arch/sparc/kernel/irq_64.c

   1 /* irq.c: UltraSparc IRQ handling/init/registry.
   2  *
   3  * Copyright (C) 1997, 2007, 2008 David S. Miller (davem@davemloft.net)
   4  * Copyright (C) 1998  Eddie C. Dost    (ecd@skynet.be)
   5  * Copyright (C) 1998  Jakub Jelinek    (jj@ultra.linux.cz)
   6  */
   7
   8 #include <linux/sched.h>
   9 #include <linux/linkage.h>
  10 #include <linux/ptrace.h>
  11 #include <linux/errno.h>
  12 #include <linux/kernel_stat.h>
  13 #include <linux/signal.h>
  14 #include <linux/mm.h>
  15 #include <linux/interrupt.h>
  16 #include <linux/slab.h>
  17 #include <linux/random.h>
  18 #include <linux/init.h>
  19 #include <linux/delay.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/seq_file.h>
  22 #include <linux/ftrace.h>
  23 #include <linux/irq.h>
  24 #include <linux/kmemleak.h>
  25
  26 #include <asm/ptrace.h>
  27 #include <asm/processor.h>
  28 #include <linux/atomic.h>
  29 #include <asm/irq.h>
  30 #include <asm/io.h>
  31 #include <asm/iommu.h>
  32 #include <asm/upa.h>
  33 #include <asm/oplib.h>
  34 #include <asm/prom.h>
  35 #include <asm/timer.h>
  36 #include <asm/smp.h>
  37 #include <asm/starfire.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/cache.h>
  40 #include <asm/cpudata.h>
  41 #include <asm/auxio.h>
  42 #include <asm/head.h>
  43 #include <asm/hypervisor.h>
  44 #include <asm/cacheflush.h>
  45
  46 #include "entry.h"
  47 #include "cpumap.h"
  48 #include "kstack.h"
  49
  50 #define NUM_IVECS       (IMAP_INR + 1)
  51
  52 struct ino_bucket *ivector_table;
  53 unsigned long ivector_table_pa;
  54
  55 /* On several sun4u processors, it is illegal to mix bypass and
  56  * non-bypass accesses.  Therefore we access all INO buckets
  57  * using bypass accesses only.
  58  */
  59 static unsigned long bucket_get_chain_pa(unsigned long bucket_pa)
  60 {
  61         unsigned long ret;
  62
  63         __asm__ __volatile__("ldxa      [%1] %2, %0"
  64                              : "=&r" (ret)
  65                              : "r" (bucket_pa +
  66                                     offsetof(struct ino_bucket,
  67                                              __irq_chain_pa)),
  68                                "i" (ASI_PHYS_USE_EC));
  69
  70         return ret;
  71 }
  72
  73 static void bucket_clear_chain_pa(unsigned long bucket_pa)
  74 {
  75         __asm__ __volatile__("stxa      %%g0, [%0] %1"
  76                              : /* no outputs */
  77                              : "r" (bucket_pa +
  78                                     offsetof(struct ino_bucket,
  79                                              __irq_chain_pa)),
  80                                "i" (ASI_PHYS_USE_EC));
  81 }
  82
  83 static unsigned int bucket_get_irq(unsigned long bucket_pa)
  84 {
  85         unsigned int ret;
  86
  87         __asm__ __volatile__("lduwa     [%1] %2, %0"
  88                              : "=&r" (ret)
  89                              : "r" (bucket_pa +
  90                                     offsetof(struct ino_bucket,
  91                                              __irq)),
  92                                "i" (ASI_PHYS_USE_EC));
  93
  94         return ret;
  95 }
  96
  97 static void bucket_set_irq(unsigned long bucket_pa, unsigned int irq)
  98 {
  99         __asm__ __volatile__("stwa      %0, [%1] %2"
 100                              : /* no outputs */
 101                              : "r" (irq),
 102                                "r" (bucket_pa +
 103                                     offsetof(struct ino_bucket,
 104                                              __irq)),
 105                                "i" (ASI_PHYS_USE_EC));
 106 }
 107
 108 #define irq_work_pa(__cpu)      &(trap_block[(__cpu)].irq_worklist_pa)
 109
 110 static struct {
 111         unsigned int dev_handle;
 112         unsigned int dev_ino;
 113         unsigned int in_use;
 114 } irq_table[NR_IRQS];
 115 static DEFINE_SPINLOCK(irq_alloc_lock);
 116
 117 unsigned char irq_alloc(unsigned int dev_handle, unsigned int dev_ino)
 118 {
 119         unsigned long flags;
 120         unsigned char ent;
 121
 122         BUILD_BUG_ON(NR_IRQS >= 256);
 123
 124         spin_lock_irqsave(&irq_alloc_lock, flags);
 125
 126         for (ent = 1; ent < NR_IRQS; ent++) {
 127                 if (!irq_table[ent].in_use)
 128                         break;
 129         }
 130         if (ent >= NR_IRQS) {
 131                 printk(KERN_ERR "IRQ: Out of virtual IRQs.\n");
 132                 ent = 0;
 133         } else {
 134                 irq_table[ent].dev_handle = dev_handle;
 135                 irq_table[ent].dev_ino = dev_ino;
 136                 irq_table[ent].in_use = 1;
 137         }
 138
 139         spin_unlock_irqrestore(&irq_alloc_lock, flags);
 140
 141         return ent;
 142 }
 143
 144 #ifdef CONFIG_PCI_MSI
 145 void irq_free(unsigned int irq)
 146 {
 147         unsigned long flags;
 148
 149         if (irq >= NR_IRQS)
 150                 return;
 151
 152         spin_lock_irqsave(&irq_alloc_lock, flags);
 153
 154         irq_table[irq].in_use = 0;
 155
 156         spin_unlock_irqrestore(&irq_alloc_lock, flags);
 157 }
 158 #endif
 159
 160 /*
 161  * /proc/interrupts printing:
 162  */
 163 int arch_show_interrupts(struct seq_file *p, int prec)
 164 {
 165         int j;
 166
 167         seq_printf(p, "NMI: ");
 168         for_each_online_cpu(j)
 169                 seq_printf(p, "%10u ", cpu_data(j).__nmi_count);
 170         seq_printf(p, "     Non-maskable interrupts\n");
 171         return 0;
 172 }
 173
 174 static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
 175 {
 176         unsigned int tid;
 177
 178         if (this_is_starfire) {
 179                 tid = starfire_translate(imap, cpuid);
 180                 tid <<= IMAP_TID_SHIFT;
 181                 tid &= IMAP_TID_UPA;
 182         } else {
 183                 if (tlb_type == cheetah || tlb_type == cheetah_plus) {
 184                         unsigned long ver;
 185
 186                         __asm__ ("rdpr %%ver, %0" : "=r" (ver));
 187                         if ((ver >> 32UL) == __JALAPENO_ID ||
 188                             (ver >> 32UL) == __SERRANO_ID) {
 189                                 tid = cpuid << IMAP_TID_SHIFT;
 190                                 tid &= IMAP_TID_JBUS;
 191                         } else {
 192                                 unsigned int a = cpuid & 0x1f;
 193                                 unsigned int n = (cpuid >> 5) & 0x1f;
 194
 195                                 tid = ((a << IMAP_AID_SHIFT) |
 196                                        (n << IMAP_NID_SHIFT));
 197                                 tid &= (IMAP_AID_SAFARI |
 198                                         IMAP_NID_SAFARI);
 199                         }
 200                 } else {
 201                         tid = cpuid << IMAP_TID_SHIFT;
 202                         tid &= IMAP_TID_UPA;
 203                 }
 204         }
 205
 206         return tid;
 207 }
 208
 209 struct irq_handler_data {
 210         unsigned long   iclr;
 211         unsigned long   imap;
 212
 213         void            (*pre_handler)(unsigned int, void *, void *);
 214         void            *arg1;
 215         void            *arg2;
 216 };
 217
 218 #ifdef CONFIG_SMP
 219 static int irq_choose_cpu(unsigned int irq, const struct cpumask *affinity)
 220 {
 221         cpumask_t mask;
 222         int cpuid;
 223
 224         cpumask_copy(&mask, affinity);
 225         if (cpumask_equal(&mask, cpu_online_mask)) {
 226                 cpuid = map_to_cpu(irq);
 227         } else {
 228                 cpumask_t tmp;
 229
 230                 cpumask_and(&tmp, cpu_online_mask, &mask);
 231                 cpuid = cpumask_empty(&tmp) ? map_to_cpu(irq) : cpumask_first(&tmp);
 232         }
 233
 234         return cpuid;
 235 }
 236 #else
 237 #define irq_choose_cpu(irq, affinity)   \
 238         real_hard_smp_processor_id()
 239 #endif
 240
 241 static void sun4u_irq_enable(struct irq_data *data)
 242 {
 243         struct irq_handler_data *handler_data = data->handler_data;
 244
 245         if (likely(handler_data)) {
 246                 unsigned long cpuid, imap, val;
 247                 unsigned int tid;
 248
 249                 cpuid = irq_choose_cpu(data->irq, data->affinity);
 250                 imap = handler_data->imap;
 251
 252                 tid = sun4u_compute_tid(imap, cpuid);
 253
 254                 val = upa_readq(imap);
 255                 val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
 256                          IMAP_AID_SAFARI | IMAP_NID_SAFARI);
 257                 val |= tid | IMAP_VALID;
 258                 upa_writeq(val, imap);
 259                 upa_writeq(ICLR_IDLE, handler_data->iclr);
 260         }
 261 }
 262
 263 static int sun4u_set_affinity(struct irq_data *data,
 264                                const struct cpumask *mask, bool force)
 265 {
 266         struct irq_handler_data *handler_data = data->handler_data;
 267
 268         if (likely(handler_data)) {
 269                 unsigned long cpuid, imap, val;
 270                 unsigned int tid;
 271
 272                 cpuid = irq_choose_cpu(data->irq, mask);
 273                 imap = handler_data->imap;
 274
 275                 tid = sun4u_compute_tid(imap, cpuid);
 276
 277                 val = upa_readq(imap);
 278                 val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
 279                          IMAP_AID_SAFARI | IMAP_NID_SAFARI);
 280                 val |= tid | IMAP_VALID;
 281                 upa_writeq(val, imap);
 282                 upa_writeq(ICLR_IDLE, handler_data->iclr);
 283         }
 284
 285         return 0;
 286 }
 287
 288 /* Don't do anything.  The desc->status check for IRQ_DISABLED in
 289  * handler_irq() will skip the handler call and that will leave the
 290  * interrupt in the sent state.  The next ->enable() call will hit the
 291  * ICLR register to reset the state machine.
 292  *
 293  * This scheme is necessary, instead of clearing the Valid bit in the
 294  * IMAP register, to handle the case of IMAP registers being shared by
 295  * multiple INOs (and thus ICLR registers).  Since we use a different
 296  * virtual IRQ for each shared IMAP instance, the generic code thinks
 297  * there is only one user so it prematurely calls ->disable() on
 298  * free_irq().
 299  *
 300  * We have to provide an explicit ->disable() method instead of using
 301  * NULL to get the default.  The reason is that if the generic code
 302  * sees that, it also hooks up a default ->shutdown method which
 303  * invokes ->mask() which we do not want.  See irq_chip_set_defaults().
 304  */
 305 static void sun4u_irq_disable(struct irq_data *data)
 306 {
 307 }
 308
 309 static void sun4u_irq_eoi(struct irq_data *data)
 310 {
 311         struct irq_handler_data *handler_data = data->handler_data;
 312
 313         if (likely(handler_data))
 314                 upa_writeq(ICLR_IDLE, handler_data->iclr);
 315 }
 316
 317 static void sun4v_irq_enable(struct irq_data *data)
 318 {
 319         unsigned int ino = irq_table[data->irq].dev_ino;
 320         unsigned long cpuid = irq_choose_cpu(data->irq, data->affinity);
 321         int err;
 322
 323         err = sun4v_intr_settarget(ino, cpuid);
 324         if (err != HV_EOK)
 325                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
 326                        "err(%d)\n", ino, cpuid, err);
 327         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
 328         if (err != HV_EOK)
 329                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
 330                        "err(%d)\n", ino, err);
 331         err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
 332         if (err != HV_EOK)
 333                 printk(KERN_ERR "sun4v_intr_setenabled(%x): err(%d)\n",
 334                        ino, err);
 335 }
 336
 337 static int sun4v_set_affinity(struct irq_data *data,
 338                                const struct cpumask *mask, bool force)
 339 {
 340         unsigned int ino = irq_table[data->irq].dev_ino;
 341         unsigned long cpuid = irq_choose_cpu(data->irq, mask);
 342         int err;
 343
 344         err = sun4v_intr_settarget(ino, cpuid);
 345         if (err != HV_EOK)
 346                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
 347                        "err(%d)\n", ino, cpuid, err);
 348
 349         return 0;
 350 }
 351
 352 static void sun4v_irq_disable(struct irq_data *data)
 353 {
 354         unsigned int ino = irq_table[data->irq].dev_ino;
 355         int err;
 356
 357         err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
 358         if (err != HV_EOK)
 359                 printk(KERN_ERR "sun4v_intr_setenabled(%x): "
 360                        "err(%d)\n", ino, err);
 361 }
 362
 363 static void sun4v_irq_eoi(struct irq_data *data)
 364 {
 365         unsigned int ino = irq_table[data->irq].dev_ino;
 366         int err;
 367
 368         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
 369         if (err != HV_EOK)
 370                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
 371                        "err(%d)\n", ino, err);
 372 }
 373
 374 static void sun4v_virq_enable(struct irq_data *data)
 375 {
 376         unsigned long cpuid, dev_handle, dev_ino;
 377         int err;
 378
 379         cpuid = irq_choose_cpu(data->irq, data->affinity);
 380
 381         dev_handle = irq_table[data->irq].dev_handle;
 382         dev_ino = irq_table[data->irq].dev_ino;
 383
 384         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
 385         if (err != HV_EOK)
 386                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
 387                        "err(%d)\n",
 388                        dev_handle, dev_ino, cpuid, err);
 389         err = sun4v_vintr_set_state(dev_handle, dev_ino,
 390                                     HV_INTR_STATE_IDLE);
 391         if (err != HV_EOK)
 392                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
 393                        "HV_INTR_STATE_IDLE): err(%d)\n",
 394                        dev_handle, dev_ino, err);
 395         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
 396                                     HV_INTR_ENABLED);
 397         if (err != HV_EOK)
 398                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
 399                        "HV_INTR_ENABLED): err(%d)\n",
 400                        dev_handle, dev_ino, err);
 401 }
 402
 403 static int sun4v_virt_set_affinity(struct irq_data *data,
 404                                     const struct cpumask *mask, bool force)
 405 {
 406         unsigned long cpuid, dev_handle, dev_ino;
 407         int err;
 408
 409         cpuid = irq_choose_cpu(data->irq, mask);
 410
 411         dev_handle = irq_table[data->irq].dev_handle;
 412         dev_ino = irq_table[data->irq].dev_ino;
 413
 414         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
 415         if (err != HV_EOK)
 416                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
 417                        "err(%d)\n",
 418                        dev_handle, dev_ino, cpuid, err);
 419
 420         return 0;
 421 }
 422
 423 static void sun4v_virq_disable(struct irq_data *data)
 424 {
 425         unsigned long dev_handle, dev_ino;
 426         int err;
 427
 428         dev_handle = irq_table[data->irq].dev_handle;
 429         dev_ino = irq_table[data->irq].dev_ino;
 430
 431         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
 432                                     HV_INTR_DISABLED);
 433         if (err != HV_EOK)
 434                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
 435                        "HV_INTR_DISABLED): err(%d)\n",
 436                        dev_handle, dev_ino, err);
 437 }
 438
 439 static void sun4v_virq_eoi(struct irq_data *data)
 440 {
 441         unsigned long dev_handle, dev_ino;
 442         int err;
 443
 444         dev_handle = irq_table[data->irq].dev_handle;
 445         dev_ino = irq_table[data->irq].dev_ino;
 446
 447         err = sun4v_vintr_set_state(dev_handle, dev_ino,
 448                                     HV_INTR_STATE_IDLE);
 449         if (err != HV_EOK)
 450                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
 451                        "HV_INTR_STATE_IDLE): err(%d)\n",
 452                        dev_handle, dev_ino, err);
 453 }
 454
 455 static struct irq_chip sun4u_irq = {
 456         .name                   = "sun4u",
 457         .irq_enable             = sun4u_irq_enable,
 458         .irq_disable            = sun4u_irq_disable,
 459         .irq_eoi                = sun4u_irq_eoi,
 460         .irq_set_affinity       = sun4u_set_affinity,
 461         .flags                  = IRQCHIP_EOI_IF_HANDLED,
 462 };
 463
 464 static struct irq_chip sun4v_irq = {
 465         .name                   = "sun4v",
 466         .irq_enable             = sun4v_irq_enable,
 467         .irq_disable            = sun4v_irq_disable,
 468         .irq_eoi                = sun4v_irq_eoi,
 469         .irq_set_affinity       = sun4v_set_affinity,
 470         .flags                  = IRQCHIP_EOI_IF_HANDLED,
 471 };
 472
 473 static struct irq_chip sun4v_virq = {
 474         .name                   = "vsun4v",
 475         .irq_enable             = sun4v_virq_enable,
 476         .irq_disable            = sun4v_virq_disable,
 477         .irq_eoi                = sun4v_virq_eoi,
 478         .irq_set_affinity       = sun4v_virt_set_affinity,
 479         .flags                  = IRQCHIP_EOI_IF_HANDLED,
 480 };
 481
 482 static void pre_flow_handler(struct irq_data *d)
 483 {
 484         struct irq_handler_data *handler_data = irq_data_get_irq_handler_data(d);
 485         unsigned int ino = irq_table[d->irq].dev_ino;
 486
 487         handler_data->pre_handler(ino, handler_data->arg1, handler_data->arg2);
 488 }
 489
 490 void irq_install_pre_handler(int irq,
 491                              void (*func)(unsigned int, void *, void *),
 492                              void *arg1, void *arg2)
 493 {
 494         struct irq_handler_data *handler_data = irq_get_handler_data(irq);
 495
 496         handler_data->pre_handler = func;
 497         handler_data->arg1 = arg1;
 498         handler_data->arg2 = arg2;
 499
 500         __irq_set_preflow_handler(irq, pre_flow_handler);
 501 }
 502
 503 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
 504 {
 505         struct ino_bucket *bucket;
 506         struct irq_handler_data *handler_data;
 507         unsigned int irq;
 508         int ino;
 509
 510         BUG_ON(tlb_type == hypervisor);
 511
 512         ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
 513         bucket = &ivector_table[ino];
 514         irq = bucket_get_irq(__pa(bucket));
 515         if (!irq) {
 516                 irq = irq_alloc(0, ino);
 517                 bucket_set_irq(__pa(bucket), irq);
 518                 irq_set_chip_and_handler_name(irq, &sun4u_irq,
 519                                               handle_fasteoi_irq, "IVEC");
 520         }
 521
 522         handler_data = irq_get_handler_data(irq);
 523         if (unlikely(handler_data))
 524                 goto out;
 525
 526         handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
 527         if (unlikely(!handler_data)) {
 528                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
 529                 prom_halt();
 530         }
 531         irq_set_handler_data(irq, handler_data);
 532
 533         handler_data->imap  = imap;
 534         handler_data->iclr  = iclr;
 535
 536 out:
 537         return irq;
 538 }
 539
 540 static unsigned int sun4v_build_common(unsigned long sysino,
 541                                        struct irq_chip *chip)
 542 {
 543         struct ino_bucket *bucket;
 544         struct irq_handler_data *handler_data;
 545         unsigned int irq;
 546
 547         BUG_ON(tlb_type != hypervisor);
 548
 549         bucket = &ivector_table[sysino];
 550         irq = bucket_get_irq(__pa(bucket));
 551         if (!irq) {
 552                 irq = irq_alloc(0, sysino);
 553                 bucket_set_irq(__pa(bucket), irq);
 554                 irq_set_chip_and_handler_name(irq, chip, handle_fasteoi_irq,
 555                                               "IVEC");
 556         }
 557
 558         handler_data = irq_get_handler_data(irq);
 559         if (unlikely(handler_data))
 560                 goto out;
 561
 562         handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
 563         if (unlikely(!handler_data)) {
 564                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
 565                 prom_halt();
 566         }
 567         irq_set_handler_data(irq, handler_data);
 568
 569         /* Catch accidental accesses to these things.  IMAP/ICLR handling
 570          * is done by hypervisor calls on sun4v platforms, not by direct
 571          * register accesses.
 572          */
 573         handler_data->imap = ~0UL;
 574         handler_data->iclr = ~0UL;
 575
 576 out:
 577         return irq;
 578 }
 579
 580 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
 581 {
 582         unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
 583
 584         return sun4v_build_common(sysino, &sun4v_irq);
 585 }
 586
 587 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
 588 {
 589         struct irq_handler_data *handler_data;
 590         unsigned long hv_err, cookie;
 591         struct ino_bucket *bucket;
 592         unsigned int irq;
 593
 594         bucket = kzalloc(sizeof(struct ino_bucket), GFP_ATOMIC);
 595         if (unlikely(!bucket))
 596                 return 0;
 597
 598         /* The only reference we store to the IRQ bucket is
 599          * by physical address which kmemleak can't see, tell
 600          * it that this object explicitly is not a leak and
 601          * should be scanned.
 602          */
 603         kmemleak_not_leak(bucket);
 604
 605         __flush_dcache_range((unsigned long) bucket,
 606                              ((unsigned long) bucket +
 607                               sizeof(struct ino_bucket)));
 608
 609         irq = irq_alloc(devhandle, devino);
 610         bucket_set_irq(__pa(bucket), irq);
 611
 612         irq_set_chip_and_handler_name(irq, &sun4v_virq, handle_fasteoi_irq,
 613                                       "IVEC");
 614
 615         handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
 616         if (unlikely(!handler_data))
 617                 return 0;
 618
 619         /* In order to make the LDC channel startup sequence easier,
 620          * especially wrt. locking, we do not let request_irq() enable
 621          * the interrupt.
 622          */
 623         irq_set_status_flags(irq, IRQ_NOAUTOEN);
 624         irq_set_handler_data(irq, handler_data);
 625
 626         /* Catch accidental accesses to these things.  IMAP/ICLR handling
 627          * is done by hypervisor calls on sun4v platforms, not by direct
 628          * register accesses.
 629          */
 630         handler_data->imap = ~0UL;
 631         handler_data->iclr = ~0UL;
 632
 633         cookie = ~__pa(bucket);
 634         hv_err = sun4v_vintr_set_cookie(devhandle, devino, cookie);
 635         if (hv_err) {
 636                 prom_printf("IRQ: Fatal, cannot set cookie for [%x:%x] "
 637                             "err=%lu\n", devhandle, devino, hv_err);
 638                 prom_halt();
 639         }
 640
 641         return irq;
 642 }
 643
 644 void ack_bad_irq(unsigned int irq)
 645 {
 646         unsigned int ino = irq_table[irq].dev_ino;
 647
 648         if (!ino)
 649                 ino = 0xdeadbeef;
 650
 651         printk(KERN_CRIT "Unexpected IRQ from ino[%x] irq[%u]\n",
 652                ino, irq);
 653 }
 654
 655 void *hardirq_stack[NR_CPUS];
 656 void *softirq_stack[NR_CPUS];
 657
 658 void __irq_entry handler_irq(int pil, struct pt_regs *regs)
 659 {
 660         unsigned long pstate, bucket_pa;
 661         struct pt_regs *old_regs;
 662         void *orig_sp;
 663
 664         clear_softint(1 << pil);
 665
 666         old_regs = set_irq_regs(regs);
 667         irq_enter();
 668
 669         /* Grab an atomic snapshot of the pending IVECs.  */
 670         __asm__ __volatile__("rdpr      %%pstate, %0\n\t"
 671                              "wrpr      %0, %3, %%pstate\n\t"
 672                              "ldx       [%2], %1\n\t"
 673                              "stx       %%g0, [%2]\n\t"
 674                              "wrpr      %0, 0x0, %%pstate\n\t"
 675                              : "=&r" (pstate), "=&r" (bucket_pa)
 676                              : "r" (irq_work_pa(smp_processor_id())),
 677                                "i" (PSTATE_IE)
 678                              : "memory");
 679
 680         orig_sp = set_hardirq_stack();
 681
 682         while (bucket_pa) {
 683                 unsigned long next_pa;
 684                 unsigned int irq;
 685
 686                 next_pa = bucket_get_chain_pa(bucket_pa);
 687                 irq = bucket_get_irq(bucket_pa);
 688                 bucket_clear_chain_pa(bucket_pa);
 689
 690                 generic_handle_irq(irq);
 691
 692                 bucket_pa = next_pa;
 693         }
 694
 695         restore_hardirq_stack(orig_sp);
 696
 697         irq_exit();
 698         set_irq_regs(old_regs);
 699 }
 700
 701 void do_softirq(void)
 702 {
 703         unsigned long flags;
 704
 705         if (in_interrupt())
 706                 return;
 707
 708         local_irq_save(flags);
 709
 710         if (local_softirq_pending()) {
 711                 void *orig_sp, *sp = softirq_stack[smp_processor_id()];
 712
 713                 sp += THREAD_SIZE - 192 - STACK_BIAS;
 714
 715                 __asm__ __volatile__("mov %%sp, %0\n\t"
 716                                      "mov %1, %%sp"
 717                                      : "=&r" (orig_sp)
 718                                      : "r" (sp));
 719                 __do_softirq();
 720                 __asm__ __volatile__("mov %0, %%sp"
 721                                      : : "r" (orig_sp));
 722         }
 723
 724         local_irq_restore(flags);
 725 }
 726
 727 #ifdef CONFIG_HOTPLUG_CPU
 728 void fixup_irqs(void)
 729 {
 730         unsigned int irq;
 731
 732         for (irq = 0; irq < NR_IRQS; irq++) {
 733                 struct irq_desc *desc = irq_to_desc(irq);
 734                 struct irq_data *data = irq_desc_get_irq_data(desc);
 735                 unsigned long flags;
 736
 737                 raw_spin_lock_irqsave(&desc->lock, flags);
 738                 if (desc->action && !irqd_is_per_cpu(data)) {
 739                         if (data->chip->irq_set_affinity)
 740                                 data->chip->irq_set_affinity(data,
 741                                                              data->affinity,
 742                                                              false);
 743                 }
 744                 raw_spin_unlock_irqrestore(&desc->lock, flags);
 745         }
 746
 747         tick_ops->disable_irq();
 748 }
 749 #endif
 750
 751 struct sun5_timer {
 752         u64     count0;
 753         u64     limit0;
 754         u64     count1;
 755         u64     limit1;
 756 };
 757
 758 static struct sun5_timer *prom_timers;
 759 static u64 prom_limit0, prom_limit1;
 760
 761 static void map_prom_timers(void)
 762 {
 763         struct device_node *dp;
 764         const unsigned int *addr;
 765
 766         /* PROM timer node hangs out in the top level of device siblings... */
 767         dp = of_find_node_by_path("/");
 768         dp = dp->child;
 769         while (dp) {
 770                 if (!strcmp(dp->name, "counter-timer"))
 771                         break;
 772                 dp = dp->sibling;
 773         }
 774
 775         /* Assume if node is not present, PROM uses different tick mechanism
 776          * which we should not care about.
 777          */
 778         if (!dp) {
 779                 prom_timers = (struct sun5_timer *) 0;
 780                 return;
 781         }
 782
 783         /* If PROM is really using this, it must be mapped by him. */
 784         addr = of_get_property(dp, "address", NULL);
 785         if (!addr) {
 786                 prom_printf("PROM does not have timer mapped, trying to continue.\n");
 787                 prom_timers = (struct sun5_timer *) 0;
 788                 return;
 789         }
 790         prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
 791 }
 792
 793 static void kill_prom_timer(void)
 794 {
 795         if (!prom_timers)
 796                 return;
 797
 798         /* Save them away for later. */
 799         prom_limit0 = prom_timers->limit0;
 800         prom_limit1 = prom_timers->limit1;
 801
 802         /* Just as in sun4c/sun4m PROM uses timer which ticks at IRQ 14.
 803          * We turn both off here just to be paranoid.
 804          */
 805         prom_timers->limit0 = 0;
 806         prom_timers->limit1 = 0;
 807
 808         /* Wheee, eat the interrupt packet too... */
 809         __asm__ __volatile__(
 810 "       mov     0x40, %%g2\n"
 811 "       ldxa    [%%g0] %0, %%g1\n"
 812 "       ldxa    [%%g2] %1, %%g1\n"
 813 "       stxa    %%g0, [%%g0] %0\n"
 814 "       membar  #Sync\n"
 815         : /* no outputs */
 816         : "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
 817         : "g1", "g2");
 818 }
 819
 820 void notrace init_irqwork_curcpu(void)
 821 {
 822         int cpu = hard_smp_processor_id();
 823
 824         trap_block[cpu].irq_worklist_pa = 0UL;
 825 }
 826
 827 /* Please be very careful with register_one_mondo() and
 828  * sun4v_register_mondo_queues().
 829  *
 830  * On SMP this gets invoked from the CPU trampoline before
 831  * the cpu has fully taken over the trap table from OBP,
 832  * and it's kernel stack + %g6 thread register state is
 833  * not fully cooked yet.
 834  *
 835  * Therefore you cannot make any OBP calls, not even prom_printf,
 836  * from these two routines.
 837  */
 838 static void __cpuinit notrace register_one_mondo(unsigned long paddr, unsigned long type, unsigned long qmask)
 839 {
 840         unsigned long num_entries = (qmask + 1) / 64;
 841         unsigned long status;
 842
 843         status = sun4v_cpu_qconf(type, paddr, num_entries);
 844         if (status != HV_EOK) {
 845                 prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
 846                             "err %lu\n", type, paddr, num_entries, status);
 847                 prom_halt();
 848         }
 849 }
 850
 851 void __cpuinit notrace sun4v_register_mondo_queues(int this_cpu)
 852 {
 853         struct trap_per_cpu *tb = &trap_block[this_cpu];
 854
 855         register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
 856                            tb->cpu_mondo_qmask);
 857         register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
 858                            tb->dev_mondo_qmask);
 859         register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
 860                            tb->resum_qmask);
 861         register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
 862                            tb->nonresum_qmask);
 863 }
 864
 865 /* Each queue region must be a power of 2 multiple of 64 bytes in
 866  * size.  The base real address must be aligned to the size of the
 867  * region.  Thus, an 8KB queue must be 8KB aligned, for example.
 868  */
 869 static void __init alloc_one_queue(unsigned long *pa_ptr, unsigned long qmask)
 870 {
 871         unsigned long size = PAGE_ALIGN(qmask + 1);
 872         unsigned long order = get_order(size);
 873         unsigned long p;
 874
 875         p = __get_free_pages(GFP_KERNEL, order);
 876         if (!p) {
 877                 prom_printf("SUN4V: Error, cannot allocate queue.\n");
 878                 prom_halt();
 879         }
 880
 881         *pa_ptr = __pa(p);
 882 }
 883
 884 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
 885 {
 886 #ifdef CONFIG_SMP
 887         unsigned long page;
 888
 889         BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
 890
 891         page = get_zeroed_page(GFP_KERNEL);
 892         if (!page) {
 893                 prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
 894                 prom_halt();
 895         }
 896
 897         tb->cpu_mondo_block_pa = __pa(page);
 898         tb->cpu_list_pa = __pa(page + 64);
 899 #endif
 900 }
 901
 902 /* Allocate mondo and error queues for all possible cpus.  */
 903 static void __init sun4v_init_mondo_queues(void)
 904 {
 905         int cpu;
 906
 907         for_each_possible_cpu(cpu) {
 908                 struct trap_per_cpu *tb = &trap_block[cpu];
 909
 910                 alloc_one_queue(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
 911                 alloc_one_queue(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
 912                 alloc_one_queue(&tb->resum_mondo_pa, tb->resum_qmask);
 913                 alloc_one_queue(&tb->resum_kernel_buf_pa, tb->resum_qmask);
 914                 alloc_one_queue(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
 915                 alloc_one_queue(&tb->nonresum_kernel_buf_pa,
 916                                 tb->nonresum_qmask);
 917         }
 918 }
 919
 920 static void __init init_send_mondo_info(void)
 921 {
 922         int cpu;
 923
 924         for_each_possible_cpu(cpu) {
 925                 struct trap_per_cpu *tb = &trap_block[cpu];
 926
 927                 init_cpu_send_mondo_info(tb);
 928         }
 929 }
 930
 931 static struct irqaction timer_irq_action = {
 932         .name = "timer",
 933 };
 934
 935 /* Only invoked on boot processor. */
 936 void __init init_IRQ(void)
 937 {
 938         unsigned long size;
 939
 940         map_prom_timers();
 941         kill_prom_timer();
 942
 943         size = sizeof(struct ino_bucket) * NUM_IVECS;
 944         ivector_table = kzalloc(size, GFP_KERNEL);
 945         if (!ivector_table) {
 946                 prom_printf("Fatal error, cannot allocate ivector_table\n");
 947                 prom_halt();
 948         }
 949         __flush_dcache_range((unsigned long) ivector_table,
 950                              ((unsigned long) ivector_table) + size);
 951
 952         ivector_table_pa = __pa(ivector_table);
 953
 954         if (tlb_type == hypervisor)
 955                 sun4v_init_mondo_queues();
 956
 957         init_send_mondo_info();
 958
 959         if (tlb_type == hypervisor) {
 960                 /* Load up the boot cpu's entries.  */
 961                 sun4v_register_mondo_queues(hard_smp_processor_id());
 962         }
 963
 964         /* We need to clear any IRQ's pending in the soft interrupt
 965          * registers, a spurious one could be left around from the
 966          * PROM timer which we just disabled.
 967          */
 968         clear_softint(get_softint());
 969
 970         /* Now that ivector table is initialized, it is safe
 971          * to receive IRQ vector traps.  We will normally take
 972          * one or two right now, in case some device PROM used
 973          * to boot us wants to speak to us.  We just ignore them.
 974          */
 975         __asm__ __volatile__("rdpr      %%pstate, %%g1\n\t"
 976                              "or        %%g1, %0, %%g1\n\t"
 977                              "wrpr      %%g1, 0x0, %%pstate"
 978                              : /* No outputs */
 979                              : "i" (PSTATE_IE)
 980                              : "g1");
 981
 982         irq_to_desc(0)->action = &timer_irq_action;
 983 }