1 /* 2 * Copyright 2018 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 24 #include "amdgpu.h" 25 #include "amdgpu_sdma.h" 26 #include "amdgpu_ras.h" 27 28 #define AMDGPU_CSA_SDMA_SIZE 64 29 /* SDMA CSA reside in the 3rd page of CSA */ 30 #define AMDGPU_CSA_SDMA_OFFSET (4096 * 2) 31 32 /* 33 * GPU SDMA IP block helpers function. 34 */ 35 36 struct amdgpu_sdma_instance *amdgpu_sdma_get_instance_from_ring(struct amdgpu_ring *ring) 37 { 38 struct amdgpu_device *adev = ring->adev; 39 int i; 40 41 for (i = 0; i < adev->sdma.num_instances; i++) 42 if (ring == &adev->sdma.instance[i].ring || 43 ring == &adev->sdma.instance[i].page) 44 return &adev->sdma.instance[i]; 45 46 return NULL; 47 } 48 49 int amdgpu_sdma_get_index_from_ring(struct amdgpu_ring *ring, uint32_t *index) 50 { 51 struct amdgpu_device *adev = ring->adev; 52 int i; 53 54 for (i = 0; i < adev->sdma.num_instances; i++) { 55 if (ring == &adev->sdma.instance[i].ring || 56 ring == &adev->sdma.instance[i].page) { 57 *index = i; 58 return 0; 59 } 60 } 61 62 return -EINVAL; 63 } 64 65 uint64_t amdgpu_sdma_get_csa_mc_addr(struct amdgpu_ring *ring, 66 unsigned vmid) 67 { 68 struct amdgpu_device *adev = ring->adev; 69 uint64_t csa_mc_addr; 70 uint32_t index = 0; 71 int r; 72 73 if (vmid == 0 || !amdgpu_mcbp) 74 return 0; 75 76 r = amdgpu_sdma_get_index_from_ring(ring, &index); 77 78 if (r || index > 31) 79 csa_mc_addr = 0; 80 else 81 csa_mc_addr = amdgpu_csa_vaddr(adev) + 82 AMDGPU_CSA_SDMA_OFFSET + 83 index * AMDGPU_CSA_SDMA_SIZE; 84 85 return csa_mc_addr; 86 } 87 88 int amdgpu_sdma_ras_late_init(struct amdgpu_device *adev, 89 void *ras_ih_info) 90 { 91 int r, i; 92 struct ras_ih_if *ih_info = (struct ras_ih_if *)ras_ih_info; 93 struct ras_fs_if fs_info = { 94 .sysfs_name = "sdma_err_count", 95 .debugfs_name = "sdma_err_inject", 96 }; 97 98 if (!ih_info) 99 return -EINVAL; 100 101 if (!adev->sdma.ras_if) { 102 adev->sdma.ras_if = kmalloc(sizeof(struct ras_common_if), GFP_KERNEL); 103 if (!adev->sdma.ras_if) 104 return -ENOMEM; 105 adev->sdma.ras_if->block = AMDGPU_RAS_BLOCK__SDMA; 106 adev->sdma.ras_if->type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE; 107 adev->sdma.ras_if->sub_block_index = 0; 108 strcpy(adev->sdma.ras_if->name, "sdma"); 109 } 110 fs_info.head = ih_info->head = *adev->sdma.ras_if; 111 112 r = amdgpu_ras_late_init(adev, adev->sdma.ras_if, 113 &fs_info, ih_info); 114 if (r) 115 goto free; 116 117 if (amdgpu_ras_is_supported(adev, adev->sdma.ras_if->block)) { 118 for (i = 0; i < adev->sdma.num_instances; i++) { 119 r = amdgpu_irq_get(adev, &adev->sdma.ecc_irq, 120 AMDGPU_SDMA_IRQ_INSTANCE0 + i); 121 if (r) 122 goto late_fini; 123 } 124 } else { 125 r = 0; 126 goto free; 127 } 128 129 return 0; 130 131 late_fini: 132 amdgpu_ras_late_fini(adev, adev->sdma.ras_if, ih_info); 133 free: 134 kfree(adev->sdma.ras_if); 135 adev->sdma.ras_if = NULL; 136 return r; 137 } 138 139 void amdgpu_sdma_ras_fini(struct amdgpu_device *adev) 140 { 141 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA) && 142 adev->sdma.ras_if) { 143 struct ras_common_if *ras_if = adev->sdma.ras_if; 144 struct ras_ih_if ih_info = { 145 .head = *ras_if, 146 /* the cb member will not be used by 147 * amdgpu_ras_interrupt_remove_handler, init it only 148 * to cheat the check in ras_late_fini 149 */ 150 .cb = amdgpu_sdma_process_ras_data_cb, 151 }; 152 153 amdgpu_ras_late_fini(adev, ras_if, &ih_info); 154 kfree(ras_if); 155 } 156 } 157 158 int amdgpu_sdma_process_ras_data_cb(struct amdgpu_device *adev, 159 void *err_data, 160 struct amdgpu_iv_entry *entry) 161 { 162 kgd2kfd_set_sram_ecc_flag(adev->kfd.dev); 163 amdgpu_ras_reset_gpu(adev); 164 165 return AMDGPU_RAS_SUCCESS; 166 } 167 168 int amdgpu_sdma_process_ecc_irq(struct amdgpu_device *adev, 169 struct amdgpu_irq_src *source, 170 struct amdgpu_iv_entry *entry) 171 { 172 struct ras_common_if *ras_if = adev->sdma.ras_if; 173 struct ras_dispatch_if ih_data = { 174 .entry = entry, 175 }; 176 177 if (!ras_if) 178 return 0; 179 180 ih_data.head = *ras_if; 181 182 amdgpu_ras_interrupt_dispatch(adev, &ih_data); 183 return 0; 184 } 185