From: Arnd Bergmann arnd@arndb.de
The tee_ioctl_object_invoke_arg structure has padding on some architectures but not on x86-32 and a few others:
include/linux/tee.h:474:32: error: padding struct to align 'params' [-Werror=padded]
I expect that all current users of this are on architectures that do have implicit padding here (arm64, arm, x86, riscv), so make the padding explicit in order to avoid surprises if this later gets used elsewhere.
Fixes: d5b8b0fa1775 ("tee: add TEE_IOCTL_PARAM_ATTR_TYPE_OBJREF") Signed-off-by: Arnd Bergmann arnd@arndb.de --- The new interface showed up in 6.18, but I only came across this after that was released. Changing it now is technically an ABI change on architectures with unusual padding rules, so please consider carefully whether we want to do it this way or not.
Working around the ABI differences without an ABI change is possible, but adds a lot of complexity for compat handling. --- include/uapi/linux/tee.h | 1 + 1 file changed, 1 insertion(+)
diff --git a/include/uapi/linux/tee.h b/include/uapi/linux/tee.h index cab5cadca8ef..5203977ed35d 100644 --- a/include/uapi/linux/tee.h +++ b/include/uapi/linux/tee.h @@ -470,6 +470,7 @@ struct tee_ioctl_object_invoke_arg { __u32 op; __u32 ret; __u32 num_params; + __u32 :32; /* num_params tells the actual number of element in params */ struct tee_ioctl_param params[]; };
Hi,
On Thu, Dec 4, 2025 at 11:17 AM Arnd Bergmann arnd@kernel.org wrote:
From: Arnd Bergmann arnd@arndb.de
The tee_ioctl_object_invoke_arg structure has padding on some architectures but not on x86-32 and a few others:
include/linux/tee.h:474:32: error: padding struct to align 'params' [-Werror=padded]
I expect that all current users of this are on architectures that do have implicit padding here (arm64, arm, x86, riscv), so make the padding explicit in order to avoid surprises if this later gets used elsewhere.
Fixes: d5b8b0fa1775 ("tee: add TEE_IOCTL_PARAM_ATTR_TYPE_OBJREF") Signed-off-by: Arnd Bergmann arnd@arndb.de
The new interface showed up in 6.18, but I only came across this after that was released. Changing it now is technically an ABI change on architectures with unusual padding rules, so please consider carefully whether we want to do it this way or not.
Working around the ABI differences without an ABI change is possible, but adds a lot of complexity for compat handling.
This is currently only used by the recently introduced qcomtee backend driver. So it's only used on a few arm64 Qualcomm platforms right now.
I think we should take this patch, but let's hear what others think.
Thanks, Jens
include/uapi/linux/tee.h | 1 + 1 file changed, 1 insertion(+)
diff --git a/include/uapi/linux/tee.h b/include/uapi/linux/tee.h index cab5cadca8ef..5203977ed35d 100644 --- a/include/uapi/linux/tee.h +++ b/include/uapi/linux/tee.h @@ -470,6 +470,7 @@ struct tee_ioctl_object_invoke_arg { __u32 op; __u32 ret; __u32 num_params;
__u32 :32; /* num_params tells the actual number of element in params */ struct tee_ioctl_param params[];};
2.39.5
Hi,
On 12/4/2025 6:57 PM, Jens Wiklander wrote:
Hi,
On Thu, Dec 4, 2025 at 11:17 AM Arnd Bergmann arnd@kernel.org wrote:
From: Arnd Bergmann arnd@arndb.de
The tee_ioctl_object_invoke_arg structure has padding on some architectures but not on x86-32 and a few others:
include/linux/tee.h:474:32: error: padding struct to align 'params' [-Werror=padded]
I expect that all current users of this are on architectures that do have implicit padding here (arm64, arm, x86, riscv), so make the padding explicit in order to avoid surprises if this later gets used elsewhere.
Fixes: d5b8b0fa1775 ("tee: add TEE_IOCTL_PARAM_ATTR_TYPE_OBJREF") Signed-off-by: Arnd Bergmann arnd@arndb.de
The new interface showed up in 6.18, but I only came across this after that was released. Changing it now is technically an ABI change on architectures with unusual padding rules, so please consider carefully whether we want to do it this way or not.
Working around the ABI differences without an ABI change is possible, but adds a lot of complexity for compat handling.
This is currently only used by the recently introduced qcomtee backend driver. So it's only used on a few arm64 Qualcomm platforms right now.
I think we should take this patch, but let's hear what others think.
Thanks, Jens
The only user-space client which is currently using this ABI (as per our knowledge) is the libqcomtee library: https://github.com/quic/quic-teec/blob/main/libqcomtee/src/linux/tee.h#L432
If I understand Arnd's concern correctly, if a compiler used to build the user-space client skips the padding for tee_ioctl_object_invoke_arg, it could lead to issues.
Let's wait for Amir's view here as well, however I do think the explicit padding would benefit here.
Regards, Harshal
include/uapi/linux/tee.h | 1 + 1 file changed, 1 insertion(+)
diff --git a/include/uapi/linux/tee.h b/include/uapi/linux/tee.h index cab5cadca8ef..5203977ed35d 100644 --- a/include/uapi/linux/tee.h +++ b/include/uapi/linux/tee.h @@ -470,6 +470,7 @@ struct tee_ioctl_object_invoke_arg { __u32 op; __u32 ret; __u32 num_params;
__u32 :32; /* num_params tells the actual number of element in params */ struct tee_ioctl_param params[];};
2.39.5
On Fri, Dec 5, 2025, at 14:45, Harshal Dev wrote:
On 12/4/2025 6:57 PM, Jens Wiklander wrote:
On Thu, Dec 4, 2025 at 11:17 AM Arnd Bergmann arnd@kernel.org wrote:
The only user-space client which is currently using this ABI (as per our knowledge) is the libqcomtee library: https://github.com/quic/quic-teec/blob/main/libqcomtee/src/linux/tee.h#L432
If I understand Arnd's concern correctly, if a compiler used to build the user-space client skips the padding for tee_ioctl_object_invoke_arg, it could lead to issues.
Let's wait for Amir's view here as well, however I do think the explicit padding would benefit here.
The problem is much narrower: as the amount of padding is determined by the architecture specific ABI, kernel and userspace on the same architecture always agree, and specifically 32-bit Arm userspace and 64-bit Arm userspace (aarch64) also have the same rules, so there is no problem on Qualcomm's platforms even with compat 32-bit userspace.
The only actual ABI problem would happen on 32-bit x86 (i386) userspace running on a 64-bit x86 kernel, since i386-linux compilers have different alignment rules from most other architectures.
Arnd
On 12/5/2025 7:26 PM, Arnd Bergmann wrote:
On Fri, Dec 5, 2025, at 14:45, Harshal Dev wrote:
On 12/4/2025 6:57 PM, Jens Wiklander wrote:
On Thu, Dec 4, 2025 at 11:17 AM Arnd Bergmann arnd@kernel.org wrote:
The only user-space client which is currently using this ABI (as per our knowledge) is the libqcomtee library: https://github.com/quic/quic-teec/blob/main/libqcomtee/src/linux/tee.h#L432
If I understand Arnd's concern correctly, if a compiler used to build the user-space client skips the padding for tee_ioctl_object_invoke_arg, it could lead to issues.
Let's wait for Amir's view here as well, however I do think the explicit padding would benefit here.
The problem is much narrower: as the amount of padding is determined by the architecture specific ABI, kernel and userspace on the same architecture always agree, and specifically 32-bit Arm userspace and 64-bit Arm userspace (aarch64) also have the same rules, so there is no problem on Qualcomm's platforms even with compat 32-bit userspace.
The only actual ABI problem would happen on 32-bit x86 (i386) userspace running on a 64-bit x86 kernel, since i386-linux compilers have different alignment rules from most other architectures.
Well, as Jens pointed out, the TEE_IOC_OBJECT_INVOKE ABI which supports object-based IPC currently has no back-end drivers that attempt to communicate with a TEE running on x86 and implementing their end of the object-IPC protocol in the firmware. So this obviously won't be an immediate issue.
But as you pointed out, the issue would pop up if someone on x86 decides to implement it at some point in the far away future.
I do not see any harm this patch would do for our existing implementation, we'll just update our user-space library ABI in-line with this.
Thanks, Harshal
Arnd
Hi,
On 12/5/2025 12:27 AM, Jens Wiklander wrote:
Hi,
On Thu, Dec 4, 2025 at 11:17 AM Arnd Bergmann arnd@kernel.org wrote:
From: Arnd Bergmann arnd@arndb.de
The tee_ioctl_object_invoke_arg structure has padding on some architectures but not on x86-32 and a few others:
include/linux/tee.h:474:32: error: padding struct to align 'params' [-Werror=padded]
I expect that all current users of this are on architectures that do have implicit padding here (arm64, arm, x86, riscv), so make the padding explicit in order to avoid surprises if this later gets used elsewhere.
Fixes: d5b8b0fa1775 ("tee: add TEE_IOCTL_PARAM_ATTR_TYPE_OBJREF") Signed-off-by: Arnd Bergmann arnd@arndb.de
The new interface showed up in 6.18, but I only came across this after that was released. Changing it now is technically an ABI change on architectures with unusual padding rules, so please consider carefully whether we want to do it this way or not.
Working around the ABI differences without an ABI change is possible, but adds a lot of complexity for compat handling.
This is currently only used by the recently introduced qcomtee backend driver. So it's only used on a few arm64 Qualcomm platforms right now.
I think we should take this patch, but let's hear what others think.
Thanks, Jens
I agree. We should take this patch. As noted, there are not many clients relying on it yet, so updating the userspace should be straightforward.
Best Regards, Amir
include/uapi/linux/tee.h | 1 + 1 file changed, 1 insertion(+)
diff --git a/include/uapi/linux/tee.h b/include/uapi/linux/tee.h index cab5cadca8ef..5203977ed35d 100644 --- a/include/uapi/linux/tee.h +++ b/include/uapi/linux/tee.h @@ -470,6 +470,7 @@ struct tee_ioctl_object_invoke_arg { __u32 op; __u32 ret; __u32 num_params;
__u32 :32; /* num_params tells the actual number of element in params */ struct tee_ioctl_param params[];};
2.39.5
On Mon, Dec 08, 2025 at 04:24:17PM +1100, Amirreza Zarrabi wrote:
Hi,
On 12/5/2025 12:27 AM, Jens Wiklander wrote:
Hi,
On Thu, Dec 4, 2025 at 11:17 AM Arnd Bergmann arnd@kernel.org wrote:
From: Arnd Bergmann arnd@arndb.de
The tee_ioctl_object_invoke_arg structure has padding on some architectures but not on x86-32 and a few others:
include/linux/tee.h:474:32: error: padding struct to align 'params' [-Werror=padded]
I expect that all current users of this are on architectures that do have implicit padding here (arm64, arm, x86, riscv), so make the padding explicit in order to avoid surprises if this later gets used elsewhere.
Fixes: d5b8b0fa1775 ("tee: add TEE_IOCTL_PARAM_ATTR_TYPE_OBJREF") Signed-off-by: Arnd Bergmann arnd@arndb.de
The new interface showed up in 6.18, but I only came across this after that was released. Changing it now is technically an ABI change on architectures with unusual padding rules, so please consider carefully whether we want to do it this way or not.
Working around the ABI differences without an ABI change is possible, but adds a lot of complexity for compat handling.
This is currently only used by the recently introduced qcomtee backend driver. So it's only used on a few arm64 Qualcomm platforms right now.
I think we should take this patch, but let's hear what others think.
Yeah since it's not an ABI issue on arm64 platforms where QTEE runs, so:
Reviewed-by: Sumit Garg sumit.garg@oss.qualcomm.com
Thanks, Jens
I agree. We should take this patch. As noted, there are not many clients relying on it yet, so updating the userspace should be straightforward.
You should rather test without any userspace library update to test it's not an ABI issue. Just for correctness sake, you can update the library too.
-Sumit
[...]
On 12/8/2025 5:50 PM, Sumit Garg via OP-TEE wrote:
On Mon, Dec 08, 2025 at 04:24:17PM +1100, Amirreza Zarrabi wrote:
Hi,
On 12/5/2025 12:27 AM, Jens Wiklander wrote:
Hi,
On Thu, Dec 4, 2025 at 11:17 AM Arnd Bergmann arnd@kernel.org wrote:
From: Arnd Bergmann arnd@arndb.de
The tee_ioctl_object_invoke_arg structure has padding on some architectures but not on x86-32 and a few others:
include/linux/tee.h:474:32: error: padding struct to align 'params' [-Werror=padded]
I expect that all current users of this are on architectures that do have implicit padding here (arm64, arm, x86, riscv), so make the padding explicit in order to avoid surprises if this later gets used elsewhere.
Fixes: d5b8b0fa1775 ("tee: add TEE_IOCTL_PARAM_ATTR_TYPE_OBJREF") Signed-off-by: Arnd Bergmann arnd@arndb.de
The new interface showed up in 6.18, but I only came across this after that was released. Changing it now is technically an ABI change on architectures with unusual padding rules, so please consider carefully whether we want to do it this way or not.
Working around the ABI differences without an ABI change is possible, but adds a lot of complexity for compat handling.
This is currently only used by the recently introduced qcomtee backend driver. So it's only used on a few arm64 Qualcomm platforms right now.
I think we should take this patch, but let's hear what others think.
Yeah since it's not an ABI issue on arm64 platforms where QTEE runs, so:
Reviewed-by: Sumit Garg sumit.garg@oss.qualcomm.com
Thanks, Jens
I agree. We should take this patch. As noted, there are not many clients relying on it yet, so updating the userspace should be straightforward.
You should rather test without any userspace library update to test it's not an ABI issue. Just for correctness sake, you can update the library too.
I'll take the time to test it at some point this week both with and without updating the library ABI.
Regards, Harshal
-Sumit
[...]
Hi,
On 12/8/2025 11:54 PM, Harshal Dev wrote:
On 12/8/2025 5:50 PM, Sumit Garg via OP-TEE wrote:
On Mon, Dec 08, 2025 at 04:24:17PM +1100, Amirreza Zarrabi wrote:
Hi,
On 12/5/2025 12:27 AM, Jens Wiklander wrote:
Hi,
On Thu, Dec 4, 2025 at 11:17 AM Arnd Bergmann arnd@kernel.org wrote:
From: Arnd Bergmann arnd@arndb.de
The tee_ioctl_object_invoke_arg structure has padding on some architectures but not on x86-32 and a few others:
include/linux/tee.h:474:32: error: padding struct to align 'params' [-Werror=padded]
I expect that all current users of this are on architectures that do have implicit padding here (arm64, arm, x86, riscv), so make the padding explicit in order to avoid surprises if this later gets used elsewhere.
Fixes: d5b8b0fa1775 ("tee: add TEE_IOCTL_PARAM_ATTR_TYPE_OBJREF") Signed-off-by: Arnd Bergmann arnd@arndb.de
The new interface showed up in 6.18, but I only came across this after that was released. Changing it now is technically an ABI change on architectures with unusual padding rules, so please consider carefully whether we want to do it this way or not.
Working around the ABI differences without an ABI change is possible, but adds a lot of complexity for compat handling.
This is currently only used by the recently introduced qcomtee backend driver. So it's only used on a few arm64 Qualcomm platforms right now.
I think we should take this patch, but let's hear what others think.
Yeah since it's not an ABI issue on arm64 platforms where QTEE runs, so:
Reviewed-by: Sumit Garg sumit.garg@oss.qualcomm.com
Thanks, Jens
I agree. We should take this patch. As noted, there are not many clients relying on it yet, so updating the userspace should be straightforward.
You should rather test without any userspace library update to test it's not an ABI issue. Just for correctness sake, you can update the library too.
I'll take the time to test it at some point this week both with and without updating the library ABI.
Summit, that was the plan from the beginning, that's why I did not add "Reviewed-by:" in the first place. Thanks Harshal for volunteering.
Best regards, Amir
Regards, Harshal
-Sumit
[...]
op-tee@lists.trustedfirmware.org
-
Amirreza Zarrabi -
Arnd Bergmann -
Arnd Bergmann -
Harshal Dev -
Jens Wiklander -
Sumit Garg