[PATCH v3] iio: adc: pac1934: validate ACPI _DSM return type and bounds

[PATCH v3] iio: adc: pac1934: validate ACPI _DSM return type and bounds

[Tristan Madani]

From: Tristan Madani <tristan@talencesecurity.com>

pac1934_acpi_parse_channel_config() calls acpi_evaluate_dsm() four
times to read channel configuration but does not validate the returned
object type or array bounds:

  1. No check that the returned object is ACPI_TYPE_PACKAGE.  If
     firmware returns a different type, accessing rez->package.count
     and rez->package.elements on a non-package object produces
     unpredictable values.

  2. The loop bounds use rez->package.count directly without capping
     to PAC1934_MAX_NUM_CHANNELS (4).  A package with more elements
     than expected causes writes past the fixed-size labels[],
     shunts[], and active_channels[] arrays.

  3. The bipolar and sample rate DSM calls access elements[0] without
     verifying package.count >= 1.

Use acpi_evaluate_dsm_typed() with ACPI_TYPE_PACKAGE to validate the
return type at evaluation time.  Cap loop iterations to
PAC1934_MAX_NUM_CHANNELS, and verify package count >= 1 before
accessing elements[0].

Signed-off-by: Tristan Madani <tristan@talencesecurity.com>
---
Changes in v3:
- Use acpi_evaluate_dsm_typed() instead of manual type checks, per
  Andy Shevchenko
- Use (count & ~1u) for the name+mohm loop bound to ensure pairs
  are complete, per Andy Shevchenko
- Drop Fixes: and Cc: stable tags since this is hardening rather
  than a fix for a specific regression, per Jonathan Cameron

 drivers/iio/adc/pac1934.c | 32 ++++++++++++++++++++++++++------
 1 file changed, 26 insertions(+), 6 deletions(-)

diff --git a/drivers/iio/adc/pac1934.c b/drivers/iio/adc/pac1934.c
index de59dc27c4405..b6315a13a353e 100644
--- a/drivers/iio/adc/pac1934.c
+++ b/drivers/iio/adc/pac1934.c
@@ -1092,17 +1092,22 @@ static int pac1934_acpi_parse_channel_config(struct i2c_client *client,
 	struct device *dev = &client->dev;
 	unsigned short bi_dir_mask;
 	int idx, i;
+	unsigned int count;
 	guid_t guid;
 
 	handle = ACPI_HANDLE(dev);
 
 	guid_parse(PAC1934_DSM_UUID, &guid);
 
-	rez = acpi_evaluate_dsm(handle, &guid, 0, PAC1934_ACPI_GET_NAMES_AND_MOHMS_VALS, NULL);
+	rez = acpi_evaluate_dsm_typed(handle, &guid, 0,
+				      PAC1934_ACPI_GET_NAMES_AND_MOHMS_VALS,
+				      NULL, ACPI_TYPE_PACKAGE);
 	if (!rez)
 		return -EINVAL;
 
-	for (i = 0; i < rez->package.count; i += 2) {
+	count = min(rez->package.count & ~1u,
+		    (unsigned int)(PAC1934_MAX_NUM_CHANNELS * 2));
+	for (i = 0; i < count; i += 2) {
 		idx = i / 2;
 		info->labels[idx] =
 			devm_kmemdup(dev, rez->package.elements[i].string.pointer,
@@ -1119,7 +1124,9 @@ static int pac1934_acpi_parse_channel_config(struct i2c_client *client,
 
 	ACPI_FREE(rez);
 
-	rez = acpi_evaluate_dsm(handle, &guid, 1, PAC1934_ACPI_GET_UOHMS_VALS, NULL);
+	rez = acpi_evaluate_dsm_typed(handle, &guid, 1,
+				      PAC1934_ACPI_GET_UOHMS_VALS,
+				      NULL, ACPI_TYPE_PACKAGE);
 	if (!rez) {
 		/*
 		 * initializing with default values
@@ -1130,7 +1137,8 @@ static int pac1934_acpi_parse_channel_config(struct i2c_client *client,
 		return 0;
 	}
 
-	for (i = 0; i < rez->package.count; i++) {
+	count = min(rez->package.count, (unsigned int)PAC1934_MAX_NUM_CHANNELS);
+	for (i = 0; i < count; i++) {
 		idx = i;
 		info->shunts[idx] = rez->package.elements[i].integer.value;
 		info->active_channels[idx] = (info->shunts[idx] != 0);
@@ -1138,9 +1146,15 @@ static int pac1934_acpi_parse_channel_config(struct i2c_client *client,
 
 	ACPI_FREE(rez);
 
-	rez = acpi_evaluate_dsm(handle, &guid, 1, PAC1934_ACPI_GET_BIPOLAR_SETTINGS, NULL);
+	rez = acpi_evaluate_dsm_typed(handle, &guid, 1,
+				      PAC1934_ACPI_GET_BIPOLAR_SETTINGS,
+				      NULL, ACPI_TYPE_PACKAGE);
 	if (!rez)
 		return -EINVAL;
+	if (rez->package.count < 1) {
+		ACPI_FREE(rez);
+		return -EINVAL;
+	}
 
 	bi_dir_mask = rez->package.elements[0].integer.value;
 	info->bi_dir[0] = ((bi_dir_mask & (1 << 3)) | (bi_dir_mask & (1 << 7))) != 0;
@@ -1150,9 +1164,15 @@ static int pac1934_acpi_parse_channel_config(struct i2c_client *client,
 
 	ACPI_FREE(rez);
 
-	rez = acpi_evaluate_dsm(handle, &guid, 1, PAC1934_ACPI_GET_SAMP, NULL);
+	rez = acpi_evaluate_dsm_typed(handle, &guid, 1,
+				      PAC1934_ACPI_GET_SAMP,
+				      NULL, ACPI_TYPE_PACKAGE);
 	if (!rez)
 		return -EINVAL;
+	if (rez->package.count < 1) {
+		ACPI_FREE(rez);
+		return -EINVAL;
+	}
 
 	info->sample_rate_value = rez->package.elements[0].integer.value;
 
-- 
2.47.3

Re: [PATCH v3] iio: adc: pac1934: validate ACPI _DSM return type and bounds

[Andy Shevchenko]

On Thu, Oct 29, 2026 at 12:14:47AM +0000, Tristan Madani wrote:
> 
> pac1934_acpi_parse_channel_config() calls acpi_evaluate_dsm() four
> times to read channel configuration but does not validate the returned
> object type or array bounds:
> 
>   1. No check that the returned object is ACPI_TYPE_PACKAGE.  If
>      firmware returns a different type, accessing rez->package.count
>      and rez->package.elements on a non-package object produces
>      unpredictable values.
> 
>   2. The loop bounds use rez->package.count directly without capping
>      to PAC1934_MAX_NUM_CHANNELS (4).  A package with more elements
>      than expected causes writes past the fixed-size labels[],
>      shunts[], and active_channels[] arrays.
> 
>   3. The bipolar and sample rate DSM calls access elements[0] without
>      verifying package.count >= 1.
> 
> Use acpi_evaluate_dsm_typed() with ACPI_TYPE_PACKAGE to validate the
> return type at evaluation time.  Cap loop iterations to
> PAC1934_MAX_NUM_CHANNELS, and verify package count >= 1 before
> accessing elements[0].

...

> static int pac1934_acpi_parse_channel_config(struct i2c_client *client,

>  	struct device *dev = &client->dev;
>  	unsigned short bi_dir_mask;
>  	int idx, i;
> +	unsigned int count;
>  	guid_t guid;

Preserve reversed xmas tree order.


> +	count = min(rez->package.count & ~1u,
> +		    (unsigned int)(PAC1934_MAX_NUM_CHANNELS * 2));

No, you are not supposed to cast parameters for min(). It should work for int
as small positive predefined number.

...

Ditto for the other case of min().

-- 
With Best Regards,
Andy Shevchenko