FreeBSD Manual Pages
FIRMWARE(9) Kernel Developer's Manual FIRMWARE(9) NAME firmware_register, firmware_unregister, firmware_get, firmware_get_flags, firmware_put -- firmware image loading and management SYNOPSIS #include <sys/param.h> #include <sys/systm.h> #include <sys/linker.h> #include <sys/firmware.h> struct firmware { const char *name; /* system-wide name */ const void *data; /* location of image */ size_t datasize; /* size of image in bytes */ unsigned int version; /* version of the image */ }; const struct firmware * firmware_register(const char *imagename, const void *data, size_t datasize, unsigned int version, const struct firmware *parent); int firmware_unregister(const char *imagename); const struct firmware * firmware_get(const char *imagename); const struct firmware * firmware_get_flags(const char *imagename, uint32_t flags); void firmware_put(const struct firmware *fp, int flags); DESCRIPTION The firmware abstraction provides a convenient interface for loading firmware images into the kernel, and for accessing such images from kernel components. A firmware image (or image for brevity) is an opaque block of data residing in kernel memory. It is associated to a unique imagename which constitutes a search key, and to an integer version number, which is also an opaque piece of information for the firmware subsystem. An image is registered with the firmware subsystem by calling the function firmware_register(), and unregistered by calling firmware_unregister(). These functions are usually (but not exclusively) called by specially crafted kernel modules that contain the firmware image. The modules can be statically compiled in the kernel, or loaded by /boot/loader, manually at runtime, or on demand by the firmware subsystem. Firmware binary files may also be loaded directly rather than embedded into kernel modules. Clients of the firmware subsystem can request access to a given image by calling the function firmware_get() with the imagename they want as an ar- gument, or by calling firmware_get_flags() with the imagename and flags they want as an arguments. If a matching image is not already registered, the firmware subsystem will try to load it using the mechanisms specified below (typically, a kernel module with firmware_register the same name as the image). API DESCRIPTION The kernel firmware_register firmware API is made of the following func- tions: firmware_register() registers with the kernel an image of size datasize lo- cated at address data, under the name imagename. The function returns NULL on error (e.g. because an image with the same name already exists, or the image table is full), or a const struct firmware * pointer to the image requested. firmware_unregister() tries to unregister the firmware image imagename from the system. The function is successful and returns 0 if there are no pend- ing references to the image, otherwise it does not unregister the image and returns EBUSY. firmware_get() and firmware_get_flags() return the requested firmware im- age. The flags argument may be set to FIRMWARE_GET_NOWARN to indicate that errors on firmware load or registration should only be logged in case of booverbose. If the image is not yet registered with the system, the func- tions try to load it. This involves the linker subsystem and disk access, so firmware_get() or firmware_get_flags() must not be called with any locks (except for Giant). Note also that if the firmware image is loaded from a filesystem it must already be mounted. In particular this means that it may be necessary to defer requests from a driver attach method unless it is known the root filesystem is already mounted. On success, firmware_get() and firmware_get_flags() return a pointer to the image description and increase the reference count for this image. On failure, the functions return NULL. firmware_put() drops a reference to a firmware image. The flags argument may be set to FIRMWARE_UNLOAD to indicate that firmware_put is free to re- claim resources associated with the firmware image if this is the last ref- erence. By default a firmware image will be deferred to a taskqueue(9) thread so the call may be done while holding a lock. In certain cases, such as on driver detach, this cannot be allowed. FIRMWARE LOADING VIA MODULES As mentioned before, any component of the system can register firmware im- ages at any time by simply calling firmware_register(). This is typically done when a module containing a firmware image is given control, whether compiled in, or preloaded by /boot/loader, or manually loaded with kldload(8). However, a system can implement additional mecha- nisms to bring these images into memory before calling firmware_register(). When firmware_get() or firmware_get_flags() does not find the requested im- age, it tries to load it using one of the available loading mechanisms. At the moment, there is only one, namely Loadable kernel modules. A firmware image named foo is looked up by trying to load the module named foo.ko, using the facilities described in kld(4). In particular, images are looked up in the directories specified by the sysctl variable kern.module_path which on most systems defaults to /boot/kernel;/boot/modules. Note that in case a module contains multiple images, the caller should first request a firmware_get() or firmware_get_flags() for the first image contained in the module, followed by requests for the other images. BUILDING FIRMWARE LOADABLE MODULES A firmware module is built by embedding the firmware image into a suitable loadable kernel module that calls firmware_register() on loading, and firmware_unregister() on unloading. Various system scripts and makefiles let you build a module by simply writ- ing a Makefile with the following entries: KMOD= imagename FIRMWS= image_file:imagename[:version] .include <bsd.kmod.mk> where KMOD is the basename of the module; FIRMWS is a list of colon-sepa- rated tuples indicating the image_file's to be embedded in the module, the imagename and version of each firmware image. If you need to embed firmware images into a system, you should write appro- priate entries in the <files.arch> or <files> file, e.g. this example is from sys/conf/files iwn1000fw.c optional iwn1000fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn1000.fw:iwn1000fw -miwn1000fw -c${.TARGET}" \ no-ctfconvert no-implicit-rule before-depend local \ clean "iwn1000fw.c" # # NB: ld encodes the path in the binary symbols generated for the # firmware image so link the file to the object directory to # get known values for reference in the _fw.c file. # iwn1000fw.fwo optional iwn1000fw | iwnfw \ dependency "iwn1000.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn1000fw.fwo" Firmware was previously committed to the source tree as uuencoded files, but this is no longer required; the binary firmware file should be commit- ted to the tree as provided by the vendor. Note that generating the firmware modules in this way requires the avail- ability of the following tools: awk(1), make(1), the compiler and the linker. LOADING BINARY FIRMWARE FILES Binary Firmware Format Binary firmware files can also be loaded, either from /boot/loader, or when firmware_get cannot find the registered firmware from a kernel module. Bi- nary firmware files are raw binary files that the creator of the firmware made. They offer an easier way to load firmware, but one that lacks the full flexibility and generality of kernel modules with the following re- strictions: * Binary firmware files only hold one set of firmware. * They do not offer kernel module dependencies to ensure they are loaded automatically by the boot loader. * They cannot be compiled into the kernel. * The imagename is identical to the full path name used to load the mod- ule. * The version number is assumed to be zero. Loading from /boot/loader Binary firmware files may be loaded either from the command line with "load -t firmware /boot/firmware/filename" or using the loader.conf(5) mechanism to load modules with a type of "firmware" For example wififw_load="YES" wififw_name="/boot/firmware/wifi2034_fw.bin" wififw_type="firmware" On demand loading from firmware_get If no kernel module with an embedded firmware image named imagename is loaded, then imagename will be appended to the module path (by default /boot/firmware/) and if that file exists, it will be loaded and registered using firmware_register using the full path to the filename as imagename. Searching for imagename firmware_get uses the following algorithm to find firmware images: * If an existing firmware image is registered for imagename, that image is returned. * If imagename matches the trailing subpath of a registered image with a full path, that image is returned. * The kernel linker searches for a kernel module named imagename. If a kernel module is found, it is loaded, and the list of registered firmware images is searched again. If a match is found, the matching image is returned. * The kernel searches for a file named imagename in the firmware image path (by default /boot/firmware/). If that file exists and can be read, it contents are registered as a firmware image with the full path as the imagename and that firmware is returned. Currently, there is an 8MB limit on the size of the firmware image. This can be changed by by the sysctl variable debug.max_firmware_size. SEE ALSO kld(4), module(9) /boot/firmware /usr/share/examples/kld/firmware HISTORY The firmware system was introduced in FreeBSD 6.1. Binary firmware loading was introduced in FreeBSD 15.0. AUTHORS This manual page was written by Max Laier <mlaier@FreeBSD.org>. FreeBSD ports 15.quarterly January 25, 2024 FIRMWARE(9)
NAME | SYNOPSIS | DESCRIPTION | API DESCRIPTION | FIRMWARE LOADING VIA MODULES | BUILDING FIRMWARE LOADABLE MODULES | LOADING BINARY FIRMWARE FILES | SEE ALSO | HISTORY | AUTHORS
Want to link to this manual page? Use this URL:
<https://man.freebsd.org/cgi/man.cgi?query=firmware&sektion=9&manpath=FreeBSD+15.1-RELEASE+and+Ports.quarterly>