[PATCH RFC/RFT 00/12] clk: samsung: Use platform_driver_probe() to avoid __refdata

[PATCH RFC/RFT 00/12] clk: samsung: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

RFT/RFC because testing needed. I tried to do the same on exynos5-subcmu
and it caused weird oopses which even KASAN did not narrow. Probably
because of multiple exynos5-subcmu devices?

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Best regards,
Krzysztof

---
Krzysztof Kozlowski (12):
      clk: samsung: exynos2200: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynos4412-isp: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynos5433: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynos7870: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynos7885: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynos850: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynos8895: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynos990: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynosautov9: Use platform_driver_probe() to avoid __refdata
      clk: samsung: exynosautov920: Use platform_driver_probe() to avoid __refdata
      clk: samsung: fsd: Use platform_driver_probe() to avoid __refdata
      clk: samsung: gs101: Use platform_driver_probe() to avoid __refdata

 drivers/clk/samsung/clk-exynos2200.c     | 7 +++----
 drivers/clk/samsung/clk-exynos4412-isp.c | 7 +++----
 drivers/clk/samsung/clk-exynos5433.c     | 7 +++----
 drivers/clk/samsung/clk-exynos7870.c     | 7 +++----
 drivers/clk/samsung/clk-exynos7885.c     | 7 +++----
 drivers/clk/samsung/clk-exynos850.c      | 7 +++----
 drivers/clk/samsung/clk-exynos8895.c     | 7 +++----
 drivers/clk/samsung/clk-exynos990.c      | 7 +++----
 drivers/clk/samsung/clk-exynosautov9.c   | 5 ++---
 drivers/clk/samsung/clk-exynosautov920.c | 7 +++----
 drivers/clk/samsung/clk-fsd.c            | 7 +++----
 drivers/clk/samsung/clk-gs101.c          | 7 +++----
 12 files changed, 35 insertions(+), 47 deletions(-)
---
base-commit: 7ec162622e66a4ff886f8f28712ea1b13069e1aa
change-id: 20250305-clk-samsung-ref-init-data-6b2cd242409f

Best regards,
-- 
Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>

[PATCH RFC/RFT 01/12] clk: samsung: exynos2200: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>

---

Testing needed.

My initial tests on Exynos5422 lead to weird, unrelated oopses, probably
because of multiple exynos5-subcmu devices. Not sure if applicable here,
but change of initcal might have actual impact.
---
 drivers/clk/samsung/clk-exynos2200.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynos2200.c b/drivers/clk/samsung/clk-exynos2200.c
index 151bdb35a46c42d0b50f48d47974065a892bbb5f..f1251d4f01781fd503f24189cc53b23e627ab5e1 100644
--- a/drivers/clk/samsung/clk-exynos2200.c
+++ b/drivers/clk/samsung/clk-exynos2200.c
@@ -3912,17 +3912,16 @@ static const struct of_device_id exynos2200_cmu_of_match[] = {
 	}, { }
 };
 
-static struct platform_driver exynos2200_cmu_driver __refdata = {
+static struct platform_driver exynos2200_cmu_driver __initdata = {
 	.driver = {
 		.name = "exynos2200-cmu",
 		.of_match_table = exynos2200_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = exynos2200_cmu_probe,
 };
 
 static int __init exynos2200_cmu_init(void)
 {
-	return platform_driver_register(&exynos2200_cmu_driver);
+	return platform_driver_probe(&exynos2200_cmu_driver, exynos2200_cmu_probe);
 }
-core_initcall(exynos2200_cmu_init);
+subsys_initcall(exynos2200_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 02/12] clk: samsung: exynos4412-isp: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynos4412-isp.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynos4412-isp.c b/drivers/clk/samsung/clk-exynos4412-isp.c
index 29c5644f059379f8a0770ce2214c32b9df7d76e3..f288c152d3f769fbb5123c3b79fe4ddc8c23aa55 100644
--- a/drivers/clk/samsung/clk-exynos4412-isp.c
+++ b/drivers/clk/samsung/clk-exynos4412-isp.c
@@ -155,18 +155,17 @@ static const struct dev_pm_ops exynos4x12_isp_pm_ops = {
 				     pm_runtime_force_resume)
 };
 
-static struct platform_driver exynos4x12_isp_clk_driver __refdata = {
+static struct platform_driver exynos4x12_isp_clk_driver __initdata = {
 	.driver	= {
 		.name = "exynos4x12-isp-clk",
 		.of_match_table = exynos4x12_isp_clk_of_match,
 		.suppress_bind_attrs = true,
 		.pm = &exynos4x12_isp_pm_ops,
 	},
-	.probe = exynos4x12_isp_clk_probe,
 };
 
 static int __init exynos4x12_isp_clk_init(void)
 {
-	return platform_driver_register(&exynos4x12_isp_clk_driver);
+	return platform_driver_probe(&exynos4x12_isp_clk_driver, exynos4x12_isp_clk_probe);
 }
-core_initcall(exynos4x12_isp_clk_init);
+subsys_initcall(exynos4x12_isp_clk_init);

-- 
2.43.0

[PATCH RFC/RFT 03/12] clk: samsung: exynos5433: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynos5433.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynos5433.c b/drivers/clk/samsung/clk-exynos5433.c
index 609d31a7aa5247e6318884609d1e651001616b73..d18b4f7b96b66e48fb94b86b663443ec691e968d 100644
--- a/drivers/clk/samsung/clk-exynos5433.c
+++ b/drivers/clk/samsung/clk-exynos5433.c
@@ -5555,18 +5555,17 @@ static const struct dev_pm_ops exynos5433_cmu_pm_ops = {
 				     pm_runtime_force_resume)
 };
 
-static struct platform_driver exynos5433_cmu_driver __refdata = {
+static struct platform_driver exynos5433_cmu_driver __initdata = {
 	.driver	= {
 		.name = "exynos5433-cmu",
 		.of_match_table = exynos5433_cmu_of_match,
 		.suppress_bind_attrs = true,
 		.pm = &exynos5433_cmu_pm_ops,
 	},
-	.probe = exynos5433_cmu_probe,
 };
 
 static int __init exynos5433_cmu_init(void)
 {
-	return platform_driver_register(&exynos5433_cmu_driver);
+	return platform_driver_probe(&exynos5433_cmu_driver, exynos5433_cmu_probe);
 }
-core_initcall(exynos5433_cmu_init);
+subsys_initcall(exynos5433_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 04/12] clk: samsung: exynos7870: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynos7870.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynos7870.c b/drivers/clk/samsung/clk-exynos7870.c
index 2ec4a4e489be30bd1cd2e6deac006bb8ac5bdc57..c3d7d816cab673518c7b01ce35b0faf249692835 100644
--- a/drivers/clk/samsung/clk-exynos7870.c
+++ b/drivers/clk/samsung/clk-exynos7870.c
@@ -1814,17 +1814,16 @@ static const struct of_device_id exynos7870_cmu_of_match[] = {
 	},
 };
 
-static struct platform_driver exynos7870_cmu_driver __refdata = {
+static struct platform_driver exynos7870_cmu_driver __initdata = {
 	.driver = {
 		.name = "exynos7870-cmu",
 		.of_match_table = exynos7870_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = exynos7870_cmu_probe,
 };
 
 static int __init exynos7870_cmu_init(void)
 {
-	return platform_driver_register(&exynos7870_cmu_driver);
+	return platform_driver_probe(&exynos7870_cmu_driver, exynos7870_cmu_probe);
 }
-core_initcall(exynos7870_cmu_init);
+subsys_initcall(exynos7870_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 05/12] clk: samsung: exynos7885: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynos7885.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynos7885.c b/drivers/clk/samsung/clk-exynos7885.c
index fc42251731edb6f8368d38638179056f8bc58335..0066c89e3515b852c7c6265322f8a1dc07af57ad 100644
--- a/drivers/clk/samsung/clk-exynos7885.c
+++ b/drivers/clk/samsung/clk-exynos7885.c
@@ -832,17 +832,16 @@ static const struct of_device_id exynos7885_cmu_of_match[] = {
 	},
 };
 
-static struct platform_driver exynos7885_cmu_driver __refdata = {
+static struct platform_driver exynos7885_cmu_driver __initdata = {
 	.driver	= {
 		.name = "exynos7885-cmu",
 		.of_match_table = exynos7885_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = exynos7885_cmu_probe,
 };
 
 static int __init exynos7885_cmu_init(void)
 {
-	return platform_driver_register(&exynos7885_cmu_driver);
+	return platform_driver_probe(&exynos7885_cmu_driver, exynos7885_cmu_probe);
 }
-core_initcall(exynos7885_cmu_init);
+subsys_initcall(exynos7885_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 06/12] clk: samsung: exynos850: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynos850.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynos850.c b/drivers/clk/samsung/clk-exynos850.c
index e00e213b1201c82ac0c70cfe23d5ccc9c04f2aa3..37c06cbbe78f87e18877d33425fcd03b97a61e7f 100644
--- a/drivers/clk/samsung/clk-exynos850.c
+++ b/drivers/clk/samsung/clk-exynos850.c
@@ -2322,17 +2322,16 @@ static const struct of_device_id exynos850_cmu_of_match[] = {
 	},
 };
 
-static struct platform_driver exynos850_cmu_driver __refdata = {
+static struct platform_driver exynos850_cmu_driver __initdata = {
 	.driver	= {
 		.name = "exynos850-cmu",
 		.of_match_table = exynos850_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = exynos850_cmu_probe,
 };
 
 static int __init exynos850_cmu_init(void)
 {
-	return platform_driver_register(&exynos850_cmu_driver);
+	return platform_driver_probe(&exynos850_cmu_driver, exynos850_cmu_probe);
 }
-core_initcall(exynos850_cmu_init);
+subsys_initcall(exynos850_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 07/12] clk: samsung: exynos8895: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynos8895.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynos8895.c b/drivers/clk/samsung/clk-exynos8895.c
index 29ec0c4a863524fe3fc5c223d762da87e5f13013..6b85d917871d20fd3a3f6768937eb5694760ffd2 100644
--- a/drivers/clk/samsung/clk-exynos8895.c
+++ b/drivers/clk/samsung/clk-exynos8895.c
@@ -2787,17 +2787,16 @@ static const struct of_device_id exynos8895_cmu_of_match[] = {
 	{ }
 };
 
-static struct platform_driver exynos8895_cmu_driver __refdata = {
+static struct platform_driver exynos8895_cmu_driver __initdata = {
 	.driver = {
 		.name = "exynos8895-cmu",
 		.of_match_table = exynos8895_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = exynos8895_cmu_probe,
 };
 
 static int __init exynos8895_cmu_init(void)
 {
-	return platform_driver_register(&exynos8895_cmu_driver);
+	return platform_driver_probe(&exynos8895_cmu_driver, exynos8895_cmu_probe);
 }
-core_initcall(exynos8895_cmu_init);
+subsys_initcall(exynos8895_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 08/12] clk: samsung: exynos990: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynos990.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynos990.c b/drivers/clk/samsung/clk-exynos990.c
index 76f22a4a4631fadf400420b977f3f59f40f840c7..d00050d8765f25042698cbe6bfa06623931c7414 100644
--- a/drivers/clk/samsung/clk-exynos990.c
+++ b/drivers/clk/samsung/clk-exynos990.c
@@ -1504,18 +1504,17 @@ static const struct of_device_id exynos990_cmu_of_match[] = {
 	{ },
 };
 
-static struct platform_driver exynos990_cmu_driver __refdata = {
+static struct platform_driver exynos990_cmu_driver __initdata = {
 	.driver	= {
 		.name = "exynos990-cmu",
 		.of_match_table = exynos990_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = exynos990_cmu_probe,
 };
 
 static int __init exynos990_cmu_init(void)
 {
-	return platform_driver_register(&exynos990_cmu_driver);
+	return platform_driver_probe(&exynos990_cmu_driver, exynos990_cmu_probe);
 }
 
-core_initcall(exynos990_cmu_init);
+subsys_initcall(exynos990_cmu_init);

-- 
2.43.0

Re: [PATCH RFC/RFT 08/12] clk: samsung: exynos990: Use platform_driver_probe() to avoid __refdata

[Igor Belwon]

On Wed Mar 5, 2025 at 10:43 PM CET, Krzysztof Kozlowski wrote:
> Switch registering platform driver to platform_driver_probe(), so the
> 'struct platform_driver' can be properly discarded after init and there
> won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.
>
> The change requires using subsys_initcall instead of core_initcall,
> because no device drivers would bound in the latter, as required by
> platform_driver_probe().
>
> Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>

Hi Krzysztof and apologies for the delay in testing.

On my Exynos990 platforms this patch causes similar oopses to those
sent by André (different places each boot). I am unable to get the logs from
UART right now though (refactoring my testing setup). Apologies.

So this bug seems to be platformwide :(

Best regards,
Igor

[PATCH RFC/RFT 09/12] clk: samsung: exynosautov9: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynosautov9.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynosautov9.c b/drivers/clk/samsung/clk-exynosautov9.c
index 5971e680e566336ecdf55bdccacc3cea09a33201..627f3d2666740fc4cba205e6e2da65f53b8df186 100644
--- a/drivers/clk/samsung/clk-exynosautov9.c
+++ b/drivers/clk/samsung/clk-exynosautov9.c
@@ -2190,17 +2190,16 @@ static const struct of_device_id exynosautov9_cmu_of_match[] = {
 	},
 };
 
-static struct platform_driver exynosautov9_cmu_driver __refdata = {
+static struct platform_driver exynosautov9_cmu_driver __initdata = {
 	.driver = {
 		.name = "exynosautov9-cmu",
 		.of_match_table = exynosautov9_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = exynosautov9_cmu_probe,
 };
 
 static int __init exynosautov9_cmu_init(void)
 {
-	return platform_driver_register(&exynosautov9_cmu_driver);
+	return platform_driver_probe(&exynosautov9_cmu_driver, exynosautov9_cmu_probe);
 }
 core_initcall(exynosautov9_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 10/12] clk: samsung: exynosautov920: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-exynosautov920.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-exynosautov920.c b/drivers/clk/samsung/clk-exynosautov920.c
index 2a8bfd5d9abc8acced4e5a3eaacefe5fa724cbd2..e0ba75fd20177022dc7f461af5968c154b6a8a37 100644
--- a/drivers/clk/samsung/clk-exynosautov920.c
+++ b/drivers/clk/samsung/clk-exynosautov920.c
@@ -1447,17 +1447,16 @@ static const struct of_device_id exynosautov920_cmu_of_match[] = {
 	{ }
 };
 
-static struct platform_driver exynosautov920_cmu_driver __refdata = {
+static struct platform_driver exynosautov920_cmu_driver __initdata = {
 	.driver = {
 		.name = "exynosautov920-cmu",
 		.of_match_table = exynosautov920_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = exynosautov920_cmu_probe,
 };
 
 static int __init exynosautov920_cmu_init(void)
 {
-	return platform_driver_register(&exynosautov920_cmu_driver);
+	return platform_driver_probe(&exynosautov920_cmu_driver, exynosautov920_cmu_probe);
 }
-core_initcall(exynosautov920_cmu_init);
+subsys_initcall(exynosautov920_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 11/12] clk: samsung: fsd: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-fsd.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-fsd.c b/drivers/clk/samsung/clk-fsd.c
index 9a6006c298c2a80da0e29aeca3b09cc827caac0c..bc0287c780bfe3a68a38f53f554440b3000b1ca7 100644
--- a/drivers/clk/samsung/clk-fsd.c
+++ b/drivers/clk/samsung/clk-fsd.c
@@ -1794,17 +1794,16 @@ static const struct of_device_id fsd_cmu_of_match[] = {
 	},
 };
 
-static struct platform_driver fsd_cmu_driver __refdata = {
+static struct platform_driver fsd_cmu_driver __initdata = {
 	.driver	= {
 		.name = "fsd-cmu",
 		.of_match_table = fsd_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = fsd_cmu_probe,
 };
 
 static int __init fsd_cmu_init(void)
 {
-	return platform_driver_register(&fsd_cmu_driver);
+	return platform_driver_probe(&fsd_cmu_driver, fsd_cmu_probe);
 }
-core_initcall(fsd_cmu_init);
+subsys_initcall(fsd_cmu_init);

-- 
2.43.0

[PATCH RFC/RFT 12/12] clk: samsung: gs101: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

Switch registering platform driver to platform_driver_probe(), so the
'struct platform_driver' can be properly discarded after init and there
won't be need of __refdata to silence DEBUG_SECTION_MISMATCH.

The change requires using subsys_initcall instead of core_initcall,
because no device drivers would bound in the latter, as required by
platform_driver_probe().

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
---
 drivers/clk/samsung/clk-gs101.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/clk/samsung/clk-gs101.c b/drivers/clk/samsung/clk-gs101.c
index 86b39edba12276745a124df6a2ea9c22d74b915a..69c121d478f3fec98dcd84e0639e204fe604e91c 100644
--- a/drivers/clk/samsung/clk-gs101.c
+++ b/drivers/clk/samsung/clk-gs101.c
@@ -4412,17 +4412,16 @@ static const struct of_device_id gs101_cmu_of_match[] = {
 	},
 };
 
-static struct platform_driver gs101_cmu_driver __refdata = {
+static struct platform_driver gs101_cmu_driver __initdata = {
 	.driver	= {
 		.name = "gs101-cmu",
 		.of_match_table = gs101_cmu_of_match,
 		.suppress_bind_attrs = true,
 	},
-	.probe = gs101_cmu_probe,
 };
 
 static int __init gs101_cmu_init(void)
 {
-	return platform_driver_register(&gs101_cmu_driver);
+	return platform_driver_probe(&gs101_cmu_driver, gs101_cmu_probe);
 }
-core_initcall(gs101_cmu_init);
+subsys_initcall(gs101_cmu_init);

-- 
2.43.0

Re: [PATCH RFC/RFT 00/12] clk: samsung: Use platform_driver_probe() to avoid __refdata

[André Draszik]

Hi Krzysztof,

Nice idea!

On Wed, 2025-03-05 at 22:43 +0100, Krzysztof Kozlowski wrote:
> RFT/RFC because testing needed. I tried to do the same on exynos5-subcmu
> and it caused weird oopses which even KASAN did not narrow. Probably
> because of multiple exynos5-subcmu devices?

I've tried this on top of next-20250225, and it doesn't work on gs101
either and OOPSes several times during boot in different places, but
I can not dig deeper right now.

[   11.502919][   T58] Unable to handle kernel paging request at virtual address ffffbfe2ab25cc30
[   11.503128][   T58] Mem abort info:
[   11.503202][   T58]   ESR = 0x0000000096000007
[   11.503295][   T58]   EC = 0x25: DABT (current EL), IL = 32 bits
[   11.503419][   T58]   SET = 0, FnV = 0
[   11.503498][   T58]   EA = 0, S1PTW = 0
[   11.503579][   T58]   FSC = 0x07: level 3 translation fault
[   11.503695][   T58] Data abort info:
[   11.503769][   T58]   ISV = 0, ISS = 0x00000007, ISS2 = 0x00000000
[   11.503896][   T58]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[   11.504058][   T58]   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[   11.504288][   T58] swapper pgtable: 4k pages, 48-bit VAs, pgdp=000000008208c000
[   11.504570][   T58] [ffffbfe2ab25cc30] pgd=0000000000000000, p4d=1000000082a58003, pud=1000000082a59003, pmd=1000000082a5d003,
pte=0000000000000000
[   11.505079][   T58] Internal error: Oops: 0000000096000007 [#1] PREEMPT SMP
[   11.505346][   T58] Modules linked in:
[   11.505495][   T58] CPU: 6 UID: 0 PID: 58 Comm: kworker/u32:1 Tainted: G                T  6.14.0-rc4-next-20250225+ #12
[   11.505902][   T58] Tainted: [T]=RANDSTRUCT
[   11.506061][   T58] Hardware name: Raven (DT)
[   11.506230][   T58] Workqueue: events_unbound deferred_probe_work_func
[   11.506477][   T58] pstate: 20400005 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[   11.506768][   T58] pc : __device_attach_driver+0x18/0x254
[   11.506978][   T58] lr : bus_for_each_drv+0xfc/0x154
[   11.507168][   T58] sp : ffff8000803abc00
[   11.507321][   T58] x29: ffff8000803abc10 x28: ffff00ed403363d0 x27: ffff00ed40336400
[   11.507620][   T58] x26: 0000000000000000 x25: ffff00ed4002380d x24: ffff00ed40028000
[   11.507919][   T58] x23: ffffbfe2aba26000 x22: 0000000000000000 x21: ffff00ed409f8400
[   11.508219][   T58] x20: ffff8000803abca8 x19: ffffbfe2a9a4cf4c x18: 0000000000000040
[   11.508518][   T58] x17: 0000000000000001 x16: 0000000000000001 x15: 0000000000000010
[   11.508817][   T58] x14: 0000000000000010 x13: 0000000000000093 x12: 0000000000000002
[   11.509117][   T58] x11: ffffffffff6070d2 x10: 0000bef56a420260 x9 : 0000000100000000
[   11.509416][   T58] x8 : 0000000100000000 x7 : ff7f7f7f7f7f7f7f x6 : 455242436f53531d
[   11.509715][   T58] x5 : 1d53536f43425245 x4 : 0000000000000000 x3 : e8c6e6e4bec6c65a
[   11.510015][   T58] x2 : 0000000000000062 x1 : ffff8000803abca8 x0 : ffffbfe2ab25cc28
[   11.510315][   T58] Call trace:
[   11.510436][   T58]  __device_attach_driver+0x18/0x254 (P)
[   11.510644][   T58]  bus_for_each_drv+0xfc/0x154
[   11.510819][   T58]  __device_attach+0x100/0x1cc
[   11.510996][   T58]  device_initial_probe+0x14/0x20
[   11.511182][   T58]  bus_probe_device+0x94/0x100
[   11.511359][   T58]  deferred_probe_work_func+0xa0/0xfc
[   11.511558][   T58]  process_scheduled_works+0x194/0x2c4
[   11.511761][   T58]  worker_thread+0x28c/0x394
[   11.511930][   T58]  kthread+0x1c0/0x204
[   11.512081][   T58]  ret_from_fork+0x10/0x20
[   11.512248][   T58] Code: a9017bfd f90013f5 a9034ff4 910043fd (f9400408) 
[   11.512506][   T58] ---[ end trace 0000000000000000 ]---

and

+ udevadm trigger '--action=add'
[  390.387986][  T394] Unable to handle kernel paging request at virtual address ffffbfe2ab25cc28
[  390.388207][  T394] Mem abort info:
[  390.388282][  T394]   ESR = 0x0000000096000007
[  390.388378][  T394]   EC = 0x25: DABT (current EL), IL = 32 bits
[  390.388506][  T394]   SET = 0, FnV = 0
[  390.388586][  T394]   EA = 0, S1PTW = 0
[  390.388669][  T394]   FSC = 0x07: level 3 translation fault
[  390.388787][  T394] Data abort info:
[  390.388862][  T394]   ISV = 0, ISS = 0x00000007, ISS2 = 0x00000000
[  390.388990][  T394]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[  390.389137][  T394]   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[  390.389367][  T394] swapper pgtable: 4k pages, 48-bit VAs, pgdp=000000008208c000
[  390.389650][  T394] [ffffbfe2ab25cc28] pgd=0000000000000000, p4d=1000000082a58003, pud=1000000082a59003, pmd=1000000082a5d003,
pte=0000000000000000
[  390.390173][  T394] Internal error: Oops: 0000000096000007 [#2] PREEMPT SMP
[  390.390426][  T394] Modules linked in:
[  390.390580][  T394] CPU: 7 UID: 0 PID: 394 Comm: udevadm Tainted: G      D         T  6.14.0-rc4-next-20250225+ #12
[  390.390969][  T394] Tainted: [D]=DIE, [T]=RANDSTRUCT
[  390.391153][  T394] Hardware name: Raven (DT)
[  390.391322][  T394] pstate: 80400005 (Nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[  390.391612][  T394] pc : dev_uevent+0x144/0x2d8
[  390.391785][  T394] lr : kobject_uevent_env+0x1e0/0x3a8
[  390.391985][  T394] sp : ffff8000810fbb40
[  390.392137][  T394] x29: ffff8000810fbb60 x28: 0000000000000002 x27: ffffbfe2aa859240
[  390.392437][  T394] x26: ffffbfe2aae1ca12 x25: 0000000000000000 x24: ffff00ed41167540
[  390.392736][  T394] x23: ffff00ed41070000 x22: ffff8000810fbbf8 x21: 0000000000000000
[  390.393035][  T394] x20: ffff00ed41070000 x19: ffff00ed41083010 x18: 0000000000000000
[  390.393335][  T394] x17: 0000000000000000 x16: 0000000000000000 x15: 0000ffffc7e40360
[  390.393634][  T394] x14: 00000000ffffffd0 x13: aaaaaaaaaaaaaaaa x12: aaaaaaaaaaaa0000
[  390.393933][  T394] x11: 0000000000000000 x10: 0000000000000004 x9 : 3bc04d1231e6a100
[  390.394233][  T394] x8 : ffffbfe2ab25cc28 x7 : 3d4d455453595342 x6 : 4554535953425553
[  390.394532][  T394] x5 : ffff00ed4107026b x4 : ffffbfe2aae64508 x3 : ffffffffffff0a00
[  390.394831][  T394] x2 : 0000000000000000 x1 : ffff00ed41070000 x0 : ffff00ed41083010
[  390.395132][  T394] Call trace:
[  390.395252][  T394]  dev_uevent+0x144/0x2d8 (P)
[  390.395425][  T394]  kobject_uevent_env+0x1e0/0x3a8
[  390.395610][  T394]  kobject_synth_uevent+0x3ec/0x440
[  390.395802][  T394]  uevent_store+0x28/0x64
[  390.395962][  T394]  dev_attr_store+0x18/0x2c
[  390.396128][  T394]  sysfs_kf_write+0x80/0xa0
[  390.396295][  T394]  kernfs_fop_write_iter+0xdc/0x178
[  390.396488][  T394]  vfs_write+0x344/0x3a0
[  390.396644][  T394]  ksys_write+0x7c/0xe8
[  390.396797][  T394]  __arm64_sys_write+0x1c/0x28
[  390.396973][  T394]  invoke_syscall+0x40/0xf8
[  390.397139][  T394]  el0_svc_common+0xac/0xdc
[  390.397306][  T394]  do_el0_svc+0x1c/0x28
[  390.397459][  T394]  el0_svc+0x34/0x9c
[  390.397602][  T394]  el0t_64_sync_handler+0x84/0x108
[  390.397791][  T394]  el0t_64_sync+0x1a4/0x1a8
[  390.397963][  T394] Code: aa1403e0 94232d4d f9403668 b40000c8 (f9400102) 
[  390.398221][  T394] ---[ end trace 0000000000000000 ]---


Cheers,
Andre'

Re: [PATCH RFC/RFT 00/12] clk: samsung: Use platform_driver_probe() to avoid __refdata

[Krzysztof Kozlowski]

On 07/03/2025 15:29, André Draszik wrote:
> Hi Krzysztof,
> 
> Nice idea!
> 
> On Wed, 2025-03-05 at 22:43 +0100, Krzysztof Kozlowski wrote:
>> RFT/RFC because testing needed. I tried to do the same on exynos5-subcmu
>> and it caused weird oopses which even KASAN did not narrow. Probably
>> because of multiple exynos5-subcmu devices?
> 
> I've tried this on top of next-20250225, and it doesn't work on gs101
> either and OOPSes several times during boot in different places, but
> I can not dig deeper right now.
> 
> [   11.502919][   T58] Unable to handle kernel paging request at virtual address ffffbfe2ab25cc30

Thanks. I'll dig more once have a bit more time. The calltrace is
exactly what I saw with exynos5-subcmu and it puzzles me.

Best regards,
Krzysztof

Re: [PATCH RFC/RFT 00/12] clk: samsung: Use platform_driver_probe() to avoid __refdata

[Stephen Boyd]

Quoting André Draszik (2025-03-07 06:29:18)
> 
> I've tried this on top of next-20250225, and it doesn't work on gs101
> either and OOPSes several times during boot in different places, but
> I can not dig deeper right now.
> 
> [   11.502919][   T58] Unable to handle kernel paging request at virtual address ffffbfe2ab25cc30
> [   11.503128][   T58] Mem abort info:
> [   11.503202][   T58]   ESR = 0x0000000096000007
> [   11.503295][   T58]   EC = 0x25: DABT (current EL), IL = 32 bits
> [   11.503419][   T58]   SET = 0, FnV = 0
> [   11.503498][   T58]   EA = 0, S1PTW = 0
> [   11.503579][   T58]   FSC = 0x07: level 3 translation fault
> [   11.503695][   T58] Data abort info:
> [   11.503769][   T58]   ISV = 0, ISS = 0x00000007, ISS2 = 0x00000000
> [   11.503896][   T58]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
> [   11.504058][   T58]   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
> [   11.504288][   T58] swapper pgtable: 4k pages, 48-bit VAs, pgdp=000000008208c000
> [   11.504570][   T58] [ffffbfe2ab25cc30] pgd=0000000000000000, p4d=1000000082a58003, pud=1000000082a59003, pmd=1000000082a5d003,
> pte=0000000000000000
> [   11.505079][   T58] Internal error: Oops: 0000000096000007 [#1] PREEMPT SMP
> [   11.505346][   T58] Modules linked in:
> [   11.505495][   T58] CPU: 6 UID: 0 PID: 58 Comm: kworker/u32:1 Tainted: G                T  6.14.0-rc4-next-20250225+ #12
> [   11.505902][   T58] Tainted: [T]=RANDSTRUCT
> [   11.506061][   T58] Hardware name: Raven (DT)
> [   11.506230][   T58] Workqueue: events_unbound deferred_probe_work_func
> [   11.506477][   T58] pstate: 20400005 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> [   11.506768][   T58] pc : __device_attach_driver+0x18/0x254
> [   11.506978][   T58] lr : bus_for_each_drv+0xfc/0x154
> [   11.507168][   T58] sp : ffff8000803abc00
> [   11.507321][   T58] x29: ffff8000803abc10 x28: ffff00ed403363d0 x27: ffff00ed40336400
> [   11.507620][   T58] x26: 0000000000000000 x25: ffff00ed4002380d x24: ffff00ed40028000
> [   11.507919][   T58] x23: ffffbfe2aba26000 x22: 0000000000000000 x21: ffff00ed409f8400
> [   11.508219][   T58] x20: ffff8000803abca8 x19: ffffbfe2a9a4cf4c x18: 0000000000000040
> [   11.508518][   T58] x17: 0000000000000001 x16: 0000000000000001 x15: 0000000000000010
> [   11.508817][   T58] x14: 0000000000000010 x13: 0000000000000093 x12: 0000000000000002
> [   11.509117][   T58] x11: ffffffffff6070d2 x10: 0000bef56a420260 x9 : 0000000100000000
> [   11.509416][   T58] x8 : 0000000100000000 x7 : ff7f7f7f7f7f7f7f x6 : 455242436f53531d
> [   11.509715][   T58] x5 : 1d53536f43425245 x4 : 0000000000000000 x3 : e8c6e6e4bec6c65a
> [   11.510015][   T58] x2 : 0000000000000062 x1 : ffff8000803abca8 x0 : ffffbfe2ab25cc28
> [   11.510315][   T58] Call trace:
> [   11.510436][   T58]  __device_attach_driver+0x18/0x254 (P)
> [   11.510644][   T58]  bus_for_each_drv+0xfc/0x154
> [   11.510819][   T58]  __device_attach+0x100/0x1cc
> [   11.510996][   T58]  device_initial_probe+0x14/0x20
> [   11.511182][   T58]  bus_probe_device+0x94/0x100
> [   11.511359][   T58]  deferred_probe_work_func+0xa0/0xfc

platform_driver_probe() is incompatible with deferred probe. It
should have set drv->prevent_deferred_probe to true so the driver must
have gotten onto the deferred probe list somehow when it shouldn't have.