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