1 /* 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> 14 #include <linux/jiffies.h> 15 #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 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; 34 off_t pg_offset; 35 u32 wbc_mask; 36 } nvidia_private; 37 38 39 static 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 66 static 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; 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); 79 80 if ((base_lo & 0xfffff000) == (base & 0xfffff000)) 81 break; 82 83 if ((mask_lo & 0x00000800) == 0) 84 free_iorr_addr = iorr_addr; 85 } 86 87 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 } 105 106 static 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 */ 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); 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 } 162 163 static 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 */ 196 extern int agp_memory_reserved; 197 198 static int nvidia_insert_memory(struct agp_memory *mem, off_t pg_start, int type) 199 { 200 int i, j; 201 202 if ((type != 0) || (mem->type != 0)) 203 return -EINVAL; 204 205 if ((pg_start + mem->page_count) > 206 (nvidia_private.num_active_entries - agp_memory_reserved/PAGE_SIZE)) 207 return -EINVAL; 208 209 for (j = pg_start; j < (pg_start + mem->page_count); j++) { 210 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 } 227 228 229 static 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 } 242 243 244 static 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); 260 if (time_before_eq(end, jiffies)) { 261 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 */ 268 for (i = 0; i < 32 + 1; i++) 269 temp = readl(nvidia_private.aperture+(i * PAGE_SIZE / sizeof(u32))); 270 for (i = 0; i < 32 + 1; i++) 271 temp = readl(nvidia_private.aperture+(i * PAGE_SIZE / sizeof(u32))); 272 } 273 274 275 static const 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 }; 284 285 286 static const struct gatt_mask nvidia_generic_masks[] = 287 { 288 { .mask = 1, .type = 0} 289 }; 290 291 292 static const struct agp_bridge_driver nvidia_driver = { 293 .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 .agp_type_to_mask_type = agp_generic_type_to_mask_type, 314 }; 315 316 static int __devinit agp_nvidia_probe(struct pci_dev *pdev, 317 const struct pci_device_id *ent) 318 { 319 struct agp_bridge_data *bridge; 320 u8 cap_ptr; 321 322 nvidia_private.dev_1 = 323 pci_get_bus_and_slot((unsigned int)pdev->bus->number, PCI_DEVFN(0, 1)); 324 nvidia_private.dev_2 = 325 pci_get_bus_and_slot((unsigned int)pdev->bus->number, PCI_DEVFN(0, 2)); 326 nvidia_private.dev_3 = 327 pci_get_bus_and_slot((unsigned int)pdev->bus->number, PCI_DEVFN(30, 0)); 328 329 if (!nvidia_private.dev_1 || !nvidia_private.dev_2 || !nvidia_private.dev_3) { 330 printk(KERN_INFO PFX "Detected an NVIDIA nForce/nForce2 " 331 "chipset, but could not find the secondary devices.\n"); 332 return -ENODEV; 333 } 334 335 cap_ptr = pci_find_capability(pdev, PCI_CAP_ID_AGP); 336 if (!cap_ptr) 337 return -ENODEV; 338 339 switch (pdev->device) { 340 case PCI_DEVICE_ID_NVIDIA_NFORCE: 341 printk(KERN_INFO PFX "Detected NVIDIA nForce chipset\n"); 342 nvidia_private.wbc_mask = 0x00010000; 343 break; 344 case PCI_DEVICE_ID_NVIDIA_NFORCE2: 345 printk(KERN_INFO PFX "Detected NVIDIA nForce2 chipset\n"); 346 nvidia_private.wbc_mask = 0x80000000; 347 break; 348 default: 349 printk(KERN_ERR PFX "Unsupported NVIDIA chipset (device id: %04x)\n", 350 pdev->device); 351 return -ENODEV; 352 } 353 354 bridge = agp_alloc_bridge(); 355 if (!bridge) 356 return -ENOMEM; 357 358 bridge->driver = &nvidia_driver; 359 bridge->dev_private_data = &nvidia_private, 360 bridge->dev = pdev; 361 bridge->capndx = cap_ptr; 362 363 /* Fill in the mode register */ 364 pci_read_config_dword(pdev, 365 bridge->capndx+PCI_AGP_STATUS, 366 &bridge->mode); 367 368 pci_set_drvdata(pdev, bridge); 369 return agp_add_bridge(bridge); 370 } 371 372 static void __devexit agp_nvidia_remove(struct pci_dev *pdev) 373 { 374 struct agp_bridge_data *bridge = pci_get_drvdata(pdev); 375 376 agp_remove_bridge(bridge); 377 agp_put_bridge(bridge); 378 } 379 380 #ifdef CONFIG_PM 381 static int agp_nvidia_suspend(struct pci_dev *pdev, pm_message_t state) 382 { 383 pci_save_state (pdev); 384 pci_set_power_state (pdev, 3); 385 386 return 0; 387 } 388 389 static int agp_nvidia_resume(struct pci_dev *pdev) 390 { 391 /* set power state 0 and restore PCI space */ 392 pci_set_power_state (pdev, 0); 393 pci_restore_state(pdev); 394 395 /* reconfigure AGP hardware again */ 396 nvidia_configure(); 397 398 return 0; 399 } 400 #endif 401 402 403 static struct pci_device_id agp_nvidia_pci_table[] = { 404 { 405 .class = (PCI_CLASS_BRIDGE_HOST << 8), 406 .class_mask = ~0, 407 .vendor = PCI_VENDOR_ID_NVIDIA, 408 .device = PCI_DEVICE_ID_NVIDIA_NFORCE, 409 .subvendor = PCI_ANY_ID, 410 .subdevice = PCI_ANY_ID, 411 }, 412 { 413 .class = (PCI_CLASS_BRIDGE_HOST << 8), 414 .class_mask = ~0, 415 .vendor = PCI_VENDOR_ID_NVIDIA, 416 .device = PCI_DEVICE_ID_NVIDIA_NFORCE2, 417 .subvendor = PCI_ANY_ID, 418 .subdevice = PCI_ANY_ID, 419 }, 420 { } 421 }; 422 423 MODULE_DEVICE_TABLE(pci, agp_nvidia_pci_table); 424 425 static struct pci_driver agp_nvidia_pci_driver = { 426 .name = "agpgart-nvidia", 427 .id_table = agp_nvidia_pci_table, 428 .probe = agp_nvidia_probe, 429 .remove = agp_nvidia_remove, 430 #ifdef CONFIG_PM 431 .suspend = agp_nvidia_suspend, 432 .resume = agp_nvidia_resume, 433 #endif 434 }; 435 436 static int __init agp_nvidia_init(void) 437 { 438 if (agp_off) 439 return -EINVAL; 440 return pci_register_driver(&agp_nvidia_pci_driver); 441 } 442 443 static void __exit agp_nvidia_cleanup(void) 444 { 445 pci_unregister_driver(&agp_nvidia_pci_driver); 446 pci_dev_put(nvidia_private.dev_1); 447 pci_dev_put(nvidia_private.dev_2); 448 pci_dev_put(nvidia_private.dev_3); 449 } 450 451 module_init(agp_nvidia_init); 452 module_exit(agp_nvidia_cleanup); 453 454 MODULE_LICENSE("GPL and additional rights"); 455 MODULE_AUTHOR("NVIDIA Corporation"); 456 457