drivers: dai: add ability to use dai.h from user threads #99811

kv2019i · 2025-11-21T13:29:13Z

Add user-space support to the dai.h interface. No functional impact to builds when CONFIG_USERSPACE is not set.

kv2019i · 2025-11-21T13:31:01Z

Draft for now, thesofproject/sof#10373 shall go in first. I'll rebase on top of that before converting to ready for review.

lyakh · 2025-11-21T16:05:40Z

drivers/dai/dai_handlers.c

+	 * merge with Adrian's PR to add bespoke_cfg size */
+	K_OOPS(K_SYSCALL_MEMORY_READ(bespoke_cfg, CONFIG_MMU_PAGE_SIZE));
+
+	/* TODO: shall a kernel copy made of cfg */


sorry, not sure what this comment means?

@lyakh So basicly, can we trust to pass an object that sits in user-space accessible memory to a kernel function. In Linux this is a no-no as another user-space thread could be active (on another CPU) and modify the data while kernel is accessing it (on another).

On a resource constrained device, this is a bit harder pill to swallow as some of these objects are rather big. But I'm thinking it we want to merge this in Zephyr, we need to do a copy of all input data to kernel, validate and run the function. In SOF side (the application using Zephyr), we could optimize this as we know there will not be other user-space threads (enforced by system design).

@kv2019i ah, so the typo is "a kernel copy be made" - that's what I couldn't guess, sorry :-)

You need to deeply copy this (nothing referring to user mode accessible memory), then validate it. This will be rather troublesome with a void * at an API level.

Ack @teburd , will at deep copies to next version. @abonislawski is added a size field in a separate PR, so that allows to handle the bespoke data objects in a sane fashion. I'll be adding a new variant of get_properties() that can be used from user-space.

softwarecki · 2025-11-25T12:42:59Z

Can we wrap these changes into a more detailed config than CONFIG_USERSPACE? I'm not entirely sure what the idea behind these changes is, but I suspect it's about running the entire pipeline in userspace. In that case, I suggest something like CONFIG_USERSPACE_PIPELINE.

kv2019i · 2025-11-25T18:00:31Z

@softwarecki

Can we wrap these changes into a more detailed config than CONFIG_USERSPACE? I'm not entirely sure what the idea behind these changes is, but I suspect it's about running the entire pipeline in userspace. In that case, I suggest something like CONFIG_USERSPACE_PIPELINE.

Good question. This is a generic capability in Zephyr, so it's not really tied to any particular use (and not related to SOF). Currently user-space is a boolean build option in Zephyr, if you enable user-space, all interfaces that have a syscall interface (semaphores, cache.h, mutexes, now dai.h) , are exposed. I'd prefer to use some standard Zephyr option if user wants to build Zephyr with a subset of all possible syscalls-enables interfaces (e.g. have semaphore syscall impelemntation built in, but noth dma.h or dai.h). Let me study this a bit (and please chime in if anyone knows how to do this the Zephyr way).

Functionally this is not a problem. Even if you enable Zephyr CONFIG_USERSPACE, but you call dai.h from kernel, nothing changes. And user threads cannot access dai.h unless permission is explicitly given, so just having dai.h syscalls built-in does not mean user threads can use this functionality. But this does add code to the binary, so a valid question for sure. Same problem should apply for other interfaces as well. Let's see if there's a solution already.

kv2019i

Thanks all to reviewers so far!

kv2019i · 2025-11-25T17:53:19Z

drivers/dai/dai_handlers.c

+	 * merge with Adrian's PR to add bespoke_cfg size */
+	K_OOPS(K_SYSCALL_MEMORY_READ(bespoke_cfg, CONFIG_MMU_PAGE_SIZE));
+
+	/* TODO: shall a kernel copy made of cfg */


Ack @teburd , will at deep copies to next version. @abonislawski is added a size field in a separate PR, so that allows to handle the bespoke data objects in a sane fashion. I'll be adding a new variant of get_properties() that can be used from user-space.

kv2019i · 2025-12-02T16:59:12Z

V2 pushed:

rebased on top of interface change from @abonislawski for set_config... now can do a more safe version of config_set
added copy of input blobs to kernel memory before validation/use as per feedback from @teburd
added new logic to leave the DAI user-space syscall handlers from build with a new Kconfig based on feedback from @softwarecki
marking as ready for review now

Add user-space support to the dai.h interface. No functional impact to builds when CONFIG_USERSPACE is not set. Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>

Add a variant of get_properties() method that writes the properties to a caller provided pointer. Unlike the old variant, this copy variant can be exported to user-space in a safe way. Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>

The DAI interface is not used from user-space in all configurations where Zephyr user-space is enabled, so it is beneficial to have a build option to contorl whether the DAI syscalls are included or not. Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>

Add support for new get_properties_copy() method. This allows to use ssp driver from user-space threads. Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>

kv2019i · 2025-12-02T17:53:26Z

V3:

fix compliance check issues in V2

sonarqubecloud · 2025-12-02T18:19:20Z

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

kv2019i · 2025-12-04T10:46:58Z

@lyakh @softwarecki @lgirdwood @nashif can you take a look?

softwarecki

Good job wrapping zephyr_syscall_header in if(CONFIG_DAI_USERSPACE).
This was missing in thesofproject/sof#10341, so it's great to see it addressed here.

Copilot

Pull request overview

This pull request adds user-space support to the DAI (Digital Audio Interface) driver interface by converting inline functions to syscalls and adding appropriate validation handlers.

Key changes:

Converted 10 DAI API functions to syscalls with __syscall declarations and z_impl_* implementations
Added new dai_get_properties_copy API function for user-space safe property retrieval
Implemented syscall validation handlers in dai_handlers.c with memory access checks and parameter validation

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
`include/zephyr/drivers/dai.h`	Converts inline functions to syscalls, adds `get_properties_copy` API and callback, includes syscall header
`drivers/dai/dai_handlers.c`	New file implementing syscall verification handlers with memory validation and bespoke config size limits
`drivers/dai/intel/ssp/ssp.c`	Implements `dai_ssp_get_properties_copy` callback for SSP driver
`drivers/dai/Kconfig`	Adds `DAI_USERSPACE` configuration option dependent on `USERSPACE`
`drivers/dai/CMakeLists.txt`	Conditionally includes syscall generation and handlers compilation

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-12-09T15:34:31Z

drivers/dai/intel/ssp/ssp.c

+	if (!prop) {
+		return -EINVAL;
+	}
+


Missing NULL check for kernel_prop before dereferencing it in memcpy. While dai_ssp_get_properties currently never returns NULL, defensive programming requires checking the return value. If it were NULL, line 2560 would cause a crash. Add a check: if (!kernel_prop) { return -ENODATA; } after line 2554.

Suggested change

if (!kernel_prop) {

return -ENODATA;

}

Copilot · 2025-12-09T15:34:32Z

include/zephyr/drivers/dai.h

+ * @brief Fetch properties of a DAI driver
+ *
+ * @param dev Pointer to the device structure for the driver instance
+ * @param dir Stream direction: RX or TX as defined by DAI_DIR_*
+ * @param stream_id Stream id: some drivers may have stream specific
+ *        properties, this id specifies the stream.
+ * @param dst address where to write properties to
+ * @retval Zero on success


The documentation for dai_get_properties_copy is incomplete. It should specify the expected error return codes (e.g., -EINVAL for invalid parameters, -ENODATA if properties are unavailable). Additionally, it should clarify the relationship between this function and dai_get_properties, explaining that this is the user-space-safe variant.

Suggested change

* @brief Fetch properties of a DAI driver

*

* @param dev Pointer to the device structure for the driver instance

* @param dir Stream direction: RX or TX as defined by DAI_DIR_*

* @param stream_id Stream id: some drivers may have stream specific

* properties, this id specifies the stream.

* @param dst address where to write properties to

* @retval Zero on success

* @brief Fetch properties of a DAI driver (user-space-safe variant)

*

* This function is the user-space-safe variant of @ref dai_get_properties.

* It copies the properties into the provided destination buffer, allowing

* safe access from user space.

*

* @param dev Pointer to the device structure for the driver instance

* @param dir Stream direction: RX or TX as defined by DAI_DIR_*

* @param stream_id Stream id: some drivers may have stream specific

* properties, this id specifies the stream.

* @param dst Address where to write properties to

*

* @retval 0 On success.

* @retval -EINVAL Invalid parameters (e.g., null pointer, invalid stream id).

* @retval -ENODATA Properties are unavailable for the specified stream.

Copilot · 2025-12-09T15:34:32Z

include/zephyr/drivers/dai.h

+						 struct dai_properties *dst)
+{
+	const struct dai_driver_api *api = (const struct dai_driver_api *)dev->api;
+


Missing NULL check for the get_properties_copy API callback before calling it. If a driver doesn't implement this callback (like alh, dmic, hda, esai, micfil, and sai drivers), this will cause a NULL pointer dereference. Add a check: if (!api->get_properties_copy) { return -ENOSYS; } before the function call.

Suggested change

if (!api->get_properties_copy) {

return -ENOSYS;

}

Copilot · 2025-12-09T15:34:33Z

drivers/dai/dai_handlers.c

+		return -EINVAL;
+	}
+
+	K_OOPS(K_SYSCALL_DRIVER_DAI(dev, config_set));


Missing validation for the cfg parameter. The syscall handler should validate that userspace has read access to the dai_config structure before passing it to the kernel implementation. Add K_OOPS(K_SYSCALL_MEMORY_READ(cfg, sizeof(*cfg))); after the driver validation check.

Suggested change

K_OOPS(K_SYSCALL_DRIVER_DAI(dev, config_set));

K_OOPS(K_SYSCALL_DRIVER_DAI(dev, config_set));

K_OOPS(K_SYSCALL_MEMORY_READ(cfg, sizeof(*cfg)));

Copilot · 2025-12-09T15:34:33Z

drivers/dai/dai_handlers.c

+	K_OOPS(k_usermode_from_copy(bespoke_cfg_kernel, bespoke_cfg, size));
+
+	return z_impl_dai_config_set(dev, cfg, bespoke_cfg_kernel, size);
+}


Missing validation for NULL or zero-size bespoke_cfg. If size is 0 and bespoke_cfg is NULL, the k_usermode_from_copy will fail. Consider checking if size is 0 and handling it appropriately, potentially skipping the copy when size == 0. This pattern should also be applied to z_vrfy_dai_config_update.

Suggested change

K_OOPS(k_usermode_from_copy(bespoke_cfg_kernel, bespoke_cfg, size));

return z_impl_dai_config_set(dev, cfg, bespoke_cfg_kernel, size);

}

if (size > 0) {

K_OOPS(k_usermode_from_copy(bespoke_cfg_kernel, bespoke_cfg, size));

return z_impl_dai_config_set(dev, cfg, bespoke_cfg_kernel, size);

} else {

/* No bespoke config to copy, pass NULL */

return z_impl_dai_config_set(dev, cfg, NULL, 0);

}

kv2019i requested review from lgirdwood, lyakh and teburd November 21, 2025 13:29

kv2019i requested review from abonislawski and softwarecki November 21, 2025 13:31

lyakh reviewed Nov 21, 2025

View reviewed changes

kv2019i commented Nov 25, 2025

View reviewed changes

kv2019i force-pushed the 202511-userspace-dai branch from c16628a to 716366c Compare December 2, 2025 16:56

kv2019i marked this pull request as ready for review December 2, 2025 16:59

kv2019i requested review from dbaluta and dcpleung December 2, 2025 16:59

zephyrbot added the area: DAI label Dec 2, 2025

zephyrbot requested review from LaurentiuM1234 and iuliana-prodan December 2, 2025 17:00

zephyrbot assigned kv2019i and LaurentiuM1234 Dec 2, 2025

kv2019i added 4 commits December 2, 2025 19:52

drivers: dai: add ability to use dai.h from user threads

de7218c

Add user-space support to the dai.h interface. No functional impact to builds when CONFIG_USERSPACE is not set. Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>

drivers: dai: intel: ssp: add get_properties_copy support

3c5607b

Add support for new get_properties_copy() method. This allows to use ssp driver from user-space threads. Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>

kv2019i force-pushed the 202511-userspace-dai branch from 716366c to 3c5607b Compare December 2, 2025 17:52

kv2019i added the platform: Intel ADSP Intel Audio platforms label Dec 3, 2025

kv2019i requested a review from nashif December 4, 2025 10:46

kv2019i mentioned this pull request Dec 9, 2025

userspace: add support to use DAIs from user threads thesofproject/sof#10421

Open

lgirdwood approved these changes Dec 9, 2025

View reviewed changes

softwarecki approved these changes Dec 9, 2025

View reviewed changes

kartben requested a review from Copilot December 9, 2025 15:28

Copilot started reviewing on behalf of kartben December 9, 2025 15:29 View session

Copilot AI reviewed Dec 9, 2025

View reviewed changes

fabiobaltieri approved these changes Dec 10, 2025

View reviewed changes

fabiobaltieri merged commit 3b7b2ae into zephyrproject-rtos:main Dec 10, 2025
34 checks passed

+	if (!kernel_prop) {
+		return -ENODATA;
+	}

+	if (!api->get_properties_copy) {
+		return -ENOSYS;
+	}

	K_OOPS(K_SYSCALL_DRIVER_DAI(dev, config_set));
	K_OOPS(K_SYSCALL_DRIVER_DAI(dev, config_set));
	K_OOPS(K_SYSCALL_MEMORY_READ(cfg, sizeof(*cfg)));

drivers: dai: add ability to use dai.h from user threads #99811

drivers: dai: add ability to use dai.h from user threads #99811

Conversation

kv2019i commented Nov 21, 2025

Uh oh!

kv2019i commented Nov 21, 2025

Uh oh!

lyakh Nov 21, 2025

Choose a reason for hiding this comment

Uh oh!

kv2019i Nov 24, 2025

Choose a reason for hiding this comment

Uh oh!

lyakh Nov 24, 2025

Choose a reason for hiding this comment

Uh oh!

teburd Nov 24, 2025

Choose a reason for hiding this comment

Uh oh!

kv2019i Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

softwarecki commented Nov 25, 2025

Uh oh!

kv2019i commented Nov 25, 2025

Uh oh!

kv2019i left a comment

Choose a reason for hiding this comment

Uh oh!

kv2019i Nov 25, 2025

Choose a reason for hiding this comment

Uh oh!

kv2019i commented Dec 2, 2025

Uh oh!

kv2019i commented Dec 2, 2025

Uh oh!

sonarqubecloud bot commented Dec 2, 2025

Quality Gate passed

Uh oh!

kv2019i commented Dec 4, 2025

Uh oh!

softwarecki left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 9, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

8 participants