xref: /freebsd/sys/dev/agp/agp_nvidia.c (revision b9f654b163bce26de79705e77b872427c9f2afa1)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2003 Matthew N. Dodd <winter@jurai.net>
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 AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 /*
33  * Written using information gleaned from the
34  * NVIDIA nForce/nForce2 AGPGART Linux Kernel Patch.
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
40 #include <sys/kernel.h>
41 #include <sys/module.h>
42 #include <sys/bus.h>
43 #include <sys/lock.h>
44 
45 #if __FreeBSD_version < 500000
46 #include "opt_pci.h"
47 #endif
48 
49 #if __FreeBSD_version > 500000
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #endif
53 
54 #include <dev/agp/agppriv.h>
55 #include <dev/agp/agpreg.h>
56 #include <dev/pci/pcivar.h>
57 #include <dev/pci/pcireg.h>
58 
59 #include <vm/vm.h>
60 #include <vm/vm_object.h>
61 #include <vm/pmap.h>
62 
63 #include <machine/bus.h>
64 #include <machine/resource.h>
65 #include <sys/rman.h>
66 
67 #define	NVIDIA_VENDORID		0x10de
68 #define	NVIDIA_DEVICEID_NFORCE	0x01a4
69 #define	NVIDIA_DEVICEID_NFORCE2	0x01e0
70 
71 struct agp_nvidia_softc {
72 	struct agp_softc	agp;
73 	u_int32_t		initial_aperture; /* aperture size at startup */
74 	struct agp_gatt *	gatt;
75 
76 	device_t		dev;		/* AGP Controller */
77 	device_t		mc1_dev;	/* Memory Controller */
78 	device_t		mc2_dev;	/* Memory Controller */
79 	device_t		bdev;		/* Bridge */
80 
81 	u_int32_t		wbc_mask;
82 	int			num_dirs;
83 	int			num_active_entries;
84 	off_t			pg_offset;
85 };
86 
87 static const char *agp_nvidia_match(device_t dev);
88 static int agp_nvidia_probe(device_t);
89 static int agp_nvidia_attach(device_t);
90 static int agp_nvidia_detach(device_t);
91 static u_int32_t agp_nvidia_get_aperture(device_t);
92 static int agp_nvidia_set_aperture(device_t, u_int32_t);
93 static int agp_nvidia_bind_page(device_t, vm_offset_t, vm_offset_t);
94 static int agp_nvidia_unbind_page(device_t, vm_offset_t);
95 
96 static int nvidia_init_iorr(u_int32_t, u_int32_t);
97 
98 static const char *
99 agp_nvidia_match (device_t dev)
100 {
101 	if (pci_get_class(dev) != PCIC_BRIDGE ||
102 	    pci_get_subclass(dev) != PCIS_BRIDGE_HOST ||
103 	    pci_get_vendor(dev) != NVIDIA_VENDORID)
104 		return (NULL);
105 
106 	switch (pci_get_device(dev)) {
107 	case NVIDIA_DEVICEID_NFORCE:
108 		return ("NVIDIA nForce AGP Controller");
109 	case NVIDIA_DEVICEID_NFORCE2:
110 		return ("NVIDIA nForce2 AGP Controller");
111 	}
112 	return (NULL);
113 }
114 
115 static int
116 agp_nvidia_probe (device_t dev)
117 {
118 	const char *desc;
119 
120 	if (resource_disabled("agp", device_get_unit(dev)))
121 		return (ENXIO);
122 	desc = agp_nvidia_match(dev);
123 	if (desc) {
124 		device_set_desc(dev, desc);
125 		return (BUS_PROBE_DEFAULT);
126 	}
127 	return (ENXIO);
128 }
129 
130 static int
131 agp_nvidia_attach (device_t dev)
132 {
133 	struct agp_nvidia_softc *sc = device_get_softc(dev);
134 	struct agp_gatt *gatt;
135 	u_int32_t apbase;
136 	u_int32_t aplimit;
137 	u_int32_t temp;
138 	int size;
139 	int i;
140 	int error;
141 
142 	switch (pci_get_device(dev)) {
143 	case NVIDIA_DEVICEID_NFORCE:
144 		sc->wbc_mask = 0x00010000;
145 		break;
146 	case NVIDIA_DEVICEID_NFORCE2:
147 		sc->wbc_mask = 0x80000000;
148 		break;
149 	default:
150 		device_printf(dev, "Bad chip id\n");
151 		return (ENODEV);
152 	}
153 
154 	/* AGP Controller */
155 	sc->dev = dev;
156 
157 	/* Memory Controller 1 */
158 	sc->mc1_dev = pci_find_bsf(pci_get_bus(dev), 0, 1);
159 	if (sc->mc1_dev == NULL) {
160 		device_printf(dev,
161 			"Unable to find NVIDIA Memory Controller 1.\n");
162 		return (ENODEV);
163 	}
164 
165 	/* Memory Controller 2 */
166 	sc->mc2_dev = pci_find_bsf(pci_get_bus(dev), 0, 2);
167 	if (sc->mc2_dev == NULL) {
168 		device_printf(dev,
169 			"Unable to find NVIDIA Memory Controller 2.\n");
170 		return (ENODEV);
171 	}
172 
173 	/* AGP Host to PCI Bridge */
174 	sc->bdev = pci_find_bsf(pci_get_bus(dev), 30, 0);
175 	if (sc->bdev == NULL) {
176 		device_printf(dev,
177 			"Unable to find NVIDIA AGP Host to PCI Bridge.\n");
178 		return (ENODEV);
179 	}
180 
181 	error = agp_generic_attach(dev);
182 	if (error)
183 		return (error);
184 
185 	sc->initial_aperture = AGP_GET_APERTURE(dev);
186 
187 	for (;;) {
188 		gatt = agp_alloc_gatt(dev);
189 		if (gatt)
190 			break;
191 		/*
192 		 * Probably contigmalloc failure. Try reducing the
193 		 * aperture so that the gatt size reduces.
194 		 */
195 		if (AGP_SET_APERTURE(dev, AGP_GET_APERTURE(dev) / 2))
196 			goto fail;
197 	}
198 	sc->gatt = gatt;
199 
200 	apbase = rman_get_start(sc->agp.as_aperture);
201 	aplimit = apbase + AGP_GET_APERTURE(dev) - 1;
202 	pci_write_config(sc->mc2_dev, AGP_NVIDIA_2_APBASE, apbase, 4);
203 	pci_write_config(sc->mc2_dev, AGP_NVIDIA_2_APLIMIT, aplimit, 4);
204 	pci_write_config(sc->bdev, AGP_NVIDIA_3_APBASE, apbase, 4);
205 	pci_write_config(sc->bdev, AGP_NVIDIA_3_APLIMIT, aplimit, 4);
206 
207 	error = nvidia_init_iorr(apbase, AGP_GET_APERTURE(dev));
208 	if (error) {
209 		device_printf(dev, "Failed to setup IORRs\n");
210 		goto fail;
211 	}
212 
213 	/* directory size is 64k */
214 	size = AGP_GET_APERTURE(dev) / 1024 / 1024;
215 	sc->num_dirs = size / 64;
216 	sc->num_active_entries = (size == 32) ? 16384 : ((size * 1024) / 4);
217 	sc->pg_offset = 0;
218 	if (sc->num_dirs == 0) {
219 		sc->num_dirs = 1;
220 		sc->num_active_entries /= (64 / size);
221 		sc->pg_offset = rounddown2(apbase & (64 * 1024 * 1024 - 1),
222 		    AGP_GET_APERTURE(dev)) / PAGE_SIZE;
223 	}
224 
225 	/* (G)ATT Base Address */
226 	for (i = 0; i < 8; i++) {
227 		pci_write_config(sc->mc2_dev, AGP_NVIDIA_2_ATTBASE(i),
228 				 (sc->gatt->ag_physical +
229 				   (i % sc->num_dirs) * 64 * 1024) | 1, 4);
230 	}
231 
232 	/* GTLB Control */
233 	temp = pci_read_config(sc->mc2_dev, AGP_NVIDIA_2_GARTCTRL, 4);
234 	pci_write_config(sc->mc2_dev, AGP_NVIDIA_2_GARTCTRL, temp | 0x11, 4);
235 
236 	/* GART Control */
237 	temp = pci_read_config(sc->dev, AGP_NVIDIA_0_APSIZE, 4);
238 	pci_write_config(sc->dev, AGP_NVIDIA_0_APSIZE, temp | 0x100, 4);
239 
240 	return (0);
241 fail:
242 	agp_generic_detach(dev);
243 	return (ENOMEM);
244 }
245 
246 static int
247 agp_nvidia_detach (device_t dev)
248 {
249 	struct agp_nvidia_softc *sc = device_get_softc(dev);
250 	u_int32_t temp;
251 
252 	agp_free_cdev(dev);
253 
254 	/* GART Control */
255 	temp = pci_read_config(sc->dev, AGP_NVIDIA_0_APSIZE, 4);
256 	pci_write_config(sc->dev, AGP_NVIDIA_0_APSIZE, temp & ~(0x100), 4);
257 
258 	/* GTLB Control */
259 	temp = pci_read_config(sc->mc2_dev, AGP_NVIDIA_2_GARTCTRL, 4);
260 	pci_write_config(sc->mc2_dev, AGP_NVIDIA_2_GARTCTRL, temp & ~(0x11), 4);
261 
262 	/* Put the aperture back the way it started. */
263 	AGP_SET_APERTURE(dev, sc->initial_aperture);
264 
265 	/* restore iorr for previous aperture size */
266 	nvidia_init_iorr(rman_get_start(sc->agp.as_aperture),
267 			 sc->initial_aperture);
268 
269 	agp_free_gatt(sc->gatt);
270 	agp_free_res(dev);
271 
272 	return (0);
273 }
274 
275 static u_int32_t
276 agp_nvidia_get_aperture(device_t dev)
277 {
278 	switch (pci_read_config(dev, AGP_NVIDIA_0_APSIZE, 1) & 0x0f) {
279 	case 0: return (512 * 1024 * 1024);
280 	case 8: return (256 * 1024 * 1024);
281 	case 12: return (128 * 1024 * 1024);
282 	case 14: return (64 * 1024 * 1024);
283 	case 15: return (32 * 1024 * 1024);
284 	default:
285 		device_printf(dev, "Invalid aperture setting 0x%x\n",
286 		    pci_read_config(dev, AGP_NVIDIA_0_APSIZE, 1));
287 		return 0;
288 	}
289 }
290 
291 static int
292 agp_nvidia_set_aperture(device_t dev, u_int32_t aperture)
293 {
294 	u_int8_t val;
295 	u_int8_t key;
296 
297 	switch (aperture) {
298 	case (512 * 1024 * 1024): key = 0; break;
299 	case (256 * 1024 * 1024): key = 8; break;
300 	case (128 * 1024 * 1024): key = 12; break;
301 	case (64 * 1024 * 1024): key = 14; break;
302 	case (32 * 1024 * 1024): key = 15; break;
303 	default:
304 		device_printf(dev, "Invalid aperture size (%dMb)\n",
305 				aperture / 1024 / 1024);
306 		return (EINVAL);
307 	}
308 	val = pci_read_config(dev, AGP_NVIDIA_0_APSIZE, 1);
309 	pci_write_config(dev, AGP_NVIDIA_0_APSIZE, ((val & ~0x0f) | key), 1);
310 
311 	return (0);
312 }
313 
314 static int
315 agp_nvidia_bind_page(device_t dev, vm_offset_t offset, vm_offset_t physical)
316 {
317 	struct agp_nvidia_softc *sc = device_get_softc(dev);
318 	u_int32_t index;
319 
320 	if (offset >= (sc->gatt->ag_entries << AGP_PAGE_SHIFT))
321 		return (EINVAL);
322 
323 	index = (sc->pg_offset + offset) >> AGP_PAGE_SHIFT;
324 	sc->gatt->ag_virtual[index] = physical | 1;
325 
326 	return (0);
327 }
328 
329 static int
330 agp_nvidia_unbind_page(device_t dev, vm_offset_t offset)
331 {
332 	struct agp_nvidia_softc *sc = device_get_softc(dev);
333 	u_int32_t index;
334 
335 	if (offset >= (sc->gatt->ag_entries << AGP_PAGE_SHIFT))
336 		return (EINVAL);
337 
338 	index = (sc->pg_offset + offset) >> AGP_PAGE_SHIFT;
339 	sc->gatt->ag_virtual[index] = 0;
340 
341 	return (0);
342 }
343 
344 static void
345 agp_nvidia_flush_tlb (device_t dev)
346 {
347 	struct agp_nvidia_softc *sc;
348 	u_int32_t wbc_reg, temp;
349 	volatile u_int32_t *ag_virtual;
350 	int i, pages;
351 
352 	sc = (struct agp_nvidia_softc *)device_get_softc(dev);
353 
354 	if (sc->wbc_mask) {
355 		wbc_reg = pci_read_config(sc->mc1_dev, AGP_NVIDIA_1_WBC, 4);
356 		wbc_reg |= sc->wbc_mask;
357 		pci_write_config(sc->mc1_dev, AGP_NVIDIA_1_WBC, wbc_reg, 4);
358 
359 		/* Wait no more than 3 seconds. */
360 		for (i = 0; i < 3000; i++) {
361 			wbc_reg = pci_read_config(sc->mc1_dev,
362 						  AGP_NVIDIA_1_WBC, 4);
363 			if ((sc->wbc_mask & wbc_reg) == 0)
364 				break;
365 			else
366 				DELAY(1000);
367 		}
368 		if (i == 3000)
369 			device_printf(dev,
370 				"TLB flush took more than 3 seconds.\n");
371 	}
372 
373 	ag_virtual = (volatile u_int32_t *)sc->gatt->ag_virtual;
374 
375 	/* Flush TLB entries. */
376 	pages = sc->gatt->ag_entries * sizeof(u_int32_t) / PAGE_SIZE;
377 	for(i = 0; i < pages; i++)
378 		temp = ag_virtual[i * PAGE_SIZE / sizeof(u_int32_t)];
379 	for(i = 0; i < pages; i++)
380 		temp = ag_virtual[i * PAGE_SIZE / sizeof(u_int32_t)];
381 }
382 
383 #define	SYSCFG		0xC0010010
384 #define	IORR_BASE0	0xC0010016
385 #define	IORR_MASK0	0xC0010017
386 #define	AMD_K7_NUM_IORR	2
387 
388 static int
389 nvidia_init_iorr(u_int32_t addr, u_int32_t size)
390 {
391 	quad_t base, mask, sys;
392 	u_int32_t iorr_addr, free_iorr_addr;
393 
394 	/* Find the iorr that is already used for the addr */
395 	/* If not found, determine the uppermost available iorr */
396 	free_iorr_addr = AMD_K7_NUM_IORR;
397 	for(iorr_addr = 0; iorr_addr < AMD_K7_NUM_IORR; iorr_addr++) {
398 		base = rdmsr(IORR_BASE0 + 2 * iorr_addr);
399 		mask = rdmsr(IORR_MASK0 + 2 * iorr_addr);
400 
401 		if ((base & 0xfffff000ULL) == (addr & 0xfffff000))
402 			break;
403 
404 		if ((mask & 0x00000800ULL) == 0)
405 			free_iorr_addr = iorr_addr;
406 	}
407 
408 	if (iorr_addr >= AMD_K7_NUM_IORR) {
409 		iorr_addr = free_iorr_addr;
410 		if (iorr_addr >= AMD_K7_NUM_IORR)
411 			return (EINVAL);
412 	}
413 
414 	base = (addr & ~0xfff) | 0x18;
415 	mask = (0xfULL << 32) | rounddown2(0xfffff000, size) | 0x800;
416 	wrmsr(IORR_BASE0 + 2 * iorr_addr, base);
417 	wrmsr(IORR_MASK0 + 2 * iorr_addr, mask);
418 
419 	sys = rdmsr(SYSCFG);
420 	sys |= 0x00100000ULL;
421 	wrmsr(SYSCFG, sys);
422 
423 	return (0);
424 }
425 
426 static device_method_t agp_nvidia_methods[] = {
427 	/* Device interface */
428 	DEVMETHOD(device_probe,		agp_nvidia_probe),
429 	DEVMETHOD(device_attach,	agp_nvidia_attach),
430 	DEVMETHOD(device_detach,	agp_nvidia_detach),
431 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
432 	DEVMETHOD(device_suspend,	bus_generic_suspend),
433 	DEVMETHOD(device_resume,	bus_generic_resume),
434 
435 	/* AGP interface */
436 	DEVMETHOD(agp_get_aperture,	agp_nvidia_get_aperture),
437 	DEVMETHOD(agp_set_aperture,	agp_nvidia_set_aperture),
438 	DEVMETHOD(agp_bind_page,	agp_nvidia_bind_page),
439 	DEVMETHOD(agp_unbind_page,	agp_nvidia_unbind_page),
440 	DEVMETHOD(agp_flush_tlb,	agp_nvidia_flush_tlb),
441 
442 	DEVMETHOD(agp_enable,		agp_generic_enable),
443 	DEVMETHOD(agp_alloc_memory,	agp_generic_alloc_memory),
444 	DEVMETHOD(agp_free_memory,	agp_generic_free_memory),
445 	DEVMETHOD(agp_bind_memory,	agp_generic_bind_memory),
446 	DEVMETHOD(agp_unbind_memory,	agp_generic_unbind_memory),
447 
448 	{ 0, 0 }
449 };
450 
451 static driver_t agp_nvidia_driver = {
452 	"agp",
453 	agp_nvidia_methods,
454 	sizeof(struct agp_nvidia_softc),
455 };
456 
457 static devclass_t agp_devclass;
458 
459 DRIVER_MODULE(agp_nvidia, hostb, agp_nvidia_driver, agp_devclass, 0, 0);
460 MODULE_DEPEND(agp_nvidia, agp, 1, 1, 1);
461 MODULE_DEPEND(agp_nvidia, pci, 1, 1, 1);
462