch 1 adds
the attribute to include/linux/compiler_attributes.h as __must_init. In
v1 it was called __require_explicit_initialization. Kees picked up that
patch with a first seq_file user [2], and in that thread Greg suggested
the shorter __must_init spelling, which Kees and Miguel agreed to. As
the attribute did not land in the v7.4 merge window, Kees asked that
this series carry the renamed attribute.

As Documentation/core-api/kobject.rst notes:
  "One important point cannot be overstated: every kobject must have a
  release() method, and the kobject must persist (in a consistent state)
  until that method is called. If these constraints are not met, the
  code is flawed."

Annotating struct kobj_type::release caught struct kobj_type definitions
across the tree that omitted .release:
- Patches 2-8 fix struct kobj_type definitions where dynamically
  allocated kobjects, ksets, or enclosing structures were either leaked
  on kobject_put(), freed prematurely before kobject release, or missing
  proper static initializers.
- Patch 9 annotates struct kobj_type::release with __must_init and
  explicitly sets .release = NULL on the remaining legacy struct
  kobj_type definitions, distinguishing built-in sites whose backing
  storage remains valid for the lifetime of the kernel (despite
  violating the kobject API contract) from sites with broken kobject
  lifetime management (marked with FIXME comments for follow-up fixes).

Danilo Krummrich offered to route the series as a whole through the
driver-core tree [3], and all subsystem maintainer Acked-by/Reviewed-by
tags for patches 1-8 have now been collected. Follow-up patches are
planned for the broken-lifetime sites documented in patch 9.

Note on tool use: An LLM coding assistant was used to help audit the
compiler diagnostics across the tree, draft the fixes and changelog
descriptions, and verify builds and review feedback.

Changes in v4:
- Rebased on v7.4-rc3.
- Patch 2 (powerpc/secvar): Add Reviewed-by from Srish Srinivasan and
  Acked-by from Madhavan Srinivasan.
- Patch 3 (sh: sq): Add Acked-by from John Paul Adrian Glaubitz.
- Patch 5 (dlm): Clarify subject ("Wait for kobject release before
  freeing dlm_ls") and update Fixes tag to commit 8a4cf500f1dd ("dlm:
  drop kobject release callback handling") (reported by Sashiko).
- Patch 6 (xen/sys-hypervisor): Add Reviewed-by from Juergen Gross.
- Patch 7 (ubifs): Update Fixes tag to commit 2e3cbf425804 ("ubifs:
  Export filesystem error counters") (reported by Sashiko).
- Patch 9 (kobject): Drop the "Pre-existing:" comment prefix and
  explicitly distinguish built-in sites whose storage outlives kobject
  cleanup (arm64 cpuinfo, ACPI hotplug profiles, s390 sclp_mem) from
  drivers and tristate modules with broken kobject lifetime management,
  documenting each lifetime hazard with a FIXME comment for follow-up
  fixes (requested by Danilo Krummrich).

Changes in v3:
- Rebased on v7.4-rc1.
- Note in cover letter that patches 2 (powerpc/secvar), 3 (sh: sq), and
  6 (xen/sys-hypervisor) are still awaiting subsystem Acked-by tags for
  merging through the driver-core tree, and CC those maintainers on the
  cover letter.
- Patch 1: Add Reviewed-by from Kees Cook; clarify that LLVM issue
  #232483 affects Clang versions prior to its upstream fix (pointed out
  by Nathan Chancellor).
- Patch 4 (cpufreq): Drop obsolete reference to cpufreq_governor_dbs in
  the commit message per Zhongqiu Han; add Reviewed-by from Zhongqiu Han
  and Acked-by from Rafael J. Wysocki and Viresh Kumar.
- Patch 5 (dlm): Add Acked-by from Alexander Aring.
- Patch 8 (drm/amdkfd): Add Reviewed-by from Felix Kuehling.
- Patch 9 (kobject): Fix omapfb manager-sysfs.c and overlay-sysfs.c
  comments to describe managers[] and overlays[] as dynamically
  allocated rather than static and note the CONFIG_DEBUG_KOBJECT_RELEASE
  hazard, fix a struct-vs-member wording nit in usnic_ib_sysfs.c, and
  generalize the commit message note on CONFIG_DEBUG_KOBJECT_RELEASE
  (reported by Sashiko).

Changes in v2:
- Rebased on v7.3-16928-g83f32c773255 (v7.4 merge window).
- Patch 1: Rename __require_explicit_initialization to __must_init per
  Greg Kroah-Hartman; document the Clang >= 20 requirement per Miguel
  Ojeda; note the Clang gap for file-scope objects without an
  initializer reported by Kees Cook; add tags from v1 and from Kees's
  posting.
- Move LLM assistance attribution from Assisted-by tags to this cover
  letter per commit 1e035297ffda ("docs: remove Assisted-by
  references").
- Patch 4 (cpufreq): Use "cpufreq: governor:" prefix, drop Fixes tag,
  fix typo, and remove the redundant runtime assignment in
  cpufreq_dbs_governor_init() per Zhongqiu Han.
- Dropped v1 patch 5 (ACPI: CPPC), already fixed upstream by
  commit 27954e29d7de ("ACPI: CPPC: Release CPC descriptors through
  kobject").
- Patch 5 (dlm): Call kobject_del() before dlm_recoverd_stop() in
  release_lockspace() and the new_lockspace() error path, and wait on a
  new ls_kobj_done completion before freeing the lockspace or dropping
  the module reference, avoiding races with delayed kobject release.
- Dropped v1 patch 8 (s390/sclp_mem) per Sumanth Korikkar: the
  /sys/firmware/memory/memoryX kobjects are permanent, so sclp_mem.c is
  folded into patch 9 with .release = NULL.
- Patch 7 (ubifs): Add Tested-by and Reviewed-by from Zhihao Cheng.
- Patch 8 (drm/amdkfd): Check q->kobj.state_in_sysfs before
  kobject_del() in kfd_procfs_del_queue().
- Patch 9 (kobject): Cover sclp_mem.c and the new amdgpu_ualink.c
  ualink_xcp_info_ktype, fix the overlay-sysfs.c comment, and prefix
  legacy .release = NULL comments with "Pre-existing:".
- Patch 9 (kobject): The ioat, kunpeng_hccs and omapfb display comments
  now say that the existing teardown can use-after-free under
  CONFIG_DEBUG_KOBJECT_RELEASE (reported by Sashiko). Fixing those
  drivers is left to follow-up patches.

v3: https://lore.kernel.org/all/20261108223913.3624571-1-irogers@google.com/
v2: https://lore.kernel.org/all/20261027223847.652884-1-irogers@google.com/
v1: https://lore.kernel.org/all/20261013223946.1836757-1-irogers@google.com/
[1] https://lpc.events/event/20/contributions/2392/
[2] https://lore.kernel.org/lkml/20261014225811.i.852-kees@kernel.org/
[3] https://lore.kernel.org/all/DMLWX8NY1N1N.2OMQMQCIVRR18@kernel.org/

Ian Rogers (9):
  Compiler Attributes: Add __must_init macro
  powerpc/secvar: Add missing kobject release callback to secvar_ktype
  sh: sq: Add missing kobject release callback to ktype_percpu_entry
  cpufreq: governor: Initialize kobj_type release callback statically
  dlm: Wait for kobject release before freeing dlm_ls
  xen/sys-hypervisor: Add release callback to hyp_sysfs_kobj_type
  ubifs: sysfs: Use kset_create_and_add() for ubifs_kset
  drm/amdkfd: Free queue in procfs_queue_type kobject release callback
  kobject: Require explicit initialization of struct kobj_type::release

 Documentation/conf.py                         |  1 +
 arch/arm64/kernel/cpuinfo.c                   |  7 +++++
 arch/powerpc/kernel/secvar-sysfs.c            |  6 +++++
 arch/sh/kernel/cpu/sh4/sq.c                   |  6 +++++
 drivers/acpi/sysfs.c                          |  7 +++++
 drivers/cpufreq/cpufreq_conservative.c        |  5 +++-
 drivers/cpufreq/cpufreq_governor.c            |  4 +--
 drivers/cpufreq/cpufreq_governor.h            |  1 +
 drivers/cpufreq/cpufreq_ondemand.c            |  5 +++-
 drivers/dma/ioat/sysfs.c                      |  8 ++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.c    |  8 ++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c       | 17 ++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_process.c      | 12 ++++++---
 drivers/gpu/drm/amd/amdkfd/kfd_queue.c        |  5 +++-
 drivers/infiniband/hw/hfi1/sysfs.c            |  7 +++++
 drivers/infiniband/hw/usnic/usnic_ib_sysfs.c  |  7 +++++
 drivers/parisc/pdc_stable.c                   |  8 ++++++
 drivers/power/supply/ab8500_fg.c              |  8 ++++++
 drivers/s390/char/sclp_mem.c                  |  6 +++++
 drivers/soc/hisilicon/kunpeng_hccs.c          | 18 +++++++++++++
 .../fbdev/omap2/omapfb/dss/display-sysfs.c    |  7 +++++
 .../fbdev/omap2/omapfb/dss/manager-sysfs.c    |  8 ++++++
 .../fbdev/omap2/omapfb/dss/overlay-sysfs.c    |  7 +++++
 drivers/xen/sys-hypervisor.c                  |  6 +++++
 fs/dlm/dlm_internal.h                         |  1 +
 fs/dlm/lockspace.c                            | 26 ++++++++++++++-----
 fs/erofs/sysfs.c                              | 14 ++++++++++
 fs/f2fs/sysfs.c                               | 21 +++++++++++++++
 fs/ocfs2/cluster/masklog.c                    |  7 +++++
 fs/ubifs/sysfs.c                              | 24 +++++------------
 include/linux/compiler_attributes.h           | 12 +++++++++
 include/linux/kobject.h                       |  2 +-
 scripts/checkpatch.pl                         |  1 +
 tools/lib/python/kdoc/xforms_lists.py         |  1 +
 34 files changed, 251 insertions(+), 32 deletions(-)


base-commit: aebb7274476403040d582c61478df8f4a7bf9cd3
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply	[flat|nested] 9+ messages in thread

* [PATCH v4 1/9] Compiler Attributes: Add __must_init macro
  2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
@ 2026-11-19 19:25 ` Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 2/9] powerpc/secvar: Add missing kobject release callback to secvar_ktype Ian Rogers
                   ` (6 subsequent siblings)
  7 siblings, 0 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

Clang supports the 'require_explicit_initialization' attribute on struct
fields to indicate that a field must be initialized explicitly by the
user when the object is constructed. When an aggregate initializer or
uninitialized variable omits an annotated field, Clang emits a
-Wuninitialized-explicit-init warning (enabled by default).

Add the __must_init macro to include/linux/compiler_attributes.h (which
is included by include/linux/compiler_types.h and
include/linux/compiler.h) so kernel structs with error-prone
initialization (such as mandatory callbacks or cleanup handlers) can
enforce explicit initialization at compile time, as discussed at the
Linux Plumbers Conference. The shorter __must_init spelling, rather than
the attribute's full name, keeps annotated struct members readable and
was suggested by Greg Kroah-Hartman.

Note that Clang versions prior to the fix for
https://github.com/llvm/llvm-project/issues/232483 do not diagnose
file-scope objects declared without any initializer (initializers at
file scope that omit the field are diagnosed).

Also teach Sphinx (Documentation/conf.py), kernel-doc
(tools/lib/python/kdoc/xforms_lists.py), and checkpatch.pl about the new
attribute macro.

Link: https://lpc.events/event/20/contributions/2392/
Link: https://lore.kernel.org/lkml/2026101532-fantastic-deafness-e04c@gregkh/
Acked-by: Miguel Ojeda <ojeda@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Reviewed-by: Kees Cook <kees@kernel.org>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 Documentation/conf.py                 |  1 +
 include/linux/compiler_attributes.h   | 12 ++++++++++++
 scripts/checkpatch.pl                 |  1 +
 tools/lib/python/kdoc/xforms_lists.py |  1 +
 4 files changed, 15 insertions(+)

diff --git a/Documentation/conf.py b/Documentation/conf.py
index 5ab26ce89daf..baaa4035258c 100644
--- a/Documentation/conf.py
+++ b/Documentation/conf.py
@@ -203,6 +203,7 @@ c_id_attributes = [
     "__noreturn",
     "__packed",
     "__pure",
+    "__must_init",
     "__section",
     "__always_unused",
     "__maybe_unused",
diff --git a/include/linux/compiler_attributes.h b/include/linux/compiler_attributes.h
index 476c4c560d17..b9c0b56dd22e 100644
--- a/include/linux/compiler_attributes.h
+++ b/include/linux/compiler_attributes.h
@@ -322,6 +322,18 @@
  */
 #define __pure                          __attribute__((__pure__))
 
+/*
+ * Optional: not supported by gcc
+ * Optional: only supported since clang >= 20.0
+ *
+ * clang: https://clang.llvm.org/docs/AttributeReference.html#require-explicit-initialization
+ */
+#if __has_attribute(__require_explicit_initialization__)
+# define __must_init __attribute__((__require_explicit_initialization__))
+#else
+# define __must_init
+#endif
+
 /*
  *   gcc: https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#index-section-function-attribute
  *   gcc: https://gcc.gnu.org/onlinedocs/gcc/Common-Variable-Attributes.html#index-section-variable-attribute
diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
index b8b64f720191..6a091c45b1a1 100755
--- a/scripts/checkpatch.pl
+++ b/scripts/checkpatch.pl
@@ -6888,6 +6888,7 @@ sub process {
 				"noreturn"			=> "__noreturn",
 				"packed"			=> "__packed",
 				"pure"				=> "__pure",
+				"require_explicit_initialization" => "__must_init",
 				"section"			=> "__section",
 				"used"				=> "__used",
 				"weak"				=> "__weak"
diff --git a/tools/lib/python/kdoc/xforms_lists.py b/tools/lib/python/kdoc/xforms_lists.py
index db23fb650ea8..42956b98aae3 100644
--- a/tools/lib/python/kdoc/xforms_lists.py
+++ b/tools/lib/python/kdoc/xforms_lists.py
@@ -36,6 +36,7 @@ class CTransforms:
         (CMatch("CRYPTO_MINALIGN_ATTR"), ""),
         (CMatch("__private"), ""),
         (CMatch("__rcu"), ""),
+        (CMatch("__must_init"), ""),
         (CMatch("____cacheline_aligned_in_smp"), ""),
         (CMatch("____cacheline_aligned"), ""),
         (CMatch("__cacheline_group_(?:begin|end)"), ""),
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply related	[flat|nested] 9+ messages in thread

* [PATCH v4 2/9] powerpc/secvar: Add missing kobject release callback to secvar_ktype
  2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 1/9] Compiler Attributes: Add __must_init macro Ian Rogers
@ 2026-11-19 19:25 ` Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 3/9] sh: sq: Add missing kobject release callback to ktype_percpu_entry Ian Rogers
                   ` (5 subsequent siblings)
  7 siblings, 0 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

In add_var(), a struct kobject is dynamically allocated with
kzalloc_obj() and initialized with secvar_ktype. If kobject_add() or
sysfs_create_file() fails, kobject_put(kobj) is called to clean up.
However, secvar_ktype does not define a .release callback, which results
in a memory leak of kobj (and a pr_debug warning in kobject_cleanup()
when CONFIG_DEBUG_KOBJECT_RELEASE is enabled).

Provide a secvar_release() callback that calls kfree(kobj).

Fixes: bd5d9c743d38 ("powerpc: expose secure variables to userspace via sysfs")
Reviewed-by: Srish Srinivasan <ssrish@linux.ibm.com>
Acked-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 arch/powerpc/kernel/secvar-sysfs.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/arch/powerpc/kernel/secvar-sysfs.c b/arch/powerpc/kernel/secvar-sysfs.c
index 395399bbab2c..f3f6ac25ea78 100644
--- a/arch/powerpc/kernel/secvar-sysfs.c
+++ b/arch/powerpc/kernel/secvar-sysfs.c
@@ -126,7 +126,13 @@ static const struct attribute_group secvar_attr_group = {
 };
 __ATTRIBUTE_GROUPS(secvar_attr);
 
+static void secvar_release(struct kobject *kobj)
+{
+	kfree(kobj);
+}
+
 static const struct kobj_type secvar_ktype = {
+	.release	= secvar_release,
 	.sysfs_ops	= &kobj_sysfs_ops,
 	.default_groups = secvar_attr_groups,
 };
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply related	[flat|nested] 9+ messages in thread

* [PATCH v4 3/9] sh: sq: Add missing kobject release callback to ktype_percpu_entry
  2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 1/9] Compiler Attributes: Add __must_init macro Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 2/9] powerpc/secvar: Add missing kobject release callback to secvar_ktype Ian Rogers
@ 2026-11-19 19:25 ` Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 4/9] cpufreq: governor: Initialize kobj_type release callback statically Ian Rogers
                   ` (4 subsequent siblings)
  7 siblings, 0 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

In sq_dev_add(), sq_kobject[cpu] is dynamically allocated with
kzalloc_obj() and initialized with ktype_percpu_entry. When
sq_dev_remove() calls kobject_put(kobj), the kobject is never freed
because ktype_percpu_entry lacks a .release callback.

Provide a sq_kobj_release() callback that calls kfree(kobj).

Fixes: d7c30c682a27 ("sh: Store Queue API rework.")
Acked-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 arch/sh/kernel/cpu/sh4/sq.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/arch/sh/kernel/cpu/sh4/sq.c b/arch/sh/kernel/cpu/sh4/sq.c
index 908a8e09113b..f33864958487 100644
--- a/arch/sh/kernel/cpu/sh4/sq.c
+++ b/arch/sh/kernel/cpu/sh4/sq.c
@@ -331,7 +331,13 @@ static const struct sysfs_ops sq_sysfs_ops = {
 	.store	= sq_sysfs_store,
 };
 
+static void sq_kobj_release(struct kobject *kobj)
+{
+	kfree(kobj);
+}
+
 static struct kobj_type ktype_percpu_entry = {
+	.release	= sq_kobj_release,
 	.sysfs_ops	= &sq_sysfs_ops,
 	.default_groups	= sq_sysfs_groups,
 };
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply related	[flat|nested] 9+ messages in thread

* [PATCH v4 4/9] cpufreq: governor: Initialize kobj_type release callback statically
  2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
                   ` (2 preceding siblings ...)
  2026-11-19 19:25 ` [PATCH v4 3/9] sh: sq: Add missing kobject release callback to ktype_percpu_entry Ian Rogers
@ 2026-11-19 19:25 ` Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 5/9] dlm: Wait for kobject release before freeing dlm_ls Ian Rogers
                   ` (3 subsequent siblings)
  7 siblings, 0 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

Export cpufreq_dbs_data_release() and initialize .kobj_type.release
statically in the ondemand and conservative governors instead of
assigning it at runtime in cpufreq_dbs_governor_init().

This prepares for annotating struct kobj_type::release with
__must_init.

Reviewed-by: Zhongqiu Han <zhongqiu.han@oss.qualcomm.com>
Acked-by: Rafael J. Wysocki (Intel) <rafael@kernel.org> # cpufreq
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 drivers/cpufreq/cpufreq_conservative.c | 5 ++++-
 drivers/cpufreq/cpufreq_governor.c     | 4 ++--
 drivers/cpufreq/cpufreq_governor.h     | 1 +
 drivers/cpufreq/cpufreq_ondemand.c     | 5 ++++-
 4 files changed, 11 insertions(+), 4 deletions(-)

diff --git a/drivers/cpufreq/cpufreq_conservative.c b/drivers/cpufreq/cpufreq_conservative.c
index 02bfd46543e9..146a2353b1a2 100644
--- a/drivers/cpufreq/cpufreq_conservative.c
+++ b/drivers/cpufreq/cpufreq_conservative.c
@@ -318,7 +318,10 @@ static void cs_limits(struct cpufreq_policy *policy)
 
 static struct dbs_governor cs_governor = {
 	.gov = CPUFREQ_DBS_GOVERNOR_INITIALIZER("conservative"),
-	.kobj_type = { .default_groups = cs_groups },
+	.kobj_type = {
+		.release = cpufreq_dbs_data_release,
+		.default_groups = cs_groups,
+	},
 	.gov_dbs_update = cs_dbs_update,
 	.alloc = cs_alloc,
 	.free = cs_free,
diff --git a/drivers/cpufreq/cpufreq_governor.c b/drivers/cpufreq/cpufreq_governor.c
index f3bd3dbe7c41..f8b5aca09b6b 100644
--- a/drivers/cpufreq/cpufreq_governor.c
+++ b/drivers/cpufreq/cpufreq_governor.c
@@ -419,7 +419,7 @@ static void free_policy_dbs_info(struct policy_dbs_info *policy_dbs,
 	gov->free(policy_dbs);
 }
 
-static void cpufreq_dbs_data_release(struct kobject *kobj)
+void cpufreq_dbs_data_release(struct kobject *kobj)
 {
 	struct dbs_data *dbs_data = to_dbs_data(to_gov_attr_set(kobj));
 	struct dbs_governor *gov = dbs_data->gov;
@@ -427,6 +427,7 @@ static void cpufreq_dbs_data_release(struct kobject *kobj)
 	gov->exit(dbs_data);
 	kfree(dbs_data);
 }
+EXPORT_SYMBOL_GPL(cpufreq_dbs_data_release);
 
 int cpufreq_dbs_governor_init(struct cpufreq_policy *policy)
 {
@@ -488,7 +489,6 @@ int cpufreq_dbs_governor_init(struct cpufreq_policy *policy)
 	policy->governor_data = policy_dbs;
 
 	gov->kobj_type.sysfs_ops = &governor_sysfs_ops;
-	gov->kobj_type.release = cpufreq_dbs_data_release;
 	ret = kobject_init_and_add(&dbs_data->attr_set.kobj, &gov->kobj_type,
 				   get_governor_parent_kobj(policy),
 				   "%s", gov->gov.name);
diff --git a/drivers/cpufreq/cpufreq_governor.h b/drivers/cpufreq/cpufreq_governor.h
index 806d8fb4dff1..1026712bb79b 100644
--- a/drivers/cpufreq/cpufreq_governor.h
+++ b/drivers/cpufreq/cpufreq_governor.h
@@ -150,6 +150,7 @@ static inline struct dbs_governor *dbs_governor_of(struct cpufreq_policy *policy
 }
 
 /* Governor callback routines */
+void cpufreq_dbs_data_release(struct kobject *kobj);
 int cpufreq_dbs_governor_init(struct cpufreq_policy *policy);
 void cpufreq_dbs_governor_exit(struct cpufreq_policy *policy);
 int cpufreq_dbs_governor_start(struct cpufreq_policy *policy);
diff --git a/drivers/cpufreq/cpufreq_ondemand.c b/drivers/cpufreq/cpufreq_ondemand.c
index 9942dbb38dae..7386d043c665 100644
--- a/drivers/cpufreq/cpufreq_ondemand.c
+++ b/drivers/cpufreq/cpufreq_ondemand.c
@@ -374,7 +374,10 @@ static struct od_ops od_ops = {
 
 static struct dbs_governor od_dbs_gov = {
 	.gov = CPUFREQ_DBS_GOVERNOR_INITIALIZER("ondemand"),
-	.kobj_type = { .default_groups = od_groups },
+	.kobj_type = {
+		.release = cpufreq_dbs_data_release,
+		.default_groups = od_groups,
+	},
 	.gov_dbs_update = od_dbs_update,
 	.alloc = od_alloc,
 	.free = od_free,
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply related	[flat|nested] 9+ messages in thread

* [PATCH v4 5/9] dlm: Wait for kobject release before freeing dlm_ls
  2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
                   ` (3 preceding siblings ...)
  2026-11-19 19:25 ` [PATCH v4 4/9] cpufreq: governor: Initialize kobj_type release callback statically Ian Rogers
@ 2026-11-19 19:25 ` Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 6/9] xen/sys-hypervisor: Add release callback to hyp_sysfs_kobj_type Ian Rogers
                   ` (2 subsequent siblings)
  7 siblings, 0 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

struct dlm_ls embeds a struct kobject (ls_kobj), but dlm_ktype lacked a
.release callback and dlm freed the lockspace structure directly via
kfree(ls). With CONFIG_DEBUG_KOBJECT_RELEASE=y, freeing ls before the
delayed kobject release runs causes a use-after-free, and unloading the
module while the delayed release is pending results in an instruction
page fault.

Add a completion ls_kobj_done and dlm_ls_release() to complete it when
the last kobject reference is dropped, waiting for it in free_lockspace()
before kfree(ls) and in the new_lockspace() error path before module_put().
Also delete the kobject via kobject_del() before stopping dlm_recoverd
to close the window where concurrent sysfs writes could wake the stopped
recovery task.

Fixes: 8a4cf500f1dd ("dlm: drop kobject release callback handling")
Acked-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 fs/dlm/dlm_internal.h |  1 +
 fs/dlm/lockspace.c    | 26 ++++++++++++++++++++------
 2 files changed, 21 insertions(+), 6 deletions(-)

diff --git a/fs/dlm/dlm_internal.h b/fs/dlm/dlm_internal.h
index 74c2e77d1b55..95b3bff0cb57 100644
--- a/fs/dlm/dlm_internal.h
+++ b/fs/dlm/dlm_internal.h
@@ -576,6 +576,7 @@ struct dlm_ls {
 	int			ls_create_count; /* create/release refcount */
 	unsigned long		ls_flags;	/* LSFL_ */
 	struct kobject		ls_kobj;
+	struct completion	ls_kobj_done;
 
 	struct xarray		ls_lkbxa;
 	rwlock_t		ls_lkbxa_lock;
diff --git a/fs/dlm/lockspace.c b/fs/dlm/lockspace.c
index a9c98b4f378f..2dccd080a10f 100644
--- a/fs/dlm/lockspace.c
+++ b/fs/dlm/lockspace.c
@@ -179,7 +179,15 @@ static const struct sysfs_ops dlm_attr_ops = {
 	.store = dlm_attr_store,
 };
 
+static void dlm_ls_release(struct kobject *kobj)
+{
+	struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj);
+
+	complete(&ls->ls_kobj_done);
+}
+
 static struct kobj_type dlm_ktype = {
+	.release        = dlm_ls_release,
 	.default_groups = dlm_groups,
 	.sysfs_ops     = &dlm_attr_ops,
 };
@@ -355,6 +363,8 @@ static void free_lockspace(struct work_struct *work)
 	 */
 	rhashtable_free_and_destroy(&ls->ls_rsbtbl, rhash_free_rsb, NULL);
 
+	kobject_put(&ls->ls_kobj);
+	wait_for_completion(&ls->ls_kobj_done);
 	kfree(ls);
 }
 
@@ -435,7 +445,10 @@ static int new_lockspace(const char *name, const char *cluster,
 	ls->ls_lvblen = lvblen;
 	atomic_set(&ls->ls_count, 0);
 	init_waitqueue_head(&ls->ls_count_wait);
+	init_completion(&ls->ls_kobj_done);
 	ls->ls_flags = 0;
+	ls->ls_kobj.kset = dlm_kset;
+	kobject_init(&ls->ls_kobj, &dlm_ktype);
 
 	if (ops && dlm_config.ci_recover_callbacks) {
 		ls->ls_ops = ops;
@@ -567,9 +580,7 @@ static int new_lockspace(const char *name, const char *cluster,
 	wait_event(ls->ls_recover_lock_wait,
 		   test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags));
 
-	ls->ls_kobj.kset = dlm_kset;
-	error = kobject_init_and_add(&ls->ls_kobj, &dlm_ktype, NULL,
-				     "%s", ls->ls_name);
+	error = kobject_add(&ls->ls_kobj, NULL, "%s", ls->ls_name);
 	if (error)
 		goto out_recoverd;
 	kobject_uevent(&ls->ls_kobj, KOBJ_ADD);
@@ -582,7 +593,7 @@ static int new_lockspace(const char *name, const char *cluster,
 
 	error = do_uevent(ls, 1, 0);
 	if (error < 0)
-		goto out_recoverd;
+		goto out_kobj;
 
 	/* wait until recovery is successful or failed */
 	wait_for_completion(&ls->ls_recovery_done);
@@ -600,6 +611,8 @@ static int new_lockspace(const char *name, const char *cluster,
 	do_uevent(ls, 0, 0);
 	dlm_clear_members(ls);
 	kfree(ls->ls_node_array);
+ out_kobj:
+	kobject_del(&ls->ls_kobj);
  out_recoverd:
 	dlm_recoverd_stop(ls);
  out_callback:
@@ -615,6 +628,7 @@ static int new_lockspace(const char *name, const char *cluster,
 	rhashtable_destroy(&ls->ls_rsbtbl);
  out_lsfree:
 	kobject_put(&ls->ls_kobj);
+	wait_for_completion(&ls->ls_kobj_done);
 	kfree(ls);
  out:
 	module_put(THIS_MODULE);
@@ -740,6 +754,8 @@ static int release_lockspace(struct dlm_ls *ls, unsigned int release_option)
 	    dlm_user_daemon_available())
 		do_uevent(ls, 0, (release_option == DLM_RELEASE_RECOVER));
 
+	kobject_del(&ls->ls_kobj);
+
 	dlm_recoverd_stop(ls);
 
 	/* clear the LSFL_RUNNING flag to fast up
@@ -759,8 +775,6 @@ static int release_lockspace(struct dlm_ls *ls, unsigned int release_option)
 
 	dlm_delete_debug_file(ls);
 
-	kobject_put(&ls->ls_kobj);
-
 	xa_destroy(&ls->ls_recover_xa);
 	kfree(ls->ls_recover_buf);
 
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply related	[flat|nested] 9+ messages in thread

* [PATCH v4 6/9] xen/sys-hypervisor: Add release callback to hyp_sysfs_kobj_type
  2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
                   ` (4 preceding siblings ...)
  2026-11-19 19:25 ` [PATCH v4 5/9] dlm: Wait for kobject release before freeing dlm_ls Ian Rogers
@ 2026-11-19 19:25 ` Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 7/9] ubifs: sysfs: Use kset_create_and_add() for ubifs_kset Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 8/9] drm/amdkfd: Free queue in procfs_queue_type kobject release callback Ian Rogers
  7 siblings, 0 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

hypervisor_init() in drivers/base/hypervisor.c creates hypervisor_kobj
with kobject_create_and_add() (which allocates a dynamic kobject using
dynamic_kobj_ktype), and hypervisor_subsys_init() in
drivers/xen/sys-hypervisor.c then overwrites hypervisor_kobj->ktype with
&hyp_sysfs_kobj_type, which had no .release callback.

Add hyp_sysfs_release() calling kfree(kobj) to preserve the dynamic
kobject release behavior.

Fixes: cff7e81b3dd7 ("xen: add /sys/hypervisor support")
Reviewed-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 drivers/xen/sys-hypervisor.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/xen/sys-hypervisor.c b/drivers/xen/sys-hypervisor.c
index 91923242a5ae..0e8a2cd69fd6 100644
--- a/drivers/xen/sys-hypervisor.c
+++ b/drivers/xen/sys-hypervisor.c
@@ -659,7 +659,13 @@ static const struct sysfs_ops hyp_sysfs_ops = {
 	.store = hyp_sysfs_store,
 };
 
+static void hyp_sysfs_release(struct kobject *kobj)
+{
+	kfree(kobj);
+}
+
 static const struct kobj_type hyp_sysfs_kobj_type = {
+	.release = hyp_sysfs_release,
 	.sysfs_ops = &hyp_sysfs_ops,
 };
 
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply related	[flat|nested] 9+ messages in thread

* [PATCH v4 7/9] ubifs: sysfs: Use kset_create_and_add() for ubifs_kset
  2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
                   ` (5 preceding siblings ...)
  2026-11-19 19:25 ` [PATCH v4 6/9] xen/sys-hypervisor: Add release callback to hyp_sysfs_kobj_type Ian Rogers
@ 2026-11-19 19:25 ` Ian Rogers
  2026-11-19 19:25 ` [PATCH v4 8/9] drm/amdkfd: Free queue in procfs_queue_type kobject release callback Ian Rogers
  7 siblings, 0 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

ubifs_kset was a static struct kset using a custom ubifs_ktype that
had no default_groups and no .release callback. With
CONFIG_DEBUG_KOBJECT_RELEASE=y, unloading the ubifs module while delayed
kobject release is pending causes a use-after-free on the static kset.

Replace the static ubifs_kset and empty ubifs_ktype with a dynamically
allocated kset via kset_create_and_add().

Fixes: 2e3cbf425804 ("ubifs: Export filesystem error counters")
Tested-by: Zhihao Cheng <chengzhihao1@huawei.com>
Reviewed-by: Zhihao Cheng <chengzhihao1@huawei.com>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 fs/ubifs/sysfs.c | 24 +++++++-----------------
 1 file changed, 7 insertions(+), 17 deletions(-)

diff --git a/fs/ubifs/sysfs.c b/fs/ubifs/sysfs.c
index dbf967dcf26e..e20931e3a62f 100644
--- a/fs/ubifs/sysfs.c
+++ b/fs/ubifs/sysfs.c
@@ -80,13 +80,7 @@ static const struct kobj_type ubifs_sb_ktype = {
 	.release	= ubifs_sb_release,
 };
 
-static const struct kobj_type ubifs_ktype = {
-	.sysfs_ops	= &ubifs_attr_ops,
-};
-
-static struct kset ubifs_kset = {
-	.kobj	= {.ktype = &ubifs_ktype},
-};
+static struct kset *ubifs_kset;
 
 int ubifs_sysfs_register(struct ubifs_info *c)
 {
@@ -107,7 +101,7 @@ int ubifs_sysfs_register(struct ubifs_info *c)
 		goto out_free;
 	}
 
-	c->kobj.kset = &ubifs_kset;
+	c->kobj.kset = ubifs_kset;
 	init_completion(&c->kobj_unregister);
 
 	ret = kobject_init_and_add(&c->kobj, &ubifs_sb_ktype, NULL,
@@ -139,18 +133,14 @@ void ubifs_sysfs_unregister(struct ubifs_info *c)
 
 int __init ubifs_sysfs_init(void)
 {
-	int ret;
+	ubifs_kset = kset_create_and_add("ubifs", NULL, fs_kobj);
+	if (!ubifs_kset)
+		return -ENOMEM;
 
-	kobject_set_name(&ubifs_kset.kobj, "ubifs");
-	ubifs_kset.kobj.parent = fs_kobj;
-	ret = kset_register(&ubifs_kset);
-	if (ret)
-		kset_put(&ubifs_kset);
-
-	return ret;
+	return 0;
 }
 
 void ubifs_sysfs_exit(void)
 {
-	kset_unregister(&ubifs_kset);
+	kset_unregister(ubifs_kset);
 }
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply related	[flat|nested] 9+ messages in thread

* [PATCH v4 8/9] drm/amdkfd: Free queue in procfs_queue_type kobject release callback
  2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
                   ` (6 preceding siblings ...)
  2026-11-19 19:25 ` [PATCH v4 7/9] ubifs: sysfs: Use kset_create_and_add() for ubifs_kset Ian Rogers
@ 2026-11-19 19:25 ` Ian Rogers
  7 siblings, 0 replies; 9+ messages in thread
From: Ian Rogers @ 2026-11-19 19:25 UTC (permalink / raw)
  To: Danilo Krummrich, Greg Kroah-Hartman, Kees Cook, Miguel Ojeda,
	Rafael J . Wysocki
  Cc: Madhavan Srinivasan, Srish Srinivasan, Michael Ellerman,
	Nicholas Piggin, Christophe Leroy, Nayna Jain, Mimi Zohar,
	Yoshinori Sato, Rich Felker, John Paul Adrian Glaubitz,
	Juergen Gross, Stefano Stabellini, Oleksandr Tyshchenko,
	Nathan Chancellor, Nick Desaulniers, Viresh Kumar, Zhongqiu Han,
	Alexander Aring, Felix Kuehling, Christian König,
	Zhihao Cheng, Sumanth Korikkar, Bill Wendling, Justin Stitt,
	driver-core, llvm, linux-hardening, linux-fsdevel, linux-doc,
	linux-acpi, linux-pm, linuxppc-dev, linux-sh, linux-s390, gfs2,
	xen-devel, linux-mtd, amd-gfx, dri-devel, linux-kernel,
	Ian Rogers

struct queue embeds a struct kobject (q->kobj) for procfs entries, but
procfs_queue_type had no .release callback and uninit_queue() freed q
immediately after kfd_procfs_del_queue() called kobject_put(&q->kobj).
With CONFIG_DEBUG_KOBJECT_RELEASE=y, this causes a use-after-free.

Add kfd_procfs_queue_release() to free struct queue from the kobject
release callback, change kfd_procfs_del_queue() to only remove the
kobject from sysfs via kobject_del(), and drop the final reference in
uninit_queue() via kobject_put(). Check q->kobj.state_in_sysfs before
calling kobject_del() in kfd_procfs_del_queue() to avoid warnings if
sysfs creation previously failed.

Fixes: 6d220a7e7971 ("drm/amdkfd: Add queue information to sysfs")
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_process.c | 12 +++++++++---
 drivers/gpu/drm/amd/amdkfd/kfd_queue.c   |  5 ++++-
 2 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 6663718fb83c..b349e0d8d47b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -512,7 +512,15 @@ static const struct sysfs_ops procfs_queue_ops = {
 	.show = kfd_procfs_queue_show,
 };
 
+static void kfd_procfs_queue_release(struct kobject *kobj)
+{
+	struct queue *q = container_of(kobj, struct queue, kobj);
+
+	kfree(q);
+}
+
 static const struct kobj_type procfs_queue_type = {
+	.release = kfd_procfs_queue_release,
 	.sysfs_ops = &procfs_queue_ops,
 	.default_groups = procfs_queue_groups,
 };
@@ -552,7 +560,6 @@ int kfd_procfs_add_queue(struct queue *q)
 	if (ret < 0) {
 		pr_warn("Creating proc/<pid>/queues/%u failed",
 			q->properties.queue_id);
-		kobject_put(&q->kobj);
 		return ret;
 	}
 
@@ -697,11 +704,10 @@ static void kfd_procfs_add_sysfs_files(struct kfd_process *p)
 
 void kfd_procfs_del_queue(struct queue *q)
 {
-	if (!q || !q->process->kobj)
+	if (!q || !q->process || !q->process->kobj || !q->kobj.state_in_sysfs)
 		return;
 
 	kobject_del(&q->kobj);
-	kobject_put(&q->kobj);
 }
 
 int kfd_process_create_wq(void)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_queue.c
index cd41b10e0873..636cc894223d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_queue.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_queue.c
@@ -83,7 +83,10 @@ int init_queue(struct queue **q, const struct queue_properties *properties)
 
 void uninit_queue(struct queue *q)
 {
-	kfree(q);
+	if (q && q->kobj.state_initialized)
+		kobject_put(&q->kobj);
+	else
+		kfree(q);
 }
 
 #if IS_ENABLED(CONFIG_HSA_AMD_SVM)
-- 
2.56.1.708.g77570a6a1d-goog


^ permalink raw reply related	[flat|nested] 9+ messages in thread

end of thread, other threads:[~2026-11-19 19:26 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-19 19:25 [PATCH v4 0/9] Add __must_init and require explicit kobj_type::release Ian Rogers
2026-11-19 19:25 ` [PATCH v4 1/9] Compiler Attributes: Add __must_init macro Ian Rogers
2026-11-19 19:25 ` [PATCH v4 2/9] powerpc/secvar: Add missing kobject release callback to secvar_ktype Ian Rogers
2026-11-19 19:25 ` [PATCH v4 3/9] sh: sq: Add missing kobject release callback to ktype_percpu_entry Ian Rogers
2026-11-19 19:25 ` [PATCH v4 4/9] cpufreq: governor: Initialize kobj_type release callback statically Ian Rogers
2026-11-19 19:25 ` [PATCH v4 5/9] dlm: Wait for kobject release before freeing dlm_ls Ian Rogers
2026-11-19 19:25 ` [PATCH v4 6/9] xen/sys-hypervisor: Add release callback to hyp_sysfs_kobj_type Ian Rogers
2026-11-19 19:25 ` [PATCH v4 7/9] ubifs: sysfs: Use kset_create_and_add() for ubifs_kset Ian Rogers
2026-11-19 19:25 ` [PATCH v4 8/9] drm/amdkfd: Free queue in procfs_queue_type kobject release callback Ian Rogers

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox