| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* leon_smp.c: Sparc-Leon SMP support. |
| 3 | * |
| 4 | * based on sun4m_smp.c |
| 5 | * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) |
| 6 | * Copyright (C) 2009 Daniel Hellstrom (daniel@gaisler.com) Aeroflex Gaisler AB |
| 7 | * Copyright (C) 2009 Konrad Eisele (konrad@gaisler.com) Aeroflex Gaisler AB |
| 8 | */ |
| 9 | |
| 10 | #include <asm/head.h> |
| 11 | |
| 12 | #include <linux/kernel.h> |
| 13 | #include <linux/sched/mm.h> |
| 14 | #include <linux/threads.h> |
| 15 | #include <linux/smp.h> |
| 16 | #include <linux/interrupt.h> |
| 17 | #include <linux/kernel_stat.h> |
| 18 | #include <linux/of.h> |
| 19 | #include <linux/init.h> |
| 20 | #include <linux/spinlock.h> |
| 21 | #include <linux/mm.h> |
| 22 | #include <linux/swap.h> |
| 23 | #include <linux/profile.h> |
| 24 | #include <linux/pm.h> |
| 25 | #include <linux/delay.h> |
| 26 | #include <linux/gfp.h> |
| 27 | #include <linux/cpu.h> |
| 28 | #include <linux/clockchips.h> |
| 29 | |
| 30 | #include <asm/cacheflush.h> |
| 31 | #include <asm/tlbflush.h> |
| 32 | |
| 33 | #include <asm/ptrace.h> |
| 34 | #include <linux/atomic.h> |
| 35 | #include <asm/irq_regs.h> |
| 36 | #include <asm/traps.h> |
| 37 | |
| 38 | #include <asm/delay.h> |
| 39 | #include <asm/irq.h> |
| 40 | #include <asm/page.h> |
| 41 | #include <asm/oplib.h> |
| 42 | #include <asm/cpudata.h> |
| 43 | #include <asm/asi.h> |
| 44 | #include <asm/leon.h> |
| 45 | #include <asm/leon_amba.h> |
| 46 | #include <asm/timer.h> |
| 47 | |
| 48 | #include "kernel.h" |
| 49 | |
| 50 | #include "irq.h" |
| 51 | |
| 52 | extern ctxd_t *srmmu_ctx_table_phys; |
| 53 | static int smp_processors_ready; |
| 54 | extern volatile unsigned long cpu_callin_map[NR_CPUS]; |
| 55 | extern cpumask_t smp_commenced_mask; |
| 56 | void leon_configure_cache_smp(void); |
| 57 | static void leon_ipi_init(void); |
| 58 | |
| 59 | /* IRQ number of LEON IPIs */ |
| 60 | int leon_ipi_irq = LEON3_IRQ_IPI_DEFAULT; |
| 61 | |
| 62 | static inline unsigned long do_swap(volatile unsigned long *ptr, |
| 63 | unsigned long val) |
| 64 | { |
| 65 | __asm__ __volatile__("swapa [%2] %3, %0\n\t" : "=&r"(val) |
| 66 | : "0"(val), "r"(ptr), "i"(ASI_LEON_DCACHE_MISS) |
| 67 | : "memory"); |
| 68 | return val; |
| 69 | } |
| 70 | |
| 71 | void leon_cpu_pre_starting(void *arg) |
| 72 | { |
| 73 | leon_configure_cache_smp(); |
| 74 | } |
| 75 | |
| 76 | void leon_cpu_pre_online(void *arg) |
| 77 | { |
| 78 | int cpuid = hard_smp_processor_id(); |
| 79 | |
| 80 | /* Allow master to continue. The master will then give us the |
| 81 | * go-ahead by setting the smp_commenced_mask and will wait without |
| 82 | * timeouts until our setup is completed fully (signified by |
| 83 | * our bit being set in the cpu_online_mask). |
| 84 | */ |
| 85 | do_swap(&cpu_callin_map[cpuid], 1); |
| 86 | |
| 87 | local_ops->cache_all(); |
| 88 | local_ops->tlb_all(); |
| 89 | |
| 90 | /* Fix idle thread fields. */ |
| 91 | __asm__ __volatile__("ld [%0], %%g6\n\t" : : "r"(¤t_set[cpuid]) |
| 92 | : "memory" /* paranoid */); |
| 93 | |
| 94 | /* Attach to the address space of init_task. */ |
| 95 | mmgrab(&init_mm); |
| 96 | current->active_mm = &init_mm; |
| 97 | |
| 98 | while (!cpumask_test_cpu(cpuid, &smp_commenced_mask)) |
| 99 | mb(); |
| 100 | } |
| 101 | |
| 102 | /* |
| 103 | * Cycle through the processors asking the PROM to start each one. |
| 104 | */ |
| 105 | |
| 106 | extern struct linux_prom_registers smp_penguin_ctable; |
| 107 | |
| 108 | void leon_configure_cache_smp(void) |
| 109 | { |
| 110 | unsigned long cfg = sparc_leon3_get_dcachecfg(); |
| 111 | int me = smp_processor_id(); |
| 112 | |
| 113 | if (ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg) > 4) { |
| 114 | printk(KERN_INFO "Note: SMP with snooping only works on 4k cache, found %dk(0x%x) on cpu %d, disabling caches\n", |
| 115 | (unsigned int)ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg), |
| 116 | (unsigned int)cfg, (unsigned int)me); |
| 117 | sparc_leon3_disable_cache(); |
| 118 | } else { |
| 119 | if (cfg & ASI_LEON3_SYSCTRL_CFG_SNOOPING) { |
| 120 | sparc_leon3_enable_snooping(); |
| 121 | } else { |
| 122 | printk(KERN_INFO "Note: You have to enable snooping in the vhdl model cpu %d, disabling caches\n", |
| 123 | me); |
| 124 | sparc_leon3_disable_cache(); |
| 125 | } |
| 126 | } |
| 127 | |
| 128 | local_ops->cache_all(); |
| 129 | local_ops->tlb_all(); |
| 130 | } |
| 131 | |
| 132 | static void leon_smp_setbroadcast(unsigned int mask) |
| 133 | { |
| 134 | int broadcast = |
| 135 | ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >> |
| 136 | LEON3_IRQMPSTATUS_BROADCAST) & 1); |
| 137 | if (!broadcast) { |
| 138 | prom_printf("######## !!!! The irqmp-ctrl must have broadcast enabled, smp wont work !!!!! ####### nr cpus: %d\n", |
| 139 | leon_smp_nrcpus()); |
| 140 | if (leon_smp_nrcpus() > 1) { |
| 141 | BUG(); |
| 142 | } else { |
| 143 | prom_printf("continue anyway\n"); |
| 144 | return; |
| 145 | } |
| 146 | } |
| 147 | LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpbroadcast), mask); |
| 148 | } |
| 149 | |
| 150 | int leon_smp_nrcpus(void) |
| 151 | { |
| 152 | int nrcpu = |
| 153 | ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >> |
| 154 | LEON3_IRQMPSTATUS_CPUNR) & 0xf) + 1; |
| 155 | return nrcpu; |
| 156 | } |
| 157 | |
| 158 | void __init leon_boot_cpus(void) |
| 159 | { |
| 160 | int nrcpu = leon_smp_nrcpus(); |
| 161 | int me = smp_processor_id(); |
| 162 | |
| 163 | /* Setup IPI */ |
| 164 | leon_ipi_init(); |
| 165 | |
| 166 | printk(KERN_INFO "%d:(%d:%d) cpus mpirq at 0x%x\n", (unsigned int)me, |
| 167 | (unsigned int)nrcpu, (unsigned int)NR_CPUS, |
| 168 | (unsigned int)&(leon3_irqctrl_regs->mpstatus)); |
| 169 | |
| 170 | leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, me); |
| 171 | leon_enable_irq_cpu(LEON3_IRQ_TICKER, me); |
| 172 | leon_enable_irq_cpu(leon_ipi_irq, me); |
| 173 | |
| 174 | leon_smp_setbroadcast(1 << LEON3_IRQ_TICKER); |
| 175 | |
| 176 | leon_configure_cache_smp(); |
| 177 | local_ops->cache_all(); |
| 178 | |
| 179 | } |
| 180 | |
| 181 | int leon_boot_one_cpu(int i, struct task_struct *idle) |
| 182 | { |
| 183 | int timeout; |
| 184 | |
| 185 | current_set[i] = task_thread_info(idle); |
| 186 | |
| 187 | /* See trampoline.S:leon_smp_cpu_startup for details... |
| 188 | * Initialize the contexts table |
| 189 | * Since the call to prom_startcpu() trashes the structure, |
| 190 | * we need to re-initialize it for each cpu |
| 191 | */ |
| 192 | smp_penguin_ctable.which_io = 0; |
| 193 | smp_penguin_ctable.phys_addr = (unsigned int)srmmu_ctx_table_phys; |
| 194 | smp_penguin_ctable.reg_size = 0; |
| 195 | |
| 196 | /* whirrr, whirrr, whirrrrrrrrr... */ |
| 197 | printk(KERN_INFO "Starting CPU %d : (irqmp: 0x%x)\n", (unsigned int)i, |
| 198 | (unsigned int)&leon3_irqctrl_regs->mpstatus); |
| 199 | local_ops->cache_all(); |
| 200 | |
| 201 | /* Make sure all IRQs are of from the start for this new CPU */ |
| 202 | LEON_BYPASS_STORE_PA(&leon3_irqctrl_regs->mask[i], 0); |
| 203 | |
| 204 | /* Wake one CPU */ |
| 205 | LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpstatus), 1 << i); |
| 206 | |
| 207 | /* wheee... it's going... */ |
| 208 | for (timeout = 0; timeout < 10000; timeout++) { |
| 209 | if (cpu_callin_map[i]) |
| 210 | break; |
| 211 | udelay(200); |
| 212 | } |
| 213 | printk(KERN_INFO "Started CPU %d\n", (unsigned int)i); |
| 214 | |
| 215 | if (!(cpu_callin_map[i])) { |
| 216 | printk(KERN_ERR "Processor %d is stuck.\n", i); |
| 217 | return -ENODEV; |
| 218 | } else { |
| 219 | leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, i); |
| 220 | leon_enable_irq_cpu(LEON3_IRQ_TICKER, i); |
| 221 | leon_enable_irq_cpu(leon_ipi_irq, i); |
| 222 | } |
| 223 | |
| 224 | local_ops->cache_all(); |
| 225 | return 0; |
| 226 | } |
| 227 | |
| 228 | void __init leon_smp_done(void) |
| 229 | { |
| 230 | |
| 231 | int i, first; |
| 232 | int *prev; |
| 233 | |
| 234 | /* setup cpu list for irq rotation */ |
| 235 | first = 0; |
| 236 | prev = &first; |
| 237 | for (i = 0; i < NR_CPUS; i++) { |
| 238 | if (cpu_online(i)) { |
| 239 | *prev = i; |
| 240 | prev = &cpu_data(i).next; |
| 241 | } |
| 242 | } |
| 243 | *prev = first; |
| 244 | local_ops->cache_all(); |
| 245 | |
| 246 | /* Free unneeded trap tables */ |
| 247 | if (!cpu_present(1)) { |
| 248 | free_reserved_page(virt_to_page(&trapbase_cpu1[0])); |
| 249 | } |
| 250 | if (!cpu_present(2)) { |
| 251 | free_reserved_page(virt_to_page(&trapbase_cpu2[0])); |
| 252 | } |
| 253 | if (!cpu_present(3)) { |
| 254 | free_reserved_page(virt_to_page(&trapbase_cpu3[0])); |
| 255 | } |
| 256 | /* Ok, they are spinning and ready to go. */ |
| 257 | smp_processors_ready = 1; |
| 258 | |
| 259 | } |
| 260 | |
| 261 | struct leon_ipi_work { |
| 262 | int single; |
| 263 | int msk; |
| 264 | int resched; |
| 265 | }; |
| 266 | |
| 267 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct leon_ipi_work, leon_ipi_work); |
| 268 | |
| 269 | /* Initialize IPIs on the LEON, in order to save IRQ resources only one IRQ |
| 270 | * is used for all three types of IPIs. |
| 271 | */ |
| 272 | static void __init leon_ipi_init(void) |
| 273 | { |
| 274 | int cpu, len; |
| 275 | struct leon_ipi_work *work; |
| 276 | struct property *pp; |
| 277 | struct device_node *rootnp; |
| 278 | struct tt_entry *trap_table; |
| 279 | unsigned long flags; |
| 280 | |
| 281 | /* Find IPI IRQ or stick with default value */ |
| 282 | rootnp = of_find_node_by_path("/ambapp0"); |
| 283 | if (rootnp) { |
| 284 | pp = of_find_property(rootnp, "ipi_num", &len); |
| 285 | if (pp && (*(int *)pp->value)) |
| 286 | leon_ipi_irq = *(int *)pp->value; |
| 287 | } |
| 288 | printk(KERN_INFO "leon: SMP IPIs at IRQ %d\n", leon_ipi_irq); |
| 289 | |
| 290 | /* Adjust so that we jump directly to smpleon_ipi */ |
| 291 | local_irq_save(flags); |
| 292 | trap_table = &sparc_ttable[SP_TRAP_IRQ1 + (leon_ipi_irq - 1)]; |
| 293 | trap_table->inst_three += smpleon_ipi - real_irq_entry; |
| 294 | local_ops->cache_all(); |
| 295 | local_irq_restore(flags); |
| 296 | |
| 297 | for_each_possible_cpu(cpu) { |
| 298 | work = &per_cpu(leon_ipi_work, cpu); |
| 299 | work->single = work->msk = work->resched = 0; |
| 300 | } |
| 301 | } |
| 302 | |
| 303 | static void leon_send_ipi(int cpu, int level) |
| 304 | { |
| 305 | unsigned long mask; |
| 306 | mask = leon_get_irqmask(level); |
| 307 | LEON3_BYPASS_STORE_PA(&leon3_irqctrl_regs->force[cpu], mask); |
| 308 | } |
| 309 | |
| 310 | static void leon_ipi_single(int cpu) |
| 311 | { |
| 312 | struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu); |
| 313 | |
| 314 | /* Mark work */ |
| 315 | work->single = 1; |
| 316 | |
| 317 | /* Generate IRQ on the CPU */ |
| 318 | leon_send_ipi(cpu, leon_ipi_irq); |
| 319 | } |
| 320 | |
| 321 | static void leon_ipi_mask_one(int cpu) |
| 322 | { |
| 323 | struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu); |
| 324 | |
| 325 | /* Mark work */ |
| 326 | work->msk = 1; |
| 327 | |
| 328 | /* Generate IRQ on the CPU */ |
| 329 | leon_send_ipi(cpu, leon_ipi_irq); |
| 330 | } |
| 331 | |
| 332 | static void leon_ipi_resched(int cpu) |
| 333 | { |
| 334 | struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu); |
| 335 | |
| 336 | /* Mark work */ |
| 337 | work->resched = 1; |
| 338 | |
| 339 | /* Generate IRQ on the CPU (any IRQ will cause resched) */ |
| 340 | leon_send_ipi(cpu, leon_ipi_irq); |
| 341 | } |
| 342 | |
| 343 | void leonsmp_ipi_interrupt(void) |
| 344 | { |
| 345 | struct leon_ipi_work *work = this_cpu_ptr(&leon_ipi_work); |
| 346 | |
| 347 | if (work->single) { |
| 348 | work->single = 0; |
| 349 | smp_call_function_single_interrupt(); |
| 350 | } |
| 351 | if (work->msk) { |
| 352 | work->msk = 0; |
| 353 | smp_call_function_interrupt(); |
| 354 | } |
| 355 | if (work->resched) { |
| 356 | work->resched = 0; |
| 357 | smp_resched_interrupt(); |
| 358 | } |
| 359 | } |
| 360 | |
| 361 | static struct smp_funcall { |
| 362 | void *func; |
| 363 | unsigned long arg1; |
| 364 | unsigned long arg2; |
| 365 | unsigned long arg3; |
| 366 | unsigned long arg4; |
| 367 | unsigned long arg5; |
| 368 | unsigned long processors_in[NR_CPUS]; /* Set when ipi entered. */ |
| 369 | unsigned long processors_out[NR_CPUS]; /* Set when ipi exited. */ |
| 370 | } ccall_info __attribute__((aligned(8))); |
| 371 | |
| 372 | static DEFINE_SPINLOCK(cross_call_lock); |
| 373 | |
| 374 | /* Cross calls must be serialized, at least currently. */ |
| 375 | static void leon_cross_call(void *func, cpumask_t mask, unsigned long arg1, |
| 376 | unsigned long arg2, unsigned long arg3, |
| 377 | unsigned long arg4) |
| 378 | { |
| 379 | if (smp_processors_ready) { |
| 380 | register int high = NR_CPUS - 1; |
| 381 | unsigned long flags; |
| 382 | |
| 383 | spin_lock_irqsave(&cross_call_lock, flags); |
| 384 | |
| 385 | { |
| 386 | /* If you make changes here, make sure gcc generates proper code... */ |
| 387 | register void *f asm("i0") = func; |
| 388 | register unsigned long a1 asm("i1") = arg1; |
| 389 | register unsigned long a2 asm("i2") = arg2; |
| 390 | register unsigned long a3 asm("i3") = arg3; |
| 391 | register unsigned long a4 asm("i4") = arg4; |
| 392 | register unsigned long a5 asm("i5") = 0; |
| 393 | |
| 394 | __asm__ __volatile__("std %0, [%6]\n\t" |
| 395 | "std %2, [%6 + 8]\n\t" |
| 396 | "std %4, [%6 + 16]\n\t" : : |
| 397 | "r"(f), "r"(a1), "r"(a2), "r"(a3), |
| 398 | "r"(a4), "r"(a5), |
| 399 | "r"(&ccall_info.func)); |
| 400 | } |
| 401 | |
| 402 | /* Init receive/complete mapping, plus fire the IPI's off. */ |
| 403 | { |
| 404 | register int i; |
| 405 | |
| 406 | cpumask_clear_cpu(smp_processor_id(), &mask); |
| 407 | cpumask_and(&mask, cpu_online_mask, &mask); |
| 408 | for (i = 0; i <= high; i++) { |
| 409 | if (cpumask_test_cpu(i, &mask)) { |
| 410 | ccall_info.processors_in[i] = 0; |
| 411 | ccall_info.processors_out[i] = 0; |
| 412 | leon_send_ipi(i, LEON3_IRQ_CROSS_CALL); |
| 413 | |
| 414 | } |
| 415 | } |
| 416 | } |
| 417 | |
| 418 | { |
| 419 | register int i; |
| 420 | |
| 421 | i = 0; |
| 422 | do { |
| 423 | if (!cpumask_test_cpu(i, &mask)) |
| 424 | continue; |
| 425 | |
| 426 | while (!ccall_info.processors_in[i]) |
| 427 | barrier(); |
| 428 | } while (++i <= high); |
| 429 | |
| 430 | i = 0; |
| 431 | do { |
| 432 | if (!cpumask_test_cpu(i, &mask)) |
| 433 | continue; |
| 434 | |
| 435 | while (!ccall_info.processors_out[i]) |
| 436 | barrier(); |
| 437 | } while (++i <= high); |
| 438 | } |
| 439 | |
| 440 | spin_unlock_irqrestore(&cross_call_lock, flags); |
| 441 | } |
| 442 | } |
| 443 | |
| 444 | /* Running cross calls. */ |
| 445 | void leon_cross_call_irq(void) |
| 446 | { |
| 447 | void (*func)(unsigned long, unsigned long, unsigned long, unsigned long, |
| 448 | unsigned long) = ccall_info.func; |
| 449 | int i = smp_processor_id(); |
| 450 | |
| 451 | ccall_info.processors_in[i] = 1; |
| 452 | func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3, ccall_info.arg4, |
| 453 | ccall_info.arg5); |
| 454 | ccall_info.processors_out[i] = 1; |
| 455 | } |
| 456 | |
| 457 | static const struct sparc32_ipi_ops leon_ipi_ops = { |
| 458 | .cross_call = leon_cross_call, |
| 459 | .resched = leon_ipi_resched, |
| 460 | .single = leon_ipi_single, |
| 461 | .mask_one = leon_ipi_mask_one, |
| 462 | }; |
| 463 | |
| 464 | void __init leon_init_smp(void) |
| 465 | { |
| 466 | /* Patch ipi15 trap table */ |
| 467 | t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_leon - linux_trap_ipi15_sun4m); |
| 468 | |
| 469 | sparc32_ipi_ops = &leon_ipi_ops; |
| 470 | } |