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 
33 /* VA hole for 48bit addresses on Vega10 */
34 #define AMDGPU_GMC_HOLE_START	0x0000800000000000ULL
35 #define AMDGPU_GMC_HOLE_END	0xffff800000000000ULL
36 
37 /*
38  * Hardware is programmed as if the hole doesn't exists with start and end
39  * address values.
40  *
41  * This mask is used to remove the upper 16bits of the VA and so come up with
42  * the linear addr value.
43  */
44 #define AMDGPU_GMC_HOLE_MASK	0x0000ffffffffffffULL
45 
46 /*
47  * Ring size as power of two for the log of recent faults.
48  */
49 #define AMDGPU_GMC_FAULT_RING_ORDER	8
50 #define AMDGPU_GMC_FAULT_RING_SIZE	(1 << AMDGPU_GMC_FAULT_RING_ORDER)
51 
52 /*
53  * Hash size as power of two for the log of recent faults
54  */
55 #define AMDGPU_GMC_FAULT_HASH_ORDER	8
56 #define AMDGPU_GMC_FAULT_HASH_SIZE	(1 << AMDGPU_GMC_FAULT_HASH_ORDER)
57 
58 /*
59  * Number of IH timestamp ticks until a fault is considered handled
60  */
61 #define AMDGPU_GMC_FAULT_TIMEOUT	5000ULL
62 
63 struct firmware;
64 
65 /*
66  * GMC page fault information
67  */
68 struct amdgpu_gmc_fault {
69 	uint64_t	timestamp;
70 	uint64_t	next:AMDGPU_GMC_FAULT_RING_ORDER;
71 	uint64_t	key:52;
72 };
73 
74 /*
75  * VMHUB structures, functions & helpers
76  */
77 struct amdgpu_vmhub {
78 	uint32_t	ctx0_ptb_addr_lo32;
79 	uint32_t	ctx0_ptb_addr_hi32;
80 	uint32_t	vm_inv_eng0_req;
81 	uint32_t	vm_inv_eng0_ack;
82 	uint32_t	vm_context0_cntl;
83 	uint32_t	vm_l2_pro_fault_status;
84 	uint32_t	vm_l2_pro_fault_cntl;
85 };
86 
87 /*
88  * GPU MC structures, functions & helpers
89  */
90 struct amdgpu_gmc_funcs {
91 	/* flush the vm tlb via mmio */
92 	void (*flush_gpu_tlb)(struct amdgpu_device *adev,
93 			      uint32_t vmid, uint32_t flush_type);
94 	/* flush the vm tlb via ring */
95 	uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid,
96 				       uint64_t pd_addr);
97 	/* Change the VMID -> PASID mapping */
98 	void (*emit_pasid_mapping)(struct amdgpu_ring *ring, unsigned vmid,
99 				   unsigned pasid);
100 	/* enable/disable PRT support */
101 	void (*set_prt)(struct amdgpu_device *adev, bool enable);
102 	/* set pte flags based per asic */
103 	uint64_t (*get_vm_pte_flags)(struct amdgpu_device *adev,
104 				     uint32_t flags);
105 	/* get the pde for a given mc addr */
106 	void (*get_vm_pde)(struct amdgpu_device *adev, int level,
107 			   u64 *dst, u64 *flags);
108 };
109 
110 struct amdgpu_xgmi {
111 	/* from psp */
112 	u64 node_id;
113 	u64 hive_id;
114 	/* fixed per family */
115 	u64 node_segment_size;
116 	/* physical node (0-3) */
117 	unsigned physical_node_id;
118 	/* number of nodes (0-4) */
119 	unsigned num_physical_nodes;
120 	/* gpu list in the same hive */
121 	struct list_head head;
122 	bool supported;
123 };
124 
125 struct amdgpu_gmc {
126 	resource_size_t		aper_size;
127 	resource_size_t		aper_base;
128 	/* for some chips with <= 32MB we need to lie
129 	 * about vram size near mc fb location */
130 	u64			mc_vram_size;
131 	u64			visible_vram_size;
132 	u64			agp_size;
133 	u64			agp_start;
134 	u64			agp_end;
135 	u64			gart_size;
136 	u64			gart_start;
137 	u64			gart_end;
138 	u64			vram_start;
139 	u64			vram_end;
140 	/* FB region , it's same as local vram region in single GPU, in XGMI
141 	 * configuration, this region covers all GPUs in the same hive ,
142 	 * each GPU in the hive has the same view of this FB region .
143 	 * GPU0's vram starts at offset (0 * segment size) ,
144 	 * GPU1 starts at offset (1 * segment size), etc.
145 	 */
146 	u64			fb_start;
147 	u64			fb_end;
148 	unsigned		vram_width;
149 	u64			real_vram_size;
150 	int			vram_mtrr;
151 	u64                     mc_mask;
152 	const struct firmware   *fw;	/* MC firmware */
153 	uint32_t                fw_version;
154 	struct amdgpu_irq_src	vm_fault;
155 	uint32_t		vram_type;
156 	uint32_t                srbm_soft_reset;
157 	bool			prt_warning;
158 	uint64_t		stolen_size;
159 	/* apertures */
160 	u64			shared_aperture_start;
161 	u64			shared_aperture_end;
162 	u64			private_aperture_start;
163 	u64			private_aperture_end;
164 	/* protects concurrent invalidation */
165 	spinlock_t		invalidate_lock;
166 	bool			translate_further;
167 	struct kfd_vm_fault_info *vm_fault_info;
168 	atomic_t		vm_fault_info_updated;
169 
170 	struct amdgpu_gmc_fault	fault_ring[AMDGPU_GMC_FAULT_RING_SIZE];
171 	struct {
172 		uint64_t	idx:AMDGPU_GMC_FAULT_RING_ORDER;
173 	} fault_hash[AMDGPU_GMC_FAULT_HASH_SIZE];
174 	uint64_t		last_fault:AMDGPU_GMC_FAULT_RING_ORDER;
175 
176 	const struct amdgpu_gmc_funcs	*gmc_funcs;
177 
178 	struct amdgpu_xgmi xgmi;
179 	struct amdgpu_irq_src	ecc_irq;
180 	struct ras_common_if    *ras_if;
181 };
182 
183 #define amdgpu_gmc_flush_gpu_tlb(adev, vmid, type) (adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (type))
184 #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
185 #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
186 #define amdgpu_gmc_get_vm_pde(adev, level, dst, flags) (adev)->gmc.gmc_funcs->get_vm_pde((adev), (level), (dst), (flags))
187 #define amdgpu_gmc_get_pte_flags(adev, flags) (adev)->gmc.gmc_funcs->get_vm_pte_flags((adev),(flags))
188 
189 /**
190  * amdgpu_gmc_vram_full_visible - Check if full VRAM is visible through the BAR
191  *
192  * @adev: amdgpu_device pointer
193  *
194  * Returns:
195  * True if full VRAM is visible through the BAR
196  */
197 static inline bool amdgpu_gmc_vram_full_visible(struct amdgpu_gmc *gmc)
198 {
199 	WARN_ON(gmc->real_vram_size < gmc->visible_vram_size);
200 
201 	return (gmc->real_vram_size == gmc->visible_vram_size);
202 }
203 
204 /**
205  * amdgpu_gmc_sign_extend - sign extend the given gmc address
206  *
207  * @addr: address to extend
208  */
209 static inline uint64_t amdgpu_gmc_sign_extend(uint64_t addr)
210 {
211 	if (addr >= AMDGPU_GMC_HOLE_START)
212 		addr |= AMDGPU_GMC_HOLE_END;
213 
214 	return addr;
215 }
216 
217 void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level,
218 			       uint64_t *addr, uint64_t *flags);
219 int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
220 				uint32_t gpu_page_idx, uint64_t addr,
221 				uint64_t flags);
222 uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo);
223 uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo);
224 void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
225 			      u64 base);
226 void amdgpu_gmc_gart_location(struct amdgpu_device *adev,
227 			      struct amdgpu_gmc *mc);
228 void amdgpu_gmc_agp_location(struct amdgpu_device *adev,
229 			     struct amdgpu_gmc *mc);
230 bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev, uint64_t addr,
231 			      uint16_t pasid, uint64_t timestamp);
232 
233 #endif
234