blob: 4c67135c12d853052b389657d1fba450e1c3dfe4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Nvidia AGPGART routines.
3 * Based upon a 2.4 agpgart diff by the folks from NVIDIA, and hacked up
4 * to work in 2.5 by Dave Jones <davej@codemonkey.org.uk>
5 */
6
7#include <linux/module.h>
8#include <linux/pci.h>
9#include <linux/init.h>
10#include <linux/agp_backend.h>
11#include <linux/gfp.h>
12#include <linux/page-flags.h>
13#include <linux/mm.h>
Marcelo Feitoza Parisi05613bd2006-03-28 01:56:45 -080014#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include "agp.h"
16
17/* NVIDIA registers */
18#define NVIDIA_0_APSIZE 0x80
19#define NVIDIA_1_WBC 0xf0
20#define NVIDIA_2_GARTCTRL 0xd0
21#define NVIDIA_2_APBASE 0xd8
22#define NVIDIA_2_APLIMIT 0xdc
23#define NVIDIA_2_ATTBASE(i) (0xe0 + (i) * 4)
24#define NVIDIA_3_APBASE 0x50
25#define NVIDIA_3_APLIMIT 0x54
26
27
28static struct _nvidia_private {
29 struct pci_dev *dev_1;
30 struct pci_dev *dev_2;
31 struct pci_dev *dev_3;
32 volatile u32 __iomem *aperture;
33 int num_active_entries;
34 off_t pg_offset;
35 u32 wbc_mask;
36} nvidia_private;
37
38
39static int nvidia_fetch_size(void)
40{
41 int i;
42 u8 size_value;
43 struct aper_size_info_8 *values;
44
45 pci_read_config_byte(agp_bridge->dev, NVIDIA_0_APSIZE, &size_value);
46 size_value &= 0x0f;
47 values = A_SIZE_8(agp_bridge->driver->aperture_sizes);
48
49 for (i = 0; i < agp_bridge->driver->num_aperture_sizes; i++) {
50 if (size_value == values[i].size_value) {
51 agp_bridge->previous_size =
52 agp_bridge->current_size = (void *) (values + i);
53 agp_bridge->aperture_size_idx = i;
54 return values[i].size;
55 }
56 }
57
58 return 0;
59}
60
61#define SYSCFG 0xC0010010
62#define IORR_BASE0 0xC0010016
63#define IORR_MASK0 0xC0010017
64#define AMD_K7_NUM_IORR 2
65
66static int nvidia_init_iorr(u32 base, u32 size)
67{
68 u32 base_hi, base_lo;
69 u32 mask_hi, mask_lo;
70 u32 sys_hi, sys_lo;
71 u32 iorr_addr, free_iorr_addr;
72
73 /* Find the iorr that is already used for the base */
74 /* If not found, determine the uppermost available iorr */
75 free_iorr_addr = AMD_K7_NUM_IORR;
Dave Jones6a92a4e2006-02-28 00:54:25 -050076 for (iorr_addr = 0; iorr_addr < AMD_K7_NUM_IORR; iorr_addr++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 rdmsr(IORR_BASE0 + 2 * iorr_addr, base_lo, base_hi);
78 rdmsr(IORR_MASK0 + 2 * iorr_addr, mask_lo, mask_hi);
79
80 if ((base_lo & 0xfffff000) == (base & 0xfffff000))
81 break;
82
83 if ((mask_lo & 0x00000800) == 0)
84 free_iorr_addr = iorr_addr;
85 }
Dave Jones6a92a4e2006-02-28 00:54:25 -050086
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 if (iorr_addr >= AMD_K7_NUM_IORR) {
88 iorr_addr = free_iorr_addr;
89 if (iorr_addr >= AMD_K7_NUM_IORR)
90 return -EINVAL;
91 }
92 base_hi = 0x0;
93 base_lo = (base & ~0xfff) | 0x18;
94 mask_hi = 0xf;
95 mask_lo = ((~(size - 1)) & 0xfffff000) | 0x800;
96 wrmsr(IORR_BASE0 + 2 * iorr_addr, base_lo, base_hi);
97 wrmsr(IORR_MASK0 + 2 * iorr_addr, mask_lo, mask_hi);
98
99 rdmsr(SYSCFG, sys_lo, sys_hi);
100 sys_lo |= 0x00100000;
101 wrmsr(SYSCFG, sys_lo, sys_hi);
102
103 return 0;
104}
105
106static int nvidia_configure(void)
107{
108 int i, rc, num_dirs;
109 u32 apbase, aplimit;
110 struct aper_size_info_8 *current_size;
111 u32 temp;
112
113 current_size = A_SIZE_8(agp_bridge->current_size);
114
115 /* aperture size */
116 pci_write_config_byte(agp_bridge->dev, NVIDIA_0_APSIZE,
117 current_size->size_value);
118
119 /* address to map to */
120 pci_read_config_dword(agp_bridge->dev, AGP_APBASE, &apbase);
121 apbase &= PCI_BASE_ADDRESS_MEM_MASK;
122 agp_bridge->gart_bus_addr = apbase;
123 aplimit = apbase + (current_size->size * 1024 * 1024) - 1;
124 pci_write_config_dword(nvidia_private.dev_2, NVIDIA_2_APBASE, apbase);
125 pci_write_config_dword(nvidia_private.dev_2, NVIDIA_2_APLIMIT, aplimit);
126 pci_write_config_dword(nvidia_private.dev_3, NVIDIA_3_APBASE, apbase);
127 pci_write_config_dword(nvidia_private.dev_3, NVIDIA_3_APLIMIT, aplimit);
128 if (0 != (rc = nvidia_init_iorr(apbase, current_size->size * 1024 * 1024)))
129 return rc;
130
131 /* directory size is 64k */
132 num_dirs = current_size->size / 64;
133 nvidia_private.num_active_entries = current_size->num_entries;
134 nvidia_private.pg_offset = 0;
135 if (num_dirs == 0) {
136 num_dirs = 1;
137 nvidia_private.num_active_entries /= (64 / current_size->size);
138 nvidia_private.pg_offset = (apbase & (64 * 1024 * 1024 - 1) &
139 ~(current_size->size * 1024 * 1024 - 1)) / PAGE_SIZE;
140 }
141
142 /* attbase */
Dave Jones6a92a4e2006-02-28 00:54:25 -0500143 for (i = 0; i < 8; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 pci_write_config_dword(nvidia_private.dev_2, NVIDIA_2_ATTBASE(i),
145 (agp_bridge->gatt_bus_addr + (i % num_dirs) * 64 * 1024) | 1);
146 }
147
148 /* gtlb control */
149 pci_read_config_dword(nvidia_private.dev_2, NVIDIA_2_GARTCTRL, &temp);
150 pci_write_config_dword(nvidia_private.dev_2, NVIDIA_2_GARTCTRL, temp | 0x11);
151
152 /* gart control */
153 pci_read_config_dword(agp_bridge->dev, NVIDIA_0_APSIZE, &temp);
154 pci_write_config_dword(agp_bridge->dev, NVIDIA_0_APSIZE, temp | 0x100);
155
156 /* map aperture */
157 nvidia_private.aperture =
158 (volatile u32 __iomem *) ioremap(apbase, 33 * PAGE_SIZE);
159
160 return 0;
161}
162
163static void nvidia_cleanup(void)
164{
165 struct aper_size_info_8 *previous_size;
166 u32 temp;
167
168 /* gart control */
169 pci_read_config_dword(agp_bridge->dev, NVIDIA_0_APSIZE, &temp);
170 pci_write_config_dword(agp_bridge->dev, NVIDIA_0_APSIZE, temp & ~(0x100));
171
172 /* gtlb control */
173 pci_read_config_dword(nvidia_private.dev_2, NVIDIA_2_GARTCTRL, &temp);
174 pci_write_config_dword(nvidia_private.dev_2, NVIDIA_2_GARTCTRL, temp & ~(0x11));
175
176 /* unmap aperture */
177 iounmap((void __iomem *) nvidia_private.aperture);
178
179 /* restore previous aperture size */
180 previous_size = A_SIZE_8(agp_bridge->previous_size);
181 pci_write_config_byte(agp_bridge->dev, NVIDIA_0_APSIZE,
182 previous_size->size_value);
183
184 /* restore iorr for previous aperture size */
185 nvidia_init_iorr(agp_bridge->gart_bus_addr,
186 previous_size->size * 1024 * 1024);
187}
188
189
190/*
191 * Note we can't use the generic routines, even though they are 99% the same.
192 * Aperture sizes <64M still requires a full 64k GART directory, but
193 * only use the portion of the TLB entries that correspond to the apertures
194 * alignment inside the surrounding 64M block.
195 */
196extern int agp_memory_reserved;
197
198static int nvidia_insert_memory(struct agp_memory *mem, off_t pg_start, int type)
199{
200 int i, j;
Dave Jones6a92a4e2006-02-28 00:54:25 -0500201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 if ((type != 0) || (mem->type != 0))
203 return -EINVAL;
Dave Jones6a92a4e2006-02-28 00:54:25 -0500204
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 if ((pg_start + mem->page_count) >
206 (nvidia_private.num_active_entries - agp_memory_reserved/PAGE_SIZE))
207 return -EINVAL;
Dave Jones6a92a4e2006-02-28 00:54:25 -0500208
209 for (j = pg_start; j < (pg_start + mem->page_count); j++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 if (!PGE_EMPTY(agp_bridge, readl(agp_bridge->gatt_table+nvidia_private.pg_offset+j)))
211 return -EBUSY;
212 }
213
214 if (mem->is_flushed == FALSE) {
215 global_cache_flush();
216 mem->is_flushed = TRUE;
217 }
218 for (i = 0, j = pg_start; i < mem->page_count; i++, j++) {
219 writel(agp_bridge->driver->mask_memory(agp_bridge,
220 mem->memory[i], mem->type),
221 agp_bridge->gatt_table+nvidia_private.pg_offset+j);
222 readl(agp_bridge->gatt_table+nvidia_private.pg_offset+j); /* PCI Posting. */
223 }
224 agp_bridge->driver->tlb_flush(mem);
225 return 0;
226}
227
228
229static int nvidia_remove_memory(struct agp_memory *mem, off_t pg_start, int type)
230{
231 int i;
232
233 if ((type != 0) || (mem->type != 0))
234 return -EINVAL;
235
236 for (i = pg_start; i < (mem->page_count + pg_start); i++)
237 writel(agp_bridge->scratch_page, agp_bridge->gatt_table+nvidia_private.pg_offset+i);
238
239 agp_bridge->driver->tlb_flush(mem);
240 return 0;
241}
242
243
244static void nvidia_tlbflush(struct agp_memory *mem)
245{
246 unsigned long end;
247 u32 wbc_reg, temp;
248 int i;
249
250 /* flush chipset */
251 if (nvidia_private.wbc_mask) {
252 pci_read_config_dword(nvidia_private.dev_1, NVIDIA_1_WBC, &wbc_reg);
253 wbc_reg |= nvidia_private.wbc_mask;
254 pci_write_config_dword(nvidia_private.dev_1, NVIDIA_1_WBC, wbc_reg);
255
256 end = jiffies + 3*HZ;
257 do {
258 pci_read_config_dword(nvidia_private.dev_1,
259 NVIDIA_1_WBC, &wbc_reg);
Marcelo Feitoza Parisi05613bd2006-03-28 01:56:45 -0800260 if (time_before_eq(end, jiffies)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 printk(KERN_ERR PFX
262 "TLB flush took more than 3 seconds.\n");
263 }
264 } while (wbc_reg & nvidia_private.wbc_mask);
265 }
266
267 /* flush TLB entries */
Dave Jones6a92a4e2006-02-28 00:54:25 -0500268 for (i = 0; i < 32 + 1; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 temp = readl(nvidia_private.aperture+(i * PAGE_SIZE / sizeof(u32)));
Dave Jones6a92a4e2006-02-28 00:54:25 -0500270 for (i = 0; i < 32 + 1; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 temp = readl(nvidia_private.aperture+(i * PAGE_SIZE / sizeof(u32)));
272}
273
274
275static struct aper_size_info_8 nvidia_generic_sizes[5] =
276{
277 {512, 131072, 7, 0},
278 {256, 65536, 6, 8},
279 {128, 32768, 5, 12},
280 {64, 16384, 4, 14},
281 /* The 32M mode still requires a 64k gatt */
282 {32, 16384, 4, 15}
283};
284
285
286static struct gatt_mask nvidia_generic_masks[] =
287{
288 { .mask = 1, .type = 0}
289};
290
291
Adrian Bunk408b6642005-05-01 08:59:29 -0700292static struct agp_bridge_driver nvidia_driver = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 .owner = THIS_MODULE,
294 .aperture_sizes = nvidia_generic_sizes,
295 .size_type = U8_APER_SIZE,
296 .num_aperture_sizes = 5,
297 .configure = nvidia_configure,
298 .fetch_size = nvidia_fetch_size,
299 .cleanup = nvidia_cleanup,
300 .tlb_flush = nvidia_tlbflush,
301 .mask_memory = agp_generic_mask_memory,
302 .masks = nvidia_generic_masks,
303 .agp_enable = agp_generic_enable,
304 .cache_flush = global_cache_flush,
305 .create_gatt_table = agp_generic_create_gatt_table,
306 .free_gatt_table = agp_generic_free_gatt_table,
307 .insert_memory = nvidia_insert_memory,
308 .remove_memory = nvidia_remove_memory,
309 .alloc_by_type = agp_generic_alloc_by_type,
310 .free_by_type = agp_generic_free_by_type,
311 .agp_alloc_page = agp_generic_alloc_page,
312 .agp_destroy_page = agp_generic_destroy_page,
313};
314
315static int __devinit agp_nvidia_probe(struct pci_dev *pdev,
316 const struct pci_device_id *ent)
317{
318 struct agp_bridge_data *bridge;
319 u8 cap_ptr;
320
321 nvidia_private.dev_1 =
322 pci_find_slot((unsigned int)pdev->bus->number, PCI_DEVFN(0, 1));
323 nvidia_private.dev_2 =
324 pci_find_slot((unsigned int)pdev->bus->number, PCI_DEVFN(0, 2));
325 nvidia_private.dev_3 =
326 pci_find_slot((unsigned int)pdev->bus->number, PCI_DEVFN(30, 0));
Dave Jones6a92a4e2006-02-28 00:54:25 -0500327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 if (!nvidia_private.dev_1 || !nvidia_private.dev_2 || !nvidia_private.dev_3) {
329 printk(KERN_INFO PFX "Detected an NVIDIA nForce/nForce2 "
330 "chipset, but could not find the secondary devices.\n");
331 return -ENODEV;
332 }
333
334 cap_ptr = pci_find_capability(pdev, PCI_CAP_ID_AGP);
335 if (!cap_ptr)
336 return -ENODEV;
337
338 switch (pdev->device) {
339 case PCI_DEVICE_ID_NVIDIA_NFORCE:
340 printk(KERN_INFO PFX "Detected NVIDIA nForce chipset\n");
341 nvidia_private.wbc_mask = 0x00010000;
342 break;
343 case PCI_DEVICE_ID_NVIDIA_NFORCE2:
344 printk(KERN_INFO PFX "Detected NVIDIA nForce2 chipset\n");
345 nvidia_private.wbc_mask = 0x80000000;
346 break;
347 default:
348 printk(KERN_ERR PFX "Unsupported NVIDIA chipset (device id: %04x)\n",
349 pdev->device);
350 return -ENODEV;
351 }
352
353 bridge = agp_alloc_bridge();
354 if (!bridge)
355 return -ENOMEM;
356
357 bridge->driver = &nvidia_driver;
358 bridge->dev_private_data = &nvidia_private,
359 bridge->dev = pdev;
360 bridge->capndx = cap_ptr;
361
362 /* Fill in the mode register */
363 pci_read_config_dword(pdev,
364 bridge->capndx+PCI_AGP_STATUS,
365 &bridge->mode);
366
367 pci_set_drvdata(pdev, bridge);
368 return agp_add_bridge(bridge);
369}
370
371static void __devexit agp_nvidia_remove(struct pci_dev *pdev)
372{
373 struct agp_bridge_data *bridge = pci_get_drvdata(pdev);
374
375 agp_remove_bridge(bridge);
376 agp_put_bridge(bridge);
377}
378
379static struct pci_device_id agp_nvidia_pci_table[] = {
380 {
381 .class = (PCI_CLASS_BRIDGE_HOST << 8),
382 .class_mask = ~0,
383 .vendor = PCI_VENDOR_ID_NVIDIA,
384 .device = PCI_DEVICE_ID_NVIDIA_NFORCE,
385 .subvendor = PCI_ANY_ID,
386 .subdevice = PCI_ANY_ID,
387 },
388 {
389 .class = (PCI_CLASS_BRIDGE_HOST << 8),
390 .class_mask = ~0,
391 .vendor = PCI_VENDOR_ID_NVIDIA,
392 .device = PCI_DEVICE_ID_NVIDIA_NFORCE2,
393 .subvendor = PCI_ANY_ID,
394 .subdevice = PCI_ANY_ID,
395 },
396 { }
397};
398
399MODULE_DEVICE_TABLE(pci, agp_nvidia_pci_table);
400
401static struct pci_driver agp_nvidia_pci_driver = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 .name = "agpgart-nvidia",
403 .id_table = agp_nvidia_pci_table,
404 .probe = agp_nvidia_probe,
405 .remove = agp_nvidia_remove,
406};
407
408static int __init agp_nvidia_init(void)
409{
410 if (agp_off)
411 return -EINVAL;
412 return pci_register_driver(&agp_nvidia_pci_driver);
413}
414
415static void __exit agp_nvidia_cleanup(void)
416{
417 pci_unregister_driver(&agp_nvidia_pci_driver);
418}
419
420module_init(agp_nvidia_init);
421module_exit(agp_nvidia_cleanup);
422
423MODULE_LICENSE("GPL and additional rights");
424MODULE_AUTHOR("NVIDIA Corporation");
425