2 * Copyright 2021 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
24 #define SWSMU_CODE_LAYER_L2
26 #include <linux/firmware.h>
28 #include "amdgpu_smu.h"
29 #include "atomfirmware.h"
30 #include "amdgpu_atomfirmware.h"
31 #include "amdgpu_atombios.h"
32 #include "smu_v13_0_6_pmfw.h"
33 #include "smu13_driver_if_v13_0_6.h"
34 #include "smu_v13_0_6_ppsmc.h"
35 #include "soc15_common.h"
37 #include "power_state.h"
38 #include "smu_v13_0.h"
39 #include "smu_v13_0_6_ppt.h"
40 #include "nbio/nbio_7_4_offset.h"
41 #include "nbio/nbio_7_4_sh_mask.h"
42 #include "thm/thm_11_0_2_offset.h"
43 #include "thm/thm_11_0_2_sh_mask.h"
44 #include "amdgpu_xgmi.h"
45 #include <linux/pci.h>
46 #include "amdgpu_ras.h"
47 #include "amdgpu_mca.h"
48 #include "amdgpu_aca.h"
50 #include "mp/mp_13_0_6_offset.h"
51 #include "mp/mp_13_0_6_sh_mask.h"
52 #include "umc_v12_0.h"
55 #undef smnMP1_FIRMWARE_FLAGS
57 /* TODO: Check final register offsets */
58 #define MP1_Public 0x03b00000
59 #define smnMP1_FIRMWARE_FLAGS 0x3010028
61 * DO NOT use these for err/warn/info/debug messages.
62 * Use dev_err, dev_warn, dev_info and dev_dbg instead.
63 * They are more MGPU friendly.
70 MODULE_FIRMWARE("amdgpu/smu_13_0_6.bin");
72 #define to_amdgpu_device(x) (container_of(x, struct amdgpu_device, pm.smu_i2c))
74 #define SMU_13_0_6_FEA_MAP(smu_feature, smu_13_0_6_feature) \
75 [smu_feature] = { 1, (smu_13_0_6_feature) }
77 #define FEATURE_MASK(feature) (1ULL << feature)
78 #define SMC_DPM_FEATURE \
79 (FEATURE_MASK(FEATURE_DATA_CALCULATION) | \
80 FEATURE_MASK(FEATURE_DPM_GFXCLK) | FEATURE_MASK(FEATURE_DPM_UCLK) | \
81 FEATURE_MASK(FEATURE_DPM_SOCCLK) | FEATURE_MASK(FEATURE_DPM_FCLK) | \
82 FEATURE_MASK(FEATURE_DPM_LCLK) | FEATURE_MASK(FEATURE_DPM_XGMI) | \
83 FEATURE_MASK(FEATURE_DPM_VCN))
85 /* possible frequency drift (1Mhz) */
88 #define smnPCIE_ESM_CTRL 0x93D0
89 #define smnPCIE_LC_LINK_WIDTH_CNTL 0x1a340288
90 #define PCIE_LC_LINK_WIDTH_CNTL__LC_LINK_WIDTH_RD_MASK 0x00000070L
91 #define PCIE_LC_LINK_WIDTH_CNTL__LC_LINK_WIDTH_RD__SHIFT 0x4
92 #define MAX_LINK_WIDTH 6
94 #define smnPCIE_LC_SPEED_CNTL 0x1a340290
95 #define PCIE_LC_SPEED_CNTL__LC_CURRENT_DATA_RATE_MASK 0xE0
96 #define PCIE_LC_SPEED_CNTL__LC_CURRENT_DATA_RATE__SHIFT 0x5
97 #define LINK_SPEED_MAX 4
99 #define SMU_13_0_6_DSCLK_THRESHOLD 140
101 #define MCA_BANK_IPID(_ip, _hwid, _type) \
102 [AMDGPU_MCA_IP_##_ip] = { .hwid = _hwid, .mcatype = _type, }
104 struct mca_bank_ipid {
105 enum amdgpu_mca_ip ip;
110 struct mca_ras_info {
111 enum amdgpu_ras_block blkid;
112 enum amdgpu_mca_ip ip;
115 int (*get_err_count)(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
116 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry, uint32_t *count);
117 bool (*bank_is_valid)(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
118 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry);
121 #define P2S_TABLE_ID_A 0x50325341
122 #define P2S_TABLE_ID_X 0x50325358
125 static const struct cmn2asic_msg_mapping smu_v13_0_6_message_map[SMU_MSG_MAX_COUNT] = {
126 MSG_MAP(TestMessage, PPSMC_MSG_TestMessage, 0),
127 MSG_MAP(GetSmuVersion, PPSMC_MSG_GetSmuVersion, 1),
128 MSG_MAP(GetDriverIfVersion, PPSMC_MSG_GetDriverIfVersion, 1),
129 MSG_MAP(EnableAllSmuFeatures, PPSMC_MSG_EnableAllSmuFeatures, 0),
130 MSG_MAP(DisableAllSmuFeatures, PPSMC_MSG_DisableAllSmuFeatures, 0),
131 MSG_MAP(RequestI2cTransaction, PPSMC_MSG_RequestI2cTransaction, 0),
132 MSG_MAP(GetMetricsTable, PPSMC_MSG_GetMetricsTable, 1),
133 MSG_MAP(GetMetricsVersion, PPSMC_MSG_GetMetricsVersion, 1),
134 MSG_MAP(GetEnabledSmuFeaturesHigh, PPSMC_MSG_GetEnabledSmuFeaturesHigh, 1),
135 MSG_MAP(GetEnabledSmuFeaturesLow, PPSMC_MSG_GetEnabledSmuFeaturesLow, 1),
136 MSG_MAP(SetDriverDramAddrHigh, PPSMC_MSG_SetDriverDramAddrHigh, 1),
137 MSG_MAP(SetDriverDramAddrLow, PPSMC_MSG_SetDriverDramAddrLow, 1),
138 MSG_MAP(SetToolsDramAddrHigh, PPSMC_MSG_SetToolsDramAddrHigh, 0),
139 MSG_MAP(SetToolsDramAddrLow, PPSMC_MSG_SetToolsDramAddrLow, 0),
140 MSG_MAP(SetSoftMinByFreq, PPSMC_MSG_SetSoftMinByFreq, 0),
141 MSG_MAP(SetSoftMaxByFreq, PPSMC_MSG_SetSoftMaxByFreq, 0),
142 MSG_MAP(GetMinDpmFreq, PPSMC_MSG_GetMinDpmFreq, 1),
143 MSG_MAP(GetMaxDpmFreq, PPSMC_MSG_GetMaxDpmFreq, 1),
144 MSG_MAP(GetDpmFreqByIndex, PPSMC_MSG_GetDpmFreqByIndex, 1),
145 MSG_MAP(SetPptLimit, PPSMC_MSG_SetPptLimit, 0),
146 MSG_MAP(GetPptLimit, PPSMC_MSG_GetPptLimit, 1),
147 MSG_MAP(GfxDeviceDriverReset, PPSMC_MSG_GfxDriverReset, 0),
148 MSG_MAP(DramLogSetDramAddrHigh, PPSMC_MSG_DramLogSetDramAddrHigh, 0),
149 MSG_MAP(DramLogSetDramAddrLow, PPSMC_MSG_DramLogSetDramAddrLow, 0),
150 MSG_MAP(DramLogSetDramSize, PPSMC_MSG_DramLogSetDramSize, 0),
151 MSG_MAP(GetDebugData, PPSMC_MSG_GetDebugData, 0),
152 MSG_MAP(SetNumBadHbmPagesRetired, PPSMC_MSG_SetNumBadHbmPagesRetired, 0),
153 MSG_MAP(DFCstateControl, PPSMC_MSG_DFCstateControl, 0),
154 MSG_MAP(GetGmiPwrDnHyst, PPSMC_MSG_GetGmiPwrDnHyst, 0),
155 MSG_MAP(SetGmiPwrDnHyst, PPSMC_MSG_SetGmiPwrDnHyst, 0),
156 MSG_MAP(GmiPwrDnControl, PPSMC_MSG_GmiPwrDnControl, 0),
157 MSG_MAP(EnterGfxoff, PPSMC_MSG_EnterGfxoff, 0),
158 MSG_MAP(ExitGfxoff, PPSMC_MSG_ExitGfxoff, 0),
159 MSG_MAP(EnableDeterminism, PPSMC_MSG_EnableDeterminism, 0),
160 MSG_MAP(DisableDeterminism, PPSMC_MSG_DisableDeterminism, 0),
161 MSG_MAP(GfxDriverResetRecovery, PPSMC_MSG_GfxDriverResetRecovery, 0),
162 MSG_MAP(GetMinGfxclkFrequency, PPSMC_MSG_GetMinGfxDpmFreq, 1),
163 MSG_MAP(GetMaxGfxclkFrequency, PPSMC_MSG_GetMaxGfxDpmFreq, 1),
164 MSG_MAP(SetSoftMinGfxclk, PPSMC_MSG_SetSoftMinGfxClk, 1),
165 MSG_MAP(SetSoftMaxGfxClk, PPSMC_MSG_SetSoftMaxGfxClk, 1),
166 MSG_MAP(PrepareMp1ForUnload, PPSMC_MSG_PrepareForDriverUnload, 0),
167 MSG_MAP(GetCTFLimit, PPSMC_MSG_GetCTFLimit, 0),
168 MSG_MAP(GetThermalLimit, PPSMC_MSG_ReadThrottlerLimit, 0),
169 MSG_MAP(ClearMcaOnRead, PPSMC_MSG_ClearMcaOnRead, 0),
170 MSG_MAP(QueryValidMcaCount, PPSMC_MSG_QueryValidMcaCount, 0),
171 MSG_MAP(QueryValidMcaCeCount, PPSMC_MSG_QueryValidMcaCeCount, 0),
172 MSG_MAP(McaBankDumpDW, PPSMC_MSG_McaBankDumpDW, 0),
173 MSG_MAP(McaBankCeDumpDW, PPSMC_MSG_McaBankCeDumpDW, 0),
174 MSG_MAP(SelectPLPDMode, PPSMC_MSG_SelectPLPDMode, 0),
175 MSG_MAP(RmaDueToBadPageThreshold, PPSMC_MSG_RmaDueToBadPageThreshold, 0),
179 static const struct cmn2asic_mapping smu_v13_0_6_clk_map[SMU_CLK_COUNT] = {
180 CLK_MAP(SOCCLK, PPCLK_SOCCLK),
181 CLK_MAP(FCLK, PPCLK_FCLK),
182 CLK_MAP(UCLK, PPCLK_UCLK),
183 CLK_MAP(MCLK, PPCLK_UCLK),
184 CLK_MAP(DCLK, PPCLK_DCLK),
185 CLK_MAP(VCLK, PPCLK_VCLK),
186 CLK_MAP(LCLK, PPCLK_LCLK),
189 static const struct cmn2asic_mapping smu_v13_0_6_feature_mask_map[SMU_FEATURE_COUNT] = {
190 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DATA_CALCULATIONS_BIT, FEATURE_DATA_CALCULATION),
191 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DPM_GFXCLK_BIT, FEATURE_DPM_GFXCLK),
192 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DPM_UCLK_BIT, FEATURE_DPM_UCLK),
193 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DPM_SOCCLK_BIT, FEATURE_DPM_SOCCLK),
194 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DPM_FCLK_BIT, FEATURE_DPM_FCLK),
195 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DPM_LCLK_BIT, FEATURE_DPM_LCLK),
196 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DPM_VCLK_BIT, FEATURE_DPM_VCN),
197 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DPM_DCLK_BIT, FEATURE_DPM_VCN),
198 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DPM_XGMI_BIT, FEATURE_DPM_XGMI),
199 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DS_GFXCLK_BIT, FEATURE_DS_GFXCLK),
200 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DS_SOCCLK_BIT, FEATURE_DS_SOCCLK),
201 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DS_LCLK_BIT, FEATURE_DS_LCLK),
202 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DS_FCLK_BIT, FEATURE_DS_FCLK),
203 SMU_13_0_6_FEA_MAP(SMU_FEATURE_VCN_DPM_BIT, FEATURE_DPM_VCN),
204 SMU_13_0_6_FEA_MAP(SMU_FEATURE_PPT_BIT, FEATURE_PPT),
205 SMU_13_0_6_FEA_MAP(SMU_FEATURE_TDC_BIT, FEATURE_TDC),
206 SMU_13_0_6_FEA_MAP(SMU_FEATURE_APCC_DFLL_BIT, FEATURE_APCC_DFLL),
207 SMU_13_0_6_FEA_MAP(SMU_FEATURE_MP1_CG_BIT, FEATURE_SMU_CG),
208 SMU_13_0_6_FEA_MAP(SMU_FEATURE_GFXOFF_BIT, FEATURE_GFXOFF),
209 SMU_13_0_6_FEA_MAP(SMU_FEATURE_FW_CTF_BIT, FEATURE_FW_CTF),
210 SMU_13_0_6_FEA_MAP(SMU_FEATURE_THERMAL_BIT, FEATURE_THERMAL),
211 SMU_13_0_6_FEA_MAP(SMU_FEATURE_XGMI_PER_LINK_PWR_DWN_BIT, FEATURE_XGMI_PER_LINK_PWR_DOWN),
212 SMU_13_0_6_FEA_MAP(SMU_FEATURE_DF_CSTATE_BIT, FEATURE_DF_CSTATE),
215 #define TABLE_PMSTATUSLOG 0
216 #define TABLE_SMU_METRICS 1
217 #define TABLE_I2C_COMMANDS 2
218 #define TABLE_COUNT 3
220 static const struct cmn2asic_mapping smu_v13_0_6_table_map[SMU_TABLE_COUNT] = {
221 TAB_MAP(PMSTATUSLOG),
222 TAB_MAP(SMU_METRICS),
223 TAB_MAP(I2C_COMMANDS),
226 static const uint8_t smu_v13_0_6_throttler_map[] = {
227 [THROTTLER_PPT_BIT] = (SMU_THROTTLER_PPT0_BIT),
228 [THROTTLER_THERMAL_SOCKET_BIT] = (SMU_THROTTLER_TEMP_GPU_BIT),
229 [THROTTLER_THERMAL_HBM_BIT] = (SMU_THROTTLER_TEMP_MEM_BIT),
230 [THROTTLER_THERMAL_VR_BIT] = (SMU_THROTTLER_TEMP_VR_GFX_BIT),
231 [THROTTLER_PROCHOT_BIT] = (SMU_THROTTLER_PROCHOT_GFX_BIT),
235 uint32_t MaxSocketPowerLimit;
236 uint32_t MaxGfxclkFrequency;
237 uint32_t MinGfxclkFrequency;
238 uint32_t FclkFrequencyTable[4];
239 uint32_t UclkFrequencyTable[4];
240 uint32_t SocclkFrequencyTable[4];
241 uint32_t VclkFrequencyTable[4];
242 uint32_t DclkFrequencyTable[4];
243 uint32_t LclkFrequencyTable[4];
244 uint32_t MaxLclkDpmRange;
245 uint32_t MinLclkDpmRange;
246 uint64_t PublicSerialNumber_AID;
250 #define SMUQ10_TO_UINT(x) ((x) >> 10)
251 #define SMUQ10_FRAC(x) ((x) & 0x3ff)
252 #define SMUQ10_ROUND(x) ((SMUQ10_TO_UINT(x)) + ((SMUQ10_FRAC(x)) >= 0x200))
253 #define GET_METRIC_FIELD(field) ((adev->flags & AMD_IS_APU) ?\
254 (metrics_a->field) : (metrics_x->field))
256 struct smu_v13_0_6_dpm_map {
257 enum smu_clk_type clk_type;
258 uint32_t feature_num;
259 struct smu_13_0_dpm_table *dpm_table;
260 uint32_t *freq_table;
263 static int smu_v13_0_6_init_microcode(struct smu_context *smu)
265 const struct smc_firmware_header_v2_1 *v2_1;
266 const struct common_firmware_header *hdr;
267 struct amdgpu_firmware_info *ucode = NULL;
268 struct smc_soft_pptable_entry *entries;
269 struct amdgpu_device *adev = smu->adev;
270 uint32_t p2s_table_id = P2S_TABLE_ID_A;
271 int ret = 0, i, p2stable_count;
272 char ucode_prefix[15];
275 /* No need to load P2S tables in IOV mode */
276 if (amdgpu_sriov_vf(adev))
279 if (!(adev->flags & AMD_IS_APU))
280 p2s_table_id = P2S_TABLE_ID_X;
282 amdgpu_ucode_ip_version_decode(adev, MP1_HWIP, ucode_prefix,
283 sizeof(ucode_prefix));
285 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s.bin", ucode_prefix);
287 ret = amdgpu_ucode_request(adev, &adev->pm.fw, fw_name);
291 hdr = (const struct common_firmware_header *)adev->pm.fw->data;
292 amdgpu_ucode_print_smc_hdr(hdr);
294 /* SMU v13.0.6 binary file doesn't carry pptables, instead the entries
295 * are used to carry p2s tables.
297 v2_1 = (const struct smc_firmware_header_v2_1 *)adev->pm.fw->data;
298 entries = (struct smc_soft_pptable_entry
300 le32_to_cpu(v2_1->pptable_entry_offset));
301 p2stable_count = le32_to_cpu(v2_1->pptable_count);
302 for (i = 0; i < p2stable_count; i++) {
303 if (le32_to_cpu(entries[i].id) == p2s_table_id) {
304 smu->pptable_firmware.data =
306 le32_to_cpu(entries[i].ppt_offset_bytes));
307 smu->pptable_firmware.size =
308 le32_to_cpu(entries[i].ppt_size_bytes);
313 if (smu->pptable_firmware.data && smu->pptable_firmware.size) {
314 ucode = &adev->firmware.ucode[AMDGPU_UCODE_ID_P2S_TABLE];
315 ucode->ucode_id = AMDGPU_UCODE_ID_P2S_TABLE;
316 ucode->fw = &smu->pptable_firmware;
317 adev->firmware.fw_size += ALIGN(ucode->fw->size, PAGE_SIZE);
322 amdgpu_ucode_release(&adev->pm.fw);
327 static int smu_v13_0_6_tables_init(struct smu_context *smu)
329 struct smu_table_context *smu_table = &smu->smu_table;
330 struct smu_table *tables = smu_table->tables;
331 struct amdgpu_device *adev = smu->adev;
333 if (!(adev->flags & AMD_IS_APU))
334 SMU_TABLE_INIT(tables, SMU_TABLE_PMSTATUSLOG, SMU13_TOOL_SIZE,
335 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM);
337 SMU_TABLE_INIT(tables, SMU_TABLE_SMU_METRICS,
338 max(sizeof(MetricsTableX_t), sizeof(MetricsTableA_t)),
340 AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT);
342 SMU_TABLE_INIT(tables, SMU_TABLE_I2C_COMMANDS, sizeof(SwI2cRequest_t),
344 AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT);
346 smu_table->metrics_table = kzalloc(max(sizeof(MetricsTableX_t),
347 sizeof(MetricsTableA_t)), GFP_KERNEL);
348 if (!smu_table->metrics_table)
350 smu_table->metrics_time = 0;
352 smu_table->gpu_metrics_table_size = sizeof(struct gpu_metrics_v1_5);
353 smu_table->gpu_metrics_table =
354 kzalloc(smu_table->gpu_metrics_table_size, GFP_KERNEL);
355 if (!smu_table->gpu_metrics_table) {
356 kfree(smu_table->metrics_table);
360 smu_table->driver_pptable =
361 kzalloc(sizeof(struct PPTable_t), GFP_KERNEL);
362 if (!smu_table->driver_pptable) {
363 kfree(smu_table->metrics_table);
364 kfree(smu_table->gpu_metrics_table);
371 static int smu_v13_0_6_allocate_dpm_context(struct smu_context *smu)
373 struct smu_dpm_context *smu_dpm = &smu->smu_dpm;
375 smu_dpm->dpm_context =
376 kzalloc(sizeof(struct smu_13_0_dpm_context), GFP_KERNEL);
377 if (!smu_dpm->dpm_context)
379 smu_dpm->dpm_context_size = sizeof(struct smu_13_0_dpm_context);
384 static int smu_v13_0_6_init_smc_tables(struct smu_context *smu)
388 ret = smu_v13_0_6_tables_init(smu);
392 ret = smu_v13_0_6_allocate_dpm_context(smu);
397 static int smu_v13_0_6_get_allowed_feature_mask(struct smu_context *smu,
398 uint32_t *feature_mask,
404 /* pptable will handle the features to enable */
405 memset(feature_mask, 0xFF, sizeof(uint32_t) * num);
410 static int smu_v13_0_6_get_metrics_table(struct smu_context *smu,
411 void *metrics_table, bool bypass_cache)
413 struct smu_table_context *smu_table = &smu->smu_table;
414 uint32_t table_size = smu_table->tables[SMU_TABLE_SMU_METRICS].size;
415 struct smu_table *table = &smu_table->driver_table;
418 if (bypass_cache || !smu_table->metrics_time ||
420 smu_table->metrics_time + msecs_to_jiffies(1))) {
421 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_GetMetricsTable, NULL);
423 dev_info(smu->adev->dev,
424 "Failed to export SMU metrics table!\n");
428 amdgpu_asic_invalidate_hdp(smu->adev, NULL);
429 memcpy(smu_table->metrics_table, table->cpu_addr, table_size);
431 smu_table->metrics_time = jiffies;
435 memcpy(metrics_table, smu_table->metrics_table, table_size);
440 static ssize_t smu_v13_0_6_get_pm_metrics(struct smu_context *smu,
441 void *metrics, size_t max_size)
443 struct smu_table_context *smu_tbl_ctxt = &smu->smu_table;
444 uint32_t table_version = smu_tbl_ctxt->tables[SMU_TABLE_SMU_METRICS].version;
445 uint32_t table_size = smu_tbl_ctxt->tables[SMU_TABLE_SMU_METRICS].size;
446 struct amdgpu_pm_metrics *pm_metrics = metrics;
447 uint32_t pmfw_version;
450 if (!pm_metrics || !max_size)
453 if (max_size < (table_size + sizeof(pm_metrics->common_header)))
456 /* Don't use cached metrics data */
457 ret = smu_v13_0_6_get_metrics_table(smu, pm_metrics->data, true);
461 smu_cmn_get_smc_version(smu, NULL, &pmfw_version);
463 memset(&pm_metrics->common_header, 0,
464 sizeof(pm_metrics->common_header));
465 pm_metrics->common_header.mp1_ip_discovery_version =
466 IP_VERSION(13, 0, 6);
467 pm_metrics->common_header.pmfw_version = pmfw_version;
468 pm_metrics->common_header.pmmetrics_version = table_version;
469 pm_metrics->common_header.structure_size =
470 sizeof(pm_metrics->common_header) + table_size;
472 return pm_metrics->common_header.structure_size;
475 static int smu_v13_0_6_setup_driver_pptable(struct smu_context *smu)
477 struct smu_table_context *smu_table = &smu->smu_table;
478 MetricsTableX_t *metrics_x = (MetricsTableX_t *)smu_table->metrics_table;
479 MetricsTableA_t *metrics_a = (MetricsTableA_t *)smu_table->metrics_table;
480 struct PPTable_t *pptable =
481 (struct PPTable_t *)smu_table->driver_pptable;
482 struct amdgpu_device *adev = smu->adev;
483 int ret, i, retry = 100;
484 uint32_t table_version;
486 /* Store one-time values in driver PPTable */
487 if (!pptable->Init) {
489 ret = smu_v13_0_6_get_metrics_table(smu, NULL, true);
493 /* Ensure that metrics have been updated */
494 if (GET_METRIC_FIELD(AccumulationCounter))
497 usleep_range(1000, 1100);
503 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_GetMetricsVersion,
507 smu_table->tables[SMU_TABLE_SMU_METRICS].version =
510 pptable->MaxSocketPowerLimit =
511 SMUQ10_ROUND(GET_METRIC_FIELD(MaxSocketPowerLimit));
512 pptable->MaxGfxclkFrequency =
513 SMUQ10_ROUND(GET_METRIC_FIELD(MaxGfxclkFrequency));
514 pptable->MinGfxclkFrequency =
515 SMUQ10_ROUND(GET_METRIC_FIELD(MinGfxclkFrequency));
517 for (i = 0; i < 4; ++i) {
518 pptable->FclkFrequencyTable[i] =
519 SMUQ10_ROUND(GET_METRIC_FIELD(FclkFrequencyTable)[i]);
520 pptable->UclkFrequencyTable[i] =
521 SMUQ10_ROUND(GET_METRIC_FIELD(UclkFrequencyTable)[i]);
522 pptable->SocclkFrequencyTable[i] = SMUQ10_ROUND(
523 GET_METRIC_FIELD(SocclkFrequencyTable)[i]);
524 pptable->VclkFrequencyTable[i] =
525 SMUQ10_ROUND(GET_METRIC_FIELD(VclkFrequencyTable)[i]);
526 pptable->DclkFrequencyTable[i] =
527 SMUQ10_ROUND(GET_METRIC_FIELD(DclkFrequencyTable)[i]);
528 pptable->LclkFrequencyTable[i] =
529 SMUQ10_ROUND(GET_METRIC_FIELD(LclkFrequencyTable)[i]);
532 /* use AID0 serial number by default */
533 pptable->PublicSerialNumber_AID = GET_METRIC_FIELD(PublicSerialNumber_AID)[0];
535 pptable->Init = true;
541 static int smu_v13_0_6_get_dpm_ultimate_freq(struct smu_context *smu,
542 enum smu_clk_type clk_type,
543 uint32_t *min, uint32_t *max)
545 struct smu_table_context *smu_table = &smu->smu_table;
546 struct PPTable_t *pptable =
547 (struct PPTable_t *)smu_table->driver_pptable;
548 uint32_t clock_limit = 0, param;
549 int ret = 0, clk_id = 0;
551 if (!smu_cmn_clk_dpm_is_enabled(smu, clk_type)) {
556 clock_limit = pptable->UclkFrequencyTable[0];
561 clock_limit = pptable->MinGfxclkFrequency;
565 clock_limit = pptable->SocclkFrequencyTable[0];
569 clock_limit = pptable->FclkFrequencyTable[0];
573 clock_limit = pptable->VclkFrequencyTable[0];
577 clock_limit = pptable->DclkFrequencyTable[0];
592 if (!(clk_type == SMU_GFXCLK || clk_type == SMU_SCLK)) {
593 clk_id = smu_cmn_to_asic_specific_index(
594 smu, CMN2ASIC_MAPPING_CLK, clk_type);
599 param = (clk_id & 0xffff) << 16;
603 if (clk_type == SMU_GFXCLK || clk_type == SMU_SCLK)
604 ret = smu_cmn_send_smc_msg(
605 smu, SMU_MSG_GetMaxGfxclkFrequency, max);
607 ret = smu_cmn_send_smc_msg_with_param(
608 smu, SMU_MSG_GetMaxDpmFreq, param, max);
614 if (clk_type == SMU_GFXCLK || clk_type == SMU_SCLK)
615 ret = smu_cmn_send_smc_msg(
616 smu, SMU_MSG_GetMinGfxclkFrequency, min);
618 ret = smu_cmn_send_smc_msg_with_param(
619 smu, SMU_MSG_GetMinDpmFreq, param, min);
626 static int smu_v13_0_6_get_dpm_level_count(struct smu_context *smu,
627 enum smu_clk_type clk_type,
632 ret = smu_v13_0_get_dpm_freq_by_index(smu, clk_type, 0xff, levels);
639 static int smu_v13_0_6_set_default_dpm_table(struct smu_context *smu)
641 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context;
642 struct smu_table_context *smu_table = &smu->smu_table;
643 struct smu_13_0_dpm_table *dpm_table = NULL;
644 struct PPTable_t *pptable =
645 (struct PPTable_t *)smu_table->driver_pptable;
646 uint32_t gfxclkmin, gfxclkmax, levels;
648 struct smu_v13_0_6_dpm_map dpm_map[] = {
649 { SMU_SOCCLK, SMU_FEATURE_DPM_SOCCLK_BIT,
650 &dpm_context->dpm_tables.soc_table,
651 pptable->SocclkFrequencyTable },
652 { SMU_UCLK, SMU_FEATURE_DPM_UCLK_BIT,
653 &dpm_context->dpm_tables.uclk_table,
654 pptable->UclkFrequencyTable },
655 { SMU_FCLK, SMU_FEATURE_DPM_FCLK_BIT,
656 &dpm_context->dpm_tables.fclk_table,
657 pptable->FclkFrequencyTable },
658 { SMU_VCLK, SMU_FEATURE_DPM_VCLK_BIT,
659 &dpm_context->dpm_tables.vclk_table,
660 pptable->VclkFrequencyTable },
661 { SMU_DCLK, SMU_FEATURE_DPM_DCLK_BIT,
662 &dpm_context->dpm_tables.dclk_table,
663 pptable->DclkFrequencyTable },
666 smu_v13_0_6_setup_driver_pptable(smu);
668 /* gfxclk dpm table setup */
669 dpm_table = &dpm_context->dpm_tables.gfx_table;
670 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT)) {
671 /* In the case of gfxclk, only fine-grained dpm is honored.
672 * Get min/max values from FW.
674 ret = smu_v13_0_6_get_dpm_ultimate_freq(smu, SMU_GFXCLK,
675 &gfxclkmin, &gfxclkmax);
679 dpm_table->count = 2;
680 dpm_table->dpm_levels[0].value = gfxclkmin;
681 dpm_table->dpm_levels[0].enabled = true;
682 dpm_table->dpm_levels[1].value = gfxclkmax;
683 dpm_table->dpm_levels[1].enabled = true;
684 dpm_table->min = dpm_table->dpm_levels[0].value;
685 dpm_table->max = dpm_table->dpm_levels[1].value;
687 dpm_table->count = 1;
688 dpm_table->dpm_levels[0].value = pptable->MinGfxclkFrequency;
689 dpm_table->dpm_levels[0].enabled = true;
690 dpm_table->min = dpm_table->dpm_levels[0].value;
691 dpm_table->max = dpm_table->dpm_levels[0].value;
694 for (j = 0; j < ARRAY_SIZE(dpm_map); j++) {
695 dpm_table = dpm_map[j].dpm_table;
697 if (smu_cmn_feature_is_enabled(smu, dpm_map[j].feature_num)) {
698 ret = smu_v13_0_6_get_dpm_level_count(
699 smu, dpm_map[j].clk_type, &levels);
703 dpm_table->count = levels;
704 for (i = 0; i < dpm_table->count; ++i) {
705 dpm_table->dpm_levels[i].value =
706 dpm_map[j].freq_table[i];
707 dpm_table->dpm_levels[i].enabled = true;
710 dpm_table->min = dpm_table->dpm_levels[0].value;
711 dpm_table->max = dpm_table->dpm_levels[levels - 1].value;
718 static int smu_v13_0_6_setup_pptable(struct smu_context *smu)
720 struct smu_table_context *table_context = &smu->smu_table;
722 /* TODO: PPTable is not available.
723 * 1) Find an alternate way to get 'PPTable values' here.
724 * 2) Check if there is SW CTF
726 table_context->thermal_controller_type = 0;
731 static int smu_v13_0_6_check_fw_status(struct smu_context *smu)
733 struct amdgpu_device *adev = smu->adev;
734 uint32_t mp1_fw_flags;
737 RREG32_PCIE(MP1_Public | (smnMP1_FIRMWARE_FLAGS & 0xffffffff));
739 if ((mp1_fw_flags & MP1_FIRMWARE_FLAGS__INTERRUPTS_ENABLED_MASK) >>
740 MP1_FIRMWARE_FLAGS__INTERRUPTS_ENABLED__SHIFT)
746 static int smu_v13_0_6_populate_umd_state_clk(struct smu_context *smu)
748 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context;
749 struct smu_13_0_dpm_table *gfx_table =
750 &dpm_context->dpm_tables.gfx_table;
751 struct smu_13_0_dpm_table *mem_table =
752 &dpm_context->dpm_tables.uclk_table;
753 struct smu_13_0_dpm_table *soc_table =
754 &dpm_context->dpm_tables.soc_table;
755 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table;
757 pstate_table->gfxclk_pstate.min = gfx_table->min;
758 pstate_table->gfxclk_pstate.peak = gfx_table->max;
759 pstate_table->gfxclk_pstate.curr.min = gfx_table->min;
760 pstate_table->gfxclk_pstate.curr.max = gfx_table->max;
762 pstate_table->uclk_pstate.min = mem_table->min;
763 pstate_table->uclk_pstate.peak = mem_table->max;
764 pstate_table->uclk_pstate.curr.min = mem_table->min;
765 pstate_table->uclk_pstate.curr.max = mem_table->max;
767 pstate_table->socclk_pstate.min = soc_table->min;
768 pstate_table->socclk_pstate.peak = soc_table->max;
769 pstate_table->socclk_pstate.curr.min = soc_table->min;
770 pstate_table->socclk_pstate.curr.max = soc_table->max;
772 if (gfx_table->count > SMU_13_0_6_UMD_PSTATE_GFXCLK_LEVEL &&
773 mem_table->count > SMU_13_0_6_UMD_PSTATE_MCLK_LEVEL &&
774 soc_table->count > SMU_13_0_6_UMD_PSTATE_SOCCLK_LEVEL) {
775 pstate_table->gfxclk_pstate.standard =
776 gfx_table->dpm_levels[SMU_13_0_6_UMD_PSTATE_GFXCLK_LEVEL].value;
777 pstate_table->uclk_pstate.standard =
778 mem_table->dpm_levels[SMU_13_0_6_UMD_PSTATE_MCLK_LEVEL].value;
779 pstate_table->socclk_pstate.standard =
780 soc_table->dpm_levels[SMU_13_0_6_UMD_PSTATE_SOCCLK_LEVEL].value;
782 pstate_table->gfxclk_pstate.standard =
783 pstate_table->gfxclk_pstate.min;
784 pstate_table->uclk_pstate.standard =
785 pstate_table->uclk_pstate.min;
786 pstate_table->socclk_pstate.standard =
787 pstate_table->socclk_pstate.min;
793 static int smu_v13_0_6_get_clk_table(struct smu_context *smu,
794 struct pp_clock_levels_with_latency *clocks,
795 struct smu_13_0_dpm_table *dpm_table)
799 count = (dpm_table->count > MAX_NUM_CLOCKS) ? MAX_NUM_CLOCKS :
801 clocks->num_levels = count;
803 for (i = 0; i < count; i++) {
804 clocks->data[i].clocks_in_khz =
805 dpm_table->dpm_levels[i].value * 1000;
806 clocks->data[i].latency_in_us = 0;
812 static int smu_v13_0_6_freqs_in_same_level(int32_t frequency1,
815 return (abs(frequency1 - frequency2) <= EPSILON);
818 static uint32_t smu_v13_0_6_get_throttler_status(struct smu_context *smu)
820 struct smu_power_context *smu_power = &smu->smu_power;
821 struct smu_13_0_power_context *power_context = smu_power->power_context;
822 uint32_t throttler_status = 0;
824 throttler_status = atomic_read(&power_context->throttle_status);
825 dev_dbg(smu->adev->dev, "SMU Throttler status: %u", throttler_status);
827 return throttler_status;
830 static int smu_v13_0_6_get_smu_metrics_data(struct smu_context *smu,
831 MetricsMember_t member,
834 struct smu_table_context *smu_table = &smu->smu_table;
835 MetricsTableX_t *metrics_x = (MetricsTableX_t *)smu_table->metrics_table;
836 MetricsTableA_t *metrics_a = (MetricsTableA_t *)smu_table->metrics_table;
837 struct amdgpu_device *adev = smu->adev;
841 ret = smu_v13_0_6_get_metrics_table(smu, NULL, false);
845 /* For clocks with multiple instances, only report the first one */
847 case METRICS_CURR_GFXCLK:
848 case METRICS_AVERAGE_GFXCLK:
849 if (smu->smc_fw_version >= 0x552F00) {
850 xcc_id = GET_INST(GC, 0);
851 *value = SMUQ10_ROUND(GET_METRIC_FIELD(GfxclkFrequency)[xcc_id]);
856 case METRICS_CURR_SOCCLK:
857 case METRICS_AVERAGE_SOCCLK:
858 *value = SMUQ10_ROUND(GET_METRIC_FIELD(SocclkFrequency)[0]);
860 case METRICS_CURR_UCLK:
861 case METRICS_AVERAGE_UCLK:
862 *value = SMUQ10_ROUND(GET_METRIC_FIELD(UclkFrequency));
864 case METRICS_CURR_VCLK:
865 *value = SMUQ10_ROUND(GET_METRIC_FIELD(VclkFrequency)[0]);
867 case METRICS_CURR_DCLK:
868 *value = SMUQ10_ROUND(GET_METRIC_FIELD(DclkFrequency)[0]);
870 case METRICS_CURR_FCLK:
871 *value = SMUQ10_ROUND(GET_METRIC_FIELD(FclkFrequency));
873 case METRICS_AVERAGE_GFXACTIVITY:
874 *value = SMUQ10_ROUND(GET_METRIC_FIELD(SocketGfxBusy));
876 case METRICS_AVERAGE_MEMACTIVITY:
877 *value = SMUQ10_ROUND(GET_METRIC_FIELD(DramBandwidthUtilization));
879 case METRICS_CURR_SOCKETPOWER:
880 *value = SMUQ10_ROUND(GET_METRIC_FIELD(SocketPower)) << 8;
882 case METRICS_TEMPERATURE_HOTSPOT:
883 *value = SMUQ10_ROUND(GET_METRIC_FIELD(MaxSocketTemperature)) *
884 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES;
886 case METRICS_TEMPERATURE_MEM:
887 *value = SMUQ10_ROUND(GET_METRIC_FIELD(MaxHbmTemperature)) *
888 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES;
890 /* This is the max of all VRs and not just SOC VR.
891 * No need to define another data type for the same.
893 case METRICS_TEMPERATURE_VRSOC:
894 *value = SMUQ10_ROUND(GET_METRIC_FIELD(MaxVrTemperature)) *
895 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES;
905 static int smu_v13_0_6_get_current_clk_freq_by_table(struct smu_context *smu,
906 enum smu_clk_type clk_type,
909 MetricsMember_t member_type;
916 member_type = METRICS_CURR_GFXCLK;
919 member_type = METRICS_CURR_UCLK;
922 member_type = METRICS_CURR_SOCCLK;
925 member_type = METRICS_CURR_VCLK;
928 member_type = METRICS_CURR_DCLK;
931 member_type = METRICS_CURR_FCLK;
937 return smu_v13_0_6_get_smu_metrics_data(smu, member_type, value);
940 static int smu_v13_0_6_print_clks(struct smu_context *smu, char *buf, int size,
941 struct smu_13_0_dpm_table *single_dpm_table,
942 uint32_t curr_clk, const char *clk_name)
944 struct pp_clock_levels_with_latency clocks;
945 int i, ret, level = -1;
948 ret = smu_v13_0_6_get_clk_table(smu, &clocks, single_dpm_table);
950 dev_err(smu->adev->dev, "Attempt to get %s clk levels failed!",
955 if (!clocks.num_levels)
958 if (curr_clk < SMU_13_0_6_DSCLK_THRESHOLD) {
959 size = sysfs_emit_at(buf, size, "S: %uMhz *\n", curr_clk);
960 for (i = 0; i < clocks.num_levels; i++)
961 size += sysfs_emit_at(buf, size, "%d: %uMhz\n", i,
962 clocks.data[i].clocks_in_khz /
966 if ((clocks.num_levels == 1) ||
967 (curr_clk < (clocks.data[0].clocks_in_khz / 1000)))
969 for (i = 0; i < clocks.num_levels; i++) {
970 clk1 = clocks.data[i].clocks_in_khz / 1000;
972 if (i < (clocks.num_levels - 1))
973 clk2 = clocks.data[i + 1].clocks_in_khz / 1000;
975 if (curr_clk == clk1) {
977 } else if (curr_clk >= clk1 && curr_clk < clk2) {
978 level = (curr_clk - clk1) <= (clk2 - curr_clk) ?
983 size += sysfs_emit_at(buf, size, "%d: %uMhz %s\n", i,
984 clk1, (level == i) ? "*" : "");
991 static int smu_v13_0_6_print_clk_levels(struct smu_context *smu,
992 enum smu_clk_type type, char *buf)
996 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table;
997 struct smu_13_0_dpm_table *single_dpm_table;
998 struct smu_dpm_context *smu_dpm = &smu->smu_dpm;
999 struct smu_13_0_dpm_context *dpm_context = NULL;
1000 uint32_t min_clk, max_clk;
1002 smu_cmn_get_sysfs_buf(&buf, &size);
1004 if (amdgpu_ras_intr_triggered()) {
1005 size += sysfs_emit_at(buf, size, "unavailable\n");
1009 dpm_context = smu_dpm->dpm_context;
1013 size += sysfs_emit_at(buf, size, "%s:\n", "GFXCLK");
1016 ret = smu_v13_0_6_get_current_clk_freq_by_table(smu, SMU_GFXCLK,
1019 dev_err(smu->adev->dev,
1020 "Attempt to get current gfx clk Failed!");
1024 min_clk = pstate_table->gfxclk_pstate.curr.min;
1025 max_clk = pstate_table->gfxclk_pstate.curr.max;
1027 if (now < SMU_13_0_6_DSCLK_THRESHOLD) {
1028 size += sysfs_emit_at(buf, size, "S: %uMhz *\n",
1030 size += sysfs_emit_at(buf, size, "0: %uMhz\n",
1032 size += sysfs_emit_at(buf, size, "1: %uMhz\n",
1035 } else if (!smu_v13_0_6_freqs_in_same_level(now, min_clk) &&
1036 !smu_v13_0_6_freqs_in_same_level(now, max_clk)) {
1037 size += sysfs_emit_at(buf, size, "0: %uMhz\n",
1039 size += sysfs_emit_at(buf, size, "1: %uMhz *\n",
1041 size += sysfs_emit_at(buf, size, "2: %uMhz\n",
1044 size += sysfs_emit_at(buf, size, "0: %uMhz %s\n",
1046 smu_v13_0_6_freqs_in_same_level(now, min_clk) ? "*" : "");
1047 size += sysfs_emit_at(buf, size, "1: %uMhz %s\n",
1049 smu_v13_0_6_freqs_in_same_level(now, max_clk) ? "*" : "");
1055 size += sysfs_emit_at(buf, size, "%s:\n", "MCLK");
1058 ret = smu_v13_0_6_get_current_clk_freq_by_table(smu, SMU_UCLK,
1061 dev_err(smu->adev->dev,
1062 "Attempt to get current mclk Failed!");
1066 single_dpm_table = &(dpm_context->dpm_tables.uclk_table);
1068 return smu_v13_0_6_print_clks(smu, buf, size, single_dpm_table,
1072 ret = smu_v13_0_6_get_current_clk_freq_by_table(smu, SMU_SOCCLK,
1075 dev_err(smu->adev->dev,
1076 "Attempt to get current socclk Failed!");
1080 single_dpm_table = &(dpm_context->dpm_tables.soc_table);
1082 return smu_v13_0_6_print_clks(smu, buf, size, single_dpm_table,
1086 ret = smu_v13_0_6_get_current_clk_freq_by_table(smu, SMU_FCLK,
1089 dev_err(smu->adev->dev,
1090 "Attempt to get current fclk Failed!");
1094 single_dpm_table = &(dpm_context->dpm_tables.fclk_table);
1096 return smu_v13_0_6_print_clks(smu, buf, size, single_dpm_table,
1100 ret = smu_v13_0_6_get_current_clk_freq_by_table(smu, SMU_VCLK,
1103 dev_err(smu->adev->dev,
1104 "Attempt to get current vclk Failed!");
1108 single_dpm_table = &(dpm_context->dpm_tables.vclk_table);
1110 return smu_v13_0_6_print_clks(smu, buf, size, single_dpm_table,
1114 ret = smu_v13_0_6_get_current_clk_freq_by_table(smu, SMU_DCLK,
1117 dev_err(smu->adev->dev,
1118 "Attempt to get current dclk Failed!");
1122 single_dpm_table = &(dpm_context->dpm_tables.dclk_table);
1124 return smu_v13_0_6_print_clks(smu, buf, size, single_dpm_table,
1134 static int smu_v13_0_6_upload_dpm_level(struct smu_context *smu, bool max,
1135 uint32_t feature_mask, uint32_t level)
1137 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context;
1141 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_GFXCLK_BIT) &&
1142 (feature_mask & FEATURE_MASK(FEATURE_DPM_GFXCLK))) {
1143 freq = dpm_context->dpm_tables.gfx_table.dpm_levels[level].value;
1144 ret = smu_cmn_send_smc_msg_with_param(
1146 (max ? SMU_MSG_SetSoftMaxGfxClk :
1147 SMU_MSG_SetSoftMinGfxclk),
1148 freq & 0xffff, NULL);
1150 dev_err(smu->adev->dev,
1151 "Failed to set soft %s gfxclk !\n",
1152 max ? "max" : "min");
1157 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_UCLK_BIT) &&
1158 (feature_mask & FEATURE_MASK(FEATURE_DPM_UCLK))) {
1159 freq = dpm_context->dpm_tables.uclk_table.dpm_levels[level]
1161 ret = smu_cmn_send_smc_msg_with_param(
1163 (max ? SMU_MSG_SetSoftMaxByFreq :
1164 SMU_MSG_SetSoftMinByFreq),
1165 (PPCLK_UCLK << 16) | (freq & 0xffff), NULL);
1167 dev_err(smu->adev->dev,
1168 "Failed to set soft %s memclk !\n",
1169 max ? "max" : "min");
1174 if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_DPM_SOCCLK_BIT) &&
1175 (feature_mask & FEATURE_MASK(FEATURE_DPM_SOCCLK))) {
1176 freq = dpm_context->dpm_tables.soc_table.dpm_levels[level].value;
1177 ret = smu_cmn_send_smc_msg_with_param(
1179 (max ? SMU_MSG_SetSoftMaxByFreq :
1180 SMU_MSG_SetSoftMinByFreq),
1181 (PPCLK_SOCCLK << 16) | (freq & 0xffff), NULL);
1183 dev_err(smu->adev->dev,
1184 "Failed to set soft %s socclk !\n",
1185 max ? "max" : "min");
1193 static int smu_v13_0_6_force_clk_levels(struct smu_context *smu,
1194 enum smu_clk_type type, uint32_t mask)
1196 struct smu_13_0_dpm_context *dpm_context = smu->smu_dpm.dpm_context;
1197 struct smu_13_0_dpm_table *single_dpm_table = NULL;
1198 uint32_t soft_min_level, soft_max_level;
1201 soft_min_level = mask ? (ffs(mask) - 1) : 0;
1202 soft_max_level = mask ? (fls(mask) - 1) : 0;
1206 single_dpm_table = &(dpm_context->dpm_tables.gfx_table);
1207 if (soft_max_level >= single_dpm_table->count) {
1208 dev_err(smu->adev->dev,
1209 "Clock level specified %d is over max allowed %d\n",
1210 soft_max_level, single_dpm_table->count - 1);
1215 ret = smu_v13_0_6_upload_dpm_level(
1216 smu, false, FEATURE_MASK(FEATURE_DPM_GFXCLK),
1219 dev_err(smu->adev->dev,
1220 "Failed to upload boot level to lowest!\n");
1224 ret = smu_v13_0_6_upload_dpm_level(
1225 smu, true, FEATURE_MASK(FEATURE_DPM_GFXCLK),
1228 dev_err(smu->adev->dev,
1229 "Failed to upload dpm max level to highest!\n");
1237 * Should not arrive here since smu_13_0_6 does not
1238 * support mclk/socclk/fclk softmin/softmax settings
1250 static int smu_v13_0_6_get_current_activity_percent(struct smu_context *smu,
1251 enum amd_pp_sensors sensor,
1260 case AMDGPU_PP_SENSOR_GPU_LOAD:
1261 ret = smu_v13_0_6_get_smu_metrics_data(
1262 smu, METRICS_AVERAGE_GFXACTIVITY, value);
1264 case AMDGPU_PP_SENSOR_MEM_LOAD:
1265 ret = smu_v13_0_6_get_smu_metrics_data(
1266 smu, METRICS_AVERAGE_MEMACTIVITY, value);
1269 dev_err(smu->adev->dev,
1270 "Invalid sensor for retrieving clock activity\n");
1277 static int smu_v13_0_6_thermal_get_temperature(struct smu_context *smu,
1278 enum amd_pp_sensors sensor,
1287 case AMDGPU_PP_SENSOR_HOTSPOT_TEMP:
1288 ret = smu_v13_0_6_get_smu_metrics_data(
1289 smu, METRICS_TEMPERATURE_HOTSPOT, value);
1291 case AMDGPU_PP_SENSOR_MEM_TEMP:
1292 ret = smu_v13_0_6_get_smu_metrics_data(
1293 smu, METRICS_TEMPERATURE_MEM, value);
1296 dev_err(smu->adev->dev, "Invalid sensor for retrieving temp\n");
1303 static int smu_v13_0_6_read_sensor(struct smu_context *smu,
1304 enum amd_pp_sensors sensor, void *data,
1309 if (amdgpu_ras_intr_triggered())
1316 case AMDGPU_PP_SENSOR_MEM_LOAD:
1317 case AMDGPU_PP_SENSOR_GPU_LOAD:
1318 ret = smu_v13_0_6_get_current_activity_percent(smu, sensor,
1322 case AMDGPU_PP_SENSOR_GPU_INPUT_POWER:
1323 ret = smu_v13_0_6_get_smu_metrics_data(smu,
1324 METRICS_CURR_SOCKETPOWER,
1328 case AMDGPU_PP_SENSOR_HOTSPOT_TEMP:
1329 case AMDGPU_PP_SENSOR_MEM_TEMP:
1330 ret = smu_v13_0_6_thermal_get_temperature(smu, sensor,
1334 case AMDGPU_PP_SENSOR_GFX_MCLK:
1335 ret = smu_v13_0_6_get_current_clk_freq_by_table(
1336 smu, SMU_UCLK, (uint32_t *)data);
1337 /* the output clock frequency in 10K unit */
1338 *(uint32_t *)data *= 100;
1341 case AMDGPU_PP_SENSOR_GFX_SCLK:
1342 ret = smu_v13_0_6_get_current_clk_freq_by_table(
1343 smu, SMU_GFXCLK, (uint32_t *)data);
1344 *(uint32_t *)data *= 100;
1347 case AMDGPU_PP_SENSOR_VDDGFX:
1348 ret = smu_v13_0_get_gfx_vdd(smu, (uint32_t *)data);
1351 case AMDGPU_PP_SENSOR_GPU_AVG_POWER:
1360 static int smu_v13_0_6_get_power_limit(struct smu_context *smu,
1361 uint32_t *current_power_limit,
1362 uint32_t *default_power_limit,
1363 uint32_t *max_power_limit,
1364 uint32_t *min_power_limit)
1366 struct smu_table_context *smu_table = &smu->smu_table;
1367 struct PPTable_t *pptable =
1368 (struct PPTable_t *)smu_table->driver_pptable;
1369 uint32_t power_limit = 0;
1372 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_GetPptLimit, &power_limit);
1375 dev_err(smu->adev->dev, "Couldn't get PPT limit");
1379 if (current_power_limit)
1380 *current_power_limit = power_limit;
1381 if (default_power_limit)
1382 *default_power_limit = power_limit;
1384 if (max_power_limit) {
1385 *max_power_limit = pptable->MaxSocketPowerLimit;
1388 if (min_power_limit)
1389 *min_power_limit = 0;
1393 static int smu_v13_0_6_set_power_limit(struct smu_context *smu,
1394 enum smu_ppt_limit_type limit_type,
1397 return smu_v13_0_set_power_limit(smu, limit_type, limit);
1400 static int smu_v13_0_6_irq_process(struct amdgpu_device *adev,
1401 struct amdgpu_irq_src *source,
1402 struct amdgpu_iv_entry *entry)
1404 struct smu_context *smu = adev->powerplay.pp_handle;
1405 struct smu_power_context *smu_power = &smu->smu_power;
1406 struct smu_13_0_power_context *power_context = smu_power->power_context;
1407 uint32_t client_id = entry->client_id;
1408 uint32_t ctxid = entry->src_data[0];
1409 uint32_t src_id = entry->src_id;
1412 if (client_id == SOC15_IH_CLIENTID_MP1) {
1413 if (src_id == IH_INTERRUPT_ID_TO_DRIVER) {
1414 /* ACK SMUToHost interrupt */
1415 data = RREG32_SOC15(MP1, 0, regMP1_SMN_IH_SW_INT_CTRL);
1416 data = REG_SET_FIELD(data, MP1_SMN_IH_SW_INT_CTRL, INT_ACK, 1);
1417 WREG32_SOC15(MP1, 0, regMP1_SMN_IH_SW_INT_CTRL, data);
1419 * ctxid is used to distinguish different events for SMCToHost
1423 case IH_INTERRUPT_CONTEXT_ID_THERMAL_THROTTLING:
1425 * Increment the throttle interrupt counter
1427 atomic64_inc(&smu->throttle_int_counter);
1429 if (!atomic_read(&adev->throttling_logging_enabled))
1432 /* This uses the new method which fixes the
1433 * incorrect throttling status reporting
1434 * through metrics table. For older FWs,
1435 * it will be ignored.
1437 if (__ratelimit(&adev->throttling_logging_rs)) {
1439 &power_context->throttle_status,
1440 entry->src_data[1]);
1441 schedule_work(&smu->throttling_logging_work);
1445 dev_dbg(adev->dev, "Unhandled context id %d from client:%d!\n",
1455 static int smu_v13_0_6_set_irq_state(struct amdgpu_device *adev,
1456 struct amdgpu_irq_src *source,
1458 enum amdgpu_interrupt_state state)
1463 case AMDGPU_IRQ_STATE_DISABLE:
1464 /* For MP1 SW irqs */
1465 val = RREG32_SOC15(MP1, 0, regMP1_SMN_IH_SW_INT_CTRL);
1466 val = REG_SET_FIELD(val, MP1_SMN_IH_SW_INT_CTRL, INT_MASK, 1);
1467 WREG32_SOC15(MP1, 0, regMP1_SMN_IH_SW_INT_CTRL, val);
1470 case AMDGPU_IRQ_STATE_ENABLE:
1471 /* For MP1 SW irqs */
1472 val = RREG32_SOC15(MP1, 0, regMP1_SMN_IH_SW_INT);
1473 val = REG_SET_FIELD(val, MP1_SMN_IH_SW_INT, ID, 0xFE);
1474 val = REG_SET_FIELD(val, MP1_SMN_IH_SW_INT, VALID, 0);
1475 WREG32_SOC15(MP1, 0, regMP1_SMN_IH_SW_INT, val);
1477 val = RREG32_SOC15(MP1, 0, regMP1_SMN_IH_SW_INT_CTRL);
1478 val = REG_SET_FIELD(val, MP1_SMN_IH_SW_INT_CTRL, INT_MASK, 0);
1479 WREG32_SOC15(MP1, 0, regMP1_SMN_IH_SW_INT_CTRL, val);
1489 static const struct amdgpu_irq_src_funcs smu_v13_0_6_irq_funcs = {
1490 .set = smu_v13_0_6_set_irq_state,
1491 .process = smu_v13_0_6_irq_process,
1494 static int smu_v13_0_6_register_irq_handler(struct smu_context *smu)
1496 struct amdgpu_device *adev = smu->adev;
1497 struct amdgpu_irq_src *irq_src = &smu->irq_source;
1500 if (amdgpu_sriov_vf(adev))
1503 irq_src->num_types = 1;
1504 irq_src->funcs = &smu_v13_0_6_irq_funcs;
1506 ret = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_MP1,
1507 IH_INTERRUPT_ID_TO_DRIVER,
1515 static int smu_v13_0_6_notify_unload(struct smu_context *smu)
1517 if (amdgpu_in_reset(smu->adev))
1520 dev_dbg(smu->adev->dev, "Notify PMFW about driver unload");
1521 /* Ignore return, just intimate FW that driver is not going to be there */
1522 smu_cmn_send_smc_msg(smu, SMU_MSG_PrepareMp1ForUnload, NULL);
1527 static int smu_v13_0_6_mca_set_debug_mode(struct smu_context *smu, bool enable)
1529 /* NOTE: this ClearMcaOnRead message is only supported for smu version 85.72.0 or higher */
1530 if (smu->smc_fw_version < 0x554800)
1533 return smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_ClearMcaOnRead,
1534 enable ? 0 : ClearMcaOnRead_UE_FLAG_MASK | ClearMcaOnRead_CE_POLL_MASK,
1538 static int smu_v13_0_6_system_features_control(struct smu_context *smu,
1541 struct amdgpu_device *adev = smu->adev;
1544 if (amdgpu_sriov_vf(adev))
1548 if (!(adev->flags & AMD_IS_APU))
1549 ret = smu_v13_0_system_features_control(smu, enable);
1551 /* Notify FW that the device is no longer driver managed */
1552 smu_v13_0_6_notify_unload(smu);
1558 static int smu_v13_0_6_set_gfx_soft_freq_limited_range(struct smu_context *smu,
1564 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_SetSoftMaxGfxClk,
1565 max & 0xffff, NULL);
1569 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_SetSoftMinGfxclk,
1570 min & 0xffff, NULL);
1575 static int smu_v13_0_6_set_performance_level(struct smu_context *smu,
1576 enum amd_dpm_forced_level level)
1578 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm);
1579 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context;
1580 struct smu_13_0_dpm_table *gfx_table =
1581 &dpm_context->dpm_tables.gfx_table;
1582 struct smu_13_0_dpm_table *uclk_table =
1583 &dpm_context->dpm_tables.uclk_table;
1584 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table;
1587 /* Disable determinism if switching to another mode */
1588 if ((smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) &&
1589 (level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM)) {
1590 smu_cmn_send_smc_msg(smu, SMU_MSG_DisableDeterminism, NULL);
1591 pstate_table->gfxclk_pstate.curr.max = gfx_table->max;
1595 case AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM:
1598 case AMD_DPM_FORCED_LEVEL_AUTO:
1599 if ((gfx_table->min != pstate_table->gfxclk_pstate.curr.min) ||
1600 (gfx_table->max != pstate_table->gfxclk_pstate.curr.max)) {
1601 ret = smu_v13_0_6_set_gfx_soft_freq_limited_range(
1602 smu, gfx_table->min, gfx_table->max);
1606 pstate_table->gfxclk_pstate.curr.min = gfx_table->min;
1607 pstate_table->gfxclk_pstate.curr.max = gfx_table->max;
1610 if (uclk_table->max != pstate_table->uclk_pstate.curr.max) {
1611 /* Min UCLK is not expected to be changed */
1612 ret = smu_v13_0_set_soft_freq_limited_range(
1613 smu, SMU_UCLK, 0, uclk_table->max);
1616 pstate_table->uclk_pstate.curr.max = uclk_table->max;
1618 pstate_table->uclk_pstate.custom.max = 0;
1621 case AMD_DPM_FORCED_LEVEL_MANUAL:
1630 static int smu_v13_0_6_set_soft_freq_limited_range(struct smu_context *smu,
1631 enum smu_clk_type clk_type,
1632 uint32_t min, uint32_t max)
1634 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm);
1635 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context;
1636 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table;
1637 struct amdgpu_device *adev = smu->adev;
1642 if (clk_type != SMU_GFXCLK && clk_type != SMU_SCLK &&
1643 clk_type != SMU_UCLK)
1646 if ((smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL) &&
1647 (smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM))
1650 if (smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_MANUAL) {
1652 dev_err(smu->adev->dev,
1653 "Minimum clk should be less than the maximum allowed clock\n");
1657 if (clk_type == SMU_GFXCLK) {
1658 if ((min == pstate_table->gfxclk_pstate.curr.min) &&
1659 (max == pstate_table->gfxclk_pstate.curr.max))
1662 ret = smu_v13_0_6_set_gfx_soft_freq_limited_range(
1665 pstate_table->gfxclk_pstate.curr.min = min;
1666 pstate_table->gfxclk_pstate.curr.max = max;
1670 if (clk_type == SMU_UCLK) {
1671 if (max == pstate_table->uclk_pstate.curr.max)
1673 /* Only max clock limiting is allowed for UCLK */
1674 ret = smu_v13_0_set_soft_freq_limited_range(
1675 smu, SMU_UCLK, 0, max);
1677 pstate_table->uclk_pstate.curr.max = max;
1683 if (smu_dpm->dpm_level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) {
1684 if (!max || (max < dpm_context->dpm_tables.gfx_table.min) ||
1685 (max > dpm_context->dpm_tables.gfx_table.max)) {
1688 "Invalid max frequency %d MHz specified for determinism\n",
1693 /* Restore default min/max clocks and enable determinism */
1694 min_clk = dpm_context->dpm_tables.gfx_table.min;
1695 max_clk = dpm_context->dpm_tables.gfx_table.max;
1696 ret = smu_v13_0_6_set_gfx_soft_freq_limited_range(smu, min_clk,
1699 usleep_range(500, 1000);
1700 ret = smu_cmn_send_smc_msg_with_param(
1701 smu, SMU_MSG_EnableDeterminism, max, NULL);
1704 "Failed to enable determinism at GFX clock %d MHz\n",
1707 pstate_table->gfxclk_pstate.curr.min = min_clk;
1708 pstate_table->gfxclk_pstate.curr.max = max;
1716 static int smu_v13_0_6_usr_edit_dpm_table(struct smu_context *smu,
1717 enum PP_OD_DPM_TABLE_COMMAND type,
1718 long input[], uint32_t size)
1720 struct smu_dpm_context *smu_dpm = &(smu->smu_dpm);
1721 struct smu_13_0_dpm_context *dpm_context = smu_dpm->dpm_context;
1722 struct smu_umd_pstate_table *pstate_table = &smu->pstate_table;
1727 /* Only allowed in manual or determinism mode */
1728 if ((smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_MANUAL) &&
1729 (smu_dpm->dpm_level != AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM))
1733 case PP_OD_EDIT_SCLK_VDDC_TABLE:
1735 dev_err(smu->adev->dev,
1736 "Input parameter number not correct\n");
1740 if (input[0] == 0) {
1741 if (input[1] < dpm_context->dpm_tables.gfx_table.min) {
1744 "Minimum GFX clk (%ld) MHz specified is less than the minimum allowed (%d) MHz\n",
1746 dpm_context->dpm_tables.gfx_table.min);
1747 pstate_table->gfxclk_pstate.custom.min =
1748 pstate_table->gfxclk_pstate.curr.min;
1752 pstate_table->gfxclk_pstate.custom.min = input[1];
1753 } else if (input[0] == 1) {
1754 if (input[1] > dpm_context->dpm_tables.gfx_table.max) {
1757 "Maximum GFX clk (%ld) MHz specified is greater than the maximum allowed (%d) MHz\n",
1759 dpm_context->dpm_tables.gfx_table.max);
1760 pstate_table->gfxclk_pstate.custom.max =
1761 pstate_table->gfxclk_pstate.curr.max;
1765 pstate_table->gfxclk_pstate.custom.max = input[1];
1770 case PP_OD_EDIT_MCLK_VDDC_TABLE:
1772 dev_err(smu->adev->dev,
1773 "Input parameter number not correct\n");
1777 if (!smu_cmn_feature_is_enabled(smu,
1778 SMU_FEATURE_DPM_UCLK_BIT)) {
1779 dev_warn(smu->adev->dev,
1780 "UCLK_LIMITS setting not supported!\n");
1784 if (input[0] == 0) {
1785 dev_info(smu->adev->dev,
1786 "Setting min UCLK level is not supported");
1788 } else if (input[0] == 1) {
1789 if (input[1] > dpm_context->dpm_tables.uclk_table.max) {
1792 "Maximum UCLK (%ld) MHz specified is greater than the maximum allowed (%d) MHz\n",
1794 dpm_context->dpm_tables.uclk_table.max);
1795 pstate_table->uclk_pstate.custom.max =
1796 pstate_table->uclk_pstate.curr.max;
1800 pstate_table->uclk_pstate.custom.max = input[1];
1804 case PP_OD_RESTORE_DEFAULT_TABLE:
1806 dev_err(smu->adev->dev,
1807 "Input parameter number not correct\n");
1810 /* Use the default frequencies for manual and determinism mode */
1811 min_clk = dpm_context->dpm_tables.gfx_table.min;
1812 max_clk = dpm_context->dpm_tables.gfx_table.max;
1814 ret = smu_v13_0_6_set_soft_freq_limited_range(
1815 smu, SMU_GFXCLK, min_clk, max_clk);
1820 min_clk = dpm_context->dpm_tables.uclk_table.min;
1821 max_clk = dpm_context->dpm_tables.uclk_table.max;
1822 ret = smu_v13_0_6_set_soft_freq_limited_range(
1823 smu, SMU_UCLK, min_clk, max_clk);
1826 pstate_table->uclk_pstate.custom.max = 0;
1829 case PP_OD_COMMIT_DPM_TABLE:
1831 dev_err(smu->adev->dev,
1832 "Input parameter number not correct\n");
1835 if (!pstate_table->gfxclk_pstate.custom.min)
1836 pstate_table->gfxclk_pstate.custom.min =
1837 pstate_table->gfxclk_pstate.curr.min;
1839 if (!pstate_table->gfxclk_pstate.custom.max)
1840 pstate_table->gfxclk_pstate.custom.max =
1841 pstate_table->gfxclk_pstate.curr.max;
1843 min_clk = pstate_table->gfxclk_pstate.custom.min;
1844 max_clk = pstate_table->gfxclk_pstate.custom.max;
1846 ret = smu_v13_0_6_set_soft_freq_limited_range(
1847 smu, SMU_GFXCLK, min_clk, max_clk);
1852 if (!pstate_table->uclk_pstate.custom.max)
1855 min_clk = pstate_table->uclk_pstate.curr.min;
1856 max_clk = pstate_table->uclk_pstate.custom.max;
1857 return smu_v13_0_6_set_soft_freq_limited_range(
1858 smu, SMU_UCLK, min_clk, max_clk);
1868 static int smu_v13_0_6_get_enabled_mask(struct smu_context *smu,
1869 uint64_t *feature_mask)
1873 ret = smu_cmn_get_enabled_mask(smu, feature_mask);
1875 if (ret == -EIO && smu->smc_fw_version < 0x552F00) {
1883 static bool smu_v13_0_6_is_dpm_running(struct smu_context *smu)
1886 uint64_t feature_enabled;
1888 ret = smu_v13_0_6_get_enabled_mask(smu, &feature_enabled);
1893 return !!(feature_enabled & SMC_DPM_FEATURE);
1896 static int smu_v13_0_6_request_i2c_xfer(struct smu_context *smu,
1899 struct smu_table_context *smu_table = &smu->smu_table;
1900 struct smu_table *table = &smu_table->driver_table;
1901 struct amdgpu_device *adev = smu->adev;
1902 uint32_t table_size;
1908 table_size = smu_table->tables[SMU_TABLE_I2C_COMMANDS].size;
1910 memcpy(table->cpu_addr, table_data, table_size);
1911 /* Flush hdp cache */
1912 amdgpu_asic_flush_hdp(adev, NULL);
1913 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_RequestI2cTransaction,
1919 static int smu_v13_0_6_i2c_xfer(struct i2c_adapter *i2c_adap,
1920 struct i2c_msg *msg, int num_msgs)
1922 struct amdgpu_smu_i2c_bus *smu_i2c = i2c_get_adapdata(i2c_adap);
1923 struct amdgpu_device *adev = smu_i2c->adev;
1924 struct smu_context *smu = adev->powerplay.pp_handle;
1925 struct smu_table_context *smu_table = &smu->smu_table;
1926 struct smu_table *table = &smu_table->driver_table;
1927 SwI2cRequest_t *req, *res = (SwI2cRequest_t *)table->cpu_addr;
1931 if (!adev->pm.dpm_enabled)
1934 req = kzalloc(sizeof(*req), GFP_KERNEL);
1938 req->I2CcontrollerPort = smu_i2c->port;
1939 req->I2CSpeed = I2C_SPEED_FAST_400K;
1940 req->SlaveAddress = msg[0].addr << 1; /* wants an 8-bit address */
1941 dir = msg[0].flags & I2C_M_RD;
1943 for (c = i = 0; i < num_msgs; i++) {
1944 for (j = 0; j < msg[i].len; j++, c++) {
1945 SwI2cCmd_t *cmd = &req->SwI2cCmds[c];
1947 if (!(msg[i].flags & I2C_M_RD)) {
1949 cmd->CmdConfig |= CMDCONFIG_READWRITE_MASK;
1950 cmd->ReadWriteData = msg[i].buf[j];
1953 if ((dir ^ msg[i].flags) & I2C_M_RD) {
1954 /* The direction changes.
1956 dir = msg[i].flags & I2C_M_RD;
1957 cmd->CmdConfig |= CMDCONFIG_RESTART_MASK;
1963 * Insert STOP if we are at the last byte of either last
1964 * message for the transaction or the client explicitly
1965 * requires a STOP at this particular message.
1967 if ((j == msg[i].len - 1) &&
1968 ((i == num_msgs - 1) || (msg[i].flags & I2C_M_STOP))) {
1969 cmd->CmdConfig &= ~CMDCONFIG_RESTART_MASK;
1970 cmd->CmdConfig |= CMDCONFIG_STOP_MASK;
1974 mutex_lock(&adev->pm.mutex);
1975 r = smu_v13_0_6_request_i2c_xfer(smu, req);
1979 for (c = i = 0; i < num_msgs; i++) {
1980 if (!(msg[i].flags & I2C_M_RD)) {
1984 for (j = 0; j < msg[i].len; j++, c++) {
1985 SwI2cCmd_t *cmd = &res->SwI2cCmds[c];
1987 msg[i].buf[j] = cmd->ReadWriteData;
1992 mutex_unlock(&adev->pm.mutex);
1997 static u32 smu_v13_0_6_i2c_func(struct i2c_adapter *adap)
1999 return I2C_FUNC_I2C | I2C_FUNC_SMBUS_EMUL;
2002 static const struct i2c_algorithm smu_v13_0_6_i2c_algo = {
2003 .master_xfer = smu_v13_0_6_i2c_xfer,
2004 .functionality = smu_v13_0_6_i2c_func,
2007 static const struct i2c_adapter_quirks smu_v13_0_6_i2c_control_quirks = {
2008 .flags = I2C_AQ_COMB | I2C_AQ_COMB_SAME_ADDR | I2C_AQ_NO_ZERO_LEN,
2009 .max_read_len = MAX_SW_I2C_COMMANDS,
2010 .max_write_len = MAX_SW_I2C_COMMANDS,
2011 .max_comb_1st_msg_len = 2,
2012 .max_comb_2nd_msg_len = MAX_SW_I2C_COMMANDS - 2,
2015 static int smu_v13_0_6_i2c_control_init(struct smu_context *smu)
2017 struct amdgpu_device *adev = smu->adev;
2020 for (i = 0; i < MAX_SMU_I2C_BUSES; i++) {
2021 struct amdgpu_smu_i2c_bus *smu_i2c = &adev->pm.smu_i2c[i];
2022 struct i2c_adapter *control = &smu_i2c->adapter;
2024 smu_i2c->adev = adev;
2026 mutex_init(&smu_i2c->mutex);
2027 control->owner = THIS_MODULE;
2028 control->dev.parent = &adev->pdev->dev;
2029 control->algo = &smu_v13_0_6_i2c_algo;
2030 snprintf(control->name, sizeof(control->name), "AMDGPU SMU %d", i);
2031 control->quirks = &smu_v13_0_6_i2c_control_quirks;
2032 i2c_set_adapdata(control, smu_i2c);
2034 res = i2c_add_adapter(control);
2036 DRM_ERROR("Failed to register hw i2c, err: %d\n", res);
2041 adev->pm.ras_eeprom_i2c_bus = &adev->pm.smu_i2c[0].adapter;
2042 adev->pm.fru_eeprom_i2c_bus = &adev->pm.smu_i2c[0].adapter;
2046 for ( ; i >= 0; i--) {
2047 struct amdgpu_smu_i2c_bus *smu_i2c = &adev->pm.smu_i2c[i];
2048 struct i2c_adapter *control = &smu_i2c->adapter;
2050 i2c_del_adapter(control);
2055 static void smu_v13_0_6_i2c_control_fini(struct smu_context *smu)
2057 struct amdgpu_device *adev = smu->adev;
2060 for (i = 0; i < MAX_SMU_I2C_BUSES; i++) {
2061 struct amdgpu_smu_i2c_bus *smu_i2c = &adev->pm.smu_i2c[i];
2062 struct i2c_adapter *control = &smu_i2c->adapter;
2064 i2c_del_adapter(control);
2066 adev->pm.ras_eeprom_i2c_bus = NULL;
2067 adev->pm.fru_eeprom_i2c_bus = NULL;
2070 static void smu_v13_0_6_get_unique_id(struct smu_context *smu)
2072 struct amdgpu_device *adev = smu->adev;
2073 struct smu_table_context *smu_table = &smu->smu_table;
2074 struct PPTable_t *pptable =
2075 (struct PPTable_t *)smu_table->driver_pptable;
2077 adev->unique_id = pptable->PublicSerialNumber_AID;
2080 static bool smu_v13_0_6_is_baco_supported(struct smu_context *smu)
2082 /* smu_13_0_6 does not support baco */
2087 static const char *const throttling_logging_label[] = {
2088 [THROTTLER_PROCHOT_BIT] = "Prochot",
2089 [THROTTLER_PPT_BIT] = "PPT",
2090 [THROTTLER_THERMAL_SOCKET_BIT] = "SOC",
2091 [THROTTLER_THERMAL_VR_BIT] = "VR",
2092 [THROTTLER_THERMAL_HBM_BIT] = "HBM"
2095 static void smu_v13_0_6_log_thermal_throttling_event(struct smu_context *smu)
2097 int throttler_idx, throttling_events = 0, buf_idx = 0;
2098 struct amdgpu_device *adev = smu->adev;
2099 uint32_t throttler_status;
2102 throttler_status = smu_v13_0_6_get_throttler_status(smu);
2103 if (!throttler_status)
2106 memset(log_buf, 0, sizeof(log_buf));
2107 for (throttler_idx = 0;
2108 throttler_idx < ARRAY_SIZE(throttling_logging_label);
2110 if (throttler_status & (1U << throttler_idx)) {
2111 throttling_events++;
2112 buf_idx += snprintf(
2113 log_buf + buf_idx, sizeof(log_buf) - buf_idx,
2114 "%s%s", throttling_events > 1 ? " and " : "",
2115 throttling_logging_label[throttler_idx]);
2116 if (buf_idx >= sizeof(log_buf)) {
2117 dev_err(adev->dev, "buffer overflow!\n");
2118 log_buf[sizeof(log_buf) - 1] = '\0';
2125 "WARN: GPU is throttled, expect performance decrease. %s.\n",
2127 kgd2kfd_smi_event_throttle(
2129 smu_cmn_get_indep_throttler_status(throttler_status,
2130 smu_v13_0_6_throttler_map));
2134 smu_v13_0_6_get_current_pcie_link_width_level(struct smu_context *smu)
2136 struct amdgpu_device *adev = smu->adev;
2138 return REG_GET_FIELD(RREG32_PCIE(smnPCIE_LC_LINK_WIDTH_CNTL),
2139 PCIE_LC_LINK_WIDTH_CNTL, LC_LINK_WIDTH_RD);
2142 static int smu_v13_0_6_get_current_pcie_link_speed(struct smu_context *smu)
2144 struct amdgpu_device *adev = smu->adev;
2145 uint32_t speed_level;
2148 /* TODO: confirm this on real target */
2149 esm_ctrl = RREG32_PCIE(smnPCIE_ESM_CTRL);
2150 if ((esm_ctrl >> 15) & 0x1FFFF)
2151 return (((esm_ctrl >> 8) & 0x3F) + 128);
2153 speed_level = (RREG32_PCIE(smnPCIE_LC_SPEED_CNTL) &
2154 PCIE_LC_SPEED_CNTL__LC_CURRENT_DATA_RATE_MASK)
2155 >> PCIE_LC_SPEED_CNTL__LC_CURRENT_DATA_RATE__SHIFT;
2156 if (speed_level > LINK_SPEED_MAX)
2159 return pcie_gen_to_speed(speed_level + 1);
2162 static ssize_t smu_v13_0_6_get_gpu_metrics(struct smu_context *smu, void **table)
2164 struct smu_table_context *smu_table = &smu->smu_table;
2165 struct gpu_metrics_v1_5 *gpu_metrics =
2166 (struct gpu_metrics_v1_5 *)smu_table->gpu_metrics_table;
2167 struct amdgpu_device *adev = smu->adev;
2168 int ret = 0, xcc_id, inst, i, j;
2169 MetricsTableX_t *metrics_x;
2170 MetricsTableA_t *metrics_a;
2171 u16 link_width_level;
2173 metrics_x = kzalloc(max(sizeof(MetricsTableX_t), sizeof(MetricsTableA_t)), GFP_KERNEL);
2174 ret = smu_v13_0_6_get_metrics_table(smu, metrics_x, true);
2180 metrics_a = (MetricsTableA_t *)metrics_x;
2182 smu_cmn_init_soft_gpu_metrics(gpu_metrics, 1, 5);
2184 gpu_metrics->temperature_hotspot =
2185 SMUQ10_ROUND(GET_METRIC_FIELD(MaxSocketTemperature));
2186 /* Individual HBM stack temperature is not reported */
2187 gpu_metrics->temperature_mem =
2188 SMUQ10_ROUND(GET_METRIC_FIELD(MaxHbmTemperature));
2189 /* Reports max temperature of all voltage rails */
2190 gpu_metrics->temperature_vrsoc =
2191 SMUQ10_ROUND(GET_METRIC_FIELD(MaxVrTemperature));
2193 gpu_metrics->average_gfx_activity =
2194 SMUQ10_ROUND(GET_METRIC_FIELD(SocketGfxBusy));
2195 gpu_metrics->average_umc_activity =
2196 SMUQ10_ROUND(GET_METRIC_FIELD(DramBandwidthUtilization));
2198 gpu_metrics->curr_socket_power =
2199 SMUQ10_ROUND(GET_METRIC_FIELD(SocketPower));
2200 /* Energy counter reported in 15.259uJ (2^-16) units */
2201 gpu_metrics->energy_accumulator = GET_METRIC_FIELD(SocketEnergyAcc);
2203 for (i = 0; i < MAX_GFX_CLKS; i++) {
2204 xcc_id = GET_INST(GC, i);
2206 gpu_metrics->current_gfxclk[i] =
2207 SMUQ10_ROUND(GET_METRIC_FIELD(GfxclkFrequency)[xcc_id]);
2210 gpu_metrics->current_socclk[i] =
2211 SMUQ10_ROUND(GET_METRIC_FIELD(SocclkFrequency)[i]);
2212 inst = GET_INST(VCN, i);
2214 gpu_metrics->current_vclk0[i] =
2215 SMUQ10_ROUND(GET_METRIC_FIELD(VclkFrequency)[inst]);
2216 gpu_metrics->current_dclk0[i] =
2217 SMUQ10_ROUND(GET_METRIC_FIELD(DclkFrequency)[inst]);
2222 gpu_metrics->current_uclk = SMUQ10_ROUND(GET_METRIC_FIELD(UclkFrequency));
2224 /* Throttle status is not reported through metrics now */
2225 gpu_metrics->throttle_status = 0;
2227 /* Clock Lock Status. Each bit corresponds to each GFXCLK instance */
2228 gpu_metrics->gfxclk_lock_status = GET_METRIC_FIELD(GfxLockXCDMak) >> GET_INST(GC, 0);
2230 if (!(adev->flags & AMD_IS_APU)) {
2231 link_width_level = smu_v13_0_6_get_current_pcie_link_width_level(smu);
2232 if (link_width_level > MAX_LINK_WIDTH)
2233 link_width_level = 0;
2235 gpu_metrics->pcie_link_width =
2236 DECODE_LANE_WIDTH(link_width_level);
2237 gpu_metrics->pcie_link_speed =
2238 smu_v13_0_6_get_current_pcie_link_speed(smu);
2239 gpu_metrics->pcie_bandwidth_acc =
2240 SMUQ10_ROUND(metrics_x->PcieBandwidthAcc[0]);
2241 gpu_metrics->pcie_bandwidth_inst =
2242 SMUQ10_ROUND(metrics_x->PcieBandwidth[0]);
2243 gpu_metrics->pcie_l0_to_recov_count_acc =
2244 metrics_x->PCIeL0ToRecoveryCountAcc;
2245 gpu_metrics->pcie_replay_count_acc =
2246 metrics_x->PCIenReplayAAcc;
2247 gpu_metrics->pcie_replay_rover_count_acc =
2248 metrics_x->PCIenReplayARolloverCountAcc;
2249 gpu_metrics->pcie_nak_sent_count_acc =
2250 metrics_x->PCIeNAKSentCountAcc;
2251 gpu_metrics->pcie_nak_rcvd_count_acc =
2252 metrics_x->PCIeNAKReceivedCountAcc;
2255 gpu_metrics->system_clock_counter = ktime_get_boottime_ns();
2257 gpu_metrics->gfx_activity_acc =
2258 SMUQ10_ROUND(GET_METRIC_FIELD(SocketGfxBusyAcc));
2259 gpu_metrics->mem_activity_acc =
2260 SMUQ10_ROUND(GET_METRIC_FIELD(DramBandwidthUtilizationAcc));
2262 for (i = 0; i < NUM_XGMI_LINKS; i++) {
2263 gpu_metrics->xgmi_read_data_acc[i] =
2264 SMUQ10_ROUND(GET_METRIC_FIELD(XgmiReadDataSizeAcc)[i]);
2265 gpu_metrics->xgmi_write_data_acc[i] =
2266 SMUQ10_ROUND(GET_METRIC_FIELD(XgmiWriteDataSizeAcc)[i]);
2269 for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
2270 inst = GET_INST(JPEG, i);
2271 for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j) {
2272 gpu_metrics->jpeg_activity[(i * adev->jpeg.num_jpeg_rings) + j] =
2273 SMUQ10_ROUND(GET_METRIC_FIELD(JpegBusy)
2274 [(inst * adev->jpeg.num_jpeg_rings) + j]);
2278 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2279 inst = GET_INST(VCN, i);
2280 gpu_metrics->vcn_activity[i] =
2281 SMUQ10_ROUND(GET_METRIC_FIELD(VcnBusy)[inst]);
2284 gpu_metrics->xgmi_link_width = SMUQ10_ROUND(GET_METRIC_FIELD(XgmiWidth));
2285 gpu_metrics->xgmi_link_speed = SMUQ10_ROUND(GET_METRIC_FIELD(XgmiBitrate));
2287 gpu_metrics->firmware_timestamp = GET_METRIC_FIELD(Timestamp);
2289 *table = (void *)gpu_metrics;
2292 return sizeof(*gpu_metrics);
2295 static int smu_v13_0_6_mode2_reset(struct smu_context *smu)
2298 struct amdgpu_device *adev = smu->adev;
2301 index = smu_cmn_to_asic_specific_index(smu, CMN2ASIC_MAPPING_MSG,
2302 SMU_MSG_GfxDeviceDriverReset);
2304 mutex_lock(&smu->message_lock);
2306 ret = smu_cmn_send_msg_without_waiting(smu, (uint16_t)index,
2309 /* This is similar to FLR, wait till max FLR timeout */
2312 dev_dbg(smu->adev->dev, "restore config space...\n");
2313 /* Restore the config space saved during init */
2314 amdgpu_device_load_pci_state(adev->pdev);
2316 dev_dbg(smu->adev->dev, "wait for reset ack\n");
2318 ret = smu_cmn_wait_for_response(smu);
2319 /* Wait a bit more time for getting ACK */
2320 if (ret == -ETIME) {
2322 usleep_range(500, 1000);
2329 } while (ret == -ETIME && timeout);
2332 mutex_unlock(&smu->message_lock);
2335 dev_err(adev->dev, "failed to send mode2 reset, error code %d",
2341 static int smu_v13_0_6_get_thermal_temperature_range(struct smu_context *smu,
2342 struct smu_temperature_range *range)
2344 struct amdgpu_device *adev = smu->adev;
2345 u32 aid_temp, xcd_temp, max_temp;
2349 if (amdgpu_sriov_vf(smu->adev))
2355 /*Check smu version, GetCtfLimit message only supported for smu version 85.69 or higher */
2356 if (smu->smc_fw_version < 0x554500)
2359 /* Get SOC Max operating temperature */
2360 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_GetCTFLimit,
2361 PPSMC_AID_THM_TYPE, &aid_temp);
2364 if (adev->flags & AMD_IS_APU) {
2365 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_GetCTFLimit,
2366 PPSMC_CCD_THM_TYPE, &ccd_temp);
2370 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_GetCTFLimit,
2371 PPSMC_XCD_THM_TYPE, &xcd_temp);
2374 range->hotspot_emergency_max = max3(aid_temp, xcd_temp, ccd_temp) *
2375 SMU_TEMPERATURE_UNITS_PER_CENTIGRADES;
2377 /* Get HBM Max operating temperature */
2378 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_GetCTFLimit,
2379 PPSMC_HBM_THM_TYPE, &max_temp);
2382 range->mem_emergency_max =
2383 max_temp * SMU_TEMPERATURE_UNITS_PER_CENTIGRADES;
2385 /* Get SOC thermal throttle limit */
2386 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_GetThermalLimit,
2387 PPSMC_THROTTLING_LIMIT_TYPE_SOCKET,
2391 range->hotspot_crit_max =
2392 max_temp * SMU_TEMPERATURE_UNITS_PER_CENTIGRADES;
2394 /* Get HBM thermal throttle limit */
2395 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_GetThermalLimit,
2396 PPSMC_THROTTLING_LIMIT_TYPE_HBM,
2401 range->mem_crit_max = max_temp * SMU_TEMPERATURE_UNITS_PER_CENTIGRADES;
2407 static int smu_v13_0_6_mode1_reset(struct smu_context *smu)
2409 struct amdgpu_device *adev = smu->adev;
2410 struct amdgpu_hive_info *hive = NULL;
2411 u32 hive_ras_recovery = 0;
2412 struct amdgpu_ras *ras;
2413 u32 fatal_err, param;
2416 hive = amdgpu_get_xgmi_hive(adev);
2417 ras = amdgpu_ras_get_context(adev);
2419 param = SMU_RESET_MODE_1;
2422 hive_ras_recovery = atomic_read(&hive->ras_recovery);
2423 amdgpu_put_xgmi_hive(hive);
2426 /* fatal error triggered by ras, PMFW supports the flag */
2427 if (ras && (atomic_read(&ras->in_recovery) || hive_ras_recovery))
2430 param |= (fatal_err << 16);
2431 ret = smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_GfxDeviceDriverReset,
2435 msleep(SMU13_MODE1_RESET_WAIT_TIME_IN_MS);
2440 static bool smu_v13_0_6_is_mode1_reset_supported(struct smu_context *smu)
2445 static bool smu_v13_0_6_is_mode2_reset_supported(struct smu_context *smu)
2450 static int smu_v13_0_6_smu_send_hbm_bad_page_num(struct smu_context *smu,
2455 /* message SMU to update the bad page number on SMUBUS */
2456 ret = smu_cmn_send_smc_msg_with_param(
2457 smu, SMU_MSG_SetNumBadHbmPagesRetired, size, NULL);
2459 dev_err(smu->adev->dev,
2460 "[%s] failed to message SMU to update HBM bad pages number\n",
2466 static int smu_v13_0_6_send_rma_reason(struct smu_context *smu)
2468 struct amdgpu_device *adev = smu->adev;
2471 /* NOTE: the message is only valid on dGPU with pmfw 85.90.0 and above */
2472 if ((adev->flags & AMD_IS_APU) || smu->smc_fw_version < 0x00555a00)
2475 ret = smu_cmn_send_smc_msg(smu, SMU_MSG_RmaDueToBadPageThreshold, NULL);
2477 dev_err(smu->adev->dev,
2478 "[%s] failed to send BadPageThreshold event to SMU\n",
2484 static int mca_smu_set_debug_mode(struct amdgpu_device *adev, bool enable)
2486 struct smu_context *smu = adev->powerplay.pp_handle;
2488 return smu_v13_0_6_mca_set_debug_mode(smu, enable);
2491 static int smu_v13_0_6_get_valid_mca_count(struct smu_context *smu, enum amdgpu_mca_error_type type, uint32_t *count)
2500 case AMDGPU_MCA_ERROR_TYPE_UE:
2501 msg = SMU_MSG_QueryValidMcaCount;
2503 case AMDGPU_MCA_ERROR_TYPE_CE:
2504 msg = SMU_MSG_QueryValidMcaCeCount;
2510 ret = smu_cmn_send_smc_msg(smu, msg, count);
2519 static int __smu_v13_0_6_mca_dump_bank(struct smu_context *smu, enum amdgpu_mca_error_type type,
2520 int idx, int offset, uint32_t *val)
2522 uint32_t msg, param;
2525 case AMDGPU_MCA_ERROR_TYPE_UE:
2526 msg = SMU_MSG_McaBankDumpDW;
2528 case AMDGPU_MCA_ERROR_TYPE_CE:
2529 msg = SMU_MSG_McaBankCeDumpDW;
2535 param = ((idx & 0xffff) << 16) | (offset & 0xfffc);
2537 return smu_cmn_send_smc_msg_with_param(smu, msg, param, val);
2540 static int smu_v13_0_6_mca_dump_bank(struct smu_context *smu, enum amdgpu_mca_error_type type,
2541 int idx, int offset, uint32_t *val, int count)
2548 for (i = 0; i < count; i++) {
2549 ret = __smu_v13_0_6_mca_dump_bank(smu, type, idx, offset + (i << 2), &val[i]);
2557 static const struct mca_bank_ipid smu_v13_0_6_mca_ipid_table[AMDGPU_MCA_IP_COUNT] = {
2558 MCA_BANK_IPID(UMC, 0x96, 0x0),
2559 MCA_BANK_IPID(SMU, 0x01, 0x1),
2560 MCA_BANK_IPID(MP5, 0x01, 0x2),
2561 MCA_BANK_IPID(PCS_XGMI, 0x50, 0x0),
2564 static void mca_bank_entry_info_decode(struct mca_bank_entry *entry, struct mca_bank_info *info)
2566 u64 ipid = entry->regs[MCA_REG_IDX_IPID];
2567 u32 instidhi, instid;
2569 /* NOTE: All MCA IPID register share the same format,
2570 * so the driver can share the MCMP1 register header file.
2573 info->hwid = REG_GET_FIELD(ipid, MCMP1_IPIDT0, HardwareID);
2574 info->mcatype = REG_GET_FIELD(ipid, MCMP1_IPIDT0, McaType);
2577 * Unfied DieID Format: SAASS. A:AID, S:Socket.
2578 * Unfied DieID[4] = InstanceId[0]
2579 * Unfied DieID[0:3] = InstanceIdHi[0:3]
2581 instidhi = REG_GET_FIELD(ipid, MCMP1_IPIDT0, InstanceIdHi);
2582 instid = REG_GET_FIELD(ipid, MCMP1_IPIDT0, InstanceIdLo);
2583 info->aid = ((instidhi >> 2) & 0x03);
2584 info->socket_id = ((instid & 0x1) << 2) | (instidhi & 0x03);
2587 static int mca_bank_read_reg(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
2588 int idx, int reg_idx, uint64_t *val)
2590 struct smu_context *smu = adev->powerplay.pp_handle;
2591 uint32_t data[2] = {0, 0};
2594 if (!val || reg_idx >= MCA_REG_IDX_COUNT)
2597 ret = smu_v13_0_6_mca_dump_bank(smu, type, idx, reg_idx * 8, data, ARRAY_SIZE(data));
2601 *val = (uint64_t)data[1] << 32 | data[0];
2603 dev_dbg(adev->dev, "mca read bank reg: type:%s, index: %d, reg_idx: %d, val: 0x%016llx\n",
2604 type == AMDGPU_MCA_ERROR_TYPE_UE ? "UE" : "CE", idx, reg_idx, *val);
2609 static int mca_get_mca_entry(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
2610 int idx, struct mca_bank_entry *entry)
2614 /* NOTE: populated all mca register by default */
2615 for (i = 0; i < ARRAY_SIZE(entry->regs); i++) {
2616 ret = mca_bank_read_reg(adev, type, idx, i, &entry->regs[i]);
2624 mca_bank_entry_info_decode(entry, &entry->info);
2629 static int mca_decode_ipid_to_hwip(uint64_t val)
2631 const struct mca_bank_ipid *ipid;
2632 uint16_t hwid, mcatype;
2635 hwid = REG_GET_FIELD(val, MCMP1_IPIDT0, HardwareID);
2636 mcatype = REG_GET_FIELD(val, MCMP1_IPIDT0, McaType);
2638 for (i = 0; i < ARRAY_SIZE(smu_v13_0_6_mca_ipid_table); i++) {
2639 ipid = &smu_v13_0_6_mca_ipid_table[i];
2644 if (ipid->hwid == hwid && ipid->mcatype == mcatype)
2648 return AMDGPU_MCA_IP_UNKNOW;
2651 static int mca_umc_mca_get_err_count(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
2652 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry, uint32_t *count)
2655 uint32_t ext_error_code;
2656 uint32_t odecc_err_cnt;
2658 status0 = entry->regs[MCA_REG_IDX_STATUS];
2659 ext_error_code = MCA_REG__STATUS__ERRORCODEEXT(status0);
2660 odecc_err_cnt = MCA_REG__MISC0__ERRCNT(entry->regs[MCA_REG_IDX_MISC0]);
2662 if (!REG_GET_FIELD(status0, MCMP1_STATUST0, Val)) {
2667 if (umc_v12_0_is_deferred_error(adev, status0) ||
2668 umc_v12_0_is_uncorrectable_error(adev, status0) ||
2669 umc_v12_0_is_correctable_error(adev, status0))
2670 *count = (ext_error_code == 0) ? odecc_err_cnt : 1;
2675 static int mca_pcs_xgmi_mca_get_err_count(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
2676 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry,
2682 ext_error_code = MCA_REG__STATUS__ERRORCODEEXT(entry->regs[MCA_REG_IDX_STATUS]);
2683 err_cnt = MCA_REG__MISC0__ERRCNT(entry->regs[MCA_REG_IDX_MISC0]);
2685 if (type == AMDGPU_MCA_ERROR_TYPE_UE && ext_error_code == 0)
2687 else if (type == AMDGPU_MCA_ERROR_TYPE_CE && ext_error_code == 6)
2693 static bool mca_smu_check_error_code(struct amdgpu_device *adev, const struct mca_ras_info *mca_ras,
2698 if (!mca_ras->err_code_count || !mca_ras->err_code_array)
2701 for (i = 0; i < mca_ras->err_code_count; i++) {
2702 if (errcode == mca_ras->err_code_array[i])
2709 static int mca_gfx_mca_get_err_count(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
2710 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry, uint32_t *count)
2712 uint64_t status0, misc0;
2714 status0 = entry->regs[MCA_REG_IDX_STATUS];
2715 if (!REG_GET_FIELD(status0, MCMP1_STATUST0, Val)) {
2720 if (type == AMDGPU_MCA_ERROR_TYPE_UE &&
2721 REG_GET_FIELD(status0, MCMP1_STATUST0, UC) == 1 &&
2722 REG_GET_FIELD(status0, MCMP1_STATUST0, PCC) == 1) {
2726 misc0 = entry->regs[MCA_REG_IDX_MISC0];
2727 *count = REG_GET_FIELD(misc0, MCMP1_MISC0T0, ErrCnt);
2733 static int mca_smu_mca_get_err_count(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
2734 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry, uint32_t *count)
2736 uint64_t status0, misc0;
2738 status0 = entry->regs[MCA_REG_IDX_STATUS];
2739 if (!REG_GET_FIELD(status0, MCMP1_STATUST0, Val)) {
2744 if (type == AMDGPU_MCA_ERROR_TYPE_UE &&
2745 REG_GET_FIELD(status0, MCMP1_STATUST0, UC) == 1 &&
2746 REG_GET_FIELD(status0, MCMP1_STATUST0, PCC) == 1) {
2752 misc0 = entry->regs[MCA_REG_IDX_MISC0];
2753 *count = REG_GET_FIELD(misc0, MCMP1_MISC0T0, ErrCnt);
2758 static bool mca_gfx_smu_bank_is_valid(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
2759 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry)
2763 instlo = REG_GET_FIELD(entry->regs[MCA_REG_IDX_IPID], MCMP1_IPIDT0, InstanceIdLo);
2764 instlo &= GENMASK(31, 1);
2766 case 0x36430400: /* SMNAID XCD 0 */
2767 case 0x38430400: /* SMNAID XCD 1 */
2768 case 0x40430400: /* SMNXCD XCD 0, NOTE: FIXME: fix this error later */
2777 static bool mca_smu_bank_is_valid(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
2778 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry)
2780 struct smu_context *smu = adev->powerplay.pp_handle;
2781 uint32_t errcode, instlo;
2783 instlo = REG_GET_FIELD(entry->regs[MCA_REG_IDX_IPID], MCMP1_IPIDT0, InstanceIdLo);
2784 instlo &= GENMASK(31, 1);
2785 if (instlo != 0x03b30400)
2788 if (!(adev->flags & AMD_IS_APU) && smu->smc_fw_version >= 0x00555600) {
2789 errcode = MCA_REG__SYND__ERRORINFORMATION(entry->regs[MCA_REG_IDX_SYND]);
2792 errcode = REG_GET_FIELD(entry->regs[MCA_REG_IDX_STATUS], MCMP1_STATUST0, ErrorCode);
2795 return mca_smu_check_error_code(adev, mca_ras, errcode);
2798 static int sdma_err_codes[] = { CODE_SDMA0, CODE_SDMA1, CODE_SDMA2, CODE_SDMA3 };
2799 static int mmhub_err_codes[] = {
2800 CODE_DAGB0, CODE_DAGB0 + 1, CODE_DAGB0 + 2, CODE_DAGB0 + 3, CODE_DAGB0 + 4, /* DAGB0-4 */
2801 CODE_EA0, CODE_EA0 + 1, CODE_EA0 + 2, CODE_EA0 + 3, CODE_EA0 + 4, /* MMEA0-4*/
2802 CODE_VML2, CODE_VML2_WALKER, CODE_MMCANE,
2805 static const struct mca_ras_info mca_ras_table[] = {
2807 .blkid = AMDGPU_RAS_BLOCK__UMC,
2808 .ip = AMDGPU_MCA_IP_UMC,
2809 .get_err_count = mca_umc_mca_get_err_count,
2811 .blkid = AMDGPU_RAS_BLOCK__GFX,
2812 .ip = AMDGPU_MCA_IP_SMU,
2813 .get_err_count = mca_gfx_mca_get_err_count,
2814 .bank_is_valid = mca_gfx_smu_bank_is_valid,
2816 .blkid = AMDGPU_RAS_BLOCK__SDMA,
2817 .ip = AMDGPU_MCA_IP_SMU,
2818 .err_code_array = sdma_err_codes,
2819 .err_code_count = ARRAY_SIZE(sdma_err_codes),
2820 .get_err_count = mca_smu_mca_get_err_count,
2821 .bank_is_valid = mca_smu_bank_is_valid,
2823 .blkid = AMDGPU_RAS_BLOCK__MMHUB,
2824 .ip = AMDGPU_MCA_IP_SMU,
2825 .err_code_array = mmhub_err_codes,
2826 .err_code_count = ARRAY_SIZE(mmhub_err_codes),
2827 .get_err_count = mca_smu_mca_get_err_count,
2828 .bank_is_valid = mca_smu_bank_is_valid,
2830 .blkid = AMDGPU_RAS_BLOCK__XGMI_WAFL,
2831 .ip = AMDGPU_MCA_IP_PCS_XGMI,
2832 .get_err_count = mca_pcs_xgmi_mca_get_err_count,
2836 static const struct mca_ras_info *mca_get_mca_ras_info(struct amdgpu_device *adev, enum amdgpu_ras_block blkid)
2840 for (i = 0; i < ARRAY_SIZE(mca_ras_table); i++) {
2841 if (mca_ras_table[i].blkid == blkid)
2842 return &mca_ras_table[i];
2848 static int mca_get_valid_mca_count(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, uint32_t *count)
2850 struct smu_context *smu = adev->powerplay.pp_handle;
2854 case AMDGPU_MCA_ERROR_TYPE_UE:
2855 case AMDGPU_MCA_ERROR_TYPE_CE:
2856 ret = smu_v13_0_6_get_valid_mca_count(smu, type, count);
2866 static bool mca_bank_is_valid(struct amdgpu_device *adev, const struct mca_ras_info *mca_ras,
2867 enum amdgpu_mca_error_type type, struct mca_bank_entry *entry)
2869 if (mca_decode_ipid_to_hwip(entry->regs[MCA_REG_IDX_IPID]) != mca_ras->ip)
2872 if (mca_ras->bank_is_valid)
2873 return mca_ras->bank_is_valid(mca_ras, adev, type, entry);
2878 static int __mca_smu_get_ras_mca_set(struct amdgpu_device *adev, const struct mca_ras_info *mca_ras,
2879 enum amdgpu_mca_error_type type, struct mca_bank_set *mca_set)
2881 struct mca_bank_entry entry;
2885 ret = mca_get_valid_mca_count(adev, type, &mca_cnt);
2889 /* if valid mca bank count is 0, the driver can return 0 directly */
2893 for (i = 0; i < mca_cnt; i++) {
2894 memset(&entry, 0, sizeof(entry));
2895 ret = mca_get_mca_entry(adev, type, i, &entry);
2899 if (mca_ras && !mca_bank_is_valid(adev, mca_ras, type, &entry))
2902 ret = amdgpu_mca_bank_set_add_entry(mca_set, &entry);
2910 static int mca_smu_get_ras_mca_set(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
2911 enum amdgpu_mca_error_type type, struct mca_bank_set *mca_set)
2913 const struct mca_ras_info *mca_ras = NULL;
2918 if (blk != AMDGPU_RAS_BLOCK_COUNT) {
2919 mca_ras = mca_get_mca_ras_info(adev, blk);
2924 return __mca_smu_get_ras_mca_set(adev, mca_ras, type, mca_set);
2927 static int mca_smu_parse_mca_error_count(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type,
2928 struct mca_bank_entry *entry, uint32_t *count)
2930 const struct mca_ras_info *mca_ras;
2932 if (!entry || !count)
2935 mca_ras = mca_get_mca_ras_info(adev, blk);
2939 if (!mca_bank_is_valid(adev, mca_ras, type, entry)) {
2944 return mca_ras->get_err_count(mca_ras, adev, type, entry, count);
2947 static int mca_smu_get_mca_entry(struct amdgpu_device *adev,
2948 enum amdgpu_mca_error_type type, int idx, struct mca_bank_entry *entry)
2950 return mca_get_mca_entry(adev, type, idx, entry);
2953 static int mca_smu_get_valid_mca_count(struct amdgpu_device *adev,
2954 enum amdgpu_mca_error_type type, uint32_t *count)
2956 return mca_get_valid_mca_count(adev, type, count);
2959 static const struct amdgpu_mca_smu_funcs smu_v13_0_6_mca_smu_funcs = {
2962 .mca_set_debug_mode = mca_smu_set_debug_mode,
2963 .mca_get_ras_mca_set = mca_smu_get_ras_mca_set,
2964 .mca_parse_mca_error_count = mca_smu_parse_mca_error_count,
2965 .mca_get_mca_entry = mca_smu_get_mca_entry,
2966 .mca_get_valid_mca_count = mca_smu_get_valid_mca_count,
2969 static int aca_smu_set_debug_mode(struct amdgpu_device *adev, bool enable)
2971 struct smu_context *smu = adev->powerplay.pp_handle;
2973 return smu_v13_0_6_mca_set_debug_mode(smu, enable);
2976 static int smu_v13_0_6_get_valid_aca_count(struct smu_context *smu, enum aca_error_type type, u32 *count)
2985 case ACA_ERROR_TYPE_UE:
2986 msg = SMU_MSG_QueryValidMcaCount;
2988 case ACA_ERROR_TYPE_CE:
2989 msg = SMU_MSG_QueryValidMcaCeCount;
2995 ret = smu_cmn_send_smc_msg(smu, msg, count);
3004 static int aca_smu_get_valid_aca_count(struct amdgpu_device *adev,
3005 enum aca_error_type type, u32 *count)
3007 struct smu_context *smu = adev->powerplay.pp_handle;
3011 case ACA_ERROR_TYPE_UE:
3012 case ACA_ERROR_TYPE_CE:
3013 ret = smu_v13_0_6_get_valid_aca_count(smu, type, count);
3023 static int __smu_v13_0_6_aca_bank_dump(struct smu_context *smu, enum aca_error_type type,
3024 int idx, int offset, u32 *val)
3026 uint32_t msg, param;
3029 case ACA_ERROR_TYPE_UE:
3030 msg = SMU_MSG_McaBankDumpDW;
3032 case ACA_ERROR_TYPE_CE:
3033 msg = SMU_MSG_McaBankCeDumpDW;
3039 param = ((idx & 0xffff) << 16) | (offset & 0xfffc);
3041 return smu_cmn_send_smc_msg_with_param(smu, msg, param, (uint32_t *)val);
3044 static int smu_v13_0_6_aca_bank_dump(struct smu_context *smu, enum aca_error_type type,
3045 int idx, int offset, u32 *val, int count)
3052 for (i = 0; i < count; i++) {
3053 ret = __smu_v13_0_6_aca_bank_dump(smu, type, idx, offset + (i << 2), &val[i]);
3061 static int aca_bank_read_reg(struct amdgpu_device *adev, enum aca_error_type type,
3062 int idx, int reg_idx, u64 *val)
3064 struct smu_context *smu = adev->powerplay.pp_handle;
3065 u32 data[2] = {0, 0};
3068 if (!val || reg_idx >= ACA_REG_IDX_COUNT)
3071 ret = smu_v13_0_6_aca_bank_dump(smu, type, idx, reg_idx * 8, data, ARRAY_SIZE(data));
3075 *val = (u64)data[1] << 32 | data[0];
3077 dev_dbg(adev->dev, "mca read bank reg: type:%s, index: %d, reg_idx: %d, val: 0x%016llx\n",
3078 type == ACA_ERROR_TYPE_UE ? "UE" : "CE", idx, reg_idx, *val);
3083 static int aca_smu_get_valid_aca_bank(struct amdgpu_device *adev,
3084 enum aca_error_type type, int idx, struct aca_bank *bank)
3088 count = min_t(int, 16, ARRAY_SIZE(bank->regs));
3089 for (i = 0; i < count; i++) {
3090 ret = aca_bank_read_reg(adev, type, idx, i, &bank->regs[i]);
3098 static const struct aca_smu_funcs smu_v13_0_6_aca_smu_funcs = {
3099 .max_ue_bank_count = 12,
3100 .max_ce_bank_count = 12,
3101 .set_debug_mode = aca_smu_set_debug_mode,
3102 .get_valid_aca_count = aca_smu_get_valid_aca_count,
3103 .get_valid_aca_bank = aca_smu_get_valid_aca_bank,
3106 static int smu_v13_0_6_select_xgmi_plpd_policy(struct smu_context *smu,
3107 enum pp_xgmi_plpd_mode mode)
3109 struct amdgpu_device *adev = smu->adev;
3113 case XGMI_PLPD_DEFAULT:
3114 param = PPSMC_PLPD_MODE_DEFAULT;
3116 case XGMI_PLPD_OPTIMIZED:
3117 param = PPSMC_PLPD_MODE_OPTIMIZED;
3119 case XGMI_PLPD_DISALLOW:
3126 if (mode == XGMI_PLPD_DISALLOW)
3127 ret = smu_cmn_send_smc_msg_with_param(smu,
3128 SMU_MSG_GmiPwrDnControl,
3131 /* change xgmi per-link power down policy */
3132 ret = smu_cmn_send_smc_msg_with_param(smu,
3133 SMU_MSG_SelectPLPDMode,
3138 "select xgmi per-link power down policy %d failed\n",
3144 static const struct pptable_funcs smu_v13_0_6_ppt_funcs = {
3146 .get_allowed_feature_mask = smu_v13_0_6_get_allowed_feature_mask,
3147 /* dpm/clk tables */
3148 .set_default_dpm_table = smu_v13_0_6_set_default_dpm_table,
3149 .populate_umd_state_clk = smu_v13_0_6_populate_umd_state_clk,
3150 .print_clk_levels = smu_v13_0_6_print_clk_levels,
3151 .force_clk_levels = smu_v13_0_6_force_clk_levels,
3152 .read_sensor = smu_v13_0_6_read_sensor,
3153 .set_performance_level = smu_v13_0_6_set_performance_level,
3154 .get_power_limit = smu_v13_0_6_get_power_limit,
3155 .is_dpm_running = smu_v13_0_6_is_dpm_running,
3156 .get_unique_id = smu_v13_0_6_get_unique_id,
3157 .init_microcode = smu_v13_0_6_init_microcode,
3158 .fini_microcode = smu_v13_0_fini_microcode,
3159 .init_smc_tables = smu_v13_0_6_init_smc_tables,
3160 .fini_smc_tables = smu_v13_0_fini_smc_tables,
3161 .init_power = smu_v13_0_init_power,
3162 .fini_power = smu_v13_0_fini_power,
3163 .check_fw_status = smu_v13_0_6_check_fw_status,
3164 /* pptable related */
3165 .check_fw_version = smu_v13_0_check_fw_version,
3166 .set_driver_table_location = smu_v13_0_set_driver_table_location,
3167 .set_tool_table_location = smu_v13_0_set_tool_table_location,
3168 .notify_memory_pool_location = smu_v13_0_notify_memory_pool_location,
3169 .system_features_control = smu_v13_0_6_system_features_control,
3170 .send_smc_msg_with_param = smu_cmn_send_smc_msg_with_param,
3171 .send_smc_msg = smu_cmn_send_smc_msg,
3172 .get_enabled_mask = smu_v13_0_6_get_enabled_mask,
3173 .feature_is_enabled = smu_cmn_feature_is_enabled,
3174 .set_power_limit = smu_v13_0_6_set_power_limit,
3175 .set_xgmi_pstate = smu_v13_0_set_xgmi_pstate,
3176 .register_irq_handler = smu_v13_0_6_register_irq_handler,
3177 .enable_thermal_alert = smu_v13_0_enable_thermal_alert,
3178 .disable_thermal_alert = smu_v13_0_disable_thermal_alert,
3179 .setup_pptable = smu_v13_0_6_setup_pptable,
3180 .baco_is_support = smu_v13_0_6_is_baco_supported,
3181 .get_dpm_ultimate_freq = smu_v13_0_6_get_dpm_ultimate_freq,
3182 .set_soft_freq_limited_range = smu_v13_0_6_set_soft_freq_limited_range,
3183 .od_edit_dpm_table = smu_v13_0_6_usr_edit_dpm_table,
3184 .select_xgmi_plpd_policy = smu_v13_0_6_select_xgmi_plpd_policy,
3185 .log_thermal_throttling_event = smu_v13_0_6_log_thermal_throttling_event,
3186 .get_pp_feature_mask = smu_cmn_get_pp_feature_mask,
3187 .get_gpu_metrics = smu_v13_0_6_get_gpu_metrics,
3188 .get_pm_metrics = smu_v13_0_6_get_pm_metrics,
3189 .get_thermal_temperature_range = smu_v13_0_6_get_thermal_temperature_range,
3190 .mode1_reset_is_support = smu_v13_0_6_is_mode1_reset_supported,
3191 .mode2_reset_is_support = smu_v13_0_6_is_mode2_reset_supported,
3192 .mode1_reset = smu_v13_0_6_mode1_reset,
3193 .mode2_reset = smu_v13_0_6_mode2_reset,
3194 .wait_for_event = smu_v13_0_wait_for_event,
3195 .i2c_init = smu_v13_0_6_i2c_control_init,
3196 .i2c_fini = smu_v13_0_6_i2c_control_fini,
3197 .send_hbm_bad_pages_num = smu_v13_0_6_smu_send_hbm_bad_page_num,
3198 .send_rma_reason = smu_v13_0_6_send_rma_reason,
3201 void smu_v13_0_6_set_ppt_funcs(struct smu_context *smu)
3203 smu->ppt_funcs = &smu_v13_0_6_ppt_funcs;
3204 smu->message_map = smu_v13_0_6_message_map;
3205 smu->clock_map = smu_v13_0_6_clk_map;
3206 smu->feature_map = smu_v13_0_6_feature_mask_map;
3207 smu->table_map = smu_v13_0_6_table_map;
3208 smu->smc_driver_if_version = SMU13_0_6_DRIVER_IF_VERSION;
3209 smu_v13_0_set_smu_mailbox_registers(smu);
3210 amdgpu_mca_smu_init_funcs(smu->adev, &smu_v13_0_6_mca_smu_funcs);
3211 amdgpu_aca_set_smu_funcs(smu->adev, &smu_v13_0_6_aca_smu_funcs);