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@redhat.com>
7 #include <linux/module.h>
9 #include <linux/init.h>
10 #include <linux/agp_backend.h>
11 #include <linux/page-flags.h>
13 #include <linux/jiffies.h>
16 /* NVIDIA registers */
17 #define NVIDIA_0_APSIZE 0x80
18 #define NVIDIA_1_WBC 0xf0
19 #define NVIDIA_2_GARTCTRL 0xd0
20 #define NVIDIA_2_APBASE 0xd8
21 #define NVIDIA_2_APLIMIT 0xdc
22 #define NVIDIA_2_ATTBASE(i) (0xe0 + (i) * 4)
23 #define NVIDIA_3_APBASE 0x50
24 #define NVIDIA_3_APLIMIT 0x54
27 static struct _nvidia_private
{
28 struct pci_dev
*dev_1
;
29 struct pci_dev
*dev_2
;
30 struct pci_dev
*dev_3
;
31 volatile u32 __iomem
*aperture
;
32 int num_active_entries
;
38 static int nvidia_fetch_size(void)
42 struct aper_size_info_8
*values
;
44 pci_read_config_byte(agp_bridge
->dev
, NVIDIA_0_APSIZE
, &size_value
);
46 values
= A_SIZE_8(agp_bridge
->driver
->aperture_sizes
);
48 for (i
= 0; i
< agp_bridge
->driver
->num_aperture_sizes
; i
++) {
49 if (size_value
== values
[i
].size_value
) {
50 agp_bridge
->previous_size
=
51 agp_bridge
->current_size
= (void *) (values
+ i
);
52 agp_bridge
->aperture_size_idx
= i
;
53 return values
[i
].size
;
60 #define SYSCFG 0xC0010010
61 #define IORR_BASE0 0xC0010016
62 #define IORR_MASK0 0xC0010017
63 #define AMD_K7_NUM_IORR 2
65 static int nvidia_init_iorr(u32 base
, u32 size
)
70 u32 iorr_addr
, free_iorr_addr
;
72 /* Find the iorr that is already used for the base */
73 /* If not found, determine the uppermost available iorr */
74 free_iorr_addr
= AMD_K7_NUM_IORR
;
75 for (iorr_addr
= 0; iorr_addr
< AMD_K7_NUM_IORR
; iorr_addr
++) {
76 rdmsr(IORR_BASE0
+ 2 * iorr_addr
, base_lo
, base_hi
);
77 rdmsr(IORR_MASK0
+ 2 * iorr_addr
, mask_lo
, mask_hi
);
79 if ((base_lo
& 0xfffff000) == (base
& 0xfffff000))
82 if ((mask_lo
& 0x00000800) == 0)
83 free_iorr_addr
= iorr_addr
;
86 if (iorr_addr
>= AMD_K7_NUM_IORR
) {
87 iorr_addr
= free_iorr_addr
;
88 if (iorr_addr
>= AMD_K7_NUM_IORR
)
92 base_lo
= (base
& ~0xfff) | 0x18;
94 mask_lo
= ((~(size
- 1)) & 0xfffff000) | 0x800;
95 wrmsr(IORR_BASE0
+ 2 * iorr_addr
, base_lo
, base_hi
);
96 wrmsr(IORR_MASK0
+ 2 * iorr_addr
, mask_lo
, mask_hi
);
98 rdmsr(SYSCFG
, sys_lo
, sys_hi
);
100 wrmsr(SYSCFG
, sys_lo
, sys_hi
);
105 static int nvidia_configure(void)
109 struct aper_size_info_8
*current_size
;
112 current_size
= A_SIZE_8(agp_bridge
->current_size
);
115 pci_write_config_byte(agp_bridge
->dev
, NVIDIA_0_APSIZE
,
116 current_size
->size_value
);
118 /* address to map to */
119 pci_read_config_dword(agp_bridge
->dev
, AGP_APBASE
, &apbase
);
120 apbase
&= PCI_BASE_ADDRESS_MEM_MASK
;
121 agp_bridge
->gart_bus_addr
= apbase
;
122 aplimit
= apbase
+ (current_size
->size
* 1024 * 1024) - 1;
123 pci_write_config_dword(nvidia_private
.dev_2
, NVIDIA_2_APBASE
, apbase
);
124 pci_write_config_dword(nvidia_private
.dev_2
, NVIDIA_2_APLIMIT
, aplimit
);
125 pci_write_config_dword(nvidia_private
.dev_3
, NVIDIA_3_APBASE
, apbase
);
126 pci_write_config_dword(nvidia_private
.dev_3
, NVIDIA_3_APLIMIT
, aplimit
);
127 if (0 != (rc
= nvidia_init_iorr(apbase
, current_size
->size
* 1024 * 1024)))
130 /* directory size is 64k */
131 num_dirs
= current_size
->size
/ 64;
132 nvidia_private
.num_active_entries
= current_size
->num_entries
;
133 nvidia_private
.pg_offset
= 0;
136 nvidia_private
.num_active_entries
/= (64 / current_size
->size
);
137 nvidia_private
.pg_offset
= (apbase
& (64 * 1024 * 1024 - 1) &
138 ~(current_size
->size
* 1024 * 1024 - 1)) / PAGE_SIZE
;
142 for (i
= 0; i
< 8; i
++) {
143 pci_write_config_dword(nvidia_private
.dev_2
, NVIDIA_2_ATTBASE(i
),
144 (agp_bridge
->gatt_bus_addr
+ (i
% num_dirs
) * 64 * 1024) | 1);
148 pci_read_config_dword(nvidia_private
.dev_2
, NVIDIA_2_GARTCTRL
, &temp
);
149 pci_write_config_dword(nvidia_private
.dev_2
, NVIDIA_2_GARTCTRL
, temp
| 0x11);
152 pci_read_config_dword(agp_bridge
->dev
, NVIDIA_0_APSIZE
, &temp
);
153 pci_write_config_dword(agp_bridge
->dev
, NVIDIA_0_APSIZE
, temp
| 0x100);
156 nvidia_private
.aperture
=
157 (volatile u32 __iomem
*) ioremap(apbase
, 33 * PAGE_SIZE
);
159 if (!nvidia_private
.aperture
)
165 static void nvidia_cleanup(void)
167 struct aper_size_info_8
*previous_size
;
171 pci_read_config_dword(agp_bridge
->dev
, NVIDIA_0_APSIZE
, &temp
);
172 pci_write_config_dword(agp_bridge
->dev
, NVIDIA_0_APSIZE
, temp
& ~(0x100));
175 pci_read_config_dword(nvidia_private
.dev_2
, NVIDIA_2_GARTCTRL
, &temp
);
176 pci_write_config_dword(nvidia_private
.dev_2
, NVIDIA_2_GARTCTRL
, temp
& ~(0x11));
179 iounmap((void __iomem
*) nvidia_private
.aperture
);
181 /* restore previous aperture size */
182 previous_size
= A_SIZE_8(agp_bridge
->previous_size
);
183 pci_write_config_byte(agp_bridge
->dev
, NVIDIA_0_APSIZE
,
184 previous_size
->size_value
);
186 /* restore iorr for previous aperture size */
187 nvidia_init_iorr(agp_bridge
->gart_bus_addr
,
188 previous_size
->size
* 1024 * 1024);
193 * Note we can't use the generic routines, even though they are 99% the same.
194 * Aperture sizes <64M still requires a full 64k GART directory, but
195 * only use the portion of the TLB entries that correspond to the apertures
196 * alignment inside the surrounding 64M block.
198 extern int agp_memory_reserved
;
200 static int nvidia_insert_memory(struct agp_memory
*mem
, off_t pg_start
, int type
)
205 mask_type
= agp_generic_type_to_mask_type(mem
->bridge
, type
);
206 if (mask_type
!= 0 || type
!= mem
->type
)
209 if (mem
->page_count
== 0)
212 if ((pg_start
+ mem
->page_count
) >
213 (nvidia_private
.num_active_entries
- agp_memory_reserved
/PAGE_SIZE
))
216 for (j
= pg_start
; j
< (pg_start
+ mem
->page_count
); j
++) {
217 if (!PGE_EMPTY(agp_bridge
, readl(agp_bridge
->gatt_table
+nvidia_private
.pg_offset
+j
)))
221 if (!mem
->is_flushed
) {
222 global_cache_flush();
223 mem
->is_flushed
= true;
225 for (i
= 0, j
= pg_start
; i
< mem
->page_count
; i
++, j
++) {
226 writel(agp_bridge
->driver
->mask_memory(agp_bridge
,
227 page_to_phys(mem
->pages
[i
]), mask_type
),
228 agp_bridge
->gatt_table
+nvidia_private
.pg_offset
+j
);
232 readl(agp_bridge
->gatt_table
+nvidia_private
.pg_offset
+j
- 1);
234 agp_bridge
->driver
->tlb_flush(mem
);
239 static int nvidia_remove_memory(struct agp_memory
*mem
, off_t pg_start
, int type
)
245 mask_type
= agp_generic_type_to_mask_type(mem
->bridge
, type
);
246 if (mask_type
!= 0 || type
!= mem
->type
)
249 if (mem
->page_count
== 0)
252 for (i
= pg_start
; i
< (mem
->page_count
+ pg_start
); i
++)
253 writel(agp_bridge
->scratch_page
, agp_bridge
->gatt_table
+nvidia_private
.pg_offset
+i
);
255 agp_bridge
->driver
->tlb_flush(mem
);
260 static void nvidia_tlbflush(struct agp_memory
*mem
)
267 if (nvidia_private
.wbc_mask
) {
268 pci_read_config_dword(nvidia_private
.dev_1
, NVIDIA_1_WBC
, &wbc_reg
);
269 wbc_reg
|= nvidia_private
.wbc_mask
;
270 pci_write_config_dword(nvidia_private
.dev_1
, NVIDIA_1_WBC
, wbc_reg
);
272 end
= jiffies
+ 3*HZ
;
274 pci_read_config_dword(nvidia_private
.dev_1
,
275 NVIDIA_1_WBC
, &wbc_reg
);
276 if (time_before_eq(end
, jiffies
)) {
278 "TLB flush took more than 3 seconds.\n");
280 } while (wbc_reg
& nvidia_private
.wbc_mask
);
283 /* flush TLB entries */
284 for (i
= 0; i
< 32 + 1; i
++)
285 temp
= readl(nvidia_private
.aperture
+(i
* PAGE_SIZE
/ sizeof(u32
)));
286 for (i
= 0; i
< 32 + 1; i
++)
287 temp
= readl(nvidia_private
.aperture
+(i
* PAGE_SIZE
/ sizeof(u32
)));
291 static const struct aper_size_info_8 nvidia_generic_sizes
[5] =
297 /* The 32M mode still requires a 64k gatt */
302 static const struct gatt_mask nvidia_generic_masks
[] =
304 { .mask
= 1, .type
= 0}
308 static const struct agp_bridge_driver nvidia_driver
= {
309 .owner
= THIS_MODULE
,
310 .aperture_sizes
= nvidia_generic_sizes
,
311 .size_type
= U8_APER_SIZE
,
312 .num_aperture_sizes
= 5,
313 .needs_scratch_page
= true,
314 .configure
= nvidia_configure
,
315 .fetch_size
= nvidia_fetch_size
,
316 .cleanup
= nvidia_cleanup
,
317 .tlb_flush
= nvidia_tlbflush
,
318 .mask_memory
= agp_generic_mask_memory
,
319 .masks
= nvidia_generic_masks
,
320 .agp_enable
= agp_generic_enable
,
321 .cache_flush
= global_cache_flush
,
322 .create_gatt_table
= agp_generic_create_gatt_table
,
323 .free_gatt_table
= agp_generic_free_gatt_table
,
324 .insert_memory
= nvidia_insert_memory
,
325 .remove_memory
= nvidia_remove_memory
,
326 .alloc_by_type
= agp_generic_alloc_by_type
,
327 .free_by_type
= agp_generic_free_by_type
,
328 .agp_alloc_page
= agp_generic_alloc_page
,
329 .agp_alloc_pages
= agp_generic_alloc_pages
,
330 .agp_destroy_page
= agp_generic_destroy_page
,
331 .agp_destroy_pages
= agp_generic_destroy_pages
,
332 .agp_type_to_mask_type
= agp_generic_type_to_mask_type
,
335 static int agp_nvidia_probe(struct pci_dev
*pdev
,
336 const struct pci_device_id
*ent
)
338 struct agp_bridge_data
*bridge
;
341 nvidia_private
.dev_1
=
342 pci_get_bus_and_slot((unsigned int)pdev
->bus
->number
, PCI_DEVFN(0, 1));
343 nvidia_private
.dev_2
=
344 pci_get_bus_and_slot((unsigned int)pdev
->bus
->number
, PCI_DEVFN(0, 2));
345 nvidia_private
.dev_3
=
346 pci_get_bus_and_slot((unsigned int)pdev
->bus
->number
, PCI_DEVFN(30, 0));
348 if (!nvidia_private
.dev_1
|| !nvidia_private
.dev_2
|| !nvidia_private
.dev_3
) {
349 printk(KERN_INFO PFX
"Detected an NVIDIA nForce/nForce2 "
350 "chipset, but could not find the secondary devices.\n");
354 cap_ptr
= pci_find_capability(pdev
, PCI_CAP_ID_AGP
);
358 switch (pdev
->device
) {
359 case PCI_DEVICE_ID_NVIDIA_NFORCE
:
360 printk(KERN_INFO PFX
"Detected NVIDIA nForce chipset\n");
361 nvidia_private
.wbc_mask
= 0x00010000;
363 case PCI_DEVICE_ID_NVIDIA_NFORCE2
:
364 printk(KERN_INFO PFX
"Detected NVIDIA nForce2 chipset\n");
365 nvidia_private
.wbc_mask
= 0x80000000;
368 printk(KERN_ERR PFX
"Unsupported NVIDIA chipset (device id: %04x)\n",
373 bridge
= agp_alloc_bridge();
377 bridge
->driver
= &nvidia_driver
;
378 bridge
->dev_private_data
= &nvidia_private
,
380 bridge
->capndx
= cap_ptr
;
382 /* Fill in the mode register */
383 pci_read_config_dword(pdev
,
384 bridge
->capndx
+PCI_AGP_STATUS
,
387 pci_set_drvdata(pdev
, bridge
);
388 return agp_add_bridge(bridge
);
391 static void agp_nvidia_remove(struct pci_dev
*pdev
)
393 struct agp_bridge_data
*bridge
= pci_get_drvdata(pdev
);
395 agp_remove_bridge(bridge
);
396 agp_put_bridge(bridge
);
400 static int agp_nvidia_suspend(struct pci_dev
*pdev
, pm_message_t state
)
402 pci_save_state(pdev
);
403 pci_set_power_state(pdev
, PCI_D3hot
);
408 static int agp_nvidia_resume(struct pci_dev
*pdev
)
410 /* set power state 0 and restore PCI space */
411 pci_set_power_state(pdev
, PCI_D0
);
412 pci_restore_state(pdev
);
414 /* reconfigure AGP hardware again */
422 static struct pci_device_id agp_nvidia_pci_table
[] = {
424 .class = (PCI_CLASS_BRIDGE_HOST
<< 8),
426 .vendor
= PCI_VENDOR_ID_NVIDIA
,
427 .device
= PCI_DEVICE_ID_NVIDIA_NFORCE
,
428 .subvendor
= PCI_ANY_ID
,
429 .subdevice
= PCI_ANY_ID
,
432 .class = (PCI_CLASS_BRIDGE_HOST
<< 8),
434 .vendor
= PCI_VENDOR_ID_NVIDIA
,
435 .device
= PCI_DEVICE_ID_NVIDIA_NFORCE2
,
436 .subvendor
= PCI_ANY_ID
,
437 .subdevice
= PCI_ANY_ID
,
442 MODULE_DEVICE_TABLE(pci
, agp_nvidia_pci_table
);
444 static struct pci_driver agp_nvidia_pci_driver
= {
445 .name
= "agpgart-nvidia",
446 .id_table
= agp_nvidia_pci_table
,
447 .probe
= agp_nvidia_probe
,
448 .remove
= agp_nvidia_remove
,
450 .suspend
= agp_nvidia_suspend
,
451 .resume
= agp_nvidia_resume
,
455 static int __init
agp_nvidia_init(void)
459 return pci_register_driver(&agp_nvidia_pci_driver
);
462 static void __exit
agp_nvidia_cleanup(void)
464 pci_unregister_driver(&agp_nvidia_pci_driver
);
465 pci_dev_put(nvidia_private
.dev_1
);
466 pci_dev_put(nvidia_private
.dev_2
);
467 pci_dev_put(nvidia_private
.dev_3
);
470 module_init(agp_nvidia_init
);
471 module_exit(agp_nvidia_cleanup
);
473 MODULE_LICENSE("GPL and additional rights");
474 MODULE_AUTHOR("NVIDIA Corporation");