Zane Shelley | 65fefb2 | 2021-10-18 15:35:26 -0500 | [diff] [blame] | 1 | #include <assert.h> |
| 2 | |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 3 | #include <analyzer/analyzer_main.hpp> |
| 4 | #include <analyzer/ras-data/ras-data-parser.hpp> |
Zane Shelley | 65fefb2 | 2021-10-18 15:35:26 -0500 | [diff] [blame] | 5 | #include <hei_main.hpp> |
Zane Shelley | 19df370 | 2021-12-16 22:32:54 -0600 | [diff] [blame] | 6 | #include <hei_util.hpp> |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 7 | #include <util/pdbg.hpp> |
Zane Shelley | 65fefb2 | 2021-10-18 15:35:26 -0500 | [diff] [blame] | 8 | |
| 9 | #include <algorithm> |
| 10 | #include <limits> |
| 11 | #include <string> |
| 12 | |
| 13 | namespace analyzer |
| 14 | { |
Zane Shelley | 65fefb2 | 2021-10-18 15:35:26 -0500 | [diff] [blame] | 15 | //------------------------------------------------------------------------------ |
| 16 | |
Zane Shelley | a7369f8 | 2021-10-18 16:52:21 -0500 | [diff] [blame] | 17 | bool __findRcsOscError(const std::vector<libhei::Signature>& i_list, |
| 18 | libhei::Signature& o_rootCause) |
| 19 | { |
| 20 | // TODO: Consider returning all of them instead of one as root cause. |
| 21 | auto itr = std::find_if(i_list.begin(), i_list.end(), [&](const auto& t) { |
Zane Shelley | 19df370 | 2021-12-16 22:32:54 -0600 | [diff] [blame] | 22 | return (libhei::hash<libhei::NodeId_t>("TP_LOCAL_FIR") == t.getId() && |
Zane Shelley | a7369f8 | 2021-10-18 16:52:21 -0500 | [diff] [blame] | 23 | (42 == t.getBit() || 43 == t.getBit())); |
| 24 | }); |
| 25 | |
| 26 | if (i_list.end() != itr) |
| 27 | { |
| 28 | o_rootCause = *itr; |
| 29 | return true; |
| 30 | } |
| 31 | |
| 32 | return false; |
| 33 | } |
| 34 | |
| 35 | //------------------------------------------------------------------------------ |
| 36 | |
| 37 | bool __findPllUnlock(const std::vector<libhei::Signature>& i_list, |
| 38 | libhei::Signature& o_rootCause) |
| 39 | { |
| 40 | // TODO: Consider returning all of them instead of one as root cause. |
| 41 | auto itr = std::find_if(i_list.begin(), i_list.end(), [&](const auto& t) { |
Zane Shelley | 19df370 | 2021-12-16 22:32:54 -0600 | [diff] [blame] | 42 | return (libhei::hash<libhei::NodeId_t>("PLL_UNLOCK") == t.getId() && |
Zane Shelley | a7369f8 | 2021-10-18 16:52:21 -0500 | [diff] [blame] | 43 | (0 == t.getBit() || 1 == t.getBit())); |
| 44 | }); |
| 45 | |
| 46 | if (i_list.end() != itr) |
| 47 | { |
| 48 | o_rootCause = *itr; |
| 49 | return true; |
| 50 | } |
| 51 | |
| 52 | return false; |
| 53 | } |
| 54 | |
| 55 | //------------------------------------------------------------------------------ |
| 56 | |
Caleb Palmer | 329dbbd | 2022-10-03 15:05:43 -0500 | [diff] [blame] | 57 | bool __findIueTh(const std::vector<libhei::Signature>& i_list, |
| 58 | libhei::Signature& o_rootCause) |
| 59 | { |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 60 | // TODO: These bit values propbably changed in Odyssey. Will need to |
| 61 | // consider flags instead of arbitrary values. |
Caleb Palmer | 329dbbd | 2022-10-03 15:05:43 -0500 | [diff] [blame] | 62 | auto itr = std::find_if(i_list.begin(), i_list.end(), [&](const auto& t) { |
| 63 | return (libhei::hash<libhei::NodeId_t>("RDFFIR") == t.getId() && |
| 64 | (17 == t.getBit() || 37 == t.getBit())); |
| 65 | }); |
| 66 | |
| 67 | if (i_list.end() != itr) |
| 68 | { |
| 69 | o_rootCause = *itr; |
| 70 | return true; |
| 71 | } |
| 72 | |
| 73 | return false; |
| 74 | } |
| 75 | |
| 76 | //------------------------------------------------------------------------------ |
| 77 | |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 78 | bool __findMemoryChannelFailure(const std::vector<libhei::Signature>& i_list, |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 79 | libhei::Signature& o_rootCause, |
| 80 | const RasDataParser& i_rasData) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 81 | { |
| 82 | using namespace util::pdbg; |
| 83 | |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 84 | using func = libhei::NodeId_t (*)(const std::string& i_str); |
Zane Shelley | 19df370 | 2021-12-16 22:32:54 -0600 | [diff] [blame] | 85 | func __hash = libhei::hash<libhei::NodeId_t>; |
| 86 | |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 87 | static const auto mc_dstl_fir = __hash("MC_DSTL_FIR"); |
| 88 | static const auto mc_ustl_fir = __hash("MC_USTL_FIR"); |
Zane Shelley | 19df370 | 2021-12-16 22:32:54 -0600 | [diff] [blame] | 89 | static const auto mc_omi_dl_err_rpt = __hash("MC_OMI_DL_ERR_RPT"); |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 90 | static const auto srqfir = __hash("SRQFIR"); |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 91 | |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 92 | // First, look for any chip checkstops from the connected OCMBs. |
| 93 | for (const auto& s : i_list) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 94 | { |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 95 | if (TYPE_OCMB != getTrgtType(getTrgt(s.getChip()))) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 96 | { |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 97 | continue; // OCMBs only |
| 98 | } |
| 99 | |
| 100 | // TODO: The chip data for Explorer chips currently report chip |
| 101 | // checkstops as unit checkstops. Once the chip data has been |
| 102 | // updated, the check for unit checkstops here will need to be |
| 103 | // removed. |
| 104 | if (libhei::ATTN_TYPE_CHIP_CS == s.getAttnType() || |
| 105 | libhei::ATTN_TYPE_UNIT_CS == s.getAttnType()) |
| 106 | { |
| 107 | // Special Case: |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 108 | // If the channel fail was specifically a firmware initiated |
| 109 | // channel fail (SRQFIR[25]) check for any IUE bits that are on |
| 110 | // that would have caused that (RDFFIR[17,37]). |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 111 | // TODO: These bit values probably changed in Odyssey. Will need to |
| 112 | // consider flags instead of arbitrary values. |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 113 | if ((srqfir == s.getId() && 25 == s.getBit()) && |
| 114 | __findIueTh(i_list, o_rootCause)) |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 115 | { |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 116 | return true; |
| 117 | } |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 118 | |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 119 | o_rootCause = s; |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 120 | return true; |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 121 | } |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 122 | } |
| 123 | |
| 124 | // Now, look for any channel failure attentions on the processor side of the |
| 125 | // memory bus. |
| 126 | for (const auto& s : i_list) |
| 127 | { |
| 128 | if (TYPE_PROC != getTrgtType(getTrgt(s.getChip()))) |
| 129 | { |
| 130 | continue; // processors only |
| 131 | } |
| 132 | |
| 133 | // Any unit checkstop attentions that originated from the MC_DSTL_FIR or |
| 134 | // MC_USTLFIR are considered a channel failure attention. |
| 135 | // TODO: The "channel failure" designation is actually configurable via |
| 136 | // other registers. We just happen to expect anything that is |
| 137 | // configured to channel failure to also be configured to unit |
| 138 | // checkstop. Eventually, we will need some mechanism to check the |
| 139 | // configuration registers for a more accurate analysis. |
| 140 | if (libhei::ATTN_TYPE_UNIT_CS == s.getAttnType() && |
| 141 | (mc_dstl_fir == s.getId() || mc_ustl_fir == s.getId()) && |
| 142 | !i_rasData.isFlagSet(s, |
| 143 | RasDataParser::RasDataFlags::ATTN_FROM_OCMB)) |
| 144 | { |
| 145 | o_rootCause = s; |
| 146 | return true; |
| 147 | } |
| 148 | // Any signatures from MC_OMI_DL_ERR_RPT feed into the only bits in |
| 149 | // MC_OMI_DL_FIR that are hardwired to channel failure. |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 150 | else if (mc_omi_dl_err_rpt == s.getId()) |
| 151 | { |
| 152 | o_rootCause = s; |
| 153 | return true; |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 154 | } |
| 155 | } |
| 156 | |
| 157 | return false; // default, nothing found |
| 158 | } |
| 159 | |
| 160 | //------------------------------------------------------------------------------ |
| 161 | |
| 162 | // Will query if a signature is a potential system checkstop root cause. |
| 163 | // attention. Note that this function excludes memory channel failure attentions |
Zane Shelley | ed3ab8f | 2022-05-24 21:08:21 -0500 | [diff] [blame] | 164 | // which are checked in __findMemoryChannelFailure(). |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 165 | bool __findCsRootCause(const libhei::Signature& i_signature, |
| 166 | const RasDataParser& i_rasData) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 167 | { |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 168 | // Check if the input signature has the CS_POSSIBLE or SUE_SOURCE flag set. |
| 169 | if (i_rasData.isFlagSet(i_signature, |
| 170 | RasDataParser::RasDataFlags::CS_POSSIBLE) || |
| 171 | i_rasData.isFlagSet(i_signature, |
| 172 | RasDataParser::RasDataFlags::SUE_SOURCE)) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 173 | { |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 174 | return true; |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 175 | } |
| 176 | |
| 177 | return false; // default, nothing found |
| 178 | } |
| 179 | |
| 180 | //------------------------------------------------------------------------------ |
| 181 | |
| 182 | bool __findCsRootCause_RE(const std::vector<libhei::Signature>& i_list, |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 183 | libhei::Signature& o_rootCause, |
| 184 | const RasDataParser& i_rasData) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 185 | { |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 186 | for (const auto& s : i_list) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 187 | { |
| 188 | // Only looking for recoverable attentions. |
| 189 | if (libhei::ATTN_TYPE_RECOVERABLE != s.getAttnType()) |
| 190 | { |
| 191 | continue; |
| 192 | } |
| 193 | |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 194 | if (__findCsRootCause(s, i_rasData)) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 195 | { |
| 196 | o_rootCause = s; |
| 197 | return true; |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | return false; // default, nothing found |
| 202 | } |
| 203 | |
| 204 | //------------------------------------------------------------------------------ |
| 205 | |
| 206 | bool __findCsRootCause_UCS(const std::vector<libhei::Signature>& i_list, |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 207 | libhei::Signature& o_rootCause, |
| 208 | const RasDataParser& i_rasData) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 209 | { |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 210 | for (const auto& s : i_list) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 211 | { |
| 212 | // Only looking for unit checkstop attentions. |
| 213 | if (libhei::ATTN_TYPE_UNIT_CS != s.getAttnType()) |
| 214 | { |
| 215 | continue; |
| 216 | } |
| 217 | |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 218 | if (__findCsRootCause(s, i_rasData)) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 219 | { |
| 220 | o_rootCause = s; |
| 221 | return true; |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | return false; // default, nothing found |
| 226 | } |
| 227 | |
| 228 | //------------------------------------------------------------------------------ |
| 229 | |
Caleb Palmer | 51f8202 | 2023-02-22 16:09:09 -0600 | [diff] [blame] | 230 | bool __findOcmbAttnBits(const std::vector<libhei::Signature>& i_list, |
| 231 | libhei::Signature& o_rootCause, |
| 232 | const RasDataParser& i_rasData) |
| 233 | { |
| 234 | using namespace util::pdbg; |
| 235 | |
| 236 | // If we have any attentions from an OCMB, assume isolation to the OCMBs |
| 237 | // was successful and the ATTN_FROM_OCMB flag does not need to be checked. |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 238 | for (const auto& s : i_list) |
Caleb Palmer | 51f8202 | 2023-02-22 16:09:09 -0600 | [diff] [blame] | 239 | { |
| 240 | if (TYPE_OCMB == getTrgtType(getTrgt(s.getChip()))) |
| 241 | { |
| 242 | return false; |
| 243 | } |
| 244 | } |
| 245 | |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 246 | for (const auto& s : i_list) |
Caleb Palmer | 51f8202 | 2023-02-22 16:09:09 -0600 | [diff] [blame] | 247 | { |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 248 | if (i_rasData.isFlagSet(s, RasDataParser::RasDataFlags::ATTN_FROM_OCMB)) |
Caleb Palmer | 51f8202 | 2023-02-22 16:09:09 -0600 | [diff] [blame] | 249 | { |
| 250 | o_rootCause = s; |
| 251 | return true; |
| 252 | } |
| 253 | } |
| 254 | |
| 255 | return false; // default, nothing found |
| 256 | } |
| 257 | |
| 258 | //------------------------------------------------------------------------------ |
| 259 | |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 260 | bool __findNonExternalCs(const std::vector<libhei::Signature>& i_list, |
| 261 | libhei::Signature& o_rootCause) |
| 262 | { |
| 263 | using namespace util::pdbg; |
| 264 | |
Zane Shelley | 19df370 | 2021-12-16 22:32:54 -0600 | [diff] [blame] | 265 | static const auto pb_ext_fir = libhei::hash<libhei::NodeId_t>("PB_EXT_FIR"); |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 266 | |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 267 | for (const auto& s : i_list) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 268 | { |
| 269 | const auto targetType = getTrgtType(getTrgt(s.getChip())); |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 270 | const auto id = s.getId(); |
| 271 | const auto attnType = s.getAttnType(); |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 272 | |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 273 | // Find any processor with chip checkstop attention that did not |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 274 | // originate from the PB_EXT_FIR. |
| 275 | if ((TYPE_PROC == targetType) && |
Zane Shelley | adda054 | 2023-04-06 16:38:02 -0500 | [diff] [blame] | 276 | (libhei::ATTN_TYPE_CHIP_CS == attnType) && (pb_ext_fir != id)) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 277 | { |
| 278 | o_rootCause = s; |
| 279 | return true; |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | return false; // default, nothing found |
| 284 | } |
| 285 | |
| 286 | //------------------------------------------------------------------------------ |
| 287 | |
Zane Shelley | baec7c0 | 2022-03-17 11:05:20 -0500 | [diff] [blame] | 288 | bool __findTiRootCause(const std::vector<libhei::Signature>& i_list, |
| 289 | libhei::Signature& o_rootCause) |
| 290 | { |
| 291 | using namespace util::pdbg; |
| 292 | |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 293 | using func = libhei::NodeId_t (*)(const std::string& i_str); |
Zane Shelley | baec7c0 | 2022-03-17 11:05:20 -0500 | [diff] [blame] | 294 | func __hash = libhei::hash<libhei::NodeId_t>; |
| 295 | |
| 296 | // PROC registers |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 297 | static const auto tp_local_fir = __hash("TP_LOCAL_FIR"); |
| 298 | static const auto occ_fir = __hash("OCC_FIR"); |
| 299 | static const auto pbao_fir = __hash("PBAO_FIR"); |
| 300 | static const auto n0_local_fir = __hash("N0_LOCAL_FIR"); |
| 301 | static const auto int_cq_fir = __hash("INT_CQ_FIR"); |
| 302 | static const auto nx_cq_fir = __hash("NX_CQ_FIR"); |
| 303 | static const auto nx_dma_eng_fir = __hash("NX_DMA_ENG_FIR"); |
| 304 | static const auto vas_fir = __hash("VAS_FIR"); |
| 305 | static const auto n1_local_fir = __hash("N1_LOCAL_FIR"); |
| 306 | static const auto mcd_fir = __hash("MCD_FIR"); |
Zane Shelley | baec7c0 | 2022-03-17 11:05:20 -0500 | [diff] [blame] | 307 | static const auto pb_station_fir_en_1 = __hash("PB_STATION_FIR_EN_1"); |
| 308 | static const auto pb_station_fir_en_2 = __hash("PB_STATION_FIR_EN_2"); |
| 309 | static const auto pb_station_fir_en_3 = __hash("PB_STATION_FIR_EN_3"); |
| 310 | static const auto pb_station_fir_en_4 = __hash("PB_STATION_FIR_EN_4"); |
| 311 | static const auto pb_station_fir_es_1 = __hash("PB_STATION_FIR_ES_1"); |
| 312 | static const auto pb_station_fir_es_2 = __hash("PB_STATION_FIR_ES_2"); |
| 313 | static const auto pb_station_fir_es_3 = __hash("PB_STATION_FIR_ES_3"); |
| 314 | static const auto pb_station_fir_es_4 = __hash("PB_STATION_FIR_ES_4"); |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 315 | static const auto pb_station_fir_eq = __hash("PB_STATION_FIR_EQ"); |
| 316 | static const auto psihb_fir = __hash("PSIHB_FIR"); |
| 317 | static const auto pbaf_fir = __hash("PBAF_FIR"); |
| 318 | static const auto lpc_fir = __hash("LPC_FIR"); |
| 319 | static const auto eq_core_fir = __hash("EQ_CORE_FIR"); |
| 320 | static const auto eq_l2_fir = __hash("EQ_L2_FIR"); |
| 321 | static const auto eq_l3_fir = __hash("EQ_L3_FIR"); |
| 322 | static const auto eq_ncu_fir = __hash("EQ_NCU_FIR"); |
| 323 | static const auto eq_local_fir = __hash("EQ_LOCAL_FIR"); |
| 324 | static const auto eq_qme_fir = __hash("EQ_QME_FIR"); |
| 325 | static const auto iohs_local_fir = __hash("IOHS_LOCAL_FIR"); |
| 326 | static const auto iohs_dlp_fir_oc = __hash("IOHS_DLP_FIR_OC"); |
| 327 | static const auto iohs_dlp_fir_smp = __hash("IOHS_DLP_FIR_SMP"); |
| 328 | static const auto mc_local_fir = __hash("MC_LOCAL_FIR"); |
| 329 | static const auto mc_fir = __hash("MC_FIR"); |
| 330 | static const auto mc_dstl_fir = __hash("MC_DSTL_FIR"); |
| 331 | static const auto mc_ustl_fir = __hash("MC_USTL_FIR"); |
| 332 | static const auto nmmu_cq_fir = __hash("NMMU_CQ_FIR"); |
| 333 | static const auto nmmu_fir = __hash("NMMU_FIR"); |
| 334 | static const auto mc_omi_dl = __hash("MC_OMI_DL"); |
| 335 | static const auto pau_local_fir = __hash("PAU_LOCAL_FIR"); |
| 336 | static const auto pau_ptl_fir = __hash("PAU_PTL_FIR"); |
| 337 | static const auto pau_phy_fir = __hash("PAU_PHY_FIR"); |
| 338 | static const auto pau_fir_0 = __hash("PAU_FIR_0"); |
| 339 | static const auto pau_fir_2 = __hash("PAU_FIR_2"); |
| 340 | static const auto pci_local_fir = __hash("PCI_LOCAL_FIR"); |
| 341 | static const auto pci_iop_fir = __hash("PCI_IOP_FIR"); |
| 342 | static const auto pci_nest_fir = __hash("PCI_NEST_FIR"); |
Zane Shelley | baec7c0 | 2022-03-17 11:05:20 -0500 | [diff] [blame] | 343 | |
| 344 | // OCMB registers |
| 345 | static const auto ocmb_lfir = __hash("OCMB_LFIR"); |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 346 | static const auto mmiofir = __hash("MMIOFIR"); |
| 347 | static const auto srqfir = __hash("SRQFIR"); |
| 348 | static const auto rdffir = __hash("RDFFIR"); |
| 349 | static const auto tlxfir = __hash("TLXFIR"); |
| 350 | static const auto omi_dl = __hash("OMI_DL"); |
Zane Shelley | baec7c0 | 2022-03-17 11:05:20 -0500 | [diff] [blame] | 351 | |
| 352 | for (const auto& signature : i_list) |
| 353 | { |
| 354 | const auto targetType = getTrgtType(getTrgt(signature.getChip())); |
Patrick Williams | 27dd636 | 2023-05-10 07:51:20 -0500 | [diff] [blame] | 355 | const auto attnType = signature.getAttnType(); |
| 356 | const auto id = signature.getId(); |
| 357 | const auto bit = signature.getBit(); |
Zane Shelley | baec7c0 | 2022-03-17 11:05:20 -0500 | [diff] [blame] | 358 | |
| 359 | // Only looking for recoverable or unit checkstop attentions. |
| 360 | if (libhei::ATTN_TYPE_RECOVERABLE != attnType && |
| 361 | libhei::ATTN_TYPE_UNIT_CS != attnType) |
| 362 | { |
| 363 | continue; |
| 364 | } |
| 365 | |
| 366 | // Ignore attentions that should not be blamed as root cause of a TI. |
| 367 | // This would include informational only FIRs or correctable errors. |
| 368 | if (TYPE_PROC == targetType) |
| 369 | { |
| 370 | if (tp_local_fir == id && |
| 371 | (0 == bit || 1 == bit || 2 == bit || 3 == bit || 4 == bit || |
| 372 | 5 == bit || 7 == bit || 8 == bit || 9 == bit || 10 == bit || |
| 373 | 11 == bit || 20 == bit || 22 == bit || 23 == bit || |
| 374 | 24 == bit || 38 == bit || 40 == bit || 41 == bit || |
| 375 | 46 == bit || 47 == bit || 48 == bit || 55 == bit || |
| 376 | 56 == bit || 57 == bit || 58 == bit || 59 == bit)) |
| 377 | { |
| 378 | continue; |
| 379 | } |
| 380 | |
| 381 | if (occ_fir == id && |
| 382 | (9 == bit || 10 == bit || 15 == bit || 20 == bit || 21 == bit || |
| 383 | 22 == bit || 23 == bit || 32 == bit || 33 == bit || |
| 384 | 34 == bit || 36 == bit || 42 == bit || 43 == bit || |
| 385 | 46 == bit || 47 == bit || 48 == bit || 51 == bit || |
| 386 | 52 == bit || 53 == bit || 54 == bit || 57 == bit)) |
| 387 | { |
| 388 | continue; |
| 389 | } |
| 390 | |
| 391 | if (pbao_fir == id && |
| 392 | (0 == bit || 1 == bit || 2 == bit || 8 == bit || 11 == bit || |
| 393 | 13 == bit || 15 == bit || 16 == bit || 17 == bit)) |
| 394 | { |
| 395 | continue; |
| 396 | } |
| 397 | |
| 398 | if ((n0_local_fir == id || n1_local_fir == id || |
| 399 | iohs_local_fir == id || mc_local_fir == id || |
| 400 | pau_local_fir == id || pci_local_fir == id) && |
| 401 | (0 == bit || 1 == bit || 2 == bit || 3 == bit || 4 == bit || |
| 402 | 5 == bit || 6 == bit || 7 == bit || 8 == bit || 9 == bit || |
| 403 | 10 == bit || 11 == bit || 20 == bit || 21 == bit)) |
| 404 | { |
| 405 | continue; |
| 406 | } |
| 407 | |
| 408 | if (int_cq_fir == id && |
| 409 | (0 == bit || 3 == bit || 5 == bit || 7 == bit || 36 == bit || |
Caleb Palmer | ecde53f | 2022-12-13 15:11:47 -0600 | [diff] [blame] | 410 | 47 == bit || 48 == bit || 49 == bit || 50 == bit || |
Zane Shelley | baec7c0 | 2022-03-17 11:05:20 -0500 | [diff] [blame] | 411 | 58 == bit || 59 == bit || 60 == bit)) |
| 412 | { |
| 413 | continue; |
| 414 | } |
| 415 | |
| 416 | if (nx_cq_fir == id && |
| 417 | (1 == bit || 4 == bit || 18 == bit || 32 == bit || 33 == bit)) |
| 418 | { |
| 419 | continue; |
| 420 | } |
| 421 | |
| 422 | if (nx_dma_eng_fir == id && |
| 423 | (4 == bit || 6 == bit || 9 == bit || 10 == bit || 11 == bit || |
| 424 | 34 == bit || 35 == bit || 36 == bit || 37 == bit || 39 == bit)) |
| 425 | { |
| 426 | continue; |
| 427 | } |
| 428 | |
| 429 | if (vas_fir == id && |
| 430 | (8 == bit || 9 == bit || 11 == bit || 12 == bit || 13 == bit)) |
| 431 | { |
| 432 | continue; |
| 433 | } |
| 434 | |
| 435 | if (mcd_fir == id && (0 == bit)) |
| 436 | { |
| 437 | continue; |
| 438 | } |
| 439 | |
| 440 | if ((pb_station_fir_en_1 == id || pb_station_fir_en_2 == id || |
| 441 | pb_station_fir_en_3 == id || pb_station_fir_en_4 == id || |
| 442 | pb_station_fir_es_1 == id || pb_station_fir_es_2 == id || |
| 443 | pb_station_fir_es_3 == id || pb_station_fir_es_4 == id || |
| 444 | pb_station_fir_eq == id) && |
| 445 | (9 == bit)) |
| 446 | { |
| 447 | continue; |
| 448 | } |
| 449 | |
| 450 | if (psihb_fir == id && (0 == bit || 23 == bit)) |
| 451 | { |
| 452 | continue; |
| 453 | } |
| 454 | |
| 455 | if (pbaf_fir == id && |
| 456 | (0 == bit || 1 == bit || 3 == bit || 4 == bit || 5 == bit || |
| 457 | 6 == bit || 7 == bit || 8 == bit || 9 == bit || 10 == bit || |
| 458 | 11 == bit || 19 == bit || 20 == bit || 21 == bit || |
| 459 | 28 == bit || 29 == bit || 30 == bit || 31 == bit || |
| 460 | 32 == bit || 33 == bit || 34 == bit || 35 == bit || 36 == bit)) |
| 461 | { |
| 462 | continue; |
| 463 | } |
| 464 | |
| 465 | if (lpc_fir == id && (5 == bit)) |
| 466 | { |
| 467 | continue; |
| 468 | } |
| 469 | |
| 470 | if (eq_core_fir == id && |
| 471 | (0 == bit || 2 == bit || 4 == bit || 7 == bit || 9 == bit || |
| 472 | 11 == bit || 13 == bit || 18 == bit || 21 == bit || |
| 473 | 24 == bit || 29 == bit || 31 == bit || 37 == bit || |
| 474 | 43 == bit || 56 == bit || 57 == bit)) |
| 475 | { |
| 476 | continue; |
| 477 | } |
| 478 | |
| 479 | if (eq_l2_fir == id && |
| 480 | (0 == bit || 6 == bit || 11 == bit || 19 == bit || 36 == bit)) |
| 481 | { |
| 482 | continue; |
| 483 | } |
| 484 | |
| 485 | if (eq_l3_fir == id && |
| 486 | (3 == bit || 4 == bit || 7 == bit || 10 == bit || 13 == bit)) |
| 487 | { |
| 488 | continue; |
| 489 | } |
| 490 | |
| 491 | if (eq_ncu_fir == id && (9 == bit)) |
| 492 | { |
| 493 | continue; |
| 494 | } |
| 495 | |
| 496 | if (eq_local_fir == id && |
| 497 | (0 == bit || 1 == bit || 2 == bit || 3 == bit || 5 == bit || |
| 498 | 6 == bit || 7 == bit || 8 == bit || 9 == bit || 10 == bit || |
| 499 | 11 == bit || 12 == bit || 13 == bit || 14 == bit || |
| 500 | 15 == bit || 16 == bit || 20 == bit || 21 == bit || |
| 501 | 22 == bit || 23 == bit || 24 == bit || 25 == bit || |
| 502 | 26 == bit || 27 == bit || 28 == bit || 29 == bit || |
| 503 | 30 == bit || 31 == bit || 32 == bit || 33 == bit || |
| 504 | 34 == bit || 35 == bit || 36 == bit || 37 == bit || |
| 505 | 38 == bit || 39 == bit)) |
| 506 | { |
| 507 | continue; |
| 508 | } |
| 509 | |
| 510 | if (eq_qme_fir == id && (7 == bit || 25 == bit)) |
| 511 | { |
| 512 | continue; |
| 513 | } |
| 514 | |
| 515 | if (iohs_dlp_fir_oc == id && |
| 516 | (6 == bit || 7 == bit || 8 == bit || 9 == bit || 10 == bit || |
| 517 | 48 == bit || 49 == bit || 52 == bit || 53 == bit)) |
| 518 | { |
| 519 | continue; |
| 520 | } |
| 521 | |
| 522 | if (iohs_dlp_fir_smp == id && |
| 523 | (6 == bit || 7 == bit || 14 == bit || 15 == bit || 16 == bit || |
| 524 | 17 == bit || 38 == bit || 39 == bit || 44 == bit || |
| 525 | 45 == bit || 50 == bit || 51 == bit)) |
| 526 | { |
| 527 | continue; |
| 528 | } |
| 529 | |
| 530 | if (mc_fir == id && |
| 531 | (5 == bit || 8 == bit || 15 == bit || 16 == bit)) |
| 532 | { |
| 533 | continue; |
| 534 | } |
| 535 | |
| 536 | if (mc_dstl_fir == id && |
| 537 | (0 == bit || 1 == bit || 2 == bit || 3 == bit || 4 == bit || |
| 538 | 5 == bit || 6 == bit || 7 == bit || 14 == bit || 15 == bit)) |
| 539 | { |
| 540 | continue; |
| 541 | } |
| 542 | |
| 543 | if (mc_ustl_fir == id && |
| 544 | (6 == bit || 20 == bit || 33 == bit || 34 == bit)) |
| 545 | { |
| 546 | continue; |
| 547 | } |
| 548 | |
| 549 | if (nmmu_cq_fir == id && (8 == bit || 11 == bit || 14 == bit)) |
| 550 | { |
| 551 | continue; |
| 552 | } |
| 553 | |
| 554 | if (nmmu_fir == id && |
| 555 | (0 == bit || 3 == bit || 8 == bit || 9 == bit || 10 == bit || |
| 556 | 11 == bit || 12 == bit || 13 == bit || 14 == bit || |
| 557 | 15 == bit || 30 == bit || 31 == bit || 41 == bit)) |
| 558 | { |
| 559 | continue; |
| 560 | } |
| 561 | |
| 562 | if (mc_omi_dl == id && (2 == bit || 3 == bit || 6 == bit || |
| 563 | 7 == bit || 9 == bit || 10 == bit)) |
| 564 | { |
| 565 | continue; |
| 566 | } |
| 567 | |
| 568 | if (pau_ptl_fir == id && (5 == bit || 9 == bit)) |
| 569 | { |
| 570 | continue; |
| 571 | } |
| 572 | |
| 573 | if (pau_phy_fir == id && |
| 574 | (2 == bit || 3 == bit || 6 == bit || 7 == bit || 15 == bit)) |
| 575 | { |
| 576 | continue; |
| 577 | } |
| 578 | |
| 579 | if (pau_fir_0 == id && (13 == bit || 30 == bit || 41 == bit)) |
| 580 | { |
| 581 | continue; |
| 582 | } |
| 583 | |
| 584 | if (pau_fir_2 == id && (19 == bit || 46 == bit || 49 == bit)) |
| 585 | { |
| 586 | continue; |
| 587 | } |
| 588 | |
| 589 | if (pci_iop_fir == id && |
| 590 | (0 == bit || 2 == bit || 4 == bit || 6 == bit || 7 == bit || |
| 591 | 8 == bit || 10 == bit)) |
| 592 | { |
| 593 | continue; |
| 594 | } |
| 595 | |
| 596 | if (pci_nest_fir == id && (2 == bit || 5 == bit)) |
| 597 | { |
| 598 | continue; |
| 599 | } |
| 600 | } |
| 601 | else if (TYPE_OCMB == targetType) |
| 602 | { |
| 603 | if (ocmb_lfir == id && |
| 604 | (0 == bit || 1 == bit || 2 == bit || 8 == bit || 23 == bit || |
| 605 | 37 == bit || 63 == bit)) |
| 606 | { |
| 607 | continue; |
| 608 | } |
| 609 | |
| 610 | if (mmiofir == id && (2 == bit)) |
| 611 | { |
| 612 | continue; |
| 613 | } |
| 614 | |
| 615 | if (srqfir == id && |
| 616 | (2 == bit || 4 == bit || 14 == bit || 15 == bit || 23 == bit || |
| 617 | 25 == bit || 28 == bit)) |
| 618 | { |
| 619 | continue; |
| 620 | } |
| 621 | |
| 622 | if (rdffir == id && |
| 623 | (0 == bit || 1 == bit || 2 == bit || 3 == bit || 4 == bit || |
| 624 | 5 == bit || 6 == bit || 7 == bit || 8 == bit || 9 == bit || |
| 625 | 18 == bit || 38 == bit || 40 == bit || 41 == bit || |
| 626 | 45 == bit || 46 == bit)) |
| 627 | { |
| 628 | continue; |
| 629 | } |
| 630 | |
| 631 | if (tlxfir == id && (0 == bit || 9 == bit || 26 == bit)) |
| 632 | { |
| 633 | continue; |
| 634 | } |
| 635 | |
| 636 | if (omi_dl == id && (2 == bit || 3 == bit || 6 == bit || 7 == bit || |
| 637 | 9 == bit || 10 == bit)) |
| 638 | { |
| 639 | continue; |
| 640 | } |
| 641 | } |
| 642 | |
| 643 | // At this point, the attention has not been explicitly ignored. So |
| 644 | // return this signature and exit. |
| 645 | o_rootCause = signature; |
| 646 | return true; |
| 647 | } |
| 648 | |
| 649 | return false; // default, nothing found |
| 650 | } |
| 651 | |
| 652 | //------------------------------------------------------------------------------ |
| 653 | |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 654 | bool filterRootCause(AnalysisType i_type, |
| 655 | const libhei::IsolationData& i_isoData, |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 656 | libhei::Signature& o_rootCause, |
| 657 | const RasDataParser& i_rasData) |
Zane Shelley | 65fefb2 | 2021-10-18 15:35:26 -0500 | [diff] [blame] | 658 | { |
| 659 | // We'll need to make a copy of the list so that the original list is |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 660 | // maintained for the PEL. |
Zane Shelley | 65fefb2 | 2021-10-18 15:35:26 -0500 | [diff] [blame] | 661 | std::vector<libhei::Signature> list{i_isoData.getSignatureList()}; |
| 662 | |
| 663 | // START WORKAROUND |
| 664 | // TODO: Filtering should be data driven. Until that support is available, |
| 665 | // use the following isolation rules. |
| 666 | |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 667 | // Ensure the list is not empty before continuing. |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 668 | if (list.empty()) |
| 669 | { |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 670 | return false; // nothing more to do |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 671 | } |
| 672 | |
| 673 | // First, look for any RCS OSC errors. This must always be first because |
| 674 | // they can cause downstream PLL unlock attentions. |
| 675 | if (__findRcsOscError(list, o_rootCause)) |
Zane Shelley | a7369f8 | 2021-10-18 16:52:21 -0500 | [diff] [blame] | 676 | { |
| 677 | return true; |
| 678 | } |
| 679 | |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 680 | // Second, look for any PLL unlock attentions. This must always be second |
| 681 | // because PLL unlock attentions can cause any number of downstream |
| 682 | // attentions, including a system checkstop. |
| 683 | if (__findPllUnlock(list, o_rootCause)) |
| 684 | { |
| 685 | return true; |
| 686 | } |
| 687 | |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 688 | // Regardless of the analysis type, always look for anything that could be |
| 689 | // blamed as the root cause of a system checkstop. |
| 690 | |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 691 | // Memory channel failure attentions will produce SUEs and likely cause |
| 692 | // downstream attentions, including a system checkstop. |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 693 | if (__findMemoryChannelFailure(list, o_rootCause, i_rasData)) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 694 | { |
| 695 | return true; |
| 696 | } |
| 697 | |
| 698 | // Look for any recoverable attentions that have been identified as a |
| 699 | // potential root cause of a system checkstop attention. These would include |
| 700 | // any attention that would generate an SUE. Note that is it possible for |
| 701 | // recoverables to generate unit checkstop attentions so we must check them |
| 702 | // first. |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 703 | if (__findCsRootCause_RE(list, o_rootCause, i_rasData)) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 704 | { |
| 705 | return true; |
| 706 | } |
| 707 | |
| 708 | // Look for any unit checkstop attentions (other than memory channel |
| 709 | // failures) that have been identified as a potential root cause of a |
| 710 | // system checkstop attention. These would include any attention that would |
| 711 | // generate an SUE. |
Caleb Palmer | 1a4f0e7 | 2022-11-07 15:08:01 -0600 | [diff] [blame] | 712 | if (__findCsRootCause_UCS(list, o_rootCause, i_rasData)) |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 713 | { |
| 714 | return true; |
| 715 | } |
| 716 | |
Zane Shelley | 93b001c | 2023-03-24 17:45:04 -0500 | [diff] [blame] | 717 | // If no other viable root cause has been found, check for any signatures |
| 718 | // with the ATTN_FROM_OCMB flag in case there was an attention from an |
| 719 | // inaccessible OCMB. |
Caleb Palmer | 51f8202 | 2023-02-22 16:09:09 -0600 | [diff] [blame] | 720 | if (__findOcmbAttnBits(list, o_rootCause, i_rasData)) |
| 721 | { |
| 722 | return true; |
| 723 | } |
| 724 | |
Zane Shelley | f4792d6 | 2021-10-28 18:08:22 -0500 | [diff] [blame] | 725 | // Look for any system checkstop attentions that originated from within the |
| 726 | // chip that reported the attention. In other words, no external checkstop |
| 727 | // attentions. |
| 728 | if (__findNonExternalCs(list, o_rootCause)) |
| 729 | { |
| 730 | return true; |
| 731 | } |
| 732 | |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 733 | if (AnalysisType::SYSTEM_CHECKSTOP != i_type) |
Zane Shelley | 65fefb2 | 2021-10-18 15:35:26 -0500 | [diff] [blame] | 734 | { |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 735 | // No system checkstop root cause attentions were found. Next, look for |
| 736 | // any recoverable or unit checkstop attentions that could be associated |
Zane Shelley | baec7c0 | 2022-03-17 11:05:20 -0500 | [diff] [blame] | 737 | // with a TI. |
| 738 | if (__findTiRootCause(list, o_rootCause)) |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 739 | { |
Zane Shelley | ec227c2 | 2021-12-09 15:54:40 -0600 | [diff] [blame] | 740 | return true; |
| 741 | } |
| 742 | |
| 743 | if (AnalysisType::TERMINATE_IMMEDIATE != i_type) |
| 744 | { |
| 745 | // No attentions associated with a system checkstop or TI were |
| 746 | // found. Simply, return the first entry in the list. |
| 747 | o_rootCause = list.front(); |
| 748 | return true; |
| 749 | } |
Zane Shelley | 65fefb2 | 2021-10-18 15:35:26 -0500 | [diff] [blame] | 750 | } |
| 751 | |
| 752 | // END WORKAROUND |
| 753 | |
| 754 | return false; // default, no active attentions found. |
| 755 | } |
| 756 | |
| 757 | //------------------------------------------------------------------------------ |
| 758 | |
| 759 | } // namespace analyzer |