| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * Copyright 2011 IBM Corporation. |
| 4 | */ |
| 5 | #include <linux/types.h> |
| 6 | #include <linux/threads.h> |
| 7 | #include <linux/kernel.h> |
| 8 | #include <linux/irq.h> |
| 9 | #include <linux/irqdomain.h> |
| 10 | #include <linux/debugfs.h> |
| 11 | #include <linux/smp.h> |
| 12 | #include <linux/interrupt.h> |
| 13 | #include <linux/seq_file.h> |
| 14 | #include <linux/init.h> |
| 15 | #include <linux/cpu.h> |
| 16 | #include <linux/of.h> |
| 17 | #include <linux/slab.h> |
| 18 | #include <linux/spinlock.h> |
| 19 | #include <linux/delay.h> |
| 20 | |
| 21 | #include <asm/io.h> |
| 22 | #include <asm/smp.h> |
| 23 | #include <asm/machdep.h> |
| 24 | #include <asm/irq.h> |
| 25 | #include <asm/errno.h> |
| 26 | #include <asm/rtas.h> |
| 27 | #include <asm/xics.h> |
| 28 | #include <asm/firmware.h> |
| 29 | |
| 30 | /* Globals common to all ICP/ICS implementations */ |
| 31 | const struct icp_ops *icp_ops; |
| 32 | |
| 33 | unsigned int xics_default_server = 0xff; |
| 34 | unsigned int xics_default_distrib_server = 0; |
| 35 | unsigned int xics_interrupt_server_size = 8; |
| 36 | |
| 37 | DEFINE_PER_CPU(struct xics_cppr, xics_cppr); |
| 38 | |
| 39 | struct irq_domain *xics_host; |
| 40 | |
| 41 | static struct ics *xics_ics; |
| 42 | |
| 43 | void xics_update_irq_servers(void) |
| 44 | { |
| 45 | int i, j; |
| 46 | struct device_node *np; |
| 47 | u32 ilen; |
| 48 | const __be32 *ireg; |
| 49 | u32 hcpuid; |
| 50 | |
| 51 | /* Find the server numbers for the boot cpu. */ |
| 52 | np = of_get_cpu_node(cpu: boot_cpuid, NULL); |
| 53 | BUG_ON(!np); |
| 54 | |
| 55 | hcpuid = get_hard_smp_processor_id(boot_cpuid); |
| 56 | xics_default_server = xics_default_distrib_server = hcpuid; |
| 57 | |
| 58 | pr_devel("xics: xics_default_server = 0x%x\n" , xics_default_server); |
| 59 | |
| 60 | ireg = of_get_property(node: np, name: "ibm,ppc-interrupt-gserver#s" , lenp: &ilen); |
| 61 | if (!ireg) { |
| 62 | of_node_put(node: np); |
| 63 | return; |
| 64 | } |
| 65 | |
| 66 | i = ilen / sizeof(int); |
| 67 | |
| 68 | /* Global interrupt distribution server is specified in the last |
| 69 | * entry of "ibm,ppc-interrupt-gserver#s" property. Get the last |
| 70 | * entry fom this property for current boot cpu id and use it as |
| 71 | * default distribution server |
| 72 | */ |
| 73 | for (j = 0; j < i; j += 2) { |
| 74 | if (be32_to_cpu(ireg[j]) == hcpuid) { |
| 75 | xics_default_distrib_server = be32_to_cpu(ireg[j+1]); |
| 76 | break; |
| 77 | } |
| 78 | } |
| 79 | pr_devel("xics: xics_default_distrib_server = 0x%x\n" , |
| 80 | xics_default_distrib_server); |
| 81 | of_node_put(node: np); |
| 82 | } |
| 83 | |
| 84 | /* GIQ stuff, currently only supported on RTAS setups, will have |
| 85 | * to be sorted properly for bare metal |
| 86 | */ |
| 87 | void xics_set_cpu_giq(unsigned int gserver, unsigned int join) |
| 88 | { |
| 89 | #ifdef CONFIG_PPC_RTAS |
| 90 | int index; |
| 91 | int status; |
| 92 | |
| 93 | if (!rtas_indicator_present(GLOBAL_INTERRUPT_QUEUE, NULL)) |
| 94 | return; |
| 95 | |
| 96 | index = (1UL << xics_interrupt_server_size) - 1 - gserver; |
| 97 | |
| 98 | status = rtas_set_indicator_fast(GLOBAL_INTERRUPT_QUEUE, index, join); |
| 99 | |
| 100 | WARN(status < 0, "set-indicator(%d, %d, %u) returned %d\n" , |
| 101 | GLOBAL_INTERRUPT_QUEUE, index, join, status); |
| 102 | #endif |
| 103 | } |
| 104 | |
| 105 | void xics_setup_cpu(void) |
| 106 | { |
| 107 | icp_ops->set_priority(LOWEST_PRIORITY); |
| 108 | |
| 109 | xics_set_cpu_giq(gserver: xics_default_distrib_server, join: 1); |
| 110 | } |
| 111 | |
| 112 | void xics_mask_unknown_vec(unsigned int vec) |
| 113 | { |
| 114 | pr_err("Interrupt 0x%x (real) is invalid, disabling it.\n" , vec); |
| 115 | |
| 116 | if (WARN_ON(!xics_ics)) |
| 117 | return; |
| 118 | xics_ics->mask_unknown(xics_ics, vec); |
| 119 | } |
| 120 | |
| 121 | |
| 122 | #ifdef CONFIG_SMP |
| 123 | |
| 124 | static void __init xics_request_ipi(void) |
| 125 | { |
| 126 | unsigned int ipi; |
| 127 | |
| 128 | ipi = irq_create_mapping(domain: xics_host, hwirq: XICS_IPI); |
| 129 | BUG_ON(!ipi); |
| 130 | |
| 131 | /* |
| 132 | * IPIs are marked IRQF_PERCPU. The handler was set in map. |
| 133 | */ |
| 134 | BUG_ON(request_irq(ipi, icp_ops->ipi_action, |
| 135 | IRQF_NO_DEBUG | IRQF_PERCPU | IRQF_NO_THREAD, "IPI" , NULL)); |
| 136 | } |
| 137 | |
| 138 | void __init xics_smp_probe(void) |
| 139 | { |
| 140 | /* Register all the IPIs */ |
| 141 | xics_request_ipi(); |
| 142 | |
| 143 | /* Setup cause_ipi callback based on which ICP is used */ |
| 144 | smp_ops->cause_ipi = icp_ops->cause_ipi; |
| 145 | } |
| 146 | |
| 147 | #endif /* CONFIG_SMP */ |
| 148 | |
| 149 | noinstr void xics_teardown_cpu(void) |
| 150 | { |
| 151 | struct xics_cppr *os_cppr = this_cpu_ptr(&xics_cppr); |
| 152 | |
| 153 | /* |
| 154 | * we have to reset the cppr index to 0 because we're |
| 155 | * not going to return from the IPI |
| 156 | */ |
| 157 | os_cppr->index = 0; |
| 158 | icp_ops->set_priority(0); |
| 159 | icp_ops->teardown_cpu(); |
| 160 | } |
| 161 | |
| 162 | noinstr void xics_kexec_teardown_cpu(int secondary) |
| 163 | { |
| 164 | xics_teardown_cpu(); |
| 165 | |
| 166 | icp_ops->flush_ipi(); |
| 167 | |
| 168 | /* |
| 169 | * Some machines need to have at least one cpu in the GIQ, |
| 170 | * so leave the master cpu in the group. |
| 171 | */ |
| 172 | if (secondary) |
| 173 | xics_set_cpu_giq(gserver: xics_default_distrib_server, join: 0); |
| 174 | } |
| 175 | |
| 176 | |
| 177 | #ifdef CONFIG_HOTPLUG_CPU |
| 178 | |
| 179 | /* Interrupts are disabled. */ |
| 180 | void xics_migrate_irqs_away(void) |
| 181 | { |
| 182 | int cpu = smp_processor_id(), hw_cpu = hard_smp_processor_id(); |
| 183 | unsigned int irq, virq; |
| 184 | struct irq_desc *desc; |
| 185 | |
| 186 | pr_debug("%s: CPU %u\n" , __func__, cpu); |
| 187 | |
| 188 | /* If we used to be the default server, move to the new "boot_cpuid" */ |
| 189 | if (hw_cpu == xics_default_server) |
| 190 | xics_update_irq_servers(); |
| 191 | |
| 192 | /* Reject any interrupt that was queued to us... */ |
| 193 | icp_ops->set_priority(0); |
| 194 | |
| 195 | /* Remove ourselves from the global interrupt queue */ |
| 196 | xics_set_cpu_giq(gserver: xics_default_distrib_server, join: 0); |
| 197 | |
| 198 | for_each_irq_desc(virq, desc) { |
| 199 | struct irq_chip *chip; |
| 200 | long server; |
| 201 | unsigned long flags; |
| 202 | struct irq_data *irqd; |
| 203 | |
| 204 | /* We can't set affinity on ISA interrupts */ |
| 205 | if (virq < NR_IRQS_LEGACY) |
| 206 | continue; |
| 207 | /* We only need to migrate enabled IRQS */ |
| 208 | if (!desc->action) |
| 209 | continue; |
| 210 | /* We need a mapping in the XICS IRQ domain */ |
| 211 | irqd = irq_domain_get_irq_data(domain: xics_host, virq); |
| 212 | if (!irqd) |
| 213 | continue; |
| 214 | irq = irqd_to_hwirq(d: irqd); |
| 215 | /* We need to get IPIs still. */ |
| 216 | if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS) |
| 217 | continue; |
| 218 | chip = irq_desc_get_chip(desc); |
| 219 | if (!chip || !chip->irq_set_affinity) |
| 220 | continue; |
| 221 | |
| 222 | raw_spin_lock_irqsave(&desc->lock, flags); |
| 223 | |
| 224 | /* Locate interrupt server */ |
| 225 | server = xics_ics->get_server(xics_ics, irq); |
| 226 | if (server < 0) { |
| 227 | pr_err("%s: Can't find server for irq %d/%x\n" , |
| 228 | __func__, virq, irq); |
| 229 | goto unlock; |
| 230 | } |
| 231 | |
| 232 | /* We only support delivery to all cpus or to one cpu. |
| 233 | * The irq has to be migrated only in the single cpu |
| 234 | * case. |
| 235 | */ |
| 236 | if (server != hw_cpu) |
| 237 | goto unlock; |
| 238 | |
| 239 | /* This is expected during cpu offline. */ |
| 240 | if (cpu_online(cpu)) |
| 241 | pr_warn("IRQ %u affinity broken off cpu %u\n" , |
| 242 | virq, cpu); |
| 243 | |
| 244 | /* Reset affinity to all cpus */ |
| 245 | raw_spin_unlock_irqrestore(&desc->lock, flags); |
| 246 | irq_set_affinity(irq: virq, cpu_all_mask); |
| 247 | continue; |
| 248 | unlock: |
| 249 | raw_spin_unlock_irqrestore(&desc->lock, flags); |
| 250 | } |
| 251 | |
| 252 | /* Allow "sufficient" time to drop any inflight IRQ's */ |
| 253 | mdelay(5); |
| 254 | |
| 255 | /* |
| 256 | * Allow IPIs again. This is done at the very end, after migrating all |
| 257 | * interrupts, the expectation is that we'll only get woken up by an IPI |
| 258 | * interrupt beyond this point, but leave externals masked just to be |
| 259 | * safe. If we're using icp-opal this may actually allow all |
| 260 | * interrupts anyway, but that should be OK. |
| 261 | */ |
| 262 | icp_ops->set_priority(DEFAULT_PRIORITY); |
| 263 | |
| 264 | } |
| 265 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 266 | |
| 267 | #ifdef CONFIG_SMP |
| 268 | /* |
| 269 | * For the moment we only implement delivery to all cpus or one cpu. |
| 270 | * |
| 271 | * If the requested affinity is cpu_all_mask, we set global affinity. |
| 272 | * If not we set it to the first cpu in the mask, even if multiple cpus |
| 273 | * are set. This is so things like irqbalance (which set core and package |
| 274 | * wide affinities) do the right thing. |
| 275 | * |
| 276 | * We need to fix this to implement support for the links |
| 277 | */ |
| 278 | int xics_get_irq_server(unsigned int virq, const struct cpumask *cpumask, |
| 279 | unsigned int strict_check) |
| 280 | { |
| 281 | |
| 282 | if (!distribute_irqs) |
| 283 | return xics_default_server; |
| 284 | |
| 285 | if (!cpumask_subset(cpu_possible_mask, src2p: cpumask)) { |
| 286 | int server = cpumask_first_and(cpu_online_mask, srcp2: cpumask); |
| 287 | |
| 288 | if (server < nr_cpu_ids) |
| 289 | return get_hard_smp_processor_id(server); |
| 290 | |
| 291 | if (strict_check) |
| 292 | return -1; |
| 293 | } |
| 294 | |
| 295 | /* |
| 296 | * Workaround issue with some versions of JS20 firmware that |
| 297 | * deliver interrupts to cpus which haven't been started. This |
| 298 | * happens when using the maxcpus= boot option. |
| 299 | */ |
| 300 | if (cpumask_equal(cpu_online_mask, cpu_present_mask)) |
| 301 | return xics_default_distrib_server; |
| 302 | |
| 303 | return xics_default_server; |
| 304 | } |
| 305 | #endif /* CONFIG_SMP */ |
| 306 | |
| 307 | static int xics_host_match(struct irq_domain *h, struct device_node *node, |
| 308 | enum irq_domain_bus_token bus_token) |
| 309 | { |
| 310 | if (WARN_ON(!xics_ics)) |
| 311 | return 0; |
| 312 | return xics_ics->host_match(xics_ics, node) ? 1 : 0; |
| 313 | } |
| 314 | |
| 315 | /* Dummies */ |
| 316 | static void xics_ipi_unmask(struct irq_data *d) { } |
| 317 | static void xics_ipi_mask(struct irq_data *d) { } |
| 318 | |
| 319 | static struct irq_chip xics_ipi_chip = { |
| 320 | .name = "XICS" , |
| 321 | .irq_eoi = NULL, /* Patched at init time */ |
| 322 | .irq_mask = xics_ipi_mask, |
| 323 | .irq_unmask = xics_ipi_unmask, |
| 324 | }; |
| 325 | |
| 326 | static int xics_host_map(struct irq_domain *domain, unsigned int virq, |
| 327 | irq_hw_number_t hwirq) |
| 328 | { |
| 329 | pr_devel("xics: map virq %d, hwirq 0x%lx\n" , virq, hwirq); |
| 330 | |
| 331 | /* |
| 332 | * Mark interrupts as edge sensitive by default so that resend |
| 333 | * actually works. The device-tree parsing will turn the LSIs |
| 334 | * back to level. |
| 335 | */ |
| 336 | irq_clear_status_flags(irq: virq, clr: IRQ_LEVEL); |
| 337 | |
| 338 | /* Don't call into ICS for IPIs */ |
| 339 | if (hwirq == XICS_IPI) { |
| 340 | irq_set_chip_and_handler(irq: virq, chip: &xics_ipi_chip, |
| 341 | handle: handle_percpu_irq); |
| 342 | return 0; |
| 343 | } |
| 344 | |
| 345 | if (WARN_ON(!xics_ics)) |
| 346 | return -EINVAL; |
| 347 | |
| 348 | if (xics_ics->check(xics_ics, hwirq)) |
| 349 | return -EINVAL; |
| 350 | |
| 351 | /* Let the ICS be the chip data for the XICS domain. For ICS native */ |
| 352 | irq_domain_set_info(domain, virq, hwirq, chip: xics_ics->chip, |
| 353 | chip_data: xics_ics, handler: handle_fasteoi_irq, NULL, NULL); |
| 354 | |
| 355 | return 0; |
| 356 | } |
| 357 | |
| 358 | static int xics_host_xlate(struct irq_domain *h, struct device_node *ct, |
| 359 | const u32 *intspec, unsigned int intsize, |
| 360 | irq_hw_number_t *out_hwirq, unsigned int *out_flags) |
| 361 | |
| 362 | { |
| 363 | *out_hwirq = intspec[0]; |
| 364 | |
| 365 | /* |
| 366 | * If intsize is at least 2, we look for the type in the second cell, |
| 367 | * we assume the LSB indicates a level interrupt. |
| 368 | */ |
| 369 | if (intsize > 1) { |
| 370 | if (intspec[1] & 1) |
| 371 | *out_flags = IRQ_TYPE_LEVEL_LOW; |
| 372 | else |
| 373 | *out_flags = IRQ_TYPE_EDGE_RISING; |
| 374 | } else |
| 375 | *out_flags = IRQ_TYPE_LEVEL_LOW; |
| 376 | |
| 377 | return 0; |
| 378 | } |
| 379 | |
| 380 | int xics_set_irq_type(struct irq_data *d, unsigned int flow_type) |
| 381 | { |
| 382 | /* |
| 383 | * We only support these. This has really no effect other than setting |
| 384 | * the corresponding descriptor bits mind you but those will in turn |
| 385 | * affect the resend function when re-enabling an edge interrupt. |
| 386 | * |
| 387 | * Set set the default to edge as explained in map(). |
| 388 | */ |
| 389 | if (flow_type == IRQ_TYPE_DEFAULT || flow_type == IRQ_TYPE_NONE) |
| 390 | flow_type = IRQ_TYPE_EDGE_RISING; |
| 391 | |
| 392 | if (flow_type != IRQ_TYPE_EDGE_RISING && |
| 393 | flow_type != IRQ_TYPE_LEVEL_LOW) |
| 394 | return -EINVAL; |
| 395 | |
| 396 | irqd_set_trigger_type(d, type: flow_type); |
| 397 | |
| 398 | return IRQ_SET_MASK_OK_NOCOPY; |
| 399 | } |
| 400 | |
| 401 | int xics_retrigger(struct irq_data *data) |
| 402 | { |
| 403 | /* |
| 404 | * We need to push a dummy CPPR when retriggering, since the subsequent |
| 405 | * EOI will try to pop it. Passing 0 works, as the function hard codes |
| 406 | * the priority value anyway. |
| 407 | */ |
| 408 | xics_push_cppr(0); |
| 409 | |
| 410 | /* Tell the core to do a soft retrigger */ |
| 411 | return 0; |
| 412 | } |
| 413 | |
| 414 | #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY |
| 415 | static int xics_host_domain_translate(struct irq_domain *d, struct irq_fwspec *fwspec, |
| 416 | unsigned long *hwirq, unsigned int *type) |
| 417 | { |
| 418 | return xics_host_xlate(h: d, to_of_node(fwspec->fwnode), intspec: fwspec->param, |
| 419 | intsize: fwspec->param_count, out_hwirq: hwirq, out_flags: type); |
| 420 | } |
| 421 | |
| 422 | static int xics_host_domain_alloc(struct irq_domain *domain, unsigned int virq, |
| 423 | unsigned int nr_irqs, void *arg) |
| 424 | { |
| 425 | struct irq_fwspec *fwspec = arg; |
| 426 | irq_hw_number_t hwirq; |
| 427 | unsigned int type = IRQ_TYPE_NONE; |
| 428 | int i, rc; |
| 429 | |
| 430 | rc = xics_host_domain_translate(d: domain, fwspec, hwirq: &hwirq, type: &type); |
| 431 | if (rc) |
| 432 | return rc; |
| 433 | |
| 434 | pr_debug("%s %d/%lx #%d\n" , __func__, virq, hwirq, nr_irqs); |
| 435 | |
| 436 | for (i = 0; i < nr_irqs; i++) |
| 437 | irq_domain_set_info(domain, virq: virq + i, hwirq: hwirq + i, chip: xics_ics->chip, |
| 438 | chip_data: xics_ics, handler: handle_fasteoi_irq, NULL, NULL); |
| 439 | |
| 440 | return 0; |
| 441 | } |
| 442 | |
| 443 | static void xics_host_domain_free(struct irq_domain *domain, |
| 444 | unsigned int virq, unsigned int nr_irqs) |
| 445 | { |
| 446 | pr_debug("%s %d #%d\n" , __func__, virq, nr_irqs); |
| 447 | } |
| 448 | #endif |
| 449 | |
| 450 | static const struct irq_domain_ops xics_host_ops = { |
| 451 | #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY |
| 452 | .alloc = xics_host_domain_alloc, |
| 453 | .free = xics_host_domain_free, |
| 454 | .translate = xics_host_domain_translate, |
| 455 | #endif |
| 456 | .match = xics_host_match, |
| 457 | .map = xics_host_map, |
| 458 | .xlate = xics_host_xlate, |
| 459 | }; |
| 460 | |
| 461 | static int __init xics_allocate_domain(void) |
| 462 | { |
| 463 | struct fwnode_handle *fn; |
| 464 | |
| 465 | fn = irq_domain_alloc_named_fwnode(name: "XICS" ); |
| 466 | if (!fn) |
| 467 | return -ENOMEM; |
| 468 | |
| 469 | xics_host = irq_domain_create_tree(fwnode: fn, ops: &xics_host_ops, NULL); |
| 470 | if (!xics_host) { |
| 471 | irq_domain_free_fwnode(fwnode: fn); |
| 472 | return -ENOMEM; |
| 473 | } |
| 474 | |
| 475 | irq_set_default_domain(domain: xics_host); |
| 476 | return 0; |
| 477 | } |
| 478 | |
| 479 | void __init xics_register_ics(struct ics *ics) |
| 480 | { |
| 481 | if (WARN_ONCE(xics_ics, "XICS: Source Controller is already defined !" )) |
| 482 | return; |
| 483 | xics_ics = ics; |
| 484 | } |
| 485 | |
| 486 | static void __init xics_get_server_size(void) |
| 487 | { |
| 488 | struct device_node *np; |
| 489 | const __be32 *isize; |
| 490 | |
| 491 | /* We fetch the interrupt server size from the first ICS node |
| 492 | * we find if any |
| 493 | */ |
| 494 | np = of_find_compatible_node(NULL, NULL, compat: "ibm,ppc-xics" ); |
| 495 | if (!np) |
| 496 | return; |
| 497 | |
| 498 | isize = of_get_property(node: np, name: "ibm,interrupt-server#-size" , NULL); |
| 499 | if (isize) |
| 500 | xics_interrupt_server_size = be32_to_cpu(*isize); |
| 501 | |
| 502 | of_node_put(node: np); |
| 503 | } |
| 504 | |
| 505 | void __init xics_init(void) |
| 506 | { |
| 507 | int rc = -1; |
| 508 | |
| 509 | /* Fist locate ICP */ |
| 510 | if (firmware_has_feature(FW_FEATURE_LPAR)) |
| 511 | rc = icp_hv_init(); |
| 512 | if (rc < 0) { |
| 513 | rc = icp_native_init(); |
| 514 | if (rc == -ENODEV) |
| 515 | rc = icp_opal_init(); |
| 516 | } |
| 517 | if (rc < 0) { |
| 518 | pr_warn("XICS: Cannot find a Presentation Controller !\n" ); |
| 519 | return; |
| 520 | } |
| 521 | |
| 522 | /* Copy get_irq callback over to ppc_md */ |
| 523 | ppc_md.get_irq = icp_ops->get_irq; |
| 524 | |
| 525 | /* Patch up IPI chip EOI */ |
| 526 | xics_ipi_chip.irq_eoi = icp_ops->eoi; |
| 527 | |
| 528 | /* Now locate ICS */ |
| 529 | rc = ics_rtas_init(); |
| 530 | if (rc < 0) |
| 531 | rc = ics_opal_init(); |
| 532 | if (rc < 0) |
| 533 | rc = ics_native_init(); |
| 534 | if (rc < 0) |
| 535 | pr_warn("XICS: Cannot find a Source Controller !\n" ); |
| 536 | |
| 537 | /* Initialize common bits */ |
| 538 | xics_get_server_size(); |
| 539 | xics_update_irq_servers(); |
| 540 | rc = xics_allocate_domain(); |
| 541 | if (rc < 0) |
| 542 | pr_err("XICS: Failed to create IRQ domain" ); |
| 543 | xics_setup_cpu(); |
| 544 | } |
| 545 | |