| 1 | // SPDX-License-Identifier: MIT |
| 2 | /* |
| 3 | * Copyright © 2014-2019 Intel Corporation |
| 4 | * |
| 5 | * Authors: |
| 6 | * Vinit Azad <vinit.azad@intel.com> |
| 7 | * Ben Widawsky <ben@bwidawsk.net> |
| 8 | * Dave Gordon <david.s.gordon@intel.com> |
| 9 | * Alex Dai <yu.dai@intel.com> |
| 10 | */ |
| 11 | |
| 12 | #include "gt/intel_gt.h" |
| 13 | #include "gt/intel_gt_mcr.h" |
| 14 | #include "gt/intel_gt_regs.h" |
| 15 | #include "gt/intel_rps.h" |
| 16 | |
| 17 | #include "i915_drv.h" |
| 18 | #include "i915_wait_util.h" |
| 19 | #include "intel_guc_fw.h" |
| 20 | #include "intel_guc_print.h" |
| 21 | |
| 22 | static void guc_prepare_xfer(struct intel_gt *gt) |
| 23 | { |
| 24 | struct intel_uncore *uncore = gt->uncore; |
| 25 | |
| 26 | u32 shim_flags = GUC_ENABLE_READ_CACHE_LOGIC | |
| 27 | GUC_ENABLE_READ_CACHE_FOR_SRAM_DATA | |
| 28 | GUC_ENABLE_READ_CACHE_FOR_WOPCM_DATA | |
| 29 | GUC_ENABLE_MIA_CLOCK_GATING; |
| 30 | |
| 31 | if (GRAPHICS_VER_FULL(uncore->i915) < IP_VER(12, 55)) |
| 32 | shim_flags |= GUC_DISABLE_SRAM_INIT_TO_ZEROES | |
| 33 | GUC_ENABLE_MIA_CACHING; |
| 34 | |
| 35 | /* Must program this register before loading the ucode with DMA */ |
| 36 | intel_uncore_write(uncore, GUC_SHIM_CONTROL, val: shim_flags); |
| 37 | |
| 38 | if (IS_GEN9_LP(uncore->i915)) |
| 39 | intel_uncore_write(uncore, GEN9LP_GT_PM_CONFIG, GT_DOORBELL_ENABLE); |
| 40 | else |
| 41 | intel_uncore_write(uncore, GEN9_GT_PM_CONFIG, GT_DOORBELL_ENABLE); |
| 42 | |
| 43 | if (GRAPHICS_VER(uncore->i915) == 9) { |
| 44 | /* DOP Clock Gating Enable for GuC clocks */ |
| 45 | intel_uncore_rmw(uncore, GEN7_MISCCPCTL, clear: 0, |
| 46 | GEN8_DOP_CLOCK_GATE_GUC_ENABLE); |
| 47 | |
| 48 | /* allows for 5us (in 10ns units) before GT can go to RC6 */ |
| 49 | intel_uncore_write(uncore, GUC_ARAT_C6DIS, val: 0x1FF); |
| 50 | } |
| 51 | |
| 52 | /* |
| 53 | * Starting from IP 12.50 we need to enable the mirroring of GuC |
| 54 | * internal state to debug registers. This is always enabled on previous |
| 55 | * IPs. |
| 56 | */ |
| 57 | if (GRAPHICS_VER_FULL(uncore->i915) >= IP_VER(12, 50)) |
| 58 | intel_uncore_rmw(uncore, GUC_SHIM_CONTROL2, clear: 0, GUC_ENABLE_DEBUG_REG); |
| 59 | } |
| 60 | |
| 61 | static int guc_xfer_rsa_mmio(struct intel_uc_fw *guc_fw, |
| 62 | struct intel_uncore *uncore) |
| 63 | { |
| 64 | u32 rsa[UOS_RSA_SCRATCH_COUNT]; |
| 65 | size_t copied; |
| 66 | int i; |
| 67 | |
| 68 | copied = intel_uc_fw_copy_rsa(uc_fw: guc_fw, dst: rsa, max_len: sizeof(rsa)); |
| 69 | if (copied < sizeof(rsa)) |
| 70 | return -ENOMEM; |
| 71 | |
| 72 | for (i = 0; i < UOS_RSA_SCRATCH_COUNT; i++) |
| 73 | intel_uncore_write(uncore, UOS_RSA_SCRATCH(i), val: rsa[i]); |
| 74 | |
| 75 | return 0; |
| 76 | } |
| 77 | |
| 78 | static int guc_xfer_rsa_vma(struct intel_uc_fw *guc_fw, |
| 79 | struct intel_uncore *uncore) |
| 80 | { |
| 81 | struct intel_guc *guc = container_of(guc_fw, struct intel_guc, fw); |
| 82 | |
| 83 | intel_uncore_write(uncore, UOS_RSA_SCRATCH(0), |
| 84 | val: intel_guc_ggtt_offset(guc, vma: guc_fw->rsa_data)); |
| 85 | |
| 86 | return 0; |
| 87 | } |
| 88 | |
| 89 | /* Copy RSA signature from the fw image to HW for verification */ |
| 90 | static int guc_xfer_rsa(struct intel_uc_fw *guc_fw, |
| 91 | struct intel_uncore *uncore) |
| 92 | { |
| 93 | if (guc_fw->rsa_data) |
| 94 | return guc_xfer_rsa_vma(guc_fw, uncore); |
| 95 | else |
| 96 | return guc_xfer_rsa_mmio(guc_fw, uncore); |
| 97 | } |
| 98 | |
| 99 | /* |
| 100 | * Read the GuC status register (GUC_STATUS) and store it in the |
| 101 | * specified location; then return a boolean indicating whether |
| 102 | * the value matches either completion or a known failure code. |
| 103 | * |
| 104 | * This is used for polling the GuC status in a wait_for() |
| 105 | * loop below. |
| 106 | */ |
| 107 | static inline bool guc_load_done(struct intel_uncore *uncore, u32 *status, bool *success) |
| 108 | { |
| 109 | u32 val = intel_uncore_read(uncore, GUC_STATUS); |
| 110 | u32 uk_val = REG_FIELD_GET(GS_UKERNEL_MASK, val); |
| 111 | u32 br_val = REG_FIELD_GET(GS_BOOTROM_MASK, val); |
| 112 | |
| 113 | *status = val; |
| 114 | switch (uk_val) { |
| 115 | case INTEL_GUC_LOAD_STATUS_READY: |
| 116 | *success = true; |
| 117 | return true; |
| 118 | |
| 119 | case INTEL_GUC_LOAD_STATUS_ERROR_DEVID_BUILD_MISMATCH: |
| 120 | case INTEL_GUC_LOAD_STATUS_GUC_PREPROD_BUILD_MISMATCH: |
| 121 | case INTEL_GUC_LOAD_STATUS_ERROR_DEVID_INVALID_GUCTYPE: |
| 122 | case INTEL_GUC_LOAD_STATUS_HWCONFIG_ERROR: |
| 123 | case INTEL_GUC_LOAD_STATUS_DPC_ERROR: |
| 124 | case INTEL_GUC_LOAD_STATUS_EXCEPTION: |
| 125 | case INTEL_GUC_LOAD_STATUS_INIT_DATA_INVALID: |
| 126 | case INTEL_GUC_LOAD_STATUS_MPU_DATA_INVALID: |
| 127 | case INTEL_GUC_LOAD_STATUS_INIT_MMIO_SAVE_RESTORE_INVALID: |
| 128 | case INTEL_GUC_LOAD_STATUS_KLV_WORKAROUND_INIT_ERROR: |
| 129 | *success = false; |
| 130 | return true; |
| 131 | } |
| 132 | |
| 133 | switch (br_val) { |
| 134 | case INTEL_BOOTROM_STATUS_NO_KEY_FOUND: |
| 135 | case INTEL_BOOTROM_STATUS_RSA_FAILED: |
| 136 | case INTEL_BOOTROM_STATUS_PAVPC_FAILED: |
| 137 | case INTEL_BOOTROM_STATUS_WOPCM_FAILED: |
| 138 | case INTEL_BOOTROM_STATUS_LOADLOC_FAILED: |
| 139 | case INTEL_BOOTROM_STATUS_JUMP_FAILED: |
| 140 | case INTEL_BOOTROM_STATUS_RC6CTXCONFIG_FAILED: |
| 141 | case INTEL_BOOTROM_STATUS_MPUMAP_INCORRECT: |
| 142 | case INTEL_BOOTROM_STATUS_EXCEPTION: |
| 143 | case INTEL_BOOTROM_STATUS_PROD_KEY_CHECK_FAILURE: |
| 144 | *success = false; |
| 145 | return true; |
| 146 | } |
| 147 | |
| 148 | return false; |
| 149 | } |
| 150 | |
| 151 | /* |
| 152 | * Use a longer timeout for debug builds so that problems can be detected |
| 153 | * and analysed. But a shorter timeout for releases so that user's don't |
| 154 | * wait forever to find out there is a problem. Note that the only reason |
| 155 | * an end user should hit the timeout is in case of extreme thermal throttling. |
| 156 | * And a system that is that hot during boot is probably dead anyway! |
| 157 | */ |
| 158 | #if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM) |
| 159 | #define GUC_LOAD_RETRY_LIMIT 20 |
| 160 | #else |
| 161 | #define GUC_LOAD_RETRY_LIMIT 3 |
| 162 | #endif |
| 163 | |
| 164 | static int guc_wait_ucode(struct intel_guc *guc) |
| 165 | { |
| 166 | struct intel_gt *gt = guc_to_gt(guc); |
| 167 | struct intel_uncore *uncore = gt->uncore; |
| 168 | ktime_t before, after, delta; |
| 169 | bool success; |
| 170 | u32 status; |
| 171 | int ret, count; |
| 172 | u64 delta_ms; |
| 173 | u32 before_freq; |
| 174 | |
| 175 | /* |
| 176 | * Wait for the GuC to start up. |
| 177 | * |
| 178 | * Measurements indicate this should take no more than 20ms |
| 179 | * (assuming the GT clock is at maximum frequency). So, a |
| 180 | * timeout here indicates that the GuC has failed and is unusable. |
| 181 | * (Higher levels of the driver may decide to reset the GuC and |
| 182 | * attempt the ucode load again if this happens.) |
| 183 | * |
| 184 | * FIXME: There is a known (but exceedingly unlikely) race condition |
| 185 | * where the asynchronous frequency management code could reduce |
| 186 | * the GT clock while a GuC reload is in progress (during a full |
| 187 | * GT reset). A fix is in progress but there are complex locking |
| 188 | * issues to be resolved. In the meantime bump the timeout to |
| 189 | * 200ms. Even at slowest clock, this should be sufficient. And |
| 190 | * in the working case, a larger timeout makes no difference. |
| 191 | * |
| 192 | * IFWI updates have also been seen to cause sporadic failures due to |
| 193 | * the requested frequency not being granted and thus the firmware |
| 194 | * load is attempted at minimum frequency. That can lead to load times |
| 195 | * in the seconds range. However, there is a limit on how long an |
| 196 | * individual wait_for() can wait. So wrap it in a loop. |
| 197 | */ |
| 198 | before_freq = intel_rps_read_actual_frequency(rps: >->rps); |
| 199 | before = ktime_get(); |
| 200 | for (count = 0; count < GUC_LOAD_RETRY_LIMIT; count++) { |
| 201 | ret = wait_for(guc_load_done(uncore, &status, &success), 1000); |
| 202 | if (!ret || !success) |
| 203 | break; |
| 204 | |
| 205 | guc_dbg(guc, "load still in progress, count = %d, freq = %dMHz, status = 0x%08X [0x%02X/%02X]\n" , |
| 206 | count, intel_rps_read_actual_frequency(>->rps), status, |
| 207 | REG_FIELD_GET(GS_BOOTROM_MASK, status), |
| 208 | REG_FIELD_GET(GS_UKERNEL_MASK, status)); |
| 209 | } |
| 210 | after = ktime_get(); |
| 211 | delta = ktime_sub(after, before); |
| 212 | delta_ms = ktime_to_ms(kt: delta); |
| 213 | if (ret || !success) { |
| 214 | u32 ukernel = REG_FIELD_GET(GS_UKERNEL_MASK, status); |
| 215 | u32 bootrom = REG_FIELD_GET(GS_BOOTROM_MASK, status); |
| 216 | |
| 217 | guc_info(guc, "load failed: status = 0x%08X, time = %lldms, freq = %dMHz, ret = %d\n" , |
| 218 | status, delta_ms, intel_rps_read_actual_frequency(>->rps), ret); |
| 219 | guc_info(guc, "load failed: status: Reset = %d, BootROM = 0x%02X, UKernel = 0x%02X, MIA = 0x%02X, Auth = 0x%02X\n" , |
| 220 | REG_FIELD_GET(GS_MIA_IN_RESET, status), |
| 221 | bootrom, ukernel, |
| 222 | REG_FIELD_GET(GS_MIA_MASK, status), |
| 223 | REG_FIELD_GET(GS_AUTH_STATUS_MASK, status)); |
| 224 | |
| 225 | switch (bootrom) { |
| 226 | case INTEL_BOOTROM_STATUS_NO_KEY_FOUND: |
| 227 | guc_info(guc, "invalid key requested, header = 0x%08X\n" , |
| 228 | intel_uncore_read(uncore, GUC_HEADER_INFO)); |
| 229 | ret = -ENOEXEC; |
| 230 | break; |
| 231 | |
| 232 | case INTEL_BOOTROM_STATUS_RSA_FAILED: |
| 233 | guc_info(guc, "firmware signature verification failed\n" ); |
| 234 | ret = -ENOEXEC; |
| 235 | break; |
| 236 | |
| 237 | case INTEL_BOOTROM_STATUS_PROD_KEY_CHECK_FAILURE: |
| 238 | guc_info(guc, "firmware production part check failure\n" ); |
| 239 | ret = -ENOEXEC; |
| 240 | break; |
| 241 | } |
| 242 | |
| 243 | switch (ukernel) { |
| 244 | case INTEL_GUC_LOAD_STATUS_EXCEPTION: |
| 245 | guc_info(guc, "firmware exception. EIP: %#x\n" , |
| 246 | intel_uncore_read(uncore, SOFT_SCRATCH(13))); |
| 247 | ret = -ENXIO; |
| 248 | break; |
| 249 | |
| 250 | case INTEL_GUC_LOAD_STATUS_INIT_MMIO_SAVE_RESTORE_INVALID: |
| 251 | guc_info(guc, "illegal register in save/restore workaround list\n" ); |
| 252 | ret = -EPERM; |
| 253 | break; |
| 254 | |
| 255 | case INTEL_GUC_LOAD_STATUS_KLV_WORKAROUND_INIT_ERROR: |
| 256 | guc_info(guc, "invalid w/a KLV entry\n" ); |
| 257 | ret = -EINVAL; |
| 258 | break; |
| 259 | |
| 260 | case INTEL_GUC_LOAD_STATUS_HWCONFIG_START: |
| 261 | guc_info(guc, "still extracting hwconfig table.\n" ); |
| 262 | ret = -ETIMEDOUT; |
| 263 | break; |
| 264 | } |
| 265 | |
| 266 | /* Uncommon/unexpected error, see earlier status code print for details */ |
| 267 | if (ret == 0) |
| 268 | ret = -ENXIO; |
| 269 | } else if (delta_ms > 200) { |
| 270 | guc_warn(guc, "excessive init time: %lldms! [status = 0x%08X, count = %d, ret = %d]\n" , |
| 271 | delta_ms, status, count, ret); |
| 272 | guc_warn(guc, "excessive init time: [freq = %dMHz -> %dMHz vs %dMHz, perf_limit_reasons = 0x%08X]\n" , |
| 273 | before_freq, intel_rps_read_actual_frequency(>->rps), |
| 274 | intel_rps_get_requested_frequency(>->rps), |
| 275 | intel_uncore_read(uncore, intel_gt_perf_limit_reasons_reg(gt))); |
| 276 | } else { |
| 277 | guc_dbg(guc, "init took %lldms, freq = %dMHz -> %dMHz vs %dMHz, status = 0x%08X, count = %d, ret = %d\n" , |
| 278 | delta_ms, before_freq, intel_rps_read_actual_frequency(>->rps), |
| 279 | intel_rps_get_requested_frequency(>->rps), status, count, ret); |
| 280 | } |
| 281 | |
| 282 | return ret; |
| 283 | } |
| 284 | |
| 285 | /** |
| 286 | * intel_guc_fw_upload() - load GuC uCode to device |
| 287 | * @guc: intel_guc structure |
| 288 | * |
| 289 | * Called from intel_uc_init_hw() during driver load, resume from sleep and |
| 290 | * after a GPU reset. |
| 291 | * |
| 292 | * The firmware image should have already been fetched into memory, so only |
| 293 | * check that fetch succeeded, and then transfer the image to the h/w. |
| 294 | * |
| 295 | * Return: non-zero code on error |
| 296 | */ |
| 297 | int intel_guc_fw_upload(struct intel_guc *guc) |
| 298 | { |
| 299 | struct intel_gt *gt = guc_to_gt(guc); |
| 300 | struct intel_uncore *uncore = gt->uncore; |
| 301 | int ret; |
| 302 | |
| 303 | guc_prepare_xfer(gt); |
| 304 | |
| 305 | /* |
| 306 | * Note that GuC needs the CSS header plus uKernel code to be copied |
| 307 | * by the DMA engine in one operation, whereas the RSA signature is |
| 308 | * loaded separately, either by copying it to the UOS_RSA_SCRATCH |
| 309 | * register (if key size <= 256) or through a ggtt-pinned vma (if key |
| 310 | * size > 256). The RSA size and therefore the way we provide it to the |
| 311 | * HW is fixed for each platform and hard-coded in the bootrom. |
| 312 | */ |
| 313 | ret = guc_xfer_rsa(guc_fw: &guc->fw, uncore); |
| 314 | if (ret) |
| 315 | goto out; |
| 316 | |
| 317 | /* |
| 318 | * Current uCode expects the code to be loaded at 8k; locations below |
| 319 | * this are used for the stack. |
| 320 | */ |
| 321 | ret = intel_uc_fw_upload(uc_fw: &guc->fw, offset: 0x2000, UOS_MOVE); |
| 322 | if (ret) |
| 323 | goto out; |
| 324 | |
| 325 | ret = guc_wait_ucode(guc); |
| 326 | if (ret) |
| 327 | goto out; |
| 328 | |
| 329 | intel_uc_fw_change_status(uc_fw: &guc->fw, status: INTEL_UC_FIRMWARE_RUNNING); |
| 330 | return 0; |
| 331 | |
| 332 | out: |
| 333 | intel_uc_fw_change_status(uc_fw: &guc->fw, status: INTEL_UC_FIRMWARE_LOAD_FAIL); |
| 334 | return ret; |
| 335 | } |
| 336 | |