1 /* 2 * Copyright 2019 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_ras.h" 25 26 static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, 27 void *ras_error_status, 28 struct amdgpu_iv_entry *entry, 29 bool reset) 30 { 31 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 32 struct amdgpu_ras *con = amdgpu_ras_get_context(adev); 33 int ret = 0; 34 35 kgd2kfd_set_sram_ecc_flag(adev->kfd.dev); 36 ret = smu_get_ecc_info(&adev->smu, (void *)&(con->umc_ecc)); 37 if (ret == -EOPNOTSUPP) { 38 if (adev->umc.ras_funcs && 39 adev->umc.ras_funcs->query_ras_error_count) 40 adev->umc.ras_funcs->query_ras_error_count(adev, ras_error_status); 41 42 if (adev->umc.ras_funcs && 43 adev->umc.ras_funcs->query_ras_error_address && 44 adev->umc.max_ras_err_cnt_per_query) { 45 err_data->err_addr = 46 kcalloc(adev->umc.max_ras_err_cnt_per_query, 47 sizeof(struct eeprom_table_record), GFP_KERNEL); 48 49 /* still call query_ras_error_address to clear error status 50 * even NOMEM error is encountered 51 */ 52 if(!err_data->err_addr) 53 dev_warn(adev->dev, "Failed to alloc memory for " 54 "umc error address record!\n"); 55 56 /* umc query_ras_error_address is also responsible for clearing 57 * error status 58 */ 59 adev->umc.ras_funcs->query_ras_error_address(adev, ras_error_status); 60 } 61 } else if (!ret) { 62 if (adev->umc.ras_funcs && 63 adev->umc.ras_funcs->ecc_info_query_ras_error_count) 64 adev->umc.ras_funcs->ecc_info_query_ras_error_count(adev, ras_error_status); 65 66 if (adev->umc.ras_funcs && 67 adev->umc.ras_funcs->ecc_info_query_ras_error_address && 68 adev->umc.max_ras_err_cnt_per_query) { 69 err_data->err_addr = 70 kcalloc(adev->umc.max_ras_err_cnt_per_query, 71 sizeof(struct eeprom_table_record), GFP_KERNEL); 72 73 /* still call query_ras_error_address to clear error status 74 * even NOMEM error is encountered 75 */ 76 if(!err_data->err_addr) 77 dev_warn(adev->dev, "Failed to alloc memory for " 78 "umc error address record!\n"); 79 80 /* umc query_ras_error_address is also responsible for clearing 81 * error status 82 */ 83 adev->umc.ras_funcs->ecc_info_query_ras_error_address(adev, ras_error_status); 84 } 85 } 86 87 /* only uncorrectable error needs gpu reset */ 88 if (err_data->ue_count) { 89 dev_info(adev->dev, "%ld uncorrectable hardware errors " 90 "detected in UMC block\n", 91 err_data->ue_count); 92 93 if ((amdgpu_bad_page_threshold != 0) && 94 err_data->err_addr_cnt) { 95 amdgpu_ras_add_bad_pages(adev, err_data->err_addr, 96 err_data->err_addr_cnt); 97 amdgpu_ras_save_bad_pages(adev); 98 99 if (adev->smu.ppt_funcs && adev->smu.ppt_funcs->send_hbm_bad_pages_num) 100 adev->smu.ppt_funcs->send_hbm_bad_pages_num(&adev->smu, con->eeprom_control.ras_num_recs); 101 } 102 103 if (reset) 104 amdgpu_ras_reset_gpu(adev); 105 } 106 107 kfree(err_data->err_addr); 108 return AMDGPU_RAS_SUCCESS; 109 } 110 111 int amdgpu_umc_poison_handler(struct amdgpu_device *adev, 112 void *ras_error_status, 113 bool reset) 114 { 115 int ret; 116 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 117 struct ras_common_if head = { 118 .block = AMDGPU_RAS_BLOCK__UMC, 119 }; 120 struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head); 121 122 ret = 123 amdgpu_umc_do_page_retirement(adev, ras_error_status, NULL, reset); 124 125 if (ret == AMDGPU_RAS_SUCCESS && obj) { 126 obj->err_data.ue_count += err_data->ue_count; 127 obj->err_data.ce_count += err_data->ce_count; 128 } 129 130 return ret; 131 } 132 133 static int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, 134 void *ras_error_status, 135 struct amdgpu_iv_entry *entry) 136 { 137 return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry, true); 138 } 139 140 int amdgpu_umc_ras_late_init(struct amdgpu_device *adev) 141 { 142 int r; 143 struct ras_fs_if fs_info = { 144 .sysfs_name = "umc_err_count", 145 }; 146 struct ras_ih_if ih_info = { 147 .cb = amdgpu_umc_process_ras_data_cb, 148 }; 149 150 if (!adev->umc.ras_if) { 151 adev->umc.ras_if = 152 kmalloc(sizeof(struct ras_common_if), GFP_KERNEL); 153 if (!adev->umc.ras_if) 154 return -ENOMEM; 155 adev->umc.ras_if->block = AMDGPU_RAS_BLOCK__UMC; 156 adev->umc.ras_if->type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE; 157 adev->umc.ras_if->sub_block_index = 0; 158 } 159 ih_info.head = fs_info.head = *adev->umc.ras_if; 160 161 r = amdgpu_ras_late_init(adev, adev->umc.ras_if, 162 &fs_info, &ih_info); 163 if (r) 164 goto free; 165 166 if (amdgpu_ras_is_supported(adev, adev->umc.ras_if->block)) { 167 r = amdgpu_irq_get(adev, &adev->gmc.ecc_irq, 0); 168 if (r) 169 goto late_fini; 170 } else { 171 r = 0; 172 goto free; 173 } 174 175 /* ras init of specific umc version */ 176 if (adev->umc.ras_funcs && 177 adev->umc.ras_funcs->err_cnt_init) 178 adev->umc.ras_funcs->err_cnt_init(adev); 179 180 return 0; 181 182 late_fini: 183 amdgpu_ras_late_fini(adev, adev->umc.ras_if, &ih_info); 184 free: 185 kfree(adev->umc.ras_if); 186 adev->umc.ras_if = NULL; 187 return r; 188 } 189 190 void amdgpu_umc_ras_fini(struct amdgpu_device *adev) 191 { 192 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC) && 193 adev->umc.ras_if) { 194 struct ras_common_if *ras_if = adev->umc.ras_if; 195 struct ras_ih_if ih_info = { 196 .head = *ras_if, 197 .cb = amdgpu_umc_process_ras_data_cb, 198 }; 199 200 amdgpu_ras_late_fini(adev, ras_if, &ih_info); 201 kfree(ras_if); 202 } 203 } 204 205 int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, 206 struct amdgpu_irq_src *source, 207 struct amdgpu_iv_entry *entry) 208 { 209 struct ras_common_if *ras_if = adev->umc.ras_if; 210 struct ras_dispatch_if ih_data = { 211 .entry = entry, 212 }; 213 214 if (!ras_if) 215 return 0; 216 217 ih_data.head = *ras_if; 218 219 amdgpu_ras_interrupt_dispatch(adev, &ih_data); 220 return 0; 221 } 222