Version 5 (modified by alain, 4 years ago) (diff)


FBF device API

A) General principles

This device provide access to an external graphic display, that is seen by the kernel as a fixed size frame buffer, mapped in the kernel address space. The only pixel encoding type in the current Almos-mkh implementation is one byte per pixel (256 levels of gray).

It defines a first user API, used by the user-level system calls, and implementing a simple, kernel controlled, windows manager. This windows manager allows any process to create and use for display one (or several) window(s). Each window defines a private buffer, dynamically allocated in user space, that can be directly accessed (for read and write) by the owner process, without system call.

These windows can overlap. They can be moved in the frame buffer, or they can be resized, using system calls. The refresh of a window in the FBF must be explicitly required by the owner process. they can overlap other windows,.

A window must be entirely contained in the frame buffer.

To avoid contention, the window descriptor, and the associated user buffer are not allocated in the cluster containing the FBF chdev, but are distributed: each window is allocated in the cluster defined by the thread that required the window creation.

Each window has a single process owner, but all the windows are registered in the FBF chdev in a windows_tbl[] array, indexed by the window identifier (wid). Each array entry contains an extended pointer on the window descriptor. All windows are also registered in a trans-cluster xlist, defining the overlapping order (last window in xlist has the highest priority).

To access the various drivers, this FBF device defines a lower-level driver API, that must be implemented by all drivers.

All FBF access functions are defined in the dev_fbf.c et dev_fbf.h files.

B) Initialisation

The dev_fbf_init() function makes two initializations :

  • it completes the FBF specific fields of the chdev descriptor.
  • It calls the driver initialisation function, to initialise the specific hardware device.

It must be called by a local thread.

C) User API

All functions defined in this user API do NOT use the the FBF device waiting queue, the associated device thread and the FBF IRQ, as the client thread does NOT deschedules,

C.1) dev_fbf_get_config()

This function implements the fbf_get_config() syscall. It returns the FBF number of lines, the number of pixels per line, and the pixel encoding type. It can be called by a client thread running in any cluster. It does not access the hardware, as the size and type have been registered in the chdev descriptor extension by the dev_fbf_init() function.

C.2) dev_fbf_create_window()

This blocking function implements the fbf_create_window() sys-call. It registers a new window in the windows_tbl[] array, and in the windows list, rooted in FBF device descriptor. It registers in the reference VSL an ANON vseg, that will be mapped in local cluster. The window index <wid> is dynamically allocated. The owner process is the calling process. The FBF window is defined by the <nlines>, <npixels>, <l_min>, <p_min> arguments. The created vseg base address in user space is returned in the <user_base> argument. It can be called by any thread running in any cluster.

As the vseg associated to the window is not directly mapped to the frame buffer, the owner process can read and write in the window buffer without syscall. As for any other vseg, the physical memory is allocated on demand, at each page fault.

C.3) dev_fbf_refresh_window()

This blocking function implements the fbf_refresh_window() sys-call. It allows the owner process to signal the windows manager that some lines of a window identified by the <wid>, <line_min>, and <line_max> arguments have been modified, and must be refreshed in the FBF. It scans all the registered FBF windows to respect the overlap order defined by the windows xlist. It can be called by any thread running in any cluster.

C.4) dev_fbf_move_window()

This blocking function implements the fbf_move_window() sys-call. It moves a window identified by the <wid> argument to a new position in the FBF, defined by the <l_min> and <p_min> arguments. It can be called by any thread running in any cluster.

C.5) dev_fbf_resize_window()

This blocking function implements the fbf_resize_window() syscall. It changes the <width> and <height> of a window identified by the <wid> argument. It updates the size of the associated vseg, but does not change the vseg "base". When the new window buffer is larger than the existing one, this buffer is 0 filled. It can be called by any thread running in any cluster.

C.6) dev_fbf_delete_window()

This function implements the fbf_delete_window() sys-call to delete a FBF window. It releases all memory allocated for the window buffer and for the window descriptor. It can be called by any thread running in any cluster.

C.7) dev_fbf_move_data()

This function implements the deprecated fbf_read() and fbf_write() sys-calls. This function allows an user application to directly access the Frame Buffer, without using any intermediate window. It calls directly the driver to synchronously move <npixels> between an user <buffer> and the FBF, starting at a given <offset> in the FBF. The transfer direction is defined by the <is_write> argument.

D) Driver API

The FBF device defines defines four command types to access FBF driver(s) :

  • FBF_DRIVER_KERNEL_WRITE : move pixels from a kernel window to the FBF.
  • FBF_DRIVER_KERNEL_READ : move pixels from the FBF to a kernel window.
  • FBF_DRIVER_USER_WRITE : move bytes from an user buffer to the FBF.
  • FBF_DRIVER_USER_READ : move bytes from the FBF to an user buffer.

These four commands use the three following arguments, that must be registered, with the command type, in the fbf_command_t structure embedded in the client thread descriptor :

  • npixels : number of pixels to be moved.
  • buffer : pointer on buffer in (can be in user space or in kernel space).
  • offset : offset in FBF (number of pixels).