]> Git Repo - linux.git/blob - drivers/gpu/drm/amd/amdgpu/umc_v6_1.c
drm/amdgpu: use 64bit operation macros for umc
[linux.git] / drivers / gpu / drm / amd / amdgpu / umc_v6_1.c
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 #include "umc_v6_1.h"
24 #include "amdgpu_ras.h"
25 #include "amdgpu.h"
26
27 #include "rsmu/rsmu_0_0_2_offset.h"
28 #include "rsmu/rsmu_0_0_2_sh_mask.h"
29 #include "umc/umc_6_1_1_offset.h"
30 #include "umc/umc_6_1_1_sh_mask.h"
31
32 static void umc_v6_1_enable_umc_index_mode(struct amdgpu_device *adev,
33                                            uint32_t umc_instance)
34 {
35         uint32_t rsmu_umc_index;
36
37         rsmu_umc_index = RREG32_SOC15(RSMU, 0,
38                         mmRSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU);
39         rsmu_umc_index = REG_SET_FIELD(rsmu_umc_index,
40                         RSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU,
41                         RSMU_UMC_INDEX_MODE_EN, 1);
42         rsmu_umc_index = REG_SET_FIELD(rsmu_umc_index,
43                         RSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU,
44                         RSMU_UMC_INDEX_INSTANCE, umc_instance);
45         rsmu_umc_index = REG_SET_FIELD(rsmu_umc_index,
46                         RSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU,
47                         RSMU_UMC_INDEX_WREN, 1 << umc_instance);
48         WREG32_SOC15(RSMU, 0, mmRSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU,
49                                 rsmu_umc_index);
50 }
51
52 static void umc_v6_1_disable_umc_index_mode(struct amdgpu_device *adev)
53 {
54         WREG32_FIELD15(RSMU, 0, RSMU_UMC_INDEX_REGISTER_NBIF_VG20_GPU,
55                         RSMU_UMC_INDEX_MODE_EN, 0);
56 }
57
58 static void umc_v6_1_query_correctable_error_count(struct amdgpu_device *adev,
59                                                    uint32_t umc_reg_offset,
60                                                    unsigned long *error_count)
61 {
62         uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr;
63         uint32_t ecc_err_cnt, ecc_err_cnt_addr;
64         uint64_t mc_umc_status;
65         uint32_t mc_umc_status_addr;
66
67         ecc_err_cnt_sel_addr =
68                 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCntSel);
69         ecc_err_cnt_addr =
70                 SOC15_REG_OFFSET(UMC, 0, mmUMCCH0_0_EccErrCnt);
71         mc_umc_status_addr =
72                 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
73
74         /* select the lower chip and check the error count */
75         ecc_err_cnt_sel = RREG32(ecc_err_cnt_sel_addr + umc_reg_offset);
76         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel,
77                                         EccErrCntCsSel, 0);
78         WREG32(ecc_err_cnt_sel_addr + umc_reg_offset, ecc_err_cnt_sel);
79         ecc_err_cnt = RREG32(ecc_err_cnt_addr + umc_reg_offset);
80         *error_count +=
81                 REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_EccErrCnt, EccErrCnt);
82         /* clear the lower chip err count */
83         WREG32(ecc_err_cnt_addr + umc_reg_offset, 0);
84
85         /* select the higher chip and check the err counter */
86         ecc_err_cnt_sel = REG_SET_FIELD(ecc_err_cnt_sel, UMCCH0_0_EccErrCntSel,
87                                         EccErrCntCsSel, 1);
88         WREG32(ecc_err_cnt_sel_addr + umc_reg_offset, ecc_err_cnt_sel);
89         ecc_err_cnt = RREG32(ecc_err_cnt_addr + umc_reg_offset);
90         *error_count +=
91                 REG_GET_FIELD(ecc_err_cnt, UMCCH0_0_EccErrCnt, EccErrCnt);
92         /* clear the higher chip err count */
93         WREG32(ecc_err_cnt_addr + umc_reg_offset, 0);
94
95         /* check for SRAM correctable error
96           MCUMC_STATUS is a 64 bit register */
97         mc_umc_status = RREG64(mc_umc_status_addr + umc_reg_offset);
98         if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 6 &&
99             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
100             REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)
101                 *error_count += 1;
102 }
103
104 static void umc_v6_1_querry_uncorrectable_error_count(struct amdgpu_device *adev,
105                                                       uint32_t umc_reg_offset,
106                                                       unsigned long *error_count)
107 {
108         uint64_t mc_umc_status;
109         uint32_t mc_umc_status_addr;
110
111         mc_umc_status_addr =
112                 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
113
114         /* check the MCUMC_STATUS */
115         mc_umc_status = RREG64(mc_umc_status_addr + umc_reg_offset);
116         if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
117                 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, ErrorCodeExt) == 6 &&
118                 (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
119                 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
120                 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
121                 REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1))
122                 *error_count += 1;
123 }
124
125 static void umc_v6_1_query_ras_error_count(struct amdgpu_device *adev,
126                                            void *ras_error_status)
127 {
128         struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
129         uint32_t umc_inst, channel_inst, umc_reg_offset, mc_umc_status_addr;
130
131         mc_umc_status_addr =
132                 SOC15_REG_OFFSET(UMC, 0, mmMCA_UMC_UMC0_MCUMC_STATUST0);
133
134         for (umc_inst = 0; umc_inst < UMC_V6_1_UMC_INSTANCE_NUM; umc_inst++) {
135                 /* enable the index mode to query eror count per channel */
136                 umc_v6_1_enable_umc_index_mode(adev, umc_inst);
137                 for (channel_inst = 0; channel_inst < UMC_V6_1_CHANNEL_INSTANCE_NUM; channel_inst++) {
138                         /* calc the register offset according to channel instance */
139                         umc_reg_offset = UMC_V6_1_PER_CHANNEL_OFFSET * channel_inst;
140                         umc_v6_1_query_correctable_error_count(adev, umc_reg_offset,
141                                                                &(err_data->ce_count));
142                         umc_v6_1_querry_uncorrectable_error_count(adev, umc_reg_offset,
143                                                                   &(err_data->ue_count));
144                         /* clear umc status */
145                         WREG64(mc_umc_status_addr + umc_reg_offset, 0x0ULL);
146                 }
147         }
148         umc_v6_1_disable_umc_index_mode(adev);
149 }
150
151 const struct amdgpu_umc_funcs umc_v6_1_funcs = {
152         .query_ras_error_count = umc_v6_1_query_ras_error_count,
153 };
This page took 0.042698 seconds and 4 git commands to generate.