linux/arch/powerpc/kvm/book3s_xics.c
<<
>>
Prefs
   1/*
   2 * Copyright 2012 Michael Ellerman, IBM Corporation.
   3 * Copyright 2012 Benjamin Herrenschmidt, IBM Corporation.
   4 *
   5 * This program is free software; you can redistribute it and/or modify
   6 * it under the terms of the GNU General Public License, version 2, as
   7 * published by the Free Software Foundation.
   8 */
   9
  10#include <linux/kernel.h>
  11#include <linux/kvm_host.h>
  12#include <linux/err.h>
  13#include <linux/gfp.h>
  14#include <linux/anon_inodes.h>
  15#include <linux/spinlock.h>
  16
  17#include <linux/uaccess.h>
  18#include <asm/kvm_book3s.h>
  19#include <asm/kvm_ppc.h>
  20#include <asm/hvcall.h>
  21#include <asm/xics.h>
  22#include <asm/debugfs.h>
  23#include <asm/time.h>
  24
  25#include <linux/seq_file.h>
  26
  27#include "book3s_xics.h"
  28
  29#if 1
  30#define XICS_DBG(fmt...) do { } while (0)
  31#else
  32#define XICS_DBG(fmt...) trace_printk(fmt)
  33#endif
  34
  35#define ENABLE_REALMODE true
  36#define DEBUG_REALMODE  false
  37
  38/*
  39 * LOCKING
  40 * =======
  41 *
  42 * Each ICS has a spin lock protecting the information about the IRQ
  43 * sources and avoiding simultaneous deliveries of the same interrupt.
  44 *
  45 * ICP operations are done via a single compare & swap transaction
  46 * (most ICP state fits in the union kvmppc_icp_state)
  47 */
  48
  49/*
  50 * TODO
  51 * ====
  52 *
  53 * - To speed up resends, keep a bitmap of "resend" set bits in the
  54 *   ICS
  55 *
  56 * - Speed up server# -> ICP lookup (array ? hash table ?)
  57 *
  58 * - Make ICS lockless as well, or at least a per-interrupt lock or hashed
  59 *   locks array to improve scalability
  60 */
  61
  62/* -- ICS routines -- */
  63
  64static void icp_deliver_irq(struct kvmppc_xics *xics, struct kvmppc_icp *icp,
  65                            u32 new_irq, bool check_resend);
  66
  67/*
  68 * Return value ideally indicates how the interrupt was handled, but no
  69 * callers look at it (given that we don't implement KVM_IRQ_LINE_STATUS),
  70 * so just return 0.
  71 */
  72static int ics_deliver_irq(struct kvmppc_xics *xics, u32 irq, u32 level)
  73{
  74        struct ics_irq_state *state;
  75        struct kvmppc_ics *ics;
  76        u16 src;
  77        u32 pq_old, pq_new;
  78
  79        XICS_DBG("ics deliver %#x (level: %d)\n", irq, level);
  80
  81        ics = kvmppc_xics_find_ics(xics, irq, &src);
  82        if (!ics) {
  83                XICS_DBG("ics_deliver_irq: IRQ 0x%06x not found !\n", irq);
  84                return -EINVAL;
  85        }
  86        state = &ics->irq_state[src];
  87        if (!state->exists)
  88                return -EINVAL;
  89
  90        if (level == KVM_INTERRUPT_SET_LEVEL || level == KVM_INTERRUPT_SET)
  91                level = 1;
  92        else if (level == KVM_INTERRUPT_UNSET)
  93                level = 0;
  94        /*
  95         * Take other values the same as 1, consistent with original code.
  96         * maybe WARN here?
  97         */
  98
  99        if (!state->lsi && level == 0) /* noop for MSI */
 100                return 0;
 101
 102        do {
 103                pq_old = state->pq_state;
 104                if (state->lsi) {
 105                        if (level) {
 106                                if (pq_old & PQ_PRESENTED)
 107                                        /* Setting already set LSI ... */
 108                                        return 0;
 109
 110                                pq_new = PQ_PRESENTED;
 111                        } else
 112                                pq_new = 0;
 113                } else
 114                        pq_new = ((pq_old << 1) & 3) | PQ_PRESENTED;
 115        } while (cmpxchg(&state->pq_state, pq_old, pq_new) != pq_old);
 116
 117        /* Test P=1, Q=0, this is the only case where we present */
 118        if (pq_new == PQ_PRESENTED)
 119                icp_deliver_irq(xics, NULL, irq, false);
 120
 121        /* Record which CPU this arrived on for passed-through interrupts */
 122        if (state->host_irq)
 123                state->intr_cpu = raw_smp_processor_id();
 124
 125        return 0;
 126}
 127
 128static void ics_check_resend(struct kvmppc_xics *xics, struct kvmppc_ics *ics,
 129                             struct kvmppc_icp *icp)
 130{
 131        int i;
 132
 133        for (i = 0; i < KVMPPC_XICS_IRQ_PER_ICS; i++) {
 134                struct ics_irq_state *state = &ics->irq_state[i];
 135                if (state->resend) {
 136                        XICS_DBG("resend %#x prio %#x\n", state->number,
 137                                      state->priority);
 138                        icp_deliver_irq(xics, icp, state->number, true);
 139                }
 140        }
 141}
 142
 143static bool write_xive(struct kvmppc_xics *xics, struct kvmppc_ics *ics,
 144                       struct ics_irq_state *state,
 145                       u32 server, u32 priority, u32 saved_priority)
 146{
 147        bool deliver;
 148        unsigned long flags;
 149
 150        local_irq_save(flags);
 151        arch_spin_lock(&ics->lock);
 152
 153        state->server = server;
 154        state->priority = priority;
 155        state->saved_priority = saved_priority;
 156        deliver = false;
 157        if ((state->masked_pending || state->resend) && priority != MASKED) {
 158                state->masked_pending = 0;
 159                state->resend = 0;
 160                deliver = true;
 161        }
 162
 163        arch_spin_unlock(&ics->lock);
 164        local_irq_restore(flags);
 165
 166        return deliver;
 167}
 168
 169int kvmppc_xics_set_xive(struct kvm *kvm, u32 irq, u32 server, u32 priority)
 170{
 171        struct kvmppc_xics *xics = kvm->arch.xics;
 172        struct kvmppc_icp *icp;
 173        struct kvmppc_ics *ics;
 174        struct ics_irq_state *state;
 175        u16 src;
 176
 177        if (!xics)
 178                return -ENODEV;
 179
 180        ics = kvmppc_xics_find_ics(xics, irq, &src);
 181        if (!ics)
 182                return -EINVAL;
 183        state = &ics->irq_state[src];
 184
 185        icp = kvmppc_xics_find_server(kvm, server);
 186        if (!icp)
 187                return -EINVAL;
 188
 189        XICS_DBG("set_xive %#x server %#x prio %#x MP:%d RS:%d\n",
 190                 irq, server, priority,
 191                 state->masked_pending, state->resend);
 192
 193        if (write_xive(xics, ics, state, server, priority, priority))
 194                icp_deliver_irq(xics, icp, irq, false);
 195
 196        return 0;
 197}
 198
 199int kvmppc_xics_get_xive(struct kvm *kvm, u32 irq, u32 *server, u32 *priority)
 200{
 201        struct kvmppc_xics *xics = kvm->arch.xics;
 202        struct kvmppc_ics *ics;
 203        struct ics_irq_state *state;
 204        u16 src;
 205        unsigned long flags;
 206
 207        if (!xics)
 208                return -ENODEV;
 209
 210        ics = kvmppc_xics_find_ics(xics, irq, &src);
 211        if (!ics)
 212                return -EINVAL;
 213        state = &ics->irq_state[src];
 214
 215        local_irq_save(flags);
 216        arch_spin_lock(&ics->lock);
 217        *server = state->server;
 218        *priority = state->priority;
 219        arch_spin_unlock(&ics->lock);
 220        local_irq_restore(flags);
 221
 222        return 0;
 223}
 224
 225int kvmppc_xics_int_on(struct kvm *kvm, u32 irq)
 226{
 227        struct kvmppc_xics *xics = kvm->arch.xics;
 228        struct kvmppc_icp *icp;
 229        struct kvmppc_ics *ics;
 230        struct ics_irq_state *state;
 231        u16 src;
 232
 233        if (!xics)
 234                return -ENODEV;
 235
 236        ics = kvmppc_xics_find_ics(xics, irq, &src);
 237        if (!ics)
 238                return -EINVAL;
 239        state = &ics->irq_state[src];
 240
 241        icp = kvmppc_xics_find_server(kvm, state->server);
 242        if (!icp)
 243                return -EINVAL;
 244
 245        if (write_xive(xics, ics, state, state->server, state->saved_priority,
 246                       state->saved_priority))
 247                icp_deliver_irq(xics, icp, irq, false);
 248
 249        return 0;
 250}
 251
 252int kvmppc_xics_int_off(struct kvm *kvm, u32 irq)
 253{
 254        struct kvmppc_xics *xics = kvm->arch.xics;
 255        struct kvmppc_ics *ics;
 256        struct ics_irq_state *state;
 257        u16 src;
 258
 259        if (!xics)
 260                return -ENODEV;
 261
 262        ics = kvmppc_xics_find_ics(xics, irq, &src);
 263        if (!ics)
 264                return -EINVAL;
 265        state = &ics->irq_state[src];
 266
 267        write_xive(xics, ics, state, state->server, MASKED, state->priority);
 268
 269        return 0;
 270}
 271
 272/* -- ICP routines, including hcalls -- */
 273
 274static inline bool icp_try_update(struct kvmppc_icp *icp,
 275                                  union kvmppc_icp_state old,
 276                                  union kvmppc_icp_state new,
 277                                  bool change_self)
 278{
 279        bool success;
 280
 281        /* Calculate new output value */
 282        new.out_ee = (new.xisr && (new.pending_pri < new.cppr));
 283
 284        /* Attempt atomic update */
 285        success = cmpxchg64(&icp->state.raw, old.raw, new.raw) == old.raw;
 286        if (!success)
 287                goto bail;
 288
 289        XICS_DBG("UPD [%04lx] - C:%02x M:%02x PP: %02x PI:%06x R:%d O:%d\n",
 290                 icp->server_num,
 291                 old.cppr, old.mfrr, old.pending_pri, old.xisr,
 292                 old.need_resend, old.out_ee);
 293        XICS_DBG("UPD        - C:%02x M:%02x PP: %02x PI:%06x R:%d O:%d\n",
 294                 new.cppr, new.mfrr, new.pending_pri, new.xisr,
 295                 new.need_resend, new.out_ee);
 296        /*
 297         * Check for output state update
 298         *
 299         * Note that this is racy since another processor could be updating
 300         * the state already. This is why we never clear the interrupt output
 301         * here, we only ever set it. The clear only happens prior to doing
 302         * an update and only by the processor itself. Currently we do it
 303         * in Accept (H_XIRR) and Up_Cppr (H_XPPR).
 304         *
 305         * We also do not try to figure out whether the EE state has changed,
 306         * we unconditionally set it if the new state calls for it. The reason
 307         * for that is that we opportunistically remove the pending interrupt
 308         * flag when raising CPPR, so we need to set it back here if an
 309         * interrupt is still pending.
 310         */
 311        if (new.out_ee) {
 312                kvmppc_book3s_queue_irqprio(icp->vcpu,
 313                                            BOOK3S_INTERRUPT_EXTERNAL_LEVEL);
 314                if (!change_self)
 315                        kvmppc_fast_vcpu_kick(icp->vcpu);
 316        }
 317 bail:
 318        return success;
 319}
 320
 321static void icp_check_resend(struct kvmppc_xics *xics,
 322                             struct kvmppc_icp *icp)
 323{
 324        u32 icsid;
 325
 326        /* Order this load with the test for need_resend in the caller */
 327        smp_rmb();
 328        for_each_set_bit(icsid, icp->resend_map, xics->max_icsid + 1) {
 329                struct kvmppc_ics *ics = xics->ics[icsid];
 330
 331                if (!test_and_clear_bit(icsid, icp->resend_map))
 332                        continue;
 333                if (!ics)
 334                        continue;
 335                ics_check_resend(xics, ics, icp);
 336        }
 337}
 338
 339static bool icp_try_to_deliver(struct kvmppc_icp *icp, u32 irq, u8 priority,
 340                               u32 *reject)
 341{
 342        union kvmppc_icp_state old_state, new_state;
 343        bool success;
 344
 345        XICS_DBG("try deliver %#x(P:%#x) to server %#lx\n", irq, priority,
 346                 icp->server_num);
 347
 348        do {
 349                old_state = new_state = READ_ONCE(icp->state);
 350
 351                *reject = 0;
 352
 353                /* See if we can deliver */
 354                success = new_state.cppr > priority &&
 355                        new_state.mfrr > priority &&
 356                        new_state.pending_pri > priority;
 357
 358                /*
 359                 * If we can, check for a rejection and perform the
 360                 * delivery
 361                 */
 362                if (success) {
 363                        *reject = new_state.xisr;
 364                        new_state.xisr = irq;
 365                        new_state.pending_pri = priority;
 366                } else {
 367                        /*
 368                         * If we failed to deliver we set need_resend
 369                         * so a subsequent CPPR state change causes us
 370                         * to try a new delivery.
 371                         */
 372                        new_state.need_resend = true;
 373                }
 374
 375        } while (!icp_try_update(icp, old_state, new_state, false));
 376
 377        return success;
 378}
 379
 380static void icp_deliver_irq(struct kvmppc_xics *xics, struct kvmppc_icp *icp,
 381                            u32 new_irq, bool check_resend)
 382{
 383        struct ics_irq_state *state;
 384        struct kvmppc_ics *ics;
 385        u32 reject;
 386        u16 src;
 387        unsigned long flags;
 388
 389        /*
 390         * This is used both for initial delivery of an interrupt and
 391         * for subsequent rejection.
 392         *
 393         * Rejection can be racy vs. resends. We have evaluated the
 394         * rejection in an atomic ICP transaction which is now complete,
 395         * so potentially the ICP can already accept the interrupt again.
 396         *
 397         * So we need to retry the delivery. Essentially the reject path
 398         * boils down to a failed delivery. Always.
 399         *
 400         * Now the interrupt could also have moved to a different target,
 401         * thus we may need to re-do the ICP lookup as well
 402         */
 403
 404 again:
 405        /* Get the ICS state and lock it */
 406        ics = kvmppc_xics_find_ics(xics, new_irq, &src);
 407        if (!ics) {
 408                XICS_DBG("icp_deliver_irq: IRQ 0x%06x not found !\n", new_irq);
 409                return;
 410        }
 411        state = &ics->irq_state[src];
 412
 413        /* Get a lock on the ICS */
 414        local_irq_save(flags);
 415        arch_spin_lock(&ics->lock);
 416
 417        /* Get our server */
 418        if (!icp || state->server != icp->server_num) {
 419                icp = kvmppc_xics_find_server(xics->kvm, state->server);
 420                if (!icp) {
 421                        pr_warn("icp_deliver_irq: IRQ 0x%06x server 0x%x not found !\n",
 422                                new_irq, state->server);
 423                        goto out;
 424                }
 425        }
 426
 427        if (check_resend)
 428                if (!state->resend)
 429                        goto out;
 430
 431        /* Clear the resend bit of that interrupt */
 432        state->resend = 0;
 433
 434        /*
 435         * If masked, bail out
 436         *
 437         * Note: PAPR doesn't mention anything about masked pending
 438         * when doing a resend, only when doing a delivery.
 439         *
 440         * However that would have the effect of losing a masked
 441         * interrupt that was rejected and isn't consistent with
 442         * the whole masked_pending business which is about not
 443         * losing interrupts that occur while masked.
 444         *
 445         * I don't differentiate normal deliveries and resends, this
 446         * implementation will differ from PAPR and not lose such
 447         * interrupts.
 448         */
 449        if (state->priority == MASKED) {
 450                XICS_DBG("irq %#x masked pending\n", new_irq);
 451                state->masked_pending = 1;
 452                goto out;
 453        }
 454
 455        /*
 456         * Try the delivery, this will set the need_resend flag
 457         * in the ICP as part of the atomic transaction if the
 458         * delivery is not possible.
 459         *
 460         * Note that if successful, the new delivery might have itself
 461         * rejected an interrupt that was "delivered" before we took the
 462         * ics spin lock.
 463         *
 464         * In this case we do the whole sequence all over again for the
 465         * new guy. We cannot assume that the rejected interrupt is less
 466         * favored than the new one, and thus doesn't need to be delivered,
 467         * because by the time we exit icp_try_to_deliver() the target
 468         * processor may well have alrady consumed & completed it, and thus
 469         * the rejected interrupt might actually be already acceptable.
 470         */
 471        if (icp_try_to_deliver(icp, new_irq, state->priority, &reject)) {
 472                /*
 473                 * Delivery was successful, did we reject somebody else ?
 474                 */
 475                if (reject && reject != XICS_IPI) {
 476                        arch_spin_unlock(&ics->lock);
 477                        local_irq_restore(flags);
 478                        new_irq = reject;
 479                        check_resend = 0;
 480                        goto again;
 481                }
 482        } else {
 483                /*
 484                 * We failed to deliver the interrupt we need to set the
 485                 * resend map bit and mark the ICS state as needing a resend
 486                 */
 487                state->resend = 1;
 488
 489                /*
 490                 * Make sure when checking resend, we don't miss the resend
 491                 * if resend_map bit is seen and cleared.
 492                 */
 493                smp_wmb();
 494                set_bit(ics->icsid, icp->resend_map);
 495
 496                /*
 497                 * If the need_resend flag got cleared in the ICP some time
 498                 * between icp_try_to_deliver() atomic update and now, then
 499                 * we know it might have missed the resend_map bit. So we
 500                 * retry
 501                 */
 502                smp_mb();
 503                if (!icp->state.need_resend) {
 504                        state->resend = 0;
 505                        arch_spin_unlock(&ics->lock);
 506                        local_irq_restore(flags);
 507                        check_resend = 0;
 508                        goto again;
 509                }
 510        }
 511 out:
 512        arch_spin_unlock(&ics->lock);
 513        local_irq_restore(flags);
 514}
 515
 516static void icp_down_cppr(struct kvmppc_xics *xics, struct kvmppc_icp *icp,
 517                          u8 new_cppr)
 518{
 519        union kvmppc_icp_state old_state, new_state;
 520        bool resend;
 521
 522        /*
 523         * This handles several related states in one operation:
 524         *
 525         * ICP State: Down_CPPR
 526         *
 527         * Load CPPR with new value and if the XISR is 0
 528         * then check for resends:
 529         *
 530         * ICP State: Resend
 531         *
 532         * If MFRR is more favored than CPPR, check for IPIs
 533         * and notify ICS of a potential resend. This is done
 534         * asynchronously (when used in real mode, we will have
 535         * to exit here).
 536         *
 537         * We do not handle the complete Check_IPI as documented
 538         * here. In the PAPR, this state will be used for both
 539         * Set_MFRR and Down_CPPR. However, we know that we aren't
 540         * changing the MFRR state here so we don't need to handle
 541         * the case of an MFRR causing a reject of a pending irq,
 542         * this will have been handled when the MFRR was set in the
 543         * first place.
 544         *
 545         * Thus we don't have to handle rejects, only resends.
 546         *
 547         * When implementing real mode for HV KVM, resend will lead to
 548         * a H_TOO_HARD return and the whole transaction will be handled
 549         * in virtual mode.
 550         */
 551        do {
 552                old_state = new_state = READ_ONCE(icp->state);
 553
 554                /* Down_CPPR */
 555                new_state.cppr = new_cppr;
 556
 557                /*
 558                 * Cut down Resend / Check_IPI / IPI
 559                 *
 560                 * The logic is that we cannot have a pending interrupt
 561                 * trumped by an IPI at this point (see above), so we
 562                 * know that either the pending interrupt is already an
 563                 * IPI (in which case we don't care to override it) or
 564                 * it's either more favored than us or non existent
 565                 */
 566                if (new_state.mfrr < new_cppr &&
 567                    new_state.mfrr <= new_state.pending_pri) {
 568                        WARN_ON(new_state.xisr != XICS_IPI &&
 569                                new_state.xisr != 0);
 570                        new_state.pending_pri = new_state.mfrr;
 571                        new_state.xisr = XICS_IPI;
 572                }
 573
 574                /* Latch/clear resend bit */
 575                resend = new_state.need_resend;
 576                new_state.need_resend = 0;
 577
 578        } while (!icp_try_update(icp, old_state, new_state, true));
 579
 580        /*
 581         * Now handle resend checks. Those are asynchronous to the ICP
 582         * state update in HW (ie bus transactions) so we can handle them
 583         * separately here too
 584         */
 585        if (resend)
 586                icp_check_resend(xics, icp);
 587}
 588
 589static noinline unsigned long kvmppc_h_xirr(struct kvm_vcpu *vcpu)
 590{
 591        union kvmppc_icp_state old_state, new_state;
 592        struct kvmppc_icp *icp = vcpu->arch.icp;
 593        u32 xirr;
 594
 595        /* First, remove EE from the processor */
 596        kvmppc_book3s_dequeue_irqprio(icp->vcpu,
 597                                      BOOK3S_INTERRUPT_EXTERNAL_LEVEL);
 598
 599        /*
 600         * ICP State: Accept_Interrupt
 601         *
 602         * Return the pending interrupt (if any) along with the
 603         * current CPPR, then clear the XISR & set CPPR to the
 604         * pending priority
 605         */
 606        do {
 607                old_state = new_state = READ_ONCE(icp->state);
 608
 609                xirr = old_state.xisr | (((u32)old_state.cppr) << 24);
 610                if (!old_state.xisr)
 611                        break;
 612                new_state.cppr = new_state.pending_pri;
 613                new_state.pending_pri = 0xff;
 614                new_state.xisr = 0;
 615
 616        } while (!icp_try_update(icp, old_state, new_state, true));
 617
 618        XICS_DBG("h_xirr vcpu %d xirr %#x\n", vcpu->vcpu_id, xirr);
 619
 620        return xirr;
 621}
 622
 623static noinline int kvmppc_h_ipi(struct kvm_vcpu *vcpu, unsigned long server,
 624                                 unsigned long mfrr)
 625{
 626        union kvmppc_icp_state old_state, new_state;
 627        struct kvmppc_xics *xics = vcpu->kvm->arch.xics;
 628        struct kvmppc_icp *icp;
 629        u32 reject;
 630        bool resend;
 631        bool local;
 632
 633        XICS_DBG("h_ipi vcpu %d to server %lu mfrr %#lx\n",
 634                 vcpu->vcpu_id, server, mfrr);
 635
 636        icp = vcpu->arch.icp;
 637        local = icp->server_num == server;
 638        if (!local) {
 639                icp = kvmppc_xics_find_server(vcpu->kvm, server);
 640                if (!icp)
 641                        return H_PARAMETER;
 642        }
 643
 644        /*
 645         * ICP state: Set_MFRR
 646         *
 647         * If the CPPR is more favored than the new MFRR, then
 648         * nothing needs to be rejected as there can be no XISR to
 649         * reject.  If the MFRR is being made less favored then
 650         * there might be a previously-rejected interrupt needing
 651         * to be resent.
 652         *
 653         * ICP state: Check_IPI
 654         *
 655         * If the CPPR is less favored, then we might be replacing
 656         * an interrupt, and thus need to possibly reject it.
 657         *
 658         * ICP State: IPI
 659         *
 660         * Besides rejecting any pending interrupts, we also
 661         * update XISR and pending_pri to mark IPI as pending.
 662         *
 663         * PAPR does not describe this state, but if the MFRR is being
 664         * made less favored than its earlier value, there might be
 665         * a previously-rejected interrupt needing to be resent.
 666         * Ideally, we would want to resend only if
 667         *      prio(pending_interrupt) < mfrr &&
 668         *      prio(pending_interrupt) < cppr
 669         * where pending interrupt is the one that was rejected. But
 670         * we don't have that state, so we simply trigger a resend
 671         * whenever the MFRR is made less favored.
 672         */
 673        do {
 674                old_state = new_state = READ_ONCE(icp->state);
 675
 676                /* Set_MFRR */
 677                new_state.mfrr = mfrr;
 678
 679                /* Check_IPI */
 680                reject = 0;
 681                resend = false;
 682                if (mfrr < new_state.cppr) {
 683                        /* Reject a pending interrupt if not an IPI */
 684                        if (mfrr <= new_state.pending_pri) {
 685                                reject = new_state.xisr;
 686                                new_state.pending_pri = mfrr;
 687                                new_state.xisr = XICS_IPI;
 688                        }
 689                }
 690
 691                if (mfrr > old_state.mfrr) {
 692                        resend = new_state.need_resend;
 693                        new_state.need_resend = 0;
 694                }
 695        } while (!icp_try_update(icp, old_state, new_state, local));
 696
 697        /* Handle reject */
 698        if (reject && reject != XICS_IPI)
 699                icp_deliver_irq(xics, icp, reject, false);
 700
 701        /* Handle resend */
 702        if (resend)
 703                icp_check_resend(xics, icp);
 704
 705        return H_SUCCESS;
 706}
 707
 708static int kvmppc_h_ipoll(struct kvm_vcpu *vcpu, unsigned long server)
 709{
 710        union kvmppc_icp_state state;
 711        struct kvmppc_icp *icp;
 712
 713        icp = vcpu->arch.icp;
 714        if (icp->server_num != server) {
 715                icp = kvmppc_xics_find_server(vcpu->kvm, server);
 716                if (!icp)
 717                        return H_PARAMETER;
 718        }
 719        state = READ_ONCE(icp->state);
 720        kvmppc_set_gpr(vcpu, 4, ((u32)state.cppr << 24) | state.xisr);
 721        kvmppc_set_gpr(vcpu, 5, state.mfrr);
 722        return H_SUCCESS;
 723}
 724
 725static noinline void kvmppc_h_cppr(struct kvm_vcpu *vcpu, unsigned long cppr)
 726{
 727        union kvmppc_icp_state old_state, new_state;
 728        struct kvmppc_xics *xics = vcpu->kvm->arch.xics;
 729        struct kvmppc_icp *icp = vcpu->arch.icp;
 730        u32 reject;
 731
 732        XICS_DBG("h_cppr vcpu %d cppr %#lx\n", vcpu->vcpu_id, cppr);
 733
 734        /*
 735         * ICP State: Set_CPPR
 736         *
 737         * We can safely compare the new value with the current
 738         * value outside of the transaction as the CPPR is only
 739         * ever changed by the processor on itself
 740         */
 741        if (cppr > icp->state.cppr)
 742                icp_down_cppr(xics, icp, cppr);
 743        else if (cppr == icp->state.cppr)
 744                return;
 745
 746        /*
 747         * ICP State: Up_CPPR
 748         *
 749         * The processor is raising its priority, this can result
 750         * in a rejection of a pending interrupt:
 751         *
 752         * ICP State: Reject_Current
 753         *
 754         * We can remove EE from the current processor, the update
 755         * transaction will set it again if needed
 756         */
 757        kvmppc_book3s_dequeue_irqprio(icp->vcpu,
 758                                      BOOK3S_INTERRUPT_EXTERNAL_LEVEL);
 759
 760        do {
 761                old_state = new_state = READ_ONCE(icp->state);
 762
 763                reject = 0;
 764                new_state.cppr = cppr;
 765
 766                if (cppr <= new_state.pending_pri) {
 767                        reject = new_state.xisr;
 768                        new_state.xisr = 0;
 769                        new_state.pending_pri = 0xff;
 770                }
 771
 772        } while (!icp_try_update(icp, old_state, new_state, true));
 773
 774        /*
 775         * Check for rejects. They are handled by doing a new delivery
 776         * attempt (see comments in icp_deliver_irq).
 777         */
 778        if (reject && reject != XICS_IPI)
 779                icp_deliver_irq(xics, icp, reject, false);
 780}
 781
 782static int ics_eoi(struct kvm_vcpu *vcpu, u32 irq)
 783{
 784        struct kvmppc_xics *xics = vcpu->kvm->arch.xics;
 785        struct kvmppc_icp *icp = vcpu->arch.icp;
 786        struct kvmppc_ics *ics;
 787        struct ics_irq_state *state;
 788        u16 src;
 789        u32 pq_old, pq_new;
 790
 791        /*
 792         * ICS EOI handling: For LSI, if P bit is still set, we need to
 793         * resend it.
 794         *
 795         * For MSI, we move Q bit into P (and clear Q). If it is set,
 796         * resend it.
 797         */
 798
 799        ics = kvmppc_xics_find_ics(xics, irq, &src);
 800        if (!ics) {
 801                XICS_DBG("ios_eoi: IRQ 0x%06x not found !\n", irq);
 802                return H_PARAMETER;
 803        }
 804        state = &ics->irq_state[src];
 805
 806        if (state->lsi)
 807                pq_new = state->pq_state;
 808        else
 809                do {
 810                        pq_old = state->pq_state;
 811                        pq_new = pq_old >> 1;
 812                } while (cmpxchg(&state->pq_state, pq_old, pq_new) != pq_old);
 813
 814        if (pq_new & PQ_PRESENTED)
 815                icp_deliver_irq(xics, icp, irq, false);
 816
 817        kvm_notify_acked_irq(vcpu->kvm, 0, irq);
 818
 819        return H_SUCCESS;
 820}
 821
 822static noinline int kvmppc_h_eoi(struct kvm_vcpu *vcpu, unsigned long xirr)
 823{
 824        struct kvmppc_xics *xics = vcpu->kvm->arch.xics;
 825        struct kvmppc_icp *icp = vcpu->arch.icp;
 826        u32 irq = xirr & 0x00ffffff;
 827
 828        XICS_DBG("h_eoi vcpu %d eoi %#lx\n", vcpu->vcpu_id, xirr);
 829
 830        /*
 831         * ICP State: EOI
 832         *
 833         * Note: If EOI is incorrectly used by SW to lower the CPPR
 834         * value (ie more favored), we do not check for rejection of
 835         * a pending interrupt, this is a SW error and PAPR sepcifies
 836         * that we don't have to deal with it.
 837         *
 838         * The sending of an EOI to the ICS is handled after the
 839         * CPPR update
 840         *
 841         * ICP State: Down_CPPR which we handle
 842         * in a separate function as it's shared with H_CPPR.
 843         */
 844        icp_down_cppr(xics, icp, xirr >> 24);
 845
 846        /* IPIs have no EOI */
 847        if (irq == XICS_IPI)
 848                return H_SUCCESS;
 849
 850        return ics_eoi(vcpu, irq);
 851}
 852
 853int kvmppc_xics_rm_complete(struct kvm_vcpu *vcpu, u32 hcall)
 854{
 855        struct kvmppc_xics *xics = vcpu->kvm->arch.xics;
 856        struct kvmppc_icp *icp = vcpu->arch.icp;
 857
 858        XICS_DBG("XICS_RM: H_%x completing, act: %x state: %lx tgt: %p\n",
 859                 hcall, icp->rm_action, icp->rm_dbgstate.raw, icp->rm_dbgtgt);
 860
 861        if (icp->rm_action & XICS_RM_KICK_VCPU) {
 862                icp->n_rm_kick_vcpu++;
 863                kvmppc_fast_vcpu_kick(icp->rm_kick_target);
 864        }
 865        if (icp->rm_action & XICS_RM_CHECK_RESEND) {
 866                icp->n_rm_check_resend++;
 867                icp_check_resend(xics, icp->rm_resend_icp);
 868        }
 869        if (icp->rm_action & XICS_RM_NOTIFY_EOI) {
 870                icp->n_rm_notify_eoi++;
 871                kvm_notify_acked_irq(vcpu->kvm, 0, icp->rm_eoied_irq);
 872        }
 873
 874        icp->rm_action = 0;
 875
 876        return H_SUCCESS;
 877}
 878EXPORT_SYMBOL_GPL(kvmppc_xics_rm_complete);
 879
 880int kvmppc_xics_hcall(struct kvm_vcpu *vcpu, u32 req)
 881{
 882        struct kvmppc_xics *xics = vcpu->kvm->arch.xics;
 883        unsigned long res;
 884        int rc = H_SUCCESS;
 885
 886        /* Check if we have an ICP */
 887        if (!xics || !vcpu->arch.icp)
 888                return H_HARDWARE;
 889
 890        /* These requests don't have real-mode implementations at present */
 891        switch (req) {
 892        case H_XIRR_X:
 893                res = kvmppc_h_xirr(vcpu);
 894                kvmppc_set_gpr(vcpu, 4, res);
 895                kvmppc_set_gpr(vcpu, 5, get_tb());
 896                return rc;
 897        case H_IPOLL:
 898                rc = kvmppc_h_ipoll(vcpu, kvmppc_get_gpr(vcpu, 4));
 899                return rc;
 900        }
 901
 902        /* Check for real mode returning too hard */
 903        if (xics->real_mode && is_kvmppc_hv_enabled(vcpu->kvm))
 904                return kvmppc_xics_rm_complete(vcpu, req);
 905
 906        switch (req) {
 907        case H_XIRR:
 908                res = kvmppc_h_xirr(vcpu);
 909                kvmppc_set_gpr(vcpu, 4, res);
 910                break;
 911        case H_CPPR:
 912                kvmppc_h_cppr(vcpu, kvmppc_get_gpr(vcpu, 4));
 913                break;
 914        case H_EOI:
 915                rc = kvmppc_h_eoi(vcpu, kvmppc_get_gpr(vcpu, 4));
 916                break;
 917        case H_IPI:
 918                rc = kvmppc_h_ipi(vcpu, kvmppc_get_gpr(vcpu, 4),
 919                                  kvmppc_get_gpr(vcpu, 5));
 920                break;
 921        }
 922
 923        return rc;
 924}
 925EXPORT_SYMBOL_GPL(kvmppc_xics_hcall);
 926
 927
 928/* -- Initialisation code etc. -- */
 929
 930static void xics_debugfs_irqmap(struct seq_file *m,
 931                                struct kvmppc_passthru_irqmap *pimap)
 932{
 933        int i;
 934
 935        if (!pimap)
 936                return;
 937        seq_printf(m, "========\nPIRQ mappings: %d maps\n===========\n",
 938                                pimap->n_mapped);
 939        for (i = 0; i < pimap->n_mapped; i++)  {
 940                seq_printf(m, "r_hwirq=%x, v_hwirq=%x\n",
 941                        pimap->mapped[i].r_hwirq, pimap->mapped[i].v_hwirq);
 942        }
 943}
 944
 945static int xics_debug_show(struct seq_file *m, void *private)
 946{
 947        struct kvmppc_xics *xics = m->private;
 948        struct kvm *kvm = xics->kvm;
 949        struct kvm_vcpu *vcpu;
 950        int icsid, i;
 951        unsigned long flags;
 952        unsigned long t_rm_kick_vcpu, t_rm_check_resend;
 953        unsigned long t_rm_notify_eoi;
 954        unsigned long t_reject, t_check_resend;
 955
 956        if (!kvm)
 957                return 0;
 958
 959        t_rm_kick_vcpu = 0;
 960        t_rm_notify_eoi = 0;
 961        t_rm_check_resend = 0;
 962        t_check_resend = 0;
 963        t_reject = 0;
 964
 965        xics_debugfs_irqmap(m, kvm->arch.pimap);
 966
 967        seq_printf(m, "=========\nICP state\n=========\n");
 968
 969        kvm_for_each_vcpu(i, vcpu, kvm) {
 970                struct kvmppc_icp *icp = vcpu->arch.icp;
 971                union kvmppc_icp_state state;
 972
 973                if (!icp)
 974                        continue;
 975
 976                state.raw = READ_ONCE(icp->state.raw);
 977                seq_printf(m, "cpu server %#lx XIRR:%#x PPRI:%#x CPPR:%#x MFRR:%#x OUT:%d NR:%d\n",
 978                           icp->server_num, state.xisr,
 979                           state.pending_pri, state.cppr, state.mfrr,
 980                           state.out_ee, state.need_resend);
 981                t_rm_kick_vcpu += icp->n_rm_kick_vcpu;
 982                t_rm_notify_eoi += icp->n_rm_notify_eoi;
 983                t_rm_check_resend += icp->n_rm_check_resend;
 984                t_check_resend += icp->n_check_resend;
 985                t_reject += icp->n_reject;
 986        }
 987
 988        seq_printf(m, "ICP Guest->Host totals: kick_vcpu=%lu check_resend=%lu notify_eoi=%lu\n",
 989                        t_rm_kick_vcpu, t_rm_check_resend,
 990                        t_rm_notify_eoi);
 991        seq_printf(m, "ICP Real Mode totals: check_resend=%lu resend=%lu\n",
 992                        t_check_resend, t_reject);
 993        for (icsid = 0; icsid <= KVMPPC_XICS_MAX_ICS_ID; icsid++) {
 994                struct kvmppc_ics *ics = xics->ics[icsid];
 995
 996                if (!ics)
 997                        continue;
 998
 999                seq_printf(m, "=========\nICS state for ICS 0x%x\n=========\n",
1000                           icsid);
1001
1002                local_irq_save(flags);
1003                arch_spin_lock(&ics->lock);
1004
1005                for (i = 0; i < KVMPPC_XICS_IRQ_PER_ICS; i++) {
1006                        struct ics_irq_state *irq = &ics->irq_state[i];
1007
1008                        seq_printf(m, "irq 0x%06x: server %#x prio %#x save prio %#x pq_state %d resend %d masked pending %d\n",
1009                                   irq->number, irq->server, irq->priority,
1010                                   irq->saved_priority, irq->pq_state,
1011                                   irq->resend, irq->masked_pending);
1012
1013                }
1014                arch_spin_unlock(&ics->lock);
1015                local_irq_restore(flags);
1016        }
1017        return 0;
1018}
1019
1020static int xics_debug_open(struct inode *inode, struct file *file)
1021{
1022        return single_open(file, xics_debug_show, inode->i_private);
1023}
1024
1025static const struct file_operations xics_debug_fops = {
1026        .open = xics_debug_open,
1027        .read = seq_read,
1028        .llseek = seq_lseek,
1029        .release = single_release,
1030};
1031
1032static void xics_debugfs_init(struct kvmppc_xics *xics)
1033{
1034        char *name;
1035
1036        name = kasprintf(GFP_KERNEL, "kvm-xics-%p", xics);
1037        if (!name) {
1038                pr_err("%s: no memory for name\n", __func__);
1039                return;
1040        }
1041
1042        xics->dentry = debugfs_create_file(name, 0444, powerpc_debugfs_root,
1043                                           xics, &xics_debug_fops);
1044
1045        pr_debug("%s: created %s\n", __func__, name);
1046        kfree(name);
1047}
1048
1049static struct kvmppc_ics *kvmppc_xics_create_ics(struct kvm *kvm,
1050                                        struct kvmppc_xics *xics, int irq)
1051{
1052        struct kvmppc_ics *ics;
1053        int i, icsid;
1054
1055        icsid = irq >> KVMPPC_XICS_ICS_SHIFT;
1056
1057        mutex_lock(&kvm->lock);
1058
1059        /* ICS already exists - somebody else got here first */
1060        if (xics->ics[icsid])
1061                goto out;
1062
1063        /* Create the ICS */
1064        ics = kzalloc(sizeof(struct kvmppc_ics), GFP_KERNEL);
1065        if (!ics)
1066                goto out;
1067
1068        ics->icsid = icsid;
1069
1070        for (i = 0; i < KVMPPC_XICS_IRQ_PER_ICS; i++) {
1071                ics->irq_state[i].number = (icsid << KVMPPC_XICS_ICS_SHIFT) | i;
1072                ics->irq_state[i].priority = MASKED;
1073                ics->irq_state[i].saved_priority = MASKED;
1074        }
1075        smp_wmb();
1076        xics->ics[icsid] = ics;
1077
1078        if (icsid > xics->max_icsid)
1079                xics->max_icsid = icsid;
1080
1081 out:
1082        mutex_unlock(&kvm->lock);
1083        return xics->ics[icsid];
1084}
1085
1086static int kvmppc_xics_create_icp(struct kvm_vcpu *vcpu, unsigned long server_num)
1087{
1088        struct kvmppc_icp *icp;
1089
1090        if (!vcpu->kvm->arch.xics)
1091                return -ENODEV;
1092
1093        if (kvmppc_xics_find_server(vcpu->kvm, server_num))
1094                return -EEXIST;
1095
1096        icp = kzalloc(sizeof(struct kvmppc_icp), GFP_KERNEL);
1097        if (!icp)
1098                return -ENOMEM;
1099
1100        icp->vcpu = vcpu;
1101        icp->server_num = server_num;
1102        icp->state.mfrr = MASKED;
1103        icp->state.pending_pri = MASKED;
1104        vcpu->arch.icp = icp;
1105
1106        XICS_DBG("created server for vcpu %d\n", vcpu->vcpu_id);
1107
1108        return 0;
1109}
1110
1111u64 kvmppc_xics_get_icp(struct kvm_vcpu *vcpu)
1112{
1113        struct kvmppc_icp *icp = vcpu->arch.icp;
1114        union kvmppc_icp_state state;
1115
1116        if (!icp)
1117                return 0;
1118        state = icp->state;
1119        return ((u64)state.cppr << KVM_REG_PPC_ICP_CPPR_SHIFT) |
1120                ((u64)state.xisr << KVM_REG_PPC_ICP_XISR_SHIFT) |
1121                ((u64)state.mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT) |
1122                ((u64)state.pending_pri << KVM_REG_PPC_ICP_PPRI_SHIFT);
1123}
1124
1125int kvmppc_xics_set_icp(struct kvm_vcpu *vcpu, u64 icpval)
1126{
1127        struct kvmppc_icp *icp = vcpu->arch.icp;
1128        struct kvmppc_xics *xics = vcpu->kvm->arch.xics;
1129        union kvmppc_icp_state old_state, new_state;
1130        struct kvmppc_ics *ics;
1131        u8 cppr, mfrr, pending_pri;
1132        u32 xisr;
1133        u16 src;
1134        bool resend;
1135
1136        if (!icp || !xics)
1137                return -ENOENT;
1138
1139        cppr = icpval >> KVM_REG_PPC_ICP_CPPR_SHIFT;
1140        xisr = (icpval >> KVM_REG_PPC_ICP_XISR_SHIFT) &
1141                KVM_REG_PPC_ICP_XISR_MASK;
1142        mfrr = icpval >> KVM_REG_PPC_ICP_MFRR_SHIFT;
1143        pending_pri = icpval >> KVM_REG_PPC_ICP_PPRI_SHIFT;
1144
1145        /* Require the new state to be internally consistent */
1146        if (xisr == 0) {
1147                if (pending_pri != 0xff)
1148                        return -EINVAL;
1149        } else if (xisr == XICS_IPI) {
1150                if (pending_pri != mfrr || pending_pri >= cppr)
1151                        return -EINVAL;
1152        } else {
1153                if (pending_pri >= mfrr || pending_pri >= cppr)
1154                        return -EINVAL;
1155                ics = kvmppc_xics_find_ics(xics, xisr, &src);
1156                if (!ics)
1157                        return -EINVAL;
1158        }
1159
1160        new_state.raw = 0;
1161        new_state.cppr = cppr;
1162        new_state.xisr = xisr;
1163        new_state.mfrr = mfrr;
1164        new_state.pending_pri = pending_pri;
1165
1166        /*
1167         * Deassert the CPU interrupt request.
1168         * icp_try_update will reassert it if necessary.
1169         */
1170        kvmppc_book3s_dequeue_irqprio(icp->vcpu,
1171                                      BOOK3S_INTERRUPT_EXTERNAL_LEVEL);
1172
1173        /*
1174         * Note that if we displace an interrupt from old_state.xisr,
1175         * we don't mark it as rejected.  We expect userspace to set
1176         * the state of the interrupt sources to be consistent with
1177         * the ICP states (either before or afterwards, which doesn't
1178         * matter).  We do handle resends due to CPPR becoming less
1179         * favoured because that is necessary to end up with a
1180         * consistent state in the situation where userspace restores
1181         * the ICS states before the ICP states.
1182         */
1183        do {
1184                old_state = READ_ONCE(icp->state);
1185
1186                if (new_state.mfrr <= old_state.mfrr) {
1187                        resend = false;
1188                        new_state.need_resend = old_state.need_resend;
1189                } else {
1190                        resend = old_state.need_resend;
1191                        new_state.need_resend = 0;
1192                }
1193        } while (!icp_try_update(icp, old_state, new_state, false));
1194
1195        if (resend)
1196                icp_check_resend(xics, icp);
1197
1198        return 0;
1199}
1200
1201static int xics_get_source(struct kvmppc_xics *xics, long irq, u64 addr)
1202{
1203        int ret;
1204        struct kvmppc_ics *ics;
1205        struct ics_irq_state *irqp;
1206        u64 __user *ubufp = (u64 __user *) addr;
1207        u16 idx;
1208        u64 val, prio;
1209        unsigned long flags;
1210
1211        ics = kvmppc_xics_find_ics(xics, irq, &idx);
1212        if (!ics)
1213                return -ENOENT;
1214
1215        irqp = &ics->irq_state[idx];
1216        local_irq_save(flags);
1217        arch_spin_lock(&ics->lock);
1218        ret = -ENOENT;
1219        if (irqp->exists) {
1220                val = irqp->server;
1221                prio = irqp->priority;
1222                if (prio == MASKED) {
1223                        val |= KVM_XICS_MASKED;
1224                        prio = irqp->saved_priority;
1225                }
1226                val |= prio << KVM_XICS_PRIORITY_SHIFT;
1227                if (irqp->lsi) {
1228                        val |= KVM_XICS_LEVEL_SENSITIVE;
1229                        if (irqp->pq_state & PQ_PRESENTED)
1230                                val |= KVM_XICS_PENDING;
1231                } else if (irqp->masked_pending || irqp->resend)
1232                        val |= KVM_XICS_PENDING;
1233
1234                if (irqp->pq_state & PQ_PRESENTED)
1235                        val |= KVM_XICS_PRESENTED;
1236
1237                if (irqp->pq_state & PQ_QUEUED)
1238                        val |= KVM_XICS_QUEUED;
1239
1240                ret = 0;
1241        }
1242        arch_spin_unlock(&ics->lock);
1243        local_irq_restore(flags);
1244
1245        if (!ret && put_user(val, ubufp))
1246                ret = -EFAULT;
1247
1248        return ret;
1249}
1250
1251static int xics_set_source(struct kvmppc_xics *xics, long irq, u64 addr)
1252{
1253        struct kvmppc_ics *ics;
1254        struct ics_irq_state *irqp;
1255        u64 __user *ubufp = (u64 __user *) addr;
1256        u16 idx;
1257        u64 val;
1258        u8 prio;
1259        u32 server;
1260        unsigned long flags;
1261
1262        if (irq < KVMPPC_XICS_FIRST_IRQ || irq >= KVMPPC_XICS_NR_IRQS)
1263                return -ENOENT;
1264
1265        ics = kvmppc_xics_find_ics(xics, irq, &idx);
1266        if (!ics) {
1267                ics = kvmppc_xics_create_ics(xics->kvm, xics, irq);
1268                if (!ics)
1269                        return -ENOMEM;
1270        }
1271        irqp = &ics->irq_state[idx];
1272        if (get_user(val, ubufp))
1273                return -EFAULT;
1274
1275        server = val & KVM_XICS_DESTINATION_MASK;
1276        prio = val >> KVM_XICS_PRIORITY_SHIFT;
1277        if (prio != MASKED &&
1278            kvmppc_xics_find_server(xics->kvm, server) == NULL)
1279                return -EINVAL;
1280
1281        local_irq_save(flags);
1282        arch_spin_lock(&ics->lock);
1283        irqp->server = server;
1284        irqp->saved_priority = prio;
1285        if (val & KVM_XICS_MASKED)
1286                prio = MASKED;
1287        irqp->priority = prio;
1288        irqp->resend = 0;
1289        irqp->masked_pending = 0;
1290        irqp->lsi = 0;
1291        irqp->pq_state = 0;
1292        if (val & KVM_XICS_LEVEL_SENSITIVE)
1293                irqp->lsi = 1;
1294        /* If PENDING, set P in case P is not saved because of old code */
1295        if (val & KVM_XICS_PRESENTED || val & KVM_XICS_PENDING)
1296                irqp->pq_state |= PQ_PRESENTED;
1297        if (val & KVM_XICS_QUEUED)
1298                irqp->pq_state |= PQ_QUEUED;
1299        irqp->exists = 1;
1300        arch_spin_unlock(&ics->lock);
1301        local_irq_restore(flags);
1302
1303        if (val & KVM_XICS_PENDING)
1304                icp_deliver_irq(xics, NULL, irqp->number, false);
1305
1306        return 0;
1307}
1308
1309int kvmppc_xics_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
1310                        bool line_status)
1311{
1312        struct kvmppc_xics *xics = kvm->arch.xics;
1313
1314        if (!xics)
1315                return -ENODEV;
1316        return ics_deliver_irq(xics, irq, level);
1317}
1318
1319static int xics_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1320{
1321        struct kvmppc_xics *xics = dev->private;
1322
1323        switch (attr->group) {
1324        case KVM_DEV_XICS_GRP_SOURCES:
1325                return xics_set_source(xics, attr->attr, attr->addr);
1326        }
1327        return -ENXIO;
1328}
1329
1330static int xics_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1331{
1332        struct kvmppc_xics *xics = dev->private;
1333
1334        switch (attr->group) {
1335        case KVM_DEV_XICS_GRP_SOURCES:
1336                return xics_get_source(xics, attr->attr, attr->addr);
1337        }
1338        return -ENXIO;
1339}
1340
1341static int xics_has_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1342{
1343        switch (attr->group) {
1344        case KVM_DEV_XICS_GRP_SOURCES:
1345                if (attr->attr >= KVMPPC_XICS_FIRST_IRQ &&
1346                    attr->attr < KVMPPC_XICS_NR_IRQS)
1347                        return 0;
1348                break;
1349        }
1350        return -ENXIO;
1351}
1352
1353static void kvmppc_xics_free(struct kvm_device *dev)
1354{
1355        struct kvmppc_xics *xics = dev->private;
1356        int i;
1357        struct kvm *kvm = xics->kvm;
1358
1359        debugfs_remove(xics->dentry);
1360
1361        if (kvm)
1362                kvm->arch.xics = NULL;
1363
1364        for (i = 0; i <= xics->max_icsid; i++)
1365                kfree(xics->ics[i]);
1366        kfree(xics);
1367        kfree(dev);
1368}
1369
1370static int kvmppc_xics_create(struct kvm_device *dev, u32 type)
1371{
1372        struct kvmppc_xics *xics;
1373        struct kvm *kvm = dev->kvm;
1374        int ret = 0;
1375
1376        xics = kzalloc(sizeof(*xics), GFP_KERNEL);
1377        if (!xics)
1378                return -ENOMEM;
1379
1380        dev->private = xics;
1381        xics->dev = dev;
1382        xics->kvm = kvm;
1383
1384        /* Already there ? */
1385        if (kvm->arch.xics)
1386                ret = -EEXIST;
1387        else
1388                kvm->arch.xics = xics;
1389
1390        if (ret) {
1391                kfree(xics);
1392                return ret;
1393        }
1394
1395#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1396        if (cpu_has_feature(CPU_FTR_ARCH_206)) {
1397                /* Enable real mode support */
1398                xics->real_mode = ENABLE_REALMODE;
1399                xics->real_mode_dbg = DEBUG_REALMODE;
1400        }
1401#endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1402
1403        return 0;
1404}
1405
1406static void kvmppc_xics_init(struct kvm_device *dev)
1407{
1408        struct kvmppc_xics *xics = (struct kvmppc_xics *)dev->private;
1409
1410        xics_debugfs_init(xics);
1411}
1412
1413struct kvm_device_ops kvm_xics_ops = {
1414        .name = "kvm-xics",
1415        .create = kvmppc_xics_create,
1416        .init = kvmppc_xics_init,
1417        .destroy = kvmppc_xics_free,
1418        .set_attr = xics_set_attr,
1419        .get_attr = xics_get_attr,
1420        .has_attr = xics_has_attr,
1421};
1422
1423int kvmppc_xics_connect_vcpu(struct kvm_device *dev, struct kvm_vcpu *vcpu,
1424                             u32 xcpu)
1425{
1426        struct kvmppc_xics *xics = dev->private;
1427        int r = -EBUSY;
1428
1429        if (dev->ops != &kvm_xics_ops)
1430                return -EPERM;
1431        if (xics->kvm != vcpu->kvm)
1432                return -EPERM;
1433        if (vcpu->arch.irq_type)
1434                return -EBUSY;
1435
1436        r = kvmppc_xics_create_icp(vcpu, xcpu);
1437        if (!r)
1438                vcpu->arch.irq_type = KVMPPC_IRQ_XICS;
1439
1440        return r;
1441}
1442
1443void kvmppc_xics_free_icp(struct kvm_vcpu *vcpu)
1444{
1445        if (!vcpu->arch.icp)
1446                return;
1447        kfree(vcpu->arch.icp);
1448        vcpu->arch.icp = NULL;
1449        vcpu->arch.irq_type = KVMPPC_IRQ_DEFAULT;
1450}
1451
1452void kvmppc_xics_set_mapped(struct kvm *kvm, unsigned long irq,
1453                            unsigned long host_irq)
1454{
1455        struct kvmppc_xics *xics = kvm->arch.xics;
1456        struct kvmppc_ics *ics;
1457        u16 idx;
1458
1459        ics = kvmppc_xics_find_ics(xics, irq, &idx);
1460        if (!ics)
1461                return;
1462
1463        ics->irq_state[idx].host_irq = host_irq;
1464        ics->irq_state[idx].intr_cpu = -1;
1465}
1466EXPORT_SYMBOL_GPL(kvmppc_xics_set_mapped);
1467
1468void kvmppc_xics_clr_mapped(struct kvm *kvm, unsigned long irq,
1469                            unsigned long host_irq)
1470{
1471        struct kvmppc_xics *xics = kvm->arch.xics;
1472        struct kvmppc_ics *ics;
1473        u16 idx;
1474
1475        ics = kvmppc_xics_find_ics(xics, irq, &idx);
1476        if (!ics)
1477                return;
1478
1479        ics->irq_state[idx].host_irq = 0;
1480}
1481EXPORT_SYMBOL_GPL(kvmppc_xics_clr_mapped);
1482