kernel/arch/powerpc/kvm/mpic.c

   1 /*
   2  * OpenPIC emulation
   3  *
   4  * Copyright (c) 2004 Jocelyn Mayer
   5  *               2011 Alexander Graf
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include <linux/slab.h>
  27 #include <linux/mutex.h>
  28 #include <linux/kvm_host.h>
  29 #include <linux/errno.h>
  30 #include <linux/fs.h>
  31 #include <linux/anon_inodes.h>
  32 #include <asm/uaccess.h>
  33 #include <asm/mpic.h>
  34 #include <asm/kvm_para.h>
  35 #include <asm/kvm_host.h>
  36 #include <asm/kvm_ppc.h>
  37 #include <kvm/iodev.h>
  38
  39 #define MAX_CPU     32
  40 #define MAX_SRC     256
  41 #define MAX_TMR     4
  42 #define MAX_IPI     4
  43 #define MAX_MSI     8
  44 #define MAX_IRQ     (MAX_SRC + MAX_IPI + MAX_TMR)
  45 #define VID         0x03        /* MPIC version ID */
  46
  47 /* OpenPIC capability flags */
  48 #define OPENPIC_FLAG_IDR_CRIT     (1 << 0)
  49 #define OPENPIC_FLAG_ILR          (2 << 0)
  50
  51 /* OpenPIC address map */
  52 #define OPENPIC_REG_SIZE             0x40000
  53 #define OPENPIC_GLB_REG_START        0x0
  54 #define OPENPIC_GLB_REG_SIZE         0x10F0
  55 #define OPENPIC_TMR_REG_START        0x10F0
  56 #define OPENPIC_TMR_REG_SIZE         0x220
  57 #define OPENPIC_MSI_REG_START        0x1600
  58 #define OPENPIC_MSI_REG_SIZE         0x200
  59 #define OPENPIC_SUMMARY_REG_START    0x3800
  60 #define OPENPIC_SUMMARY_REG_SIZE     0x800
  61 #define OPENPIC_SRC_REG_START        0x10000
  62 #define OPENPIC_SRC_REG_SIZE         (MAX_SRC * 0x20)
  63 #define OPENPIC_CPU_REG_START        0x20000
  64 #define OPENPIC_CPU_REG_SIZE         (0x100 + ((MAX_CPU - 1) * 0x1000))
  65
  66 struct fsl_mpic_info {
  67         int max_ext;
  68 };
  69
  70 static struct fsl_mpic_info fsl_mpic_20 = {
  71         .max_ext = 12,
  72 };
  73
  74 static struct fsl_mpic_info fsl_mpic_42 = {
  75         .max_ext = 12,
  76 };
  77
  78 #define FRR_NIRQ_SHIFT    16
  79 #define FRR_NCPU_SHIFT     8
  80 #define FRR_VID_SHIFT      0
  81
  82 #define VID_REVISION_1_2   2
  83 #define VID_REVISION_1_3   3
  84
  85 #define VIR_GENERIC      0x00000000     /* Generic Vendor ID */
  86
  87 #define GCR_RESET        0x80000000
  88 #define GCR_MODE_PASS    0x00000000
  89 #define GCR_MODE_MIXED   0x20000000
  90 #define GCR_MODE_PROXY   0x60000000
  91
  92 #define TBCR_CI           0x80000000    /* count inhibit */
  93 #define TCCR_TOG          0x80000000    /* toggles when decrement to zero */
  94
  95 #define IDR_EP_SHIFT      31
  96 #define IDR_EP_MASK       (1 << IDR_EP_SHIFT)
  97 #define IDR_CI0_SHIFT     30
  98 #define IDR_CI1_SHIFT     29
  99 #define IDR_P1_SHIFT      1
 100 #define IDR_P0_SHIFT      0
 101
 102 #define ILR_INTTGT_MASK   0x000000ff
 103 #define ILR_INTTGT_INT    0x00
 104 #define ILR_INTTGT_CINT   0x01  /* critical */
 105 #define ILR_INTTGT_MCP    0x02  /* machine check */
 106 #define NUM_OUTPUTS       3
 107
 108 #define MSIIR_OFFSET       0x140
 109 #define MSIIR_SRS_SHIFT    29
 110 #define MSIIR_SRS_MASK     (0x7 << MSIIR_SRS_SHIFT)
 111 #define MSIIR_IBS_SHIFT    24
 112 #define MSIIR_IBS_MASK     (0x1f << MSIIR_IBS_SHIFT)
 113
 114 static int get_current_cpu(void)
 115 {
 116 #if defined(CONFIG_KVM) && defined(CONFIG_BOOKE)
 117         struct kvm_vcpu *vcpu = current->thread.kvm_vcpu;
 118         return vcpu ? vcpu->arch.irq_cpu_id : -1;
 119 #else
 120         /* XXX */
 121         return -1;
 122 #endif
 123 }
 124
 125 static int openpic_cpu_write_internal(void *opaque, gpa_t addr,
 126                                       u32 val, int idx);
 127 static int openpic_cpu_read_internal(void *opaque, gpa_t addr,
 128                                      u32 *ptr, int idx);
 129 static inline void write_IRQreg_idr(struct openpic *opp, int n_IRQ,
 130                                     uint32_t val);
 131
 132 enum irq_type {
 133         IRQ_TYPE_NORMAL = 0,
 134         IRQ_TYPE_FSLINT,        /* FSL internal interrupt -- level only */
 135         IRQ_TYPE_FSLSPECIAL,    /* FSL timer/IPI interrupt, edge, no polarity */
 136 };
 137
 138 struct irq_queue {
 139         /* Round up to the nearest 64 IRQs so that the queue length
 140          * won't change when moving between 32 and 64 bit hosts.
 141          */
 142         unsigned long queue[BITS_TO_LONGS((MAX_IRQ + 63) & ~63)];
 143         int next;
 144         int priority;
 145 };
 146
 147 struct irq_source {
 148         uint32_t ivpr;          /* IRQ vector/priority register */
 149         uint32_t idr;           /* IRQ destination register */
 150         uint32_t destmask;      /* bitmap of CPU destinations */
 151         int last_cpu;
 152         int output;             /* IRQ level, e.g. ILR_INTTGT_INT */
 153         int pending;            /* TRUE if IRQ is pending */
 154         enum irq_type type;
 155         bool level:1;           /* level-triggered */
 156         bool nomask:1;  /* critical interrupts ignore mask on some FSL MPICs */
 157 };
 158
 159 #define IVPR_MASK_SHIFT       31
 160 #define IVPR_MASK_MASK        (1 << IVPR_MASK_SHIFT)
 161 #define IVPR_ACTIVITY_SHIFT   30
 162 #define IVPR_ACTIVITY_MASK    (1 << IVPR_ACTIVITY_SHIFT)
 163 #define IVPR_MODE_SHIFT       29
 164 #define IVPR_MODE_MASK        (1 << IVPR_MODE_SHIFT)
 165 #define IVPR_POLARITY_SHIFT   23
 166 #define IVPR_POLARITY_MASK    (1 << IVPR_POLARITY_SHIFT)
 167 #define IVPR_SENSE_SHIFT      22
 168 #define IVPR_SENSE_MASK       (1 << IVPR_SENSE_SHIFT)
 169
 170 #define IVPR_PRIORITY_MASK     (0xF << 16)
 171 #define IVPR_PRIORITY(_ivprr_) ((int)(((_ivprr_) & IVPR_PRIORITY_MASK) >> 16))
 172 #define IVPR_VECTOR(opp, _ivprr_) ((_ivprr_) & (opp)->vector_mask)
 173
 174 /* IDR[EP/CI] are only for FSL MPIC prior to v4.0 */
 175 #define IDR_EP      0x80000000  /* external pin */
 176 #define IDR_CI      0x40000000  /* critical interrupt */
 177
 178 struct irq_dest {
 179         struct kvm_vcpu *vcpu;
 180
 181         int32_t ctpr;           /* CPU current task priority */
 182         struct irq_queue raised;
 183         struct irq_queue servicing;
 184
 185         /* Count of IRQ sources asserting on non-INT outputs */
 186         uint32_t outputs_active[NUM_OUTPUTS];
 187 };
 188
 189 #define MAX_MMIO_REGIONS 10
 190
 191 struct openpic {
 192         struct kvm *kvm;
 193         struct kvm_device *dev;
 194         struct kvm_io_device mmio;
 195         const struct mem_reg *mmio_regions[MAX_MMIO_REGIONS];
 196         int num_mmio_regions;
 197
 198         gpa_t reg_base;
 199         spinlock_t lock;
 200
 201         /* Behavior control */
 202         struct fsl_mpic_info *fsl;
 203         uint32_t model;
 204         uint32_t flags;
 205         uint32_t nb_irqs;
 206         uint32_t vid;
 207         uint32_t vir;           /* Vendor identification register */
 208         uint32_t vector_mask;
 209         uint32_t tfrr_reset;
 210         uint32_t ivpr_reset;
 211         uint32_t idr_reset;
 212         uint32_t brr1;
 213         uint32_t mpic_mode_mask;
 214
 215         /* Global registers */
 216         uint32_t frr;           /* Feature reporting register */
 217         uint32_t gcr;           /* Global configuration register  */
 218         uint32_t pir;           /* Processor initialization register */
 219         uint32_t spve;          /* Spurious vector register */
 220         uint32_t tfrr;          /* Timer frequency reporting register */
 221         /* Source registers */
 222         struct irq_source src[MAX_IRQ];
 223         /* Local registers per output pin */
 224         struct irq_dest dst[MAX_CPU];
 225         uint32_t nb_cpus;
 226         /* Timer registers */
 227         struct {
 228                 uint32_t tccr;  /* Global timer current count register */
 229                 uint32_t tbcr;  /* Global timer base count register */
 230         } timers[MAX_TMR];
 231         /* Shared MSI registers */
 232         struct {
 233                 uint32_t msir;  /* Shared Message Signaled Interrupt Register */
 234         } msi[MAX_MSI];
 235         uint32_t max_irq;
 236         uint32_t irq_ipi0;
 237         uint32_t irq_tim0;
 238         uint32_t irq_msi;
 239 };
 240
 241
 242 static void mpic_irq_raise(struct openpic *opp, struct irq_dest *dst,
 243                            int output)
 244 {
 245         struct kvm_interrupt irq = {
 246                 .irq = KVM_INTERRUPT_SET_LEVEL,
 247         };
 248
 249         if (!dst->vcpu) {
 250                 pr_debug("%s: destination cpu %d does not exist\n",
 251                          __func__, (int)(dst - &opp->dst[0]));
 252                 return;
 253         }
 254
 255         pr_debug("%s: cpu %d output %d\n", __func__, dst->vcpu->arch.irq_cpu_id,
 256                 output);
 257
 258         if (output != ILR_INTTGT_INT)   /* TODO */
 259                 return;
 260
 261         kvm_vcpu_ioctl_interrupt(dst->vcpu, &irq);
 262 }
 263
 264 static void mpic_irq_lower(struct openpic *opp, struct irq_dest *dst,
 265                            int output)
 266 {
 267         if (!dst->vcpu) {
 268                 pr_debug("%s: destination cpu %d does not exist\n",
 269                          __func__, (int)(dst - &opp->dst[0]));
 270                 return;
 271         }
 272
 273         pr_debug("%s: cpu %d output %d\n", __func__, dst->vcpu->arch.irq_cpu_id,
 274                 output);
 275
 276         if (output != ILR_INTTGT_INT)   /* TODO */
 277                 return;
 278
 279         kvmppc_core_dequeue_external(dst->vcpu);
 280 }
 281
 282 static inline void IRQ_setbit(struct irq_queue *q, int n_IRQ)
 283 {
 284         set_bit(n_IRQ, q->queue);
 285 }
 286
 287 static inline void IRQ_resetbit(struct irq_queue *q, int n_IRQ)
 288 {
 289         clear_bit(n_IRQ, q->queue);
 290 }
 291
 292 static void IRQ_check(struct openpic *opp, struct irq_queue *q)
 293 {
 294         int irq = -1;
 295         int next = -1;
 296         int priority = -1;
 297
 298         for (;;) {
 299                 irq = find_next_bit(q->queue, opp->max_irq, irq + 1);
 300                 if (irq == opp->max_irq)
 301                         break;
 302
 303                 pr_debug("IRQ_check: irq %d set ivpr_pr=%d pr=%d\n",
 304                         irq, IVPR_PRIORITY(opp->src[irq].ivpr), priority);
 305
 306                 if (IVPR_PRIORITY(opp->src[irq].ivpr) > priority) {
 307                         next = irq;
 308                         priority = IVPR_PRIORITY(opp->src[irq].ivpr);
 309                 }
 310         }
 311
 312         q->next = next;
 313         q->priority = priority;
 314 }
 315
 316 static int IRQ_get_next(struct openpic *opp, struct irq_queue *q)
 317 {
 318         /* XXX: optimize */
 319         IRQ_check(opp, q);
 320
 321         return q->next;
 322 }
 323
 324 static void IRQ_local_pipe(struct openpic *opp, int n_CPU, int n_IRQ,
 325                            bool active, bool was_active)
 326 {
 327         struct irq_dest *dst;
 328         struct irq_source *src;
 329         int priority;
 330
 331         dst = &opp->dst[n_CPU];
 332         src = &opp->src[n_IRQ];
 333
 334         pr_debug("%s: IRQ %d active %d was %d\n",
 335                 __func__, n_IRQ, active, was_active);
 336
 337         if (src->output != ILR_INTTGT_INT) {
 338                 pr_debug("%s: output %d irq %d active %d was %d count %d\n",
 339                         __func__, src->output, n_IRQ, active, was_active,
 340                         dst->outputs_active[src->output]);
 341
 342                 /* On Freescale MPIC, critical interrupts ignore priority,
 343                  * IACK, EOI, etc.  Before MPIC v4.1 they also ignore
 344                  * masking.
 345                  */
 346                 if (active) {
 347                         if (!was_active &&
 348                             dst->outputs_active[src->output]++ == 0) {
 349                                 pr_debug("%s: Raise OpenPIC output %d cpu %d irq %d\n",
 350                                         __func__, src->output, n_CPU, n_IRQ);
 351                                 mpic_irq_raise(opp, dst, src->output);
 352                         }
 353                 } else {
 354                         if (was_active &&
 355                             --dst->outputs_active[src->output] == 0) {
 356                                 pr_debug("%s: Lower OpenPIC output %d cpu %d irq %d\n",
 357                                         __func__, src->output, n_CPU, n_IRQ);
 358                                 mpic_irq_lower(opp, dst, src->output);
 359                         }
 360                 }
 361
 362                 return;
 363         }
 364
 365         priority = IVPR_PRIORITY(src->ivpr);
 366
 367         /* Even if the interrupt doesn't have enough priority,
 368          * it is still raised, in case ctpr is lowered later.
 369          */
 370         if (active)
 371                 IRQ_setbit(&dst->raised, n_IRQ);
 372         else
 373                 IRQ_resetbit(&dst->raised, n_IRQ);
 374
 375         IRQ_check(opp, &dst->raised);
 376
 377         if (active && priority <= dst->ctpr) {
 378                 pr_debug("%s: IRQ %d priority %d too low for ctpr %d on CPU %d\n",
 379                         __func__, n_IRQ, priority, dst->ctpr, n_CPU);
 380                 active = 0;
 381         }
 382
 383         if (active) {
 384                 if (IRQ_get_next(opp, &dst->servicing) >= 0 &&
 385                     priority <= dst->servicing.priority) {
 386                         pr_debug("%s: IRQ %d is hidden by servicing IRQ %d on CPU %d\n",
 387                                 __func__, n_IRQ, dst->servicing.next, n_CPU);
 388                 } else {
 389                         pr_debug("%s: Raise OpenPIC INT output cpu %d irq %d/%d\n",
 390                                 __func__, n_CPU, n_IRQ, dst->raised.next);
 391                         mpic_irq_raise(opp, dst, ILR_INTTGT_INT);
 392                 }
 393         } else {
 394                 IRQ_get_next(opp, &dst->servicing);
 395                 if (dst->raised.priority > dst->ctpr &&
 396                     dst->raised.priority > dst->servicing.priority) {
 397                         pr_debug("%s: IRQ %d inactive, IRQ %d prio %d above %d/%d, CPU %d\n",
 398                                 __func__, n_IRQ, dst->raised.next,
 399                                 dst->raised.priority, dst->ctpr,
 400                                 dst->servicing.priority, n_CPU);
 401                         /* IRQ line stays asserted */
 402                 } else {
 403                         pr_debug("%s: IRQ %d inactive, current prio %d/%d, CPU %d\n",
 404                                 __func__, n_IRQ, dst->ctpr,
 405                                 dst->servicing.priority, n_CPU);
 406                         mpic_irq_lower(opp, dst, ILR_INTTGT_INT);
 407                 }
 408         }
 409 }
 410
 411 /* update pic state because registers for n_IRQ have changed value */
 412 static void openpic_update_irq(struct openpic *opp, int n_IRQ)
 413 {
 414         struct irq_source *src;
 415         bool active, was_active;
 416         int i;
 417
 418         src = &opp->src[n_IRQ];
 419         active = src->pending;
 420
 421         if ((src->ivpr & IVPR_MASK_MASK) && !src->nomask) {
 422                 /* Interrupt source is disabled */
 423                 pr_debug("%s: IRQ %d is disabled\n", __func__, n_IRQ);
 424                 active = false;
 425         }
 426
 427         was_active = !!(src->ivpr & IVPR_ACTIVITY_MASK);
 428
 429         /*
 430          * We don't have a similar check for already-active because
 431          * ctpr may have changed and we need to withdraw the interrupt.
 432          */
 433         if (!active && !was_active) {
 434                 pr_debug("%s: IRQ %d is already inactive\n", __func__, n_IRQ);
 435                 return;
 436         }
 437
 438         if (active)
 439                 src->ivpr |= IVPR_ACTIVITY_MASK;
 440         else
 441                 src->ivpr &= ~IVPR_ACTIVITY_MASK;
 442
 443         if (src->destmask == 0) {
 444                 /* No target */
 445                 pr_debug("%s: IRQ %d has no target\n", __func__, n_IRQ);
 446                 return;
 447         }
 448
 449         if (src->destmask == (1 << src->last_cpu)) {
 450                 /* Only one CPU is allowed to receive this IRQ */
 451                 IRQ_local_pipe(opp, src->last_cpu, n_IRQ, active, was_active);
 452         } else if (!(src->ivpr & IVPR_MODE_MASK)) {
 453                 /* Directed delivery mode */
 454                 for (i = 0; i < opp->nb_cpus; i++) {
 455                         if (src->destmask & (1 << i)) {
 456                                 IRQ_local_pipe(opp, i, n_IRQ, active,
 457                                                was_active);
 458                         }
 459                 }
 460         } else {
 461                 /* Distributed delivery mode */
 462                 for (i = src->last_cpu + 1; i != src->last_cpu; i++) {
 463                         if (i == opp->nb_cpus)
 464                                 i = 0;
 465
 466                         if (src->destmask & (1 << i)) {
 467                                 IRQ_local_pipe(opp, i, n_IRQ, active,
 468                                                was_active);
 469                                 src->last_cpu = i;
 470                                 break;
 471                         }
 472                 }
 473         }
 474 }
 475
 476 static void openpic_set_irq(void *opaque, int n_IRQ, int level)
 477 {
 478         struct openpic *opp = opaque;
 479         struct irq_source *src;
 480
 481         if (n_IRQ >= MAX_IRQ) {
 482                 WARN_ONCE(1, "%s: IRQ %d out of range\n", __func__, n_IRQ);
 483                 return;
 484         }
 485
 486         src = &opp->src[n_IRQ];
 487         pr_debug("openpic: set irq %d = %d ivpr=0x%08x\n",
 488                 n_IRQ, level, src->ivpr);
 489         if (src->level) {
 490                 /* level-sensitive irq */
 491                 src->pending = level;
 492                 openpic_update_irq(opp, n_IRQ);
 493         } else {
 494                 /* edge-sensitive irq */
 495                 if (level) {
 496                         src->pending = 1;
 497                         openpic_update_irq(opp, n_IRQ);
 498                 }
 499
 500                 if (src->output != ILR_INTTGT_INT) {
 501                         /* Edge-triggered interrupts shouldn't be used
 502                          * with non-INT delivery, but just in case,
 503                          * try to make it do something sane rather than
 504                          * cause an interrupt storm.  This is close to
 505                          * what you'd probably see happen in real hardware.
 506                          */
 507                         src->pending = 0;
 508                         openpic_update_irq(opp, n_IRQ);
 509                 }
 510         }
 511 }
 512
 513 static void openpic_reset(struct openpic *opp)
 514 {
 515         int i;
 516
 517         opp->gcr = GCR_RESET;
 518         /* Initialise controller registers */
 519         opp->frr = ((opp->nb_irqs - 1) << FRR_NIRQ_SHIFT) |
 520             (opp->vid << FRR_VID_SHIFT);
 521
 522         opp->pir = 0;
 523         opp->spve = -1 & opp->vector_mask;
 524         opp->tfrr = opp->tfrr_reset;
 525         /* Initialise IRQ sources */
 526         for (i = 0; i < opp->max_irq; i++) {
 527                 opp->src[i].ivpr = opp->ivpr_reset;
 528
 529                 switch (opp->src[i].type) {
 530                 case IRQ_TYPE_NORMAL:
 531                         opp->src[i].level =
 532                             !!(opp->ivpr_reset & IVPR_SENSE_MASK);
 533                         break;
 534
 535                 case IRQ_TYPE_FSLINT:
 536                         opp->src[i].ivpr |= IVPR_POLARITY_MASK;
 537                         break;
 538
 539                 case IRQ_TYPE_FSLSPECIAL:
 540                         break;
 541                 }
 542
 543                 write_IRQreg_idr(opp, i, opp->idr_reset);
 544         }
 545         /* Initialise IRQ destinations */
 546         for (i = 0; i < MAX_CPU; i++) {
 547                 opp->dst[i].ctpr = 15;
 548                 memset(&opp->dst[i].raised, 0, sizeof(struct irq_queue));
 549                 opp->dst[i].raised.next = -1;
 550                 memset(&opp->dst[i].servicing, 0, sizeof(struct irq_queue));
 551                 opp->dst[i].servicing.next = -1;
 552         }
 553         /* Initialise timers */
 554         for (i = 0; i < MAX_TMR; i++) {
 555                 opp->timers[i].tccr = 0;
 556                 opp->timers[i].tbcr = TBCR_CI;
 557         }
 558         /* Go out of RESET state */
 559         opp->gcr = 0;
 560 }
 561
 562 static inline uint32_t read_IRQreg_idr(struct openpic *opp, int n_IRQ)
 563 {
 564         return opp->src[n_IRQ].idr;
 565 }
 566
 567 static inline uint32_t read_IRQreg_ilr(struct openpic *opp, int n_IRQ)
 568 {
 569         if (opp->flags & OPENPIC_FLAG_ILR)
 570                 return opp->src[n_IRQ].output;
 571
 572         return 0xffffffff;
 573 }
 574
 575 static inline uint32_t read_IRQreg_ivpr(struct openpic *opp, int n_IRQ)
 576 {
 577         return opp->src[n_IRQ].ivpr;
 578 }
 579
 580 static inline void write_IRQreg_idr(struct openpic *opp, int n_IRQ,
 581                                     uint32_t val)
 582 {
 583         struct irq_source *src = &opp->src[n_IRQ];
 584         uint32_t normal_mask = (1UL << opp->nb_cpus) - 1;
 585         uint32_t crit_mask = 0;
 586         uint32_t mask = normal_mask;
 587         int crit_shift = IDR_EP_SHIFT - opp->nb_cpus;
 588         int i;
 589
 590         if (opp->flags & OPENPIC_FLAG_IDR_CRIT) {
 591                 crit_mask = mask << crit_shift;
 592                 mask |= crit_mask | IDR_EP;
 593         }
 594
 595         src->idr = val & mask;
 596         pr_debug("Set IDR %d to 0x%08x\n", n_IRQ, src->idr);
 597
 598         if (opp->flags & OPENPIC_FLAG_IDR_CRIT) {
 599                 if (src->idr & crit_mask) {
 600                         if (src->idr & normal_mask) {
 601                                 pr_debug("%s: IRQ configured for multiple output types, using critical\n",
 602                                         __func__);
 603                         }
 604
 605                         src->output = ILR_INTTGT_CINT;
 606                         src->nomask = true;
 607                         src->destmask = 0;
 608
 609                         for (i = 0; i < opp->nb_cpus; i++) {
 610                                 int n_ci = IDR_CI0_SHIFT - i;
 611
 612                                 if (src->idr & (1UL << n_ci))
 613                                         src->destmask |= 1UL << i;
 614                         }
 615                 } else {
 616                         src->output = ILR_INTTGT_INT;
 617                         src->nomask = false;
 618                         src->destmask = src->idr & normal_mask;
 619                 }
 620         } else {
 621                 src->destmask = src->idr;
 622         }
 623 }
 624
 625 static inline void write_IRQreg_ilr(struct openpic *opp, int n_IRQ,
 626                                     uint32_t val)
 627 {
 628         if (opp->flags & OPENPIC_FLAG_ILR) {
 629                 struct irq_source *src = &opp->src[n_IRQ];
 630
 631                 src->output = val & ILR_INTTGT_MASK;
 632                 pr_debug("Set ILR %d to 0x%08x, output %d\n", n_IRQ, src->idr,
 633                         src->output);
 634
 635                 /* TODO: on MPIC v4.0 only, set nomask for non-INT */
 636         }
 637 }
 638
 639 static inline void write_IRQreg_ivpr(struct openpic *opp, int n_IRQ,
 640                                      uint32_t val)
 641 {
 642         uint32_t mask;
 643
 644         /* NOTE when implementing newer FSL MPIC models: starting with v4.0,
 645          * the polarity bit is read-only on internal interrupts.
 646          */
 647         mask = IVPR_MASK_MASK | IVPR_PRIORITY_MASK | IVPR_SENSE_MASK |
 648             IVPR_POLARITY_MASK | opp->vector_mask;
 649
 650         /* ACTIVITY bit is read-only */
 651         opp->src[n_IRQ].ivpr =
 652             (opp->src[n_IRQ].ivpr & IVPR_ACTIVITY_MASK) | (val & mask);
 653
 654         /* For FSL internal interrupts, The sense bit is reserved and zero,
 655          * and the interrupt is always level-triggered.  Timers and IPIs
 656          * have no sense or polarity bits, and are edge-triggered.
 657          */
 658         switch (opp->src[n_IRQ].type) {
 659         case IRQ_TYPE_NORMAL:
 660                 opp->src[n_IRQ].level =
 661                     !!(opp->src[n_IRQ].ivpr & IVPR_SENSE_MASK);
 662                 break;
 663
 664         case IRQ_TYPE_FSLINT:
 665                 opp->src[n_IRQ].ivpr &= ~IVPR_SENSE_MASK;
 666                 break;
 667
 668         case IRQ_TYPE_FSLSPECIAL:
 669                 opp->src[n_IRQ].ivpr &= ~(IVPR_POLARITY_MASK | IVPR_SENSE_MASK);
 670                 break;
 671         }
 672
 673         openpic_update_irq(opp, n_IRQ);
 674         pr_debug("Set IVPR %d to 0x%08x -> 0x%08x\n", n_IRQ, val,
 675                 opp->src[n_IRQ].ivpr);
 676 }
 677
 678 static void openpic_gcr_write(struct openpic *opp, uint64_t val)
 679 {
 680         if (val & GCR_RESET) {
 681                 openpic_reset(opp);
 682                 return;
 683         }
 684
 685         opp->gcr &= ~opp->mpic_mode_mask;
 686         opp->gcr |= val & opp->mpic_mode_mask;
 687 }
 688
 689 static int openpic_gbl_write(void *opaque, gpa_t addr, u32 val)
 690 {
 691         struct openpic *opp = opaque;
 692         int err = 0;
 693
 694         pr_debug("%s: addr %#llx <= %08x\n", __func__, addr, val);
 695         if (addr & 0xF)
 696                 return 0;
 697
 698         switch (addr) {
 699         case 0x00:      /* Block Revision Register1 (BRR1) is Readonly */
 700                 break;
 701         case 0x40:
 702         case 0x50:
 703         case 0x60:
 704         case 0x70:
 705         case 0x80:
 706         case 0x90:
 707         case 0xA0:
 708         case 0xB0:
 709                 err = openpic_cpu_write_internal(opp, addr, val,
 710                                                  get_current_cpu());
 711                 break;
 712         case 0x1000:            /* FRR */
 713                 break;
 714         case 0x1020:            /* GCR */
 715                 openpic_gcr_write(opp, val);
 716                 break;
 717         case 0x1080:            /* VIR */
 718                 break;
 719         case 0x1090:            /* PIR */
 720                 /*
 721                  * This register is used to reset a CPU core --
 722                  * let userspace handle it.
 723                  */
 724                 err = -ENXIO;
 725                 break;
 726         case 0x10A0:            /* IPI_IVPR */
 727         case 0x10B0:
 728         case 0x10C0:
 729         case 0x10D0: {
 730                 int idx;
 731                 idx = (addr - 0x10A0) >> 4;
 732                 write_IRQreg_ivpr(opp, opp->irq_ipi0 + idx, val);
 733                 break;
 734         }
 735         case 0x10E0:            /* SPVE */
 736                 opp->spve = val & opp->vector_mask;
 737                 break;
 738         default:
 739                 break;
 740         }
 741
 742         return err;
 743 }
 744
 745 static int openpic_gbl_read(void *opaque, gpa_t addr, u32 *ptr)
 746 {
 747         struct openpic *opp = opaque;
 748         u32 retval;
 749         int err = 0;
 750
 751         pr_debug("%s: addr %#llx\n", __func__, addr);
 752         retval = 0xFFFFFFFF;
 753         if (addr & 0xF)
 754                 goto out;
 755
 756         switch (addr) {
 757         case 0x1000:            /* FRR */
 758                 retval = opp->frr;
 759                 retval |= (opp->nb_cpus - 1) << FRR_NCPU_SHIFT;
 760                 break;
 761         case 0x1020:            /* GCR */
 762                 retval = opp->gcr;
 763                 break;
 764         case 0x1080:            /* VIR */
 765                 retval = opp->vir;
 766                 break;
 767         case 0x1090:            /* PIR */
 768                 retval = 0x00000000;
 769                 break;
 770         case 0x00:              /* Block Revision Register1 (BRR1) */
 771                 retval = opp->brr1;
 772                 break;
 773         case 0x40:
 774         case 0x50:
 775         case 0x60:
 776         case 0x70:
 777         case 0x80:
 778         case 0x90:
 779         case 0xA0:
 780         case 0xB0:
 781                 err = openpic_cpu_read_internal(opp, addr,
 782                         &retval, get_current_cpu());
 783                 break;
 784         case 0x10A0:            /* IPI_IVPR */
 785         case 0x10B0:
 786         case 0x10C0:
 787         case 0x10D0:
 788                 {
 789                         int idx;
 790                         idx = (addr - 0x10A0) >> 4;
 791                         retval = read_IRQreg_ivpr(opp, opp->irq_ipi0 + idx);
 792                 }
 793                 break;
 794         case 0x10E0:            /* SPVE */
 795                 retval = opp->spve;
 796                 break;
 797         default:
 798                 break;
 799         }
 800
 801 out:
 802         pr_debug("%s: => 0x%08x\n", __func__, retval);
 803         *ptr = retval;
 804         return err;
 805 }
 806
 807 static int openpic_tmr_write(void *opaque, gpa_t addr, u32 val)
 808 {
 809         struct openpic *opp = opaque;
 810         int idx;
 811
 812         addr += 0x10f0;
 813
 814         pr_debug("%s: addr %#llx <= %08x\n", __func__, addr, val);
 815         if (addr & 0xF)
 816                 return 0;
 817
 818         if (addr == 0x10f0) {
 819                 /* TFRR */
 820                 opp->tfrr = val;
 821                 return 0;
 822         }
 823
 824         idx = (addr >> 6) & 0x3;
 825         addr = addr & 0x30;
 826
 827         switch (addr & 0x30) {
 828         case 0x00:              /* TCCR */
 829                 break;
 830         case 0x10:              /* TBCR */
 831                 if ((opp->timers[idx].tccr & TCCR_TOG) != 0 &&
 832                     (val & TBCR_CI) == 0 &&
 833                     (opp->timers[idx].tbcr & TBCR_CI) != 0)
 834                         opp->timers[idx].tccr &= ~TCCR_TOG;
 835
 836                 opp->timers[idx].tbcr = val;
 837                 break;
 838         case 0x20:              /* TVPR */
 839                 write_IRQreg_ivpr(opp, opp->irq_tim0 + idx, val);
 840                 break;
 841         case 0x30:              /* TDR */
 842                 write_IRQreg_idr(opp, opp->irq_tim0 + idx, val);
 843                 break;
 844         }
 845
 846         return 0;
 847 }
 848
 849 static int openpic_tmr_read(void *opaque, gpa_t addr, u32 *ptr)
 850 {
 851         struct openpic *opp = opaque;
 852         uint32_t retval = -1;
 853         int idx;
 854
 855         pr_debug("%s: addr %#llx\n", __func__, addr);
 856         if (addr & 0xF)
 857                 goto out;
 858
 859         idx = (addr >> 6) & 0x3;
 860         if (addr == 0x0) {
 861                 /* TFRR */
 862                 retval = opp->tfrr;
 863                 goto out;
 864         }
 865
 866         switch (addr & 0x30) {
 867         case 0x00:              /* TCCR */
 868                 retval = opp->timers[idx].tccr;
 869                 break;
 870         case 0x10:              /* TBCR */
 871                 retval = opp->timers[idx].tbcr;
 872                 break;
 873         case 0x20:              /* TIPV */
 874                 retval = read_IRQreg_ivpr(opp, opp->irq_tim0 + idx);
 875                 break;
 876         case 0x30:              /* TIDE (TIDR) */
 877                 retval = read_IRQreg_idr(opp, opp->irq_tim0 + idx);
 878                 break;
 879         }
 880
 881 out:
 882         pr_debug("%s: => 0x%08x\n", __func__, retval);
 883         *ptr = retval;
 884         return 0;
 885 }
 886
 887 static int openpic_src_write(void *opaque, gpa_t addr, u32 val)
 888 {
 889         struct openpic *opp = opaque;
 890         int idx;
 891
 892         pr_debug("%s: addr %#llx <= %08x\n", __func__, addr, val);
 893
 894         addr = addr & 0xffff;
 895         idx = addr >> 5;
 896
 897         switch (addr & 0x1f) {
 898         case 0x00:
 899                 write_IRQreg_ivpr(opp, idx, val);
 900                 break;
 901         case 0x10:
 902                 write_IRQreg_idr(opp, idx, val);
 903                 break;
 904         case 0x18:
 905                 write_IRQreg_ilr(opp, idx, val);
 906                 break;
 907         }
 908
 909         return 0;
 910 }
 911
 912 static int openpic_src_read(void *opaque, gpa_t addr, u32 *ptr)
 913 {
 914         struct openpic *opp = opaque;
 915         uint32_t retval;
 916         int idx;
 917
 918         pr_debug("%s: addr %#llx\n", __func__, addr);
 919         retval = 0xFFFFFFFF;
 920
 921         addr = addr & 0xffff;
 922         idx = addr >> 5;
 923
 924         switch (addr & 0x1f) {
 925         case 0x00:
 926                 retval = read_IRQreg_ivpr(opp, idx);
 927                 break;
 928         case 0x10:
 929                 retval = read_IRQreg_idr(opp, idx);
 930                 break;
 931         case 0x18:
 932                 retval = read_IRQreg_ilr(opp, idx);
 933                 break;
 934         }
 935
 936         pr_debug("%s: => 0x%08x\n", __func__, retval);
 937         *ptr = retval;
 938         return 0;
 939 }
 940
 941 static int openpic_msi_write(void *opaque, gpa_t addr, u32 val)
 942 {
 943         struct openpic *opp = opaque;
 944         int idx = opp->irq_msi;
 945         int srs, ibs;
 946
 947         pr_debug("%s: addr %#llx <= 0x%08x\n", __func__, addr, val);
 948         if (addr & 0xF)
 949                 return 0;
 950
 951         switch (addr) {
 952         case MSIIR_OFFSET:
 953                 srs = val >> MSIIR_SRS_SHIFT;
 954                 idx += srs;
 955                 ibs = (val & MSIIR_IBS_MASK) >> MSIIR_IBS_SHIFT;
 956                 opp->msi[srs].msir |= 1 << ibs;
 957                 openpic_set_irq(opp, idx, 1);
 958                 break;
 959         default:
 960                 /* most registers are read-only, thus ignored */
 961                 break;
 962         }
 963
 964         return 0;
 965 }
 966
 967 static int openpic_msi_read(void *opaque, gpa_t addr, u32 *ptr)
 968 {
 969         struct openpic *opp = opaque;
 970         uint32_t r = 0;
 971         int i, srs;
 972
 973         pr_debug("%s: addr %#llx\n", __func__, addr);
 974         if (addr & 0xF)
 975                 return -ENXIO;
 976
 977         srs = addr >> 4;
 978
 979         switch (addr) {
 980         case 0x00:
 981         case 0x10:
 982         case 0x20:
 983         case 0x30:
 984         case 0x40:
 985         case 0x50:
 986         case 0x60:
 987         case 0x70:              /* MSIRs */
 988                 r = opp->msi[srs].msir;
 989                 /* Clear on read */
 990                 opp->msi[srs].msir = 0;
 991                 openpic_set_irq(opp, opp->irq_msi + srs, 0);
 992                 break;
 993         case 0x120:             /* MSISR */
 994                 for (i = 0; i < MAX_MSI; i++)
 995                         r |= (opp->msi[i].msir ? 1 : 0) << i;
 996                 break;
 997         }
 998
 999         pr_debug("%s: => 0x%08x\n", __func__, r);
1000         *ptr = r;
1001         return 0;
1002 }
1003
1004 static int openpic_summary_read(void *opaque, gpa_t addr, u32 *ptr)
1005 {
1006         uint32_t r = 0;
1007
1008         pr_debug("%s: addr %#llx\n", __func__, addr);
1009
1010         /* TODO: EISR/EIMR */
1011
1012         *ptr = r;
1013         return 0;
1014 }
1015
1016 static int openpic_summary_write(void *opaque, gpa_t addr, u32 val)
1017 {
1018         pr_debug("%s: addr %#llx <= 0x%08x\n", __func__, addr, val);
1019
1020         /* TODO: EISR/EIMR */
1021         return 0;
1022 }
1023
1024 static int openpic_cpu_write_internal(void *opaque, gpa_t addr,
1025                                       u32 val, int idx)
1026 {
1027         struct openpic *opp = opaque;
1028         struct irq_source *src;
1029         struct irq_dest *dst;
1030         int s_IRQ, n_IRQ;
1031
1032         pr_debug("%s: cpu %d addr %#llx <= 0x%08x\n", __func__, idx,
1033                 addr, val);
1034
1035         if (idx < 0)
1036                 return 0;
1037
1038         if (addr & 0xF)
1039                 return 0;
1040
1041         dst = &opp->dst[idx];
1042         addr &= 0xFF0;
1043         switch (addr) {
1044         case 0x40:              /* IPIDR */
1045         case 0x50:
1046         case 0x60:
1047         case 0x70:
1048                 idx = (addr - 0x40) >> 4;
1049                 /* we use IDE as mask which CPUs to deliver the IPI to still. */
1050                 opp->src[opp->irq_ipi0 + idx].destmask |= val;
1051                 openpic_set_irq(opp, opp->irq_ipi0 + idx, 1);
1052                 openpic_set_irq(opp, opp->irq_ipi0 + idx, 0);
1053                 break;
1054         case 0x80:              /* CTPR */
1055                 dst->ctpr = val & 0x0000000F;
1056
1057                 pr_debug("%s: set CPU %d ctpr to %d, raised %d servicing %d\n",
1058                         __func__, idx, dst->ctpr, dst->raised.priority,
1059                         dst->servicing.priority);
1060
1061                 if (dst->raised.priority <= dst->ctpr) {
1062                         pr_debug("%s: Lower OpenPIC INT output cpu %d due to ctpr\n",
1063                                 __func__, idx);
1064                         mpic_irq_lower(opp, dst, ILR_INTTGT_INT);
1065                 } else if (dst->raised.priority > dst->servicing.priority) {
1066                         pr_debug("%s: Raise OpenPIC INT output cpu %d irq %d\n",
1067                                 __func__, idx, dst->raised.next);
1068                         mpic_irq_raise(opp, dst, ILR_INTTGT_INT);
1069                 }
1070
1071                 break;
1072         case 0x90:              /* WHOAMI */
1073                 /* Read-only register */
1074                 break;
1075         case 0xA0:              /* IACK */
1076                 /* Read-only register */
1077                 break;
1078         case 0xB0: {            /* EOI */
1079                 int notify_eoi;
1080
1081                 pr_debug("EOI\n");
1082                 s_IRQ = IRQ_get_next(opp, &dst->servicing);
1083
1084                 if (s_IRQ < 0) {
1085                         pr_debug("%s: EOI with no interrupt in service\n",
1086                                 __func__);
1087                         break;
1088                 }
1089
1090                 IRQ_resetbit(&dst->servicing, s_IRQ);
1091                 /* Notify listeners that the IRQ is over */
1092                 notify_eoi = s_IRQ;
1093                 /* Set up next servicing IRQ */
1094                 s_IRQ = IRQ_get_next(opp, &dst->servicing);
1095                 /* Check queued interrupts. */
1096                 n_IRQ = IRQ_get_next(opp, &dst->raised);
1097                 src = &opp->src[n_IRQ];
1098                 if (n_IRQ != -1 &&
1099                     (s_IRQ == -1 ||
1100                      IVPR_PRIORITY(src->ivpr) > dst->servicing.priority)) {
1101                         pr_debug("Raise OpenPIC INT output cpu %d irq %d\n",
1102                                 idx, n_IRQ);
1103                         mpic_irq_raise(opp, dst, ILR_INTTGT_INT);
1104                 }
1105
1106                 spin_unlock(&opp->lock);
1107                 kvm_notify_acked_irq(opp->kvm, 0, notify_eoi);
1108                 spin_lock(&opp->lock);
1109
1110                 break;
1111         }
1112         default:
1113                 break;
1114         }
1115
1116         return 0;
1117 }
1118
1119 static int openpic_cpu_write(void *opaque, gpa_t addr, u32 val)
1120 {
1121         struct openpic *opp = opaque;
1122
1123         return openpic_cpu_write_internal(opp, addr, val,
1124                                          (addr & 0x1f000) >> 12);
1125 }
1126
1127 static uint32_t openpic_iack(struct openpic *opp, struct irq_dest *dst,
1128                              int cpu)
1129 {
1130         struct irq_source *src;
1131         int retval, irq;
1132
1133         pr_debug("Lower OpenPIC INT output\n");
1134         mpic_irq_lower(opp, dst, ILR_INTTGT_INT);
1135
1136         irq = IRQ_get_next(opp, &dst->raised);
1137         pr_debug("IACK: irq=%d\n", irq);
1138
1139         if (irq == -1)
1140                 /* No more interrupt pending */
1141                 return opp->spve;
1142
1143         src = &opp->src[irq];
1144         if (!(src->ivpr & IVPR_ACTIVITY_MASK) ||
1145             !(IVPR_PRIORITY(src->ivpr) > dst->ctpr)) {
1146                 pr_err("%s: bad raised IRQ %d ctpr %d ivpr 0x%08x\n",
1147                         __func__, irq, dst->ctpr, src->ivpr);
1148                 openpic_update_irq(opp, irq);
1149                 retval = opp->spve;
1150         } else {
1151                 /* IRQ enter servicing state */
1152                 IRQ_setbit(&dst->servicing, irq);
1153                 retval = IVPR_VECTOR(opp, src->ivpr);
1154         }
1155
1156         if (!src->level) {
1157                 /* edge-sensitive IRQ */
1158                 src->ivpr &= ~IVPR_ACTIVITY_MASK;
1159                 src->pending = 0;
1160                 IRQ_resetbit(&dst->raised, irq);
1161         }
1162
1163         if ((irq >= opp->irq_ipi0) && (irq < (opp->irq_ipi0 + MAX_IPI))) {
1164                 src->destmask &= ~(1 << cpu);
1165                 if (src->destmask && !src->level) {
1166                         /* trigger on CPUs that didn't know about it yet */
1167                         openpic_set_irq(opp, irq, 1);
1168                         openpic_set_irq(opp, irq, 0);
1169                         /* if all CPUs knew about it, set active bit again */
1170                         src->ivpr |= IVPR_ACTIVITY_MASK;
1171                 }
1172         }
1173
1174         return retval;
1175 }
1176
1177 void kvmppc_mpic_set_epr(struct kvm_vcpu *vcpu)
1178 {
1179         struct openpic *opp = vcpu->arch.mpic;
1180         int cpu = vcpu->arch.irq_cpu_id;
1181         unsigned long flags;
1182
1183         spin_lock_irqsave(&opp->lock, flags);
1184
1185         if ((opp->gcr & opp->mpic_mode_mask) == GCR_MODE_PROXY)
1186                 kvmppc_set_epr(vcpu, openpic_iack(opp, &opp->dst[cpu], cpu));
1187
1188         spin_unlock_irqrestore(&opp->lock, flags);
1189 }
1190
1191 static int openpic_cpu_read_internal(void *opaque, gpa_t addr,
1192                                      u32 *ptr, int idx)
1193 {
1194         struct openpic *opp = opaque;
1195         struct irq_dest *dst;
1196         uint32_t retval;
1197
1198         pr_debug("%s: cpu %d addr %#llx\n", __func__, idx, addr);
1199         retval = 0xFFFFFFFF;
1200
1201         if (idx < 0)
1202                 goto out;
1203
1204         if (addr & 0xF)
1205                 goto out;
1206
1207         dst = &opp->dst[idx];
1208         addr &= 0xFF0;
1209         switch (addr) {
1210         case 0x80:              /* CTPR */
1211                 retval = dst->ctpr;
1212                 break;
1213         case 0x90:              /* WHOAMI */
1214                 retval = idx;
1215                 break;
1216         case 0xA0:              /* IACK */
1217                 retval = openpic_iack(opp, dst, idx);
1218                 break;
1219         case 0xB0:              /* EOI */
1220                 retval = 0;
1221                 break;
1222         default:
1223                 break;
1224         }
1225         pr_debug("%s: => 0x%08x\n", __func__, retval);
1226
1227 out:
1228         *ptr = retval;
1229         return 0;
1230 }
1231
1232 static int openpic_cpu_read(void *opaque, gpa_t addr, u32 *ptr)
1233 {
1234         struct openpic *opp = opaque;
1235
1236         return openpic_cpu_read_internal(opp, addr, ptr,
1237                                          (addr & 0x1f000) >> 12);
1238 }
1239
1240 struct mem_reg {
1241         int (*read)(void *opaque, gpa_t addr, u32 *ptr);
1242         int (*write)(void *opaque, gpa_t addr, u32 val);
1243         gpa_t start_addr;
1244         int size;
1245 };
1246
1247 static const struct mem_reg openpic_gbl_mmio = {
1248         .write = openpic_gbl_write,
1249         .read = openpic_gbl_read,
1250         .start_addr = OPENPIC_GLB_REG_START,
1251         .size = OPENPIC_GLB_REG_SIZE,
1252 };
1253
1254 static const struct mem_reg openpic_tmr_mmio = {
1255         .write = openpic_tmr_write,
1256         .read = openpic_tmr_read,
1257         .start_addr = OPENPIC_TMR_REG_START,
1258         .size = OPENPIC_TMR_REG_SIZE,
1259 };
1260
1261 static const struct mem_reg openpic_cpu_mmio = {
1262         .write = openpic_cpu_write,
1263         .read = openpic_cpu_read,
1264         .start_addr = OPENPIC_CPU_REG_START,
1265         .size = OPENPIC_CPU_REG_SIZE,
1266 };
1267
1268 static const struct mem_reg openpic_src_mmio = {
1269         .write = openpic_src_write,
1270         .read = openpic_src_read,
1271         .start_addr = OPENPIC_SRC_REG_START,
1272         .size = OPENPIC_SRC_REG_SIZE,
1273 };
1274
1275 static const struct mem_reg openpic_msi_mmio = {
1276         .read = openpic_msi_read,
1277         .write = openpic_msi_write,
1278         .start_addr = OPENPIC_MSI_REG_START,
1279         .size = OPENPIC_MSI_REG_SIZE,
1280 };
1281
1282 static const struct mem_reg openpic_summary_mmio = {
1283         .read = openpic_summary_read,
1284         .write = openpic_summary_write,
1285         .start_addr = OPENPIC_SUMMARY_REG_START,
1286         .size = OPENPIC_SUMMARY_REG_SIZE,
1287 };
1288
1289 static void add_mmio_region(struct openpic *opp, const struct mem_reg *mr)
1290 {
1291         if (opp->num_mmio_regions >= MAX_MMIO_REGIONS) {
1292                 WARN(1, "kvm mpic: too many mmio regions\n");
1293                 return;
1294         }
1295
1296         opp->mmio_regions[opp->num_mmio_regions++] = mr;
1297 }
1298
1299 static void fsl_common_init(struct openpic *opp)
1300 {
1301         int i;
1302         int virq = MAX_SRC;
1303
1304         add_mmio_region(opp, &openpic_msi_mmio);
1305         add_mmio_region(opp, &openpic_summary_mmio);
1306
1307         opp->vid = VID_REVISION_1_2;
1308         opp->vir = VIR_GENERIC;
1309         opp->vector_mask = 0xFFFF;
1310         opp->tfrr_reset = 0;
1311         opp->ivpr_reset = IVPR_MASK_MASK;
1312         opp->idr_reset = 1 << 0;
1313         opp->max_irq = MAX_IRQ;
1314
1315         opp->irq_ipi0 = virq;
1316         virq += MAX_IPI;
1317         opp->irq_tim0 = virq;
1318         virq += MAX_TMR;
1319
1320         BUG_ON(virq > MAX_IRQ);
1321
1322         opp->irq_msi = 224;
1323
1324         for (i = 0; i < opp->fsl->max_ext; i++)
1325                 opp->src[i].level = false;
1326
1327         /* Internal interrupts, including message and MSI */
1328         for (i = 16; i < MAX_SRC; i++) {
1329                 opp->src[i].type = IRQ_TYPE_FSLINT;
1330                 opp->src[i].level = true;
1331         }
1332
1333         /* timers and IPIs */
1334         for (i = MAX_SRC; i < virq; i++) {
1335                 opp->src[i].type = IRQ_TYPE_FSLSPECIAL;
1336                 opp->src[i].level = false;
1337         }
1338 }
1339
1340 static int kvm_mpic_read_internal(struct openpic *opp, gpa_t addr, u32 *ptr)
1341 {
1342         int i;
1343
1344         for (i = 0; i < opp->num_mmio_regions; i++) {
1345                 const struct mem_reg *mr = opp->mmio_regions[i];
1346
1347                 if (mr->start_addr > addr || addr >= mr->start_addr + mr->size)
1348                         continue;
1349
1350                 return mr->read(opp, addr - mr->start_addr, ptr);
1351         }
1352
1353         return -ENXIO;
1354 }
1355
1356 static int kvm_mpic_write_internal(struct openpic *opp, gpa_t addr, u32 val)
1357 {
1358         int i;
1359
1360         for (i = 0; i < opp->num_mmio_regions; i++) {
1361                 const struct mem_reg *mr = opp->mmio_regions[i];
1362
1363                 if (mr->start_addr > addr || addr >= mr->start_addr + mr->size)
1364                         continue;
1365
1366                 return mr->write(opp, addr - mr->start_addr, val);
1367         }
1368
1369         return -ENXIO;
1370 }
1371
1372 static int kvm_mpic_read(struct kvm_vcpu *vcpu,
1373                          struct kvm_io_device *this,
1374                          gpa_t addr, int len, void *ptr)
1375 {
1376         struct openpic *opp = container_of(this, struct openpic, mmio);
1377         int ret;
1378         union {
1379                 u32 val;
1380                 u8 bytes[4];
1381         } u;
1382
1383         if (addr & (len - 1)) {
1384                 pr_debug("%s: bad alignment %llx/%d\n",
1385                          __func__, addr, len);
1386                 return -EINVAL;
1387         }
1388
1389         spin_lock_irq(&opp->lock);
1390         ret = kvm_mpic_read_internal(opp, addr - opp->reg_base, &u.val);
1391         spin_unlock_irq(&opp->lock);
1392
1393         /*
1394          * Technically only 32-bit accesses are allowed, but be nice to
1395          * people dumping registers a byte at a time -- it works in real
1396          * hardware (reads only, not writes).
1397          */
1398         if (len == 4) {
1399                 *(u32 *)ptr = u.val;
1400                 pr_debug("%s: addr %llx ret %d len 4 val %x\n",
1401                          __func__, addr, ret, u.val);
1402         } else if (len == 1) {
1403                 *(u8 *)ptr = u.bytes[addr & 3];
1404                 pr_debug("%s: addr %llx ret %d len 1 val %x\n",
1405                          __func__, addr, ret, u.bytes[addr & 3]);
1406         } else {
1407                 pr_debug("%s: bad length %d\n", __func__, len);
1408                 return -EINVAL;
1409         }
1410
1411         return ret;
1412 }
1413
1414 static int kvm_mpic_write(struct kvm_vcpu *vcpu,
1415                           struct kvm_io_device *this,
1416                           gpa_t addr, int len, const void *ptr)
1417 {
1418         struct openpic *opp = container_of(this, struct openpic, mmio);
1419         int ret;
1420
1421         if (len != 4) {
1422                 pr_debug("%s: bad length %d\n", __func__, len);
1423                 return -EOPNOTSUPP;
1424         }
1425         if (addr & 3) {
1426                 pr_debug("%s: bad alignment %llx/%d\n", __func__, addr, len);
1427                 return -EOPNOTSUPP;
1428         }
1429
1430         spin_lock_irq(&opp->lock);
1431         ret = kvm_mpic_write_internal(opp, addr - opp->reg_base,
1432                                       *(const u32 *)ptr);
1433         spin_unlock_irq(&opp->lock);
1434
1435         pr_debug("%s: addr %llx ret %d val %x\n",
1436                  __func__, addr, ret, *(const u32 *)ptr);
1437
1438         return ret;
1439 }
1440
1441 static const struct kvm_io_device_ops mpic_mmio_ops = {
1442         .read = kvm_mpic_read,
1443         .write = kvm_mpic_write,
1444 };
1445
1446 static void map_mmio(struct openpic *opp)
1447 {
1448         kvm_iodevice_init(&opp->mmio, &mpic_mmio_ops);
1449
1450         kvm_io_bus_register_dev(opp->kvm, KVM_MMIO_BUS,
1451                                 opp->reg_base, OPENPIC_REG_SIZE,
1452                                 &opp->mmio);
1453 }
1454
1455 static void unmap_mmio(struct openpic *opp)
1456 {
1457         kvm_io_bus_unregister_dev(opp->kvm, KVM_MMIO_BUS, &opp->mmio);
1458 }
1459
1460 static int set_base_addr(struct openpic *opp, struct kvm_device_attr *attr)
1461 {
1462         u64 base;
1463
1464         if (copy_from_user(&base, (u64 __user *)(long)attr->addr, sizeof(u64)))
1465                 return -EFAULT;
1466
1467         if (base & 0x3ffff) {
1468                 pr_debug("kvm mpic %s: KVM_DEV_MPIC_BASE_ADDR %08llx not aligned\n",
1469                          __func__, base);
1470                 return -EINVAL;
1471         }
1472
1473         if (base == opp->reg_base)
1474                 return 0;
1475
1476         mutex_lock(&opp->kvm->slots_lock);
1477
1478         unmap_mmio(opp);
1479         opp->reg_base = base;
1480
1481         pr_debug("kvm mpic %s: KVM_DEV_MPIC_BASE_ADDR %08llx\n",
1482                  __func__, base);
1483
1484         if (base == 0)
1485                 goto out;
1486
1487         map_mmio(opp);
1488
1489 out:
1490         mutex_unlock(&opp->kvm->slots_lock);
1491         return 0;
1492 }
1493
1494 #define ATTR_SET                0
1495 #define ATTR_GET                1
1496
1497 static int access_reg(struct openpic *opp, gpa_t addr, u32 *val, int type)
1498 {
1499         int ret;
1500
1501         if (addr & 3)
1502                 return -ENXIO;
1503
1504         spin_lock_irq(&opp->lock);
1505
1506         if (type == ATTR_SET)
1507                 ret = kvm_mpic_write_internal(opp, addr, *val);
1508         else
1509                 ret = kvm_mpic_read_internal(opp, addr, val);
1510
1511         spin_unlock_irq(&opp->lock);
1512
1513         pr_debug("%s: type %d addr %llx val %x\n", __func__, type, addr, *val);
1514
1515         return ret;
1516 }
1517
1518 static int mpic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1519 {
1520         struct openpic *opp = dev->private;
1521         u32 attr32;
1522
1523         switch (attr->group) {
1524         case KVM_DEV_MPIC_GRP_MISC:
1525                 switch (attr->attr) {
1526                 case KVM_DEV_MPIC_BASE_ADDR:
1527                         return set_base_addr(opp, attr);
1528                 }
1529
1530                 break;
1531
1532         case KVM_DEV_MPIC_GRP_REGISTER:
1533                 if (get_user(attr32, (u32 __user *)(long)attr->addr))
1534                         return -EFAULT;
1535
1536                 return access_reg(opp, attr->attr, &attr32, ATTR_SET);
1537
1538         case KVM_DEV_MPIC_GRP_IRQ_ACTIVE:
1539                 if (attr->attr > MAX_SRC)
1540                         return -EINVAL;
1541
1542                 if (get_user(attr32, (u32 __user *)(long)attr->addr))
1543                         return -EFAULT;
1544
1545                 if (attr32 != 0 && attr32 != 1)
1546                         return -EINVAL;
1547
1548                 spin_lock_irq(&opp->lock);
1549                 openpic_set_irq(opp, attr->attr, attr32);
1550                 spin_unlock_irq(&opp->lock);
1551                 return 0;
1552         }
1553
1554         return -ENXIO;
1555 }
1556
1557 static int mpic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1558 {
1559         struct openpic *opp = dev->private;
1560         u64 attr64;
1561         u32 attr32;
1562         int ret;
1563
1564         switch (attr->group) {
1565         case KVM_DEV_MPIC_GRP_MISC:
1566                 switch (attr->attr) {
1567                 case KVM_DEV_MPIC_BASE_ADDR:
1568                         mutex_lock(&opp->kvm->slots_lock);
1569                         attr64 = opp->reg_base;
1570                         mutex_unlock(&opp->kvm->slots_lock);
1571
1572                         if (copy_to_user((u64 __user *)(long)attr->addr,
1573                                          &attr64, sizeof(u64)))
1574                                 return -EFAULT;
1575
1576                         return 0;
1577                 }
1578
1579                 break;
1580
1581         case KVM_DEV_MPIC_GRP_REGISTER:
1582                 ret = access_reg(opp, attr->attr, &attr32, ATTR_GET);
1583                 if (ret)
1584                         return ret;
1585
1586                 if (put_user(attr32, (u32 __user *)(long)attr->addr))
1587                         return -EFAULT;
1588
1589                 return 0;
1590
1591         case KVM_DEV_MPIC_GRP_IRQ_ACTIVE:
1592                 if (attr->attr > MAX_SRC)
1593                         return -EINVAL;
1594
1595                 spin_lock_irq(&opp->lock);
1596                 attr32 = opp->src[attr->attr].pending;
1597                 spin_unlock_irq(&opp->lock);
1598
1599                 if (put_user(attr32, (u32 __user *)(long)attr->addr))
1600                         return -EFAULT;
1601
1602                 return 0;
1603         }
1604
1605         return -ENXIO;
1606 }
1607
1608 static int mpic_has_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1609 {
1610         switch (attr->group) {
1611         case KVM_DEV_MPIC_GRP_MISC:
1612                 switch (attr->attr) {
1613                 case KVM_DEV_MPIC_BASE_ADDR:
1614                         return 0;
1615                 }
1616
1617                 break;
1618
1619         case KVM_DEV_MPIC_GRP_REGISTER:
1620                 return 0;
1621
1622         case KVM_DEV_MPIC_GRP_IRQ_ACTIVE:
1623                 if (attr->attr > MAX_SRC)
1624                         break;
1625
1626                 return 0;
1627         }
1628
1629         return -ENXIO;
1630 }
1631
1632 static void mpic_destroy(struct kvm_device *dev)
1633 {
1634         struct openpic *opp = dev->private;
1635
1636         dev->kvm->arch.mpic = NULL;
1637         kfree(opp);
1638         kfree(dev);
1639 }
1640
1641 static int mpic_set_default_irq_routing(struct openpic *opp)
1642 {
1643         struct kvm_irq_routing_entry *routing;
1644
1645         /* Create a nop default map, so that dereferencing it still works */
1646         routing = kzalloc((sizeof(*routing)), GFP_KERNEL);
1647         if (!routing)
1648                 return -ENOMEM;
1649
1650         kvm_set_irq_routing(opp->kvm, routing, 0, 0);
1651
1652         kfree(routing);
1653         return 0;
1654 }
1655
1656 static int mpic_create(struct kvm_device *dev, u32 type)
1657 {
1658         struct openpic *opp;
1659         int ret;
1660
1661         /* We only support one MPIC at a time for now */
1662         if (dev->kvm->arch.mpic)
1663                 return -EINVAL;
1664
1665         opp = kzalloc(sizeof(struct openpic), GFP_KERNEL);
1666         if (!opp)
1667                 return -ENOMEM;
1668
1669         dev->private = opp;
1670         opp->kvm = dev->kvm;
1671         opp->dev = dev;
1672         opp->model = type;
1673         spin_lock_init(&opp->lock);
1674
1675         add_mmio_region(opp, &openpic_gbl_mmio);
1676         add_mmio_region(opp, &openpic_tmr_mmio);
1677         add_mmio_region(opp, &openpic_src_mmio);
1678         add_mmio_region(opp, &openpic_cpu_mmio);
1679
1680         switch (opp->model) {
1681         case KVM_DEV_TYPE_FSL_MPIC_20:
1682                 opp->fsl = &fsl_mpic_20;
1683                 opp->brr1 = 0x00400200;
1684                 opp->flags |= OPENPIC_FLAG_IDR_CRIT;
1685                 opp->nb_irqs = 80;
1686                 opp->mpic_mode_mask = GCR_MODE_MIXED;
1687
1688                 fsl_common_init(opp);
1689
1690                 break;
1691
1692         case KVM_DEV_TYPE_FSL_MPIC_42:
1693                 opp->fsl = &fsl_mpic_42;
1694                 opp->brr1 = 0x00400402;
1695                 opp->flags |= OPENPIC_FLAG_ILR;
1696                 opp->nb_irqs = 196;
1697                 opp->mpic_mode_mask = GCR_MODE_PROXY;
1698
1699                 fsl_common_init(opp);
1700
1701                 break;
1702
1703         default:
1704                 ret = -ENODEV;
1705                 goto err;
1706         }
1707
1708         ret = mpic_set_default_irq_routing(opp);
1709         if (ret)
1710                 goto err;
1711
1712         openpic_reset(opp);
1713
1714         smp_wmb();
1715         dev->kvm->arch.mpic = opp;
1716
1717         return 0;
1718
1719 err:
1720         kfree(opp);
1721         return ret;
1722 }
1723
1724 struct kvm_device_ops kvm_mpic_ops = {
1725         .name = "kvm-mpic",
1726         .create = mpic_create,
1727         .destroy = mpic_destroy,
1728         .set_attr = mpic_set_attr,
1729         .get_attr = mpic_get_attr,
1730         .has_attr = mpic_has_attr,
1731 };
1732
1733 int kvmppc_mpic_connect_vcpu(struct kvm_device *dev, struct kvm_vcpu *vcpu,
1734                              u32 cpu)
1735 {
1736         struct openpic *opp = dev->private;
1737         int ret = 0;
1738
1739         if (dev->ops != &kvm_mpic_ops)
1740                 return -EPERM;
1741         if (opp->kvm != vcpu->kvm)
1742                 return -EPERM;
1743         if (cpu < 0 || cpu >= MAX_CPU)
1744                 return -EPERM;
1745
1746         spin_lock_irq(&opp->lock);
1747
1748         if (opp->dst[cpu].vcpu) {
1749                 ret = -EEXIST;
1750                 goto out;
1751         }
1752         if (vcpu->arch.irq_type) {
1753                 ret = -EBUSY;
1754                 goto out;
1755         }
1756
1757         opp->dst[cpu].vcpu = vcpu;
1758         opp->nb_cpus = max(opp->nb_cpus, cpu + 1);
1759
1760         vcpu->arch.mpic = opp;
1761         vcpu->arch.irq_cpu_id = cpu;
1762         vcpu->arch.irq_type = KVMPPC_IRQ_MPIC;
1763
1764         /* This might need to be changed if GCR gets extended */
1765         if (opp->mpic_mode_mask == GCR_MODE_PROXY)
1766                 vcpu->arch.epr_flags |= KVMPPC_EPR_KERNEL;
1767
1768 out:
1769         spin_unlock_irq(&opp->lock);
1770         return ret;
1771 }
1772
1773 /*
1774  * This should only happen immediately before the mpic is destroyed,
1775  * so we shouldn't need to worry about anything still trying to
1776  * access the vcpu pointer.
1777  */
1778 void kvmppc_mpic_disconnect_vcpu(struct openpic *opp, struct kvm_vcpu *vcpu)
1779 {
1780         BUG_ON(!opp->dst[vcpu->arch.irq_cpu_id].vcpu);
1781
1782         opp->dst[vcpu->arch.irq_cpu_id].vcpu = NULL;
1783 }
1784
1785 /*
1786  * Return value:
1787  *  < 0   Interrupt was ignored (masked or not delivered for other reasons)
1788  *  = 0   Interrupt was coalesced (previous irq is still pending)
1789  *  > 0   Number of CPUs interrupt was delivered to
1790  */
1791 static int mpic_set_irq(struct kvm_kernel_irq_routing_entry *e,
1792                         struct kvm *kvm, int irq_source_id, int level,
1793                         bool line_status)
1794 {
1795         u32 irq = e->irqchip.pin;
1796         struct openpic *opp = kvm->arch.mpic;
1797         unsigned long flags;
1798
1799         spin_lock_irqsave(&opp->lock, flags);
1800         openpic_set_irq(opp, irq, level);
1801         spin_unlock_irqrestore(&opp->lock, flags);
1802
1803         /* All code paths we care about don't check for the return value */
1804         return 0;
1805 }
1806
1807 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
1808                 struct kvm *kvm, int irq_source_id, int level, bool line_status)
1809 {
1810         struct openpic *opp = kvm->arch.mpic;
1811         unsigned long flags;
1812
1813         spin_lock_irqsave(&opp->lock, flags);
1814
1815         /*
1816          * XXX We ignore the target address for now, as we only support
1817          *     a single MSI bank.
1818          */
1819         openpic_msi_write(kvm->arch.mpic, MSIIR_OFFSET, e->msi.data);
1820         spin_unlock_irqrestore(&opp->lock, flags);
1821
1822         /* All code paths we care about don't check for the return value */
1823         return 0;
1824 }
1825
1826 int kvm_set_routing_entry(struct kvm_kernel_irq_routing_entry *e,
1827                           const struct kvm_irq_routing_entry *ue)
1828 {
1829         int r = -EINVAL;
1830
1831         switch (ue->type) {
1832         case KVM_IRQ_ROUTING_IRQCHIP:
1833                 e->set = mpic_set_irq;
1834                 e->irqchip.irqchip = ue->u.irqchip.irqchip;
1835                 e->irqchip.pin = ue->u.irqchip.pin;
1836                 if (e->irqchip.pin >= KVM_IRQCHIP_NUM_PINS)
1837                         goto out;
1838                 break;
1839         case KVM_IRQ_ROUTING_MSI:
1840                 e->set = kvm_set_msi;
1841                 e->msi.address_lo = ue->u.msi.address_lo;
1842                 e->msi.address_hi = ue->u.msi.address_hi;
1843                 e->msi.data = ue->u.msi.data;
1844                 break;
1845         default:
1846                 goto out;
1847         }
1848
1849         r = 0;
1850 out:
1851         return r;
1852 }