[PATCH 0/3] soc: qcom: several fixes for SAW handling

[PATCH 0/3] soc: qcom: several fixes for SAW handling

[Dmitry Baryshkov]

The Qualcomm MSM8996 / MSM8998 / SDM660 platforms use SPM / SAW core to
handle lower-power modes and to program core voltage regulators. Apply
several small fixes to drivers for that core.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
---
Dmitry Baryshkov (3):
      soc: qcom: spm: fix the SDM660 silver L2 AVS control value
      soc: qcom: spm: don't clobber the SAW control registers
      regulator: qcom_spmi: don't take a sleeping lock in the SAW IPI

 drivers/regulator/Kconfig               |  1 +
 drivers/regulator/qcom_spmi-regulator.c | 35 +++++++++++++++++++++++++--------
 drivers/soc/qcom/spm.c                  | 10 ++++++----
 3 files changed, 34 insertions(+), 12 deletions(-)
---
base-commit: 2efb7a11f1aa4d2dc285fe2680cca22cb9f72f64
change-id: 20261019-fix-qcom-saw-e72f93e40991

Best regards,
--  
With best wishes
Dmitry

[PATCH 1/3] soc: qcom: spm: fix the SDM660 silver L2 AVS control value

[Dmitry Baryshkov]

The power cluster L2 SAW of SDM660 is programmed with an AVS control
value that differs from the one the vendor kernel uses for this SAW
(0x1010031), so the AVS hardware runs with the wrong configuration.

Use the vendor value.

Fixes: 13e72c3e2261 ("soc: qcom: spm: Implement support for SAWv4.1, SDM630/660 L2 AVS")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
---
 drivers/soc/qcom/spm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/soc/qcom/spm.c b/drivers/soc/qcom/spm.c
index f75659fff287..d41c94d32b51 100644
--- a/drivers/soc/qcom/spm.c
+++ b/drivers/soc/qcom/spm.c
@@ -100,7 +100,7 @@ static const struct spm_reg_data spm_reg_660_gold_l2  = {
 
 static const struct spm_reg_data spm_reg_660_silver_l2  = {
 	.reg_offset = spm_reg_offset_v4_1,
-	.avs_ctl = 0x101c031,
+	.avs_ctl = 0x1010031,
 	.avs_limit = 0x4580458,
 };
 

-- 
2.47.3

Re: [PATCH 1/3] soc: qcom: spm: fix the SDM660 silver L2 AVS control value

[Konrad Dybcio]

On 10/19/26 4:39 PM, Dmitry Baryshkov wrote:
> The power cluster L2 SAW of SDM660 is programmed with an AVS control
> value that differs from the one the vendor kernel uses for this SAW
> (0x1010031), so the AVS hardware runs with the wrong configuration.
> 
> Use the vendor value.
> 
> Fixes: 13e72c3e2261 ("soc: qcom: spm: Implement support for SAWv4.1, SDM630/660 L2 AVS")
> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
> ---

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad

[PATCH 2/3] soc: qcom: spm: don't clobber the SAW control registers

[Dmitry Baryshkov]

The SAWv4.1 and SAWv2.3 register tables have no sequence entry, so
probe copies the all-zero sequence buffer to offset 0 instead, wiping
the SPM_CTL, SPM_DLY and SAW_CFG values the boot firmware programmed.
On MSM8998 the firmware then still powers the gold cluster down once
its last CPU goes offline, but never brings it back: PSCI CPU_ON leaves
the first CPU pending and the SoC hangs or resets.

Only write the sequence when the SAW has a sequence entry, like the
other registers.

Fixes: 13e72c3e2261 ("soc: qcom: spm: Implement support for SAWv4.1, SDM630/660 L2 AVS")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
---
 drivers/soc/qcom/spm.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/soc/qcom/spm.c b/drivers/soc/qcom/spm.c
index d41c94d32b51..6f2f14eb9022 100644
--- a/drivers/soc/qcom/spm.c
+++ b/drivers/soc/qcom/spm.c
@@ -528,9 +528,11 @@ static int spm_dev_probe(struct platform_device *pdev)
 	platform_set_drvdata(pdev, drv);
 
 	/* Write the SPM sequences first.. */
-	addr = drv->reg_base + drv->reg_data->reg_offset[SPM_REG_SEQ_ENTRY];
-	__iowrite32_copy(addr, drv->reg_data->seq,
-			ARRAY_SIZE(drv->reg_data->seq) / 4);
+	if (drv->reg_data->reg_offset[SPM_REG_SEQ_ENTRY]) {
+		addr = drv->reg_base + drv->reg_data->reg_offset[SPM_REG_SEQ_ENTRY];
+		__iowrite32_copy(addr, drv->reg_data->seq,
+				 ARRAY_SIZE(drv->reg_data->seq) / 4);
+	}
 
 	/*
 	 * ..and then the control registers.

-- 
2.47.3

Re: [PATCH 2/3] soc: qcom: spm: don't clobber the SAW control registers

[Konrad Dybcio]

On 10/19/26 4:39 PM, Dmitry Baryshkov wrote:
> The SAWv4.1 and SAWv2.3 register tables have no sequence entry, so
> probe copies the all-zero sequence buffer to offset 0 instead, wiping
> the SPM_CTL, SPM_DLY and SAW_CFG values the boot firmware programmed.
> On MSM8998 the firmware then still powers the gold cluster down once
> its last CPU goes offline, but never brings it back: PSCI CPU_ON leaves
> the first CPU pending and the SoC hangs or resets.
> 
> Only write the sequence when the SAW has a sequence entry, like the
> other registers.
> 
> Fixes: 13e72c3e2261 ("soc: qcom: spm: Implement support for SAWv4.1, SDM630/660 L2 AVS")
> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
> ---

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad

>  drivers/soc/qcom/spm.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/soc/qcom/spm.c b/drivers/soc/qcom/spm.c
> index d41c94d32b51..6f2f14eb9022 100644
> --- a/drivers/soc/qcom/spm.c
> +++ b/drivers/soc/qcom/spm.c
> @@ -528,9 +528,11 @@ static int spm_dev_probe(struct platform_device *pdev)
>  	platform_set_drvdata(pdev, drv);
>  
>  	/* Write the SPM sequences first.. */
> -	addr = drv->reg_base + drv->reg_data->reg_offset[SPM_REG_SEQ_ENTRY];
> -	__iowrite32_copy(addr, drv->reg_data->seq,
> -			ARRAY_SIZE(drv->reg_data->seq) / 4);
> +	if (drv->reg_data->reg_offset[SPM_REG_SEQ_ENTRY]) {
> +		addr = drv->reg_base + drv->reg_data->reg_offset[SPM_REG_SEQ_ENTRY];
> +		__iowrite32_copy(addr, drv->reg_data->seq,
> +				 ARRAY_SIZE(drv->reg_data->seq) / 4);
> +	}
>  
>  	/*
>  	 * ..and then the control registers.
> 

[PATCH 3/3] regulator: qcom_spmi: don't take a sleeping lock in the SAW IPI

[Dmitry Baryshkov]

The SAW voltage change runs on CPU 0 from an IPI, but it reaches the SAW
registers through the syscon regmap, which is protected by a spinlock_t.
That is a sleeping lock on PREEMPT_RT, and lockdep complains as soon as
VDD_APCC actually changes voltage:

  BUG: Invalid wait context
  ...
  regmap_lock_spinlock+0x18/0x30
  regmap_read+0x3c/0x78
  spmi_saw_set_vdd+0x50/0x1ac
  __flush_smp_call_function_queue+0x1ac/0x334

Map the SAW with a regmap of our own that uses a raw spinlock, and fail
the probe when that cannot be done instead of carrying an error pointer
around as the SAW regmap.

Fixes: 0caecaa87202 ("regulator: qcom_spmi: Add support for SAW")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
---
 drivers/regulator/Kconfig               |  1 +
 drivers/regulator/qcom_spmi-regulator.c | 35 +++++++++++++++++++++++++--------
 2 files changed, 28 insertions(+), 8 deletions(-)

diff --git a/drivers/regulator/Kconfig b/drivers/regulator/Kconfig
index be6b72daa76f..83a76b78e58c 100644
--- a/drivers/regulator/Kconfig
+++ b/drivers/regulator/Kconfig
@@ -1210,6 +1210,7 @@ config REGULATOR_QCOM_SMD_RPM
 config REGULATOR_QCOM_SPMI
 	tristate "Qualcomm SPMI regulator driver"
 	depends on SPMI || COMPILE_TEST
+	select REGMAP_MMIO
 	help
 	  If you say yes to this option, support will be included for the
 	  regulators found in Qualcomm SPMI PMICs.
diff --git a/drivers/regulator/qcom_spmi-regulator.c b/drivers/regulator/qcom_spmi-regulator.c
index c1a41ce70b36..a04ddb1eac94 100644
--- a/drivers/regulator/qcom_spmi-regulator.c
+++ b/drivers/regulator/qcom_spmi-regulator.c
@@ -12,13 +12,13 @@
 #include <linux/bitops.h>
 #include <linux/slab.h>
 #include <linux/of.h>
+#include <linux/of_address.h>
 #include <linux/of_device.h>
 #include <linux/platform_device.h>
 #include <linux/ktime.h>
 #include <linux/regulator/driver.h>
 #include <linux/regmap.h>
 #include <linux/list.h>
-#include <linux/mfd/syscon.h>
 #include <linux/io.h>
 
 /* Pin control enable input pins. */
@@ -1335,6 +1335,14 @@ static irqreturn_t spmi_regulator_vs_ocp_isr(int irq, void *data)
 
 static struct regmap *saw_regmap;
 
+static const struct regmap_config saw_regmap_config = {
+	.name = "saw",
+	.reg_bits = 32,
+	.val_bits = 32,
+	.reg_stride = 4,
+	.use_raw_spinlock = true,
+};
+
 static void spmi_saw_set_vdd(void *data)
 {
 	u32 vctl, data3, avs_ctl, pmic_sts;
@@ -2504,7 +2512,8 @@ static int qcom_spmi_regulator_probe(struct platform_device *pdev)
 	const char *name;
 	struct device *dev = &pdev->dev;
 	struct device_node *node = pdev->dev.of_node;
-	struct device_node *syscon, *reg_node;
+	struct device_node *saw, *reg_node;
+	void __iomem *saw_base;
 	struct property *reg_prop;
 	int ret, lenp;
 	struct list_head *vreg_list;
@@ -2523,12 +2532,22 @@ static int qcom_spmi_regulator_probe(struct platform_device *pdev)
 	if (!reg)
 		return -ENODEV;
 
-	syscon = of_parse_phandle(node, "qcom,saw-reg", 0);
-	if (syscon) {
-		saw_regmap = syscon_node_to_regmap(syscon);
-		of_node_put(syscon);
-		if (IS_ERR(saw_regmap))
-			dev_err(dev, "ERROR reading SAW regmap\n");
+	saw = of_parse_phandle(node, "qcom,saw-reg", 0);
+	if (saw) {
+		saw_base = devm_of_iomap(dev, saw, 0, NULL);
+		of_node_put(saw);
+		if (IS_ERR(saw_base))
+			return dev_err_probe(dev, PTR_ERR(saw_base),
+					     "failed to map the SAW\n");
+
+		saw_regmap = devm_regmap_init_mmio(dev, saw_base,
+						   &saw_regmap_config);
+		if (IS_ERR(saw_regmap)) {
+			ret = PTR_ERR(saw_regmap);
+			saw_regmap = NULL;
+			return dev_err_probe(dev, ret,
+					     "failed to set the SAW regmap up\n");
+		}
 	}
 
 	for (; reg->name; reg++) {

-- 
2.47.3

Re: [PATCH 3/3] regulator: qcom_spmi: don't take a sleeping lock in the SAW IPI

[Konrad Dybcio]

On 10/19/26 4:39 PM, Dmitry Baryshkov wrote:
> The SAW voltage change runs on CPU 0 from an IPI, but it reaches the SAW
> registers through the syscon regmap, which is protected by a spinlock_t.
> That is a sleeping lock on PREEMPT_RT, and lockdep complains as soon as
> VDD_APCC actually changes voltage:

The SAW handling barely belongs in that driver to be honest.. but
separating it out would make a mess out of sharing the of_node

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad

Re: [PATCH 3/3] regulator: qcom_spmi: don't take a sleeping lock in the SAW IPI

[Sebastian Andrzej Siewior]

On 2026-10-19 17:39:21 [+0300], Dmitry Baryshkov wrote:
> --- a/drivers/regulator/qcom_spmi-regulator.c
> +++ b/drivers/regulator/qcom_spmi-regulator.c
> @@ -2523,12 +2532,22 @@ static int qcom_spmi_regulator_probe(struct platform_device *pdev)
>  	if (!reg)
>  		return -ENODEV;
>  
> -	syscon = of_parse_phandle(node, "qcom,saw-reg", 0);
> -	if (syscon) {
> -		saw_regmap = syscon_node_to_regmap(syscon);
> -		of_node_put(syscon);
> -		if (IS_ERR(saw_regmap))
> -			dev_err(dev, "ERROR reading SAW regmap\n");
> +	saw = of_parse_phandle(node, "qcom,saw-reg", 0);
> +	if (saw) {
> +		saw_base = devm_of_iomap(dev, saw, 0, NULL);

I was asking if the syscon could be made using a raw_spinlock_t if all
users do simple reads/writes.
	https://lore.kernel.org/all/20260929-rbrue-suez-upstreaming-mtk-pinctrl-raw-regmap-v1-1-db92943f42cb@gmail.com/T/#u

If the qcom,saw-reg registers are not shared anywhere, then this should
work, too.

> +		of_node_put(saw);

Sebastian

Re: [PATCH 0/3] soc: qcom: several fixes for SAW handling

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 781 bytes --]

On Mon, Oct 19, 2026 at 05:39:18PM +0300, Dmitry Baryshkov wrote:
> The Qualcomm MSM8996 / MSM8998 / SDM660 platforms use SPM / SAW core to
> handle lower-power modes and to program core voltage regulators. Apply
> several small fixes to drivers for that core.
> 
> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
> ---
> Dmitry Baryshkov (3):
>       soc: qcom: spm: fix the SDM660 silver L2 AVS control value
>       soc: qcom: spm: don't clobber the SAW control registers
>       regulator: qcom_spmi: don't take a sleeping lock in the SAW IPI

Are there any actual dependencies between the patches here?  Please
don't create cross tree serieses unless there are dependencies, it just
makes it harder to figure out how things will get merged.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]