Home | History | Annotate | Line # | Download | only in pci
gffb.c revision 1.26
      1 /*	$NetBSD: gffb.c,v 1.26 2025/08/30 07:25:10 macallan Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2013 Michael Lorenz
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  *
     16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     26  */
     27 
     28 /*
     29  * A console driver for nvidia geforce graphics controllers
     30  * tested on macppc only so far, should work on other hardware as long as
     31  * something sets up a usable graphics mode and sets the right device properties
     32  * This driver should work with all NV1x hardware but so far it's been tested
     33  * only on NV11 / GeForce2 MX. Needs testing with more hardware and if
     34  * successful, PCI IDs need to be added to gffb_match()
     35  */
     36 
     37 #include <sys/cdefs.h>
     38 __KERNEL_RCSID(0, "$NetBSD: gffb.c,v 1.26 2025/08/30 07:25:10 macallan Exp $");
     39 
     40 #include <sys/param.h>
     41 #include <sys/systm.h>
     42 #include <sys/kernel.h>
     43 #include <sys/device.h>
     44 #include <sys/lwp.h>
     45 #include <sys/kauth.h>
     46 #include <sys/atomic.h>
     47 
     48 #include <dev/pci/pcivar.h>
     49 #include <dev/pci/pcireg.h>
     50 #include <dev/pci/pcidevs.h>
     51 #include <dev/pci/pciio.h>
     52 #include <dev/pci/gffbreg.h>
     53 
     54 #include <dev/wscons/wsdisplayvar.h>
     55 #include <dev/wscons/wsconsio.h>
     56 #include <dev/wsfont/wsfont.h>
     57 #include <dev/rasops/rasops.h>
     58 #include <dev/wscons/wsdisplay_vconsvar.h>
     59 #include <dev/pci/wsdisplay_pci.h>
     60 #include <dev/wscons/wsdisplay_glyphcachevar.h>
     61 
     62 #include "opt_gffb.h"
     63 #include "opt_vcons.h"
     64 
     65 #ifdef GFFB_DEBUG
     66 #define DPRINTF printf
     67 #else
     68 #define DPRINTF while(0) printf
     69 #endif
     70 
     71 struct gffb_softc {
     72 	device_t sc_dev;
     73 
     74 	pci_chipset_tag_t sc_pc;
     75 	pcitag_t sc_pcitag;
     76 
     77 	bus_space_tag_t sc_memt;
     78 	bus_space_tag_t sc_iot;
     79 
     80 	bus_space_handle_t sc_regh, sc_fbh;
     81 	bus_addr_t sc_fb, sc_reg;
     82 	bus_size_t sc_fbsize, sc_regsize;
     83 	uint8_t *sc_fbaddr;
     84 	size_t sc_vramsize;
     85 	uint32_t sc_fboffset;
     86 
     87 	int sc_width, sc_height, sc_depth, sc_stride;
     88 	int sc_locked, sc_accel;
     89 	struct vcons_screen sc_console_screen;
     90 	struct wsscreen_descr sc_defaultscreen_descr;
     91 	const struct wsscreen_descr *sc_screens[1];
     92 	struct wsscreen_list sc_screenlist;
     93 	struct vcons_data vd;
     94 	int sc_mode, sc_arch;
     95 	u_char sc_cmap_red[256];
     96 	u_char sc_cmap_green[256];
     97 	u_char sc_cmap_blue[256];
     98 	int sc_put, sc_current, sc_free;
     99 	uint32_t sc_rop;
    100 	void (*sc_putchar)(void *, int, int, u_int, long);
    101 	kmutex_t sc_lock;
    102 	glyphcache sc_gc;
    103 };
    104 
    105 static int	gffb_match(device_t, cfdata_t, void *);
    106 static void	gffb_attach(device_t, device_t, void *);
    107 
    108 CFATTACH_DECL_NEW(gffb, sizeof(struct gffb_softc),
    109     gffb_match, gffb_attach, NULL, NULL);
    110 
    111 static int	gffb_ioctl(void *, void *, u_long, void *, int, struct lwp *);
    112 static paddr_t	gffb_mmap(void *, void *, off_t, int);
    113 static void	gffb_init_screen(void *, struct vcons_screen *, int, long *);
    114 
    115 static int	gffb_putcmap(struct gffb_softc *, struct wsdisplay_cmap *);
    116 static int 	gffb_getcmap(struct gffb_softc *, struct wsdisplay_cmap *);
    117 static void	gffb_restore_palette(struct gffb_softc *);
    118 static int 	gffb_putpalreg(struct gffb_softc *, uint8_t, uint8_t,
    119 			    uint8_t, uint8_t);
    120 
    121  void	gffb_init(struct gffb_softc *);
    122 
    123 static void	gffb_make_room(struct gffb_softc *, int);
    124 static void	gffb_sync(struct gffb_softc *);
    125 
    126 static void	gffb_rectfill(struct gffb_softc *, int, int, int, int,
    127 			    uint32_t);
    128 static void	gffb_bitblt(void *, int, int, int, int, int, int, int);
    129 static void	gffb_rop(struct gffb_softc *, int);
    130 
    131 static void	gffb_cursor(void *, int, int, int);
    132 static void	gffb_putchar(void *, int, int, u_int, long);
    133 static void	gffb_copycols(void *, int, int, int, int);
    134 static void	gffb_erasecols(void *, int, int, int, long);
    135 static void	gffb_copyrows(void *, int, int, int);
    136 static void	gffb_eraserows(void *, int, int, long);
    137 
    138 #define GFFB_READ_4(o) bus_space_read_stream_4(sc->sc_memt, sc->sc_regh, (o))
    139 #define GFFB_WRITE_4(o, v) bus_space_write_stream_4(sc->sc_memt, sc->sc_regh, (o), (v))
    140 #define GFFB_WRITE_1(o, v) bus_space_write_1(sc->sc_memt, sc->sc_regh, (o), (v))
    141 
    142 struct wsdisplay_accessops gffb_accessops = {
    143 	gffb_ioctl,
    144 	gffb_mmap,
    145 	NULL,	/* alloc_screen */
    146 	NULL,	/* free_screen */
    147 	NULL,	/* show_screen */
    148 	NULL, 	/* load_font */
    149 	NULL,	/* pollc */
    150 	NULL	/* scroll */
    151 };
    152 
    153 static int
    154 gffb_match(device_t parent, cfdata_t match, void *aux)
    155 {
    156 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
    157 
    158 	if (PCI_CLASS(pa->pa_class) != PCI_CLASS_DISPLAY)
    159 		return 0;
    160 	if (PCI_VENDOR(pa->pa_id) != PCI_VENDOR_NVIDIA)
    161 		return 0;
    162 
    163 	/* only cards tested on so far - likely needs a list */
    164 	if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_NVIDIA_GEFORCE2MX)
    165 		return 100;
    166 	if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_NVIDIA_GEFORCE_6800U)
    167 		return 100;
    168 	if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_NVIDIA_GF_FXGO5200)
    169 		return 100;
    170 	if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_NVIDIA_GF_FX5200U)
    171 		return 100;
    172 	return (0);
    173 }
    174 
    175 static void
    176 gffb_attach(device_t parent, device_t self, void *aux)
    177 {
    178 	struct gffb_softc	*sc = device_private(self);
    179 	struct pci_attach_args	*pa = aux;
    180 	struct rasops_info	*ri;
    181 	bus_space_tag_t		tag;
    182 	struct wsemuldisplaydev_attach_args aa;
    183 	prop_dictionary_t	dict;
    184 	unsigned long		defattr;
    185 	pcireg_t		reg;
    186 	bool			is_console = FALSE;
    187 	uint32_t		addr;
    188 	int			i, j, f;
    189 	uint8_t			cmap[768];
    190 
    191 	sc->sc_pc = pa->pa_pc;
    192 	sc->sc_pcitag = pa->pa_tag;
    193 	sc->sc_memt = pa->pa_memt;
    194 	sc->sc_iot = pa->pa_iot;
    195 	sc->sc_dev = self;
    196 
    197 	/* first, see what kind of chip we've got */
    198 	reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, PCI_ID_REG);
    199 	switch (PCI_PRODUCT(reg)) {
    200 		case PCI_PRODUCT_NVIDIA_GEFORCE2MX:
    201 			sc->sc_accel = true;
    202 			sc->sc_arch = 10;
    203 			break;
    204 		case PCI_PRODUCT_NVIDIA_GEFORCE_6800U:
    205 			sc->sc_accel = false;
    206 			sc->sc_arch = 40;
    207 			break;
    208 		case PCI_PRODUCT_NVIDIA_GF_FXGO5200:
    209 		case PCI_PRODUCT_NVIDIA_GF_FX5200U:
    210 			sc->sc_accel = true;
    211 			sc->sc_arch = 30;
    212 			break;
    213 		default:
    214 			sc->sc_accel = false;
    215 			sc->sc_arch = 0;
    216 	}
    217 
    218 	pci_aprint_devinfo(pa, NULL);
    219 	DPRINTF("%s accel %d arch %d\n", __func__, sc->sc_accel, sc->sc_arch);
    220 	/* fill in parameters from properties */
    221 	dict = device_properties(self);
    222 	if (!prop_dictionary_get_uint32(dict, "width", &sc->sc_width)) {
    223 		aprint_error("%s: no width property\n", device_xname(self));
    224 		return;
    225 	}
    226 	if (!prop_dictionary_get_uint32(dict, "height", &sc->sc_height)) {
    227 		aprint_error("%s: no height property\n", device_xname(self));
    228 		return;
    229 	}
    230 
    231 #ifdef GLYPHCACHE_DEBUG
    232 	/* leave some visible VRAM unused so we can see the glyph cache */
    233 	sc->sc_height -= 300;
    234 #endif
    235 
    236 	if (!prop_dictionary_get_uint32(dict, "depth", &sc->sc_depth)) {
    237 		aprint_error("%s: no depth property\n", device_xname(self));
    238 		return;
    239 	}
    240 	if (!prop_dictionary_get_uint32(dict, "linebytes", &sc->sc_stride)) {
    241 		aprint_error("%s: no linebytes property\n",
    242 		    device_xname(self));
    243 		return;
    244 	}
    245 
    246 	/*
    247 	 * on !2MX we need to use the firmware's offset - for some reason
    248 	 * register writes to anything other than the DACs go wrong
    249 	 */
    250 	sc->sc_fboffset = 0;
    251 	if (prop_dictionary_get_uint32(dict, "address", &addr)) {
    252 		sc->sc_fboffset = addr & 0x000fffff;	/* XXX */
    253 	}
    254 	DPRINTF("%s: fboffset %8x\n", __func__, sc->sc_fboffset);
    255 	prop_dictionary_get_bool(dict, "is_console", &is_console);
    256 
    257 	if (pci_mapreg_map(pa, 0x10, PCI_MAPREG_TYPE_MEM, 0,
    258 	    &tag, &sc->sc_regh, &sc->sc_reg, &sc->sc_regsize)) {
    259 		aprint_error("%s: failed to map registers.\n",
    260 		    device_xname(sc->sc_dev));
    261 	}
    262 	/*
    263 	 * first thing we need to make sure register access uses host byte order
    264 	 * so we can recycle as much of xf86-video-nv as possible
    265 	 */
    266 #if BYTE_ORDER == BIG_ENDIAN
    267 	uint32_t mreg = GFFB_READ_4(GFFB_PMC + 4);
    268 	if ((mreg & 0x01000001) == 0) {
    269 		GFFB_WRITE_4(GFFB_PMC + 4, 0x01000001);
    270 	}
    271 #endif
    272 	sc->sc_vramsize = GFFB_READ_4(GFFB_VRAM) & 0xfff00000;
    273 
    274 	/* don't map more VRAM than we actually have */
    275 	if (pci_mapreg_info(sc->sc_pc, sc->sc_pcitag,
    276 	    0x14, PCI_MAPREG_TYPE_MEM, &sc->sc_fb, &sc->sc_fbsize, &f)) {
    277 		aprint_error("%s: can't find the framebuffer?!\n",
    278 		    device_xname(sc->sc_dev));
    279 	}
    280 	if (sc->sc_vramsize == 0) sc->sc_vramsize = sc->sc_fbsize;
    281 
    282 	/* don't map (much) more than we actually need */
    283 	if (bus_space_map(sc->sc_memt, sc->sc_fb, 0x1000000,
    284 	    BUS_SPACE_MAP_PREFETCHABLE | BUS_SPACE_MAP_LINEAR,
    285 	    &sc->sc_fbh)) {
    286 		aprint_error("%s: failed to map the framebuffer.\n",
    287 		    device_xname(sc->sc_dev));
    288 	}
    289 	sc->sc_fbaddr = bus_space_vaddr(tag, sc->sc_fbh);
    290 
    291 	aprint_normal("%s: %d MB aperture at 0x%08x\n", device_xname(self),
    292 	    (int)(sc->sc_fbsize >> 20), (uint32_t)sc->sc_fb);
    293 	aprint_normal_dev(sc->sc_dev, "%d MB video memory\n",
    294 	    (int)(sc->sc_vramsize >> 20));
    295 
    296 	sc->sc_defaultscreen_descr = (struct wsscreen_descr){
    297 		"default",
    298 		0, 0,
    299 		NULL,
    300 		8, 16,
    301 		WSSCREEN_WSCOLORS | WSSCREEN_HILIT | WSSCREEN_RESIZE,
    302 		NULL
    303 	};
    304 	sc->sc_screens[0] = &sc->sc_defaultscreen_descr;
    305 	sc->sc_screenlist = (struct wsscreen_list){1, sc->sc_screens};
    306 	sc->sc_mode = WSDISPLAYIO_MODE_EMUL;
    307 	sc->sc_locked = 0;
    308 
    309 #ifdef GFFB_DEBUG
    310 	printf("put: %08x\n", GFFB_READ_4(GFFB_FIFO_PUT));
    311 	printf("get: %08x\n", GFFB_READ_4(GFFB_FIFO_GET));
    312 #endif
    313 
    314 	/*
    315 	 * we don't have hardware synchronization so we need a lock to serialize
    316 	 * access to the DMA buffer between normal and kernel output
    317 	 * actually it might be enough to use atomic ops on sc_current, sc_free
    318 	 * etc. but for now we'll play it safe
    319 	 * XXX we will probably deadlock if we take an interrupt while sc_lock
    320 	 * is held and then try to printf()
    321 	 */
    322 	mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE);
    323 
    324 	/* init engine here */
    325 	gffb_init(sc);
    326 
    327 #ifdef GFFB_DEBUG
    328 	printf("put: %08x\n", GFFB_READ_4(GFFB_FIFO_PUT));
    329 	printf("get: %08x\n", GFFB_READ_4(GFFB_FIFO_GET));
    330 #endif
    331 	vcons_init(&sc->vd, sc, &sc->sc_defaultscreen_descr,
    332 	    &gffb_accessops);
    333 	sc->vd.init_screen = gffb_init_screen;
    334 
    335 
    336 	ri = &sc->sc_console_screen.scr_ri;
    337 
    338 	if (sc->sc_accel) {
    339 		sc->sc_gc.gc_bitblt = gffb_bitblt;
    340 		sc->sc_gc.gc_blitcookie = sc;
    341 		sc->sc_gc.gc_rop = 0xcc;
    342 		sc->vd.show_screen_cookie = &sc->sc_gc;
    343 		sc->vd.show_screen_cb = glyphcache_adapt;
    344 	}
    345 
    346 	if (is_console) {
    347 		vcons_init_screen(&sc->vd, &sc->sc_console_screen, 1,
    348 		    &defattr);
    349 		sc->sc_console_screen.scr_flags |= VCONS_SCREEN_IS_STATIC;
    350 
    351 		if (sc->sc_accel) {
    352 			gffb_rectfill(sc, 0, 0, sc->sc_width, sc->sc_height,
    353 		    		ri->ri_devcmap[(defattr >> 16) & 0xf]);
    354 		} else {
    355 			memset(sc->sc_fbaddr + sc->sc_fboffset,
    356 			       ri->ri_devcmap[(defattr >> 16) & 0xf],
    357 			       sc->sc_stride * sc->sc_height);
    358 		}
    359 		sc->sc_defaultscreen_descr.textops = &ri->ri_ops;
    360 		sc->sc_defaultscreen_descr.capabilities = ri->ri_caps;
    361 		sc->sc_defaultscreen_descr.nrows = ri->ri_rows;
    362 		sc->sc_defaultscreen_descr.ncols = ri->ri_cols;
    363 
    364 		if (sc->sc_accel)
    365 			glyphcache_init(&sc->sc_gc, sc->sc_height + 5,
    366 				(0x800000 / sc->sc_stride) - sc->sc_height - 5,
    367 				sc->sc_width,
    368 				ri->ri_font->fontwidth,
    369 				ri->ri_font->fontheight,
    370 				defattr);
    371 
    372 		wsdisplay_cnattach(&sc->sc_defaultscreen_descr, ri, 0, 0,
    373 		    defattr);
    374 		vcons_replay_msgbuf(&sc->sc_console_screen);
    375 	} else {
    376 		/*
    377 		 * since we're not the console we can postpone the rest
    378 		 * until someone actually allocates a screen for us
    379 		 */
    380 		if (sc->sc_console_screen.scr_ri.ri_rows == 0) {
    381 			/* do some minimal setup to avoid weirdnesses later */
    382 			vcons_init_screen(&sc->vd, &sc->sc_console_screen, 1,
    383 			    &defattr);
    384 		} else
    385 			(*ri->ri_ops.allocattr)(ri, 0, 0, 0, &defattr);
    386 
    387 		if (sc->sc_accel)
    388 			glyphcache_init(&sc->sc_gc, sc->sc_height + 5,
    389 				(0x800000 / sc->sc_stride) - sc->sc_height - 5,
    390 				sc->sc_width,
    391 				ri->ri_font->fontwidth,
    392 				ri->ri_font->fontheight,
    393 				defattr);
    394 	}
    395 
    396 	j = 0;
    397 	rasops_get_cmap(ri, cmap, sizeof(cmap));
    398 	for (i = 0; i < 256; i++) {
    399 		sc->sc_cmap_red[i] = cmap[j];
    400 		sc->sc_cmap_green[i] = cmap[j + 1];
    401 		sc->sc_cmap_blue[i] = cmap[j + 2];
    402 		gffb_putpalreg(sc, i, cmap[j], cmap[j + 1], cmap[j + 2]);
    403 		j += 3;
    404 	}
    405 
    406 	/* no suspend/resume support yet */
    407 	if (!pmf_device_register(sc->sc_dev, NULL, NULL))
    408 		aprint_error_dev(sc->sc_dev,
    409 		    "couldn't establish power handler\n");
    410 
    411 	aa.console = is_console;
    412 	aa.scrdata = &sc->sc_screenlist;
    413 	aa.accessops = &gffb_accessops;
    414 	aa.accesscookie = &sc->vd;
    415 
    416 	config_found(sc->sc_dev, &aa, wsemuldisplaydevprint, CFARGS_NONE);
    417 }
    418 
    419 static int
    420 gffb_ioctl(void *v, void *vs, u_long cmd, void *data, int flag, struct lwp *l)
    421 {
    422 	struct vcons_data *vd = v;
    423 	struct gffb_softc *sc = vd->cookie;
    424 	struct wsdisplay_fbinfo *wdf;
    425 	struct vcons_screen *ms = vd->active;
    426 
    427 	switch (cmd) {
    428 	case WSDISPLAYIO_GTYPE:
    429 		*(u_int *)data = WSDISPLAY_TYPE_PCIMISC;
    430 		return 0;
    431 
    432 	/* PCI config read/write passthrough. */
    433 	case PCI_IOC_CFGREAD:
    434 	case PCI_IOC_CFGWRITE:
    435 		return pci_devioctl(sc->sc_pc, sc->sc_pcitag,
    436 		    cmd, data, flag, l);
    437 
    438 	case WSDISPLAYIO_GET_BUSID:
    439 		return wsdisplayio_busid_pci(sc->sc_dev, sc->sc_pc,
    440 		    sc->sc_pcitag, data);
    441 
    442 	case WSDISPLAYIO_GINFO:
    443 		if (ms == NULL)
    444 			return ENODEV;
    445 		wdf = (void *)data;
    446 		wdf->height = ms->scr_ri.ri_height;
    447 		wdf->width = ms->scr_ri.ri_width;
    448 		wdf->depth = ms->scr_ri.ri_depth;
    449 		wdf->cmsize = 256;
    450 		return 0;
    451 
    452 	case WSDISPLAYIO_GETCMAP:
    453 		return gffb_getcmap(sc,
    454 		    (struct wsdisplay_cmap *)data);
    455 
    456 	case WSDISPLAYIO_PUTCMAP:
    457 		return gffb_putcmap(sc,
    458 		    (struct wsdisplay_cmap *)data);
    459 
    460 	case WSDISPLAYIO_LINEBYTES:
    461 		*(u_int *)data = sc->sc_stride;
    462 		return 0;
    463 
    464 	case WSDISPLAYIO_SMODE: {
    465 		int new_mode = *(int*)data;
    466 		if (new_mode != sc->sc_mode) {
    467 			sc->sc_mode = new_mode;
    468 			if(new_mode == WSDISPLAYIO_MODE_EMUL) {
    469 				gffb_init(sc);
    470 				gffb_restore_palette(sc);
    471 				if (sc->sc_accel) {
    472 					glyphcache_wipe(&sc->sc_gc);
    473 					gffb_rectfill(sc, 0, 0, sc->sc_width,
    474 					    sc->sc_height, ms->scr_ri.ri_devcmap[
    475 					    (ms->scr_defattr >> 16) & 0xf]);
    476 				} else {
    477 					memset(sc->sc_fbaddr + sc->sc_fboffset,
    478 					       ms->scr_ri.ri_devcmap[
    479 					         (ms->scr_defattr >> 16) & 0xf],
    480 					       sc->sc_stride * sc->sc_height);
    481 				}
    482 				vcons_redraw_screen(ms);
    483 			}
    484 		}
    485 		}
    486 		return 0;
    487 
    488 	case WSDISPLAYIO_GET_EDID: {
    489 		struct wsdisplayio_edid_info *d = data;
    490 		return wsdisplayio_get_edid(sc->sc_dev, d);
    491 	}
    492 
    493 	case WSDISPLAYIO_GET_FBINFO: {
    494 		struct wsdisplayio_fbinfo *fbi = data;
    495 		return wsdisplayio_get_fbinfo(&ms->scr_ri, fbi);
    496 	}
    497 	}
    498 	return EPASSTHROUGH;
    499 }
    500 
    501 static paddr_t
    502 gffb_mmap(void *v, void *vs, off_t offset, int prot)
    503 {
    504 	struct vcons_data *vd = v;
    505 	struct gffb_softc *sc = vd->cookie;
    506 	paddr_t pa;
    507 
    508 	/* 'regular' framebuffer mmap()ing */
    509 	if (offset < sc->sc_vramsize) {
    510 		pa = bus_space_mmap(sc->sc_memt, sc->sc_fb + offset + 0x2000,
    511 		    0, prot, BUS_SPACE_MAP_LINEAR);
    512 		return pa;
    513 	}
    514 
    515 	/*
    516 	 * restrict all other mappings to processes with superuser privileges
    517 	 * or the kernel itself
    518 	 */
    519 	if (kauth_authorize_machdep(kauth_cred_get(),
    520 	    KAUTH_MACHDEP_UNMANAGEDMEM,
    521 	    NULL, NULL, NULL, NULL) != 0) {
    522 		aprint_normal("%s: mmap() rejected.\n",
    523 		    device_xname(sc->sc_dev));
    524 		return -1;
    525 	}
    526 
    527 	if ((offset >= sc->sc_fb) && (offset < (sc->sc_fb + sc->sc_fbsize))) {
    528 		pa = bus_space_mmap(sc->sc_memt, offset, 0, prot,
    529 		    BUS_SPACE_MAP_LINEAR);
    530 		return pa;
    531 	}
    532 
    533 	if ((offset >= sc->sc_reg) &&
    534 	    (offset < (sc->sc_reg + sc->sc_regsize))) {
    535 		pa = bus_space_mmap(sc->sc_memt, offset, 0, prot,
    536 		    BUS_SPACE_MAP_LINEAR);
    537 		return pa;
    538 	}
    539 
    540 #ifdef PCI_MAGIC_IO_RANGE
    541 	/* allow mapping of IO space */
    542 	if ((offset >= PCI_MAGIC_IO_RANGE) &&
    543 	    (offset < PCI_MAGIC_IO_RANGE + 0x10000)) {
    544 		pa = bus_space_mmap(sc->sc_iot, offset - PCI_MAGIC_IO_RANGE,
    545 		    0, prot, BUS_SPACE_MAP_LINEAR);
    546 		return pa;
    547 	}
    548 #endif
    549 
    550 	return -1;
    551 }
    552 
    553 static void
    554 gffb_init_screen(void *cookie, struct vcons_screen *scr,
    555     int existing, long *defattr)
    556 {
    557 	struct gffb_softc *sc = cookie;
    558 	struct rasops_info *ri = &scr->scr_ri;
    559 
    560 	ri->ri_depth = sc->sc_depth;
    561 	ri->ri_width = sc->sc_width;
    562 	ri->ri_height = sc->sc_height;
    563 	ri->ri_stride = sc->sc_stride;
    564 	if (sc->sc_depth == 8)
    565 	ri->ri_bits = sc->sc_fbaddr + sc->sc_fboffset;
    566 	ri->ri_flg = RI_CENTER;
    567 		ri->ri_flg |= RI_8BIT_IS_RGB | RI_ENABLE_ALPHA;
    568 
    569 	rasops_init(ri, 0, 0);
    570 	ri->ri_caps = WSSCREEN_WSCOLORS | WSSCREEN_RESIZE;
    571 	scr->scr_flags |= VCONS_LOADFONT;
    572 
    573 	rasops_reconfig(ri, sc->sc_height / ri->ri_font->fontheight,
    574 		    sc->sc_width / ri->ri_font->fontwidth);
    575 
    576 	ri->ri_hw = scr;
    577 
    578 	if (sc->sc_accel) {
    579 		sc->sc_putchar = ri->ri_ops.putchar;
    580 		ri->ri_ops.copyrows = gffb_copyrows;
    581 		ri->ri_ops.copycols = gffb_copycols;
    582 		ri->ri_ops.eraserows = gffb_eraserows;
    583 		ri->ri_ops.erasecols = gffb_erasecols;
    584 		ri->ri_ops.cursor = gffb_cursor;
    585 		ri->ri_ops.putchar = gffb_putchar;
    586 	} else {
    587 		scr->scr_flags |= VCONS_DONT_READ;
    588 	}
    589 }
    590 
    591 static int
    592 gffb_putcmap(struct gffb_softc *sc, struct wsdisplay_cmap *cm)
    593 {
    594 	u_char *r, *g, *b;
    595 	u_int index = cm->index;
    596 	u_int count = cm->count;
    597 	int i, error;
    598 	u_char rbuf[256], gbuf[256], bbuf[256];
    599 
    600 #ifdef GFFB_DEBUG
    601 	aprint_debug("putcmap: %d %d\n",index, count);
    602 #endif
    603 	if (cm->index >= 256 || cm->count > 256 ||
    604 	    (cm->index + cm->count) > 256)
    605 		return EINVAL;
    606 	error = copyin(cm->red, &rbuf[index], count);
    607 	if (error)
    608 		return error;
    609 	error = copyin(cm->green, &gbuf[index], count);
    610 	if (error)
    611 		return error;
    612 	error = copyin(cm->blue, &bbuf[index], count);
    613 	if (error)
    614 		return error;
    615 
    616 	memcpy(&sc->sc_cmap_red[index], &rbuf[index], count);
    617 	memcpy(&sc->sc_cmap_green[index], &gbuf[index], count);
    618 	memcpy(&sc->sc_cmap_blue[index], &bbuf[index], count);
    619 
    620 	r = &sc->sc_cmap_red[index];
    621 	g = &sc->sc_cmap_green[index];
    622 	b = &sc->sc_cmap_blue[index];
    623 
    624 	for (i = 0; i < count; i++) {
    625 		gffb_putpalreg(sc, index, *r, *g, *b);
    626 		index++;
    627 		r++, g++, b++;
    628 	}
    629 	return 0;
    630 }
    631 
    632 static int
    633 gffb_getcmap(struct gffb_softc *sc, struct wsdisplay_cmap *cm)
    634 {
    635 	u_int index = cm->index;
    636 	u_int count = cm->count;
    637 	int error;
    638 
    639 	if (index >= 255 || count > 256 || index + count > 256)
    640 		return EINVAL;
    641 
    642 	error = copyout(&sc->sc_cmap_red[index],   cm->red,   count);
    643 	if (error)
    644 		return error;
    645 	error = copyout(&sc->sc_cmap_green[index], cm->green, count);
    646 	if (error)
    647 		return error;
    648 	error = copyout(&sc->sc_cmap_blue[index],  cm->blue,  count);
    649 	if (error)
    650 		return error;
    651 
    652 	return 0;
    653 }
    654 
    655 static void
    656 gffb_restore_palette(struct gffb_softc *sc)
    657 {
    658 	int i;
    659 
    660 	for (i = 0; i < (1 << sc->sc_depth); i++) {
    661 		gffb_putpalreg(sc, i, sc->sc_cmap_red[i],
    662 		    sc->sc_cmap_green[i], sc->sc_cmap_blue[i]);
    663 	}
    664 }
    665 
    666 static int
    667 gffb_putpalreg(struct gffb_softc *sc, uint8_t idx, uint8_t r, uint8_t g,
    668     uint8_t b)
    669 {
    670 	/* port 0 */
    671 	GFFB_WRITE_1(GFFB_PDIO0 + GFFB_PEL_IW, idx);
    672 	GFFB_WRITE_1(GFFB_PDIO0 + GFFB_PEL_D, r);
    673 	GFFB_WRITE_1(GFFB_PDIO0 + GFFB_PEL_D, g);
    674 	GFFB_WRITE_1(GFFB_PDIO0 + GFFB_PEL_D, b);
    675 
    676 	/* port 1 */
    677 	GFFB_WRITE_1(GFFB_PDIO1 + GFFB_PEL_IW, idx);
    678 	GFFB_WRITE_1(GFFB_PDIO1 + GFFB_PEL_D, r);
    679 	GFFB_WRITE_1(GFFB_PDIO1 + GFFB_PEL_D, g);
    680 	GFFB_WRITE_1(GFFB_PDIO1 + GFFB_PEL_D, b);
    681 
    682 	return 0;
    683 }
    684 
    685 
    686 static void
    687 gffb_dma_kickoff(struct gffb_softc *sc)
    688 {
    689 	volatile uint32_t junk;
    690 	if (sc->sc_current != sc->sc_put) {
    691 		sc->sc_put = sc->sc_current;
    692 		bus_space_barrier(sc->sc_memt, sc->sc_fbh, 0, 0x1000000,
    693 		    BUS_SPACE_BARRIER_WRITE);
    694 		junk = *sc->sc_fbaddr;
    695 		__USE(junk);
    696 		GFFB_WRITE_4(GFFB_FIFO_PUT, sc->sc_put);
    697 		bus_space_barrier(sc->sc_memt, sc->sc_regh, GFFB_FIFO_PUT, 4,
    698 		    BUS_SPACE_BARRIER_WRITE);
    699 	}
    700 }
    701 
    702 static void
    703 gffb_dmanext(struct gffb_softc *sc, uint32_t data)
    704 {
    705 	bus_space_write_stream_4(sc->sc_memt, sc->sc_fbh, sc->sc_current, data);
    706 	sc->sc_current += 4;
    707 }
    708 
    709 static void
    710 gffb_dmastart(struct gffb_softc *sc, uint32_t tag, int size)
    711 {
    712 	if(sc->sc_free <= (size << 2))
    713 		gffb_make_room(sc, size);
    714 	gffb_dmanext(sc, ((size) << 18) | (tag));
    715 	sc->sc_free -= ((size + 1) << 2);
    716 }
    717 
    718 /*
    719  * from xf86_video_nv/nv_xaa.c:
    720  * There is a HW race condition with videoram command buffers.
    721  * You can't jump to the location of your put offset.  We write put
    722  * at the jump offset + SKIPS dwords with noop padding in between
    723  * to solve this problem
    724  */
    725 
    726 #define SKIPS  8
    727 
    728 static void
    729 gffb_make_room(struct gffb_softc *sc, int size)
    730 {
    731 	uint32_t get;
    732 
    733 	size = (size + 1) << 2;	/* slots -> offset */
    734 
    735 	while (sc->sc_free < size) {
    736 		get = GFFB_READ_4(GFFB_FIFO_GET);
    737 
    738 		if (sc->sc_put >= get) {
    739 			sc->sc_free = 0x2000 - sc->sc_current;
    740 			if (sc->sc_free < size) {
    741 				gffb_dmanext(sc, 0x20000000);
    742 				if(get <= (SKIPS << 2)) {
    743 					if (sc->sc_put <= (SKIPS << 2)) {
    744 						/* corner case - will be idle */
    745 						GFFB_WRITE_4(GFFB_FIFO_PUT,
    746 						    (SKIPS + 1) << 2);
    747 					}
    748 					do {
    749 						get =GFFB_READ_4(GFFB_FIFO_GET);
    750 					} while (get <= (SKIPS << 2));
    751 				}
    752 				GFFB_WRITE_4(GFFB_FIFO_PUT, SKIPS << 2);
    753 				sc->sc_current = sc->sc_put = (SKIPS << 2);
    754 				sc->sc_free = get - ((SKIPS + 1) << 2);
    755 			}
    756 		} else
    757 			sc->sc_free = get - sc->sc_current - 4;
    758 	}
    759 }
    760 
    761 static void
    762 gffb_sync(struct gffb_softc *sc)
    763 {
    764 	int bail;
    765 	int i;
    766 
    767 	/*
    768 	 * if there are commands in the buffer make sure the chip is actually
    769 	 * trying to run them
    770 	 */
    771 	gffb_dma_kickoff(sc);
    772 
    773 	/* now wait for the command buffer to drain... */
    774 	bail = 100000000;
    775 	while ((GFFB_READ_4(GFFB_FIFO_GET) != sc->sc_put) && (bail > 0)) {
    776 		bail--;
    777 	}
    778 	if (bail == 0) {
    779 		printf("FIFO isn't moving\n");
    780 		goto crap;
    781 	}
    782 
    783 	/* ... and for the engine to go idle */
    784 	bail = 100000000;
    785 	while((GFFB_READ_4(GFFB_BUSY) != 0) && (bail > 0)) {
    786 		bail--;
    787 	}
    788 	if (bail == 0) goto crap;
    789 	return;
    790 crap:
    791 	/* if we time out fill the buffer with NOPs and cross fingers */
    792 	DPRINTF("GET %08x\n", GFFB_READ_4(GFFB_FIFO_GET));
    793 	sc->sc_put = 0;
    794 	sc->sc_current = 0;
    795 	for (i = 0; i < 0x2000; i += 4)
    796 		bus_space_write_stream_4(sc->sc_memt, sc->sc_fbh, i, 0);
    797 	aprint_error_dev(sc->sc_dev, "DMA lockup\n");
    798 }
    799 
    800 void
    801 gffb_init(struct gffb_softc *sc)
    802 {
    803 	int i;
    804 	uint32_t foo;
    805 
    806 	if (!sc->sc_accel) return;
    807 	DPRINTF("%s offset %08x %08x\n", __func__,
    808 	    GFFB_READ_4(GFFB_CRTC0 + GFFB_DISPLAYSTART),
    809 	    GFFB_READ_4(GFFB_CRTC1 + GFFB_DISPLAYSTART));
    810 
    811 	sc->sc_fboffset = 0x2000;
    812 
    813 	/* init display start */
    814 	GFFB_WRITE_4(GFFB_CRTC0 + GFFB_DISPLAYSTART, sc->sc_fboffset);
    815 	GFFB_WRITE_4(GFFB_CRTC1 + GFFB_DISPLAYSTART, sc->sc_fboffset);
    816 
    817 	/* make sure we do 8bit per channel */
    818 	GFFB_WRITE_1(GFFB_PDIO0 + GFFB_PEL_MASK, 0xff);
    819 	GFFB_WRITE_1(GFFB_PDIO1 + GFFB_PEL_MASK, 0xff);
    820 
    821 	/* DMA stuff. A whole lot of magic number voodoo from xf86-video-nv */
    822 	GFFB_WRITE_4(GFFB_PMC + 0x140, 0);
    823 	GFFB_WRITE_4(GFFB_PMC + 0x200, 0xffff00ff);
    824 	GFFB_WRITE_4(GFFB_PMC + 0x200, 0xffffffff);
    825 	GFFB_WRITE_4(GFFB_PTIMER + 0x800, 8);
    826 	GFFB_WRITE_4(GFFB_PTIMER + 0x840, 3);
    827 	GFFB_WRITE_4(GFFB_PTIMER + 0x500, 0);
    828 	GFFB_WRITE_4(GFFB_PTIMER + 0x400, 0xffffffff);
    829 	for (i = 0; i < 8; i++) {
    830 		GFFB_WRITE_4(GFFB_PFB + 0x0240 + (i * 0x10), 0);
    831 		GFFB_WRITE_4(GFFB_PFB + 0x0244 + (i * 0x10),
    832 		    sc->sc_vramsize - 1);
    833 	}
    834 
    835 	GFFB_WRITE_4(GFFB_PRAMIN, 0x80000010);
    836 	GFFB_WRITE_4(GFFB_PRAMIN + 0x04, 0x80011201);
    837 	GFFB_WRITE_4(GFFB_PRAMIN + 0x08, 0x80000011);
    838 	GFFB_WRITE_4(GFFB_PRAMIN + 0x0c, 0x80011202);
    839 	GFFB_WRITE_4(GFFB_PRAMIN + 0x10, 0x80000012);
    840 	GFFB_WRITE_4(GFFB_PRAMIN + 0x14, 0x80011203);
    841 	GFFB_WRITE_4(GFFB_PRAMIN + 0x18, 0x80000013);
    842 	GFFB_WRITE_4(GFFB_PRAMIN + 0x1c, 0x80011204);
    843 	GFFB_WRITE_4(GFFB_PRAMIN + 0x20, 0x80000014);
    844 	GFFB_WRITE_4(GFFB_PRAMIN + 0x24, 0x80011205);
    845 	GFFB_WRITE_4(GFFB_PRAMIN + 0x28, 0x80000015);
    846 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2c, 0x80011206);
    847 	GFFB_WRITE_4(GFFB_PRAMIN + 0x30, 0x80000016);
    848 	GFFB_WRITE_4(GFFB_PRAMIN + 0x34, 0x80011207);
    849 	GFFB_WRITE_4(GFFB_PRAMIN + 0x38, 0x80000017);
    850 	GFFB_WRITE_4(GFFB_PRAMIN + 0x3c, 0x80011208);
    851 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2000, 0x00003000);
    852 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2004, sc->sc_vramsize - 1);
    853 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2008, 0x00000002);
    854 	GFFB_WRITE_4(GFFB_PRAMIN + 0x200c, 0x00000002);
    855 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2010, 0x01008062);	/* nv10+ */
    856 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2014, 0);
    857 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2018, 0x12001200);
    858 	GFFB_WRITE_4(GFFB_PRAMIN + 0x201c, 0);
    859 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2020, 0x01008043);
    860 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2024, 0);
    861 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2028, 0);
    862 	GFFB_WRITE_4(GFFB_PRAMIN + 0x202c, 0);
    863 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2030, 0x01008044);
    864 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2034, 0x00000002);
    865 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2038, 0);
    866 	GFFB_WRITE_4(GFFB_PRAMIN + 0x203c, 0);
    867 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2040, 0x01008019);
    868 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2044, 0);
    869 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2048, 0);
    870 	GFFB_WRITE_4(GFFB_PRAMIN + 0x204c, 0);
    871 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2050, 0x0100a05c);
    872 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2054, 0);
    873 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2058, 0);
    874 	GFFB_WRITE_4(GFFB_PRAMIN + 0x205c, 0);
    875 	/* XXX 0x0100805f if !WaitVSynvPossible */
    876 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2060, 0x0100805f);
    877 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2064, 0);
    878 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2068, 0x12001200);
    879 	GFFB_WRITE_4(GFFB_PRAMIN + 0x206c, 0);
    880 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2070, 0x0100804a);
    881 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2074, 0x00000002);
    882 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2078, 0);
    883 	GFFB_WRITE_4(GFFB_PRAMIN + 0x207c, 0);
    884 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2080, 0x01018077);
    885 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2084, 0);
    886 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2088, 0x12001200);
    887 	GFFB_WRITE_4(GFFB_PRAMIN + 0x208c, 0);
    888 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2090, 0x00003002);
    889 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2094, 0x00007fff);
    890 	/* command buffer start with some flag in the lower bits */
    891 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2098, sc->sc_vramsize | 0x00000002);
    892 	GFFB_WRITE_4(GFFB_PRAMIN + 0x209c, 0x00000002);
    893 #if BYTE_ORDER == BIG_ENDIAN
    894 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2010, 0x01088062);
    895 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2020, 0x01088043);
    896 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2030, 0x01088044);
    897 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2040, 0x01088019);
    898 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2050, 0x0108a05c);
    899 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2060, 0x0108805f);
    900 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2070, 0x0108804a);
    901 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2080, 0x01098077);
    902 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2034, 0x00000001);
    903 	GFFB_WRITE_4(GFFB_PRAMIN + 0x2074, 0x00000001);
    904 #endif
    905 	/* PGRAPH setup */
    906 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0080, 0xFFFFFFFF);
    907 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0080, 0x00000000);
    908 
    909 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0140, 0x00000000);
    910 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0100, 0xFFFFFFFF);
    911 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0144, 0x10010100);
    912 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0714, 0xFFFFFFFF);
    913 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0720, 0x00000001);
    914 	/*
    915 	 * xf86_video_nv does this in two writes,
    916 	 * not sure if they can be combined
    917 	 */
    918 	foo = GFFB_READ_4(GFFB_PGRAPH + 0x0710);
    919 	foo &= 0x0007ff00;
    920 	foo |= 0x00020100;
    921 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0710, foo);
    922 
    923 	/* NV_ARCH_10 */
    924 	if(sc->sc_arch == 10) {
    925 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0084, 0x00118700);
    926 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0088, 0x24E00810);
    927 		GFFB_WRITE_4(GFFB_PGRAPH + 0x008C, 0x55DE0030);
    928 
    929 		for(i = 0; i < 128; i += 4) {
    930 			GFFB_WRITE_4(GFFB_PGRAPH + 0x0B00 + i,
    931 			    GFFB_READ_4(GFFB_PFB + 0x0240 + i));
    932 		}
    933 
    934 		GFFB_WRITE_4(GFFB_PGRAPH + 0x640, 0);
    935 		GFFB_WRITE_4(GFFB_PGRAPH + 0x644, 0);
    936 		GFFB_WRITE_4(GFFB_PGRAPH + 0x684, sc->sc_vramsize - 1);
    937 		GFFB_WRITE_4(GFFB_PGRAPH + 0x688, sc->sc_vramsize - 1);
    938 
    939 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0810, 0x00000000);
    940 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0608, 0xFFFFFFFF);
    941 	} else {
    942 		/* nv30 */
    943 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0084, 0x40108700);
    944 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0890, 0x00140000);
    945 		GFFB_WRITE_4(GFFB_PGRAPH + 0x008C, 0xf00e0431);
    946 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0090, 0x00008000);
    947 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0610, 0xf04b1f36);
    948 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0B80, 0x1002d888);
    949 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0B88, 0x62ff007f);
    950 
    951 		for (i = 0; i < 128; i += 4) {
    952 			GFFB_WRITE_4(GFFB_PGRAPH + 0x0900 + i,
    953 			    GFFB_READ_4(GFFB_PFB + 0x0240 + i));
    954   			GFFB_WRITE_4(GFFB_PGRAPH + 0x6900 + i,
    955 			    GFFB_READ_4(GFFB_PFB + 0x0240 + i));
    956 		}
    957 
    958 		GFFB_WRITE_4(GFFB_PGRAPH + 0x09A4,
    959 			GFFB_READ_4(GFFB_PFB + 0x0200));
    960 		GFFB_WRITE_4(GFFB_PGRAPH + 0x09A8,
    961 			GFFB_READ_4(GFFB_PFB + 0x0204));
    962 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0750, 0x00EA0000);
    963 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0754,
    964 			GFFB_READ_4(GFFB_PFB + 0x0200));
    965 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0750, 0x00EA0004);
    966 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0754,
    967 			GFFB_READ_4(GFFB_PFB + 0x0204));
    968 
    969 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0820, 0);
    970 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0824, 0);
    971 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0864, sc->sc_vramsize - 1);
    972 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0868, sc->sc_vramsize - 1);
    973 
    974 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0B20, 0x00000000);
    975 		GFFB_WRITE_4(GFFB_PGRAPH + 0x0B04, 0xFFFFFFFF);
    976 	}
    977 
    978 	/* PFIFO setup */
    979 	GFFB_WRITE_4(GFFB_PGRAPH + 0x053C, 0);
    980 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0540, 0);
    981 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0544, 0x00007FFF);
    982 	GFFB_WRITE_4(GFFB_PGRAPH + 0x0548, 0x00007FFF);
    983 
    984 	GFFB_WRITE_4(GFFB_PFIFO + 0x0500, 0);
    985 	GFFB_WRITE_4(GFFB_PFIFO + 0x0504, 0x00000001);
    986 	GFFB_WRITE_4(GFFB_PFIFO + 0x1200, 0);
    987 	GFFB_WRITE_4(GFFB_PFIFO + 0x1250, 0);
    988 	GFFB_WRITE_4(GFFB_PFIFO + 0x1204, 0x00000100);	/* different on nv40 */
    989 	GFFB_WRITE_4(GFFB_PFIFO + 0x1240, 0);
    990 	GFFB_WRITE_4(GFFB_PFIFO + 0x1244, 0);
    991 	GFFB_WRITE_4(GFFB_PFIFO + 0x122c, 0x00001209);	/* different on nv40 */
    992 	GFFB_WRITE_4(GFFB_PFIFO + 0x1000, 0);
    993 	GFFB_WRITE_4(GFFB_PFIFO + 0x1050, 0);
    994 	GFFB_WRITE_4(GFFB_PFIFO + 0x0210, 0x03000100);
    995 	GFFB_WRITE_4(GFFB_PFIFO + 0x0214, 0x00000110);
    996 	GFFB_WRITE_4(GFFB_PFIFO + 0x0218, 0x00000112);
    997 	GFFB_WRITE_4(GFFB_PFIFO + 0x050c, 0x0000ffff);
    998 	GFFB_WRITE_4(GFFB_PFIFO + 0x1258, 0x0000ffff);
    999 	GFFB_WRITE_4(GFFB_PFIFO + 0x0140, 0);
   1000 	GFFB_WRITE_4(GFFB_PFIFO + 0x0100, 0xffffffff);
   1001 	GFFB_WRITE_4(GFFB_PFIFO + 0x1054, 0x00000001);
   1002 	GFFB_WRITE_4(GFFB_PFIFO + 0x1230, 0);
   1003 	GFFB_WRITE_4(GFFB_PFIFO + 0x1280, 0);
   1004 #if BYTE_ORDER == BIG_ENDIAN
   1005 	GFFB_WRITE_4(GFFB_PFIFO + 0x1224, 0x800f0078);
   1006 #else
   1007 	GFFB_WRITE_4(GFFB_PFIFO + 0x1224, 0x000f0078);
   1008 #endif
   1009 	GFFB_WRITE_4(GFFB_PFIFO + 0x1220, 0x00000001);
   1010 	GFFB_WRITE_4(GFFB_PFIFO + 0x1200, 0x00000001);
   1011 	GFFB_WRITE_4(GFFB_PFIFO + 0x1250, 0x00000001);
   1012 	GFFB_WRITE_4(GFFB_PFIFO + 0x1254, 0x00000001);
   1013 	GFFB_WRITE_4(GFFB_PFIFO + 0x0500, 0x00000001);
   1014 
   1015 	GFFB_WRITE_4(GFFB_PMC + 0x8704, 1);
   1016 	GFFB_WRITE_4(GFFB_PMC + 0x8140, 0);
   1017 	GFFB_WRITE_4(GFFB_PMC + 0x8920, 0);
   1018 	GFFB_WRITE_4(GFFB_PMC + 0x8924, 0);
   1019 	GFFB_WRITE_4(GFFB_PMC + 0x8908, sc->sc_vramsize - 1);
   1020 	GFFB_WRITE_4(GFFB_PMC + 0x890C, sc->sc_vramsize - 1);
   1021 	GFFB_WRITE_4(GFFB_PMC + 0x1588, 0);
   1022 
   1023 	__asm("eieio; sync;");
   1024 	GFFB_WRITE_4(GFFB_FIFO_GET, 0);
   1025 	GFFB_WRITE_4(GFFB_CMDSTART, 0x00000002);
   1026 	sc->sc_put = 0;
   1027 	sc->sc_current = 0;
   1028 	sc->sc_free = 0x2000;
   1029 
   1030 	for(i = 0; i < SKIPS; i++)
   1031 		gffb_dmanext(sc, 0);
   1032 
   1033 	gffb_dmanext(sc, 0x00040000);
   1034 	gffb_dmanext(sc, 0x80000010);
   1035 	gffb_dmanext(sc, 0x00042000);
   1036 	gffb_dmanext(sc, 0x80000011);
   1037 	gffb_dmanext(sc, 0x00044000);
   1038 	gffb_dmanext(sc, 0x80000012);
   1039 	gffb_dmanext(sc, 0x00046000);
   1040 	gffb_dmanext(sc, 0x80000013);
   1041 	gffb_dmanext(sc, 0x00048000);
   1042 	gffb_dmanext(sc, 0x80000014);
   1043 	gffb_dmanext(sc, 0x0004A000);
   1044 	gffb_dmanext(sc, 0x80000015);
   1045 	gffb_dmanext(sc, 0x0004C000);
   1046 	gffb_dmanext(sc, 0x80000016);
   1047 	gffb_dmanext(sc, 0x0004E000);
   1048 	gffb_dmanext(sc, 0x80000017);
   1049 	sc->sc_free = 0x2000 - sc->sc_current;
   1050 
   1051 	gffb_dmastart(sc, SURFACE_FORMAT, 4);
   1052 	gffb_dmanext(sc, SURFACE_FORMAT_DEPTH8);
   1053 	gffb_dmanext(sc, sc->sc_stride | (sc->sc_stride << 16));
   1054 	gffb_dmanext(sc, sc->sc_fboffset);	/* src offset */
   1055 	gffb_dmanext(sc, sc->sc_fboffset);	/* dst offset */
   1056 
   1057 	gffb_dmastart(sc, RECT_FORMAT, 1);
   1058 	gffb_dmanext(sc, RECT_FORMAT_DEPTH8);
   1059 
   1060 	gffb_dmastart(sc, PATTERN_FORMAT, 1);
   1061 	gffb_dmanext(sc, PATTERN_FORMAT_DEPTH8);
   1062 
   1063 	gffb_dmastart(sc, PATTERN_COLOR_0, 4);
   1064 	gffb_dmanext(sc, 0xffffffff);
   1065 	gffb_dmanext(sc, 0xffffffff);
   1066 	gffb_dmanext(sc, 0xffffffff);
   1067 	gffb_dmanext(sc, 0xffffffff);
   1068 
   1069 	gffb_dmastart(sc, ROP_SET, 1);
   1070 	gffb_dmanext(sc, 0xcc);
   1071 	sc->sc_rop = 0xcc;
   1072 	DPRINTF("put %x current %x\n", sc->sc_put, sc->sc_current);
   1073 
   1074 	gffb_dma_kickoff(sc);
   1075 	DPRINTF("put %x current %x\n", sc->sc_put, sc->sc_current);
   1076 #ifdef GFFB_DEBUG
   1077 	printf("put: %08x\n", GFFB_READ_4(GFFB_FIFO_PUT));
   1078 	printf("get: %08x\n", GFFB_READ_4(GFFB_FIFO_GET));
   1079 #endif
   1080 	gffb_sync(sc);
   1081 	DPRINTF("put %x current %x\n", sc->sc_put, sc->sc_current);
   1082 }
   1083 
   1084 static void
   1085 gffb_rop(struct gffb_softc *sc, int rop)
   1086 {
   1087 	if (rop == sc->sc_rop)
   1088 		return;
   1089 	sc->sc_rop = rop;
   1090 	gffb_dmastart(sc, ROP_SET, 1);
   1091 	gffb_dmanext(sc, rop);
   1092 }
   1093 
   1094 static void
   1095 gffb_rectfill(struct gffb_softc *sc, int x, int y, int wi, int he,
   1096      uint32_t colour)
   1097 {
   1098 	if (!sc->sc_accel) return;
   1099 	mutex_enter(&sc->sc_lock);
   1100 	gffb_rop(sc, 0xcc);
   1101 
   1102 	gffb_dmastart(sc, RECT_SOLID_COLOR, 1);
   1103 	gffb_dmanext(sc, colour);
   1104 
   1105 	gffb_dmastart(sc, RECT_SOLID_RECTS(0), 2);
   1106 	gffb_dmanext(sc, (x << 16) | y);
   1107 	gffb_dmanext(sc, (wi << 16) | he);
   1108 	gffb_dma_kickoff(sc);
   1109 	mutex_exit(&sc->sc_lock);
   1110 }
   1111 
   1112 static void
   1113 gffb_bitblt(void *cookie, int xs, int ys, int xd, int yd,
   1114     int wi, int he, int rop)
   1115 {
   1116 	struct gffb_softc *sc = cookie;
   1117 
   1118 	if (!sc->sc_accel) return;
   1119 	mutex_enter(&sc->sc_lock);
   1120 
   1121 	gffb_rop(sc, rop);
   1122 
   1123 	gffb_dmastart(sc, BLIT_POINT_SRC, 3);
   1124 	gffb_dmanext(sc, (ys << 16) | xs);
   1125 	gffb_dmanext(sc, (yd << 16) | xd);
   1126 	gffb_dmanext(sc, (he << 16) | wi);
   1127 	gffb_dma_kickoff(sc);
   1128 	mutex_exit(&sc->sc_lock);
   1129 }
   1130 
   1131 static void
   1132 gffb_cursor(void *cookie, int on, int row, int col)
   1133 {
   1134 	struct rasops_info *ri = cookie;
   1135 	struct vcons_screen *scr = ri->ri_hw;
   1136 	struct gffb_softc *sc = scr->scr_cookie;
   1137 	int x, y, wi, he;
   1138 
   1139 	wi = ri->ri_font->fontwidth;
   1140 	he = ri->ri_font->fontheight;
   1141 
   1142 	if (sc->sc_mode == WSDISPLAYIO_MODE_EMUL) {
   1143 		x = ri->ri_ccol * wi + ri->ri_xorigin;
   1144 		y = ri->ri_crow * he + ri->ri_yorigin;
   1145 		if (ri->ri_flg & RI_CURSOR) {
   1146 			gffb_bitblt(sc, x, y, x, y, wi, he, 0x33);
   1147 			ri->ri_flg &= ~RI_CURSOR;
   1148 		}
   1149 		ri->ri_crow = row;
   1150 		ri->ri_ccol = col;
   1151 		if (on) {
   1152 			x = ri->ri_ccol * wi + ri->ri_xorigin;
   1153 			y = ri->ri_crow * he + ri->ri_yorigin;
   1154 			gffb_bitblt(sc, x, y, x, y, wi, he, 0x33);
   1155 			ri->ri_flg |= RI_CURSOR;
   1156 		}
   1157 	} else {
   1158 		scr->scr_ri.ri_crow = row;
   1159 		scr->scr_ri.ri_ccol = col;
   1160 		scr->scr_ri.ri_flg &= ~RI_CURSOR;
   1161 	}
   1162 
   1163 }
   1164 
   1165 static void
   1166 gffb_putchar(void *cookie, int row, int col, u_int c, long attr)
   1167 {
   1168 	struct rasops_info *ri = cookie;
   1169 	struct wsdisplay_font *font = PICK_FONT(ri, c);
   1170 	struct vcons_screen *scr = ri->ri_hw;
   1171 	struct gffb_softc *sc = scr->scr_cookie;
   1172 	int x, y, wi, he, rv = GC_NOPE;
   1173 	uint32_t bg;
   1174 
   1175 	if (sc->sc_mode != WSDISPLAYIO_MODE_EMUL)
   1176 		return;
   1177 
   1178 	if (!CHAR_IN_FONT(c, font))
   1179 		return;
   1180 
   1181 	wi = font->fontwidth;
   1182 	he = font->fontheight;
   1183 
   1184 	x = ri->ri_xorigin + col * wi;
   1185 	y = ri->ri_yorigin + row * he;
   1186 	bg = ri->ri_devcmap[(attr >> 16) & 0xf];
   1187 
   1188 	if (c == 0x20) {
   1189 		gffb_rectfill(sc, x, y, wi, he, bg);
   1190 		return;
   1191 	}
   1192 	rv = glyphcache_try(&sc->sc_gc, c, x, y, attr);
   1193 	if (rv == GC_OK)
   1194 		return;
   1195 
   1196 	/*
   1197 	 * Use gffb_sync to wait for the engine to become idle before
   1198 	 * we start scribbling into VRAM -- we wouldn't want to stomp on
   1199 	 * a scroll in progress or a prior glyphcache_add that hasn't
   1200 	 * completed yet on the GPU.
   1201 	 */
   1202 	mutex_enter(&sc->sc_lock);
   1203 	gffb_sync(sc);
   1204 	sc->sc_putchar(cookie, row, col, c, attr);
   1205 	mutex_exit(&sc->sc_lock);
   1206 
   1207 	/*
   1208 	 * If glyphcache_try asked us to, cache the newly written
   1209 	 * character.  This will issue a gffb_bitblt which will wait
   1210 	 * for our CPU writes to the framebuffer in VRAM to complete
   1211 	 * before triggering GPU reads from the framebuffer in VRAM.
   1212 	 */
   1213 	if (rv == GC_ADD) {
   1214 		glyphcache_add(&sc->sc_gc, c, x, y);
   1215 	}
   1216 }
   1217 
   1218 static void
   1219 gffb_copycols(void *cookie, int row, int srccol, int dstcol, int ncols)
   1220 {
   1221 	struct rasops_info *ri = cookie;
   1222 	struct vcons_screen *scr = ri->ri_hw;
   1223 	struct gffb_softc *sc = scr->scr_cookie;
   1224 	int32_t xs, xd, y, width, height;
   1225 
   1226 	if ((sc->sc_locked == 0) && (sc->sc_mode == WSDISPLAYIO_MODE_EMUL)) {
   1227 		xs = ri->ri_xorigin + ri->ri_font->fontwidth * srccol;
   1228 		xd = ri->ri_xorigin + ri->ri_font->fontwidth * dstcol;
   1229 		y = ri->ri_yorigin + ri->ri_font->fontheight * row;
   1230 		width = ri->ri_font->fontwidth * ncols;
   1231 		height = ri->ri_font->fontheight;
   1232 		gffb_bitblt(sc, xs, y, xd, y, width, height, 0xcc);
   1233 	}
   1234 }
   1235 
   1236 static void
   1237 gffb_erasecols(void *cookie, int row, int startcol, int ncols, long fillattr)
   1238 {
   1239 	struct rasops_info *ri = cookie;
   1240 	struct vcons_screen *scr = ri->ri_hw;
   1241 	struct gffb_softc *sc = scr->scr_cookie;
   1242 	int32_t x, y, width, height, fg, bg, ul;
   1243 
   1244 	if ((sc->sc_locked == 0) && (sc->sc_mode == WSDISPLAYIO_MODE_EMUL)) {
   1245 		x = ri->ri_xorigin + ri->ri_font->fontwidth * startcol;
   1246 		y = ri->ri_yorigin + ri->ri_font->fontheight * row;
   1247 		width = ri->ri_font->fontwidth * ncols;
   1248 		height = ri->ri_font->fontheight;
   1249 		rasops_unpack_attr(fillattr, &fg, &bg, &ul);
   1250 
   1251 		gffb_rectfill(sc, x, y, width, height, ri->ri_devcmap[bg]);
   1252 	}
   1253 }
   1254 
   1255 static void
   1256 gffb_copyrows(void *cookie, int srcrow, int dstrow, int nrows)
   1257 {
   1258 	struct rasops_info *ri = cookie;
   1259 	struct vcons_screen *scr = ri->ri_hw;
   1260 	struct gffb_softc *sc = scr->scr_cookie;
   1261 	int32_t x, ys, yd, width, height;
   1262 
   1263 	if ((sc->sc_locked == 0) && (sc->sc_mode == WSDISPLAYIO_MODE_EMUL)) {
   1264 		x = ri->ri_xorigin;
   1265 		ys = ri->ri_yorigin + ri->ri_font->fontheight * srcrow;
   1266 		yd = ri->ri_yorigin + ri->ri_font->fontheight * dstrow;
   1267 		width = ri->ri_emuwidth;
   1268 		height = ri->ri_font->fontheight * nrows;
   1269 		gffb_bitblt(sc, x, ys, x, yd, width, height, 0xcc);
   1270 	}
   1271 }
   1272 
   1273 static void
   1274 gffb_eraserows(void *cookie, int row, int nrows, long fillattr)
   1275 {
   1276 	struct rasops_info *ri = cookie;
   1277 	struct vcons_screen *scr = ri->ri_hw;
   1278 	struct gffb_softc *sc = scr->scr_cookie;
   1279 	int32_t x, y, width, height, fg, bg, ul;
   1280 
   1281 	if ((sc->sc_locked == 0) && (sc->sc_mode == WSDISPLAYIO_MODE_EMUL)) {
   1282 		x = ri->ri_xorigin;
   1283 		y = ri->ri_yorigin + ri->ri_font->fontheight * row;
   1284 		width = ri->ri_emuwidth;
   1285 		height = ri->ri_font->fontheight * nrows;
   1286 		rasops_unpack_attr(fillattr, &fg, &bg, &ul);
   1287 
   1288 		gffb_rectfill(sc, x, y, width, height, ri->ri_devcmap[bg]);
   1289 	}
   1290 }
   1291