* [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes @ 2026-10-02 4:15 Omokefe Emmanuel Onanaroghene 2026-10-02 4:28 ` Randy Dunlap 2026-10-14 17:22 ` Jonathan Corbet 0 siblings, 2 replies; 12+ messages in thread From: Omokefe Emmanuel Onanaroghene @ 2026-10-02 4:15 UTC (permalink / raw) To: linux-doc, linux-kernel Cc: corbet, skhan, rdunlap, mchehab, aleksandr.loktionov, bvanassche, emmaonana18, tudor.ambarus kernel-doc fails to parse functions whose names are wrapped in the DOMAIN_NS() namespace macro, e.g. in drivers/iommu/generic_pt/iommu_pt.h:217: ... expecting prototype for iova_to_phys(). Prototype was for DOMAIN_NS() instead function_xforms is the designated place for unwrapping function-name macros, but it has no entry for DOMAIN_NS(), so the macro name is kept as the prototype and the real name leaks into the parameter list. Add a transform expanding DOMAIN_NS(op) to op, following the existing DEFINE_IDTENTRY_IRQ precedent. Add a regression test for the wrapped iova_to_phys() declaration. Signed-off-by: Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> --- tools/lib/python/kdoc/xforms_lists.py | 6 ++++++ tools/unittests/test_kdoc_parser.py | 22 ++++++++++++++++++++++ 2 files changed, 28 insertions(+) diff --git a/tools/lib/python/kdoc/xforms_lists.py b/tools/lib/python/kdoc/xforms_lists.py index e3dda2fe8a53..5e8bd9117895 100644 --- a/tools/lib/python/kdoc/xforms_lists.py +++ b/tools/lib/python/kdoc/xforms_lists.py @@ -91,6 +91,12 @@ class CTransforms: (CMatch("__diagnose_as"), ""), (CMatch("DECL_BUCKET_PARAMS"), r"\1, \2"), (CMatch("DEFINE_IDTENTRY_IRQ"), r"static void \1(struct pt_regs *regs, u32 vector)"), + # + # DOMAIN_NS() wraps a function name with a namespace prefix + # (see drivers/iommu/generic_pt/iommu_pt.h). Unwrap it so the + # real function name is parsed. + # + (CMatch("DOMAIN_NS"), r"\1"), (CMatch("__cond_acquires"), ""), (CMatch("__cond_releases"), ""), (CMatch("__acquires"), ""), diff --git a/tools/unittests/test_kdoc_parser.py b/tools/unittests/test_kdoc_parser.py index c4a76ed13dbc..4b47a858824a 100755 --- a/tools/unittests/test_kdoc_parser.py +++ b/tools/unittests/test_kdoc_parser.py @@ -418,6 +418,28 @@ class TestSelfValidate(GenerateKdocItem): """ self.run_test(self.SOURCE, [self.DEFAULT.copy()], self.EXPORTS) +class TestDomainNs(GenerateKdocItem): + def test_domain_ns_wrapped_function(self): + source = """ + /** + * iova_to_phys() - Return the output address for the given IOVA + * @domain: Table to query + * @iova: IO virtual address to query + */ + phys_addr_t DOMAIN_NS(iova_to_phys)(struct iommu_domain *domain, + dma_addr_t iova); + """ + + kernel_doc = KernelDoc(self.config, "test.c", self.xforms) + patcher = patch('builtins.open', new_callable=mock_open, + read_data=dedent(source)) + with patcher: + _, entries = kernel_doc.parse_kdoc() + + self.assertEqual(len(entries), 1) + self.assertEqual(entries[0].name, "iova_to_phys") + self.assertEqual(entries[0].warnings, []) + # # Class and logic to create dynamic tests from YAML # -- 2.43.0 ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes 2026-10-02 4:15 [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes Omokefe Emmanuel Onanaroghene @ 2026-10-02 4:28 ` Randy Dunlap 2026-10-14 17:22 ` Jonathan Corbet 1 sibling, 0 replies; 12+ messages in thread From: Randy Dunlap @ 2026-10-02 4:28 UTC (permalink / raw) To: Omokefe Emmanuel Onanaroghene, linux-doc, linux-kernel Cc: corbet, skhan, mchehab, aleksandr.loktionov, bvanassche, tudor.ambarus On 10/1/26 9:15 PM, Omokefe Emmanuel Onanaroghene wrote: > kernel-doc fails to parse functions whose names are wrapped in the > DOMAIN_NS() namespace macro, e.g. in > drivers/iommu/generic_pt/iommu_pt.h:217: > > ... expecting prototype for iova_to_phys(). Prototype was for > DOMAIN_NS() instead > > function_xforms is the designated place for unwrapping function-name > macros, but it has no entry for DOMAIN_NS(), so the macro name is kept > as the prototype and the real name leaks into the parameter list. > > Add a transform expanding DOMAIN_NS(op) to op, following the existing > DEFINE_IDTENTRY_IRQ precedent. > > Add a regression test for the wrapped iova_to_phys() declaration. > > Signed-off-by: Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> Tested-by: Randy Dunlap <rdunlap@infradead.org> Acked-by: Randy Dunlap <rdunlap@infradead.org> Thanks. > --- > tools/lib/python/kdoc/xforms_lists.py | 6 ++++++ > tools/unittests/test_kdoc_parser.py | 22 ++++++++++++++++++++++ > 2 files changed, 28 insertions(+) > > diff --git a/tools/lib/python/kdoc/xforms_lists.py b/tools/lib/python/kdoc/xforms_lists.py > index e3dda2fe8a53..5e8bd9117895 100644 > --- a/tools/lib/python/kdoc/xforms_lists.py > +++ b/tools/lib/python/kdoc/xforms_lists.py > @@ -91,6 +91,12 @@ class CTransforms: > (CMatch("__diagnose_as"), ""), > (CMatch("DECL_BUCKET_PARAMS"), r"\1, \2"), > (CMatch("DEFINE_IDTENTRY_IRQ"), r"static void \1(struct pt_regs *regs, u32 vector)"), > + # > + # DOMAIN_NS() wraps a function name with a namespace prefix > + # (see drivers/iommu/generic_pt/iommu_pt.h). Unwrap it so the > + # real function name is parsed. > + # > + (CMatch("DOMAIN_NS"), r"\1"), > (CMatch("__cond_acquires"), ""), > (CMatch("__cond_releases"), ""), > (CMatch("__acquires"), ""), > diff --git a/tools/unittests/test_kdoc_parser.py b/tools/unittests/test_kdoc_parser.py > index c4a76ed13dbc..4b47a858824a 100755 > --- a/tools/unittests/test_kdoc_parser.py > +++ b/tools/unittests/test_kdoc_parser.py > @@ -418,6 +418,28 @@ class TestSelfValidate(GenerateKdocItem): > """ > self.run_test(self.SOURCE, [self.DEFAULT.copy()], self.EXPORTS) > > +class TestDomainNs(GenerateKdocItem): > + def test_domain_ns_wrapped_function(self): > + source = """ > + /** > + * iova_to_phys() - Return the output address for the given IOVA > + * @domain: Table to query > + * @iova: IO virtual address to query > + */ > + phys_addr_t DOMAIN_NS(iova_to_phys)(struct iommu_domain *domain, > + dma_addr_t iova); > + """ > + > + kernel_doc = KernelDoc(self.config, "test.c", self.xforms) > + patcher = patch('builtins.open', new_callable=mock_open, > + read_data=dedent(source)) > + with patcher: > + _, entries = kernel_doc.parse_kdoc() > + > + self.assertEqual(len(entries), 1) > + self.assertEqual(entries[0].name, "iova_to_phys") > + self.assertEqual(entries[0].warnings, []) > + > # > # Class and logic to create dynamic tests from YAML > # -- ~Randy ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes 2026-10-02 4:15 [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes Omokefe Emmanuel Onanaroghene 2026-10-02 4:28 ` Randy Dunlap @ 2026-10-14 17:22 ` Jonathan Corbet 2026-10-14 22:47 ` Omokefe Emmanuel Onanaroghene 1 sibling, 1 reply; 12+ messages in thread From: Jonathan Corbet @ 2026-10-14 17:22 UTC (permalink / raw) To: Omokefe Emmanuel Onanaroghene, linux-doc, linux-kernel Cc: skhan, rdunlap, mchehab, aleksandr.loktionov, bvanassche, emmaonana18, tudor.ambarus Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> writes: > kernel-doc fails to parse functions whose names are wrapped in the > DOMAIN_NS() namespace macro, e.g. in > drivers/iommu/generic_pt/iommu_pt.h:217: > > ... expecting prototype for iova_to_phys(). Prototype was for > DOMAIN_NS() instead > > function_xforms is the designated place for unwrapping function-name > macros, but it has no entry for DOMAIN_NS(), so the macro name is kept > as the prototype and the real name leaks into the parameter list. > > Add a transform expanding DOMAIN_NS(op) to op, following the existing > DEFINE_IDTENTRY_IRQ precedent. > > Add a regression test for the wrapped iova_to_phys() declaration. > > Signed-off-by: Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> > --- > tools/lib/python/kdoc/xforms_lists.py | 6 ++++++ > tools/unittests/test_kdoc_parser.py | 22 ++++++++++++++++++++++ > 2 files changed, 28 insertions(+) Neither of these patches applies to docs-next...what did you base them on? jon ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes 2026-10-14 17:22 ` Jonathan Corbet @ 2026-10-14 22:47 ` Omokefe Emmanuel Onanaroghene 2026-10-15 0:53 ` Omokefe Emmanuel Onanaroghene 0 siblings, 1 reply; 12+ messages in thread From: Omokefe Emmanuel Onanaroghene @ 2026-10-14 22:47 UTC (permalink / raw) To: Jonathan Corbet Cc: linux-doc, linux-kernel, skhan, rdunlap, mchehab, aleksandr.loktionov, bvanassche, tudor.ambarus, emmaonana18 On Wed, 14 Oct 2026 11:22:02 -0600, Jonathan Corbet <corbet@lwn.net> wrote: > Neither of these patches applies to docs-next...what did you base them > on? Hi Jon, I had based both patches on mainline rather than docs-next. I have rebased both patches on top of docs-next and sent a v2 for both: - https://lore.kernel.org/all/20261014223134.14772-1-emmaonana18@gmail.com/ - https://lore.kernel.org/all/20261014223437.16164-1-emmaonana18@gmail.com/ Apologies for the inconvenience. Thanks, Omokefe ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes 2026-10-14 22:47 ` Omokefe Emmanuel Onanaroghene @ 2026-10-15 0:53 ` Omokefe Emmanuel Onanaroghene 2026-10-15 13:00 ` Jonathan Corbet 0 siblings, 1 reply; 12+ messages in thread From: Omokefe Emmanuel Onanaroghene @ 2026-10-15 0:53 UTC (permalink / raw) To: Jonathan Corbet Cc: linux-doc, linux-kernel, skhan, rdunlap, mchehab, aleksandr.loktionov, bvanassche, tudor.ambarus, emmaonana18 On Wed, 14 Oct 2026 11:22:02 -0600, Jonathan Corbet <corbet@lwn.net> wrote: >> Neither of these patches applies to docs-next...what did you base them >> on? > > Hi Jon, > > I had mistakenly based both patches on mainline rather than docs-next. > I have rebased both on top of docs-next and sent a v2 for each: > > - docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser: > https://lore.kernel.org/all/20261014223134.14772-1-emmaonana18@gmail.com/ > > - docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes: > https://lore.kernel.org/all/20261014223437.16164-1-emmaonana18@gmail.com/ > I just noticed another mistake on my part, because I prepared the two v2 patches on separate independent branches off docs-next, both patches add their regression test at the exact same location in tools/unittests/test_kdoc_parser.py (line 444). Applying them sequentially will cause a merge conflict I didn't realize that patches modifying the exact same context in a file should be stacked sequentially on a single branch as a patch series rather than developed on separate branches. Should I respin them as a 2-patch series against docs-next, or would you prefer to resolve when applying? Thanks, Omokefe ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes 2026-10-15 0:53 ` Omokefe Emmanuel Onanaroghene @ 2026-10-15 13:00 ` Jonathan Corbet 2026-10-15 16:44 ` [PATCH v3 0/2] docs: kdoc: fix TRACE_EVENT_FN and DOMAIN_NS parsing Omokefe Emmanuel Onanaroghene 0 siblings, 1 reply; 12+ messages in thread From: Jonathan Corbet @ 2026-10-15 13:00 UTC (permalink / raw) To: Omokefe Emmanuel Onanaroghene Cc: linux-doc, linux-kernel, skhan, rdunlap, mchehab, aleksandr.loktionov, bvanassche, tudor.ambarus, emmaonana18 Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> writes: > On Wed, 14 Oct 2026 11:22:02 -0600, Jonathan Corbet <corbet@lwn.net> wrote: >>> Neither of these patches applies to docs-next...what did you base them >>> on? >> >> Hi Jon, >> >> I had mistakenly based both patches on mainline rather than docs-next. >> I have rebased both on top of docs-next and sent a v2 for each: >> >> - docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser: >> https://lore.kernel.org/all/20261014223134.14772-1-emmaonana18@gmail.com/ >> >> - docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes: >> https://lore.kernel.org/all/20261014223437.16164-1-emmaonana18@gmail.com/ >> > > I just noticed another mistake on my part, because I prepared > the two v2 patches on separate independent branches off docs-next, both > patches add their regression test at the exact same location in > tools/unittests/test_kdoc_parser.py (line 444). Applying them sequentially will > cause a merge conflict > > I didn't realize that patches modifying the exact same context in a file > should be stacked sequentially on a single branch as a patch series > rather than developed on separate branches. > > Should I respin them as a 2-patch series against docs-next, > or would you prefer to resolve when applying? Redo them as a two-patch series if you could. Thanks, jon ^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH v3 0/2] docs: kdoc: fix TRACE_EVENT_FN and DOMAIN_NS parsing 2026-10-15 13:00 ` Jonathan Corbet @ 2026-10-15 16:44 ` Omokefe Emmanuel Onanaroghene 2026-10-15 16:44 ` [PATCH v3 1/2] docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser Omokefe Emmanuel Onanaroghene 2026-10-15 16:44 ` [PATCH v3 2/2] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes Omokefe Emmanuel Onanaroghene 0 siblings, 2 replies; 12+ messages in thread From: Omokefe Emmanuel Onanaroghene @ 2026-10-15 16:44 UTC (permalink / raw) To: Jonathan Corbet Cc: linux-doc, linux-kernel, skhan, rdunlap, mchehab, paulmck, linusw, Julia.Lawall, gary, rmk+kernel, hemanth.selam, emmaonana18 This series fixes two kernel-doc parsing failures where the prototype was misidentified and parameters leaked into garbage: - TRACE_EVENT_FN() family tracepoints (e.g. gpu_power_status) - DOMAIN_NS() wrapped function names (e.g. iova_to_phys) Rebased onto docs-next and stacked as a series per Jon's feedback to avoid both tests landing at the same spot in test_kdoc_parser.py. Omokefe Emmanuel Onanaroghene (2): docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes tools/lib/python/kdoc/kdoc_parser.py | 18 +++++++++- tools/lib/python/kdoc/xforms_lists.py | 6 ++++ tools/unittests/test_kdoc_parser.py | 50 +++++++++++++++++++++++++++ 3 files changed, 73 insertions(+), 1 deletion(-) -- 2.43.0 ^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH v3 1/2] docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser 2026-10-15 16:44 ` [PATCH v3 0/2] docs: kdoc: fix TRACE_EVENT_FN and DOMAIN_NS parsing Omokefe Emmanuel Onanaroghene @ 2026-10-15 16:44 ` Omokefe Emmanuel Onanaroghene 2026-10-15 17:06 ` Mauro Carvalho Chehab 2026-10-15 16:44 ` [PATCH v3 2/2] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes Omokefe Emmanuel Onanaroghene 1 sibling, 1 reply; 12+ messages in thread From: Omokefe Emmanuel Onanaroghene @ 2026-10-15 16:44 UTC (permalink / raw) To: Jonathan Corbet Cc: linux-doc, linux-kernel, skhan, rdunlap, mchehab, paulmck, linusw, Julia.Lawall, gary, rmk+kernel, hemanth.selam, emmaonana18 kernel-doc fails to parse tracepoint declarations that use TRACE_EVENT_FN(), e.g. gpu_power_status: drivers/gpu/drm/panthor/panthor_trace.h:28 ... Unrecognized tracepoint format ... ... expecting prototype for gpu_power_status(). Prototype was for TRACE_EVENT_FN() instead tracepoint_munge() only matches TRACE_EVENT(), so the macro name is kept as the prototype and the TP_PROTO() arguments leak into the parameter list as garbage. Match TRACE_EVENT_FN(), TRACE_EVENT_FN_COND(), TRACE_EVENT_CONDITION() and TRACE_EVENT_SYSCALL() the same way as TRACE_EVENT() and list them explicitly at the call site in process_proto_function(). Add a regression test for the gpu_power_status declaration. Signed-off-by: Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> --- tools/lib/python/kdoc/kdoc_parser.py | 18 +++++++++++++++++- tools/unittests/test_kdoc_parser.py | 28 ++++++++++++++++++++++++++++ 2 files changed, 45 insertions(+), 1 deletion(-) diff --git a/tools/lib/python/kdoc/kdoc_parser.py b/tools/lib/python/kdoc/kdoc_parser.py index 80d4245acd17..a65bda3e81b9 100644 --- a/tools/lib/python/kdoc/kdoc_parser.py +++ b/tools/lib/python/kdoc/kdoc_parser.py @@ -1545,6 +1545,22 @@ class KernelDoc: if r.search(proto): tracepointname = r.group(1) + r = KernRe(r'TRACE_EVENT_FN\((.*?),') + if r.search(proto): + tracepointname = r.group(1) + + r = KernRe(r'TRACE_EVENT_FN_COND\((.*?),') + if r.search(proto): + tracepointname = r.group(1) + + r = KernRe(r'TRACE_EVENT_CONDITION\((.*?),') + if r.search(proto): + tracepointname = r.group(1) + + r = KernRe(r'TRACE_EVENT_SYSCALL\((.*?),') + if r.search(proto): + tracepointname = r.group(1) + r = KernRe(r'DEFINE_SINGLE_EVENT\((.*?),') if r.search(proto): tracepointname = r.group(1) @@ -1604,7 +1620,7 @@ class KernelDoc: self.entry.prototype = self.syscall_munge(ln, self.entry.prototype) else: - r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT') + r = KernRe(r'TRACE_EVENT|TRACE_EVENT_FN|TRACE_EVENT_FN_COND|TRACE_EVENT_CONDITION|TRACE_EVENT_SYSCALL|DEFINE_EVENT|DEFINE_SINGLE_EVENT') if r.search(self.entry.prototype): self.entry.prototype = self.tracepoint_munge(ln, self.entry.prototype) diff --git a/tools/unittests/test_kdoc_parser.py b/tools/unittests/test_kdoc_parser.py index 4baa78ca02b3..04dafae412e4 100755 --- a/tools/unittests/test_kdoc_parser.py +++ b/tools/unittests/test_kdoc_parser.py @@ -444,6 +444,34 @@ class TestContextLockStruct(GenerateKdocItem): self.assertEqual(entries[0].name, "debugfs_cancellation") self.assertEqual(entries[0].warnings, []) +class TestTraceEventFn(GenerateKdocItem): + def test_trace_event_fn_declaration(self): + source = """ + /** + * gpu_power_status - called when hardware turns on or off + * @dev: pointer to device + * @shader_bitmap: bitmap + */ + TRACE_EVENT_FN(gpu_power_status, + TP_PROTO(const struct device *dev, u64 shader_bitmap), + TP_ARGS(dev, shader_bitmap), + TP_STRUCT__entry(__field(u64, shader_bitmap)), + TP_fast_assign(__entry->shader_bitmap = shader_bitmap;), + TP_printk("%s", __get_str(dev_name)), + my_register, my_unregister + ); + """ + + kernel_doc = KernelDoc(self.config, "test.c", self.xforms) + patcher = patch('builtins.open', new_callable=mock_open, + read_data=dedent(source)) + with patcher: + _, entries = kernel_doc.parse_kdoc() + + self.assertEqual(len(entries), 1) + self.assertEqual(entries[0].name, "trace_gpu_power_status") + self.assertEqual(entries[0].warnings, []) + # # Class and logic to create dynamic tests from YAML # -- 2.43.0 ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v3 1/2] docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser 2026-10-15 16:44 ` [PATCH v3 1/2] docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser Omokefe Emmanuel Onanaroghene @ 2026-10-15 17:06 ` Mauro Carvalho Chehab 2026-10-16 20:43 ` [PATCH] " Omokefe Emmanuel Onanaroghene 0 siblings, 1 reply; 12+ messages in thread From: Mauro Carvalho Chehab @ 2026-10-15 17:06 UTC (permalink / raw) To: Omokefe Emmanuel Onanaroghene Cc: Jonathan Corbet, linux-doc, linux-kernel, skhan, rdunlap, mchehab, paulmck, linusw, Julia.Lawall, gary, rmk+kernel, hemanth.selam On Thu, 15 Oct 2026 17:44:47 +0100 Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> wrote: > kernel-doc fails to parse tracepoint declarations that use > TRACE_EVENT_FN(), e.g. gpu_power_status: > > drivers/gpu/drm/panthor/panthor_trace.h:28 ... Unrecognized > tracepoint format ... > ... expecting prototype for gpu_power_status(). Prototype was for > TRACE_EVENT_FN() instead > > tracepoint_munge() only matches TRACE_EVENT(), so the macro name is > kept as the prototype and the TP_PROTO() arguments leak into the > parameter list as garbage. > > Match TRACE_EVENT_FN(), TRACE_EVENT_FN_COND(), > TRACE_EVENT_CONDITION() and TRACE_EVENT_SYSCALL() the same way as > TRACE_EVENT() and list them explicitly at the call site in > process_proto_function(). > > Add a regression test for the gpu_power_status declaration. > > Signed-off-by: Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> > --- > tools/lib/python/kdoc/kdoc_parser.py | 18 +++++++++++++++++- > tools/unittests/test_kdoc_parser.py | 28 ++++++++++++++++++++++++++++ > 2 files changed, 45 insertions(+), 1 deletion(-) > > diff --git a/tools/lib/python/kdoc/kdoc_parser.py b/tools/lib/python/kdoc/kdoc_parser.py > index 80d4245acd17..a65bda3e81b9 100644 > --- a/tools/lib/python/kdoc/kdoc_parser.py > +++ b/tools/lib/python/kdoc/kdoc_parser.py > @@ -1545,6 +1545,22 @@ class KernelDoc: > if r.search(proto): > tracepointname = r.group(1) > > + r = KernRe(r'TRACE_EVENT_FN\((.*?),') > + if r.search(proto): > + tracepointname = r.group(1) > + > + r = KernRe(r'TRACE_EVENT_FN_COND\((.*?),') > + if r.search(proto): > + tracepointname = r.group(1) > + > + r = KernRe(r'TRACE_EVENT_CONDITION\((.*?),') > + if r.search(proto): > + tracepointname = r.group(1) > + > + r = KernRe(r'TRACE_EVENT_SYSCALL\((.*?),') > + if r.search(proto): > + tracepointname = r.group(1) > + > r = KernRe(r'DEFINE_SINGLE_EVENT\((.*?),') > if r.search(proto): > tracepointname = r.group(1) > @@ -1604,7 +1620,7 @@ class KernelDoc: > self.entry.prototype = self.syscall_munge(ln, > self.entry.prototype) > else: > - r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT') > + r = KernRe(r'TRACE_EVENT|TRACE_EVENT_FN|TRACE_EVENT_FN_COND|TRACE_EVENT_CONDITION|TRACE_EVENT_SYSCALL|DEFINE_EVENT|DEFINE_SINGLE_EVENT') The best would be instead to move all of them to xforms_lists, together with SYSCALL_DEFINE. Regards, Mauro > if r.search(self.entry.prototype): > self.entry.prototype = self.tracepoint_munge(ln, > self.entry.prototype) > diff --git a/tools/unittests/test_kdoc_parser.py b/tools/unittests/test_kdoc_parser.py > index 4baa78ca02b3..04dafae412e4 100755 > --- a/tools/unittests/test_kdoc_parser.py > +++ b/tools/unittests/test_kdoc_parser.py > @@ -444,6 +444,34 @@ class TestContextLockStruct(GenerateKdocItem): > self.assertEqual(entries[0].name, "debugfs_cancellation") > self.assertEqual(entries[0].warnings, []) > > +class TestTraceEventFn(GenerateKdocItem): > + def test_trace_event_fn_declaration(self): > + source = """ > + /** > + * gpu_power_status - called when hardware turns on or off > + * @dev: pointer to device > + * @shader_bitmap: bitmap > + */ > + TRACE_EVENT_FN(gpu_power_status, > + TP_PROTO(const struct device *dev, u64 shader_bitmap), > + TP_ARGS(dev, shader_bitmap), > + TP_STRUCT__entry(__field(u64, shader_bitmap)), > + TP_fast_assign(__entry->shader_bitmap = shader_bitmap;), > + TP_printk("%s", __get_str(dev_name)), > + my_register, my_unregister > + ); > + """ > + > + kernel_doc = KernelDoc(self.config, "test.c", self.xforms) > + patcher = patch('builtins.open', new_callable=mock_open, > + read_data=dedent(source)) > + with patcher: > + _, entries = kernel_doc.parse_kdoc() > + > + self.assertEqual(len(entries), 1) > + self.assertEqual(entries[0].name, "trace_gpu_power_status") > + self.assertEqual(entries[0].warnings, []) > + > # > # Class and logic to create dynamic tests from YAML > # Thanks, Mauro ^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH] docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser 2026-10-15 17:06 ` Mauro Carvalho Chehab @ 2026-10-16 20:43 ` Omokefe Emmanuel Onanaroghene 2026-10-16 21:11 ` Mauro Carvalho Chehab 0 siblings, 1 reply; 12+ messages in thread From: Omokefe Emmanuel Onanaroghene @ 2026-10-16 20:43 UTC (permalink / raw) To: Mauro Carvalho Chehab, Jonathan Corbet, linux-doc Cc: linux-kernel, skhan, rdunlap, mchehab, paulmck, linusw, Julia.Lawall, gary, rmk+kernel, hemanth.selam, Omokefe Emmanuel Onanaroghene kernel-doc fails to parse tracepoint declarations that use TRACE_EVENT_FN(), e.g. gpu_power_status: drivers/gpu/drm/panthor/panthor_trace.h:28 ... Unrecognized tracepoint format ... ... expecting prototype for gpu_power_status(). Prototype was for TRACE_EVENT_FN() instead tracepoint_munge() only matches TRACE_EVENT(), so the macro name is kept as the prototype and the TP_PROTO() arguments leak into the parameter list as garbage. Match TRACE_EVENT_FN(), TRACE_EVENT_FN_COND(), TRACE_EVENT_CONDITION() and TRACE_EVENT_SYSCALL() the same way as TRACE_EVENT() Add a regression test for the gpu_power_status declaration. Signed-off-by: Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> --- Move tracepoint handling to xforms_lists, together with SYSCALL_DEFINE as requested by Mauro. Question for Mauro: is adding syscall_munge() and tracepoint_munge() in apply() what you meant, or did you want a pure CMatch? A pure CMatch looks like it would need a tokenizer fix for nested commas, so I'd prefer to keep it this way unless you want the pure CMatch rework. tools/lib/python/kdoc/kdoc_parser.py | 110 ++++++-------------------- tools/lib/python/kdoc/xforms_lists.py | 109 +++++++++++++++++++++++++ tools/unittests/test_kdoc_parser.py | 28 +++++++ 3 files changed, 159 insertions(+), 88 deletions(-) diff --git a/tools/lib/python/kdoc/kdoc_parser.py b/tools/lib/python/kdoc/kdoc_parser.py index 80d4245acd17..a274e06b9f23 100644 --- a/tools/lib/python/kdoc/kdoc_parser.py +++ b/tools/lib/python/kdoc/kdoc_parser.py @@ -1059,8 +1059,27 @@ class KernelDoc: # # Apply the initial transformations. # + # This also handles SYSCALL_DEFINE*() and TRACE_EVENT*()/ + # DEFINE_EVENT*() via CTransforms (see xforms_lists.py). + # prototype = self.xforms.apply("func", prototype) + # + # Tracepoints are documented with their short name, e.g. + # gpu_power_status for trace_gpu_power_status(). + # Accept the short name and store the full trace_ name. + # + r = KernRe(r'\btrace_(\w+)') + if r.search(prototype): + if self.entry.identifier == r.group(1): + self.entry.identifier = f"trace_{self.entry.identifier}" + + + r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT') + if r.search(prototype): + self.emit_msg(ln, + f"Unrecognized tracepoint format:\n{prototype}\n") + # Yes, this truly is vile. We are looking for: # 1. Return type (may be nothing if we're looking at a macro) # 2. Function name @@ -1493,82 +1512,6 @@ class KernelDoc: self.entry.add_text(doc_content.group(1)) # else ... ?? - def syscall_munge(self, ln, proto): # pylint: disable=W0613 - """ - Handle syscall definitions. - """ - - is_void = False - - # Strip newlines/CR's - proto = re.sub(r'[\r\n]+', ' ', proto) - - # Check if it's a SYSCALL_DEFINE0 - if 'SYSCALL_DEFINE0' in proto: - is_void = True - - # Replace SYSCALL_DEFINE with correct return type & function name - proto = KernRe(r'SYSCALL_DEFINE.*\(').sub('long sys_', proto) - - r = KernRe(r'long\s+(sys_.*?),') - if r.search(proto): - proto = KernRe(',').sub('(', proto, count=1) - elif is_void: - proto = KernRe(r'\)').sub('(void)', proto, count=1) - - # Now delete all of the odd-numbered commas in the proto - # so that argument types & names don't have a comma between them - count = 0 - length = len(proto) - - if is_void: - length = 0 # skip the loop if is_void - - for ix in range(length): - if proto[ix] == ',': - count += 1 - if count % 2 == 1: - proto = proto[:ix] + ' ' + proto[ix + 1:] - - return proto - - def tracepoint_munge(self, ln, proto): - """ - Handle tracepoint definitions. - """ - - tracepointname = None - tracepointargs = None - - # Match tracepoint name based on different patterns - r = KernRe(r'TRACE_EVENT\((.*?),') - if r.search(proto): - tracepointname = r.group(1) - - r = KernRe(r'DEFINE_SINGLE_EVENT\((.*?),') - if r.search(proto): - tracepointname = r.group(1) - - r = KernRe(r'DEFINE_EVENT\((.*?),(.*?),') - if r.search(proto): - tracepointname = r.group(2) - - if tracepointname: - tracepointname = tracepointname.lstrip() - - r = KernRe(r'TP_PROTO\((.*?)\)') - if r.search(proto): - tracepointargs = r.group(1) - - if not tracepointname or not tracepointargs: - self.emit_msg(ln, - f"Unrecognized tracepoint format:\n{proto}\n") - else: - proto = f"static inline void trace_{tracepointname}({tracepointargs})" - self.entry.identifier = f"trace_{self.entry.identifier}" - - return proto - def process_proto_function(self, ln, line, source): """Ancillary routine to process a function prototype.""" @@ -1598,18 +1541,9 @@ class KernelDoc: r = KernRe(r'^(\S+\s+)\(\s*\*(\S+)\)') self.entry.prototype = r.sub(r'\1\2', self.entry.prototype) # - # Handle special declaration syntaxes - # - if 'SYSCALL_DEFINE' in self.entry.prototype: - self.entry.prototype = self.syscall_munge(ln, - self.entry.prototype) - else: - r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT') - if r.search(self.entry.prototype): - self.entry.prototype = self.tracepoint_munge(ln, - self.entry.prototype) - # - # ... and we're done + # ... and we're done. SYSCALL_DEFINE*() and TRACE_EVENT*()/ + # DEFINE_EVENT*() are handled via xforms_lists.CTransforms + # inside dump_function(). # self.dump_function(ln, self.entry.prototype, source) self.reset_state(ln) diff --git a/tools/lib/python/kdoc/xforms_lists.py b/tools/lib/python/kdoc/xforms_lists.py index 15f6f21c7a9e..aa82dc927ef1 100644 --- a/tools/lib/python/kdoc/xforms_lists.py +++ b/tools/lib/python/kdoc/xforms_lists.py @@ -17,6 +17,98 @@ class CTransforms: into something we can parse and generate kdoc for. """ + @staticmethod + def syscall_munge(proto): + """ + Handle syscall definitions. + + Turns SYSCALL_DEFINE*() into a plain C prototype, e.g.:: + + SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval) + -> long sys_setpriority(int which, int who, int niceval) + """ + + is_void = False + + # Strip newlines/CR's + proto = re.sub(r'[\r\n]+', ' ', proto) + + # Check if it's a SYSCALL_DEFINE0 + if 'SYSCALL_DEFINE0' in proto: + is_void = True + + # Replace SYSCALL_DEFINE with correct return type & function name + proto = KernRe(r'SYSCALL_DEFINE.*\(').sub('long sys_', proto) + + r = KernRe(r'long\s+(sys_.*?),') + if r.search(proto): + proto = KernRe(',').sub('(', proto, count=1) + elif is_void: + proto = KernRe(r'\)').sub('(void)', proto, count=1) + + # Now delete all of the odd-numbered commas in the proto + # so that argument types & names don't have a comma between them + count = 0 + length = len(proto) + + if is_void: + length = 0 # skip the loop if is_void + + for ix in range(length): + if proto[ix] == ',': + count += 1 + if count % 2 == 1: + proto = proto[:ix] + ' ' + proto[ix + 1:] + + return proto + + @staticmethod + def tracepoint_munge(proto): + """ + Handle tracepoint definitions. + + Turns TRACE_EVENT*()/DEFINE_EVENT*() into a plain C prototype, + e.g.:: + + TRACE_EVENT_FN(gpu_power_status, TP_PROTO(...), ...) + -> static inline void trace_gpu_power_status(...) + + Returns the original string if the tracepoint format is not + recognized, letting the caller emit a warning. + """ + + tracepointname = None + tracepointargs = None + + # Match tracepoint name based on different patterns. + # NOTE: DEFINE_EVENT*() takes a template name first, so the + # event name is the second argument. + for pattern, group in [ + (r'TRACE_EVENT\((.*?),', 1), + (r'TRACE_EVENT_FN\((.*?),', 1), + (r'TRACE_EVENT_FN_COND\((.*?),', 1), + (r'TRACE_EVENT_CONDITION\((.*?),', 1), + (r'TRACE_EVENT_SYSCALL\((.*?),', 1), + (r'DEFINE_SINGLE_EVENT\((.*?),', 1), + (r'DEFINE_EVENT\((.*?),(.*?),', 2), + ]: + r = KernRe(pattern) + if r.search(proto): + tracepointname = r.group(group) + + if tracepointname: + tracepointname = tracepointname.lstrip() + + r = KernRe(r'TP_PROTO\((.*?)\)') + if r.search(proto): + tracepointargs = r.group(1) + + if not tracepointname or not tracepointargs: + return proto + + return f"static inline void trace_{tracepointname}({tracepointargs})" + + # # NOTE: # Due to performance reasons, place CMatch rules before KernRe, @@ -147,6 +239,23 @@ class CTransforms: if xforms_type not in self.xforms: return source + # + # Handle special declaration syntaxes first, so that the + # declarative transforms below see plain C prototypes. + # + if xforms_type == "func": + source_str = str(source) + if 'SYSCALL_DEFINE' in source_str: + source_str = self.syscall_munge(source_str) + source = source_str + else: + r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT') + if r.search(source_str): + munged = self.tracepoint_munge(source_str) + # tracepoint_munge() returns the input unchanged + # when the format is not recognized. + source = munged + if isinstance(source, str): source = CTokenizer(source) diff --git a/tools/unittests/test_kdoc_parser.py b/tools/unittests/test_kdoc_parser.py index 4baa78ca02b3..04dafae412e4 100755 --- a/tools/unittests/test_kdoc_parser.py +++ b/tools/unittests/test_kdoc_parser.py @@ -444,6 +444,34 @@ class TestContextLockStruct(GenerateKdocItem): self.assertEqual(entries[0].name, "debugfs_cancellation") self.assertEqual(entries[0].warnings, []) +class TestTraceEventFn(GenerateKdocItem): + def test_trace_event_fn_declaration(self): + source = """ + /** + * gpu_power_status - called when hardware turns on or off + * @dev: pointer to device + * @shader_bitmap: bitmap + */ + TRACE_EVENT_FN(gpu_power_status, + TP_PROTO(const struct device *dev, u64 shader_bitmap), + TP_ARGS(dev, shader_bitmap), + TP_STRUCT__entry(__field(u64, shader_bitmap)), + TP_fast_assign(__entry->shader_bitmap = shader_bitmap;), + TP_printk("%s", __get_str(dev_name)), + my_register, my_unregister + ); + """ + + kernel_doc = KernelDoc(self.config, "test.c", self.xforms) + patcher = patch('builtins.open', new_callable=mock_open, + read_data=dedent(source)) + with patcher: + _, entries = kernel_doc.parse_kdoc() + + self.assertEqual(len(entries), 1) + self.assertEqual(entries[0].name, "trace_gpu_power_status") + self.assertEqual(entries[0].warnings, []) + # # Class and logic to create dynamic tests from YAML # -- 2.43.0 ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser 2026-10-16 20:43 ` [PATCH] " Omokefe Emmanuel Onanaroghene @ 2026-10-16 21:11 ` Mauro Carvalho Chehab 0 siblings, 0 replies; 12+ messages in thread From: Mauro Carvalho Chehab @ 2026-10-16 21:11 UTC (permalink / raw) To: Omokefe Emmanuel Onanaroghene Cc: Jonathan Corbet, linux-doc, linux-kernel, skhan, rdunlap, mchehab, paulmck, linusw, Julia.Lawall, gary, rmk+kernel, hemanth.selam On Fri, 16 Oct 2026 21:43:35 +0100 Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> wrote: > kernel-doc fails to parse tracepoint declarations that use > TRACE_EVENT_FN(), e.g. gpu_power_status: > > drivers/gpu/drm/panthor/panthor_trace.h:28 ... Unrecognized > tracepoint format ... > ... expecting prototype for gpu_power_status(). Prototype was for > TRACE_EVENT_FN() instead > > tracepoint_munge() only matches TRACE_EVENT(), so the macro name is > kept as the prototype and the TP_PROTO() arguments leak into the > parameter list as garbage. > > Match TRACE_EVENT_FN(), TRACE_EVENT_FN_COND(), > TRACE_EVENT_CONDITION() and TRACE_EVENT_SYSCALL() the same way as > TRACE_EVENT() > > Add a regression test for the gpu_power_status declaration. > > Signed-off-by: Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> > --- > Move tracepoint handling to xforms_lists, together with SYSCALL_DEFINE > as requested by Mauro. > > Question for Mauro: is adding syscall_munge() and > tracepoint_munge() in apply() what you meant, or > did you want a pure CMatch? I meant a pure CMatch. One of the hardest things to maintain at kernel-doc are those complex regex logic spread all over it. The goal for CMatch and the C tokenizer is to avoid adding more - and instead - to reduce the existing ones over time. In the specific case of commas, the tokenizer, ideally double commas should be generating a singe STRING token and single commas a single CHAR. > A pure CMatch looks > like it would need a tokenizer fix for nested > commas, so I'd prefer to keep it this way unless > you want the pure CMatch rework. If the nested commas logic there is broken, indeed it needs a fix, on a separate patch. > > tools/lib/python/kdoc/kdoc_parser.py | 110 ++++++-------------------- > tools/lib/python/kdoc/xforms_lists.py | 109 +++++++++++++++++++++++++ > tools/unittests/test_kdoc_parser.py | 28 +++++++ > 3 files changed, 159 insertions(+), 88 deletions(-) > > diff --git a/tools/lib/python/kdoc/kdoc_parser.py b/tools/lib/python/kdoc/kdoc_parser.py > index 80d4245acd17..a274e06b9f23 100644 > --- a/tools/lib/python/kdoc/kdoc_parser.py > +++ b/tools/lib/python/kdoc/kdoc_parser.py > @@ -1059,8 +1059,27 @@ class KernelDoc: > # > # Apply the initial transformations. > # > + # This also handles SYSCALL_DEFINE*() and TRACE_EVENT*()/ > + # DEFINE_EVENT*() via CTransforms (see xforms_lists.py). > + # > prototype = self.xforms.apply("func", prototype) > > + # > + # Tracepoints are documented with their short name, e.g. > + # gpu_power_status for trace_gpu_power_status(). > + # Accept the short name and store the full trace_ name. > + # > + r = KernRe(r'\btrace_(\w+)') > + if r.search(prototype): > + if self.entry.identifier == r.group(1): > + self.entry.identifier = f"trace_{self.entry.identifier}" > + > + > + r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT') > + if r.search(prototype): > + self.emit_msg(ln, > + f"Unrecognized tracepoint format:\n{prototype}\n") > + > # Yes, this truly is vile. We are looking for: > # 1. Return type (may be nothing if we're looking at a macro) > # 2. Function name > @@ -1493,82 +1512,6 @@ class KernelDoc: > self.entry.add_text(doc_content.group(1)) > # else ... ?? > > - def syscall_munge(self, ln, proto): # pylint: disable=W0613 > - """ > - Handle syscall definitions. > - """ > - > - is_void = False > - > - # Strip newlines/CR's > - proto = re.sub(r'[\r\n]+', ' ', proto) > - > - # Check if it's a SYSCALL_DEFINE0 > - if 'SYSCALL_DEFINE0' in proto: > - is_void = True > - > - # Replace SYSCALL_DEFINE with correct return type & function name > - proto = KernRe(r'SYSCALL_DEFINE.*\(').sub('long sys_', proto) > - > - r = KernRe(r'long\s+(sys_.*?),') > - if r.search(proto): > - proto = KernRe(',').sub('(', proto, count=1) > - elif is_void: > - proto = KernRe(r'\)').sub('(void)', proto, count=1) > - > - # Now delete all of the odd-numbered commas in the proto > - # so that argument types & names don't have a comma between them > - count = 0 > - length = len(proto) > - > - if is_void: > - length = 0 # skip the loop if is_void > - > - for ix in range(length): > - if proto[ix] == ',': > - count += 1 > - if count % 2 == 1: > - proto = proto[:ix] + ' ' + proto[ix + 1:] > - > - return proto > - > - def tracepoint_munge(self, ln, proto): > - """ > - Handle tracepoint definitions. > - """ > - > - tracepointname = None > - tracepointargs = None > - > - # Match tracepoint name based on different patterns > - r = KernRe(r'TRACE_EVENT\((.*?),') > - if r.search(proto): > - tracepointname = r.group(1) > - > - r = KernRe(r'DEFINE_SINGLE_EVENT\((.*?),') > - if r.search(proto): > - tracepointname = r.group(1) > - > - r = KernRe(r'DEFINE_EVENT\((.*?),(.*?),') > - if r.search(proto): > - tracepointname = r.group(2) > - > - if tracepointname: > - tracepointname = tracepointname.lstrip() > - > - r = KernRe(r'TP_PROTO\((.*?)\)') > - if r.search(proto): > - tracepointargs = r.group(1) > - > - if not tracepointname or not tracepointargs: > - self.emit_msg(ln, > - f"Unrecognized tracepoint format:\n{proto}\n") > - else: > - proto = f"static inline void trace_{tracepointname}({tracepointargs})" > - self.entry.identifier = f"trace_{self.entry.identifier}" > - > - return proto > - > def process_proto_function(self, ln, line, source): > """Ancillary routine to process a function prototype.""" > > @@ -1598,18 +1541,9 @@ class KernelDoc: > r = KernRe(r'^(\S+\s+)\(\s*\*(\S+)\)') > self.entry.prototype = r.sub(r'\1\2', self.entry.prototype) > # > - # Handle special declaration syntaxes > - # > - if 'SYSCALL_DEFINE' in self.entry.prototype: > - self.entry.prototype = self.syscall_munge(ln, > - self.entry.prototype) > - else: > - r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT') > - if r.search(self.entry.prototype): > - self.entry.prototype = self.tracepoint_munge(ln, > - self.entry.prototype) > - # > - # ... and we're done > + # ... and we're done. SYSCALL_DEFINE*() and TRACE_EVENT*()/ > + # DEFINE_EVENT*() are handled via xforms_lists.CTransforms > + # inside dump_function(). > # > self.dump_function(ln, self.entry.prototype, source) > self.reset_state(ln) > diff --git a/tools/lib/python/kdoc/xforms_lists.py b/tools/lib/python/kdoc/xforms_lists.py > index 15f6f21c7a9e..aa82dc927ef1 100644 > --- a/tools/lib/python/kdoc/xforms_lists.py > +++ b/tools/lib/python/kdoc/xforms_lists.py > @@ -17,6 +17,98 @@ class CTransforms: > into something we can parse and generate kdoc for. > """ > > + @staticmethod > + def syscall_munge(proto): > + """ > + Handle syscall definitions. > + > + Turns SYSCALL_DEFINE*() into a plain C prototype, e.g.:: > + > + SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval) > + -> long sys_setpriority(int which, int who, int niceval) > + """ > + > + is_void = False > + > + # Strip newlines/CR's > + proto = re.sub(r'[\r\n]+', ' ', proto) > + > + # Check if it's a SYSCALL_DEFINE0 > + if 'SYSCALL_DEFINE0' in proto: > + is_void = True > + > + # Replace SYSCALL_DEFINE with correct return type & function name > + proto = KernRe(r'SYSCALL_DEFINE.*\(').sub('long sys_', proto) > + > + r = KernRe(r'long\s+(sys_.*?),') > + if r.search(proto): > + proto = KernRe(',').sub('(', proto, count=1) > + elif is_void: > + proto = KernRe(r'\)').sub('(void)', proto, count=1) > + > + # Now delete all of the odd-numbered commas in the proto > + # so that argument types & names don't have a comma between them > + count = 0 > + length = len(proto) > + > + if is_void: > + length = 0 # skip the loop if is_void > + > + for ix in range(length): > + if proto[ix] == ',': > + count += 1 > + if count % 2 == 1: > + proto = proto[:ix] + ' ' + proto[ix + 1:] > + > + return proto > + > + @staticmethod > + def tracepoint_munge(proto): > + """ > + Handle tracepoint definitions. > + > + Turns TRACE_EVENT*()/DEFINE_EVENT*() into a plain C prototype, > + e.g.:: > + > + TRACE_EVENT_FN(gpu_power_status, TP_PROTO(...), ...) > + -> static inline void trace_gpu_power_status(...) > + > + Returns the original string if the tracepoint format is not > + recognized, letting the caller emit a warning. > + """ > + > + tracepointname = None > + tracepointargs = None > + > + # Match tracepoint name based on different patterns. > + # NOTE: DEFINE_EVENT*() takes a template name first, so the > + # event name is the second argument. > + for pattern, group in [ > + (r'TRACE_EVENT\((.*?),', 1), > + (r'TRACE_EVENT_FN\((.*?),', 1), > + (r'TRACE_EVENT_FN_COND\((.*?),', 1), > + (r'TRACE_EVENT_CONDITION\((.*?),', 1), > + (r'TRACE_EVENT_SYSCALL\((.*?),', 1), > + (r'DEFINE_SINGLE_EVENT\((.*?),', 1), > + (r'DEFINE_EVENT\((.*?),(.*?),', 2), > + ]: > + r = KernRe(pattern) > + if r.search(proto): > + tracepointname = r.group(group) > + > + if tracepointname: > + tracepointname = tracepointname.lstrip() > + > + r = KernRe(r'TP_PROTO\((.*?)\)') > + if r.search(proto): > + tracepointargs = r.group(1) > + > + if not tracepointname or not tracepointargs: > + return proto > + > + return f"static inline void trace_{tracepointname}({tracepointargs})" > + > + > # > # NOTE: > # Due to performance reasons, place CMatch rules before KernRe, > @@ -147,6 +239,23 @@ class CTransforms: > if xforms_type not in self.xforms: > return source > > + # > + # Handle special declaration syntaxes first, so that the > + # declarative transforms below see plain C prototypes. > + # > + if xforms_type == "func": > + source_str = str(source) > + if 'SYSCALL_DEFINE' in source_str: > + source_str = self.syscall_munge(source_str) > + source = source_str > + else: > + r = KernRe(r'TRACE_EVENT|DEFINE_EVENT|DEFINE_SINGLE_EVENT') > + if r.search(source_str): > + munged = self.tracepoint_munge(source_str) > + # tracepoint_munge() returns the input unchanged > + # when the format is not recognized. > + source = munged > + > if isinstance(source, str): > source = CTokenizer(source) > > diff --git a/tools/unittests/test_kdoc_parser.py b/tools/unittests/test_kdoc_parser.py > index 4baa78ca02b3..04dafae412e4 100755 > --- a/tools/unittests/test_kdoc_parser.py > +++ b/tools/unittests/test_kdoc_parser.py > @@ -444,6 +444,34 @@ class TestContextLockStruct(GenerateKdocItem): > self.assertEqual(entries[0].name, "debugfs_cancellation") > self.assertEqual(entries[0].warnings, []) > > +class TestTraceEventFn(GenerateKdocItem): > + def test_trace_event_fn_declaration(self): > + source = """ > + /** > + * gpu_power_status - called when hardware turns on or off > + * @dev: pointer to device > + * @shader_bitmap: bitmap > + */ > + TRACE_EVENT_FN(gpu_power_status, > + TP_PROTO(const struct device *dev, u64 shader_bitmap), > + TP_ARGS(dev, shader_bitmap), > + TP_STRUCT__entry(__field(u64, shader_bitmap)), > + TP_fast_assign(__entry->shader_bitmap = shader_bitmap;), > + TP_printk("%s", __get_str(dev_name)), > + my_register, my_unregister > + ); > + """ > + > + kernel_doc = KernelDoc(self.config, "test.c", self.xforms) > + patcher = patch('builtins.open', new_callable=mock_open, > + read_data=dedent(source)) > + with patcher: > + _, entries = kernel_doc.parse_kdoc() > + > + self.assertEqual(len(entries), 1) > + self.assertEqual(entries[0].name, "trace_gpu_power_status") > + self.assertEqual(entries[0].warnings, []) > + > # > # Class and logic to create dynamic tests from YAML > # Thanks, Mauro ^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH v3 2/2] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes 2026-10-15 16:44 ` [PATCH v3 0/2] docs: kdoc: fix TRACE_EVENT_FN and DOMAIN_NS parsing Omokefe Emmanuel Onanaroghene 2026-10-15 16:44 ` [PATCH v3 1/2] docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser Omokefe Emmanuel Onanaroghene @ 2026-10-15 16:44 ` Omokefe Emmanuel Onanaroghene 1 sibling, 0 replies; 12+ messages in thread From: Omokefe Emmanuel Onanaroghene @ 2026-10-15 16:44 UTC (permalink / raw) To: Jonathan Corbet Cc: linux-doc, linux-kernel, skhan, rdunlap, mchehab, paulmck, linusw, Julia.Lawall, gary, rmk+kernel, hemanth.selam, emmaonana18 kernel-doc fails to parse functions whose names are wrapped in the DOMAIN_NS() namespace macro, e.g. in drivers/iommu/generic_pt/iommu_pt.h:217: ... expecting prototype for iova_to_phys(). Prototype was for DOMAIN_NS() instead function_xforms is the designated place for unwrapping function-name macros, but it has no entry for DOMAIN_NS(), so the macro name is kept as the prototype and the real name leaks into the parameter list. Add a transform expanding DOMAIN_NS(op) to op, following the existing DEFINE_IDTENTRY_IRQ precedent. Add a regression test for the wrapped iova_to_phys() declaration. Signed-off-by: Omokefe Emmanuel Onanaroghene <emmaonana18@gmail.com> --- tools/lib/python/kdoc/xforms_lists.py | 6 ++++++ tools/unittests/test_kdoc_parser.py | 22 ++++++++++++++++++++++ 2 files changed, 28 insertions(+) diff --git a/tools/lib/python/kdoc/xforms_lists.py b/tools/lib/python/kdoc/xforms_lists.py index 15f6f21c7a9e..c9ce9bddebba 100644 --- a/tools/lib/python/kdoc/xforms_lists.py +++ b/tools/lib/python/kdoc/xforms_lists.py @@ -92,6 +92,12 @@ class CTransforms: (CMatch("__diagnose_as"), ""), (CMatch("DECL_BUCKET_PARAMS"), r"\1, \2"), (CMatch("DEFINE_IDTENTRY_IRQ"), r"static void \1(struct pt_regs *regs, u32 vector)"), + # + # DOMAIN_NS() wraps a function name with a namespace prefix + # (see drivers/iommu/generic_pt/iommu_pt.h). Unwrap it so the + # real function name is parsed. + # + (CMatch("DOMAIN_NS"), r"\1"), (CMatch("__cond_acquires"), ""), (CMatch("__cond_releases"), ""), (CMatch("__acquires"), ""), diff --git a/tools/unittests/test_kdoc_parser.py b/tools/unittests/test_kdoc_parser.py index 04dafae412e4..e4f44535d0dc 100755 --- a/tools/unittests/test_kdoc_parser.py +++ b/tools/unittests/test_kdoc_parser.py @@ -472,6 +472,28 @@ class TestTraceEventFn(GenerateKdocItem): self.assertEqual(entries[0].name, "trace_gpu_power_status") self.assertEqual(entries[0].warnings, []) +class TestDomainNs(GenerateKdocItem): + def test_domain_ns_wrapped_function(self): + source = """ + /** + * iova_to_phys() - Return the output address for the given IOVA + * @domain: Table to query + * @iova: IO virtual address to query + */ + phys_addr_t DOMAIN_NS(iova_to_phys)(struct iommu_domain *domain, + dma_addr_t iova); + """ + + kernel_doc = KernelDoc(self.config, "test.c", self.xforms) + patcher = patch('builtins.open', new_callable=mock_open, + read_data=dedent(source)) + with patcher: + _, entries = kernel_doc.parse_kdoc() + + self.assertEqual(len(entries), 1) + self.assertEqual(entries[0].name, "iova_to_phys") + self.assertEqual(entries[0].warnings, []) + # # Class and logic to create dynamic tests from YAML # -- 2.43.0 ^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2026-10-16 21:11 UTC | newest] Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2026-10-02 4:15 [PATCH] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes Omokefe Emmanuel Onanaroghene 2026-10-02 4:28 ` Randy Dunlap 2026-10-14 17:22 ` Jonathan Corbet 2026-10-14 22:47 ` Omokefe Emmanuel Onanaroghene 2026-10-15 0:53 ` Omokefe Emmanuel Onanaroghene 2026-10-15 13:00 ` Jonathan Corbet 2026-10-15 16:44 ` [PATCH v3 0/2] docs: kdoc: fix TRACE_EVENT_FN and DOMAIN_NS parsing Omokefe Emmanuel Onanaroghene 2026-10-15 16:44 ` [PATCH v3 1/2] docs: kdoc: handle TRACE_EVENT_FN() family in tracepoint parser Omokefe Emmanuel Onanaroghene 2026-10-15 17:06 ` Mauro Carvalho Chehab 2026-10-16 20:43 ` [PATCH] " Omokefe Emmanuel Onanaroghene 2026-10-16 21:11 ` Mauro Carvalho Chehab 2026-10-15 16:44 ` [PATCH v3 2/2] docs: kdoc: strip DOMAIN_NS() wrapper in function prototypes Omokefe Emmanuel Onanaroghene
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®