linux/arch/powerpc/kernel/irq.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *  Derived from arch/i386/kernel/irq.c
   4 *    Copyright (C) 1992 Linus Torvalds
   5 *  Adapted from arch/i386 by Gary Thomas
   6 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   7 *  Updated and modified by Cort Dougan <cort@fsmlabs.com>
   8 *    Copyright (C) 1996-2001 Cort Dougan
   9 *  Adapted for Power Macintosh by Paul Mackerras
  10 *    Copyright (C) 1996 Paul Mackerras (paulus@cs.anu.edu.au)
  11 *
  12 * This file contains the code used by various IRQ handling routines:
  13 * asking for different IRQ's should be done through these routines
  14 * instead of just grabbing them. Thus setups with different IRQ numbers
  15 * shouldn't result in any weird surprises, and installing new handlers
  16 * should be easier.
  17 *
  18 * The MPC8xx has an interrupt mask in the SIU.  If a bit is set, the
  19 * interrupt is _enabled_.  As expected, IRQ0 is bit 0 in the 32-bit
  20 * mask register (of which only 16 are defined), hence the weird shifting
  21 * and complement of the cached_irq_mask.  I want to be able to stuff
  22 * this right into the SIU SMASK register.
  23 * Many of the prep/chrp functions are conditional compiled on CONFIG_PPC_8xx
  24 * to reduce code space and undefined function references.
  25 */
  26
  27#undef DEBUG
  28
  29#include <linux/export.h>
  30#include <linux/threads.h>
  31#include <linux/kernel_stat.h>
  32#include <linux/signal.h>
  33#include <linux/sched.h>
  34#include <linux/ptrace.h>
  35#include <linux/ioport.h>
  36#include <linux/interrupt.h>
  37#include <linux/timex.h>
  38#include <linux/init.h>
  39#include <linux/slab.h>
  40#include <linux/delay.h>
  41#include <linux/irq.h>
  42#include <linux/seq_file.h>
  43#include <linux/cpumask.h>
  44#include <linux/profile.h>
  45#include <linux/bitops.h>
  46#include <linux/list.h>
  47#include <linux/radix-tree.h>
  48#include <linux/mutex.h>
  49#include <linux/pci.h>
  50#include <linux/debugfs.h>
  51#include <linux/of.h>
  52#include <linux/of_irq.h>
  53#include <linux/vmalloc.h>
  54#include <linux/pgtable.h>
  55
  56#include <linux/uaccess.h>
  57#include <asm/interrupt.h>
  58#include <asm/io.h>
  59#include <asm/irq.h>
  60#include <asm/cache.h>
  61#include <asm/prom.h>
  62#include <asm/ptrace.h>
  63#include <asm/machdep.h>
  64#include <asm/udbg.h>
  65#include <asm/smp.h>
  66#include <asm/livepatch.h>
  67#include <asm/asm-prototypes.h>
  68#include <asm/hw_irq.h>
  69#include <asm/softirq_stack.h>
  70
  71#ifdef CONFIG_PPC64
  72#include <asm/paca.h>
  73#include <asm/firmware.h>
  74#include <asm/lv1call.h>
  75#include <asm/dbell.h>
  76#endif
  77#define CREATE_TRACE_POINTS
  78#include <asm/trace.h>
  79#include <asm/cpu_has_feature.h>
  80
  81DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
  82EXPORT_PER_CPU_SYMBOL(irq_stat);
  83
  84#ifdef CONFIG_PPC32
  85atomic_t ppc_n_lost_interrupts;
  86
  87#ifdef CONFIG_TAU_INT
  88extern int tau_initialized;
  89u32 tau_interrupts(unsigned long cpu);
  90#endif
  91#endif /* CONFIG_PPC32 */
  92
  93#ifdef CONFIG_PPC64
  94
  95int distribute_irqs = 1;
  96
  97static inline notrace unsigned long get_irq_happened(void)
  98{
  99        unsigned long happened;
 100
 101        __asm__ __volatile__("lbz %0,%1(13)"
 102        : "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
 103
 104        return happened;
 105}
 106
 107void replay_soft_interrupts(void)
 108{
 109        struct pt_regs regs;
 110
 111        /*
 112         * Be careful here, calling these interrupt handlers can cause
 113         * softirqs to be raised, which they may run when calling irq_exit,
 114         * which will cause local_irq_enable() to be run, which can then
 115         * recurse into this function. Don't keep any state across
 116         * interrupt handler calls which may change underneath us.
 117         *
 118         * We use local_paca rather than get_paca() to avoid all the
 119         * debug_smp_processor_id() business in this low level function.
 120         */
 121
 122        ppc_save_regs(&regs);
 123        regs.softe = IRQS_ENABLED;
 124
 125again:
 126        if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
 127                WARN_ON_ONCE(mfmsr() & MSR_EE);
 128
 129        /*
 130         * Force the delivery of pending soft-disabled interrupts on PS3.
 131         * Any HV call will have this side effect.
 132         */
 133        if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
 134                u64 tmp, tmp2;
 135                lv1_get_version_info(&tmp, &tmp2);
 136        }
 137
 138        /*
 139         * Check if an hypervisor Maintenance interrupt happened.
 140         * This is a higher priority interrupt than the others, so
 141         * replay it first.
 142         */
 143        if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (local_paca->irq_happened & PACA_IRQ_HMI)) {
 144                local_paca->irq_happened &= ~PACA_IRQ_HMI;
 145                regs.trap = INTERRUPT_HMI;
 146                handle_hmi_exception(&regs);
 147                if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
 148                        hard_irq_disable();
 149        }
 150
 151        if (local_paca->irq_happened & PACA_IRQ_DEC) {
 152                local_paca->irq_happened &= ~PACA_IRQ_DEC;
 153                regs.trap = INTERRUPT_DECREMENTER;
 154                timer_interrupt(&regs);
 155                if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
 156                        hard_irq_disable();
 157        }
 158
 159        if (local_paca->irq_happened & PACA_IRQ_EE) {
 160                local_paca->irq_happened &= ~PACA_IRQ_EE;
 161                regs.trap = INTERRUPT_EXTERNAL;
 162                do_IRQ(&regs);
 163                if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
 164                        hard_irq_disable();
 165        }
 166
 167        if (IS_ENABLED(CONFIG_PPC_DOORBELL) && (local_paca->irq_happened & PACA_IRQ_DBELL)) {
 168                local_paca->irq_happened &= ~PACA_IRQ_DBELL;
 169                regs.trap = INTERRUPT_DOORBELL;
 170                doorbell_exception(&regs);
 171                if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
 172                        hard_irq_disable();
 173        }
 174
 175        /* Book3E does not support soft-masking PMI interrupts */
 176        if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (local_paca->irq_happened & PACA_IRQ_PMI)) {
 177                local_paca->irq_happened &= ~PACA_IRQ_PMI;
 178                regs.trap = INTERRUPT_PERFMON;
 179                performance_monitor_exception(&regs);
 180                if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
 181                        hard_irq_disable();
 182        }
 183
 184        if (local_paca->irq_happened & ~PACA_IRQ_HARD_DIS) {
 185                /*
 186                 * We are responding to the next interrupt, so interrupt-off
 187                 * latencies should be reset here.
 188                 */
 189                trace_hardirqs_on();
 190                trace_hardirqs_off();
 191                goto again;
 192        }
 193}
 194
 195#if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_KUAP)
 196static inline void replay_soft_interrupts_irqrestore(void)
 197{
 198        unsigned long kuap_state = get_kuap();
 199
 200        /*
 201         * Check if anything calls local_irq_enable/restore() when KUAP is
 202         * disabled (user access enabled). We handle that case here by saving
 203         * and re-locking AMR but we shouldn't get here in the first place,
 204         * hence the warning.
 205         */
 206        kuap_assert_locked();
 207
 208        if (kuap_state != AMR_KUAP_BLOCKED)
 209                set_kuap(AMR_KUAP_BLOCKED);
 210
 211        replay_soft_interrupts();
 212
 213        if (kuap_state != AMR_KUAP_BLOCKED)
 214                set_kuap(kuap_state);
 215}
 216#else
 217#define replay_soft_interrupts_irqrestore() replay_soft_interrupts()
 218#endif
 219
 220notrace void arch_local_irq_restore(unsigned long mask)
 221{
 222        unsigned char irq_happened;
 223
 224        /* Write the new soft-enabled value */
 225        irq_soft_mask_set(mask);
 226        if (mask)
 227                return;
 228
 229        /*
 230         * From this point onward, we can take interrupts, preempt,
 231         * etc... unless we got hard-disabled. We check if an event
 232         * happened. If none happened, we know we can just return.
 233         *
 234         * We may have preempted before the check below, in which case
 235         * we are checking the "new" CPU instead of the old one. This
 236         * is only a problem if an event happened on the "old" CPU.
 237         *
 238         * External interrupt events will have caused interrupts to
 239         * be hard-disabled, so there is no problem, we
 240         * cannot have preempted.
 241         */
 242        irq_happened = get_irq_happened();
 243        if (!irq_happened) {
 244                if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
 245                        WARN_ON_ONCE(!(mfmsr() & MSR_EE));
 246                return;
 247        }
 248
 249        /* We need to hard disable to replay. */
 250        if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
 251                if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
 252                        WARN_ON_ONCE(!(mfmsr() & MSR_EE));
 253                __hard_irq_disable();
 254                local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
 255        } else {
 256                /*
 257                 * We should already be hard disabled here. We had bugs
 258                 * where that wasn't the case so let's dbl check it and
 259                 * warn if we are wrong. Only do that when IRQ tracing
 260                 * is enabled as mfmsr() can be costly.
 261                 */
 262                if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
 263                        if (WARN_ON_ONCE(mfmsr() & MSR_EE))
 264                                __hard_irq_disable();
 265                }
 266
 267                if (irq_happened == PACA_IRQ_HARD_DIS) {
 268                        local_paca->irq_happened = 0;
 269                        __hard_irq_enable();
 270                        return;
 271                }
 272        }
 273
 274        /*
 275         * Disable preempt here, so that the below preempt_enable will
 276         * perform resched if required (a replayed interrupt may set
 277         * need_resched).
 278         */
 279        preempt_disable();
 280        irq_soft_mask_set(IRQS_ALL_DISABLED);
 281        trace_hardirqs_off();
 282
 283        replay_soft_interrupts_irqrestore();
 284        local_paca->irq_happened = 0;
 285
 286        trace_hardirqs_on();
 287        irq_soft_mask_set(IRQS_ENABLED);
 288        __hard_irq_enable();
 289        preempt_enable();
 290}
 291EXPORT_SYMBOL(arch_local_irq_restore);
 292
 293/*
 294 * This is a helper to use when about to go into idle low-power
 295 * when the latter has the side effect of re-enabling interrupts
 296 * (such as calling H_CEDE under pHyp).
 297 *
 298 * You call this function with interrupts soft-disabled (this is
 299 * already the case when ppc_md.power_save is called). The function
 300 * will return whether to enter power save or just return.
 301 *
 302 * In the former case, it will have notified lockdep of interrupts
 303 * being re-enabled and generally sanitized the lazy irq state,
 304 * and in the latter case it will leave with interrupts hard
 305 * disabled and marked as such, so the local_irq_enable() call
 306 * in arch_cpu_idle() will properly re-enable everything.
 307 */
 308bool prep_irq_for_idle(void)
 309{
 310        /*
 311         * First we need to hard disable to ensure no interrupt
 312         * occurs before we effectively enter the low power state
 313         */
 314        __hard_irq_disable();
 315        local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
 316
 317        /*
 318         * If anything happened while we were soft-disabled,
 319         * we return now and do not enter the low power state.
 320         */
 321        if (lazy_irq_pending())
 322                return false;
 323
 324        /* Tell lockdep we are about to re-enable */
 325        trace_hardirqs_on();
 326
 327        /*
 328         * Mark interrupts as soft-enabled and clear the
 329         * PACA_IRQ_HARD_DIS from the pending mask since we
 330         * are about to hard enable as well as a side effect
 331         * of entering the low power state.
 332         */
 333        local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
 334        irq_soft_mask_set(IRQS_ENABLED);
 335
 336        /* Tell the caller to enter the low power state */
 337        return true;
 338}
 339
 340#ifdef CONFIG_PPC_BOOK3S
 341/*
 342 * This is for idle sequences that return with IRQs off, but the
 343 * idle state itself wakes on interrupt. Tell the irq tracer that
 344 * IRQs are enabled for the duration of idle so it does not get long
 345 * off times. Must be paired with fini_irq_for_idle_irqsoff.
 346 */
 347bool prep_irq_for_idle_irqsoff(void)
 348{
 349        WARN_ON(!irqs_disabled());
 350
 351        /*
 352         * First we need to hard disable to ensure no interrupt
 353         * occurs before we effectively enter the low power state
 354         */
 355        __hard_irq_disable();
 356        local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
 357
 358        /*
 359         * If anything happened while we were soft-disabled,
 360         * we return now and do not enter the low power state.
 361         */
 362        if (lazy_irq_pending())
 363                return false;
 364
 365        /* Tell lockdep we are about to re-enable */
 366        trace_hardirqs_on();
 367
 368        return true;
 369}
 370
 371/*
 372 * Take the SRR1 wakeup reason, index into this table to find the
 373 * appropriate irq_happened bit.
 374 *
 375 * Sytem reset exceptions taken in idle state also come through here,
 376 * but they are NMI interrupts so do not need to wait for IRQs to be
 377 * restored, and should be taken as early as practical. These are marked
 378 * with 0xff in the table. The Power ISA specifies 0100b as the system
 379 * reset interrupt reason.
 380 */
 381#define IRQ_SYSTEM_RESET        0xff
 382
 383static const u8 srr1_to_lazyirq[0x10] = {
 384        0, 0, 0,
 385        PACA_IRQ_DBELL,
 386        IRQ_SYSTEM_RESET,
 387        PACA_IRQ_DBELL,
 388        PACA_IRQ_DEC,
 389        0,
 390        PACA_IRQ_EE,
 391        PACA_IRQ_EE,
 392        PACA_IRQ_HMI,
 393        0, 0, 0, 0, 0 };
 394
 395void replay_system_reset(void)
 396{
 397        struct pt_regs regs;
 398
 399        ppc_save_regs(&regs);
 400        regs.trap = 0x100;
 401        get_paca()->in_nmi = 1;
 402        system_reset_exception(&regs);
 403        get_paca()->in_nmi = 0;
 404}
 405EXPORT_SYMBOL_GPL(replay_system_reset);
 406
 407void irq_set_pending_from_srr1(unsigned long srr1)
 408{
 409        unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
 410        u8 reason = srr1_to_lazyirq[idx];
 411
 412        /*
 413         * Take the system reset now, which is immediately after registers
 414         * are restored from idle. It's an NMI, so interrupts need not be
 415         * re-enabled before it is taken.
 416         */
 417        if (unlikely(reason == IRQ_SYSTEM_RESET)) {
 418                replay_system_reset();
 419                return;
 420        }
 421
 422        if (reason == PACA_IRQ_DBELL) {
 423                /*
 424                 * When doorbell triggers a system reset wakeup, the message
 425                 * is not cleared, so if the doorbell interrupt is replayed
 426                 * and the IPI handled, the doorbell interrupt would still
 427                 * fire when EE is enabled.
 428                 *
 429                 * To avoid taking the superfluous doorbell interrupt,
 430                 * execute a msgclr here before the interrupt is replayed.
 431                 */
 432                ppc_msgclr(PPC_DBELL_MSGTYPE);
 433        }
 434
 435        /*
 436         * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
 437         * so this can be called unconditionally with the SRR1 wake
 438         * reason as returned by the idle code, which uses 0 to mean no
 439         * interrupt.
 440         *
 441         * If a future CPU was to designate this as an interrupt reason,
 442         * then a new index for no interrupt must be assigned.
 443         */
 444        local_paca->irq_happened |= reason;
 445}
 446#endif /* CONFIG_PPC_BOOK3S */
 447
 448/*
 449 * Force a replay of the external interrupt handler on this CPU.
 450 */
 451void force_external_irq_replay(void)
 452{
 453        /*
 454         * This must only be called with interrupts soft-disabled,
 455         * the replay will happen when re-enabling.
 456         */
 457        WARN_ON(!arch_irqs_disabled());
 458
 459        /*
 460         * Interrupts must always be hard disabled before irq_happened is
 461         * modified (to prevent lost update in case of interrupt between
 462         * load and store).
 463         */
 464        __hard_irq_disable();
 465        local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
 466
 467        /* Indicate in the PACA that we have an interrupt to replay */
 468        local_paca->irq_happened |= PACA_IRQ_EE;
 469}
 470
 471#endif /* CONFIG_PPC64 */
 472
 473int arch_show_interrupts(struct seq_file *p, int prec)
 474{
 475        int j;
 476
 477#if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
 478        if (tau_initialized) {
 479                seq_printf(p, "%*s: ", prec, "TAU");
 480                for_each_online_cpu(j)
 481                        seq_printf(p, "%10u ", tau_interrupts(j));
 482                seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
 483        }
 484#endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
 485
 486        seq_printf(p, "%*s: ", prec, "LOC");
 487        for_each_online_cpu(j)
 488                seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
 489        seq_printf(p, "  Local timer interrupts for timer event device\n");
 490
 491        seq_printf(p, "%*s: ", prec, "BCT");
 492        for_each_online_cpu(j)
 493                seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
 494        seq_printf(p, "  Broadcast timer interrupts for timer event device\n");
 495
 496        seq_printf(p, "%*s: ", prec, "LOC");
 497        for_each_online_cpu(j)
 498                seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
 499        seq_printf(p, "  Local timer interrupts for others\n");
 500
 501        seq_printf(p, "%*s: ", prec, "SPU");
 502        for_each_online_cpu(j)
 503                seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
 504        seq_printf(p, "  Spurious interrupts\n");
 505
 506        seq_printf(p, "%*s: ", prec, "PMI");
 507        for_each_online_cpu(j)
 508                seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
 509        seq_printf(p, "  Performance monitoring interrupts\n");
 510
 511        seq_printf(p, "%*s: ", prec, "MCE");
 512        for_each_online_cpu(j)
 513                seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
 514        seq_printf(p, "  Machine check exceptions\n");
 515
 516#ifdef CONFIG_PPC_BOOK3S_64
 517        if (cpu_has_feature(CPU_FTR_HVMODE)) {
 518                seq_printf(p, "%*s: ", prec, "HMI");
 519                for_each_online_cpu(j)
 520                        seq_printf(p, "%10u ", paca_ptrs[j]->hmi_irqs);
 521                seq_printf(p, "  Hypervisor Maintenance Interrupts\n");
 522        }
 523#endif
 524
 525        seq_printf(p, "%*s: ", prec, "NMI");
 526        for_each_online_cpu(j)
 527                seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
 528        seq_printf(p, "  System Reset interrupts\n");
 529
 530#ifdef CONFIG_PPC_WATCHDOG
 531        seq_printf(p, "%*s: ", prec, "WDG");
 532        for_each_online_cpu(j)
 533                seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
 534        seq_printf(p, "  Watchdog soft-NMI interrupts\n");
 535#endif
 536
 537#ifdef CONFIG_PPC_DOORBELL
 538        if (cpu_has_feature(CPU_FTR_DBELL)) {
 539                seq_printf(p, "%*s: ", prec, "DBL");
 540                for_each_online_cpu(j)
 541                        seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
 542                seq_printf(p, "  Doorbell interrupts\n");
 543        }
 544#endif
 545
 546        return 0;
 547}
 548
 549/*
 550 * /proc/stat helpers
 551 */
 552u64 arch_irq_stat_cpu(unsigned int cpu)
 553{
 554        u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
 555
 556        sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
 557        sum += per_cpu(irq_stat, cpu).pmu_irqs;
 558        sum += per_cpu(irq_stat, cpu).mce_exceptions;
 559        sum += per_cpu(irq_stat, cpu).spurious_irqs;
 560        sum += per_cpu(irq_stat, cpu).timer_irqs_others;
 561#ifdef CONFIG_PPC_BOOK3S_64
 562        sum += paca_ptrs[cpu]->hmi_irqs;
 563#endif
 564        sum += per_cpu(irq_stat, cpu).sreset_irqs;
 565#ifdef CONFIG_PPC_WATCHDOG
 566        sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
 567#endif
 568#ifdef CONFIG_PPC_DOORBELL
 569        sum += per_cpu(irq_stat, cpu).doorbell_irqs;
 570#endif
 571
 572        return sum;
 573}
 574
 575static inline void check_stack_overflow(void)
 576{
 577        long sp;
 578
 579        if (!IS_ENABLED(CONFIG_DEBUG_STACKOVERFLOW))
 580                return;
 581
 582        sp = current_stack_pointer & (THREAD_SIZE - 1);
 583
 584        /* check for stack overflow: is there less than 2KB free? */
 585        if (unlikely(sp < 2048)) {
 586                pr_err("do_IRQ: stack overflow: %ld\n", sp);
 587                dump_stack();
 588        }
 589}
 590
 591static __always_inline void call_do_softirq(const void *sp)
 592{
 593        /* Temporarily switch r1 to sp, call __do_softirq() then restore r1. */
 594        asm volatile (
 595                 PPC_STLU "     %%r1, %[offset](%[sp])  ;"
 596                "mr             %%r1, %[sp]             ;"
 597                "bl             %[callee]               ;"
 598                 PPC_LL "       %%r1, 0(%%r1)           ;"
 599                 : // Outputs
 600                 : // Inputs
 601                   [sp] "b" (sp), [offset] "i" (THREAD_SIZE - STACK_FRAME_OVERHEAD),
 602                   [callee] "i" (__do_softirq)
 603                 : // Clobbers
 604                   "lr", "xer", "ctr", "memory", "cr0", "cr1", "cr5", "cr6",
 605                   "cr7", "r0", "r3", "r4", "r5", "r6", "r7", "r8", "r9", "r10",
 606                   "r11", "r12"
 607        );
 608}
 609
 610static __always_inline void call_do_irq(struct pt_regs *regs, void *sp)
 611{
 612        register unsigned long r3 asm("r3") = (unsigned long)regs;
 613
 614        /* Temporarily switch r1 to sp, call __do_irq() then restore r1. */
 615        asm volatile (
 616                 PPC_STLU "     %%r1, %[offset](%[sp])  ;"
 617                "mr             %%r1, %[sp]             ;"
 618                "bl             %[callee]               ;"
 619                 PPC_LL "       %%r1, 0(%%r1)           ;"
 620                 : // Outputs
 621                   "+r" (r3)
 622                 : // Inputs
 623                   [sp] "b" (sp), [offset] "i" (THREAD_SIZE - STACK_FRAME_OVERHEAD),
 624                   [callee] "i" (__do_irq)
 625                 : // Clobbers
 626                   "lr", "xer", "ctr", "memory", "cr0", "cr1", "cr5", "cr6",
 627                   "cr7", "r0", "r4", "r5", "r6", "r7", "r8", "r9", "r10",
 628                   "r11", "r12"
 629        );
 630}
 631
 632void __do_irq(struct pt_regs *regs)
 633{
 634        unsigned int irq;
 635
 636        trace_irq_entry(regs);
 637
 638        /*
 639         * Query the platform PIC for the interrupt & ack it.
 640         *
 641         * This will typically lower the interrupt line to the CPU
 642         */
 643        irq = ppc_md.get_irq();
 644
 645        /* We can hard enable interrupts now to allow perf interrupts */
 646        may_hard_irq_enable();
 647
 648        /* And finally process it */
 649        if (unlikely(!irq))
 650                __this_cpu_inc(irq_stat.spurious_irqs);
 651        else
 652                generic_handle_irq(irq);
 653
 654        trace_irq_exit(regs);
 655}
 656
 657DEFINE_INTERRUPT_HANDLER_ASYNC(do_IRQ)
 658{
 659        struct pt_regs *old_regs = set_irq_regs(regs);
 660        void *cursp, *irqsp, *sirqsp;
 661
 662        /* Switch to the irq stack to handle this */
 663        cursp = (void *)(current_stack_pointer & ~(THREAD_SIZE - 1));
 664        irqsp = hardirq_ctx[raw_smp_processor_id()];
 665        sirqsp = softirq_ctx[raw_smp_processor_id()];
 666
 667        check_stack_overflow();
 668
 669        /* Already there ? */
 670        if (unlikely(cursp == irqsp || cursp == sirqsp)) {
 671                __do_irq(regs);
 672                set_irq_regs(old_regs);
 673                return;
 674        }
 675        /* Switch stack and call */
 676        call_do_irq(regs, irqsp);
 677
 678        set_irq_regs(old_regs);
 679}
 680
 681static void *__init alloc_vm_stack(void)
 682{
 683        return __vmalloc_node(THREAD_SIZE, THREAD_ALIGN, THREADINFO_GFP,
 684                              NUMA_NO_NODE, (void *)_RET_IP_);
 685}
 686
 687static void __init vmap_irqstack_init(void)
 688{
 689        int i;
 690
 691        for_each_possible_cpu(i) {
 692                softirq_ctx[i] = alloc_vm_stack();
 693                hardirq_ctx[i] = alloc_vm_stack();
 694        }
 695}
 696
 697
 698void __init init_IRQ(void)
 699{
 700        if (IS_ENABLED(CONFIG_VMAP_STACK))
 701                vmap_irqstack_init();
 702
 703        if (ppc_md.init_IRQ)
 704                ppc_md.init_IRQ();
 705}
 706
 707#if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
 708void   *critirq_ctx[NR_CPUS] __read_mostly;
 709void    *dbgirq_ctx[NR_CPUS] __read_mostly;
 710void *mcheckirq_ctx[NR_CPUS] __read_mostly;
 711#endif
 712
 713void *softirq_ctx[NR_CPUS] __read_mostly;
 714void *hardirq_ctx[NR_CPUS] __read_mostly;
 715
 716void do_softirq_own_stack(void)
 717{
 718        call_do_softirq(softirq_ctx[smp_processor_id()]);
 719}
 720
 721irq_hw_number_t virq_to_hw(unsigned int virq)
 722{
 723        struct irq_data *irq_data = irq_get_irq_data(virq);
 724        return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
 725}
 726EXPORT_SYMBOL_GPL(virq_to_hw);
 727
 728#ifdef CONFIG_SMP
 729int irq_choose_cpu(const struct cpumask *mask)
 730{
 731        int cpuid;
 732
 733        if (cpumask_equal(mask, cpu_online_mask)) {
 734                static int irq_rover;
 735                static DEFINE_RAW_SPINLOCK(irq_rover_lock);
 736                unsigned long flags;
 737
 738                /* Round-robin distribution... */
 739do_round_robin:
 740                raw_spin_lock_irqsave(&irq_rover_lock, flags);
 741
 742                irq_rover = cpumask_next(irq_rover, cpu_online_mask);
 743                if (irq_rover >= nr_cpu_ids)
 744                        irq_rover = cpumask_first(cpu_online_mask);
 745
 746                cpuid = irq_rover;
 747
 748                raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
 749        } else {
 750                cpuid = cpumask_first_and(mask, cpu_online_mask);
 751                if (cpuid >= nr_cpu_ids)
 752                        goto do_round_robin;
 753        }
 754
 755        return get_hard_smp_processor_id(cpuid);
 756}
 757#else
 758int irq_choose_cpu(const struct cpumask *mask)
 759{
 760        return hard_smp_processor_id();
 761}
 762#endif
 763
 764#ifdef CONFIG_PPC64
 765static int __init setup_noirqdistrib(char *str)
 766{
 767        distribute_irqs = 0;
 768        return 1;
 769}
 770
 771__setup("noirqdistrib", setup_noirqdistrib);
 772#endif /* CONFIG_PPC64 */
 773