Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/*
2 * Copyright 2018 Advanced Micro Devices, Inc.
3 * All Rights Reserved.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sub license, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19 * USE OR OTHER DEALINGS IN THE SOFTWARE.
20 *
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
23 * of the Software.
24 *
25 */
26#ifndef __AMDGPU_GMC_H__
27#define __AMDGPU_GMC_H__
28
29#include <linux/types.h>
30
31#include "amdgpu_irq.h"
32#include "amdgpu_xgmi.h"
33#include "amdgpu_ras.h"
34
35/* VA hole for 48bit and 57bit addresses */
36#define AMDGPU_GMC_HOLE_START (adev->vm_manager.max_level == 4 ?\
37 0x0100000000000000ULL : 0x0000800000000000ULL)
38#define AMDGPU_GMC_HOLE_END (adev->vm_manager.max_level == 4 ?\
39 0xff00000000000000ULL : 0xffff800000000000ULL)
40
41/*
42 * Hardware is programmed as if the hole doesn't exists with start and end
43 * address values.
44 *
45 * This mask is used to remove the upper 16bits of the VA and so come up with
46 * the linear addr value.
47 */
48#define AMDGPU_GMC_HOLE_MASK (adev->vm_manager.max_level == 4 ?\
49 0x01ffffffffffffffULL : 0x0000ffffffffffffULL)
50
51/*
52 * Ring size as power of two for the log of recent faults.
53 */
54#define AMDGPU_GMC_FAULT_RING_ORDER 8
55#define AMDGPU_GMC_FAULT_RING_SIZE (1 << AMDGPU_GMC_FAULT_RING_ORDER)
56
57/*
58 * Hash size as power of two for the log of recent faults
59 */
60#define AMDGPU_GMC_FAULT_HASH_ORDER 8
61#define AMDGPU_GMC_FAULT_HASH_SIZE (1 << AMDGPU_GMC_FAULT_HASH_ORDER)
62
63/*
64 * Number of IH timestamp ticks until a fault is considered handled
65 */
66#define AMDGPU_GMC_FAULT_TIMEOUT 5000ULL
67
68/* XNACK flags */
69#define AMDGPU_GMC_XNACK_FLAG_CHAIN BIT(0)
70
71struct firmware;
72
73enum amdgpu_memory_partition {
74 UNKNOWN_MEMORY_PARTITION_MODE = 0,
75 AMDGPU_NPS1_PARTITION_MODE = 1,
76 AMDGPU_NPS2_PARTITION_MODE = 2,
77 AMDGPU_NPS3_PARTITION_MODE = 3,
78 AMDGPU_NPS4_PARTITION_MODE = 4,
79 AMDGPU_NPS6_PARTITION_MODE = 6,
80 AMDGPU_NPS8_PARTITION_MODE = 8,
81};
82
83#define AMDGPU_ALL_NPS_MASK \
84 (BIT(AMDGPU_NPS1_PARTITION_MODE) | BIT(AMDGPU_NPS2_PARTITION_MODE) | \
85 BIT(AMDGPU_NPS3_PARTITION_MODE) | BIT(AMDGPU_NPS4_PARTITION_MODE) | \
86 BIT(AMDGPU_NPS6_PARTITION_MODE) | BIT(AMDGPU_NPS8_PARTITION_MODE))
87
88#define AMDGPU_GMC_INIT_RESET_NPS BIT(0)
89
90#define AMDGPU_MAX_MEM_RANGES 8
91
92#define AMDGPU_GMC9_FAULT_SOURCE_DATA_RETRY 0x80
93#define AMDGPU_GMC9_FAULT_SOURCE_DATA_READ 0x40
94#define AMDGPU_GMC9_FAULT_SOURCE_DATA_WRITE 0x20
95#define AMDGPU_GMC9_FAULT_SOURCE_DATA_EXE 0x10
96
97#define AMDGPU_GMC121_FAULT_SOURCE_DATA_READ 0x400000
98#define AMDGPU_GMC121_FAULT_SOURCE_DATA_WRITE 0x200000
99#define AMDGPU_GMC121_FAULT_SOURCE_DATA_EXE 0x100000
100
101/*
102 * GMC page fault information
103 */
104struct amdgpu_gmc_fault {
105 uint64_t timestamp:48;
106 uint64_t next:AMDGPU_GMC_FAULT_RING_ORDER;
107 atomic64_t key;
108 uint64_t timestamp_expiry:48;
109};
110
111/*
112 * VMHUB structures, functions & helpers
113 */
114struct amdgpu_vmhub_funcs {
115 void (*print_l2_protection_fault_status)(struct amdgpu_device *adev,
116 uint32_t status);
117 uint32_t (*get_invalidate_req)(unsigned int vmid, uint32_t flush_type);
118};
119
120struct amdgpu_vmhub {
121 uint32_t ctx0_ptb_addr_lo32;
122 uint32_t ctx0_ptb_addr_hi32;
123 uint32_t vm_inv_eng0_sem;
124 uint32_t vm_inv_eng0_req;
125 uint32_t vm_inv_eng0_ack;
126 uint32_t vm_context0_cntl;
127 uint32_t vm_l2_pro_fault_status;
128 uint32_t vm_l2_pro_fault_cntl;
129
130 /*
131 * store the register distances between two continuous context domain
132 * and invalidation engine.
133 */
134 uint32_t ctx_distance;
135 uint32_t ctx_addr_distance; /* include LO32/HI32 */
136 uint32_t eng_distance;
137 uint32_t eng_addr_distance; /* include LO32/HI32 */
138
139 uint32_t vm_cntx_cntl;
140 uint32_t vm_cntx_cntl_vm_fault;
141 uint32_t vm_l2_bank_select_reserved_cid2;
142
143 uint32_t vm_contexts_disable;
144
145 bool sdma_invalidation_workaround;
146
147 const struct amdgpu_vmhub_funcs *vmhub_funcs;
148};
149
150/*
151 * GPU MC structures, functions & helpers
152 */
153struct amdgpu_gmc_funcs {
154 /* flush the vm tlb via mmio */
155 void (*flush_gpu_tlb)(struct amdgpu_device *adev, uint32_t vmid,
156 uint32_t vmhub, uint32_t flush_type);
157 /* flush the vm tlb via pasid */
158 void (*flush_gpu_tlb_pasid)(struct amdgpu_device *adev, uint16_t pasid,
159 uint32_t flush_type, bool all_hub,
160 uint32_t inst);
161 /* flush the vm tlb via ring */
162 uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid,
163 uint64_t pd_addr);
164 /* Change the VMID -> PASID mapping */
165 void (*emit_pasid_mapping)(struct amdgpu_ring *ring, unsigned vmid,
166 unsigned pasid);
167 /* enable/disable PRT support */
168 void (*set_prt)(struct amdgpu_device *adev, bool enable);
169 /* get the pde for a given mc addr */
170 void (*get_vm_pde)(struct amdgpu_device *adev, int level,
171 u64 *dst, u64 *flags);
172 /* get the pte flags to use for PTEs */
173 void (*get_vm_pte)(struct amdgpu_device *adev,
174 struct amdgpu_vm *vm,
175 struct amdgpu_bo *bo,
176 uint32_t vm_flags,
177 uint64_t *pte_flags);
178 /* override per-page pte flags */
179 void (*override_vm_pte_flags)(struct amdgpu_device *dev,
180 struct amdgpu_vm *vm,
181 uint64_t addr, uint64_t *flags);
182 /* get the amount of memory used by the vbios for pre-OS console */
183 unsigned int (*get_vbios_fb_size)(struct amdgpu_device *adev);
184 /* get the DCC buffer alignment */
185 unsigned int (*get_dcc_alignment)(struct amdgpu_device *adev);
186
187 enum amdgpu_memory_partition (*query_mem_partition_mode)(
188 struct amdgpu_device *adev);
189 /* Request NPS mode */
190 int (*request_mem_partition_mode)(struct amdgpu_device *adev,
191 int nps_mode);
192 bool (*need_reset_on_init)(struct amdgpu_device *adev);
193};
194
195struct amdgpu_mem_partition_info {
196 union {
197 struct {
198 uint32_t fpfn;
199 uint32_t lpfn;
200 } range;
201 struct {
202 int node;
203 } numa;
204 };
205 uint64_t size;
206};
207
208#define INVALID_PFN -1
209
210struct amdgpu_gmc_memrange {
211 uint64_t base_address;
212 uint64_t limit_address;
213 uint32_t flags;
214 int nid_mask;
215};
216
217enum amdgpu_gart_placement {
218 AMDGPU_GART_PLACEMENT_BEST_FIT = 0,
219 AMDGPU_GART_PLACEMENT_HIGH,
220 AMDGPU_GART_PLACEMENT_LOW,
221};
222
223struct amdgpu_gmc {
224 /* FB's physical address in MMIO space (for CPU to
225 * map FB). This is different compared to the agp/
226 * gart/vram_start/end field as the later is from
227 * GPU's view and aper_base is from CPU's view.
228 */
229 resource_size_t aper_size;
230 resource_size_t aper_base;
231 /* for some chips with <= 32MB we need to lie
232 * about vram size near mc fb location */
233 u64 mc_vram_size;
234 u64 visible_vram_size;
235 /* AGP aperture start and end in MC address space
236 * Driver find a hole in the MC address space
237 * to place AGP by setting MC_VM_AGP_BOT/TOP registers
238 * Under VMID0, logical address == MC address. AGP
239 * aperture maps to physical bus or IOVA addressed.
240 * AGP aperture is used to simulate FB in ZFB case.
241 * AGP aperture is also used for page table in system
242 * memory (mainly for APU).
243 *
244 */
245 u64 agp_size;
246 u64 agp_start;
247 u64 agp_end;
248 /* GART aperture start and end in MC address space
249 * Driver find a hole in the MC address space
250 * to place GART by setting VM_CONTEXT0_PAGE_TABLE_START/END_ADDR
251 * registers
252 * Under VMID0, logical address inside GART aperture will
253 * be translated through gpuvm gart page table to access
254 * paged system memory
255 */
256 u64 gart_size;
257 u64 gart_start;
258 u64 gart_end;
259 /* Frame buffer aperture of this GPU device. Different from
260 * fb_start (see below), this only covers the local GPU device.
261 * If driver uses FB aperture to access FB, driver get fb_start from
262 * MC_VM_FB_LOCATION_BASE (set by vbios) and calculate vram_start
263 * of this local device by adding an offset inside the XGMI hive.
264 * If driver uses GART table for VMID0 FB access, driver finds a hole in
265 * VMID0's virtual address space to place the SYSVM aperture inside
266 * which the first part is vram and the second part is gart (covering
267 * system ram).
268 */
269 u64 vram_start;
270 u64 vram_end;
271 /* FB region , it's same as local vram region in single GPU, in XGMI
272 * configuration, this region covers all GPUs in the same hive ,
273 * each GPU in the hive has the same view of this FB region .
274 * GPU0's vram starts at offset (0 * segment size) ,
275 * GPU1 starts at offset (1 * segment size), etc.
276 */
277 u64 fb_start;
278 u64 fb_end;
279 unsigned vram_width;
280 u64 real_vram_size;
281 int vram_mtrr;
282 u64 mc_mask;
283 const struct firmware *fw; /* MC firmware */
284 uint32_t fw_version;
285 struct amdgpu_irq_src vm_fault;
286 uint32_t vram_type;
287 uint8_t vram_vendor;
288 uint32_t srbm_soft_reset;
289 bool prt_warning;
290 uint32_t sdpif_register;
291 /* apertures */
292 u64 shared_aperture_start;
293 u64 shared_aperture_end;
294 u64 private_aperture_start;
295 u64 private_aperture_end;
296 /* protects concurrent invalidation */
297 spinlock_t invalidate_lock;
298 bool translate_further;
299 struct kfd_vm_fault_info *vm_fault_info;
300 atomic_t vm_fault_info_updated;
301
302 struct amdgpu_gmc_fault fault_ring[AMDGPU_GMC_FAULT_RING_SIZE];
303 struct {
304 uint64_t idx:AMDGPU_GMC_FAULT_RING_ORDER;
305 } fault_hash[AMDGPU_GMC_FAULT_HASH_SIZE];
306 uint64_t last_fault:AMDGPU_GMC_FAULT_RING_ORDER;
307
308 bool tmz_enabled;
309 bool is_app_apu;
310
311 struct amdgpu_mem_partition_info *mem_partitions;
312 uint8_t num_mem_partitions;
313 const struct amdgpu_gmc_funcs *gmc_funcs;
314 enum amdgpu_memory_partition requested_nps_mode;
315 uint32_t supported_nps_modes;
316 uint32_t reset_flags;
317
318 struct amdgpu_xgmi xgmi;
319 struct amdgpu_irq_src ecc_irq;
320 int noretry;
321 uint32_t xnack_flags;
322
323 uint32_t vmid0_page_table_block_size;
324 uint32_t vmid0_page_table_depth;
325 struct amdgpu_bo *pdb0_bo;
326 /* CPU kmapped address of pdb0*/
327 void *ptr_pdb0;
328
329 /* MALL size */
330 u64 mall_size;
331 uint32_t m_half_use;
332
333 /* number of UMC instances */
334 int num_umc;
335 /* mode2 save restore */
336 u64 VM_L2_CNTL;
337 u64 VM_L2_CNTL2;
338 u64 VM_DUMMY_PAGE_FAULT_CNTL;
339 u64 VM_DUMMY_PAGE_FAULT_ADDR_LO32;
340 u64 VM_DUMMY_PAGE_FAULT_ADDR_HI32;
341 u64 VM_L2_PROTECTION_FAULT_CNTL;
342 u64 VM_L2_PROTECTION_FAULT_CNTL2;
343 u64 VM_L2_PROTECTION_FAULT_MM_CNTL3;
344 u64 VM_L2_PROTECTION_FAULT_MM_CNTL4;
345 u64 VM_L2_PROTECTION_FAULT_ADDR_LO32;
346 u64 VM_L2_PROTECTION_FAULT_ADDR_HI32;
347 u64 VM_DEBUG;
348 u64 VM_L2_MM_GROUP_RT_CLASSES;
349 u64 VM_L2_BANK_SELECT_RESERVED_CID;
350 u64 VM_L2_BANK_SELECT_RESERVED_CID2;
351 u64 VM_L2_CACHE_PARITY_CNTL;
352 u64 VM_L2_IH_LOG_CNTL;
353 u64 VM_CONTEXT_CNTL[16];
354 u64 VM_CONTEXT_PAGE_TABLE_BASE_ADDR_LO32[16];
355 u64 VM_CONTEXT_PAGE_TABLE_BASE_ADDR_HI32[16];
356 u64 VM_CONTEXT_PAGE_TABLE_START_ADDR_LO32[16];
357 u64 VM_CONTEXT_PAGE_TABLE_START_ADDR_HI32[16];
358 u64 VM_CONTEXT_PAGE_TABLE_END_ADDR_LO32[16];
359 u64 VM_CONTEXT_PAGE_TABLE_END_ADDR_HI32[16];
360 u64 MC_VM_MX_L1_TLB_CNTL;
361
362 u64 noretry_flags;
363 u64 init_pte_flags;
364
365 bool flush_tlb_needs_extra_type_0;
366 bool flush_tlb_needs_extra_type_2;
367 bool flush_pasid_uses_kiq;
368};
369
370#define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
371#define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
372#define amdgpu_gmc_get_vm_pde(adev, level, dst, flags) (adev)->gmc.gmc_funcs->get_vm_pde((adev), (level), (dst), (flags))
373#define amdgpu_gmc_get_vm_pte(adev, vm, bo, vm_flags, pte_flags) \
374 ((adev)->gmc.gmc_funcs->get_vm_pte((adev), (vm), (bo), (vm_flags), \
375 (pte_flags)))
376#define amdgpu_gmc_override_vm_pte_flags(adev, vm, addr, pte_flags) \
377 (adev)->gmc.gmc_funcs->override_vm_pte_flags \
378 ((adev), (vm), (addr), (pte_flags))
379#define amdgpu_gmc_get_vbios_fb_size(adev) (adev)->gmc.gmc_funcs->get_vbios_fb_size((adev))
380#define amdgpu_gmc_get_dcc_alignment(adev) ({ \
381 typeof(adev) _adev = (adev); \
382 _adev->gmc.gmc_funcs->get_dcc_alignment(_adev); \
383})
384
385/**
386 * amdgpu_gmc_vram_full_visible - Check if full VRAM is visible through the BAR
387 *
388 * @adev: amdgpu_device pointer
389 *
390 * Returns:
391 * True if full VRAM is visible through the BAR
392 */
393static inline bool amdgpu_gmc_vram_full_visible(struct amdgpu_gmc *gmc)
394{
395 WARN_ON(gmc->real_vram_size < gmc->visible_vram_size);
396
397 return (gmc->real_vram_size == gmc->visible_vram_size);
398}
399
400/**
401 * amdgpu_gmc_sign_extend - sign extend the given gmc address
402 *
403 * @addr: address to extend
404 */
405#define amdgpu_gmc_sign_extend(addr) ((addr) >= AMDGPU_GMC_HOLE_START ?\
406 ((addr) | AMDGPU_GMC_HOLE_END) : (addr))
407
408bool amdgpu_gmc_is_pdb0_enabled(struct amdgpu_device *adev);
409int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev);
410void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level,
411 uint64_t *addr, uint64_t *flags);
412int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
413 uint32_t gpu_page_idx, uint64_t addr,
414 uint64_t flags);
415uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo);
416uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo);
417void amdgpu_gmc_sysvm_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc);
418void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
419 u64 base);
420void amdgpu_gmc_gart_location(struct amdgpu_device *adev,
421 struct amdgpu_gmc *mc,
422 enum amdgpu_gart_placement gart_placement);
423void amdgpu_gmc_agp_location(struct amdgpu_device *adev,
424 struct amdgpu_gmc *mc);
425void amdgpu_gmc_set_agp_default(struct amdgpu_device *adev,
426 struct amdgpu_gmc *mc);
427bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev,
428 struct amdgpu_ih_ring *ih, uint64_t addr,
429 uint16_t pasid, uint64_t timestamp);
430void amdgpu_gmc_filter_faults_remove(struct amdgpu_device *adev, uint64_t addr,
431 uint16_t pasid);
432int amdgpu_gmc_handle_retry_fault(struct amdgpu_device *adev,
433 struct amdgpu_iv_entry *entry,
434 u64 addr,
435 u32 cam_index,
436 u32 node_id,
437 bool write_fault);
438int amdgpu_gmc_ras_sw_init(struct amdgpu_device *adev);
439int amdgpu_gmc_ras_late_init(struct amdgpu_device *adev);
440void amdgpu_gmc_ras_fini(struct amdgpu_device *adev);
441int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev);
442void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
443 uint32_t vmhub, uint32_t flush_type);
444int amdgpu_gmc_flush_gpu_tlb_pasid(struct amdgpu_device *adev, uint16_t pasid,
445 uint32_t flush_type, bool all_hub,
446 uint32_t inst);
447void amdgpu_gmc_fw_reg_write_reg_wait(struct amdgpu_device *adev,
448 uint32_t reg0, uint32_t reg1,
449 uint32_t ref, uint32_t mask,
450 uint32_t xcc_inst);
451
452extern void amdgpu_gmc_tmz_set(struct amdgpu_device *adev);
453extern void amdgpu_gmc_noretry_set(struct amdgpu_device *adev);
454
455extern void
456amdgpu_gmc_set_vm_fault_masks(struct amdgpu_device *adev, int hub_type,
457 bool enable);
458
459void amdgpu_gmc_init_vga_resv_regions(struct amdgpu_device *adev);
460
461void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev);
462uint64_t amdgpu_gmc_vram_mc2pa(struct amdgpu_device *adev, uint64_t mc_addr);
463uint64_t amdgpu_gmc_vram_pa(struct amdgpu_device *adev, struct amdgpu_bo *bo);
464int amdgpu_gmc_vram_checking(struct amdgpu_device *adev);
465int amdgpu_gmc_sysfs_init(struct amdgpu_device *adev);
466void amdgpu_gmc_sysfs_fini(struct amdgpu_device *adev);
467
468int amdgpu_gmc_get_nps_memranges(struct amdgpu_device *adev,
469 struct amdgpu_mem_partition_info *mem_ranges,
470 uint8_t *exp_ranges);
471
472int amdgpu_gmc_request_memory_partition(struct amdgpu_device *adev,
473 int nps_mode);
474void amdgpu_gmc_prepare_nps_mode_change(struct amdgpu_device *adev);
475bool amdgpu_gmc_need_reset_on_init(struct amdgpu_device *adev);
476enum amdgpu_memory_partition
477amdgpu_gmc_get_vf_memory_partition(struct amdgpu_device *adev);
478enum amdgpu_memory_partition
479amdgpu_gmc_get_memory_partition(struct amdgpu_device *adev, u32 *supp_modes);
480enum amdgpu_memory_partition
481amdgpu_gmc_query_memory_partition(struct amdgpu_device *adev);
482int amdgpu_gmc_init_mem_ranges(struct amdgpu_device *adev);
483void amdgpu_gmc_init_sw_mem_ranges(struct amdgpu_device *adev,
484 struct amdgpu_mem_partition_info *mem_ranges);
485int amdgpu_gmc_get_vram_info(struct amdgpu_device *adev,
486 int *vram_width, int *vram_type, int *vram_vendor);
487#endif