xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h (revision cb8fef4b07f71da01843457b66e579ad248a625e)
1 /*
2  * Copyright (C) 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
12  * in all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
15  * OR 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) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
18  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
19  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
20  */
21 #ifndef __AMDGPU_UMC_H__
22 #define __AMDGPU_UMC_H__
23 #include "amdgpu_ras.h"
24 /*
25  * (addr / 256) * 4096, the higher 26 bits in ErrorAddr
26  * is the index of 4KB block
27  */
28 #define ADDR_OF_4KB_BLOCK(addr)			(((addr) & ~0xffULL) << 4)
29 /*
30  * (addr / 256) * 8192, the higher 26 bits in ErrorAddr
31  * is the index of 8KB block
32  */
33 #define ADDR_OF_8KB_BLOCK(addr)			(((addr) & ~0xffULL) << 5)
34 /*
35  * (addr / 256) * 32768, the higher 26 bits in ErrorAddr
36  * is the index of 8KB block
37  */
38 #define ADDR_OF_32KB_BLOCK(addr)			(((addr) & ~0xffULL) << 7)
39 /* channel index is the index of 256B block */
40 #define ADDR_OF_256B_BLOCK(channel_index)	((channel_index) << 8)
41 /* offset in 256B block */
42 #define OFFSET_IN_256B_BLOCK(addr)		((addr) & 0xffULL)
43 
44 #define LOOP_UMC_INST(umc_inst) for ((umc_inst) = 0; (umc_inst) < adev->umc.umc_inst_num; (umc_inst)++)
45 #define LOOP_UMC_CH_INST(ch_inst) for ((ch_inst) = 0; (ch_inst) < adev->umc.channel_inst_num; (ch_inst)++)
46 #define LOOP_UMC_INST_AND_CH(umc_inst, ch_inst) LOOP_UMC_INST((umc_inst)) LOOP_UMC_CH_INST((ch_inst))
47 
48 /* Page retirement tag */
49 #define UMC_ECC_NEW_DETECTED_TAG       0x1
50 /*
51  * a flag to indicate v2 of channel index stored in eeprom
52  *
53  * v1 (legacy way): store channel index within a umc instance in eeprom
54  *    range in UMC v12: 0 ~ 7
55  * v2: store global channel index in eeprom
56  *    range in UMC v12: 0 ~ 127
57  *
58  * NOTE: it's better to store it in eeprom_table_record.mem_channel,
59  * but there is only 8 bits in mem_channel, and the channel number may
60  * increase in the future, we decide to save it in
61  * eeprom_table_record.retired_page. retired_page is useless in v2,
62  * we depend on eeprom_table_record.address instead of retired_page in v2.
63  * Only 48 bits are saved on eeprom, use bit 47 here.
64  */
65 #define UMC_CHANNEL_IDX_V2	BIT_ULL(47)
66 
67 /*
68  * save nps value to eeprom_table_record.retired_page[47:40],
69  * the channel index flag above will be retired.
70  */
71 #define UMC_NPS_SHIFT 40
72 #define UMC_NPS_MASK 0xffULL
73 
74 typedef int (*umc_func)(struct amdgpu_device *adev, uint32_t node_inst,
75 			uint32_t umc_inst, uint32_t ch_inst, void *data);
76 
77 struct amdgpu_umc_ras {
78 	struct amdgpu_ras_block_object ras_block;
79 	void (*err_cnt_init)(struct amdgpu_device *adev);
80 	bool (*query_ras_poison_mode)(struct amdgpu_device *adev);
81 	void (*ecc_info_query_ras_error_count)(struct amdgpu_device *adev,
82 				      void *ras_error_status);
83 	void (*ecc_info_query_ras_error_address)(struct amdgpu_device *adev,
84 					void *ras_error_status);
85 };
86 
87 struct amdgpu_umc_funcs {
88 	void (*init_registers)(struct amdgpu_device *adev);
89 };
90 
91 struct amdgpu_umc {
92 	/* max error count in one ras query call */
93 	uint32_t max_ras_err_cnt_per_query;
94 	/* number of umc channel instance with memory map register access */
95 	uint32_t channel_inst_num;
96 	/* number of umc instance with memory map register access */
97 	uint32_t umc_inst_num;
98 
99 	/* Total number of umc node instance including harvest one */
100 	uint32_t node_inst_num;
101 
102 	/* UMC regiser per channel offset */
103 	uint32_t channel_offs;
104 	/* how many pages are retired in one UE */
105 	uint32_t retire_unit;
106 	/* channel index table of interleaved memory */
107 	const uint32_t *channel_idx_tbl;
108 	struct ras_common_if *ras_if;
109 
110 	const struct amdgpu_umc_funcs *funcs;
111 	struct amdgpu_umc_ras *ras;
112 
113 	/* active mask for umc node instance */
114 	u64 active_mask;
115 
116 	unsigned long err_addr_cnt;
117 };
118 
119 int amdgpu_umc_ras_sw_init(struct amdgpu_device *adev);
120 int amdgpu_umc_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block);
121 int amdgpu_umc_poison_handler(struct amdgpu_device *adev,
122 			enum amdgpu_ras_block block, uint32_t reset);
123 int amdgpu_umc_pasid_poison_handler(struct amdgpu_device *adev,
124 			enum amdgpu_ras_block block, uint16_t pasid,
125 			pasid_notify pasid_fn, void *data, uint32_t reset);
126 int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev,
127 		struct amdgpu_irq_src *source,
128 		struct amdgpu_iv_entry *entry);
129 int amdgpu_umc_uniras_process_ecc_irq(struct amdgpu_device *adev,
130 		struct amdgpu_irq_src *source,
131 		struct amdgpu_iv_entry *entry);
132 int amdgpu_umc_fill_error_record(struct ras_err_data *err_data,
133 		uint64_t err_addr,
134 		uint64_t retired_page,
135 		uint32_t channel_index,
136 		uint32_t umc_inst);
137 
138 int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev,
139 		void *ras_error_status,
140 		struct amdgpu_iv_entry *entry);
141 int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
142 			uint64_t err_addr, uint32_t ch_inst, uint32_t umc_inst);
143 
144 int amdgpu_umc_loop_channels(struct amdgpu_device *adev,
145 			umc_func func, void *data);
146 
147 void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
148 			void *ras_error_status);
149 #endif
150