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.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 = amdgpu_dpm_get_ecc_info(adev, (void *)&(con->umc_ecc)); 37 if (ret == -EOPNOTSUPP) { 38 if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && 39 adev->umc.ras->ras_block.hw_ops->query_ras_error_count) 40 adev->umc.ras->ras_block.hw_ops->query_ras_error_count(adev, ras_error_status); 41 42 if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && 43 adev->umc.ras->ras_block.hw_ops->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->ras_block.hw_ops->query_ras_error_address(adev, ras_error_status); 60 } 61 } else if (!ret) { 62 if (adev->umc.ras && 63 adev->umc.ras->ecc_info_query_ras_error_count) 64 adev->umc.ras->ecc_info_query_ras_error_count(adev, ras_error_status); 65 66 if (adev->umc.ras && 67 adev->umc.ras->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->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 amdgpu_dpm_send_hbm_bad_pages_num(adev, con->eeprom_control.ras_num_recs); 100 } 101 102 if (reset) 103 amdgpu_ras_reset_gpu(adev); 104 } 105 106 kfree(err_data->err_addr); 107 return AMDGPU_RAS_SUCCESS; 108 } 109 110 int amdgpu_umc_poison_handler(struct amdgpu_device *adev, 111 void *ras_error_status, 112 bool reset) 113 { 114 int ret; 115 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 116 struct ras_common_if head = { 117 .block = AMDGPU_RAS_BLOCK__UMC, 118 }; 119 struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head); 120 121 ret = 122 amdgpu_umc_do_page_retirement(adev, ras_error_status, NULL, reset); 123 124 if (ret == AMDGPU_RAS_SUCCESS && obj) { 125 obj->err_data.ue_count += err_data->ue_count; 126 obj->err_data.ce_count += err_data->ce_count; 127 } 128 129 return ret; 130 } 131 132 int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, 133 void *ras_error_status, 134 struct amdgpu_iv_entry *entry) 135 { 136 return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry, true); 137 } 138 139 int amdgpu_umc_ras_late_init(struct amdgpu_device *adev, void *ras_info) 140 { 141 int r; 142 143 r = amdgpu_ras_block_late_init(adev, adev->umc.ras_if); 144 if (r) 145 return r; 146 147 if (amdgpu_ras_is_supported(adev, adev->umc.ras_if->block)) { 148 r = amdgpu_irq_get(adev, &adev->gmc.ecc_irq, 0); 149 if (r) 150 goto late_fini; 151 } 152 153 /* ras init of specific umc version */ 154 if (adev->umc.ras && 155 adev->umc.ras->err_cnt_init) 156 adev->umc.ras->err_cnt_init(adev); 157 158 return 0; 159 160 late_fini: 161 amdgpu_ras_block_late_fini(adev, adev->umc.ras_if); 162 return r; 163 } 164 165 void amdgpu_umc_ras_fini(struct amdgpu_device *adev) 166 { 167 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC) && 168 adev->umc.ras_if) 169 amdgpu_ras_block_late_fini(adev, adev->umc.ras_if); 170 } 171 172 int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, 173 struct amdgpu_irq_src *source, 174 struct amdgpu_iv_entry *entry) 175 { 176 struct ras_common_if *ras_if = adev->umc.ras_if; 177 struct ras_dispatch_if ih_data = { 178 .entry = entry, 179 }; 180 181 if (!ras_if) 182 return 0; 183 184 ih_data.head = *ras_if; 185 186 amdgpu_ras_interrupt_dispatch(adev, &ih_data); 187 return 0; 188 } 189 190 void amdgpu_umc_fill_error_record(struct ras_err_data *err_data, 191 uint64_t err_addr, 192 uint64_t retired_page, 193 uint32_t channel_index, 194 uint32_t umc_inst) 195 { 196 struct eeprom_table_record *err_rec = 197 &err_data->err_addr[err_data->err_addr_cnt]; 198 199 err_rec->address = err_addr; 200 /* page frame address is saved */ 201 err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT; 202 err_rec->ts = (uint64_t)ktime_get_real_seconds(); 203 err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE; 204 err_rec->cu = 0; 205 err_rec->mem_channel = channel_index; 206 err_rec->mcumc_id = umc_inst; 207 208 err_data->err_addr_cnt++; 209 } 210