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 struct firmware; 47 48 /* 49 * VMHUB structures, functions & helpers 50 */ 51 struct amdgpu_vmhub { 52 uint32_t ctx0_ptb_addr_lo32; 53 uint32_t ctx0_ptb_addr_hi32; 54 uint32_t vm_inv_eng0_req; 55 uint32_t vm_inv_eng0_ack; 56 uint32_t vm_context0_cntl; 57 uint32_t vm_l2_pro_fault_status; 58 uint32_t vm_l2_pro_fault_cntl; 59 }; 60 61 /* 62 * GPU MC structures, functions & helpers 63 */ 64 struct amdgpu_gmc_funcs { 65 /* flush the vm tlb via mmio */ 66 void (*flush_gpu_tlb)(struct amdgpu_device *adev, 67 uint32_t vmid); 68 /* flush the vm tlb via ring */ 69 uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid, 70 uint64_t pd_addr); 71 /* Change the VMID -> PASID mapping */ 72 void (*emit_pasid_mapping)(struct amdgpu_ring *ring, unsigned vmid, 73 unsigned pasid); 74 /* write pte/pde updates using the cpu */ 75 int (*set_pte_pde)(struct amdgpu_device *adev, 76 void *cpu_pt_addr, /* cpu addr of page table */ 77 uint32_t gpu_page_idx, /* pte/pde to update */ 78 uint64_t addr, /* addr to write into pte/pde */ 79 uint64_t flags); /* access flags */ 80 /* enable/disable PRT support */ 81 void (*set_prt)(struct amdgpu_device *adev, bool enable); 82 /* set pte flags based per asic */ 83 uint64_t (*get_vm_pte_flags)(struct amdgpu_device *adev, 84 uint32_t flags); 85 /* get the pde for a given mc addr */ 86 void (*get_vm_pde)(struct amdgpu_device *adev, int level, 87 u64 *dst, u64 *flags); 88 }; 89 90 struct amdgpu_xgmi { 91 /* from psp */ 92 u64 device_id; 93 u64 hive_id; 94 /* fixed per family */ 95 u64 node_segment_size; 96 /* physical node (0-3) */ 97 unsigned physical_node_id; 98 /* number of nodes (0-4) */ 99 unsigned num_physical_nodes; 100 /* gpu list in the same hive */ 101 struct list_head head; 102 }; 103 104 struct amdgpu_gmc { 105 resource_size_t aper_size; 106 resource_size_t aper_base; 107 /* for some chips with <= 32MB we need to lie 108 * about vram size near mc fb location */ 109 u64 mc_vram_size; 110 u64 visible_vram_size; 111 u64 agp_size; 112 u64 agp_start; 113 u64 agp_end; 114 u64 gart_size; 115 u64 gart_start; 116 u64 gart_end; 117 u64 vram_start; 118 u64 vram_end; 119 /* FB region , it's same as local vram region in single GPU, in XGMI 120 * configuration, this region covers all GPUs in the same hive , 121 * each GPU in the hive has the same view of this FB region . 122 * GPU0's vram starts at offset (0 * segment size) , 123 * GPU1 starts at offset (1 * segment size), etc. 124 */ 125 u64 fb_start; 126 u64 fb_end; 127 unsigned vram_width; 128 u64 real_vram_size; 129 int vram_mtrr; 130 u64 mc_mask; 131 const struct firmware *fw; /* MC firmware */ 132 uint32_t fw_version; 133 struct amdgpu_irq_src vm_fault; 134 uint32_t vram_type; 135 uint32_t srbm_soft_reset; 136 bool prt_warning; 137 uint64_t stolen_size; 138 /* apertures */ 139 u64 shared_aperture_start; 140 u64 shared_aperture_end; 141 u64 private_aperture_start; 142 u64 private_aperture_end; 143 /* protects concurrent invalidation */ 144 spinlock_t invalidate_lock; 145 bool translate_further; 146 struct kfd_vm_fault_info *vm_fault_info; 147 atomic_t vm_fault_info_updated; 148 149 const struct amdgpu_gmc_funcs *gmc_funcs; 150 151 struct amdgpu_xgmi xgmi; 152 }; 153 154 #define amdgpu_gmc_flush_gpu_tlb(adev, vmid) (adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid)) 155 #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr)) 156 #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid)) 157 #define amdgpu_gmc_set_pte_pde(adev, pt, idx, addr, flags) (adev)->gmc.gmc_funcs->set_pte_pde((adev), (pt), (idx), (addr), (flags)) 158 #define amdgpu_gmc_get_vm_pde(adev, level, dst, flags) (adev)->gmc.gmc_funcs->get_vm_pde((adev), (level), (dst), (flags)) 159 #define amdgpu_gmc_get_pte_flags(adev, flags) (adev)->gmc.gmc_funcs->get_vm_pte_flags((adev),(flags)) 160 161 /** 162 * amdgpu_gmc_vram_full_visible - Check if full VRAM is visible through the BAR 163 * 164 * @adev: amdgpu_device pointer 165 * 166 * Returns: 167 * True if full VRAM is visible through the BAR 168 */ 169 static inline bool amdgpu_gmc_vram_full_visible(struct amdgpu_gmc *gmc) 170 { 171 WARN_ON(gmc->real_vram_size < gmc->visible_vram_size); 172 173 return (gmc->real_vram_size == gmc->visible_vram_size); 174 } 175 176 /** 177 * amdgpu_gmc_sign_extend - sign extend the given gmc address 178 * 179 * @addr: address to extend 180 */ 181 static inline uint64_t amdgpu_gmc_sign_extend(uint64_t addr) 182 { 183 if (addr >= AMDGPU_GMC_HOLE_START) 184 addr |= AMDGPU_GMC_HOLE_END; 185 186 return addr; 187 } 188 189 void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level, 190 uint64_t *addr, uint64_t *flags); 191 uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo); 192 uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo); 193 void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc, 194 u64 base); 195 void amdgpu_gmc_gart_location(struct amdgpu_device *adev, 196 struct amdgpu_gmc *mc); 197 void amdgpu_gmc_agp_location(struct amdgpu_device *adev, 198 struct amdgpu_gmc *mc); 199 200 #endif 201