Re: SS20 cg14 help

[Christian Corti <corti%informatik.uni-stuttgart.de@localhost>]

Are the following files (and more of this kind) of any interest, or is the 
information already known?

cgfourteen.c
------------
[...]
#pragma ident   "@(#)cgfourteen.c 1.112 96/09/24 SMI"

/*
 * Driver for Campus-II MBus framebuffer (CG14).
 * The CG14 is made up of the MDI, the VBC, the RAMDAC, the PCG, and various
 * amounts of VRAM.  Together, these make up the VSIMM.
 * The CG14 provides access to 24 bit frame buffers and compatibility modes
 * for 8-bit and 16-bit (8+8) frame buffer access. The main functions
 * of this driver are to provide:
 *
 *      a) mappings to the MDI control address space which contains the cursor
 *         color look up tables, the DAC etc.
 *      b) mappings to the frame buffer a.k.a the VRAM in 8 different modes
 *         of access.
 *      c) a cgthree emulation mode to provide backwards binary compatibility
 *         for Pixrect/SunView applications.
 */
[...]

tcx.c
-----
#ifdef  SYSV
#pragma ident   "@(#)tcx.c
#else
#if !defined(lint) && !defined(NOID)
static char sccsid[] = "@(#)tcx.c 1.1 94/10/31 SMI";
#endif
#endif

/*
 * Copyright 1993 1994, Sun Microsystems, Inc.
 */

/*
 * A24 24-bit framebuffer.
 */

/*
 * A24 theory of operation:
 *
 * The A24 is driven by mapping framebuffer memory into process address
 * space and writing to it directly.  There are some control registers
 * that are also accessed by mapping into process address space, but
 * access to these registers are normally limited to the window system
 * only.
 *
 * The A24 is a (mostly) stateless framebuffer.  This means that
 * most operations are atomic to the extent that if a process accessing
 * the framebuffer is interrupted, and then resumes later after the
 * framebuffer state has been altered, there are no bad effects.
 *
 * Different operations on the A24 are achieved by referencing different
 * address spaces.  There is a 24-bit address space where 32-bit writes
 * reference individual 24-bit pixels.  There is an 8-bit address space
 * where 8-bit writes reference the red channel of individual pixels
 * (and 32-bit writes reference four pixels at a time.)
 *
 * In addition, there are special address spaces:  The "stipple" address
 * space allows a 64-bit control word to be written containing a color (24
 * or 8 bit) and a 32-bit pixel mask.  This allows 32 pixels to be written
 * in a single operation.  The address of the 64-bit write determines the
 * starting pixel for the operation.  The "blit" address space is similar
 * to the "stipple" address space, except that the 64-bit control word
 * specifies a source pixel address and a pixel mask rather than a color
 * and a pixel mask.
 *
 * In addition, there are "raw 32", "raw stipple" and "raw blit" address
 * spaces.  These are identical to the 24-bit, stipple and blit address
 * spaces, except that they also access the 2 control planes which determine
 * how underlying pixels are interpreted by the dacs (modes are: 8-bit
 * indexed, 24-bit indexed, 24-bit gamma-corrected and 24-bit raw).
 *
 * Virtual addresses and lengths for mmap() are defined in tcxreg.h.  Since
 * the framebuffer size is not engraved in stone, applications should
 * determine mmap() sizes by multiplying linebytes*height*depth.  Pixel depth
 * is as follows:
 *
 *      dfb8    1
 *      dfb24   4
 *      stip    8
 *      blit    8
 *      rdfb32  4
 *      rstip   8
 *      rblit   8
 *
 * linebytes is returned via the FBIOGXINFO ioctl.
 *
 * Registers used by the driver:
 *
 *      HCMISC          chiprev, reset, video enable, vblank, vbank interrupt
 *      LINECOUNT       current line counter
 *      CURSOR_ADDRESS  cursor address register
 *
 *
 *
 *
 * Note on softc locking:  Bug 1142190 requires that the softc struct
 * *not* be locked during copyin/copyout.  This is because copyin/copyout
 * could generate disk I/O to swap in the user address space.  If
 * a vertical retrace interrupt should occur during this I/O, the tcx
 * interrupt routine would block if softc were locked.  This could block
 * the disk I/O, causing system deadlock.
 */
[...]

Christian