| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * Copyright (C) 2013 Imagination Technologies |
| 4 | * Author: Paul Burton <paul.burton@mips.com> |
| 5 | */ |
| 6 | |
| 7 | #include <linux/cpu.h> |
| 8 | #include <linux/delay.h> |
| 9 | #include <linux/io.h> |
| 10 | #include <linux/memblock.h> |
| 11 | #include <linux/sched/task_stack.h> |
| 12 | #include <linux/sched/hotplug.h> |
| 13 | #include <linux/slab.h> |
| 14 | #include <linux/smp.h> |
| 15 | #include <linux/types.h> |
| 16 | #include <linux/irq.h> |
| 17 | |
| 18 | #include <asm/bcache.h> |
| 19 | #include <asm/mips-cps.h> |
| 20 | #include <asm/mips_mt.h> |
| 21 | #include <asm/mipsregs.h> |
| 22 | #include <asm/pm-cps.h> |
| 23 | #include <asm/r4kcache.h> |
| 24 | #include <asm/regdef.h> |
| 25 | #include <asm/smp.h> |
| 26 | #include <asm/smp-cps.h> |
| 27 | #include <asm/time.h> |
| 28 | #include <asm/uasm.h> |
| 29 | |
| 30 | #define BEV_VEC_SIZE 0x500 |
| 31 | #define BEV_VEC_ALIGN 0x1000 |
| 32 | |
| 33 | enum label_id { |
| 34 | label_not_nmi = 1, |
| 35 | }; |
| 36 | |
| 37 | UASM_L_LA(_not_nmi) |
| 38 | |
| 39 | static u64 core_entry_reg; |
| 40 | static phys_addr_t cps_vec_pa; |
| 41 | |
| 42 | struct cluster_boot_config *mips_cps_cluster_bootcfg; |
| 43 | |
| 44 | static void power_up_other_cluster(unsigned int cluster) |
| 45 | { |
| 46 | u32 stat, seq_state; |
| 47 | unsigned int timeout; |
| 48 | |
| 49 | mips_cm_lock_other(cluster, CM_GCR_Cx_OTHER_CORE_CM, 0, |
| 50 | CM_GCR_Cx_OTHER_BLOCK_LOCAL); |
| 51 | stat = read_cpc_co_stat_conf(); |
| 52 | mips_cm_unlock_other(); |
| 53 | |
| 54 | seq_state = stat & CPC_Cx_STAT_CONF_SEQSTATE; |
| 55 | seq_state >>= __ffs(CPC_Cx_STAT_CONF_SEQSTATE); |
| 56 | if (seq_state == CPC_Cx_STAT_CONF_SEQSTATE_U5) |
| 57 | return; |
| 58 | |
| 59 | /* Set endianness & power up the CM */ |
| 60 | mips_cm_lock_other(cluster, 0, 0, CM_GCR_Cx_OTHER_BLOCK_GLOBAL); |
| 61 | write_cpc_redir_sys_config(IS_ENABLED(CONFIG_CPU_BIG_ENDIAN)); |
| 62 | write_cpc_redir_pwrup_ctl(1); |
| 63 | mips_cm_unlock_other(); |
| 64 | |
| 65 | /* Wait for the CM to start up */ |
| 66 | timeout = 1000; |
| 67 | mips_cm_lock_other(cluster, CM_GCR_Cx_OTHER_CORE_CM, 0, |
| 68 | CM_GCR_Cx_OTHER_BLOCK_LOCAL); |
| 69 | while (1) { |
| 70 | stat = read_cpc_co_stat_conf(); |
| 71 | seq_state = stat & CPC_Cx_STAT_CONF_SEQSTATE; |
| 72 | seq_state >>= __ffs(CPC_Cx_STAT_CONF_SEQSTATE); |
| 73 | if (seq_state == CPC_Cx_STAT_CONF_SEQSTATE_U5) |
| 74 | break; |
| 75 | |
| 76 | if (timeout) { |
| 77 | mdelay(1); |
| 78 | timeout--; |
| 79 | } else { |
| 80 | pr_warn("Waiting for cluster %u CM to power up... STAT_CONF=0x%x\n" , |
| 81 | cluster, stat); |
| 82 | mdelay(1000); |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | mips_cm_unlock_other(); |
| 87 | } |
| 88 | |
| 89 | static unsigned __init core_vpe_count(unsigned int cluster, unsigned core) |
| 90 | { |
| 91 | return min(smp_max_threads, mips_cps_numvps(cluster, core)); |
| 92 | } |
| 93 | |
| 94 | static void __init *mips_cps_build_core_entry(void *addr) |
| 95 | { |
| 96 | extern void (*nmi_handler)(void); |
| 97 | u32 *p = addr; |
| 98 | u32 val; |
| 99 | struct uasm_label labels[2]; |
| 100 | struct uasm_reloc relocs[2]; |
| 101 | struct uasm_label *l = labels; |
| 102 | struct uasm_reloc *r = relocs; |
| 103 | |
| 104 | memset(labels, 0, sizeof(labels)); |
| 105 | memset(relocs, 0, sizeof(relocs)); |
| 106 | |
| 107 | uasm_i_mfc0(&p, GPR_K0, C0_STATUS); |
| 108 | UASM_i_LA(&p, GPR_T9, ST0_NMI); |
| 109 | uasm_i_and(&p, GPR_K0, GPR_K0, GPR_T9); |
| 110 | |
| 111 | uasm_il_bnez(&p, &r, GPR_K0, label_not_nmi); |
| 112 | uasm_i_nop(&p); |
| 113 | UASM_i_LA(&p, GPR_K0, (long)&nmi_handler); |
| 114 | |
| 115 | uasm_l_not_nmi(&l, p); |
| 116 | |
| 117 | val = CAUSEF_IV; |
| 118 | uasm_i_lui(&p, GPR_K0, val >> 16); |
| 119 | uasm_i_ori(&p, GPR_K0, GPR_K0, val & 0xffff); |
| 120 | uasm_i_mtc0(&p, GPR_K0, C0_CAUSE); |
| 121 | val = ST0_CU1 | ST0_CU0 | ST0_BEV | ST0_KX_IF_64; |
| 122 | uasm_i_lui(&p, GPR_K0, val >> 16); |
| 123 | uasm_i_ori(&p, GPR_K0, GPR_K0, val & 0xffff); |
| 124 | uasm_i_mtc0(&p, GPR_K0, C0_STATUS); |
| 125 | uasm_i_ehb(&p); |
| 126 | uasm_i_ori(&p, GPR_A0, 0, read_c0_config() & CONF_CM_CMASK); |
| 127 | UASM_i_LA(&p, GPR_A1, (long)mips_gcr_base); |
| 128 | #if defined(KBUILD_64BIT_SYM32) || defined(CONFIG_32BIT) |
| 129 | UASM_i_LA(&p, GPR_T9, CKSEG1ADDR(__pa_symbol(mips_cps_core_boot))); |
| 130 | #else |
| 131 | UASM_i_LA(&p, GPR_T9, TO_UNCAC(__pa_symbol(mips_cps_core_boot))); |
| 132 | #endif |
| 133 | uasm_i_jr(&p, GPR_T9); |
| 134 | uasm_i_nop(&p); |
| 135 | |
| 136 | uasm_resolve_relocs(relocs, labels); |
| 137 | |
| 138 | return p; |
| 139 | } |
| 140 | |
| 141 | static bool __init check_64bit_reset(void) |
| 142 | { |
| 143 | bool cx_64bit_reset = false; |
| 144 | |
| 145 | mips_cm_lock_other(0, 0, 0, CM_GCR_Cx_OTHER_BLOCK_LOCAL); |
| 146 | write_gcr_co_reset64_base(CM_GCR_Cx_RESET64_BASE_BEVEXCBASE); |
| 147 | if ((read_gcr_co_reset64_base() & CM_GCR_Cx_RESET64_BASE_BEVEXCBASE) == |
| 148 | CM_GCR_Cx_RESET64_BASE_BEVEXCBASE) |
| 149 | cx_64bit_reset = true; |
| 150 | mips_cm_unlock_other(); |
| 151 | |
| 152 | return cx_64bit_reset; |
| 153 | } |
| 154 | |
| 155 | static int __init allocate_cps_vecs(void) |
| 156 | { |
| 157 | /* Try to allocate in KSEG1 first */ |
| 158 | cps_vec_pa = memblock_phys_alloc_range(BEV_VEC_SIZE, BEV_VEC_ALIGN, |
| 159 | 0x0, CSEGX_SIZE - 1); |
| 160 | |
| 161 | if (cps_vec_pa) |
| 162 | core_entry_reg = CKSEG1ADDR(cps_vec_pa) & |
| 163 | CM_GCR_Cx_RESET_BASE_BEVEXCBASE; |
| 164 | |
| 165 | if (!cps_vec_pa && mips_cm_is64) { |
| 166 | phys_addr_t end; |
| 167 | |
| 168 | if (check_64bit_reset()) { |
| 169 | pr_info("VP Local Reset Exception Base support 47 bits address\n" ); |
| 170 | end = MEMBLOCK_ALLOC_ANYWHERE; |
| 171 | } else { |
| 172 | end = SZ_4G - 1; |
| 173 | } |
| 174 | cps_vec_pa = memblock_phys_alloc_range(BEV_VEC_SIZE, BEV_VEC_ALIGN, 0, end); |
| 175 | if (cps_vec_pa) { |
| 176 | if (check_64bit_reset()) |
| 177 | core_entry_reg = (cps_vec_pa & CM_GCR_Cx_RESET64_BASE_BEVEXCBASE) | |
| 178 | CM_GCR_Cx_RESET_BASE_MODE; |
| 179 | else |
| 180 | core_entry_reg = (cps_vec_pa & CM_GCR_Cx_RESET_BASE_BEVEXCBASE) | |
| 181 | CM_GCR_Cx_RESET_BASE_MODE; |
| 182 | } |
| 183 | } |
| 184 | |
| 185 | if (!cps_vec_pa) |
| 186 | return -ENOMEM; |
| 187 | |
| 188 | return 0; |
| 189 | } |
| 190 | |
| 191 | static void __init setup_cps_vecs(void) |
| 192 | { |
| 193 | void *cps_vec; |
| 194 | |
| 195 | cps_vec = (void *)CKSEG1ADDR_OR_64BIT(cps_vec_pa); |
| 196 | mips_cps_build_core_entry(addr: cps_vec); |
| 197 | |
| 198 | memcpy(cps_vec + 0x200, &excep_tlbfill, 0x80); |
| 199 | memcpy(cps_vec + 0x280, &excep_xtlbfill, 0x80); |
| 200 | memcpy(cps_vec + 0x300, &excep_cache, 0x80); |
| 201 | memcpy(cps_vec + 0x380, &excep_genex, 0x80); |
| 202 | memcpy(cps_vec + 0x400, &excep_intex, 0x80); |
| 203 | memcpy(cps_vec + 0x480, &excep_ejtag, 0x80); |
| 204 | |
| 205 | /* Make sure no prefetched data in cache */ |
| 206 | blast_inv_dcache_range(CKSEG0ADDR_OR_64BIT(cps_vec_pa), CKSEG0ADDR_OR_64BIT(cps_vec_pa) + BEV_VEC_SIZE); |
| 207 | bc_inv(CKSEG0ADDR_OR_64BIT(cps_vec_pa), BEV_VEC_SIZE); |
| 208 | __sync(); |
| 209 | } |
| 210 | |
| 211 | static void __init cps_smp_setup(void) |
| 212 | { |
| 213 | unsigned int nclusters, ncores, nvpes, core_vpes; |
| 214 | int cl, c, v; |
| 215 | |
| 216 | /* Detect & record VPE topology */ |
| 217 | nvpes = 0; |
| 218 | nclusters = mips_cps_numclusters(); |
| 219 | pr_info("%s topology " , cpu_has_mips_r6 ? "VP" : "VPE" ); |
| 220 | for (cl = 0; cl < nclusters; cl++) { |
| 221 | if (cl > 0) |
| 222 | pr_cont("," ); |
| 223 | pr_cont("{" ); |
| 224 | |
| 225 | if (mips_cm_revision() >= CM_REV_CM3_5) |
| 226 | power_up_other_cluster(cl); |
| 227 | |
| 228 | ncores = mips_cps_numcores(cl); |
| 229 | for (c = 0; c < ncores; c++) { |
| 230 | core_vpes = core_vpe_count(cluster: cl, core: c); |
| 231 | |
| 232 | if (c > 0) |
| 233 | pr_cont("," ); |
| 234 | pr_cont("%u" , core_vpes); |
| 235 | |
| 236 | /* Use the number of VPEs in cluster 0 core 0 for smp_num_siblings */ |
| 237 | if (!cl && !c) |
| 238 | smp_num_siblings = core_vpes; |
| 239 | cpumask_set_cpu(cpu: nvpes, dstp: &__cpu_primary_thread_mask); |
| 240 | |
| 241 | for (v = 0; v < min_t(int, core_vpes, NR_CPUS - nvpes); v++) { |
| 242 | cpu_set_cluster(&cpu_data[nvpes + v], cl); |
| 243 | cpu_set_core(&cpu_data[nvpes + v], c); |
| 244 | cpu_set_vpe_id(&cpu_data[nvpes + v], v); |
| 245 | } |
| 246 | |
| 247 | nvpes += core_vpes; |
| 248 | } |
| 249 | |
| 250 | pr_cont("}" ); |
| 251 | } |
| 252 | pr_cont(" total %u\n" , nvpes); |
| 253 | |
| 254 | /* Indicate present CPUs (CPU being synonymous with VPE) */ |
| 255 | for (v = 0; v < min_t(unsigned, nvpes, NR_CPUS); v++) { |
| 256 | set_cpu_possible(cpu: v, possible: true); |
| 257 | set_cpu_present(v, true); |
| 258 | __cpu_number_map[v] = v; |
| 259 | __cpu_logical_map[v] = v; |
| 260 | } |
| 261 | |
| 262 | /* Set a coherent default CCA (CWB) */ |
| 263 | change_c0_config(CONF_CM_CMASK, 0x5); |
| 264 | |
| 265 | /* Initialise core 0 */ |
| 266 | mips_cps_core_init(); |
| 267 | |
| 268 | /* Make core 0 coherent with everything */ |
| 269 | write_gcr_cl_coherence(0xff); |
| 270 | |
| 271 | if (allocate_cps_vecs()) |
| 272 | pr_err("Failed to allocate CPS vectors\n" ); |
| 273 | |
| 274 | if (core_entry_reg && mips_cm_revision() >= CM_REV_CM3) |
| 275 | write_gcr_bev_base(core_entry_reg); |
| 276 | |
| 277 | #ifdef CONFIG_MIPS_MT_FPAFF |
| 278 | /* If we have an FPU, enroll ourselves in the FPU-full mask */ |
| 279 | if (cpu_has_fpu) |
| 280 | cpumask_set_cpu(0, &mt_fpu_cpumask); |
| 281 | #endif /* CONFIG_MIPS_MT_FPAFF */ |
| 282 | } |
| 283 | |
| 284 | unsigned long calibrate_delay_is_known(void) |
| 285 | { |
| 286 | int first_cpu_cluster = 0; |
| 287 | |
| 288 | /* The calibration has to be done on the primary CPU of the cluster */ |
| 289 | if (mips_cps_first_online_in_cluster(&first_cpu_cluster)) |
| 290 | return 0; |
| 291 | |
| 292 | return cpu_data[first_cpu_cluster].udelay_val; |
| 293 | } |
| 294 | |
| 295 | static void __init cps_prepare_cpus(unsigned int max_cpus) |
| 296 | { |
| 297 | unsigned int nclusters, ncores, core_vpes, nvpe = 0, c, cl, cca; |
| 298 | bool cca_unsuitable, cores_limited; |
| 299 | struct cluster_boot_config *cluster_bootcfg; |
| 300 | struct core_boot_config *core_bootcfg; |
| 301 | |
| 302 | mips_mt_set_cpuoptions(); |
| 303 | |
| 304 | if (!core_entry_reg) { |
| 305 | pr_err("core_entry address unsuitable, disabling smp-cps\n" ); |
| 306 | goto err_out; |
| 307 | } |
| 308 | |
| 309 | /* Detect whether the CCA is unsuited to multi-core SMP */ |
| 310 | cca = read_c0_config() & CONF_CM_CMASK; |
| 311 | switch (cca) { |
| 312 | case 0x4: /* CWBE */ |
| 313 | case 0x5: /* CWB */ |
| 314 | /* The CCA is coherent, multi-core is fine */ |
| 315 | cca_unsuitable = false; |
| 316 | break; |
| 317 | |
| 318 | default: |
| 319 | /* CCA is not coherent, multi-core is not usable */ |
| 320 | cca_unsuitable = true; |
| 321 | } |
| 322 | |
| 323 | /* Warn the user if the CCA prevents multi-core */ |
| 324 | cores_limited = false; |
| 325 | if (cca_unsuitable || cpu_has_dc_aliases) { |
| 326 | for_each_present_cpu(c) { |
| 327 | if (cpus_are_siblings(smp_processor_id(), c)) |
| 328 | continue; |
| 329 | |
| 330 | set_cpu_present(c, false); |
| 331 | cores_limited = true; |
| 332 | } |
| 333 | } |
| 334 | if (cores_limited) |
| 335 | pr_warn("Using only one core due to %s%s%s\n" , |
| 336 | cca_unsuitable ? "unsuitable CCA" : "" , |
| 337 | (cca_unsuitable && cpu_has_dc_aliases) ? " & " : "" , |
| 338 | cpu_has_dc_aliases ? "dcache aliasing" : "" ); |
| 339 | |
| 340 | setup_cps_vecs(); |
| 341 | |
| 342 | /* Allocate cluster boot configuration structs */ |
| 343 | nclusters = mips_cps_numclusters(); |
| 344 | mips_cps_cluster_bootcfg = kcalloc(nclusters, |
| 345 | sizeof(*mips_cps_cluster_bootcfg), |
| 346 | GFP_KERNEL); |
| 347 | if (!mips_cps_cluster_bootcfg) |
| 348 | goto err_out; |
| 349 | |
| 350 | if (nclusters > 1) |
| 351 | mips_cm_update_property(); |
| 352 | |
| 353 | for (cl = 0; cl < nclusters; cl++) { |
| 354 | /* Allocate core boot configuration structs */ |
| 355 | ncores = mips_cps_numcores(cl); |
| 356 | core_bootcfg = kcalloc(ncores, sizeof(*core_bootcfg), |
| 357 | GFP_KERNEL); |
| 358 | if (!core_bootcfg) |
| 359 | goto err_out; |
| 360 | mips_cps_cluster_bootcfg[cl].core_config = core_bootcfg; |
| 361 | |
| 362 | mips_cps_cluster_bootcfg[cl].core_power = |
| 363 | kcalloc(BITS_TO_LONGS(ncores), sizeof(unsigned long), |
| 364 | GFP_KERNEL); |
| 365 | if (!mips_cps_cluster_bootcfg[cl].core_power) |
| 366 | goto err_out; |
| 367 | |
| 368 | /* Allocate VPE boot configuration structs */ |
| 369 | for (c = 0; c < ncores; c++) { |
| 370 | int v; |
| 371 | core_vpes = core_vpe_count(cluster: cl, core: c); |
| 372 | core_bootcfg[c].vpe_config = kcalloc(core_vpes, |
| 373 | sizeof(*core_bootcfg[c].vpe_config), |
| 374 | GFP_KERNEL); |
| 375 | for (v = 0; v < core_vpes; v++) |
| 376 | cpumask_set_cpu(nvpe++, &mips_cps_cluster_bootcfg[cl].cpumask); |
| 377 | if (!core_bootcfg[c].vpe_config) |
| 378 | goto err_out; |
| 379 | } |
| 380 | } |
| 381 | |
| 382 | /* Mark this CPU as powered up & booted */ |
| 383 | cl = cpu_cluster(¤t_cpu_data); |
| 384 | c = cpu_core(¤t_cpu_data); |
| 385 | cluster_bootcfg = &mips_cps_cluster_bootcfg[cl]; |
| 386 | cpu_smt_set_num_threads(num_threads: core_vpes, max_threads: core_vpes); |
| 387 | core_bootcfg = &cluster_bootcfg->core_config[c]; |
| 388 | bitmap_set(cluster_bootcfg->core_power, cpu_core(¤t_cpu_data), 1); |
| 389 | atomic_set(&core_bootcfg->vpe_mask, 1 << cpu_vpe_id(¤t_cpu_data)); |
| 390 | |
| 391 | return; |
| 392 | err_out: |
| 393 | /* Clean up allocations */ |
| 394 | if (mips_cps_cluster_bootcfg) { |
| 395 | for (cl = 0; cl < nclusters; cl++) { |
| 396 | cluster_bootcfg = &mips_cps_cluster_bootcfg[cl]; |
| 397 | ncores = mips_cps_numcores(cl); |
| 398 | for (c = 0; c < ncores; c++) { |
| 399 | core_bootcfg = &cluster_bootcfg->core_config[c]; |
| 400 | kfree(objp: core_bootcfg->vpe_config); |
| 401 | } |
| 402 | kfree(mips_cps_cluster_bootcfg[c].core_config); |
| 403 | } |
| 404 | kfree(mips_cps_cluster_bootcfg); |
| 405 | mips_cps_cluster_bootcfg = NULL; |
| 406 | } |
| 407 | |
| 408 | /* Effectively disable SMP by declaring CPUs not present */ |
| 409 | for_each_possible_cpu(c) { |
| 410 | if (c == 0) |
| 411 | continue; |
| 412 | set_cpu_present(c, false); |
| 413 | } |
| 414 | } |
| 415 | |
| 416 | static void init_cluster_l2(void) |
| 417 | { |
| 418 | u32 l2_cfg, l2sm_cop, result; |
| 419 | |
| 420 | while (!mips_cm_is_l2_hci_broken) { |
| 421 | l2_cfg = read_gcr_redir_l2_ram_config(); |
| 422 | |
| 423 | /* If HCI is not supported, use the state machine below */ |
| 424 | if (!(l2_cfg & CM_GCR_L2_RAM_CONFIG_PRESENT)) |
| 425 | break; |
| 426 | if (!(l2_cfg & CM_GCR_L2_RAM_CONFIG_HCI_SUPPORTED)) |
| 427 | break; |
| 428 | |
| 429 | /* If the HCI_DONE bit is set, we're finished */ |
| 430 | if (l2_cfg & CM_GCR_L2_RAM_CONFIG_HCI_DONE) |
| 431 | return; |
| 432 | } |
| 433 | |
| 434 | l2sm_cop = read_gcr_redir_l2sm_cop(); |
| 435 | if (WARN(!(l2sm_cop & CM_GCR_L2SM_COP_PRESENT), |
| 436 | "L2 init not supported on this system yet" )) |
| 437 | return; |
| 438 | |
| 439 | /* Clear L2 tag registers */ |
| 440 | write_gcr_redir_l2_tag_state(0); |
| 441 | write_gcr_redir_l2_ecc(0); |
| 442 | |
| 443 | /* Ensure the L2 tag writes complete before the state machine starts */ |
| 444 | mb(); |
| 445 | |
| 446 | /* Wait for the L2 state machine to be idle */ |
| 447 | do { |
| 448 | l2sm_cop = read_gcr_redir_l2sm_cop(); |
| 449 | } while (l2sm_cop & CM_GCR_L2SM_COP_RUNNING); |
| 450 | |
| 451 | /* Start a store tag operation */ |
| 452 | l2sm_cop = CM_GCR_L2SM_COP_TYPE_IDX_STORETAG; |
| 453 | l2sm_cop <<= __ffs(CM_GCR_L2SM_COP_TYPE); |
| 454 | l2sm_cop |= CM_GCR_L2SM_COP_CMD_START; |
| 455 | write_gcr_redir_l2sm_cop(l2sm_cop); |
| 456 | |
| 457 | /* Ensure the state machine starts before we poll for completion */ |
| 458 | mb(); |
| 459 | |
| 460 | /* Wait for the operation to be complete */ |
| 461 | do { |
| 462 | l2sm_cop = read_gcr_redir_l2sm_cop(); |
| 463 | result = l2sm_cop & CM_GCR_L2SM_COP_RESULT; |
| 464 | result >>= __ffs(CM_GCR_L2SM_COP_RESULT); |
| 465 | } while (!result); |
| 466 | |
| 467 | WARN(result != CM_GCR_L2SM_COP_RESULT_DONE_OK, |
| 468 | "L2 state machine failed cache init with error %u\n" , result); |
| 469 | } |
| 470 | |
| 471 | static void boot_core(unsigned int cluster, unsigned int core, |
| 472 | unsigned int vpe_id) |
| 473 | { |
| 474 | struct cluster_boot_config *cluster_cfg; |
| 475 | u32 access, stat, seq_state; |
| 476 | unsigned int timeout, ncores; |
| 477 | |
| 478 | cluster_cfg = &mips_cps_cluster_bootcfg[cluster]; |
| 479 | ncores = mips_cps_numcores(cluster); |
| 480 | |
| 481 | if ((cluster != cpu_cluster(¤t_cpu_data)) && |
| 482 | bitmap_empty(cluster_cfg->core_power, ncores)) { |
| 483 | power_up_other_cluster(cluster); |
| 484 | |
| 485 | mips_cm_lock_other(cluster, core, 0, |
| 486 | CM_GCR_Cx_OTHER_BLOCK_GLOBAL); |
| 487 | |
| 488 | /* Ensure cluster GCRs are where we expect */ |
| 489 | write_gcr_redir_base(read_gcr_base()); |
| 490 | write_gcr_redir_cpc_base(read_gcr_cpc_base()); |
| 491 | write_gcr_redir_gic_base(read_gcr_gic_base()); |
| 492 | |
| 493 | init_cluster_l2(); |
| 494 | |
| 495 | /* Mirror L2 configuration */ |
| 496 | write_gcr_redir_l2_only_sync_base(read_gcr_l2_only_sync_base()); |
| 497 | write_gcr_redir_l2_pft_control(read_gcr_l2_pft_control()); |
| 498 | write_gcr_redir_l2_pft_control_b(read_gcr_l2_pft_control_b()); |
| 499 | |
| 500 | /* Mirror ECC/parity setup */ |
| 501 | write_gcr_redir_err_control(read_gcr_err_control()); |
| 502 | |
| 503 | /* Set BEV base */ |
| 504 | write_gcr_redir_bev_base(core_entry_reg); |
| 505 | |
| 506 | mips_cm_unlock_other(); |
| 507 | } |
| 508 | |
| 509 | if (cluster != cpu_cluster(¤t_cpu_data)) { |
| 510 | mips_cm_lock_other(cluster, core, 0, |
| 511 | CM_GCR_Cx_OTHER_BLOCK_GLOBAL); |
| 512 | |
| 513 | /* Ensure the core can access the GCRs */ |
| 514 | access = read_gcr_redir_access(); |
| 515 | access |= BIT(core); |
| 516 | write_gcr_redir_access(access); |
| 517 | |
| 518 | mips_cm_unlock_other(); |
| 519 | } else { |
| 520 | /* Ensure the core can access the GCRs */ |
| 521 | access = read_gcr_access(); |
| 522 | access |= BIT(core); |
| 523 | write_gcr_access(access); |
| 524 | } |
| 525 | |
| 526 | /* Select the appropriate core */ |
| 527 | mips_cm_lock_other(cluster, core, 0, CM_GCR_Cx_OTHER_BLOCK_LOCAL); |
| 528 | |
| 529 | /* Set its reset vector */ |
| 530 | if (mips_cm_is64) |
| 531 | write_gcr_co_reset64_base(core_entry_reg); |
| 532 | else |
| 533 | write_gcr_co_reset_base(core_entry_reg); |
| 534 | |
| 535 | /* Ensure its coherency is disabled */ |
| 536 | write_gcr_co_coherence(0); |
| 537 | |
| 538 | /* Start it with the legacy memory map and exception base */ |
| 539 | write_gcr_co_reset_ext_base(CM_GCR_Cx_RESET_EXT_BASE_UEB); |
| 540 | |
| 541 | /* Ensure the core can access the GCRs */ |
| 542 | if (mips_cm_revision() < CM_REV_CM3) |
| 543 | set_gcr_access(1 << core); |
| 544 | else |
| 545 | set_gcr_access_cm3(1 << core); |
| 546 | |
| 547 | if (mips_cpc_present()) { |
| 548 | /* Reset the core */ |
| 549 | mips_cpc_lock_other(core); |
| 550 | |
| 551 | if (mips_cm_revision() >= CM_REV_CM3) { |
| 552 | /* Run only the requested VP following the reset */ |
| 553 | write_cpc_co_vp_stop(0xf); |
| 554 | write_cpc_co_vp_run(1 << vpe_id); |
| 555 | |
| 556 | /* |
| 557 | * Ensure that the VP_RUN register is written before the |
| 558 | * core leaves reset. |
| 559 | */ |
| 560 | wmb(); |
| 561 | } |
| 562 | |
| 563 | write_cpc_co_cmd(CPC_Cx_CMD_RESET); |
| 564 | |
| 565 | timeout = 100; |
| 566 | while (true) { |
| 567 | stat = read_cpc_co_stat_conf(); |
| 568 | seq_state = stat & CPC_Cx_STAT_CONF_SEQSTATE; |
| 569 | seq_state >>= __ffs(CPC_Cx_STAT_CONF_SEQSTATE); |
| 570 | |
| 571 | /* U6 == coherent execution, ie. the core is up */ |
| 572 | if (seq_state == CPC_Cx_STAT_CONF_SEQSTATE_U6) |
| 573 | break; |
| 574 | |
| 575 | /* Delay a little while before we start warning */ |
| 576 | if (timeout) { |
| 577 | timeout--; |
| 578 | mdelay(10); |
| 579 | continue; |
| 580 | } |
| 581 | |
| 582 | pr_warn("Waiting for core %u to start... STAT_CONF=0x%x\n" , |
| 583 | core, stat); |
| 584 | mdelay(1000); |
| 585 | } |
| 586 | |
| 587 | mips_cpc_unlock_other(); |
| 588 | } else { |
| 589 | /* Take the core out of reset */ |
| 590 | write_gcr_co_reset_release(0); |
| 591 | } |
| 592 | |
| 593 | mips_cm_unlock_other(); |
| 594 | |
| 595 | /* The core is now powered up */ |
| 596 | bitmap_set(map: cluster_cfg->core_power, start: core, nbits: 1); |
| 597 | |
| 598 | /* |
| 599 | * Restore CM_PWRUP=0 so that the CM can power down if all the cores in |
| 600 | * the cluster do (eg. if they're all removed via hotplug. |
| 601 | */ |
| 602 | if (mips_cm_revision() >= CM_REV_CM3_5) { |
| 603 | mips_cm_lock_other(cluster, 0, 0, CM_GCR_Cx_OTHER_BLOCK_GLOBAL); |
| 604 | write_cpc_redir_pwrup_ctl(0); |
| 605 | mips_cm_unlock_other(); |
| 606 | } |
| 607 | } |
| 608 | |
| 609 | static void remote_vpe_boot(void *dummy) |
| 610 | { |
| 611 | unsigned int cluster = cpu_cluster(¤t_cpu_data); |
| 612 | unsigned core = cpu_core(¤t_cpu_data); |
| 613 | struct cluster_boot_config *cluster_cfg = |
| 614 | &mips_cps_cluster_bootcfg[cluster]; |
| 615 | struct core_boot_config *core_cfg = &cluster_cfg->core_config[core]; |
| 616 | |
| 617 | mips_cps_boot_vpes(core_cfg, cpu_vpe_id(¤t_cpu_data)); |
| 618 | } |
| 619 | |
| 620 | static int cps_boot_secondary(int cpu, struct task_struct *idle) |
| 621 | { |
| 622 | unsigned int cluster = cpu_cluster(&cpu_data[cpu]); |
| 623 | unsigned core = cpu_core(&cpu_data[cpu]); |
| 624 | unsigned vpe_id = cpu_vpe_id(&cpu_data[cpu]); |
| 625 | struct cluster_boot_config *cluster_cfg = |
| 626 | &mips_cps_cluster_bootcfg[cluster]; |
| 627 | struct core_boot_config *core_cfg = &cluster_cfg->core_config[core]; |
| 628 | struct vpe_boot_config *vpe_cfg = &core_cfg->vpe_config[vpe_id]; |
| 629 | unsigned int remote; |
| 630 | int err; |
| 631 | |
| 632 | vpe_cfg->pc = (unsigned long)&smp_bootstrap; |
| 633 | vpe_cfg->sp = __KSTK_TOS(idle); |
| 634 | vpe_cfg->gp = (unsigned long)task_thread_info(idle); |
| 635 | |
| 636 | atomic_or(1 << cpu_vpe_id(&cpu_data[cpu]), &core_cfg->vpe_mask); |
| 637 | |
| 638 | preempt_disable(); |
| 639 | |
| 640 | if (!test_bit(core, cluster_cfg->core_power)) { |
| 641 | /* Boot a VPE on a powered down core */ |
| 642 | boot_core(cluster, core, vpe_id); |
| 643 | goto out; |
| 644 | } |
| 645 | |
| 646 | if (cpu_has_vp) { |
| 647 | mips_cm_lock_other(cluster, core, vpe_id, |
| 648 | CM_GCR_Cx_OTHER_BLOCK_LOCAL); |
| 649 | if (mips_cm_is64) |
| 650 | write_gcr_co_reset64_base(core_entry_reg); |
| 651 | else |
| 652 | write_gcr_co_reset_base(core_entry_reg); |
| 653 | mips_cm_unlock_other(); |
| 654 | } |
| 655 | |
| 656 | if (!cpus_are_siblings(cpu, smp_processor_id())) { |
| 657 | /* Boot a VPE on another powered up core */ |
| 658 | for (remote = 0; remote < NR_CPUS; remote++) { |
| 659 | if (!cpus_are_siblings(cpu, remote)) |
| 660 | continue; |
| 661 | if (cpu_online(cpu: remote)) |
| 662 | break; |
| 663 | } |
| 664 | if (remote >= NR_CPUS) { |
| 665 | pr_crit("No online CPU in core %u to start CPU%d\n" , |
| 666 | core, cpu); |
| 667 | goto out; |
| 668 | } |
| 669 | |
| 670 | err = smp_call_function_single(cpuid: remote, func: remote_vpe_boot, |
| 671 | NULL, wait: 1); |
| 672 | if (err) |
| 673 | panic(fmt: "Failed to call remote CPU\n" ); |
| 674 | goto out; |
| 675 | } |
| 676 | |
| 677 | BUG_ON(!cpu_has_mipsmt && !cpu_has_vp); |
| 678 | |
| 679 | /* Boot a VPE on this core */ |
| 680 | mips_cps_boot_vpes(core_cfg, vpe_id); |
| 681 | out: |
| 682 | preempt_enable(); |
| 683 | return 0; |
| 684 | } |
| 685 | |
| 686 | static void cps_init_secondary(void) |
| 687 | { |
| 688 | int core = cpu_core(¤t_cpu_data); |
| 689 | |
| 690 | /* Disable MT - we only want to run 1 TC per VPE */ |
| 691 | if (cpu_has_mipsmt) |
| 692 | dmt(); |
| 693 | |
| 694 | if (mips_cm_revision() >= CM_REV_CM3) { |
| 695 | unsigned int ident = read_gic_vl_ident(); |
| 696 | |
| 697 | /* |
| 698 | * Ensure that our calculation of the VP ID matches up with |
| 699 | * what the GIC reports, otherwise we'll have configured |
| 700 | * interrupts incorrectly. |
| 701 | */ |
| 702 | BUG_ON(ident != mips_cm_vp_id(smp_processor_id())); |
| 703 | } |
| 704 | |
| 705 | if (core > 0 && !read_gcr_cl_coherence()) |
| 706 | pr_warn("Core %u is not in coherent domain\n" , core); |
| 707 | |
| 708 | if (cpu_has_veic) |
| 709 | clear_c0_status(ST0_IM); |
| 710 | else |
| 711 | change_c0_status(ST0_IM, STATUSF_IP2 | STATUSF_IP3 | |
| 712 | STATUSF_IP4 | STATUSF_IP5 | |
| 713 | STATUSF_IP6 | STATUSF_IP7); |
| 714 | } |
| 715 | |
| 716 | static void cps_smp_finish(void) |
| 717 | { |
| 718 | write_c0_compare(read_c0_count() + (8 * mips_hpt_frequency / HZ)); |
| 719 | |
| 720 | #ifdef CONFIG_MIPS_MT_FPAFF |
| 721 | /* If we have an FPU, enroll ourselves in the FPU-full mask */ |
| 722 | if (cpu_has_fpu) |
| 723 | cpumask_set_cpu(smp_processor_id(), &mt_fpu_cpumask); |
| 724 | #endif /* CONFIG_MIPS_MT_FPAFF */ |
| 725 | |
| 726 | local_irq_enable(); |
| 727 | } |
| 728 | |
| 729 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_KEXEC_CORE) |
| 730 | |
| 731 | enum cpu_death { |
| 732 | CPU_DEATH_HALT, |
| 733 | CPU_DEATH_POWER, |
| 734 | }; |
| 735 | |
| 736 | static void cps_shutdown_this_cpu(enum cpu_death death) |
| 737 | { |
| 738 | unsigned int cpu, core, vpe_id; |
| 739 | |
| 740 | cpu = smp_processor_id(); |
| 741 | core = cpu_core(&cpu_data[cpu]); |
| 742 | |
| 743 | if (death == CPU_DEATH_HALT) { |
| 744 | vpe_id = cpu_vpe_id(&cpu_data[cpu]); |
| 745 | |
| 746 | pr_debug("Halting core %d VP%d\n" , core, vpe_id); |
| 747 | if (cpu_has_mipsmt) { |
| 748 | /* Halt this TC */ |
| 749 | write_c0_tchalt(TCHALT_H); |
| 750 | instruction_hazard(); |
| 751 | } else if (cpu_has_vp) { |
| 752 | write_cpc_cl_vp_stop(1 << vpe_id); |
| 753 | |
| 754 | /* Ensure that the VP_STOP register is written */ |
| 755 | wmb(); |
| 756 | } |
| 757 | } else { |
| 758 | if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) { |
| 759 | pr_debug("Gating power to core %d\n" , core); |
| 760 | /* Power down the core */ |
| 761 | cps_pm_enter_state(CPS_PM_POWER_GATED); |
| 762 | } |
| 763 | } |
| 764 | } |
| 765 | |
| 766 | #ifdef CONFIG_KEXEC_CORE |
| 767 | |
| 768 | static void cps_kexec_nonboot_cpu(void) |
| 769 | { |
| 770 | if (cpu_has_mipsmt || cpu_has_vp) |
| 771 | cps_shutdown_this_cpu(death: CPU_DEATH_HALT); |
| 772 | else |
| 773 | cps_shutdown_this_cpu(death: CPU_DEATH_POWER); |
| 774 | } |
| 775 | |
| 776 | #endif /* CONFIG_KEXEC_CORE */ |
| 777 | |
| 778 | #endif /* CONFIG_HOTPLUG_CPU || CONFIG_KEXEC_CORE */ |
| 779 | |
| 780 | #ifdef CONFIG_HOTPLUG_CPU |
| 781 | |
| 782 | static int cps_cpu_disable(void) |
| 783 | { |
| 784 | unsigned cpu = smp_processor_id(); |
| 785 | struct cluster_boot_config *cluster_cfg; |
| 786 | struct core_boot_config *core_cfg; |
| 787 | |
| 788 | if (!cps_pm_support_state(CPS_PM_POWER_GATED)) |
| 789 | return -EINVAL; |
| 790 | |
| 791 | cluster_cfg = &mips_cps_cluster_bootcfg[cpu_cluster(¤t_cpu_data)]; |
| 792 | core_cfg = &cluster_cfg->core_config[cpu_core(¤t_cpu_data)]; |
| 793 | atomic_sub(1 << cpu_vpe_id(¤t_cpu_data), &core_cfg->vpe_mask); |
| 794 | smp_mb__after_atomic(); |
| 795 | set_cpu_online(cpu, online: false); |
| 796 | calculate_cpu_foreign_map(); |
| 797 | irq_migrate_all_off_this_cpu(); |
| 798 | |
| 799 | return 0; |
| 800 | } |
| 801 | |
| 802 | static unsigned cpu_death_sibling; |
| 803 | static enum cpu_death cpu_death; |
| 804 | |
| 805 | void play_dead(void) |
| 806 | { |
| 807 | unsigned int cpu; |
| 808 | |
| 809 | local_irq_disable(); |
| 810 | idle_task_exit(); |
| 811 | cpu = smp_processor_id(); |
| 812 | cpu_death = CPU_DEATH_POWER; |
| 813 | |
| 814 | pr_debug("CPU%d going offline\n" , cpu); |
| 815 | |
| 816 | if (cpu_has_mipsmt || cpu_has_vp) { |
| 817 | /* Look for another online VPE within the core */ |
| 818 | for_each_online_cpu(cpu_death_sibling) { |
| 819 | if (!cpus_are_siblings(cpu, cpu_death_sibling)) |
| 820 | continue; |
| 821 | |
| 822 | /* |
| 823 | * There is an online VPE within the core. Just halt |
| 824 | * this TC and leave the core alone. |
| 825 | */ |
| 826 | cpu_death = CPU_DEATH_HALT; |
| 827 | break; |
| 828 | } |
| 829 | } |
| 830 | |
| 831 | cpuhp_ap_report_dead(); |
| 832 | |
| 833 | cps_shutdown_this_cpu(death: cpu_death); |
| 834 | |
| 835 | /* This should never be reached */ |
| 836 | panic(fmt: "Failed to offline CPU %u" , cpu); |
| 837 | } |
| 838 | |
| 839 | static void wait_for_sibling_halt(void *ptr_cpu) |
| 840 | { |
| 841 | unsigned cpu = (unsigned long)ptr_cpu; |
| 842 | unsigned vpe_id = cpu_vpe_id(&cpu_data[cpu]); |
| 843 | unsigned halted; |
| 844 | unsigned long flags; |
| 845 | |
| 846 | do { |
| 847 | local_irq_save(flags); |
| 848 | settc(vpe_id); |
| 849 | halted = read_tc_c0_tchalt(); |
| 850 | local_irq_restore(flags); |
| 851 | } while (!(halted & TCHALT_H)); |
| 852 | } |
| 853 | |
| 854 | static void cps_cpu_die(unsigned int cpu) { } |
| 855 | |
| 856 | static void cps_cleanup_dead_cpu(unsigned cpu) |
| 857 | { |
| 858 | unsigned int cluster = cpu_cluster(&cpu_data[cpu]); |
| 859 | unsigned core = cpu_core(&cpu_data[cpu]); |
| 860 | unsigned int vpe_id = cpu_vpe_id(&cpu_data[cpu]); |
| 861 | ktime_t fail_time; |
| 862 | unsigned stat; |
| 863 | int err; |
| 864 | struct cluster_boot_config *cluster_cfg; |
| 865 | |
| 866 | cluster_cfg = &mips_cps_cluster_bootcfg[cluster]; |
| 867 | |
| 868 | /* |
| 869 | * Now wait for the CPU to actually offline. Without doing this that |
| 870 | * offlining may race with one or more of: |
| 871 | * |
| 872 | * - Onlining the CPU again. |
| 873 | * - Powering down the core if another VPE within it is offlined. |
| 874 | * - A sibling VPE entering a non-coherent state. |
| 875 | * |
| 876 | * In the non-MT halt case (ie. infinite loop) the CPU is doing nothing |
| 877 | * with which we could race, so do nothing. |
| 878 | */ |
| 879 | if (cpu_death == CPU_DEATH_POWER) { |
| 880 | /* |
| 881 | * Wait for the core to enter a powered down or clock gated |
| 882 | * state, the latter happening when a JTAG probe is connected |
| 883 | * in which case the CPC will refuse to power down the core. |
| 884 | */ |
| 885 | fail_time = ktime_add_ms(kt: ktime_get(), msec: 2000); |
| 886 | do { |
| 887 | mips_cm_lock_other(0, core, 0, CM_GCR_Cx_OTHER_BLOCK_LOCAL); |
| 888 | mips_cpc_lock_other(core); |
| 889 | stat = read_cpc_co_stat_conf(); |
| 890 | stat &= CPC_Cx_STAT_CONF_SEQSTATE; |
| 891 | stat >>= __ffs(CPC_Cx_STAT_CONF_SEQSTATE); |
| 892 | mips_cpc_unlock_other(); |
| 893 | mips_cm_unlock_other(); |
| 894 | |
| 895 | if (stat == CPC_Cx_STAT_CONF_SEQSTATE_D0 || |
| 896 | stat == CPC_Cx_STAT_CONF_SEQSTATE_D2 || |
| 897 | stat == CPC_Cx_STAT_CONF_SEQSTATE_U2) |
| 898 | break; |
| 899 | |
| 900 | /* |
| 901 | * The core ought to have powered down, but didn't & |
| 902 | * now we don't really know what state it's in. It's |
| 903 | * likely that its _pwr_up pin has been wired to logic |
| 904 | * 1 & it powered back up as soon as we powered it |
| 905 | * down... |
| 906 | * |
| 907 | * The best we can do is warn the user & continue in |
| 908 | * the hope that the core is doing nothing harmful & |
| 909 | * might behave properly if we online it later. |
| 910 | */ |
| 911 | if (WARN(ktime_after(ktime_get(), fail_time), |
| 912 | "CPU%u hasn't powered down, seq. state %u\n" , |
| 913 | cpu, stat)) |
| 914 | break; |
| 915 | } while (1); |
| 916 | |
| 917 | /* Indicate the core is powered off */ |
| 918 | bitmap_clear(map: cluster_cfg->core_power, start: core, nbits: 1); |
| 919 | } else if (cpu_has_mipsmt) { |
| 920 | /* |
| 921 | * Have a CPU with access to the offlined CPUs registers wait |
| 922 | * for its TC to halt. |
| 923 | */ |
| 924 | err = smp_call_function_single(cpuid: cpu_death_sibling, |
| 925 | func: wait_for_sibling_halt, |
| 926 | info: (void *)(unsigned long)cpu, wait: 1); |
| 927 | if (err) |
| 928 | panic(fmt: "Failed to call remote sibling CPU\n" ); |
| 929 | } else if (cpu_has_vp) { |
| 930 | do { |
| 931 | mips_cm_lock_other(0, core, vpe_id, CM_GCR_Cx_OTHER_BLOCK_LOCAL); |
| 932 | stat = read_cpc_co_vp_running(); |
| 933 | mips_cm_unlock_other(); |
| 934 | } while (stat & (1 << vpe_id)); |
| 935 | } |
| 936 | } |
| 937 | |
| 938 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 939 | |
| 940 | static const struct plat_smp_ops cps_smp_ops = { |
| 941 | .smp_setup = cps_smp_setup, |
| 942 | .prepare_cpus = cps_prepare_cpus, |
| 943 | .boot_secondary = cps_boot_secondary, |
| 944 | .init_secondary = cps_init_secondary, |
| 945 | .smp_finish = cps_smp_finish, |
| 946 | .send_ipi_single = mips_smp_send_ipi_single, |
| 947 | .send_ipi_mask = mips_smp_send_ipi_mask, |
| 948 | #ifdef CONFIG_HOTPLUG_CPU |
| 949 | .cpu_disable = cps_cpu_disable, |
| 950 | .cpu_die = cps_cpu_die, |
| 951 | .cleanup_dead_cpu = cps_cleanup_dead_cpu, |
| 952 | #endif |
| 953 | #ifdef CONFIG_KEXEC_CORE |
| 954 | .kexec_nonboot_cpu = cps_kexec_nonboot_cpu, |
| 955 | #endif |
| 956 | }; |
| 957 | |
| 958 | bool mips_cps_smp_in_use(void) |
| 959 | { |
| 960 | extern const struct plat_smp_ops *mp_ops; |
| 961 | return mp_ops == &cps_smp_ops; |
| 962 | } |
| 963 | |
| 964 | int register_cps_smp_ops(void) |
| 965 | { |
| 966 | if (!mips_cm_present()) { |
| 967 | pr_warn("MIPS CPS SMP unable to proceed without a CM\n" ); |
| 968 | return -ENODEV; |
| 969 | } |
| 970 | |
| 971 | /* check we have a GIC - we need one for IPIs */ |
| 972 | if (!(read_gcr_gic_status() & CM_GCR_GIC_STATUS_EX)) { |
| 973 | pr_warn("MIPS CPS SMP unable to proceed without a GIC\n" ); |
| 974 | return -ENODEV; |
| 975 | } |
| 976 | |
| 977 | register_smp_ops(&cps_smp_ops); |
| 978 | return 0; |
| 979 | } |
| 980 | |