UFM Background

UFMs come in different flavors and styles that vary from device to device. "Firmware" generally refers to some form of software that runs on a device and is often packaged up as a binary payload. However, many things that aren't always called "firmware", such as EEPROM images, CPU microcode, flash based configuration, and more, are all just as important here. Take for example a hard drive. While it is a field replaceable unit (FRU), it also contains some amount of firmware that manages the drive which can be updated independently of replacing the drive.

The motherboard often has a UFM in the form of the BIOS or UEFI. The Lights Out Management controller on a system has a UFM, which is usually the entire system image. CPUs also have a UFM in the form of microcode.

An important property of a UFM is that it is a persistent part of the device itself. For example, many WiFi device drivers are required to send a binary blob of firmware to the device after every reset. Because these images are not persistent parts of the device and must be upgraded by either changing the device driver or related system files, we do not consider these UFMs.

There are also devices that have firmware which is a part of the device, but may not be upgradable from the running OS. This may be because the vendor doesn't have tooling to upgrade the image or because the firmware image itself cannot be upgraded in the field at all. For example, a YubiKey has a firmware image that's burned into it in the factory, but there is no way to change the firmware on it short of replacing the device in its entirety. However, because these images are a permanent and persistent part of the device, we also consider them a UFM.

Images and Slots

A device that supports UFMs is made up of one or more distinct firmware images. Each image has its own unique purpose. For example, a motherboard may have both a BIOS and a CPLD image, each of which has independent firmware revisions.

A given image may have a number of slots. A slot represents a particular version of the image. Only one slot is considered the active slot. It represents the currently running version of the image. Devices support multiple slots so that an image can be downloaded to an inactive slot without risking damage to the active slot. This ensures that a power-loss or failure halfway through writing to a slot doesn't leave the device with corrupted firmware.

The various entry points are designed such that all a driver has to do is provide information about the image and its slots to the kernel, it does not have to wrangle with how that is marshalled to users and the appearance of those structures.

Registering with the UFM Subsystem

During a device driver's attach(9E) entry point, a device driver should register with the UFM subsystem by filling out a UFM operations vector and then calling ddi_ufm_init(9F). The driver may pass in a value, usually a pointer to its soft state pointer, which it will then receive when its subsequent entry points are called.

Once the driver has finished initializing, it must call ddi_ufm_update(9F) to indicate that the driver is in a state where it's ready to receive calls to the entry points.

The various UFM entry points may be called from an arbitrary kernel context. However, they will only ever be called from a single thread at a given time.

UFM operations vector

The UFM operations vector is a structure that has the following members:

typedef struct ddi_ufm_ops {
	int (*ddi_ufm_op_nimages)(ddi_ufm_handle_t *uhp, void *drv_arg,
	    uint_t *nimgp);
	int (*ddi_ufm_op_fill_image)(ddi_ufm_handle_t *uhp, void *drv_arg,
            uint_t imgno, ddi_ufm_image_t *imgp);
	int (*ddi_ufm_op_fill_slot)(ddi_ufm_handle_t *uhp, void *drv_arg,
            int imgno, ddi_ufm_image_t *img, uint_t slotno,
	    ddi_ufm_slot_t *slotp);
	int (*ddi_ufm_op_getcaps)(ddi_ufm_handle_t *uhp, void *drv_arg,
	    ddi_ufm_cap_t *caps);
	int (*ddi_ufm_op_readimg)(ddi_ufm_handle_t *uhp, void *drv_arg,
	    uint_t imgno, uint_t slotno, uint64_t len, uint64_t offset,
	    void *buf, uint64_t *nreadp);
} ddi_ufm_ops_t;

The ddi_ufm_op_nimages() and ddi_ufm_op_readimg() entry points are optional. If a device only has a single image, then there is no requirement to implement the ddi_ufm_op_nimages() entry point and it may be set to NULL. The system will assume that there is only a single image.

Slots and images are numbered starting at zero. If a driver indicates support for multiple images, through the ddi_ufm_op_nimages() entry point, or slots, by using the ddi_ufm_image_set_nslots(9F) function in the ddi_fum_op_fill_image() callback then the images or slots will be numbered sequentially going from 0 to the number of images or slots minus one. These values will be passed to the various entry points to indicate which image and slot the system is interested in. It is up to the driver to maintain a consistent view of the images and slots for a given UFM.

ddi_ufm_op_nimages()

The ddi_ufm_op_nimages() entry point is an optional entry point that answers the question of how many different, distinct firmware images are present on the device. Once the driver determines how many are present, it should set the value in nimgp to the determined value.

It is legal for a device to pass in zero for this value, which indicates that there are none present.

Upon successful completion, the driver should return 0. Otherwise, the driver should return the appropriate error number. For a full list of error numbers, see Intro(2). Common values are:

ddi_ufm_op_fill_image()

The ddi_ufm_op_fill_image() entry point is used to fill in information about a given image. The value in imgno is used to indicate which image the system is asking to fill information about. If the driver does not recognize the image ID in imgno then it should return an error.

The ddi_ufm_image_t structure passed in imgp is opaque. To fill in information about the image, the driver should call the functions described in ddi_ufm_image(9F).

The driver must call the ddi_ufm_image_set_desc(9F) function to set a description of the image which indicates its purpose. This should be a human-readable string. In addition, the driver must call the ddi_ufm_image_set_nslots(9F) function to indicate the number of slots that the device supports for that particular firmware image. The driver may also set any ancillary data that it deems may be useful with the ddi_ufm_image_set_misc(9F) function. This function takes an nvlist, allowing the driver to set arbitrary keys and values.

Once the driver has finished setting all of the information about the image then the driver should return 0. Otherwise, the driver should return the appropriate error number. For a full list of error numbers, see Intro(2). Common values are:

ddi_ufm_op_fill_slot()

The ddi_ufm_op_fill_slot() function is used to fill in information about a specific slot for a specific image. The value in imgno indicates the image the system wants slot information for and the value in slotno indicates which slot of that image the system is interested in. If the device driver does not recognize the value in either or imgno or slotno, then it should return an error.

The ddi_ufm_slot_t structure passed in slotp is opaque. To fill in information about the image the driver should call the functions described in ddi_ufm_slot(9F).

The driver should call the ddi_ufm_slot_set_version(9F) function to indicate the version of the UFM. The version is a device-specific character string. It should contain the current version of the UFM as a human can understand it and it should try to match the format used by device vendor.

The ddi_ufm_slot_set_attrs(9F) function should be used to set the attributes of the UFM slot. These attributes include the following enumeration values:

DDI_UFM_ATTR_READABLE

The DDI_UFM_ATTR_READABLE attribute indicates that the firmware image in the specified slot may be read, even if the device driver does not currently support such functionality.

DDI_UFM_ATTR_WRITEABLE

The DDI_UFM_ATTR_WRITEABLE attribute indicates that the firmware image in the specified slot may be updated, even if the driver does not currently support such functionality.

DDI_UFM_ATTR_ACTIVE

The DDI_UFM_ATTR_ACTIVE attribute indicates that the firmware image in the specified slot is the active (i.e. currently running) firmware. Only one slot should be marked active.

DDI_UFM_ATTR_EMPTY

The DDI_UFM_ATTR_EMPTY attribute indicates that the specified slot does not currently contain any firmware image.

If the driver supports the ddi_ufm_op_readimg() entry point, then the driver should attempt to determine the size in bytes of the image in the slot and indicate that by calling the ddi_ufm_slot_set_imgsize(9F) function.

Finally, if there are any device-specific key-value pairs that form useful, ancillary data, then the driver should assemble an nvlist and pass it to the ddi_ufm_slot_set_misc(9F) function.

Once the driver has finished setting all of the information about the slot then the driver should return 0. Otherwise, the driver should return the appropriate error number. For a full list of error numbers, see Intro(2). Common values are:

ddi_ufm_op_getcaps()

The ddi_ufm_op_getcaps() function is used to indicate which DDI UFM capabilities are supported by this driver instance. Currently, all UFM-capable drivers are required to implement the DDI_UFM_CAP_REPORT capability. The following capabilities are supported and the drivers should return a bitwise-inclusive-OR of the following values:

The driver should indicate the supported capabilities by setting the value in the caps parameter. Once the driver has populated caps with an appropriate value, then the driver should return 0. Otherwise, the driver should return the appropriate error number. For a full list of error numbers, see Intro(2). Common values are:

ddi_ufm_op_readimg()

The ddi_ufm_op_readimg() is an optional entry point that allows the system to read a binary firmware payload from the device. The driver should read the firmware payload indicated by both imgno and slotno. The driver should check to make sure that the region requested, starting at offset bytes into the image and len bytes long is valid for the image and if not, return the error EINVAL. Data from the device should be copied into buf and the number of bytes successfully read should be placed into nreadp.

Upon successfully reading this data, the driver should return 0. Otherwise the driver should return the appropriate error number. For a full list of error numbers, see Intro(2). Common values are:

Caching and Updates

The system will fetch firmware and slot information on an as-needed basis. Once it obtains some information, it may end up caching this information on behalf of the driver. Whenever the driver believes that something could have changed then the driver must call ddi_ufm_update(9F). The driver does not need to know for certain that something has changed. For example, after a device reset or firmware upgrade, the driver doesn't need to check if the firmware revision changed at all, it can simply call ddi_ufm_update(9F).

Locking

All UFM operations on a single UFM handle will always be run serially. However, the device driver may still need to apply adequate locking to its structure members as other entry points may be called on the device in parallel, which could access the same data structures and try to communicate with the device.

Unregistering from the UFM subsystem

When a device driver is detached, it should unregister from the UFM subsystem. To do so, the driver should call ddi_ufm_fini(9F). By the time this function returns, the driver is guaranteed that no UFM entry points will be called. However, if there are outstanding UFM related activity, the function will block until it is terminated.

ioctl Interface

Userland consumers can access UFM information via a set of ioctls that are implemented by the ufm(4D) driver.