* [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-16 10:49 ` Junyi Zhao via B4 Relay
0 siblings, 0 replies; 30+ messages in thread
From: Junyi Zhao via B4 Relay @ 2026-11-16 10:49 UTC (permalink / raw)
To: Alexandre Belloni, Frank Li, Rob Herring, Krzysztof Kozlowski,
Conor Dooley, Neil Armstrong, Kevin Hilman, Jerome Brunet,
Martin Blumenstingl, Miquel Raynal, Kees Cook,
Gustavo A. R. Silva
Cc: linux-i3c, devicetree, linux-kernel, linux-arm-kernel,
linux-amlogic, imx, linux-hardening, xianwei.zhao, Kelvin.Zhang,
Junyi Zhao
From: Junyi Zhao <junyi.zhao@amlogic.com>
Amlogic a9 SoCs integrate the Silvaco I3C IP, and their peripheral
clocks vary from the vanilla implementation:
- The EE-domain controller (amlogic,a9-i3c) takes two clocks,
"pclk" and "fast_clk", and the fast clock rate must be programmed
before the controller can be used.
- The AO-domain controller (amlogic,a9-ao-i3c) exposes only "pclk":
its bus clock is generated by an internal divider located in the
controller's own register space.
Both variants need setup that the generic driver does not perform:
the fast clock rate must be rounded and set at probe time, and on the
AO variant the internal divider must be registered with the clock
framework so that "fast_clk" even exists. The divider registers sit
in the controller's own register space, so they can only be accessed
while pclk is running.
Add a vendor_init callback, invoked from the probe with the clocks
enabled, to handle this. On the AO variant it registers the internal
divider chain (div_en -> div -> gate) with the common clock framework
and appends the resulting clock to the clk bulk array as "fast_clk";
for both variants it then rounds and sets the fast clock rate
(target 25 MHz). As vendor_init now runs after
clk_bulk_prepare_enable(), the probe error paths that follow disable
the clocks on failure.
Extend the quirks bitmask with an Amlogic-specific bit
(AML_I3C_QUIRK_AO) marking the AO variant's internal-divider
behaviour. Since the driver now registers clocks through the common
clock framework, SVC_I3C_MASTER also depends on COMMON_CLK.
Signed-off-by: Junyi Zhao <junyi.zhao@amlogic.com>
---
drivers/i3c/master/Kconfig | 1 +
drivers/i3c/master/svc-i3c-master.c | 276 ++++++++++++++++++++++++++++++++++--
2 files changed, 265 insertions(+), 12 deletions(-)
diff --git a/drivers/i3c/master/Kconfig b/drivers/i3c/master/Kconfig
index da96d2aaa..ac7d246c2 100644
--- a/drivers/i3c/master/Kconfig
+++ b/drivers/i3c/master/Kconfig
@@ -47,6 +47,7 @@ config AST2600_I3C_MASTER
config SVC_I3C_MASTER
tristate "Silvaco I3C Dual-Role Master driver"
+ depends on COMMON_CLK
depends on HAS_IOMEM
depends on !(ALPHA || PARISC)
help
diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
index e700392f0..0268fc713 100644
--- a/drivers/i3c/master/svc-i3c-master.c
+++ b/drivers/i3c/master/svc-i3c-master.c
@@ -9,6 +9,7 @@
#include <linux/bitfield.h>
#include <linux/clk.h>
+#include <linux/clk-provider.h>
#include <linux/completion.h>
#include <linux/errno.h>
#include <linux/i3c/master.h>
@@ -166,6 +167,31 @@
*/
#define SVC_I3C_QUIRK_DAA_CORRUPT BIT(2)
+/*
+ * AML_I3C_QUIRK_AO:
+ * AO is the Always-On domain on Amlogic SoCs (remains powered in suspend).
+ * Its clock is generated by an internal divider located in the I3C
+ * controller's own register space.
+ */
+#define AML_I3C_QUIRK_AO BIT(3)
+
+/*
+ * Quirks shared by all the Amlogic A9 I3C instances.
+ */
+#define AML_A9_I3C_QUIRKS (SVC_I3C_QUIRK_FIFO_EMPTY | \
+ SVC_I3C_QUIRK_FALSE_SLVSTART | \
+ SVC_I3C_QUIRK_DAA_CORRUPT)
+
+/* Amlogic specific registers */
+#define AML_AO_I3C_FCLK_DIV_CTRL 0x304
+#define AML_AO_I3C_FCLK_DIV_EN_SHIFT 8
+#define AML_AO_I3C_FCLK_DIV_SHIFT 0
+#define AML_AO_I3C_FCLK_DIV_WIDTH 8
+#define AML_AO_I3C_FCLK_EN_SHIFT 9
+
+/* Target fast clock rate of the Amlogic A9 I3C controllers, in Hz */
+#define AML_I3C_FAST_CLK_RATE 25000000
+
struct svc_i3c_cmd {
u8 addr;
union {
@@ -195,8 +221,11 @@ struct svc_i3c_regs_save {
u32 mdynaddr;
};
+struct svc_i3c_master;
+
struct svc_i3c_drvdata {
u32 quirks;
+ int (*vendor_init)(struct svc_i3c_master *master);
};
/**
@@ -2011,6 +2040,7 @@ static int svc_i3c_master_probe(struct platform_device *pdev)
if (!master->drvdata)
return -EINVAL;
+ master->dev = dev;
master->regs = devm_platform_ioremap_resource(pdev, 0);
if (IS_ERR(master->regs))
return PTR_ERR(master->regs);
@@ -2019,27 +2049,43 @@ static int svc_i3c_master_probe(struct platform_device *pdev)
if (master->num_clks < 0)
return dev_err_probe(dev, -EINVAL, "can't get I3C clocks\n");
+ ret = clk_bulk_prepare_enable(master->num_clks, master->clks);
+ if (ret)
+ return dev_err_probe(dev, ret, "can't enable I3C clocks\n");
+
+ /*
+ * Vendor specific setup may register additional clocks (e.g. the
+ * internal divider chain of the AO variant) and program clock rates,
+ * so it must run with the clocks enabled.
+ */
+ if (master->drvdata->vendor_init) {
+ ret = master->drvdata->vendor_init(master);
+ if (ret)
+ goto err_disable_clks;
+ }
+
for (i = 0; i < master->num_clks; i++) {
if (!strcmp(master->clks[i].id, "fast_clk"))
break;
}
- if (i == master->num_clks)
- return dev_err_probe(dev, -EINVAL,
- "can't get I3C peripheral clock\n");
+ if (i == master->num_clks) {
+ ret = dev_err_probe(dev, -EINVAL,
+ "can't get I3C peripheral clock\n");
+ goto err_disable_clks;
+ }
master->fclk = master->clks[i].clk;
- if (IS_ERR(master->fclk))
- return PTR_ERR(master->fclk);
+ if (IS_ERR(master->fclk)) {
+ ret = PTR_ERR(master->fclk);
+ goto err_disable_clks;
+ }
master->irq = platform_get_irq(pdev, 0);
- if (master->irq < 0)
- return master->irq;
-
- master->dev = dev;
- ret = clk_bulk_prepare_enable(master->num_clks, master->clks);
- if (ret)
- return dev_err_probe(dev, ret, "can't enable I3C clocks\n");
+ if (master->irq < 0) {
+ ret = master->irq;
+ goto err_disable_clks;
+ }
mutex_init(&master->lock);
@@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
static const struct svc_i3c_drvdata svc_default_drvdata = {};
+/**
+ * struct aml_i3c_ao_clk - container for the AO variant's internal fast clock
+ * @div_en: enables the internal divider
+ * @div: bus clock divider
+ * @gate: output gate of the generated fast clock
+ * @lock: spinlock shared by the gate and divider register operations
+ *
+ * The clock HW structures are registered with the clock framework and must
+ * stay alive for the whole driver lifetime, so they are kept along with the
+ * shared lock used by their gate/divider operations.
+ */
+struct aml_i3c_ao_clk {
+ struct clk_gate div_en;
+ struct clk_divider div;
+ struct clk_gate gate;
+ spinlock_t lock;
+};
+
+/**
+ * aml_ao_i3c_tail_clocks - register the AO variant's internal fast clock
+ * @master: I3C master controller
+ * @parent: parent clock of the internal divider chain; must be "pclk", the
+ * only clock the AO variant exposes (see silvaco,i3c-master.yaml)
+ *
+ * Register the internal divider chain that produces the AO variant's
+ * fast_clk and append it to the clk bulk array as "fast_clk". The caller
+ * must run with the clocks enabled since the divider registers are
+ * accessed during registration. The appended clock is prepared and
+ * enabled here as well, because clk_bulk_prepare_enable() already ran
+ * in the probe with the original clock list.
+ *
+ * Return: 0 on success, or a negative error code.
+ */
+static int aml_ao_i3c_tail_clocks(struct svc_i3c_master *master, struct clk *parent)
+{
+ struct device *dev = master->dev;
+ struct clk_bulk_data *new_clks;
+ struct clk_init_data init = {};
+ struct aml_i3c_ao_clk *ao;
+ struct clk *fclk;
+ char name[64];
+ int ret;
+
+ ao = devm_kzalloc(dev, sizeof(*ao), GFP_KERNEL);
+ if (!ao)
+ return -ENOMEM;
+
+ spin_lock_init(&ao->lock);
+
+ snprintf(name, sizeof(name), "%s#div_en", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_gate_ops;
+ init.flags = CLK_IGNORE_UNUSED | CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = __clk_get_hw(parent),
+ };
+ init.num_parents = 1;
+
+ ao->div_en.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->div_en.bit_idx = AML_AO_I3C_FCLK_DIV_EN_SHIFT;
+ ao->div_en.hw.init = &init;
+ ao->div_en.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->div_en.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ snprintf(name, sizeof(name), "%s#div", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_divider_ops;
+ init.flags = CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = &ao->div_en.hw,
+ };
+ init.num_parents = 1;
+
+ ao->div.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->div.shift = AML_AO_I3C_FCLK_DIV_SHIFT;
+ ao->div.width = AML_AO_I3C_FCLK_DIV_WIDTH;
+ ao->div.hw.init = &init;
+ ao->div.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->div.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ snprintf(name, sizeof(name), "%s#gate", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_gate_ops;
+ init.flags = CLK_SET_RATE_PARENT | CLK_IGNORE_UNUSED | CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = &ao->div.hw,
+ };
+ init.num_parents = 1;
+
+ ao->gate.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->gate.bit_idx = AML_AO_I3C_FCLK_EN_SHIFT;
+ ao->gate.hw.init = &init;
+ ao->gate.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->gate.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ fclk = devm_clk_hw_get_clk(dev, &ao->gate.hw, NULL);
+ if (IS_ERR(fclk))
+ return dev_err_probe(dev, PTR_ERR(fclk),
+ "tail i3c fast clock failed\n");
+
+ /*
+ * Keep the clk bulk state balanced for the following
+ * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
+ */
+ ret = clk_prepare_enable(fclk);
+ if (ret)
+ return dev_err_probe(dev, ret, "can't enable fast_clk\n");
+
+ new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
+ GFP_KERNEL);
+ if (!new_clks) {
+ clk_disable_unprepare(fclk);
+ return -ENOMEM;
+ }
+
+ memcpy(new_clks, master->clks,
+ master->num_clks * sizeof(*master->clks));
+ new_clks[master->num_clks].id = "fast_clk";
+ new_clks[master->num_clks].clk = fclk;
+ master->clks = new_clks;
+ master->num_clks++;
+
+ return 0;
+}
+
+/**
+ * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
+ * @master: I3C master controller
+ *
+ * Register the AO variant's internal divider chain and round/set the fast
+ * clock rate for both variants.
+ *
+ * Return: 0 on success, or a negative error code.
+ */
+static int aml_i3c_init(struct svc_i3c_master *master)
+{
+ struct device *dev = master->dev;
+ struct clk *fclk;
+ long fin_freq;
+ int ret, i;
+
+ if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
+ /*
+ * clk_bulk_get_all() preserves the DT order, and the
+ * binding restricts the AO variant to exactly one clock,
+ * which is always "pclk". So clks[0] is the input of the
+ * internal divider chain.
+ */
+ if (strcmp(master->clks[0].id, "pclk"))
+ return dev_err_probe(dev, -EINVAL,
+ "invalid AO input clock\n");
+
+ ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
+ if (ret)
+ return ret;
+ }
+
+ for (i = 0; i < master->num_clks; i++) {
+ if (!strcmp(master->clks[i].id, "fast_clk"))
+ break;
+ }
+
+ if (i == master->num_clks)
+ return dev_err_probe(dev, -EINVAL,
+ "can't get I3C fast clock\n");
+
+ fclk = master->clks[i].clk;
+ if (IS_ERR(fclk))
+ return PTR_ERR(fclk);
+
+ fin_freq = clk_round_rate(fclk, AML_I3C_FAST_CLK_RATE);
+ if (fin_freq <= 0)
+ return dev_err_probe(dev, fin_freq ? fin_freq : -EINVAL,
+ "invalid fast_clk\n");
+
+ ret = clk_set_rate(fclk, fin_freq);
+ if (ret)
+ return dev_err_probe(dev, ret, "setting fast_clk rate failed\n");
+
+ return 0;
+}
+
+static const struct svc_i3c_drvdata aml_a9_drvdata = {
+ .quirks = AML_A9_I3C_QUIRKS,
+ .vendor_init = aml_i3c_init,
+};
+
+static const struct svc_i3c_drvdata aml_a9_ao_drvdata = {
+ .quirks = AML_A9_I3C_QUIRKS | AML_I3C_QUIRK_AO,
+ .vendor_init = aml_i3c_init,
+};
+
static const struct of_device_id svc_i3c_master_of_match_tbl[] = {
{ .compatible = "nuvoton,npcm845-i3c", .data = &npcm845_drvdata },
{ .compatible = "silvaco,i3c-master-v1", .data = &svc_default_drvdata },
+ { .compatible = "amlogic,a9-i3c", .data = &aml_a9_drvdata },
+ { .compatible = "amlogic,a9-ao-i3c", .data = &aml_a9_ao_drvdata },
{ /* sentinel */ },
};
MODULE_DEVICE_TABLE(of, svc_i3c_master_of_match_tbl);
--
2.42.0
_______________________________________________
linux-amlogic mailing list
linux-amlogic@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-amlogic
^ permalink raw reply related [flat|nested] 30+ messages in thread* [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-16 10:49 ` Junyi Zhao via B4 Relay
0 siblings, 0 replies; 30+ messages in thread
From: Junyi Zhao via B4 Relay @ 2026-11-16 10:49 UTC (permalink / raw)
To: Alexandre Belloni, Frank Li, Rob Herring, Krzysztof Kozlowski,
Conor Dooley, Neil Armstrong, Kevin Hilman, Jerome Brunet,
Martin Blumenstingl, Miquel Raynal, Kees Cook,
Gustavo A. R. Silva
Cc: linux-i3c, devicetree, linux-kernel, linux-arm-kernel,
linux-amlogic, imx, linux-hardening, xianwei.zhao, Kelvin.Zhang,
Junyi Zhao
From: Junyi Zhao <junyi.zhao@amlogic.com>
Amlogic a9 SoCs integrate the Silvaco I3C IP, and their peripheral
clocks vary from the vanilla implementation:
- The EE-domain controller (amlogic,a9-i3c) takes two clocks,
"pclk" and "fast_clk", and the fast clock rate must be programmed
before the controller can be used.
- The AO-domain controller (amlogic,a9-ao-i3c) exposes only "pclk":
its bus clock is generated by an internal divider located in the
controller's own register space.
Both variants need setup that the generic driver does not perform:
the fast clock rate must be rounded and set at probe time, and on the
AO variant the internal divider must be registered with the clock
framework so that "fast_clk" even exists. The divider registers sit
in the controller's own register space, so they can only be accessed
while pclk is running.
Add a vendor_init callback, invoked from the probe with the clocks
enabled, to handle this. On the AO variant it registers the internal
divider chain (div_en -> div -> gate) with the common clock framework
and appends the resulting clock to the clk bulk array as "fast_clk";
for both variants it then rounds and sets the fast clock rate
(target 25 MHz). As vendor_init now runs after
clk_bulk_prepare_enable(), the probe error paths that follow disable
the clocks on failure.
Extend the quirks bitmask with an Amlogic-specific bit
(AML_I3C_QUIRK_AO) marking the AO variant's internal-divider
behaviour. Since the driver now registers clocks through the common
clock framework, SVC_I3C_MASTER also depends on COMMON_CLK.
Signed-off-by: Junyi Zhao <junyi.zhao@amlogic.com>
---
drivers/i3c/master/Kconfig | 1 +
drivers/i3c/master/svc-i3c-master.c | 276 ++++++++++++++++++++++++++++++++++--
2 files changed, 265 insertions(+), 12 deletions(-)
diff --git a/drivers/i3c/master/Kconfig b/drivers/i3c/master/Kconfig
index da96d2aaa..ac7d246c2 100644
--- a/drivers/i3c/master/Kconfig
+++ b/drivers/i3c/master/Kconfig
@@ -47,6 +47,7 @@ config AST2600_I3C_MASTER
config SVC_I3C_MASTER
tristate "Silvaco I3C Dual-Role Master driver"
+ depends on COMMON_CLK
depends on HAS_IOMEM
depends on !(ALPHA || PARISC)
help
diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
index e700392f0..0268fc713 100644
--- a/drivers/i3c/master/svc-i3c-master.c
+++ b/drivers/i3c/master/svc-i3c-master.c
@@ -9,6 +9,7 @@
#include <linux/bitfield.h>
#include <linux/clk.h>
+#include <linux/clk-provider.h>
#include <linux/completion.h>
#include <linux/errno.h>
#include <linux/i3c/master.h>
@@ -166,6 +167,31 @@
*/
#define SVC_I3C_QUIRK_DAA_CORRUPT BIT(2)
+/*
+ * AML_I3C_QUIRK_AO:
+ * AO is the Always-On domain on Amlogic SoCs (remains powered in suspend).
+ * Its clock is generated by an internal divider located in the I3C
+ * controller's own register space.
+ */
+#define AML_I3C_QUIRK_AO BIT(3)
+
+/*
+ * Quirks shared by all the Amlogic A9 I3C instances.
+ */
+#define AML_A9_I3C_QUIRKS (SVC_I3C_QUIRK_FIFO_EMPTY | \
+ SVC_I3C_QUIRK_FALSE_SLVSTART | \
+ SVC_I3C_QUIRK_DAA_CORRUPT)
+
+/* Amlogic specific registers */
+#define AML_AO_I3C_FCLK_DIV_CTRL 0x304
+#define AML_AO_I3C_FCLK_DIV_EN_SHIFT 8
+#define AML_AO_I3C_FCLK_DIV_SHIFT 0
+#define AML_AO_I3C_FCLK_DIV_WIDTH 8
+#define AML_AO_I3C_FCLK_EN_SHIFT 9
+
+/* Target fast clock rate of the Amlogic A9 I3C controllers, in Hz */
+#define AML_I3C_FAST_CLK_RATE 25000000
+
struct svc_i3c_cmd {
u8 addr;
union {
@@ -195,8 +221,11 @@ struct svc_i3c_regs_save {
u32 mdynaddr;
};
+struct svc_i3c_master;
+
struct svc_i3c_drvdata {
u32 quirks;
+ int (*vendor_init)(struct svc_i3c_master *master);
};
/**
@@ -2011,6 +2040,7 @@ static int svc_i3c_master_probe(struct platform_device *pdev)
if (!master->drvdata)
return -EINVAL;
+ master->dev = dev;
master->regs = devm_platform_ioremap_resource(pdev, 0);
if (IS_ERR(master->regs))
return PTR_ERR(master->regs);
@@ -2019,27 +2049,43 @@ static int svc_i3c_master_probe(struct platform_device *pdev)
if (master->num_clks < 0)
return dev_err_probe(dev, -EINVAL, "can't get I3C clocks\n");
+ ret = clk_bulk_prepare_enable(master->num_clks, master->clks);
+ if (ret)
+ return dev_err_probe(dev, ret, "can't enable I3C clocks\n");
+
+ /*
+ * Vendor specific setup may register additional clocks (e.g. the
+ * internal divider chain of the AO variant) and program clock rates,
+ * so it must run with the clocks enabled.
+ */
+ if (master->drvdata->vendor_init) {
+ ret = master->drvdata->vendor_init(master);
+ if (ret)
+ goto err_disable_clks;
+ }
+
for (i = 0; i < master->num_clks; i++) {
if (!strcmp(master->clks[i].id, "fast_clk"))
break;
}
- if (i == master->num_clks)
- return dev_err_probe(dev, -EINVAL,
- "can't get I3C peripheral clock\n");
+ if (i == master->num_clks) {
+ ret = dev_err_probe(dev, -EINVAL,
+ "can't get I3C peripheral clock\n");
+ goto err_disable_clks;
+ }
master->fclk = master->clks[i].clk;
- if (IS_ERR(master->fclk))
- return PTR_ERR(master->fclk);
+ if (IS_ERR(master->fclk)) {
+ ret = PTR_ERR(master->fclk);
+ goto err_disable_clks;
+ }
master->irq = platform_get_irq(pdev, 0);
- if (master->irq < 0)
- return master->irq;
-
- master->dev = dev;
- ret = clk_bulk_prepare_enable(master->num_clks, master->clks);
- if (ret)
- return dev_err_probe(dev, ret, "can't enable I3C clocks\n");
+ if (master->irq < 0) {
+ ret = master->irq;
+ goto err_disable_clks;
+ }
mutex_init(&master->lock);
@@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
static const struct svc_i3c_drvdata svc_default_drvdata = {};
+/**
+ * struct aml_i3c_ao_clk - container for the AO variant's internal fast clock
+ * @div_en: enables the internal divider
+ * @div: bus clock divider
+ * @gate: output gate of the generated fast clock
+ * @lock: spinlock shared by the gate and divider register operations
+ *
+ * The clock HW structures are registered with the clock framework and must
+ * stay alive for the whole driver lifetime, so they are kept along with the
+ * shared lock used by their gate/divider operations.
+ */
+struct aml_i3c_ao_clk {
+ struct clk_gate div_en;
+ struct clk_divider div;
+ struct clk_gate gate;
+ spinlock_t lock;
+};
+
+/**
+ * aml_ao_i3c_tail_clocks - register the AO variant's internal fast clock
+ * @master: I3C master controller
+ * @parent: parent clock of the internal divider chain; must be "pclk", the
+ * only clock the AO variant exposes (see silvaco,i3c-master.yaml)
+ *
+ * Register the internal divider chain that produces the AO variant's
+ * fast_clk and append it to the clk bulk array as "fast_clk". The caller
+ * must run with the clocks enabled since the divider registers are
+ * accessed during registration. The appended clock is prepared and
+ * enabled here as well, because clk_bulk_prepare_enable() already ran
+ * in the probe with the original clock list.
+ *
+ * Return: 0 on success, or a negative error code.
+ */
+static int aml_ao_i3c_tail_clocks(struct svc_i3c_master *master, struct clk *parent)
+{
+ struct device *dev = master->dev;
+ struct clk_bulk_data *new_clks;
+ struct clk_init_data init = {};
+ struct aml_i3c_ao_clk *ao;
+ struct clk *fclk;
+ char name[64];
+ int ret;
+
+ ao = devm_kzalloc(dev, sizeof(*ao), GFP_KERNEL);
+ if (!ao)
+ return -ENOMEM;
+
+ spin_lock_init(&ao->lock);
+
+ snprintf(name, sizeof(name), "%s#div_en", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_gate_ops;
+ init.flags = CLK_IGNORE_UNUSED | CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = __clk_get_hw(parent),
+ };
+ init.num_parents = 1;
+
+ ao->div_en.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->div_en.bit_idx = AML_AO_I3C_FCLK_DIV_EN_SHIFT;
+ ao->div_en.hw.init = &init;
+ ao->div_en.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->div_en.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ snprintf(name, sizeof(name), "%s#div", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_divider_ops;
+ init.flags = CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = &ao->div_en.hw,
+ };
+ init.num_parents = 1;
+
+ ao->div.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->div.shift = AML_AO_I3C_FCLK_DIV_SHIFT;
+ ao->div.width = AML_AO_I3C_FCLK_DIV_WIDTH;
+ ao->div.hw.init = &init;
+ ao->div.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->div.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ snprintf(name, sizeof(name), "%s#gate", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_gate_ops;
+ init.flags = CLK_SET_RATE_PARENT | CLK_IGNORE_UNUSED | CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = &ao->div.hw,
+ };
+ init.num_parents = 1;
+
+ ao->gate.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->gate.bit_idx = AML_AO_I3C_FCLK_EN_SHIFT;
+ ao->gate.hw.init = &init;
+ ao->gate.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->gate.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ fclk = devm_clk_hw_get_clk(dev, &ao->gate.hw, NULL);
+ if (IS_ERR(fclk))
+ return dev_err_probe(dev, PTR_ERR(fclk),
+ "tail i3c fast clock failed\n");
+
+ /*
+ * Keep the clk bulk state balanced for the following
+ * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
+ */
+ ret = clk_prepare_enable(fclk);
+ if (ret)
+ return dev_err_probe(dev, ret, "can't enable fast_clk\n");
+
+ new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
+ GFP_KERNEL);
+ if (!new_clks) {
+ clk_disable_unprepare(fclk);
+ return -ENOMEM;
+ }
+
+ memcpy(new_clks, master->clks,
+ master->num_clks * sizeof(*master->clks));
+ new_clks[master->num_clks].id = "fast_clk";
+ new_clks[master->num_clks].clk = fclk;
+ master->clks = new_clks;
+ master->num_clks++;
+
+ return 0;
+}
+
+/**
+ * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
+ * @master: I3C master controller
+ *
+ * Register the AO variant's internal divider chain and round/set the fast
+ * clock rate for both variants.
+ *
+ * Return: 0 on success, or a negative error code.
+ */
+static int aml_i3c_init(struct svc_i3c_master *master)
+{
+ struct device *dev = master->dev;
+ struct clk *fclk;
+ long fin_freq;
+ int ret, i;
+
+ if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
+ /*
+ * clk_bulk_get_all() preserves the DT order, and the
+ * binding restricts the AO variant to exactly one clock,
+ * which is always "pclk". So clks[0] is the input of the
+ * internal divider chain.
+ */
+ if (strcmp(master->clks[0].id, "pclk"))
+ return dev_err_probe(dev, -EINVAL,
+ "invalid AO input clock\n");
+
+ ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
+ if (ret)
+ return ret;
+ }
+
+ for (i = 0; i < master->num_clks; i++) {
+ if (!strcmp(master->clks[i].id, "fast_clk"))
+ break;
+ }
+
+ if (i == master->num_clks)
+ return dev_err_probe(dev, -EINVAL,
+ "can't get I3C fast clock\n");
+
+ fclk = master->clks[i].clk;
+ if (IS_ERR(fclk))
+ return PTR_ERR(fclk);
+
+ fin_freq = clk_round_rate(fclk, AML_I3C_FAST_CLK_RATE);
+ if (fin_freq <= 0)
+ return dev_err_probe(dev, fin_freq ? fin_freq : -EINVAL,
+ "invalid fast_clk\n");
+
+ ret = clk_set_rate(fclk, fin_freq);
+ if (ret)
+ return dev_err_probe(dev, ret, "setting fast_clk rate failed\n");
+
+ return 0;
+}
+
+static const struct svc_i3c_drvdata aml_a9_drvdata = {
+ .quirks = AML_A9_I3C_QUIRKS,
+ .vendor_init = aml_i3c_init,
+};
+
+static const struct svc_i3c_drvdata aml_a9_ao_drvdata = {
+ .quirks = AML_A9_I3C_QUIRKS | AML_I3C_QUIRK_AO,
+ .vendor_init = aml_i3c_init,
+};
+
static const struct of_device_id svc_i3c_master_of_match_tbl[] = {
{ .compatible = "nuvoton,npcm845-i3c", .data = &npcm845_drvdata },
{ .compatible = "silvaco,i3c-master-v1", .data = &svc_default_drvdata },
+ { .compatible = "amlogic,a9-i3c", .data = &aml_a9_drvdata },
+ { .compatible = "amlogic,a9-ao-i3c", .data = &aml_a9_ao_drvdata },
{ /* sentinel */ },
};
MODULE_DEVICE_TABLE(of, svc_i3c_master_of_match_tbl);
--
2.42.0
--
linux-i3c mailing list
linux-i3c@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-i3c
^ permalink raw reply related [flat|nested] 30+ messages in thread* [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-16 10:49 ` Junyi Zhao via B4 Relay
0 siblings, 0 replies; 30+ messages in thread
From: Junyi Zhao via B4 Relay @ 2026-11-16 10:49 UTC (permalink / raw)
To: Alexandre Belloni, Frank Li, Rob Herring, Krzysztof Kozlowski,
Conor Dooley, Neil Armstrong, Kevin Hilman, Jerome Brunet,
Martin Blumenstingl, Miquel Raynal, Kees Cook,
Gustavo A. R. Silva
Cc: linux-i3c, devicetree, linux-kernel, linux-arm-kernel,
linux-amlogic, imx, linux-hardening, xianwei.zhao, Kelvin.Zhang,
Junyi Zhao
From: Junyi Zhao <junyi.zhao@amlogic.com>
Amlogic a9 SoCs integrate the Silvaco I3C IP, and their peripheral
clocks vary from the vanilla implementation:
- The EE-domain controller (amlogic,a9-i3c) takes two clocks,
"pclk" and "fast_clk", and the fast clock rate must be programmed
before the controller can be used.
- The AO-domain controller (amlogic,a9-ao-i3c) exposes only "pclk":
its bus clock is generated by an internal divider located in the
controller's own register space.
Both variants need setup that the generic driver does not perform:
the fast clock rate must be rounded and set at probe time, and on the
AO variant the internal divider must be registered with the clock
framework so that "fast_clk" even exists. The divider registers sit
in the controller's own register space, so they can only be accessed
while pclk is running.
Add a vendor_init callback, invoked from the probe with the clocks
enabled, to handle this. On the AO variant it registers the internal
divider chain (div_en -> div -> gate) with the common clock framework
and appends the resulting clock to the clk bulk array as "fast_clk";
for both variants it then rounds and sets the fast clock rate
(target 25 MHz). As vendor_init now runs after
clk_bulk_prepare_enable(), the probe error paths that follow disable
the clocks on failure.
Extend the quirks bitmask with an Amlogic-specific bit
(AML_I3C_QUIRK_AO) marking the AO variant's internal-divider
behaviour. Since the driver now registers clocks through the common
clock framework, SVC_I3C_MASTER also depends on COMMON_CLK.
Signed-off-by: Junyi Zhao <junyi.zhao@amlogic.com>
---
drivers/i3c/master/Kconfig | 1 +
drivers/i3c/master/svc-i3c-master.c | 276 ++++++++++++++++++++++++++++++++++--
2 files changed, 265 insertions(+), 12 deletions(-)
diff --git a/drivers/i3c/master/Kconfig b/drivers/i3c/master/Kconfig
index da96d2aaa..ac7d246c2 100644
--- a/drivers/i3c/master/Kconfig
+++ b/drivers/i3c/master/Kconfig
@@ -47,6 +47,7 @@ config AST2600_I3C_MASTER
config SVC_I3C_MASTER
tristate "Silvaco I3C Dual-Role Master driver"
+ depends on COMMON_CLK
depends on HAS_IOMEM
depends on !(ALPHA || PARISC)
help
diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
index e700392f0..0268fc713 100644
--- a/drivers/i3c/master/svc-i3c-master.c
+++ b/drivers/i3c/master/svc-i3c-master.c
@@ -9,6 +9,7 @@
#include <linux/bitfield.h>
#include <linux/clk.h>
+#include <linux/clk-provider.h>
#include <linux/completion.h>
#include <linux/errno.h>
#include <linux/i3c/master.h>
@@ -166,6 +167,31 @@
*/
#define SVC_I3C_QUIRK_DAA_CORRUPT BIT(2)
+/*
+ * AML_I3C_QUIRK_AO:
+ * AO is the Always-On domain on Amlogic SoCs (remains powered in suspend).
+ * Its clock is generated by an internal divider located in the I3C
+ * controller's own register space.
+ */
+#define AML_I3C_QUIRK_AO BIT(3)
+
+/*
+ * Quirks shared by all the Amlogic A9 I3C instances.
+ */
+#define AML_A9_I3C_QUIRKS (SVC_I3C_QUIRK_FIFO_EMPTY | \
+ SVC_I3C_QUIRK_FALSE_SLVSTART | \
+ SVC_I3C_QUIRK_DAA_CORRUPT)
+
+/* Amlogic specific registers */
+#define AML_AO_I3C_FCLK_DIV_CTRL 0x304
+#define AML_AO_I3C_FCLK_DIV_EN_SHIFT 8
+#define AML_AO_I3C_FCLK_DIV_SHIFT 0
+#define AML_AO_I3C_FCLK_DIV_WIDTH 8
+#define AML_AO_I3C_FCLK_EN_SHIFT 9
+
+/* Target fast clock rate of the Amlogic A9 I3C controllers, in Hz */
+#define AML_I3C_FAST_CLK_RATE 25000000
+
struct svc_i3c_cmd {
u8 addr;
union {
@@ -195,8 +221,11 @@ struct svc_i3c_regs_save {
u32 mdynaddr;
};
+struct svc_i3c_master;
+
struct svc_i3c_drvdata {
u32 quirks;
+ int (*vendor_init)(struct svc_i3c_master *master);
};
/**
@@ -2011,6 +2040,7 @@ static int svc_i3c_master_probe(struct platform_device *pdev)
if (!master->drvdata)
return -EINVAL;
+ master->dev = dev;
master->regs = devm_platform_ioremap_resource(pdev, 0);
if (IS_ERR(master->regs))
return PTR_ERR(master->regs);
@@ -2019,27 +2049,43 @@ static int svc_i3c_master_probe(struct platform_device *pdev)
if (master->num_clks < 0)
return dev_err_probe(dev, -EINVAL, "can't get I3C clocks\n");
+ ret = clk_bulk_prepare_enable(master->num_clks, master->clks);
+ if (ret)
+ return dev_err_probe(dev, ret, "can't enable I3C clocks\n");
+
+ /*
+ * Vendor specific setup may register additional clocks (e.g. the
+ * internal divider chain of the AO variant) and program clock rates,
+ * so it must run with the clocks enabled.
+ */
+ if (master->drvdata->vendor_init) {
+ ret = master->drvdata->vendor_init(master);
+ if (ret)
+ goto err_disable_clks;
+ }
+
for (i = 0; i < master->num_clks; i++) {
if (!strcmp(master->clks[i].id, "fast_clk"))
break;
}
- if (i == master->num_clks)
- return dev_err_probe(dev, -EINVAL,
- "can't get I3C peripheral clock\n");
+ if (i == master->num_clks) {
+ ret = dev_err_probe(dev, -EINVAL,
+ "can't get I3C peripheral clock\n");
+ goto err_disable_clks;
+ }
master->fclk = master->clks[i].clk;
- if (IS_ERR(master->fclk))
- return PTR_ERR(master->fclk);
+ if (IS_ERR(master->fclk)) {
+ ret = PTR_ERR(master->fclk);
+ goto err_disable_clks;
+ }
master->irq = platform_get_irq(pdev, 0);
- if (master->irq < 0)
- return master->irq;
-
- master->dev = dev;
- ret = clk_bulk_prepare_enable(master->num_clks, master->clks);
- if (ret)
- return dev_err_probe(dev, ret, "can't enable I3C clocks\n");
+ if (master->irq < 0) {
+ ret = master->irq;
+ goto err_disable_clks;
+ }
mutex_init(&master->lock);
@@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
static const struct svc_i3c_drvdata svc_default_drvdata = {};
+/**
+ * struct aml_i3c_ao_clk - container for the AO variant's internal fast clock
+ * @div_en: enables the internal divider
+ * @div: bus clock divider
+ * @gate: output gate of the generated fast clock
+ * @lock: spinlock shared by the gate and divider register operations
+ *
+ * The clock HW structures are registered with the clock framework and must
+ * stay alive for the whole driver lifetime, so they are kept along with the
+ * shared lock used by their gate/divider operations.
+ */
+struct aml_i3c_ao_clk {
+ struct clk_gate div_en;
+ struct clk_divider div;
+ struct clk_gate gate;
+ spinlock_t lock;
+};
+
+/**
+ * aml_ao_i3c_tail_clocks - register the AO variant's internal fast clock
+ * @master: I3C master controller
+ * @parent: parent clock of the internal divider chain; must be "pclk", the
+ * only clock the AO variant exposes (see silvaco,i3c-master.yaml)
+ *
+ * Register the internal divider chain that produces the AO variant's
+ * fast_clk and append it to the clk bulk array as "fast_clk". The caller
+ * must run with the clocks enabled since the divider registers are
+ * accessed during registration. The appended clock is prepared and
+ * enabled here as well, because clk_bulk_prepare_enable() already ran
+ * in the probe with the original clock list.
+ *
+ * Return: 0 on success, or a negative error code.
+ */
+static int aml_ao_i3c_tail_clocks(struct svc_i3c_master *master, struct clk *parent)
+{
+ struct device *dev = master->dev;
+ struct clk_bulk_data *new_clks;
+ struct clk_init_data init = {};
+ struct aml_i3c_ao_clk *ao;
+ struct clk *fclk;
+ char name[64];
+ int ret;
+
+ ao = devm_kzalloc(dev, sizeof(*ao), GFP_KERNEL);
+ if (!ao)
+ return -ENOMEM;
+
+ spin_lock_init(&ao->lock);
+
+ snprintf(name, sizeof(name), "%s#div_en", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_gate_ops;
+ init.flags = CLK_IGNORE_UNUSED | CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = __clk_get_hw(parent),
+ };
+ init.num_parents = 1;
+
+ ao->div_en.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->div_en.bit_idx = AML_AO_I3C_FCLK_DIV_EN_SHIFT;
+ ao->div_en.hw.init = &init;
+ ao->div_en.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->div_en.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ snprintf(name, sizeof(name), "%s#div", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_divider_ops;
+ init.flags = CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = &ao->div_en.hw,
+ };
+ init.num_parents = 1;
+
+ ao->div.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->div.shift = AML_AO_I3C_FCLK_DIV_SHIFT;
+ ao->div.width = AML_AO_I3C_FCLK_DIV_WIDTH;
+ ao->div.hw.init = &init;
+ ao->div.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->div.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ snprintf(name, sizeof(name), "%s#gate", dev_name(dev));
+ init.name = name;
+ init.ops = &clk_gate_ops;
+ init.flags = CLK_SET_RATE_PARENT | CLK_IGNORE_UNUSED | CLK_OPS_PARENT_ENABLE;
+ init.parent_data = &(struct clk_parent_data) {
+ .index = -1,
+ .hw = &ao->div.hw,
+ };
+ init.num_parents = 1;
+
+ ao->gate.reg = master->regs + AML_AO_I3C_FCLK_DIV_CTRL;
+ ao->gate.bit_idx = AML_AO_I3C_FCLK_EN_SHIFT;
+ ao->gate.hw.init = &init;
+ ao->gate.lock = &ao->lock;
+
+ ret = devm_clk_hw_register(dev, &ao->gate.hw);
+ if (ret)
+ return dev_err_probe(dev, ret, "failed to register %s\n", name);
+
+ fclk = devm_clk_hw_get_clk(dev, &ao->gate.hw, NULL);
+ if (IS_ERR(fclk))
+ return dev_err_probe(dev, PTR_ERR(fclk),
+ "tail i3c fast clock failed\n");
+
+ /*
+ * Keep the clk bulk state balanced for the following
+ * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
+ */
+ ret = clk_prepare_enable(fclk);
+ if (ret)
+ return dev_err_probe(dev, ret, "can't enable fast_clk\n");
+
+ new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
+ GFP_KERNEL);
+ if (!new_clks) {
+ clk_disable_unprepare(fclk);
+ return -ENOMEM;
+ }
+
+ memcpy(new_clks, master->clks,
+ master->num_clks * sizeof(*master->clks));
+ new_clks[master->num_clks].id = "fast_clk";
+ new_clks[master->num_clks].clk = fclk;
+ master->clks = new_clks;
+ master->num_clks++;
+
+ return 0;
+}
+
+/**
+ * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
+ * @master: I3C master controller
+ *
+ * Register the AO variant's internal divider chain and round/set the fast
+ * clock rate for both variants.
+ *
+ * Return: 0 on success, or a negative error code.
+ */
+static int aml_i3c_init(struct svc_i3c_master *master)
+{
+ struct device *dev = master->dev;
+ struct clk *fclk;
+ long fin_freq;
+ int ret, i;
+
+ if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
+ /*
+ * clk_bulk_get_all() preserves the DT order, and the
+ * binding restricts the AO variant to exactly one clock,
+ * which is always "pclk". So clks[0] is the input of the
+ * internal divider chain.
+ */
+ if (strcmp(master->clks[0].id, "pclk"))
+ return dev_err_probe(dev, -EINVAL,
+ "invalid AO input clock\n");
+
+ ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
+ if (ret)
+ return ret;
+ }
+
+ for (i = 0; i < master->num_clks; i++) {
+ if (!strcmp(master->clks[i].id, "fast_clk"))
+ break;
+ }
+
+ if (i == master->num_clks)
+ return dev_err_probe(dev, -EINVAL,
+ "can't get I3C fast clock\n");
+
+ fclk = master->clks[i].clk;
+ if (IS_ERR(fclk))
+ return PTR_ERR(fclk);
+
+ fin_freq = clk_round_rate(fclk, AML_I3C_FAST_CLK_RATE);
+ if (fin_freq <= 0)
+ return dev_err_probe(dev, fin_freq ? fin_freq : -EINVAL,
+ "invalid fast_clk\n");
+
+ ret = clk_set_rate(fclk, fin_freq);
+ if (ret)
+ return dev_err_probe(dev, ret, "setting fast_clk rate failed\n");
+
+ return 0;
+}
+
+static const struct svc_i3c_drvdata aml_a9_drvdata = {
+ .quirks = AML_A9_I3C_QUIRKS,
+ .vendor_init = aml_i3c_init,
+};
+
+static const struct svc_i3c_drvdata aml_a9_ao_drvdata = {
+ .quirks = AML_A9_I3C_QUIRKS | AML_I3C_QUIRK_AO,
+ .vendor_init = aml_i3c_init,
+};
+
static const struct of_device_id svc_i3c_master_of_match_tbl[] = {
{ .compatible = "nuvoton,npcm845-i3c", .data = &npcm845_drvdata },
{ .compatible = "silvaco,i3c-master-v1", .data = &svc_default_drvdata },
+ { .compatible = "amlogic,a9-i3c", .data = &aml_a9_drvdata },
+ { .compatible = "amlogic,a9-ao-i3c", .data = &aml_a9_ao_drvdata },
{ /* sentinel */ },
};
MODULE_DEVICE_TABLE(of, svc_i3c_master_of_match_tbl);
--
2.42.0
^ permalink raw reply related [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
2026-11-16 10:49 ` Junyi Zhao via B4 Relay
(?)
@ 2026-11-16 11:12 ` sashiko-bot
-1 siblings, 0 replies; 30+ messages in thread
From: sashiko-bot @ 2026-11-16 11:12 UTC (permalink / raw)
To: Junyi Zhao
Cc: linux-i3c, devicetree, linux-amlogic, imx, robh, conor+dt,
Frank.Li, neil.armstrong, Alexandre Belloni
Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [Medium] i3c: master: svc: NULL dereference of clks[0].id in
aml_i3c_init()
- [Medium] i3c: master: svc: clocks left enabled in
svc_i3c_master_remove()
- [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
--
commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
Author: Junyi Zhao <junyi.zhao@amlogic.com>
i3c: master: svc: Add amlogic a9 support
This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
I3C controllers, introducing a vendor_init callback to register the AO
variant's internal clock divider chain and configure the fast_clk rate.
> diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> index e700392f0801f..0268fc713afdf 100644
> --- a/drivers/i3c/master/svc-i3c-master.c
> +++ b/drivers/i3c/master/svc-i3c-master.c
[ ... ]
> @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
[ ... ]
> + /*
> + * Keep the clk bulk state balanced for the following
> + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> + */
> + ret = clk_prepare_enable(fclk);
> + if (ret)
> + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> +
> + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> + GFP_KERNEL);
> + if (!new_clks) {
> + clk_disable_unprepare(fclk);
> + return -ENOMEM;
> + }
> +
> + memcpy(new_clks, master->clks,
> + master->num_clks * sizeof(*master->clks));
> + new_clks[master->num_clks].id = "fast_clk";
> + new_clks[master->num_clks].clk = fclk;
> + master->clks = new_clks;
> + master->num_clks++;
> +
> + return 0;
> +}
[Severity: Medium]
Can unbinding the driver leave fclk and the input clocks enabled when the
device is not runtime-suspended?
svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
reference not released by i3c_master_unregister()), runtime suspend does
not run during remove. When devres then unregisters ao->gate.hw,
ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
clk_unregister() triggers a kernel warning and leaks the prepare and
enable counts on the parent pclk.
> +
> +/**
> + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> + * @master: I3C master controller
> + *
> + * Register the AO variant's internal divider chain and round/set the fast
> + * clock rate for both variants.
> + *
> + * Return: 0 on success, or a negative error code.
> + */
> +static int aml_i3c_init(struct svc_i3c_master *master)
> +{
> + struct device *dev = master->dev;
> + struct clk *fclk;
> + long fin_freq;
> + int ret, i;
> +
> + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> + /*
> + * clk_bulk_get_all() preserves the DT order, and the
> + * binding restricts the AO variant to exactly one clock,
> + * which is always "pclk". So clks[0] is the input of the
> + * internal divider chain.
> + */
> + if (strcmp(master->clks[0].id, "pclk"))
> + return dev_err_probe(dev, -EINVAL,
> + "invalid AO input clock\n");
[Severity: Medium]
Can this trigger a NULL pointer dereference on master->clks or
master->clks[0].id?
When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
and leaves master->clks as NULL, which passes the master->num_clks < 0
check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
first and indexes master->clks[0] without checking that master->num_clks
is at least 1 (or equal to 1 as required by the AO binding). In addition,
of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
shorter than clocks, which would cause strcmp() here and in the fast_clk
loop below to dereference NULL.
> +
> + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> + if (ret)
> + return ret;
> + }
> +
> + for (i = 0; i < master->num_clks; i++) {
> + if (!strcmp(master->clks[i].id, "fast_clk"))
> + break;
> + }
> +
> + if (i == master->num_clks)
> + return dev_err_probe(dev, -EINVAL,
> + "can't get I3C fast clock\n");
> +
> + fclk = master->clks[i].clk;
> + if (IS_ERR(fclk))
> + return PTR_ERR(fclk);
[Severity: Low]
This isn't a bug, but this fast_clk lookup and validation duplicates the
identical loop in svc_i3c_master_probe() that runs right after
vendor_init() returns. Could a shared helper or resolving master->fclk
once avoid repeating the lookup?
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-16 11:12 ` sashiko-bot
0 siblings, 0 replies; 30+ messages in thread
From: sashiko-bot @ 2026-11-16 11:12 UTC (permalink / raw)
To: Junyi Zhao
Cc: linux-i3c, devicetree, linux-amlogic, imx, robh, conor+dt,
Frank.Li, neil.armstrong, Alexandre Belloni
Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [Medium] i3c: master: svc: NULL dereference of clks[0].id in
aml_i3c_init()
- [Medium] i3c: master: svc: clocks left enabled in
svc_i3c_master_remove()
- [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
--
commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
Author: Junyi Zhao <junyi.zhao@amlogic.com>
i3c: master: svc: Add amlogic a9 support
This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
I3C controllers, introducing a vendor_init callback to register the AO
variant's internal clock divider chain and configure the fast_clk rate.
> diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> index e700392f0801f..0268fc713afdf 100644
> --- a/drivers/i3c/master/svc-i3c-master.c
> +++ b/drivers/i3c/master/svc-i3c-master.c
[ ... ]
> @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
[ ... ]
> + /*
> + * Keep the clk bulk state balanced for the following
> + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> + */
> + ret = clk_prepare_enable(fclk);
> + if (ret)
> + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> +
> + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> + GFP_KERNEL);
> + if (!new_clks) {
> + clk_disable_unprepare(fclk);
> + return -ENOMEM;
> + }
> +
> + memcpy(new_clks, master->clks,
> + master->num_clks * sizeof(*master->clks));
> + new_clks[master->num_clks].id = "fast_clk";
> + new_clks[master->num_clks].clk = fclk;
> + master->clks = new_clks;
> + master->num_clks++;
> +
> + return 0;
> +}
[Severity: Medium]
Can unbinding the driver leave fclk and the input clocks enabled when the
device is not runtime-suspended?
svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
reference not released by i3c_master_unregister()), runtime suspend does
not run during remove. When devres then unregisters ao->gate.hw,
ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
clk_unregister() triggers a kernel warning and leaks the prepare and
enable counts on the parent pclk.
> +
> +/**
> + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> + * @master: I3C master controller
> + *
> + * Register the AO variant's internal divider chain and round/set the fast
> + * clock rate for both variants.
> + *
> + * Return: 0 on success, or a negative error code.
> + */
> +static int aml_i3c_init(struct svc_i3c_master *master)
> +{
> + struct device *dev = master->dev;
> + struct clk *fclk;
> + long fin_freq;
> + int ret, i;
> +
> + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> + /*
> + * clk_bulk_get_all() preserves the DT order, and the
> + * binding restricts the AO variant to exactly one clock,
> + * which is always "pclk". So clks[0] is the input of the
> + * internal divider chain.
> + */
> + if (strcmp(master->clks[0].id, "pclk"))
> + return dev_err_probe(dev, -EINVAL,
> + "invalid AO input clock\n");
[Severity: Medium]
Can this trigger a NULL pointer dereference on master->clks or
master->clks[0].id?
When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
and leaves master->clks as NULL, which passes the master->num_clks < 0
check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
first and indexes master->clks[0] without checking that master->num_clks
is at least 1 (or equal to 1 as required by the AO binding). In addition,
of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
shorter than clocks, which would cause strcmp() here and in the fast_clk
loop below to dereference NULL.
> +
> + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> + if (ret)
> + return ret;
> + }
> +
> + for (i = 0; i < master->num_clks; i++) {
> + if (!strcmp(master->clks[i].id, "fast_clk"))
> + break;
> + }
> +
> + if (i == master->num_clks)
> + return dev_err_probe(dev, -EINVAL,
> + "can't get I3C fast clock\n");
> +
> + fclk = master->clks[i].clk;
> + if (IS_ERR(fclk))
> + return PTR_ERR(fclk);
[Severity: Low]
This isn't a bug, but this fast_clk lookup and validation duplicates the
identical loop in svc_i3c_master_probe() that runs right after
vendor_init() returns. Could a shared helper or resolving master->fclk
once avoid repeating the lookup?
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
_______________________________________________
linux-amlogic mailing list
linux-amlogic@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-amlogic
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-16 11:12 ` sashiko-bot
0 siblings, 0 replies; 30+ messages in thread
From: sashiko-bot @ 2026-11-16 11:12 UTC (permalink / raw)
To: Junyi Zhao
Cc: linux-i3c, devicetree, linux-amlogic, imx, robh, conor+dt,
Frank.Li, neil.armstrong, Alexandre Belloni
Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [Medium] i3c: master: svc: NULL dereference of clks[0].id in
aml_i3c_init()
- [Medium] i3c: master: svc: clocks left enabled in
svc_i3c_master_remove()
- [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
--
commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
Author: Junyi Zhao <junyi.zhao@amlogic.com>
i3c: master: svc: Add amlogic a9 support
This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
I3C controllers, introducing a vendor_init callback to register the AO
variant's internal clock divider chain and configure the fast_clk rate.
> diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> index e700392f0801f..0268fc713afdf 100644
> --- a/drivers/i3c/master/svc-i3c-master.c
> +++ b/drivers/i3c/master/svc-i3c-master.c
[ ... ]
> @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
[ ... ]
> + /*
> + * Keep the clk bulk state balanced for the following
> + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> + */
> + ret = clk_prepare_enable(fclk);
> + if (ret)
> + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> +
> + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> + GFP_KERNEL);
> + if (!new_clks) {
> + clk_disable_unprepare(fclk);
> + return -ENOMEM;
> + }
> +
> + memcpy(new_clks, master->clks,
> + master->num_clks * sizeof(*master->clks));
> + new_clks[master->num_clks].id = "fast_clk";
> + new_clks[master->num_clks].clk = fclk;
> + master->clks = new_clks;
> + master->num_clks++;
> +
> + return 0;
> +}
[Severity: Medium]
Can unbinding the driver leave fclk and the input clocks enabled when the
device is not runtime-suspended?
svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
reference not released by i3c_master_unregister()), runtime suspend does
not run during remove. When devres then unregisters ao->gate.hw,
ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
clk_unregister() triggers a kernel warning and leaks the prepare and
enable counts on the parent pclk.
> +
> +/**
> + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> + * @master: I3C master controller
> + *
> + * Register the AO variant's internal divider chain and round/set the fast
> + * clock rate for both variants.
> + *
> + * Return: 0 on success, or a negative error code.
> + */
> +static int aml_i3c_init(struct svc_i3c_master *master)
> +{
> + struct device *dev = master->dev;
> + struct clk *fclk;
> + long fin_freq;
> + int ret, i;
> +
> + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> + /*
> + * clk_bulk_get_all() preserves the DT order, and the
> + * binding restricts the AO variant to exactly one clock,
> + * which is always "pclk". So clks[0] is the input of the
> + * internal divider chain.
> + */
> + if (strcmp(master->clks[0].id, "pclk"))
> + return dev_err_probe(dev, -EINVAL,
> + "invalid AO input clock\n");
[Severity: Medium]
Can this trigger a NULL pointer dereference on master->clks or
master->clks[0].id?
When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
and leaves master->clks as NULL, which passes the master->num_clks < 0
check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
first and indexes master->clks[0] without checking that master->num_clks
is at least 1 (or equal to 1 as required by the AO binding). In addition,
of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
shorter than clocks, which would cause strcmp() here and in the fast_clk
loop below to dereference NULL.
> +
> + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> + if (ret)
> + return ret;
> + }
> +
> + for (i = 0; i < master->num_clks; i++) {
> + if (!strcmp(master->clks[i].id, "fast_clk"))
> + break;
> + }
> +
> + if (i == master->num_clks)
> + return dev_err_probe(dev, -EINVAL,
> + "can't get I3C fast clock\n");
> +
> + fclk = master->clks[i].clk;
> + if (IS_ERR(fclk))
> + return PTR_ERR(fclk);
[Severity: Low]
This isn't a bug, but this fast_clk lookup and validation duplicates the
identical loop in svc_i3c_master_probe() that runs right after
vendor_init() returns. Could a shared helper or resolving master->fclk
once avoid repeating the lookup?
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
--
linux-i3c mailing list
linux-i3c@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-i3c
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
2026-11-16 11:12 ` sashiko-bot
(?)
@ 2026-11-16 16:28 ` Frank Li
-1 siblings, 0 replies; 30+ messages in thread
From: Frank Li @ 2026-11-16 16:28 UTC (permalink / raw)
To: sashiko-reviews
Cc: Junyi Zhao, linux-i3c, devicetree, linux-amlogic, imx, robh,
conor+dt, Frank.Li, neil.armstrong, Alexandre Belloni
On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 3 potential
> issues to consider:
> - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
> aml_i3c_init()
> - [Medium] i3c: master: svc: clocks left enabled in
> svc_i3c_master_remove()
> - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
> --
>
> commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
> Author: Junyi Zhao <junyi.zhao@amlogic.com>
>
> i3c: master: svc: Add amlogic a9 support
>
> This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
> I3C controllers, introducing a vendor_init callback to register the AO
> variant's internal clock divider chain and configure the fast_clk rate.
>
> > diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> > index e700392f0801f..0268fc713afdf 100644
> > --- a/drivers/i3c/master/svc-i3c-master.c
> > +++ b/drivers/i3c/master/svc-i3c-master.c
>
> [ ... ]
>
> > @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
>
> [ ... ]
>
> > + /*
> > + * Keep the clk bulk state balanced for the following
> > + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> > + */
> > + ret = clk_prepare_enable(fclk);
> > + if (ret)
> > + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> > +
> > + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> > + GFP_KERNEL);
> > + if (!new_clks) {
> > + clk_disable_unprepare(fclk);
> > + return -ENOMEM;
> > + }
> > +
> > + memcpy(new_clks, master->clks,
> > + master->num_clks * sizeof(*master->clks));
> > + new_clks[master->num_clks].id = "fast_clk";
> > + new_clks[master->num_clks].clk = fclk;
> > + master->clks = new_clks;
> > + master->num_clks++;
> > +
> > + return 0;
> > +}
>
> [Severity: Medium]
> Can unbinding the driver leave fclk and the input clocks enabled when the
> device is not runtime-suspended?
>
> svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
> pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
> clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
> forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
> reference not released by i3c_master_unregister()), runtime suspend does
> not run during remove. When devres then unregisters ao->gate.hw,
> ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
> clk_unregister() triggers a kernel warning and leaks the prepare and
> enable counts on the parent pclk.
Suppose it is existing problem.
>
> > +
> > +/**
> > + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> > + * @master: I3C master controller
> > + *
> > + * Register the AO variant's internal divider chain and round/set the fast
> > + * clock rate for both variants.
> > + *
> > + * Return: 0 on success, or a negative error code.
> > + */
> > +static int aml_i3c_init(struct svc_i3c_master *master)
> > +{
> > + struct device *dev = master->dev;
> > + struct clk *fclk;
> > + long fin_freq;
> > + int ret, i;
> > +
> > + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> > + /*
> > + * clk_bulk_get_all() preserves the DT order, and the
> > + * binding restricts the AO variant to exactly one clock,
> > + * which is always "pclk". So clks[0] is the input of the
> > + * internal divider chain.
> > + */
> > + if (strcmp(master->clks[0].id, "pclk"))
> > + return dev_err_probe(dev, -EINVAL,
> > + "invalid AO input clock\n");
>
> [Severity: Medium]
> Can this trigger a NULL pointer dereference on master->clks or
> master->clks[0].id?
>
> When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
> and leaves master->clks as NULL, which passes the master->num_clks < 0
> check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
> ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
> first and indexes master->clks[0] without checking that master->num_clks
> is at least 1 (or equal to 1 as required by the AO binding). In addition,
> of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
> shorter than clocks, which would cause strcmp() here and in the fast_clk
> loop below to dereference NULL.
This one need check.
Frank
>
> > +
> > + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> > + if (ret)
> > + return ret;
> > + }
> > +
> > + for (i = 0; i < master->num_clks; i++) {
> > + if (!strcmp(master->clks[i].id, "fast_clk"))
> > + break;
> > + }
> > +
> > + if (i == master->num_clks)
> > + return dev_err_probe(dev, -EINVAL,
> > + "can't get I3C fast clock\n");
> > +
> > + fclk = master->clks[i].clk;
> > + if (IS_ERR(fclk))
> > + return PTR_ERR(fclk);
>
> [Severity: Low]
> This isn't a bug, but this fast_clk lookup and validation duplicates the
> identical loop in svc_i3c_master_probe() that runs right after
> vendor_init() returns. Could a shared helper or resolving master->fclk
> once avoid repeating the lookup?
>
> --
> Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-16 16:28 ` Frank Li
0 siblings, 0 replies; 30+ messages in thread
From: Frank Li @ 2026-11-16 16:28 UTC (permalink / raw)
To: sashiko-reviews
Cc: Junyi Zhao, linux-i3c, devicetree, linux-amlogic, imx, robh,
conor+dt, Frank.Li, neil.armstrong, Alexandre Belloni
On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 3 potential
> issues to consider:
> - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
> aml_i3c_init()
> - [Medium] i3c: master: svc: clocks left enabled in
> svc_i3c_master_remove()
> - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
> --
>
> commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
> Author: Junyi Zhao <junyi.zhao@amlogic.com>
>
> i3c: master: svc: Add amlogic a9 support
>
> This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
> I3C controllers, introducing a vendor_init callback to register the AO
> variant's internal clock divider chain and configure the fast_clk rate.
>
> > diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> > index e700392f0801f..0268fc713afdf 100644
> > --- a/drivers/i3c/master/svc-i3c-master.c
> > +++ b/drivers/i3c/master/svc-i3c-master.c
>
> [ ... ]
>
> > @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
>
> [ ... ]
>
> > + /*
> > + * Keep the clk bulk state balanced for the following
> > + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> > + */
> > + ret = clk_prepare_enable(fclk);
> > + if (ret)
> > + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> > +
> > + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> > + GFP_KERNEL);
> > + if (!new_clks) {
> > + clk_disable_unprepare(fclk);
> > + return -ENOMEM;
> > + }
> > +
> > + memcpy(new_clks, master->clks,
> > + master->num_clks * sizeof(*master->clks));
> > + new_clks[master->num_clks].id = "fast_clk";
> > + new_clks[master->num_clks].clk = fclk;
> > + master->clks = new_clks;
> > + master->num_clks++;
> > +
> > + return 0;
> > +}
>
> [Severity: Medium]
> Can unbinding the driver leave fclk and the input clocks enabled when the
> device is not runtime-suspended?
>
> svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
> pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
> clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
> forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
> reference not released by i3c_master_unregister()), runtime suspend does
> not run during remove. When devres then unregisters ao->gate.hw,
> ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
> clk_unregister() triggers a kernel warning and leaks the prepare and
> enable counts on the parent pclk.
Suppose it is existing problem.
>
> > +
> > +/**
> > + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> > + * @master: I3C master controller
> > + *
> > + * Register the AO variant's internal divider chain and round/set the fast
> > + * clock rate for both variants.
> > + *
> > + * Return: 0 on success, or a negative error code.
> > + */
> > +static int aml_i3c_init(struct svc_i3c_master *master)
> > +{
> > + struct device *dev = master->dev;
> > + struct clk *fclk;
> > + long fin_freq;
> > + int ret, i;
> > +
> > + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> > + /*
> > + * clk_bulk_get_all() preserves the DT order, and the
> > + * binding restricts the AO variant to exactly one clock,
> > + * which is always "pclk". So clks[0] is the input of the
> > + * internal divider chain.
> > + */
> > + if (strcmp(master->clks[0].id, "pclk"))
> > + return dev_err_probe(dev, -EINVAL,
> > + "invalid AO input clock\n");
>
> [Severity: Medium]
> Can this trigger a NULL pointer dereference on master->clks or
> master->clks[0].id?
>
> When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
> and leaves master->clks as NULL, which passes the master->num_clks < 0
> check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
> ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
> first and indexes master->clks[0] without checking that master->num_clks
> is at least 1 (or equal to 1 as required by the AO binding). In addition,
> of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
> shorter than clocks, which would cause strcmp() here and in the fast_clk
> loop below to dereference NULL.
This one need check.
Frank
>
> > +
> > + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> > + if (ret)
> > + return ret;
> > + }
> > +
> > + for (i = 0; i < master->num_clks; i++) {
> > + if (!strcmp(master->clks[i].id, "fast_clk"))
> > + break;
> > + }
> > +
> > + if (i == master->num_clks)
> > + return dev_err_probe(dev, -EINVAL,
> > + "can't get I3C fast clock\n");
> > +
> > + fclk = master->clks[i].clk;
> > + if (IS_ERR(fclk))
> > + return PTR_ERR(fclk);
>
> [Severity: Low]
> This isn't a bug, but this fast_clk lookup and validation duplicates the
> identical loop in svc_i3c_master_probe() that runs right after
> vendor_init() returns. Could a shared helper or resolving master->fclk
> once avoid repeating the lookup?
>
> --
> Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
--
linux-i3c mailing list
linux-i3c@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-i3c
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-16 16:28 ` Frank Li
0 siblings, 0 replies; 30+ messages in thread
From: Frank Li @ 2026-11-16 16:28 UTC (permalink / raw)
To: sashiko-reviews
Cc: Junyi Zhao, linux-i3c, devicetree, linux-amlogic, imx, robh,
conor+dt, Frank.Li, neil.armstrong, Alexandre Belloni
On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 3 potential
> issues to consider:
> - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
> aml_i3c_init()
> - [Medium] i3c: master: svc: clocks left enabled in
> svc_i3c_master_remove()
> - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
> --
>
> commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
> Author: Junyi Zhao <junyi.zhao@amlogic.com>
>
> i3c: master: svc: Add amlogic a9 support
>
> This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
> I3C controllers, introducing a vendor_init callback to register the AO
> variant's internal clock divider chain and configure the fast_clk rate.
>
> > diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> > index e700392f0801f..0268fc713afdf 100644
> > --- a/drivers/i3c/master/svc-i3c-master.c
> > +++ b/drivers/i3c/master/svc-i3c-master.c
>
> [ ... ]
>
> > @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
>
> [ ... ]
>
> > + /*
> > + * Keep the clk bulk state balanced for the following
> > + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> > + */
> > + ret = clk_prepare_enable(fclk);
> > + if (ret)
> > + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> > +
> > + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> > + GFP_KERNEL);
> > + if (!new_clks) {
> > + clk_disable_unprepare(fclk);
> > + return -ENOMEM;
> > + }
> > +
> > + memcpy(new_clks, master->clks,
> > + master->num_clks * sizeof(*master->clks));
> > + new_clks[master->num_clks].id = "fast_clk";
> > + new_clks[master->num_clks].clk = fclk;
> > + master->clks = new_clks;
> > + master->num_clks++;
> > +
> > + return 0;
> > +}
>
> [Severity: Medium]
> Can unbinding the driver leave fclk and the input clocks enabled when the
> device is not runtime-suspended?
>
> svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
> pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
> clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
> forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
> reference not released by i3c_master_unregister()), runtime suspend does
> not run during remove. When devres then unregisters ao->gate.hw,
> ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
> clk_unregister() triggers a kernel warning and leaks the prepare and
> enable counts on the parent pclk.
Suppose it is existing problem.
>
> > +
> > +/**
> > + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> > + * @master: I3C master controller
> > + *
> > + * Register the AO variant's internal divider chain and round/set the fast
> > + * clock rate for both variants.
> > + *
> > + * Return: 0 on success, or a negative error code.
> > + */
> > +static int aml_i3c_init(struct svc_i3c_master *master)
> > +{
> > + struct device *dev = master->dev;
> > + struct clk *fclk;
> > + long fin_freq;
> > + int ret, i;
> > +
> > + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> > + /*
> > + * clk_bulk_get_all() preserves the DT order, and the
> > + * binding restricts the AO variant to exactly one clock,
> > + * which is always "pclk". So clks[0] is the input of the
> > + * internal divider chain.
> > + */
> > + if (strcmp(master->clks[0].id, "pclk"))
> > + return dev_err_probe(dev, -EINVAL,
> > + "invalid AO input clock\n");
>
> [Severity: Medium]
> Can this trigger a NULL pointer dereference on master->clks or
> master->clks[0].id?
>
> When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
> and leaves master->clks as NULL, which passes the master->num_clks < 0
> check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
> ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
> first and indexes master->clks[0] without checking that master->num_clks
> is at least 1 (or equal to 1 as required by the AO binding). In addition,
> of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
> shorter than clocks, which would cause strcmp() here and in the fast_clk
> loop below to dereference NULL.
This one need check.
Frank
>
> > +
> > + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> > + if (ret)
> > + return ret;
> > + }
> > +
> > + for (i = 0; i < master->num_clks; i++) {
> > + if (!strcmp(master->clks[i].id, "fast_clk"))
> > + break;
> > + }
> > +
> > + if (i == master->num_clks)
> > + return dev_err_probe(dev, -EINVAL,
> > + "can't get I3C fast clock\n");
> > +
> > + fclk = master->clks[i].clk;
> > + if (IS_ERR(fclk))
> > + return PTR_ERR(fclk);
>
> [Severity: Low]
> This isn't a bug, but this fast_clk lookup and validation duplicates the
> identical loop in svc_i3c_master_probe() that runs right after
> vendor_init() returns. Could a shared helper or resolving master->fclk
> once avoid repeating the lookup?
>
> --
> Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
_______________________________________________
linux-amlogic mailing list
linux-amlogic@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-amlogic
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
2026-11-16 16:28 ` Frank Li
(?)
@ 2026-11-17 5:45 ` Junyi Zhao
-1 siblings, 0 replies; 30+ messages in thread
From: Junyi Zhao @ 2026-11-17 5:45 UTC (permalink / raw)
To: Frank Li, sashiko-reviews
Cc: linux-i3c, devicetree, linux-amlogic, imx, robh, conor+dt,
Frank.Li, neil.armstrong, Alexandre Belloni
Hi frank,
thks for the review.
On 17/11/2026 00:28, Frank Li wrote:
> [ EXTERNAL EMAIL ]
>
> On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
>> Thank you for your contribution! Sashiko AI review found 3 potential
>> issues to consider:
>> - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
>> aml_i3c_init()
>> - [Medium] i3c: master: svc: clocks left enabled in
>> svc_i3c_master_remove()
>> - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
>> --
>>
>> commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
>> Author: Junyi Zhao <junyi.zhao@amlogic.com>
>>
>> i3c: master: svc: Add amlogic a9 support
>>
>> This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
>> I3C controllers, introducing a vendor_init callback to register the AO
>> variant's internal clock divider chain and configure the fast_clk rate.
>>
>>> diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
>>> index e700392f0801f..0268fc713afdf 100644
>>> --- a/drivers/i3c/master/svc-i3c-master.c
>>> +++ b/drivers/i3c/master/svc-i3c-master.c
>>
>> [ ... ]
>>
>>> @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
>>
>> [ ... ]
>>
>>> + /*
>>> + * Keep the clk bulk state balanced for the following
>>> + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
>>> + */
>>> + ret = clk_prepare_enable(fclk);
>>> + if (ret)
>>> + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
>>> +
>>> + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
>>> + GFP_KERNEL);
>>> + if (!new_clks) {
>>> + clk_disable_unprepare(fclk);
>>> + return -ENOMEM;
>>> + }
>>> +
>>> + memcpy(new_clks, master->clks,
>>> + master->num_clks * sizeof(*master->clks));
>>> + new_clks[master->num_clks].id = "fast_clk";
>>> + new_clks[master->num_clks].clk = fclk;
>>> + master->clks = new_clks;
>>> + master->num_clks++;
>>> +
>>> + return 0;
>>> +}
>>
>> [Severity: Medium]
>> Can unbinding the driver leave fclk and the input clocks enabled when the
>> device is not runtime-suspended?
>>
>> svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
>> pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
>> clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
>> forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
>> reference not released by i3c_master_unregister()), runtime suspend does
>> not run during remove. When devres then unregisters ao->gate.hw,
>> ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
>> clk_unregister() triggers a kernel warning and leaks the prepare and
>> enable counts on the parent pclk.
>
> Suppose it is existing problem.
>
Ok, not introduced by this patch.
Address it in a separate patch follow this series or create new change
series?
>>
>>> +
>>> +/**
>>> + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
>>> + * @master: I3C master controller
>>> + *
>>> + * Register the AO variant's internal divider chain and round/set the fast
>>> + * clock rate for both variants.
>>> + *
>>> + * Return: 0 on success, or a negative error code.
>>> + */
>>> +static int aml_i3c_init(struct svc_i3c_master *master)
>>> +{
>>> + struct device *dev = master->dev;
>>> + struct clk *fclk;
>>> + long fin_freq;
>>> + int ret, i;
>>> +
>>> + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
>>> + /*
>>> + * clk_bulk_get_all() preserves the DT order, and the
>>> + * binding restricts the AO variant to exactly one clock,
>>> + * which is always "pclk". So clks[0] is the input of the
>>> + * internal divider chain.
>>> + */
>>> + if (strcmp(master->clks[0].id, "pclk"))
>>> + return dev_err_probe(dev, -EINVAL,
>>> + "invalid AO input clock\n");
>>
>> [Severity: Medium]
>> Can this trigger a NULL pointer dereference on master->clks or
>> master->clks[0].id?
>>
>> When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
>> and leaves master->clks as NULL, which passes the master->num_clks < 0
>> check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
>> ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
>> first and indexes master->clks[0] without checking that master->num_clks
>> is at least 1 (or equal to 1 as required by the AO binding). In addition,
>> of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
>> shorter than clocks, which would cause strcmp() here and in the fast_clk
>> loop below to dereference NULL.
>
> This one need check.
>
will add check.
> Frank
>
>>
>>> +
>>> + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
>>> + if (ret)
>>> + return ret;
>>> + }
>>> +
>>> + for (i = 0; i < master->num_clks; i++) {
>>> + if (!strcmp(master->clks[i].id, "fast_clk"))
>>> + break;
>>> + }
>>> +
>>> + if (i == master->num_clks)
>>> + return dev_err_probe(dev, -EINVAL,
>>> + "can't get I3C fast clock\n");
>>> +
>>> + fclk = master->clks[i].clk;
>>> + if (IS_ERR(fclk))
>>> + return PTR_ERR(fclk);
>>
>> [Severity: Low]
>> This isn't a bug, but this fast_clk lookup and validation duplicates the
>> identical loop in svc_i3c_master_probe() that runs right after
>> vendor_init() returns. Could a shared helper or resolving master->fclk
>> once avoid repeating the lookup?
>>
Will try.
And guards against a NULL clks[i].id, and reused by both probe() and
aml_i3c_init().
>> --
>> Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
_______________________________________________
linux-amlogic mailing list
linux-amlogic@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-amlogic
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-17 5:45 ` Junyi Zhao
0 siblings, 0 replies; 30+ messages in thread
From: Junyi Zhao @ 2026-11-17 5:45 UTC (permalink / raw)
To: Frank Li, sashiko-reviews
Cc: linux-i3c, devicetree, linux-amlogic, imx, robh, conor+dt,
Frank.Li, neil.armstrong, Alexandre Belloni
Hi frank,
thks for the review.
On 17/11/2026 00:28, Frank Li wrote:
> [ EXTERNAL EMAIL ]
>
> On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
>> Thank you for your contribution! Sashiko AI review found 3 potential
>> issues to consider:
>> - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
>> aml_i3c_init()
>> - [Medium] i3c: master: svc: clocks left enabled in
>> svc_i3c_master_remove()
>> - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
>> --
>>
>> commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
>> Author: Junyi Zhao <junyi.zhao@amlogic.com>
>>
>> i3c: master: svc: Add amlogic a9 support
>>
>> This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
>> I3C controllers, introducing a vendor_init callback to register the AO
>> variant's internal clock divider chain and configure the fast_clk rate.
>>
>>> diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
>>> index e700392f0801f..0268fc713afdf 100644
>>> --- a/drivers/i3c/master/svc-i3c-master.c
>>> +++ b/drivers/i3c/master/svc-i3c-master.c
>>
>> [ ... ]
>>
>>> @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
>>
>> [ ... ]
>>
>>> + /*
>>> + * Keep the clk bulk state balanced for the following
>>> + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
>>> + */
>>> + ret = clk_prepare_enable(fclk);
>>> + if (ret)
>>> + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
>>> +
>>> + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
>>> + GFP_KERNEL);
>>> + if (!new_clks) {
>>> + clk_disable_unprepare(fclk);
>>> + return -ENOMEM;
>>> + }
>>> +
>>> + memcpy(new_clks, master->clks,
>>> + master->num_clks * sizeof(*master->clks));
>>> + new_clks[master->num_clks].id = "fast_clk";
>>> + new_clks[master->num_clks].clk = fclk;
>>> + master->clks = new_clks;
>>> + master->num_clks++;
>>> +
>>> + return 0;
>>> +}
>>
>> [Severity: Medium]
>> Can unbinding the driver leave fclk and the input clocks enabled when the
>> device is not runtime-suspended?
>>
>> svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
>> pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
>> clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
>> forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
>> reference not released by i3c_master_unregister()), runtime suspend does
>> not run during remove. When devres then unregisters ao->gate.hw,
>> ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
>> clk_unregister() triggers a kernel warning and leaks the prepare and
>> enable counts on the parent pclk.
>
> Suppose it is existing problem.
>
Ok, not introduced by this patch.
Address it in a separate patch follow this series or create new change
series?
>>
>>> +
>>> +/**
>>> + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
>>> + * @master: I3C master controller
>>> + *
>>> + * Register the AO variant's internal divider chain and round/set the fast
>>> + * clock rate for both variants.
>>> + *
>>> + * Return: 0 on success, or a negative error code.
>>> + */
>>> +static int aml_i3c_init(struct svc_i3c_master *master)
>>> +{
>>> + struct device *dev = master->dev;
>>> + struct clk *fclk;
>>> + long fin_freq;
>>> + int ret, i;
>>> +
>>> + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
>>> + /*
>>> + * clk_bulk_get_all() preserves the DT order, and the
>>> + * binding restricts the AO variant to exactly one clock,
>>> + * which is always "pclk". So clks[0] is the input of the
>>> + * internal divider chain.
>>> + */
>>> + if (strcmp(master->clks[0].id, "pclk"))
>>> + return dev_err_probe(dev, -EINVAL,
>>> + "invalid AO input clock\n");
>>
>> [Severity: Medium]
>> Can this trigger a NULL pointer dereference on master->clks or
>> master->clks[0].id?
>>
>> When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
>> and leaves master->clks as NULL, which passes the master->num_clks < 0
>> check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
>> ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
>> first and indexes master->clks[0] without checking that master->num_clks
>> is at least 1 (or equal to 1 as required by the AO binding). In addition,
>> of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
>> shorter than clocks, which would cause strcmp() here and in the fast_clk
>> loop below to dereference NULL.
>
> This one need check.
>
will add check.
> Frank
>
>>
>>> +
>>> + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
>>> + if (ret)
>>> + return ret;
>>> + }
>>> +
>>> + for (i = 0; i < master->num_clks; i++) {
>>> + if (!strcmp(master->clks[i].id, "fast_clk"))
>>> + break;
>>> + }
>>> +
>>> + if (i == master->num_clks)
>>> + return dev_err_probe(dev, -EINVAL,
>>> + "can't get I3C fast clock\n");
>>> +
>>> + fclk = master->clks[i].clk;
>>> + if (IS_ERR(fclk))
>>> + return PTR_ERR(fclk);
>>
>> [Severity: Low]
>> This isn't a bug, but this fast_clk lookup and validation duplicates the
>> identical loop in svc_i3c_master_probe() that runs right after
>> vendor_init() returns. Could a shared helper or resolving master->fclk
>> once avoid repeating the lookup?
>>
Will try.
And guards against a NULL clks[i].id, and reused by both probe() and
aml_i3c_init().
>> --
>> Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
--
linux-i3c mailing list
linux-i3c@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-i3c
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-17 5:45 ` Junyi Zhao
0 siblings, 0 replies; 30+ messages in thread
From: Junyi Zhao @ 2026-11-17 5:45 UTC (permalink / raw)
To: Frank Li, sashiko-reviews
Cc: linux-i3c, devicetree, linux-amlogic, imx, robh, conor+dt,
Frank.Li, neil.armstrong, Alexandre Belloni
Hi frank,
thks for the review.
On 17/11/2026 00:28, Frank Li wrote:
> [ EXTERNAL EMAIL ]
>
> On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
>> Thank you for your contribution! Sashiko AI review found 3 potential
>> issues to consider:
>> - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
>> aml_i3c_init()
>> - [Medium] i3c: master: svc: clocks left enabled in
>> svc_i3c_master_remove()
>> - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
>> --
>>
>> commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
>> Author: Junyi Zhao <junyi.zhao@amlogic.com>
>>
>> i3c: master: svc: Add amlogic a9 support
>>
>> This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
>> I3C controllers, introducing a vendor_init callback to register the AO
>> variant's internal clock divider chain and configure the fast_clk rate.
>>
>>> diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
>>> index e700392f0801f..0268fc713afdf 100644
>>> --- a/drivers/i3c/master/svc-i3c-master.c
>>> +++ b/drivers/i3c/master/svc-i3c-master.c
>>
>> [ ... ]
>>
>>> @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
>>
>> [ ... ]
>>
>>> + /*
>>> + * Keep the clk bulk state balanced for the following
>>> + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
>>> + */
>>> + ret = clk_prepare_enable(fclk);
>>> + if (ret)
>>> + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
>>> +
>>> + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
>>> + GFP_KERNEL);
>>> + if (!new_clks) {
>>> + clk_disable_unprepare(fclk);
>>> + return -ENOMEM;
>>> + }
>>> +
>>> + memcpy(new_clks, master->clks,
>>> + master->num_clks * sizeof(*master->clks));
>>> + new_clks[master->num_clks].id = "fast_clk";
>>> + new_clks[master->num_clks].clk = fclk;
>>> + master->clks = new_clks;
>>> + master->num_clks++;
>>> +
>>> + return 0;
>>> +}
>>
>> [Severity: Medium]
>> Can unbinding the driver leave fclk and the input clocks enabled when the
>> device is not runtime-suspended?
>>
>> svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
>> pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
>> clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
>> forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
>> reference not released by i3c_master_unregister()), runtime suspend does
>> not run during remove. When devres then unregisters ao->gate.hw,
>> ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
>> clk_unregister() triggers a kernel warning and leaks the prepare and
>> enable counts on the parent pclk.
>
> Suppose it is existing problem.
>
Ok, not introduced by this patch.
Address it in a separate patch follow this series or create new change
series?
>>
>>> +
>>> +/**
>>> + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
>>> + * @master: I3C master controller
>>> + *
>>> + * Register the AO variant's internal divider chain and round/set the fast
>>> + * clock rate for both variants.
>>> + *
>>> + * Return: 0 on success, or a negative error code.
>>> + */
>>> +static int aml_i3c_init(struct svc_i3c_master *master)
>>> +{
>>> + struct device *dev = master->dev;
>>> + struct clk *fclk;
>>> + long fin_freq;
>>> + int ret, i;
>>> +
>>> + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
>>> + /*
>>> + * clk_bulk_get_all() preserves the DT order, and the
>>> + * binding restricts the AO variant to exactly one clock,
>>> + * which is always "pclk". So clks[0] is the input of the
>>> + * internal divider chain.
>>> + */
>>> + if (strcmp(master->clks[0].id, "pclk"))
>>> + return dev_err_probe(dev, -EINVAL,
>>> + "invalid AO input clock\n");
>>
>> [Severity: Medium]
>> Can this trigger a NULL pointer dereference on master->clks or
>> master->clks[0].id?
>>
>> When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
>> and leaves master->clks as NULL, which passes the master->num_clks < 0
>> check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
>> ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
>> first and indexes master->clks[0] without checking that master->num_clks
>> is at least 1 (or equal to 1 as required by the AO binding). In addition,
>> of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
>> shorter than clocks, which would cause strcmp() here and in the fast_clk
>> loop below to dereference NULL.
>
> This one need check.
>
will add check.
> Frank
>
>>
>>> +
>>> + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
>>> + if (ret)
>>> + return ret;
>>> + }
>>> +
>>> + for (i = 0; i < master->num_clks; i++) {
>>> + if (!strcmp(master->clks[i].id, "fast_clk"))
>>> + break;
>>> + }
>>> +
>>> + if (i == master->num_clks)
>>> + return dev_err_probe(dev, -EINVAL,
>>> + "can't get I3C fast clock\n");
>>> +
>>> + fclk = master->clks[i].clk;
>>> + if (IS_ERR(fclk))
>>> + return PTR_ERR(fclk);
>>
>> [Severity: Low]
>> This isn't a bug, but this fast_clk lookup and validation duplicates the
>> identical loop in svc_i3c_master_probe() that runs right after
>> vendor_init() returns. Could a shared helper or resolving master->fclk
>> once avoid repeating the lookup?
>>
Will try.
And guards against a NULL clks[i].id, and reused by both probe() and
aml_i3c_init().
>> --
>> Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
2026-11-17 5:45 ` Junyi Zhao
(?)
@ 2026-11-17 17:41 ` Frank Li
-1 siblings, 0 replies; 30+ messages in thread
From: Frank Li @ 2026-11-17 17:41 UTC (permalink / raw)
To: Junyi Zhao
Cc: sashiko-reviews, linux-i3c, devicetree, linux-amlogic, imx, robh,
conor+dt, Frank.Li, neil.armstrong, Alexandre Belloni
On Tue, Nov 17, 2026 at 01:45:41PM +0800, Junyi Zhao wrote:
> Hi frank,
> thks for the review.
>
> On 17/11/2026 00:28, Frank Li wrote:
> > [ EXTERNAL EMAIL ]
> >
> > On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
> > > Thank you for your contribution! Sashiko AI review found 3 potential
> > > issues to consider:
> > > - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
> > > aml_i3c_init()
> > > - [Medium] i3c: master: svc: clocks left enabled in
> > > svc_i3c_master_remove()
> > > - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
> > > --
> > >
> > > commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
> > > Author: Junyi Zhao <junyi.zhao@amlogic.com>
> > >
> > > i3c: master: svc: Add amlogic a9 support
> > >
> > > This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
> > > I3C controllers, introducing a vendor_init callback to register the AO
> > > variant's internal clock divider chain and configure the fast_clk rate.
> > >
> > > > diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> > > > index e700392f0801f..0268fc713afdf 100644
> > > > --- a/drivers/i3c/master/svc-i3c-master.c
> > > > +++ b/drivers/i3c/master/svc-i3c-master.c
> > >
> > > [ ... ]
> > >
> > > > @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
> > >
> > > [ ... ]
> > >
> > > > + /*
> > > > + * Keep the clk bulk state balanced for the following
> > > > + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> > > > + */
> > > > + ret = clk_prepare_enable(fclk);
> > > > + if (ret)
> > > > + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> > > > +
> > > > + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> > > > + GFP_KERNEL);
> > > > + if (!new_clks) {
> > > > + clk_disable_unprepare(fclk);
> > > > + return -ENOMEM;
> > > > + }
> > > > +
> > > > + memcpy(new_clks, master->clks,
> > > > + master->num_clks * sizeof(*master->clks));
> > > > + new_clks[master->num_clks].id = "fast_clk";
> > > > + new_clks[master->num_clks].clk = fclk;
> > > > + master->clks = new_clks;
> > > > + master->num_clks++;
> > > > +
> > > > + return 0;
> > > > +}
> > >
> > > [Severity: Medium]
> > > Can unbinding the driver leave fclk and the input clocks enabled when the
> > > device is not runtime-suspended?
> > >
> > > svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
> > > pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
> > > clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
> > > forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
> > > reference not released by i3c_master_unregister()), runtime suspend does
> > > not run during remove. When devres then unregisters ao->gate.hw,
> > > ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
> > > clk_unregister() triggers a kernel warning and leaks the prepare and
> > > enable counts on the parent pclk.
> >
> > Suppose it is existing problem.
> >
> Ok, not introduced by this patch.
>
> Address it in a separate patch follow this series or create new change
> series?
New series. Many codes can be removed because framework already support
call pm_runtime_get() before tranfer data after simple set a flag.
Frank
>
> > >
> > > > +
> > > > +/**
> > > > + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> > > > + * @master: I3C master controller
> > > > + *
> > > > + * Register the AO variant's internal divider chain and round/set the fast
> > > > + * clock rate for both variants.
> > > > + *
> > > > + * Return: 0 on success, or a negative error code.
> > > > + */
> > > > +static int aml_i3c_init(struct svc_i3c_master *master)
> > > > +{
> > > > + struct device *dev = master->dev;
> > > > + struct clk *fclk;
> > > > + long fin_freq;
> > > > + int ret, i;
> > > > +
> > > > + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> > > > + /*
> > > > + * clk_bulk_get_all() preserves the DT order, and the
> > > > + * binding restricts the AO variant to exactly one clock,
> > > > + * which is always "pclk". So clks[0] is the input of the
> > > > + * internal divider chain.
> > > > + */
> > > > + if (strcmp(master->clks[0].id, "pclk"))
> > > > + return dev_err_probe(dev, -EINVAL,
> > > > + "invalid AO input clock\n");
> > >
> > > [Severity: Medium]
> > > Can this trigger a NULL pointer dereference on master->clks or
> > > master->clks[0].id?
> > >
> > > When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
> > > and leaves master->clks as NULL, which passes the master->num_clks < 0
> > > check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
> > > ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
> > > first and indexes master->clks[0] without checking that master->num_clks
> > > is at least 1 (or equal to 1 as required by the AO binding). In addition,
> > > of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
> > > shorter than clocks, which would cause strcmp() here and in the fast_clk
> > > loop below to dereference NULL.
> >
> > This one need check.
> >
>
> will add check.
>
> > Frank
> >
> > >
> > > > +
> > > > + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> > > > + if (ret)
> > > > + return ret;
> > > > + }
> > > > +
> > > > + for (i = 0; i < master->num_clks; i++) {
> > > > + if (!strcmp(master->clks[i].id, "fast_clk"))
> > > > + break;
> > > > + }
> > > > +
> > > > + if (i == master->num_clks)
> > > > + return dev_err_probe(dev, -EINVAL,
> > > > + "can't get I3C fast clock\n");
> > > > +
> > > > + fclk = master->clks[i].clk;
> > > > + if (IS_ERR(fclk))
> > > > + return PTR_ERR(fclk);
> > >
> > > [Severity: Low]
> > > This isn't a bug, but this fast_clk lookup and validation duplicates the
> > > identical loop in svc_i3c_master_probe() that runs right after
> > > vendor_init() returns. Could a shared helper or resolving master->fclk
> > > once avoid repeating the lookup?
> > >
> Will try.
> And guards against a NULL clks[i].id, and reused by both probe() and
> aml_i3c_init().
>
> > > --
> > > Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
>
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-17 17:41 ` Frank Li
0 siblings, 0 replies; 30+ messages in thread
From: Frank Li @ 2026-11-17 17:41 UTC (permalink / raw)
To: Junyi Zhao
Cc: sashiko-reviews, linux-i3c, devicetree, linux-amlogic, imx, robh,
conor+dt, Frank.Li, neil.armstrong, Alexandre Belloni
On Tue, Nov 17, 2026 at 01:45:41PM +0800, Junyi Zhao wrote:
> Hi frank,
> thks for the review.
>
> On 17/11/2026 00:28, Frank Li wrote:
> > [ EXTERNAL EMAIL ]
> >
> > On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
> > > Thank you for your contribution! Sashiko AI review found 3 potential
> > > issues to consider:
> > > - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
> > > aml_i3c_init()
> > > - [Medium] i3c: master: svc: clocks left enabled in
> > > svc_i3c_master_remove()
> > > - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
> > > --
> > >
> > > commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
> > > Author: Junyi Zhao <junyi.zhao@amlogic.com>
> > >
> > > i3c: master: svc: Add amlogic a9 support
> > >
> > > This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
> > > I3C controllers, introducing a vendor_init callback to register the AO
> > > variant's internal clock divider chain and configure the fast_clk rate.
> > >
> > > > diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> > > > index e700392f0801f..0268fc713afdf 100644
> > > > --- a/drivers/i3c/master/svc-i3c-master.c
> > > > +++ b/drivers/i3c/master/svc-i3c-master.c
> > >
> > > [ ... ]
> > >
> > > > @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
> > >
> > > [ ... ]
> > >
> > > > + /*
> > > > + * Keep the clk bulk state balanced for the following
> > > > + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> > > > + */
> > > > + ret = clk_prepare_enable(fclk);
> > > > + if (ret)
> > > > + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> > > > +
> > > > + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> > > > + GFP_KERNEL);
> > > > + if (!new_clks) {
> > > > + clk_disable_unprepare(fclk);
> > > > + return -ENOMEM;
> > > > + }
> > > > +
> > > > + memcpy(new_clks, master->clks,
> > > > + master->num_clks * sizeof(*master->clks));
> > > > + new_clks[master->num_clks].id = "fast_clk";
> > > > + new_clks[master->num_clks].clk = fclk;
> > > > + master->clks = new_clks;
> > > > + master->num_clks++;
> > > > +
> > > > + return 0;
> > > > +}
> > >
> > > [Severity: Medium]
> > > Can unbinding the driver leave fclk and the input clocks enabled when the
> > > device is not runtime-suspended?
> > >
> > > svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
> > > pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
> > > clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
> > > forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
> > > reference not released by i3c_master_unregister()), runtime suspend does
> > > not run during remove. When devres then unregisters ao->gate.hw,
> > > ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
> > > clk_unregister() triggers a kernel warning and leaks the prepare and
> > > enable counts on the parent pclk.
> >
> > Suppose it is existing problem.
> >
> Ok, not introduced by this patch.
>
> Address it in a separate patch follow this series or create new change
> series?
New series. Many codes can be removed because framework already support
call pm_runtime_get() before tranfer data after simple set a flag.
Frank
>
> > >
> > > > +
> > > > +/**
> > > > + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> > > > + * @master: I3C master controller
> > > > + *
> > > > + * Register the AO variant's internal divider chain and round/set the fast
> > > > + * clock rate for both variants.
> > > > + *
> > > > + * Return: 0 on success, or a negative error code.
> > > > + */
> > > > +static int aml_i3c_init(struct svc_i3c_master *master)
> > > > +{
> > > > + struct device *dev = master->dev;
> > > > + struct clk *fclk;
> > > > + long fin_freq;
> > > > + int ret, i;
> > > > +
> > > > + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> > > > + /*
> > > > + * clk_bulk_get_all() preserves the DT order, and the
> > > > + * binding restricts the AO variant to exactly one clock,
> > > > + * which is always "pclk". So clks[0] is the input of the
> > > > + * internal divider chain.
> > > > + */
> > > > + if (strcmp(master->clks[0].id, "pclk"))
> > > > + return dev_err_probe(dev, -EINVAL,
> > > > + "invalid AO input clock\n");
> > >
> > > [Severity: Medium]
> > > Can this trigger a NULL pointer dereference on master->clks or
> > > master->clks[0].id?
> > >
> > > When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
> > > and leaves master->clks as NULL, which passes the master->num_clks < 0
> > > check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
> > > ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
> > > first and indexes master->clks[0] without checking that master->num_clks
> > > is at least 1 (or equal to 1 as required by the AO binding). In addition,
> > > of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
> > > shorter than clocks, which would cause strcmp() here and in the fast_clk
> > > loop below to dereference NULL.
> >
> > This one need check.
> >
>
> will add check.
>
> > Frank
> >
> > >
> > > > +
> > > > + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> > > > + if (ret)
> > > > + return ret;
> > > > + }
> > > > +
> > > > + for (i = 0; i < master->num_clks; i++) {
> > > > + if (!strcmp(master->clks[i].id, "fast_clk"))
> > > > + break;
> > > > + }
> > > > +
> > > > + if (i == master->num_clks)
> > > > + return dev_err_probe(dev, -EINVAL,
> > > > + "can't get I3C fast clock\n");
> > > > +
> > > > + fclk = master->clks[i].clk;
> > > > + if (IS_ERR(fclk))
> > > > + return PTR_ERR(fclk);
> > >
> > > [Severity: Low]
> > > This isn't a bug, but this fast_clk lookup and validation duplicates the
> > > identical loop in svc_i3c_master_probe() that runs right after
> > > vendor_init() returns. Could a shared helper or resolving master->fclk
> > > once avoid repeating the lookup?
> > >
> Will try.
> And guards against a NULL clks[i].id, and reused by both probe() and
> aml_i3c_init().
>
> > > --
> > > Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
>
_______________________________________________
linux-amlogic mailing list
linux-amlogic@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-amlogic
^ permalink raw reply [flat|nested] 30+ messages in thread* Re: [PATCH v4 2/2] i3c: master: svc: Add amlogic a9 support
@ 2026-11-17 17:41 ` Frank Li
0 siblings, 0 replies; 30+ messages in thread
From: Frank Li @ 2026-11-17 17:41 UTC (permalink / raw)
To: Junyi Zhao
Cc: sashiko-reviews, linux-i3c, devicetree, linux-amlogic, imx, robh,
conor+dt, Frank.Li, neil.armstrong, Alexandre Belloni
On Tue, Nov 17, 2026 at 01:45:41PM +0800, Junyi Zhao wrote:
> Hi frank,
> thks for the review.
>
> On 17/11/2026 00:28, Frank Li wrote:
> > [ EXTERNAL EMAIL ]
> >
> > On Mon, Nov 16, 2026 at 11:12:37AM +0000, sashiko-bot@kernel.org wrote:
> > > Thank you for your contribution! Sashiko AI review found 3 potential
> > > issues to consider:
> > > - [Medium] i3c: master: svc: NULL dereference of clks[0].id in
> > > aml_i3c_init()
> > > - [Medium] i3c: master: svc: clocks left enabled in
> > > svc_i3c_master_remove()
> > > - [Low] i3c: master: svc: duplicate fast_clk lookup in aml_i3c_init()
> > > --
> > >
> > > commit 260aeb5a4d5c58d3bea32674062960231e4d4d2c
> > > Author: Junyi Zhao <junyi.zhao@amlogic.com>
> > >
> > > i3c: master: svc: Add amlogic a9 support
> > >
> > > This commit adds support for the Amlogic A9 EE-domain and AO-domain Silvaco
> > > I3C controllers, introducing a vendor_init callback to register the AO
> > > variant's internal clock divider chain and configure the fast_clk rate.
> > >
> > > > diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> > > > index e700392f0801f..0268fc713afdf 100644
> > > > --- a/drivers/i3c/master/svc-i3c-master.c
> > > > +++ b/drivers/i3c/master/svc-i3c-master.c
> > >
> > > [ ... ]
> > >
> > > > @@ -2163,9 +2209,215 @@ static const struct svc_i3c_drvdata npcm845_drvdata = {
> > >
> > > [ ... ]
> > >
> > > > + /*
> > > > + * Keep the clk bulk state balanced for the following
> > > > + * clk_bulk_disable_unprepare() calls (error paths and runtime PM).
> > > > + */
> > > > + ret = clk_prepare_enable(fclk);
> > > > + if (ret)
> > > > + return dev_err_probe(dev, ret, "can't enable fast_clk\n");
> > > > +
> > > > + new_clks = devm_kcalloc(dev, master->num_clks + 1, sizeof(*new_clks),
> > > > + GFP_KERNEL);
> > > > + if (!new_clks) {
> > > > + clk_disable_unprepare(fclk);
> > > > + return -ENOMEM;
> > > > + }
> > > > +
> > > > + memcpy(new_clks, master->clks,
> > > > + master->num_clks * sizeof(*master->clks));
> > > > + new_clks[master->num_clks].id = "fast_clk";
> > > > + new_clks[master->num_clks].clk = fclk;
> > > > + master->clks = new_clks;
> > > > + master->num_clks++;
> > > > +
> > > > + return 0;
> > > > +}
> > >
> > > [Severity: Medium]
> > > Can unbinding the driver leave fclk and the input clocks enabled when the
> > > device is not runtime-suspended?
> > >
> > > svc_i3c_master_remove() calls pm_runtime_dont_use_autosuspend() and
> > > pm_runtime_disable(), relying on svc_i3c_runtime_suspend() to call
> > > clk_bulk_disable_unprepare(). If CONFIG_PM is disabled, runtime PM is
> > > forbidden via sysfs, or hotjoin is enabled (which holds a runtime PM
> > > reference not released by i3c_master_unregister()), runtime suspend does
> > > not run during remove. When devres then unregisters ao->gate.hw,
> > > ao->div.hw, and ao->div_en.hw while they are still prepared and enabled,
> > > clk_unregister() triggers a kernel warning and leaks the prepare and
> > > enable counts on the parent pclk.
> >
> > Suppose it is existing problem.
> >
> Ok, not introduced by this patch.
>
> Address it in a separate patch follow this series or create new change
> series?
New series. Many codes can be removed because framework already support
call pm_runtime_get() before tranfer data after simple set a flag.
Frank
>
> > >
> > > > +
> > > > +/**
> > > > + * aml_i3c_init - Amlogic A9 specific init, called with the clocks enabled
> > > > + * @master: I3C master controller
> > > > + *
> > > > + * Register the AO variant's internal divider chain and round/set the fast
> > > > + * clock rate for both variants.
> > > > + *
> > > > + * Return: 0 on success, or a negative error code.
> > > > + */
> > > > +static int aml_i3c_init(struct svc_i3c_master *master)
> > > > +{
> > > > + struct device *dev = master->dev;
> > > > + struct clk *fclk;
> > > > + long fin_freq;
> > > > + int ret, i;
> > > > +
> > > > + if (svc_has_quirk(master, AML_I3C_QUIRK_AO)) {
> > > > + /*
> > > > + * clk_bulk_get_all() preserves the DT order, and the
> > > > + * binding restricts the AO variant to exactly one clock,
> > > > + * which is always "pclk". So clks[0] is the input of the
> > > > + * internal divider chain.
> > > > + */
> > > > + if (strcmp(master->clks[0].id, "pclk"))
> > > > + return dev_err_probe(dev, -EINVAL,
> > > > + "invalid AO input clock\n");
> > >
> > > [Severity: Medium]
> > > Can this trigger a NULL pointer dereference on master->clks or
> > > master->clks[0].id?
> > >
> > > When a device tree node has no clocks, devm_clk_bulk_get_all() returns 0
> > > and leaves master->clks as NULL, which passes the master->num_clks < 0
> > > check in svc_i3c_master_probe(). Previously, the fast_clk loop in probe
> > > ran zero iterations and returned -EINVAL, but aml_i3c_init() now runs
> > > first and indexes master->clks[0] without checking that master->num_clks
> > > is at least 1 (or equal to 1 as required by the AO binding). In addition,
> > > of_clk_bulk_get() leaves clks[i].id as NULL if clock-names is missing or
> > > shorter than clocks, which would cause strcmp() here and in the fast_clk
> > > loop below to dereference NULL.
> >
> > This one need check.
> >
>
> will add check.
>
> > Frank
> >
> > >
> > > > +
> > > > + ret = aml_ao_i3c_tail_clocks(master, master->clks[0].clk);
> > > > + if (ret)
> > > > + return ret;
> > > > + }
> > > > +
> > > > + for (i = 0; i < master->num_clks; i++) {
> > > > + if (!strcmp(master->clks[i].id, "fast_clk"))
> > > > + break;
> > > > + }
> > > > +
> > > > + if (i == master->num_clks)
> > > > + return dev_err_probe(dev, -EINVAL,
> > > > + "can't get I3C fast clock\n");
> > > > +
> > > > + fclk = master->clks[i].clk;
> > > > + if (IS_ERR(fclk))
> > > > + return PTR_ERR(fclk);
> > >
> > > [Severity: Low]
> > > This isn't a bug, but this fast_clk lookup and validation duplicates the
> > > identical loop in svc_i3c_master_probe() that runs right after
> > > vendor_init() returns. Could a shared helper or resolving master->fclk
> > > once avoid repeating the lookup?
> > >
> Will try.
> And guards against a NULL clks[i].id, and reused by both probe() and
> aml_i3c_init().
>
> > > --
> > > Sashiko AI review · https://sashiko.dev/#/patchset/20261116-a9_i3c-v4-0-3d3222b7128d@amlogic.com?part=2
>
--
linux-i3c mailing list
linux-i3c@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-i3c
^ permalink raw reply [flat|nested] 30+ messages in thread