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>
7 #include <linux/module.h>
9 #include <linux/init.h>
10 #include <linux/agp_backend.h>
11 #include <linux/gfp.h>
12 #include <linux/page-flags.h>
14 #include <linux/jiffies.h>
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
28 static 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;
39 static int nvidia_fetch_size(void)
43 struct aper_size_info_8 *values;
45 pci_read_config_byte(agp_bridge->dev, NVIDIA_0_APSIZE, &size_value);
47 values = A_SIZE_8(agp_bridge->driver->aperture_sizes);
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;
61 #define SYSCFG 0xC0010010
62 #define IORR_BASE0 0xC0010016
63 #define IORR_MASK0 0xC0010017
64 #define AMD_K7_NUM_IORR 2
66 static int nvidia_init_iorr(u32 base, u32 size)
71 u32 iorr_addr, free_iorr_addr;
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;
76 for (iorr_addr = 0; iorr_addr < AMD_K7_NUM_IORR; iorr_addr++) {
77 rdmsr(IORR_BASE0 + 2 * iorr_addr, base_lo, base_hi);
78 rdmsr(IORR_MASK0 + 2 * iorr_addr, mask_lo, mask_hi);
80 if ((base_lo & 0xfffff000) == (base & 0xfffff000))
83 if ((mask_lo & 0x00000800) == 0)
84 free_iorr_addr = iorr_addr;
87 if (iorr_addr >= AMD_K7_NUM_IORR) {
88 iorr_addr = free_iorr_addr;
89 if (iorr_addr >= AMD_K7_NUM_IORR)
93 base_lo = (base & ~0xfff) | 0x18;
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);
99 rdmsr(SYSCFG, sys_lo, sys_hi);
100 sys_lo |= 0x00100000;
101 wrmsr(SYSCFG, sys_lo, sys_hi);
106 static int nvidia_configure(void)
110 struct aper_size_info_8 *current_size;
113 current_size = A_SIZE_8(agp_bridge->current_size);
116 pci_write_config_byte(agp_bridge->dev, NVIDIA_0_APSIZE,
117 current_size->size_value);
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)))
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;
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;
143 for (i = 0; i < 8; i++) {
144 pci_write_config_dword(nvidia_private.dev_2, NVIDIA_2_ATTBASE(i),
145 (agp_bridge->gatt_bus_addr + (i % num_dirs) * 64 * 1024) | 1);
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);
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);
157 nvidia_private.aperture =
158 (volatile u32 __iomem *) ioremap(apbase, 33 * PAGE_SIZE);
160 if (!nvidia_private.aperture)
166 static void nvidia_cleanup(void)
168 struct aper_size_info_8 *previous_size;
172 pci_read_config_dword(agp_bridge->dev, NVIDIA_0_APSIZE, &temp);
173 pci_write_config_dword(agp_bridge->dev, NVIDIA_0_APSIZE, temp & ~(0x100));
176 pci_read_config_dword(nvidia_private.dev_2, NVIDIA_2_GARTCTRL, &temp);
177 pci_write_config_dword(nvidia_private.dev_2, NVIDIA_2_GARTCTRL, temp & ~(0x11));
180 iounmap((void __iomem *) nvidia_private.aperture);
182 /* restore previous aperture size */
183 previous_size = A_SIZE_8(agp_bridge->previous_size);
184 pci_write_config_byte(agp_bridge->dev, NVIDIA_0_APSIZE,
185 previous_size->size_value);
187 /* restore iorr for previous aperture size */
188 nvidia_init_iorr(agp_bridge->gart_bus_addr,
189 previous_size->size * 1024 * 1024);
194 * Note we can't use the generic routines, even though they are 99% the same.
195 * Aperture sizes <64M still requires a full 64k GART directory, but
196 * only use the portion of the TLB entries that correspond to the apertures
197 * alignment inside the surrounding 64M block.
199 extern int agp_memory_reserved;
201 static int nvidia_insert_memory(struct agp_memory *mem, off_t pg_start, int type)
205 if ((type != 0) || (mem->type != 0))
208 if ((pg_start + mem->page_count) >
209 (nvidia_private.num_active_entries - agp_memory_reserved/PAGE_SIZE))
212 for (j = pg_start; j < (pg_start + mem->page_count); j++) {
213 if (!PGE_EMPTY(agp_bridge, readl(agp_bridge->gatt_table+nvidia_private.pg_offset+j)))
217 if (!mem->is_flushed) {
218 global_cache_flush();
219 mem->is_flushed = true;
221 for (i = 0, j = pg_start; i < mem->page_count; i++, j++) {
222 writel(agp_bridge->driver->mask_memory(agp_bridge,
223 mem->memory[i], mem->type),
224 agp_bridge->gatt_table+nvidia_private.pg_offset+j);
225 readl(agp_bridge->gatt_table+nvidia_private.pg_offset+j); /* PCI Posting. */
227 agp_bridge->driver->tlb_flush(mem);
232 static int nvidia_remove_memory(struct agp_memory *mem, off_t pg_start, int type)
236 if ((type != 0) || (mem->type != 0))
239 for (i = pg_start; i < (mem->page_count + pg_start); i++)
240 writel(agp_bridge->scratch_page, agp_bridge->gatt_table+nvidia_private.pg_offset+i);
242 agp_bridge->driver->tlb_flush(mem);
247 static void nvidia_tlbflush(struct agp_memory *mem)
254 if (nvidia_private.wbc_mask) {
255 pci_read_config_dword(nvidia_private.dev_1, NVIDIA_1_WBC, &wbc_reg);
256 wbc_reg |= nvidia_private.wbc_mask;
257 pci_write_config_dword(nvidia_private.dev_1, NVIDIA_1_WBC, wbc_reg);
259 end = jiffies + 3*HZ;
261 pci_read_config_dword(nvidia_private.dev_1,
262 NVIDIA_1_WBC, &wbc_reg);
263 if (time_before_eq(end, jiffies)) {
265 "TLB flush took more than 3 seconds.\n");
267 } while (wbc_reg & nvidia_private.wbc_mask);
270 /* flush TLB entries */
271 for (i = 0; i < 32 + 1; i++)
272 temp = readl(nvidia_private.aperture+(i * PAGE_SIZE / sizeof(u32)));
273 for (i = 0; i < 32 + 1; i++)
274 temp = readl(nvidia_private.aperture+(i * PAGE_SIZE / sizeof(u32)));
278 static const struct aper_size_info_8 nvidia_generic_sizes[5] =
284 /* The 32M mode still requires a 64k gatt */
289 static const struct gatt_mask nvidia_generic_masks[] =
291 { .mask = 1, .type = 0}
295 static const struct agp_bridge_driver nvidia_driver = {
296 .owner = THIS_MODULE,
297 .aperture_sizes = nvidia_generic_sizes,
298 .size_type = U8_APER_SIZE,
299 .num_aperture_sizes = 5,
300 .configure = nvidia_configure,
301 .fetch_size = nvidia_fetch_size,
302 .cleanup = nvidia_cleanup,
303 .tlb_flush = nvidia_tlbflush,
304 .mask_memory = agp_generic_mask_memory,
305 .masks = nvidia_generic_masks,
306 .agp_enable = agp_generic_enable,
307 .cache_flush = global_cache_flush,
308 .create_gatt_table = agp_generic_create_gatt_table,
309 .free_gatt_table = agp_generic_free_gatt_table,
310 .insert_memory = nvidia_insert_memory,
311 .remove_memory = nvidia_remove_memory,
312 .alloc_by_type = agp_generic_alloc_by_type,
313 .free_by_type = agp_generic_free_by_type,
314 .agp_alloc_page = agp_generic_alloc_page,
315 .agp_destroy_page = agp_generic_destroy_page,
316 .agp_type_to_mask_type = agp_generic_type_to_mask_type,
319 static int __devinit agp_nvidia_probe(struct pci_dev *pdev,
320 const struct pci_device_id *ent)
322 struct agp_bridge_data *bridge;
325 nvidia_private.dev_1 =
326 pci_get_bus_and_slot((unsigned int)pdev->bus->number, PCI_DEVFN(0, 1));
327 nvidia_private.dev_2 =
328 pci_get_bus_and_slot((unsigned int)pdev->bus->number, PCI_DEVFN(0, 2));
329 nvidia_private.dev_3 =
330 pci_get_bus_and_slot((unsigned int)pdev->bus->number, PCI_DEVFN(30, 0));
332 if (!nvidia_private.dev_1 || !nvidia_private.dev_2 || !nvidia_private.dev_3) {
333 printk(KERN_INFO PFX "Detected an NVIDIA nForce/nForce2 "
334 "chipset, but could not find the secondary devices.\n");
338 cap_ptr = pci_find_capability(pdev, PCI_CAP_ID_AGP);
342 switch (pdev->device) {
343 case PCI_DEVICE_ID_NVIDIA_NFORCE:
344 printk(KERN_INFO PFX "Detected NVIDIA nForce chipset\n");
345 nvidia_private.wbc_mask = 0x00010000;
347 case PCI_DEVICE_ID_NVIDIA_NFORCE2:
348 printk(KERN_INFO PFX "Detected NVIDIA nForce2 chipset\n");
349 nvidia_private.wbc_mask = 0x80000000;
352 printk(KERN_ERR PFX "Unsupported NVIDIA chipset (device id: %04x)\n",
357 bridge = agp_alloc_bridge();
361 bridge->driver = &nvidia_driver;
362 bridge->dev_private_data = &nvidia_private,
364 bridge->capndx = cap_ptr;
366 /* Fill in the mode register */
367 pci_read_config_dword(pdev,
368 bridge->capndx+PCI_AGP_STATUS,
371 pci_set_drvdata(pdev, bridge);
372 return agp_add_bridge(bridge);
375 static void __devexit agp_nvidia_remove(struct pci_dev *pdev)
377 struct agp_bridge_data *bridge = pci_get_drvdata(pdev);
379 agp_remove_bridge(bridge);
380 agp_put_bridge(bridge);
384 static int agp_nvidia_suspend(struct pci_dev *pdev, pm_message_t state)
386 pci_save_state (pdev);
387 pci_set_power_state (pdev, 3);
392 static int agp_nvidia_resume(struct pci_dev *pdev)
394 /* set power state 0 and restore PCI space */
395 pci_set_power_state (pdev, 0);
396 pci_restore_state(pdev);
398 /* reconfigure AGP hardware again */
406 static struct pci_device_id agp_nvidia_pci_table[] = {
408 .class = (PCI_CLASS_BRIDGE_HOST << 8),
410 .vendor = PCI_VENDOR_ID_NVIDIA,
411 .device = PCI_DEVICE_ID_NVIDIA_NFORCE,
412 .subvendor = PCI_ANY_ID,
413 .subdevice = PCI_ANY_ID,
416 .class = (PCI_CLASS_BRIDGE_HOST << 8),
418 .vendor = PCI_VENDOR_ID_NVIDIA,
419 .device = PCI_DEVICE_ID_NVIDIA_NFORCE2,
420 .subvendor = PCI_ANY_ID,
421 .subdevice = PCI_ANY_ID,
426 MODULE_DEVICE_TABLE(pci, agp_nvidia_pci_table);
428 static struct pci_driver agp_nvidia_pci_driver = {
429 .name = "agpgart-nvidia",
430 .id_table = agp_nvidia_pci_table,
431 .probe = agp_nvidia_probe,
432 .remove = agp_nvidia_remove,
434 .suspend = agp_nvidia_suspend,
435 .resume = agp_nvidia_resume,
439 static int __init agp_nvidia_init(void)
443 return pci_register_driver(&agp_nvidia_pci_driver);
446 static void __exit agp_nvidia_cleanup(void)
448 pci_unregister_driver(&agp_nvidia_pci_driver);
449 pci_dev_put(nvidia_private.dev_1);
450 pci_dev_put(nvidia_private.dev_2);
451 pci_dev_put(nvidia_private.dev_3);
454 module_init(agp_nvidia_init);
455 module_exit(agp_nvidia_cleanup);
457 MODULE_LICENSE("GPL and additional rights");
458 MODULE_AUTHOR("NVIDIA Corporation");