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 addresses on Vega10 */
36 #define AMDGPU_GMC_HOLE_START	0x0000800000000000ULL
37 #define AMDGPU_GMC_HOLE_END	0xffff800000000000ULL
38 
39 /*
40  * Hardware is programmed as if the hole doesn't exists with start and end
41  * address values.
42  *
43  * This mask is used to remove the upper 16bits of the VA and so come up with
44  * the linear addr value.
45  */
46 #define AMDGPU_GMC_HOLE_MASK	0x0000ffffffffffffULL
47 
48 /*
49  * Ring size as power of two for the log of recent faults.
50  */
51 #define AMDGPU_GMC_FAULT_RING_ORDER	8
52 #define AMDGPU_GMC_FAULT_RING_SIZE	(1 << AMDGPU_GMC_FAULT_RING_ORDER)
53 
54 /*
55  * Hash size as power of two for the log of recent faults
56  */
57 #define AMDGPU_GMC_FAULT_HASH_ORDER	8
58 #define AMDGPU_GMC_FAULT_HASH_SIZE	(1 << AMDGPU_GMC_FAULT_HASH_ORDER)
59 
60 /*
61  * Number of IH timestamp ticks until a fault is considered handled
62  */
63 #define AMDGPU_GMC_FAULT_TIMEOUT	5000ULL
64 
65 struct firmware;
66 
67 enum amdgpu_memory_partition {
68 	UNKNOWN_MEMORY_PARTITION_MODE = 0,
69 	AMDGPU_NPS1_PARTITION_MODE = 1,
70 	AMDGPU_NPS2_PARTITION_MODE = 2,
71 	AMDGPU_NPS3_PARTITION_MODE = 3,
72 	AMDGPU_NPS4_PARTITION_MODE = 4,
73 	AMDGPU_NPS6_PARTITION_MODE = 6,
74 	AMDGPU_NPS8_PARTITION_MODE = 8,
75 };
76 
77 #define AMDGPU_ALL_NPS_MASK                                                  \
78 	(BIT(AMDGPU_NPS1_PARTITION_MODE) | BIT(AMDGPU_NPS2_PARTITION_MODE) | \
79 	 BIT(AMDGPU_NPS3_PARTITION_MODE) | BIT(AMDGPU_NPS4_PARTITION_MODE) | \
80 	 BIT(AMDGPU_NPS6_PARTITION_MODE) | BIT(AMDGPU_NPS8_PARTITION_MODE))
81 
82 #define AMDGPU_GMC_INIT_RESET_NPS  BIT(0)
83 
84 /*
85  * GMC page fault information
86  */
87 struct amdgpu_gmc_fault {
88 	uint64_t	timestamp:48;
89 	uint64_t	next:AMDGPU_GMC_FAULT_RING_ORDER;
90 	atomic64_t	key;
91 	uint64_t	timestamp_expiry:48;
92 };
93 
94 /*
95  * VMHUB structures, functions & helpers
96  */
97 struct amdgpu_vmhub_funcs {
98 	void (*print_l2_protection_fault_status)(struct amdgpu_device *adev,
99 						 uint32_t status);
100 	uint32_t (*get_invalidate_req)(unsigned int vmid, uint32_t flush_type);
101 };
102 
103 struct amdgpu_vmhub {
104 	uint32_t	ctx0_ptb_addr_lo32;
105 	uint32_t	ctx0_ptb_addr_hi32;
106 	uint32_t	vm_inv_eng0_sem;
107 	uint32_t	vm_inv_eng0_req;
108 	uint32_t	vm_inv_eng0_ack;
109 	uint32_t	vm_context0_cntl;
110 	uint32_t	vm_l2_pro_fault_status;
111 	uint32_t	vm_l2_pro_fault_cntl;
112 
113 	/*
114 	 * store the register distances between two continuous context domain
115 	 * and invalidation engine.
116 	 */
117 	uint32_t	ctx_distance;
118 	uint32_t	ctx_addr_distance; /* include LO32/HI32 */
119 	uint32_t	eng_distance;
120 	uint32_t	eng_addr_distance; /* include LO32/HI32 */
121 
122 	uint32_t        vm_cntx_cntl;
123 	uint32_t	vm_cntx_cntl_vm_fault;
124 	uint32_t	vm_l2_bank_select_reserved_cid2;
125 
126 	uint32_t	vm_contexts_disable;
127 
128 	bool		sdma_invalidation_workaround;
129 
130 	const struct amdgpu_vmhub_funcs *vmhub_funcs;
131 };
132 
133 /*
134  * GPU MC structures, functions & helpers
135  */
136 struct amdgpu_gmc_funcs {
137 	/* flush the vm tlb via mmio */
138 	void (*flush_gpu_tlb)(struct amdgpu_device *adev, uint32_t vmid,
139 				uint32_t vmhub, uint32_t flush_type);
140 	/* flush the vm tlb via pasid */
141 	void (*flush_gpu_tlb_pasid)(struct amdgpu_device *adev, uint16_t pasid,
142 				    uint32_t flush_type, bool all_hub,
143 				    uint32_t inst);
144 	/* flush the vm tlb via ring */
145 	uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid,
146 				       uint64_t pd_addr);
147 	/* Change the VMID -> PASID mapping */
148 	void (*emit_pasid_mapping)(struct amdgpu_ring *ring, unsigned vmid,
149 				   unsigned pasid);
150 	/* enable/disable PRT support */
151 	void (*set_prt)(struct amdgpu_device *adev, bool enable);
152 	/* map mtype to hardware flags */
153 	uint64_t (*map_mtype)(struct amdgpu_device *adev, uint32_t flags);
154 	/* get the pde for a given mc addr */
155 	void (*get_vm_pde)(struct amdgpu_device *adev, int level,
156 			   u64 *dst, u64 *flags);
157 	/* get the pte flags to use for a BO VA mapping */
158 	void (*get_vm_pte)(struct amdgpu_device *adev,
159 			   struct amdgpu_bo_va_mapping *mapping,
160 			   uint64_t *flags);
161 	/* override per-page pte flags */
162 	void (*override_vm_pte_flags)(struct amdgpu_device *dev,
163 				      struct amdgpu_vm *vm,
164 				      uint64_t addr, uint64_t *flags);
165 	/* get the amount of memory used by the vbios for pre-OS console */
166 	unsigned int (*get_vbios_fb_size)(struct amdgpu_device *adev);
167 	/* get the DCC buffer alignment */
168 	unsigned int (*get_dcc_alignment)(struct amdgpu_device *adev);
169 
170 	enum amdgpu_memory_partition (*query_mem_partition_mode)(
171 		struct amdgpu_device *adev);
172 	/* Request NPS mode */
173 	int (*request_mem_partition_mode)(struct amdgpu_device *adev,
174 					  int nps_mode);
175 	bool (*need_reset_on_init)(struct amdgpu_device *adev);
176 };
177 
178 struct amdgpu_mem_partition_info {
179 	union {
180 		struct {
181 			uint32_t fpfn;
182 			uint32_t lpfn;
183 		} range;
184 		struct {
185 			int node;
186 		} numa;
187 	};
188 	uint64_t size;
189 };
190 
191 #define INVALID_PFN    -1
192 
193 struct amdgpu_gmc_memrange {
194 	uint64_t base_address;
195 	uint64_t limit_address;
196 	uint32_t flags;
197 	int nid_mask;
198 };
199 
200 enum amdgpu_gart_placement {
201 	AMDGPU_GART_PLACEMENT_BEST_FIT = 0,
202 	AMDGPU_GART_PLACEMENT_HIGH,
203 	AMDGPU_GART_PLACEMENT_LOW,
204 };
205 
206 struct amdgpu_gmc {
207 	/* FB's physical address in MMIO space (for CPU to
208 	 * map FB). This is different compared to the agp/
209 	 * gart/vram_start/end field as the later is from
210 	 * GPU's view and aper_base is from CPU's view.
211 	 */
212 	resource_size_t		aper_size;
213 	resource_size_t		aper_base;
214 	/* for some chips with <= 32MB we need to lie
215 	 * about vram size near mc fb location */
216 	u64			mc_vram_size;
217 	u64			visible_vram_size;
218 	/* AGP aperture start and end in MC address space
219 	 * Driver find a hole in the MC address space
220 	 * to place AGP by setting MC_VM_AGP_BOT/TOP registers
221 	 * Under VMID0, logical address == MC address. AGP
222 	 * aperture maps to physical bus or IOVA addressed.
223 	 * AGP aperture is used to simulate FB in ZFB case.
224 	 * AGP aperture is also used for page table in system
225 	 * memory (mainly for APU).
226 	 *
227 	 */
228 	u64			agp_size;
229 	u64			agp_start;
230 	u64			agp_end;
231 	/* GART aperture start and end in MC address space
232 	 * Driver find a hole in the MC address space
233 	 * to place GART by setting VM_CONTEXT0_PAGE_TABLE_START/END_ADDR
234 	 * registers
235 	 * Under VMID0, logical address inside GART aperture will
236 	 * be translated through gpuvm gart page table to access
237 	 * paged system memory
238 	 */
239 	u64			gart_size;
240 	u64			gart_start;
241 	u64			gart_end;
242 	/* Frame buffer aperture of this GPU device. Different from
243 	 * fb_start (see below), this only covers the local GPU device.
244 	 * If driver uses FB aperture to access FB, driver get fb_start from
245 	 * MC_VM_FB_LOCATION_BASE (set by vbios) and calculate vram_start
246 	 * of this local device by adding an offset inside the XGMI hive.
247 	 * If driver uses GART table for VMID0 FB access, driver finds a hole in
248 	 * VMID0's virtual address space to place the SYSVM aperture inside
249 	 * which the first part is vram and the second part is gart (covering
250 	 * system ram).
251 	 */
252 	u64			vram_start;
253 	u64			vram_end;
254 	/* FB region , it's same as local vram region in single GPU, in XGMI
255 	 * configuration, this region covers all GPUs in the same hive ,
256 	 * each GPU in the hive has the same view of this FB region .
257 	 * GPU0's vram starts at offset (0 * segment size) ,
258 	 * GPU1 starts at offset (1 * segment size), etc.
259 	 */
260 	u64			fb_start;
261 	u64			fb_end;
262 	unsigned		vram_width;
263 	u64			real_vram_size;
264 	int			vram_mtrr;
265 	u64                     mc_mask;
266 	const struct firmware   *fw;	/* MC firmware */
267 	uint32_t                fw_version;
268 	struct amdgpu_irq_src	vm_fault;
269 	uint32_t		vram_type;
270 	uint8_t			vram_vendor;
271 	uint32_t                srbm_soft_reset;
272 	bool			prt_warning;
273 	uint32_t		sdpif_register;
274 	/* apertures */
275 	u64			shared_aperture_start;
276 	u64			shared_aperture_end;
277 	u64			private_aperture_start;
278 	u64			private_aperture_end;
279 	/* protects concurrent invalidation */
280 	spinlock_t		invalidate_lock;
281 	bool			translate_further;
282 	struct kfd_vm_fault_info *vm_fault_info;
283 	atomic_t		vm_fault_info_updated;
284 
285 	struct amdgpu_gmc_fault	fault_ring[AMDGPU_GMC_FAULT_RING_SIZE];
286 	struct {
287 		uint64_t	idx:AMDGPU_GMC_FAULT_RING_ORDER;
288 	} fault_hash[AMDGPU_GMC_FAULT_HASH_SIZE];
289 	uint64_t		last_fault:AMDGPU_GMC_FAULT_RING_ORDER;
290 
291 	bool tmz_enabled;
292 	bool is_app_apu;
293 
294 	struct amdgpu_mem_partition_info *mem_partitions;
295 	uint8_t num_mem_partitions;
296 	const struct amdgpu_gmc_funcs	*gmc_funcs;
297 	enum amdgpu_memory_partition	requested_nps_mode;
298 	uint32_t supported_nps_modes;
299 	uint32_t reset_flags;
300 
301 	struct amdgpu_xgmi xgmi;
302 	struct amdgpu_irq_src	ecc_irq;
303 	int noretry;
304 
305 	uint32_t	vmid0_page_table_block_size;
306 	uint32_t	vmid0_page_table_depth;
307 	struct amdgpu_bo		*pdb0_bo;
308 	/* CPU kmapped address of pdb0*/
309 	void				*ptr_pdb0;
310 
311 	/* MALL size */
312 	u64 mall_size;
313 	uint32_t m_half_use;
314 
315 	/* number of UMC instances */
316 	int num_umc;
317 	/* mode2 save restore */
318 	u64 VM_L2_CNTL;
319 	u64 VM_L2_CNTL2;
320 	u64 VM_DUMMY_PAGE_FAULT_CNTL;
321 	u64 VM_DUMMY_PAGE_FAULT_ADDR_LO32;
322 	u64 VM_DUMMY_PAGE_FAULT_ADDR_HI32;
323 	u64 VM_L2_PROTECTION_FAULT_CNTL;
324 	u64 VM_L2_PROTECTION_FAULT_CNTL2;
325 	u64 VM_L2_PROTECTION_FAULT_MM_CNTL3;
326 	u64 VM_L2_PROTECTION_FAULT_MM_CNTL4;
327 	u64 VM_L2_PROTECTION_FAULT_ADDR_LO32;
328 	u64 VM_L2_PROTECTION_FAULT_ADDR_HI32;
329 	u64 VM_DEBUG;
330 	u64 VM_L2_MM_GROUP_RT_CLASSES;
331 	u64 VM_L2_BANK_SELECT_RESERVED_CID;
332 	u64 VM_L2_BANK_SELECT_RESERVED_CID2;
333 	u64 VM_L2_CACHE_PARITY_CNTL;
334 	u64 VM_L2_IH_LOG_CNTL;
335 	u64 VM_CONTEXT_CNTL[16];
336 	u64 VM_CONTEXT_PAGE_TABLE_BASE_ADDR_LO32[16];
337 	u64 VM_CONTEXT_PAGE_TABLE_BASE_ADDR_HI32[16];
338 	u64 VM_CONTEXT_PAGE_TABLE_START_ADDR_LO32[16];
339 	u64 VM_CONTEXT_PAGE_TABLE_START_ADDR_HI32[16];
340 	u64 VM_CONTEXT_PAGE_TABLE_END_ADDR_LO32[16];
341 	u64 VM_CONTEXT_PAGE_TABLE_END_ADDR_HI32[16];
342 	u64 MC_VM_MX_L1_TLB_CNTL;
343 
344 	u64 noretry_flags;
345 
346 	bool flush_tlb_needs_extra_type_0;
347 	bool flush_tlb_needs_extra_type_2;
348 	bool flush_pasid_uses_kiq;
349 };
350 
351 #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
352 #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
353 #define amdgpu_gmc_map_mtype(adev, flags) (adev)->gmc.gmc_funcs->map_mtype((adev),(flags))
354 #define amdgpu_gmc_get_vm_pde(adev, level, dst, flags) (adev)->gmc.gmc_funcs->get_vm_pde((adev), (level), (dst), (flags))
355 #define amdgpu_gmc_get_vm_pte(adev, mapping, flags) (adev)->gmc.gmc_funcs->get_vm_pte((adev), (mapping), (flags))
356 #define amdgpu_gmc_override_vm_pte_flags(adev, vm, addr, pte_flags)	\
357 	(adev)->gmc.gmc_funcs->override_vm_pte_flags			\
358 		((adev), (vm), (addr), (pte_flags))
359 #define amdgpu_gmc_get_vbios_fb_size(adev) (adev)->gmc.gmc_funcs->get_vbios_fb_size((adev))
360 #define amdgpu_gmc_get_dcc_alignment(adev) ({			\
361 	typeof(adev) _adev = (adev);				\
362 	_adev->gmc.gmc_funcs->get_dcc_alignment(_adev);		\
363 })
364 
365 /**
366  * amdgpu_gmc_vram_full_visible - Check if full VRAM is visible through the BAR
367  *
368  * @adev: amdgpu_device pointer
369  *
370  * Returns:
371  * True if full VRAM is visible through the BAR
372  */
amdgpu_gmc_vram_full_visible(struct amdgpu_gmc * gmc)373 static inline bool amdgpu_gmc_vram_full_visible(struct amdgpu_gmc *gmc)
374 {
375 	WARN_ON(gmc->real_vram_size < gmc->visible_vram_size);
376 
377 	return (gmc->real_vram_size == gmc->visible_vram_size);
378 }
379 
380 /**
381  * amdgpu_gmc_sign_extend - sign extend the given gmc address
382  *
383  * @addr: address to extend
384  */
amdgpu_gmc_sign_extend(uint64_t addr)385 static inline uint64_t amdgpu_gmc_sign_extend(uint64_t addr)
386 {
387 	if (addr >= AMDGPU_GMC_HOLE_START)
388 		addr |= AMDGPU_GMC_HOLE_END;
389 
390 	return addr;
391 }
392 
393 int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev);
394 void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level,
395 			       uint64_t *addr, uint64_t *flags);
396 int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
397 				uint32_t gpu_page_idx, uint64_t addr,
398 				uint64_t flags);
399 uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo);
400 uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo);
401 void amdgpu_gmc_sysvm_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc);
402 void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
403 			      u64 base);
404 void amdgpu_gmc_gart_location(struct amdgpu_device *adev,
405 			      struct amdgpu_gmc *mc,
406 			      enum amdgpu_gart_placement gart_placement);
407 void amdgpu_gmc_agp_location(struct amdgpu_device *adev,
408 			     struct amdgpu_gmc *mc);
409 void amdgpu_gmc_set_agp_default(struct amdgpu_device *adev,
410 				struct amdgpu_gmc *mc);
411 bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev,
412 			      struct amdgpu_ih_ring *ih, uint64_t addr,
413 			      uint16_t pasid, uint64_t timestamp);
414 void amdgpu_gmc_filter_faults_remove(struct amdgpu_device *adev, uint64_t addr,
415 				     uint16_t pasid);
416 int amdgpu_gmc_ras_sw_init(struct amdgpu_device *adev);
417 int amdgpu_gmc_ras_late_init(struct amdgpu_device *adev);
418 void amdgpu_gmc_ras_fini(struct amdgpu_device *adev);
419 int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev);
420 void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
421 			      uint32_t vmhub, uint32_t flush_type);
422 int amdgpu_gmc_flush_gpu_tlb_pasid(struct amdgpu_device *adev, uint16_t pasid,
423 				   uint32_t flush_type, bool all_hub,
424 				   uint32_t inst);
425 void amdgpu_gmc_fw_reg_write_reg_wait(struct amdgpu_device *adev,
426 				      uint32_t reg0, uint32_t reg1,
427 				      uint32_t ref, uint32_t mask,
428 				      uint32_t xcc_inst);
429 
430 extern void amdgpu_gmc_tmz_set(struct amdgpu_device *adev);
431 extern void amdgpu_gmc_noretry_set(struct amdgpu_device *adev);
432 
433 extern void
434 amdgpu_gmc_set_vm_fault_masks(struct amdgpu_device *adev, int hub_type,
435 			      bool enable);
436 
437 void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev);
438 
439 void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev);
440 uint64_t amdgpu_gmc_vram_mc2pa(struct amdgpu_device *adev, uint64_t mc_addr);
441 uint64_t amdgpu_gmc_vram_pa(struct amdgpu_device *adev, struct amdgpu_bo *bo);
442 int amdgpu_gmc_vram_checking(struct amdgpu_device *adev);
443 int amdgpu_gmc_sysfs_init(struct amdgpu_device *adev);
444 void amdgpu_gmc_sysfs_fini(struct amdgpu_device *adev);
445 
446 int amdgpu_gmc_get_nps_memranges(struct amdgpu_device *adev,
447 				 struct amdgpu_mem_partition_info *mem_ranges,
448 				 uint8_t *exp_ranges);
449 
450 int amdgpu_gmc_request_memory_partition(struct amdgpu_device *adev,
451 					int nps_mode);
452 void amdgpu_gmc_prepare_nps_mode_change(struct amdgpu_device *adev);
453 bool amdgpu_gmc_need_reset_on_init(struct amdgpu_device *adev);
454 
455 #endif
456