Skip to contentSkip to conversation

hwmon: corsair-psu: fix unintentional sign extension issue

PATCH In review Colin King 3 participants 6 messages Nov 5, 2020 – Nov 8, 2020

Subsystems (3)
lore
Share & tools
lore

Colin King

author
#0 lore raw

PATCH 1/1 hwmon: corsair-psu: fix unintentional sign extension issue

From: Colin Ian King <[email protected]>

The shifting of the u8 integer data[3] by 24 bits to the left will be promoted to a 32 bit signed int and then sign-extended to a long. In the event that the top bit of data[3] is set then all then all the upper 32 bits of a 64 bit long end up as also being set because of the sign-extension. Fix this by casting data[3] to a long before the shift.

Addresses-Coverity: ("Unintended sign extension")

Fixes ce15cd2cee8b ("hwmon: add Corsair PSU HID controller driver")
---
 drivers/hwmon/corsair-psu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/corsair-psu.c +1 -1
index e92d0376e7ac..5d19a888231a 100644
@@ -241,7 +241,7 @@ static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8 rail, l
 	 * the LINEAR11 conversion are the watts values which are about 1200 for the strongest psu
 	 * supported (HX1200i)
 	 */
-	tmp = (data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
+	tmp = ((long)data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
 	switch (cmd) {
 	case PSU_CMD_IN_VOLTS:
 	case PSU_CMD_IN_AMPS:
Signature · 1 line
2.27.0
maintainer
#1 lore raw
Quoted 24 lines · On Thu, 5 Nov 2020 11:50:19 +0000 Colin King <[email protected]> wrote:

From: Colin Ian King <[email protected]>

The shifting of the u8 integer data[3] by 24 bits to the left will be promoted to a 32 bit signed int and then sign-extended to a long. In the event that the top bit of data[3] is set then all then all the upper 32 bits of a 64 bit long end up as also being set because of the sign-extension. Fix this by casting data[3] to a long before the shift.

Addresses-Coverity: ("Unintended sign extension")

Fixes ce15cd2cee8b ("hwmon: add Corsair PSU HID controller driver")
---
 drivers/hwmon/corsair-psu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/corsair-psu.c +0 -0
index e92d0376e7ac..5d19a888231a 100644
@@ -241,7 +241,7 @@ static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8

rail, l
* the LINEAR11 conversion are the watts values which are about 1200 for the strongest
psu

drivers/hwmon/corsair-psu.c static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8 rail, l · line 247

Quoted 7 lines
drivers/hwmon/corsair-psu.c +1 -1
 	 * supported (HX1200i)
 	 */
-	tmp = (data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
+	tmp = ((long)data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
 	switch (cmd) {
 	case PSU_CMD_IN_VOLTS:
 	case PSU_CMD_IN_AMPS:

Yeah, this could happen if the uptime value in the micro-controller gets bigger than 68 years (in seconds), and it is the only value which actually uses more than 2 bytes for the representation. So what about architectures which are 32 bit wide and where a long has 32 bits? I guess this simple cast is not enough.

greetings,
Wilken

#2 lore raw
Quoted 31 lines · Wilken Gottwalt
Quoted 28 lines · On Thu, 5 Nov 2020 11:50:19 +0000 Colin King <[email protected]> wrote:

From: Colin Ian King <[email protected]>

The shifting of the u8 integer data[3] by 24 bits to the left will be promoted to a 32 bit signed int and then sign-extended to a long. In the event that the top bit of data[3] is set then all then all the upper 32 bits of a 64 bit long end up as also being set because of the sign-extension. Fix this by casting data[3] to a long before the shift.

Addresses-Coverity: ("Unintended sign extension")

Fixes ce15cd2cee8b ("hwmon: add Corsair PSU HID controller driver")
---
 drivers/hwmon/corsair-psu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/corsair-psu.c +0 -0
index e92d0376e7ac..5d19a888231a 100644
@@ -241,7 +241,7 @@ static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8

rail, l
* the LINEAR11 conversion are the watts values which are about 1200 for the strongest
psu

drivers/hwmon/corsair-psu.c +1 -1
 	 * supported (HX1200i)
 	 */
-	tmp = (data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
+	tmp = ((long)data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
Quoted 8 lines

drivers/hwmon/corsair-psu.c static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8 rail, l · line 247

Quoted 3 lines
drivers/hwmon/corsair-psu.c +0 -0
 	switch (cmd) {
 	case PSU_CMD_IN_VOLTS:
 	case PSU_CMD_IN_AMPS:

Yeah, this could happen if the uptime value in the micro-controller gets bigger than 68 years (in seconds), and it is the only value which actually uses more than 2 bytes for the representation. So what about architectures which are 32 bit wide and where a long has 32 bits? I guess this simple cast is not enough.

For 32 bits (unsigned) the 4 u8 values in data represents ~136 years no matter if we use a 32 or 64 bit long for tmp. The cast to long is signed, so yes, that's ~68 years in seconds. So perhaps corsairpsu_get_value() should be passing a unsigned long for the *val arg and tmp should be unsigned long too?

Quoted 4 lines

greetings,
Wilken

maintainer
#3 lore raw
In reply to Wilken Gottwalt
Quoted 31 lines · Wilken Gottwalt
Quoted 28 lines · On Thu, 5 Nov 2020 11:50:19 +0000 Colin King <[email protected]> wrote:

From: Colin Ian King <[email protected]>

The shifting of the u8 integer data[3] by 24 bits to the left will be promoted to a 32 bit signed int and then sign-extended to a long. In the event that the top bit of data[3] is set then all then all the upper 32 bits of a 64 bit long end up as also being set because of the sign-extension. Fix this by casting data[3] to a long before the shift.

Addresses-Coverity: ("Unintended sign extension")

Fixes ce15cd2cee8b ("hwmon: add Corsair PSU HID controller driver")
---
 drivers/hwmon/corsair-psu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/corsair-psu.c +0 -0
index e92d0376e7ac..5d19a888231a 100644
@@ -241,7 +241,7 @@ static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8

rail, l
* the LINEAR11 conversion are the watts values which are about 1200 for the strongest
psu

drivers/hwmon/corsair-psu.c +1 -1
 	 * supported (HX1200i)
 	 */
-	tmp = (data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
+	tmp = ((long)data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
Quoted 9 lines

drivers/hwmon/corsair-psu.c static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8 rail, l · line 247

Quoted 3 lines
drivers/hwmon/corsair-psu.c +0 -0
 	switch (cmd) {
 	case PSU_CMD_IN_VOLTS:
 	case PSU_CMD_IN_AMPS:

Yeah, this could happen if the uptime value in the micro-controller gets bigger than 68 years (in seconds), and it is the only value which actually uses more than 2 bytes for the representation. So what about architectures which are 32 bit wide and where a long has 32 bits? I guess this simple cast is not enough.

The hwmon subsystem uses 'long' to pass values back to the core. While that may be a bit unfortunate as it doesn't support uptimes of more than 68 years in seconds, we are not going to change the hwmon core to accommodate it. If the incoming data is always expected to be positive, and you don't want to risk the overflow, feel free to either use "data[3] & 0x7f" instead, or max out the result at INT_MAX (from <linux/limits.h>).

Thanks,
Guenter

maintainer
#4 lore raw
Quoted 50 lines · On Thu, 5 Nov 2020 06:15:49 -0800 Guenter Roeck <[email protected]> wrote:
Quoted 31 lines · Wilken Gottwalt
Quoted 28 lines · On Thu, 5 Nov 2020 11:50:19 +0000 Colin King <[email protected]> wrote:

From: Colin Ian King <[email protected]>

The shifting of the u8 integer data[3] by 24 bits to the left will be promoted to a 32 bit signed int and then sign-extended to a long. In the event that the top bit of data[3] is set then all then all the upper 32 bits of a 64 bit long end up as also being set because of the sign-extension. Fix this by casting data[3] to a long before the shift.

Addresses-Coverity: ("Unintended sign extension")

Fixes ce15cd2cee8b ("hwmon: add Corsair PSU HID controller driver")
---
 drivers/hwmon/corsair-psu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/corsair-psu.c +0 -0
index e92d0376e7ac..5d19a888231a 100644
@@ -241,7 +241,7 @@ static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8

rail, l
* the LINEAR11 conversion are the watts values which are about 1200 for the
strongest psu

drivers/hwmon/corsair-psu.c +1 -1
 	 * supported (HX1200i)
 	 */
-	tmp = (data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
+	tmp = ((long)data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
Quoted 9 lines

drivers/hwmon/corsair-psu.c static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8 rail, l · line 247

Quoted 3 lines
drivers/hwmon/corsair-psu.c +0 -0
 	switch (cmd) {
 	case PSU_CMD_IN_VOLTS:
 	case PSU_CMD_IN_AMPS:

Yeah, this could happen if the uptime value in the micro-controller gets bigger than 68 years (in seconds), and it is the only value which actually uses more than 2 bytes for the representation. So what about architectures which are 32 bit wide and where a long has 32 bits? I guess this simple cast is not enough.

The hwmon subsystem uses 'long' to pass values back to the core. While that may be a bit unfortunate as it doesn't support uptimes of more than 68 years in seconds, we are not going to change the hwmon core to accommodate it. If the incoming data is always expected to be positive, and you don't want to risk the overflow, feel free to either use "data[3] & 0x7f" instead, or max out the result at INT_MAX (from <linux/limits.h>).

Quoted 2 lines

Thanks,
Guenter

And it is basically impossible to hit the 68 years. For example here are the values for my psu, which I got about 6 years ago (2014 is the first year it was sold) and which runs 8-20 hours nearly every day: 656 day(s), 21:26:34, that are not even 2 years. And we are talking about gaming PSUs here, which no sane person would put into a server where it runs 24/7. And if so, it wouldn't survive 10 uptime years, not to mention 68 years. I'm pretty sure that the fan in my PSU is somewhat damaged already (hence the best guess fan value calculation). So yes, technically it is not perfect but practically it doesn't make much sense here trying to find the perfect solution. It will only introduce unnecessary complexity for a scenario which is so unlikely to happen at all. I actually spent a lot of time thinking about it, and this is the most practical solution I found - and which made Guenter happy ;-)

Though, I must admit there is a way to solve this problem - by making chained requests and get all values at once and parse/split the response. You can easily join all requests in this 64 bytes hid message. But this may not work reliable for all supported devices. I already encountered issues while joining rail select and read a rail value. It would at least increase the complexity to quite a new level.

greetings,
Wilken

maintainer
#5 lore raw
In reply to Colin King
Quoted 9 lines · Colin King

From: Colin Ian King <[email protected]>

The shifting of the u8 integer data[3] by 24 bits to the left will be promoted to a 32 bit signed int and then sign-extended to a long. In the event that the top bit of data[3] is set then all then all the upper 32 bits of a 64 bit long end up as also being set because of the sign-extension. Fix this by casting data[3] to a long before the shift.

Quoted 3 lines

Addresses-Coverity: ("Unintended sign extension")

Fixes ce15cd2cee8b ("hwmon: add Corsair PSU HID controller driver")

I wasn't sure if I am going to get another version of this patch. Either case, I applied this now.

Guenter

Quoted 17 lines
---
 drivers/hwmon/corsair-psu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/corsair-psu.c +1 -1
index e92d0376e7ac..5d19a888231a 100644
@@ -241,7 +241,7 @@ static int corsairpsu_get_value(struct corsairpsu_data *priv, u8 cmd, u8 rail, l
 	 * the LINEAR11 conversion are the watts values which are about 1200 for the strongest psu
 	 * supported (HX1200i)
 	 */
-	tmp = (data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
+	tmp = ((long)data[3] << 24) + (data[2] << 16) + (data[1] << 8) + data[0];
 	switch (cmd) {
 	case PSU_CMD_IN_VOLTS:
 	case PSU_CMD_IN_AMPS: