arch/s390/kvm/interrupt.c

   1 /*
   2  * handling kvm guest interrupts
   3  *
   4  * Copyright IBM Corp. 2008, 2015
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License (version 2 only)
   8  * as published by the Free Software Foundation.
   9  *
  10  *    Author(s): Carsten Otte <cotte@de.ibm.com>
  11  */
  12
  13 #include <linux/interrupt.h>
  14 #include <linux/kvm_host.h>
  15 #include <linux/hrtimer.h>
  16 #include <linux/mmu_context.h>
  17 #include <linux/signal.h>
  18 #include <linux/slab.h>
  19 #include <linux/bitmap.h>
  20 #include <linux/vmalloc.h>
  21 #include <asm/asm-offsets.h>
  22 #include <asm/dis.h>
  23 #include <linux/uaccess.h>
  24 #include <asm/sclp.h>
  25 #include <asm/isc.h>
  26 #include <asm/gmap.h>
  27 #include <asm/switch_to.h>
  28 #include <asm/nmi.h>
  29 #include "kvm-s390.h"
  30 #include "gaccess.h"
  31 #include "trace-s390.h"
  32
  33 #define PFAULT_INIT 0x0600
  34 #define PFAULT_DONE 0x0680
  35 #define VIRTIO_PARAM 0x0d00
  36
  37 /* handle external calls via sigp interpretation facility */
  38 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
  39 {
  40         int c, scn;
  41
  42         if (!(atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_ECALL_PEND))
  43                 return 0;
  44
  45         BUG_ON(!kvm_s390_use_sca_entries());
  46         read_lock(&vcpu->kvm->arch.sca_lock);
  47         if (vcpu->kvm->arch.use_esca) {
  48                 struct esca_block *sca = vcpu->kvm->arch.sca;
  49                 union esca_sigp_ctrl sigp_ctrl =
  50                         sca->cpu[vcpu->vcpu_id].sigp_ctrl;
  51
  52                 c = sigp_ctrl.c;
  53                 scn = sigp_ctrl.scn;
  54         } else {
  55                 struct bsca_block *sca = vcpu->kvm->arch.sca;
  56                 union bsca_sigp_ctrl sigp_ctrl =
  57                         sca->cpu[vcpu->vcpu_id].sigp_ctrl;
  58
  59                 c = sigp_ctrl.c;
  60                 scn = sigp_ctrl.scn;
  61         }
  62         read_unlock(&vcpu->kvm->arch.sca_lock);
  63
  64         if (src_id)
  65                 *src_id = scn;
  66
  67         return c;
  68 }
  69
  70 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
  71 {
  72         int expect, rc;
  73
  74         BUG_ON(!kvm_s390_use_sca_entries());
  75         read_lock(&vcpu->kvm->arch.sca_lock);
  76         if (vcpu->kvm->arch.use_esca) {
  77                 struct esca_block *sca = vcpu->kvm->arch.sca;
  78                 union esca_sigp_ctrl *sigp_ctrl =
  79                         &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
  80                 union esca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
  81
  82                 new_val.scn = src_id;
  83                 new_val.c = 1;
  84                 old_val.c = 0;
  85
  86                 expect = old_val.value;
  87                 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
  88         } else {
  89                 struct bsca_block *sca = vcpu->kvm->arch.sca;
  90                 union bsca_sigp_ctrl *sigp_ctrl =
  91                         &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
  92                 union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
  93
  94                 new_val.scn = src_id;
  95                 new_val.c = 1;
  96                 old_val.c = 0;
  97
  98                 expect = old_val.value;
  99                 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
 100         }
 101         read_unlock(&vcpu->kvm->arch.sca_lock);
 102
 103         if (rc != expect) {
 104                 /* another external call is pending */
 105                 return -EBUSY;
 106         }
 107         atomic_or(CPUSTAT_ECALL_PEND, &vcpu->arch.sie_block->cpuflags);
 108         return 0;
 109 }
 110
 111 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
 112 {
 113         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 114         int rc, expect;
 115
 116         if (!kvm_s390_use_sca_entries())
 117                 return;
 118         atomic_andnot(CPUSTAT_ECALL_PEND, li->cpuflags);
 119         read_lock(&vcpu->kvm->arch.sca_lock);
 120         if (vcpu->kvm->arch.use_esca) {
 121                 struct esca_block *sca = vcpu->kvm->arch.sca;
 122                 union esca_sigp_ctrl *sigp_ctrl =
 123                         &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
 124                 union esca_sigp_ctrl old = *sigp_ctrl;
 125
 126                 expect = old.value;
 127                 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
 128         } else {
 129                 struct bsca_block *sca = vcpu->kvm->arch.sca;
 130                 union bsca_sigp_ctrl *sigp_ctrl =
 131                         &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
 132                 union bsca_sigp_ctrl old = *sigp_ctrl;
 133
 134                 expect = old.value;
 135                 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
 136         }
 137         read_unlock(&vcpu->kvm->arch.sca_lock);
 138         WARN_ON(rc != expect); /* cannot clear? */
 139 }
 140
 141 int psw_extint_disabled(struct kvm_vcpu *vcpu)
 142 {
 143         return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
 144 }
 145
 146 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
 147 {
 148         return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
 149 }
 150
 151 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
 152 {
 153         return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
 154 }
 155
 156 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
 157 {
 158         return psw_extint_disabled(vcpu) &&
 159                psw_ioint_disabled(vcpu) &&
 160                psw_mchk_disabled(vcpu);
 161 }
 162
 163 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
 164 {
 165         if (psw_extint_disabled(vcpu) ||
 166             !(vcpu->arch.sie_block->gcr[0] & 0x800ul))
 167                 return 0;
 168         if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
 169                 /* No timer interrupts when single stepping */
 170                 return 0;
 171         return 1;
 172 }
 173
 174 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
 175 {
 176         if (vcpu->arch.sie_block->ckc >= kvm_s390_get_tod_clock_fast(vcpu->kvm))
 177                 return 0;
 178         return ckc_interrupts_enabled(vcpu);
 179 }
 180
 181 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
 182 {
 183         return !psw_extint_disabled(vcpu) &&
 184                (vcpu->arch.sie_block->gcr[0] & 0x400ul);
 185 }
 186
 187 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
 188 {
 189         if (!cpu_timer_interrupts_enabled(vcpu))
 190                 return 0;
 191         return kvm_s390_get_cpu_timer(vcpu) >> 63;
 192 }
 193
 194 static inline int is_ioirq(unsigned long irq_type)
 195 {
 196         return ((irq_type >= IRQ_PEND_IO_ISC_0) &&
 197                 (irq_type <= IRQ_PEND_IO_ISC_7));
 198 }
 199
 200 static uint64_t isc_to_isc_bits(int isc)
 201 {
 202         return (0x80 >> isc) << 24;
 203 }
 204
 205 static inline u8 int_word_to_isc(u32 int_word)
 206 {
 207         return (int_word & 0x38000000) >> 27;
 208 }
 209
 210 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
 211 {
 212         return vcpu->kvm->arch.float_int.pending_irqs |
 213                vcpu->arch.local_int.pending_irqs;
 214 }
 215
 216 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
 217                                    unsigned long active_mask)
 218 {
 219         int i;
 220
 221         for (i = 0; i <= MAX_ISC; i++)
 222                 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
 223                         active_mask &= ~(1UL << (IRQ_PEND_IO_ISC_0 + i));
 224
 225         return active_mask;
 226 }
 227
 228 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
 229 {
 230         unsigned long active_mask;
 231
 232         active_mask = pending_irqs(vcpu);
 233         if (!active_mask)
 234                 return 0;
 235
 236         if (psw_extint_disabled(vcpu))
 237                 active_mask &= ~IRQ_PEND_EXT_MASK;
 238         if (psw_ioint_disabled(vcpu))
 239                 active_mask &= ~IRQ_PEND_IO_MASK;
 240         else
 241                 active_mask = disable_iscs(vcpu, active_mask);
 242         if (!(vcpu->arch.sie_block->gcr[0] & 0x2000ul))
 243                 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
 244         if (!(vcpu->arch.sie_block->gcr[0] & 0x4000ul))
 245                 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
 246         if (!(vcpu->arch.sie_block->gcr[0] & 0x800ul))
 247                 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
 248         if (!(vcpu->arch.sie_block->gcr[0] & 0x400ul))
 249                 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
 250         if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
 251                 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
 252         if (psw_mchk_disabled(vcpu))
 253                 active_mask &= ~IRQ_PEND_MCHK_MASK;
 254         if (!(vcpu->arch.sie_block->gcr[14] &
 255               vcpu->kvm->arch.float_int.mchk.cr14))
 256                 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
 257
 258         /*
 259          * STOP irqs will never be actively delivered. They are triggered via
 260          * intercept requests and cleared when the stop intercept is performed.
 261          */
 262         __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
 263
 264         return active_mask;
 265 }
 266
 267 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
 268 {
 269         atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
 270         set_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
 271 }
 272
 273 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
 274 {
 275         atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
 276         clear_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
 277 }
 278
 279 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
 280 {
 281         atomic_andnot(CPUSTAT_IO_INT | CPUSTAT_EXT_INT | CPUSTAT_STOP_INT,
 282                     &vcpu->arch.sie_block->cpuflags);
 283         vcpu->arch.sie_block->lctl = 0x0000;
 284         vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
 285
 286         if (guestdbg_enabled(vcpu)) {
 287                 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
 288                                                LCTL_CR10 | LCTL_CR11);
 289                 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
 290         }
 291 }
 292
 293 static void __set_cpuflag(struct kvm_vcpu *vcpu, u32 flag)
 294 {
 295         atomic_or(flag, &vcpu->arch.sie_block->cpuflags);
 296 }
 297
 298 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
 299 {
 300         if (!(pending_irqs(vcpu) & IRQ_PEND_IO_MASK))
 301                 return;
 302         else if (psw_ioint_disabled(vcpu))
 303                 __set_cpuflag(vcpu, CPUSTAT_IO_INT);
 304         else
 305                 vcpu->arch.sie_block->lctl |= LCTL_CR6;
 306 }
 307
 308 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
 309 {
 310         if (!(pending_irqs(vcpu) & IRQ_PEND_EXT_MASK))
 311                 return;
 312         if (psw_extint_disabled(vcpu))
 313                 __set_cpuflag(vcpu, CPUSTAT_EXT_INT);
 314         else
 315                 vcpu->arch.sie_block->lctl |= LCTL_CR0;
 316 }
 317
 318 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
 319 {
 320         if (!(pending_irqs(vcpu) & IRQ_PEND_MCHK_MASK))
 321                 return;
 322         if (psw_mchk_disabled(vcpu))
 323                 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
 324         else
 325                 vcpu->arch.sie_block->lctl |= LCTL_CR14;
 326 }
 327
 328 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
 329 {
 330         if (kvm_s390_is_stop_irq_pending(vcpu))
 331                 __set_cpuflag(vcpu, CPUSTAT_STOP_INT);
 332 }
 333
 334 /* Set interception request for non-deliverable interrupts */
 335 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
 336 {
 337         set_intercept_indicators_io(vcpu);
 338         set_intercept_indicators_ext(vcpu);
 339         set_intercept_indicators_mchk(vcpu);
 340         set_intercept_indicators_stop(vcpu);
 341 }
 342
 343 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
 344 {
 345         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 346         int rc;
 347
 348         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
 349                                          0, 0);
 350
 351         rc  = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
 352                            (u16 *)__LC_EXT_INT_CODE);
 353         rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
 354         rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
 355                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 356         rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
 357                             &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 358         clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
 359         return rc ? -EFAULT : 0;
 360 }
 361
 362 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
 363 {
 364         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 365         int rc;
 366
 367         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
 368                                          0, 0);
 369
 370         rc  = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
 371                            (u16 __user *)__LC_EXT_INT_CODE);
 372         rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
 373         rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
 374                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 375         rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
 376                             &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 377         clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
 378         return rc ? -EFAULT : 0;
 379 }
 380
 381 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
 382 {
 383         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 384         struct kvm_s390_ext_info ext;
 385         int rc;
 386
 387         spin_lock(&li->lock);
 388         ext = li->irq.ext;
 389         clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
 390         li->irq.ext.ext_params2 = 0;
 391         spin_unlock(&li->lock);
 392
 393         VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
 394                    ext.ext_params2);
 395         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
 396                                          KVM_S390_INT_PFAULT_INIT,
 397                                          0, ext.ext_params2);
 398
 399         rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
 400         rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
 401         rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
 402                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 403         rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
 404                             &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 405         rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
 406         return rc ? -EFAULT : 0;
 407 }
 408
 409 static int __write_machine_check(struct kvm_vcpu *vcpu,
 410                                  struct kvm_s390_mchk_info *mchk)
 411 {
 412         unsigned long ext_sa_addr;
 413         unsigned long lc;
 414         freg_t fprs[NUM_FPRS];
 415         union mci mci;
 416         int rc;
 417
 418         mci.val = mchk->mcic;
 419         /* take care of lazy register loading */
 420         save_fpu_regs();
 421         save_access_regs(vcpu->run->s.regs.acrs);
 422
 423         /* Extended save area */
 424         rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
 425                            sizeof(unsigned long));
 426         /* Only bits 0 through 63-LC are used for address formation */
 427         lc = ext_sa_addr & MCESA_LC_MASK;
 428         if (test_kvm_facility(vcpu->kvm, 133)) {
 429                 switch (lc) {
 430                 case 0:
 431                 case 10:
 432                         ext_sa_addr &= ~0x3ffUL;
 433                         break;
 434                 case 11:
 435                         ext_sa_addr &= ~0x7ffUL;
 436                         break;
 437                 case 12:
 438                         ext_sa_addr &= ~0xfffUL;
 439                         break;
 440                 default:
 441                         ext_sa_addr = 0;
 442                         break;
 443                 }
 444         } else {
 445                 ext_sa_addr &= ~0x3ffUL;
 446         }
 447
 448         if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
 449                 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
 450                                     512))
 451                         mci.vr = 0;
 452         } else {
 453                 mci.vr = 0;
 454         }
 455         if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
 456             && (lc == 11 || lc == 12)) {
 457                 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
 458                                     &vcpu->run->s.regs.gscb, 32))
 459                         mci.gs = 0;
 460         } else {
 461                 mci.gs = 0;
 462         }
 463
 464         /* General interruption information */
 465         rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
 466         rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
 467                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 468         rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
 469                             &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 470         rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
 471
 472         /* Register-save areas */
 473         if (MACHINE_HAS_VX) {
 474                 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
 475                 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
 476         } else {
 477                 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
 478                                      vcpu->run->s.regs.fprs, 128);
 479         }
 480         rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
 481                              vcpu->run->s.regs.gprs, 128);
 482         rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
 483                            (u32 __user *) __LC_FP_CREG_SAVE_AREA);
 484         rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
 485                            (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
 486         rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
 487                            (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
 488         rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
 489                            (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
 490         rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
 491                              &vcpu->run->s.regs.acrs, 64);
 492         rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
 493                              &vcpu->arch.sie_block->gcr, 128);
 494
 495         /* Extended interruption information */
 496         rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
 497                            (u32 __user *) __LC_EXT_DAMAGE_CODE);
 498         rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
 499                            (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
 500         rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
 501                              sizeof(mchk->fixed_logout));
 502         return rc ? -EFAULT : 0;
 503 }
 504
 505 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
 506 {
 507         struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
 508         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 509         struct kvm_s390_mchk_info mchk = {};
 510         int deliver = 0;
 511         int rc = 0;
 512
 513         spin_lock(&fi->lock);
 514         spin_lock(&li->lock);
 515         if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
 516             test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
 517                 /*
 518                  * If there was an exigent machine check pending, then any
 519                  * repressible machine checks that might have been pending
 520                  * are indicated along with it, so always clear bits for
 521                  * repressible and exigent interrupts
 522                  */
 523                 mchk = li->irq.mchk;
 524                 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
 525                 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
 526                 memset(&li->irq.mchk, 0, sizeof(mchk));
 527                 deliver = 1;
 528         }
 529         /*
 530          * We indicate floating repressible conditions along with
 531          * other pending conditions. Channel Report Pending and Channel
 532          * Subsystem damage are the only two and and are indicated by
 533          * bits in mcic and masked in cr14.
 534          */
 535         if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
 536                 mchk.mcic |= fi->mchk.mcic;
 537                 mchk.cr14 |= fi->mchk.cr14;
 538                 memset(&fi->mchk, 0, sizeof(mchk));
 539                 deliver = 1;
 540         }
 541         spin_unlock(&li->lock);
 542         spin_unlock(&fi->lock);
 543
 544         if (deliver) {
 545                 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
 546                            mchk.mcic);
 547                 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
 548                                                  KVM_S390_MCHK,
 549                                                  mchk.cr14, mchk.mcic);
 550                 rc = __write_machine_check(vcpu, &mchk);
 551         }
 552         return rc;
 553 }
 554
 555 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
 556 {
 557         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 558         int rc;
 559
 560         VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
 561         vcpu->stat.deliver_restart_signal++;
 562         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
 563
 564         rc  = write_guest_lc(vcpu,
 565                              offsetof(struct lowcore, restart_old_psw),
 566                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 567         rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
 568                             &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 569         clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
 570         return rc ? -EFAULT : 0;
 571 }
 572
 573 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
 574 {
 575         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 576         struct kvm_s390_prefix_info prefix;
 577
 578         spin_lock(&li->lock);
 579         prefix = li->irq.prefix;
 580         li->irq.prefix.address = 0;
 581         clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
 582         spin_unlock(&li->lock);
 583
 584         vcpu->stat.deliver_prefix_signal++;
 585         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
 586                                          KVM_S390_SIGP_SET_PREFIX,
 587                                          prefix.address, 0);
 588
 589         kvm_s390_set_prefix(vcpu, prefix.address);
 590         return 0;
 591 }
 592
 593 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
 594 {
 595         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 596         int rc;
 597         int cpu_addr;
 598
 599         spin_lock(&li->lock);
 600         cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
 601         clear_bit(cpu_addr, li->sigp_emerg_pending);
 602         if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
 603                 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
 604         spin_unlock(&li->lock);
 605
 606         VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
 607         vcpu->stat.deliver_emergency_signal++;
 608         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
 609                                          cpu_addr, 0);
 610
 611         rc  = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
 612                            (u16 *)__LC_EXT_INT_CODE);
 613         rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
 614         rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
 615                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 616         rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
 617                             &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 618         return rc ? -EFAULT : 0;
 619 }
 620
 621 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
 622 {
 623         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 624         struct kvm_s390_extcall_info extcall;
 625         int rc;
 626
 627         spin_lock(&li->lock);
 628         extcall = li->irq.extcall;
 629         li->irq.extcall.code = 0;
 630         clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
 631         spin_unlock(&li->lock);
 632
 633         VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
 634         vcpu->stat.deliver_external_call++;
 635         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
 636                                          KVM_S390_INT_EXTERNAL_CALL,
 637                                          extcall.code, 0);
 638
 639         rc  = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
 640                            (u16 *)__LC_EXT_INT_CODE);
 641         rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
 642         rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
 643                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 644         rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
 645                             sizeof(psw_t));
 646         return rc ? -EFAULT : 0;
 647 }
 648
 649 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
 650 {
 651         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 652         struct kvm_s390_pgm_info pgm_info;
 653         int rc = 0, nullifying = false;
 654         u16 ilen;
 655
 656         spin_lock(&li->lock);
 657         pgm_info = li->irq.pgm;
 658         clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
 659         memset(&li->irq.pgm, 0, sizeof(pgm_info));
 660         spin_unlock(&li->lock);
 661
 662         ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
 663         VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
 664                    pgm_info.code, ilen);
 665         vcpu->stat.deliver_program_int++;
 666         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
 667                                          pgm_info.code, 0);
 668
 669         switch (pgm_info.code & ~PGM_PER) {
 670         case PGM_AFX_TRANSLATION:
 671         case PGM_ASX_TRANSLATION:
 672         case PGM_EX_TRANSLATION:
 673         case PGM_LFX_TRANSLATION:
 674         case PGM_LSTE_SEQUENCE:
 675         case PGM_LSX_TRANSLATION:
 676         case PGM_LX_TRANSLATION:
 677         case PGM_PRIMARY_AUTHORITY:
 678         case PGM_SECONDARY_AUTHORITY:
 679                 nullifying = true;
 680                 /* fall through */
 681         case PGM_SPACE_SWITCH:
 682                 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
 683                                   (u64 *)__LC_TRANS_EXC_CODE);
 684                 break;
 685         case PGM_ALEN_TRANSLATION:
 686         case PGM_ALE_SEQUENCE:
 687         case PGM_ASTE_INSTANCE:
 688         case PGM_ASTE_SEQUENCE:
 689         case PGM_ASTE_VALIDITY:
 690         case PGM_EXTENDED_AUTHORITY:
 691                 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
 692                                   (u8 *)__LC_EXC_ACCESS_ID);
 693                 nullifying = true;
 694                 break;
 695         case PGM_ASCE_TYPE:
 696         case PGM_PAGE_TRANSLATION:
 697         case PGM_REGION_FIRST_TRANS:
 698         case PGM_REGION_SECOND_TRANS:
 699         case PGM_REGION_THIRD_TRANS:
 700         case PGM_SEGMENT_TRANSLATION:
 701                 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
 702                                   (u64 *)__LC_TRANS_EXC_CODE);
 703                 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
 704                                    (u8 *)__LC_EXC_ACCESS_ID);
 705                 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
 706                                    (u8 *)__LC_OP_ACCESS_ID);
 707                 nullifying = true;
 708                 break;
 709         case PGM_MONITOR:
 710                 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
 711                                   (u16 *)__LC_MON_CLASS_NR);
 712                 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
 713                                    (u64 *)__LC_MON_CODE);
 714                 break;
 715         case PGM_VECTOR_PROCESSING:
 716         case PGM_DATA:
 717                 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
 718                                   (u32 *)__LC_DATA_EXC_CODE);
 719                 break;
 720         case PGM_PROTECTION:
 721                 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
 722                                   (u64 *)__LC_TRANS_EXC_CODE);
 723                 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
 724                                    (u8 *)__LC_EXC_ACCESS_ID);
 725                 break;
 726         case PGM_STACK_FULL:
 727         case PGM_STACK_EMPTY:
 728         case PGM_STACK_SPECIFICATION:
 729         case PGM_STACK_TYPE:
 730         case PGM_STACK_OPERATION:
 731         case PGM_TRACE_TABEL:
 732         case PGM_CRYPTO_OPERATION:
 733                 nullifying = true;
 734                 break;
 735         }
 736
 737         if (pgm_info.code & PGM_PER) {
 738                 rc |= put_guest_lc(vcpu, pgm_info.per_code,
 739                                    (u8 *) __LC_PER_CODE);
 740                 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
 741                                    (u8 *)__LC_PER_ATMID);
 742                 rc |= put_guest_lc(vcpu, pgm_info.per_address,
 743                                    (u64 *) __LC_PER_ADDRESS);
 744                 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
 745                                    (u8 *) __LC_PER_ACCESS_ID);
 746         }
 747
 748         if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
 749                 kvm_s390_rewind_psw(vcpu, ilen);
 750
 751         /* bit 1+2 of the target are the ilc, so we can directly use ilen */
 752         rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
 753         rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
 754                                  (u64 *) __LC_LAST_BREAK);
 755         rc |= put_guest_lc(vcpu, pgm_info.code,
 756                            (u16 *)__LC_PGM_INT_CODE);
 757         rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
 758                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 759         rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
 760                             &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 761         return rc ? -EFAULT : 0;
 762 }
 763
 764 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
 765 {
 766         struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
 767         struct kvm_s390_ext_info ext;
 768         int rc = 0;
 769
 770         spin_lock(&fi->lock);
 771         if (!(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
 772                 spin_unlock(&fi->lock);
 773                 return 0;
 774         }
 775         ext = fi->srv_signal;
 776         memset(&fi->srv_signal, 0, sizeof(ext));
 777         clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
 778         spin_unlock(&fi->lock);
 779
 780         VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
 781                    ext.ext_params);
 782         vcpu->stat.deliver_service_signal++;
 783         trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
 784                                          ext.ext_params, 0);
 785
 786         rc  = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
 787         rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
 788         rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
 789                              &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 790         rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
 791                             &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
 792         rc |= put_guest_lc(vcpu, ext.ext_params,
 793                            (u32 *)__LC_EXT_PARAMS);
 794
 795         return rc ? -EFAULT : 0;
 796 }
 797
 798 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
 799 {
 800         struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
 801         struct kvm_s390_interrupt_info *inti;
 802         int rc = 0;
 803
 804         spin_lock(&fi->lock);
 805         inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
 806                                         struct kvm_s390_interrupt_info,
 807                                         list);
 808         if (inti) {
 809                 list_del(&inti->list);
 810                 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
 811         }
 812         if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
 813                 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
 814         spin_unlock(&fi->lock);
 815
 816         if (inti) {
 817                 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
 818                                                  KVM_S390_INT_PFAULT_DONE, 0,
 819                                                  inti->ext.ext_params2);
 820                 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
 821                            inti->ext.ext_params2);
 822
 823                 rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
 824                                 (u16 *)__LC_EXT_INT_CODE);
 825                 rc |= put_guest_lc(vcpu, PFAULT_DONE,
 826                                 (u16 *)__LC_EXT_CPU_ADDR);
 827                 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
 828                                 &vcpu->arch.sie_block->gpsw,
 829                                 sizeof(psw_t));
 830                 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
 831                                 &vcpu->arch.sie_block->gpsw,
 832                                 sizeof(psw_t));
 833                 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
 834                                 (u64 *)__LC_EXT_PARAMS2);
 835                 kfree(inti);
 836         }
 837         return rc ? -EFAULT : 0;
 838 }
 839
 840 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
 841 {
 842         struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
 843         struct kvm_s390_interrupt_info *inti;
 844         int rc = 0;
 845
 846         spin_lock(&fi->lock);
 847         inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
 848                                         struct kvm_s390_interrupt_info,
 849                                         list);
 850         if (inti) {
 851                 VCPU_EVENT(vcpu, 4,
 852                            "deliver: virtio parm: 0x%x,parm64: 0x%llx",
 853                            inti->ext.ext_params, inti->ext.ext_params2);
 854                 vcpu->stat.deliver_virtio_interrupt++;
 855                 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
 856                                 inti->type,
 857                                 inti->ext.ext_params,
 858                                 inti->ext.ext_params2);
 859                 list_del(&inti->list);
 860                 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
 861         }
 862         if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
 863                 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
 864         spin_unlock(&fi->lock);
 865
 866         if (inti) {
 867                 rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
 868                                 (u16 *)__LC_EXT_INT_CODE);
 869                 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
 870                                 (u16 *)__LC_EXT_CPU_ADDR);
 871                 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
 872                                 &vcpu->arch.sie_block->gpsw,
 873                                 sizeof(psw_t));
 874                 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
 875                                 &vcpu->arch.sie_block->gpsw,
 876                                 sizeof(psw_t));
 877                 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
 878                                 (u32 *)__LC_EXT_PARAMS);
 879                 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
 880                                 (u64 *)__LC_EXT_PARAMS2);
 881                 kfree(inti);
 882         }
 883         return rc ? -EFAULT : 0;
 884 }
 885
 886 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
 887                                      unsigned long irq_type)
 888 {
 889         struct list_head *isc_list;
 890         struct kvm_s390_float_interrupt *fi;
 891         struct kvm_s390_interrupt_info *inti = NULL;
 892         int rc = 0;
 893
 894         fi = &vcpu->kvm->arch.float_int;
 895
 896         spin_lock(&fi->lock);
 897         isc_list = &fi->lists[irq_type - IRQ_PEND_IO_ISC_0];
 898         inti = list_first_entry_or_null(isc_list,
 899                                         struct kvm_s390_interrupt_info,
 900                                         list);
 901         if (inti) {
 902                 if (inti->type & KVM_S390_INT_IO_AI_MASK)
 903                         VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
 904                 else
 905                         VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
 906                         inti->io.subchannel_id >> 8,
 907                         inti->io.subchannel_id >> 1 & 0x3,
 908                         inti->io.subchannel_nr);
 909
 910                 vcpu->stat.deliver_io_int++;
 911                 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
 912                                 inti->type,
 913                                 ((__u32)inti->io.subchannel_id << 16) |
 914                                 inti->io.subchannel_nr,
 915                                 ((__u64)inti->io.io_int_parm << 32) |
 916                                 inti->io.io_int_word);
 917                 list_del(&inti->list);
 918                 fi->counters[FIRQ_CNTR_IO] -= 1;
 919         }
 920         if (list_empty(isc_list))
 921                 clear_bit(irq_type, &fi->pending_irqs);
 922         spin_unlock(&fi->lock);
 923
 924         if (inti) {
 925                 rc  = put_guest_lc(vcpu, inti->io.subchannel_id,
 926                                 (u16 *)__LC_SUBCHANNEL_ID);
 927                 rc |= put_guest_lc(vcpu, inti->io.subchannel_nr,
 928                                 (u16 *)__LC_SUBCHANNEL_NR);
 929                 rc |= put_guest_lc(vcpu, inti->io.io_int_parm,
 930                                 (u32 *)__LC_IO_INT_PARM);
 931                 rc |= put_guest_lc(vcpu, inti->io.io_int_word,
 932                                 (u32 *)__LC_IO_INT_WORD);
 933                 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
 934                                 &vcpu->arch.sie_block->gpsw,
 935                                 sizeof(psw_t));
 936                 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
 937                                 &vcpu->arch.sie_block->gpsw,
 938                                 sizeof(psw_t));
 939                 kfree(inti);
 940         }
 941
 942         return rc ? -EFAULT : 0;
 943 }
 944
 945 typedef int (*deliver_irq_t)(struct kvm_vcpu *vcpu);
 946
 947 static const deliver_irq_t deliver_irq_funcs[] = {
 948         [IRQ_PEND_MCHK_EX]        = __deliver_machine_check,
 949         [IRQ_PEND_MCHK_REP]       = __deliver_machine_check,
 950         [IRQ_PEND_PROG]           = __deliver_prog,
 951         [IRQ_PEND_EXT_EMERGENCY]  = __deliver_emergency_signal,
 952         [IRQ_PEND_EXT_EXTERNAL]   = __deliver_external_call,
 953         [IRQ_PEND_EXT_CLOCK_COMP] = __deliver_ckc,
 954         [IRQ_PEND_EXT_CPU_TIMER]  = __deliver_cpu_timer,
 955         [IRQ_PEND_RESTART]        = __deliver_restart,
 956         [IRQ_PEND_SET_PREFIX]     = __deliver_set_prefix,
 957         [IRQ_PEND_PFAULT_INIT]    = __deliver_pfault_init,
 958         [IRQ_PEND_EXT_SERVICE]    = __deliver_service,
 959         [IRQ_PEND_PFAULT_DONE]    = __deliver_pfault_done,
 960         [IRQ_PEND_VIRTIO]         = __deliver_virtio,
 961 };
 962
 963 /* Check whether an external call is pending (deliverable or not) */
 964 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
 965 {
 966         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 967
 968         if (!sclp.has_sigpif)
 969                 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
 970
 971         return sca_ext_call_pending(vcpu, NULL);
 972 }
 973
 974 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
 975 {
 976         if (deliverable_irqs(vcpu))
 977                 return 1;
 978
 979         if (kvm_cpu_has_pending_timer(vcpu))
 980                 return 1;
 981
 982         /* external call pending and deliverable */
 983         if (kvm_s390_ext_call_pending(vcpu) &&
 984             !psw_extint_disabled(vcpu) &&
 985             (vcpu->arch.sie_block->gcr[0] & 0x2000ul))
 986                 return 1;
 987
 988         if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
 989                 return 1;
 990         return 0;
 991 }
 992
 993 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 994 {
 995         return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
 996 }
 997
 998 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
 999 {
1000         u64 now, cputm, sltime = 0;
1001
1002         if (ckc_interrupts_enabled(vcpu)) {
1003                 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1004                 sltime = tod_to_ns(vcpu->arch.sie_block->ckc - now);
1005                 /* already expired or overflow? */
1006                 if (!sltime || vcpu->arch.sie_block->ckc <= now)
1007                         return 0;
1008                 if (cpu_timer_interrupts_enabled(vcpu)) {
1009                         cputm = kvm_s390_get_cpu_timer(vcpu);
1010                         /* already expired? */
1011                         if (cputm >> 63)
1012                                 return 0;
1013                         return min(sltime, tod_to_ns(cputm));
1014                 }
1015         } else if (cpu_timer_interrupts_enabled(vcpu)) {
1016                 sltime = kvm_s390_get_cpu_timer(vcpu);
1017                 /* already expired? */
1018                 if (sltime >> 63)
1019                         return 0;
1020         }
1021         return sltime;
1022 }
1023
1024 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1025 {
1026         u64 sltime;
1027
1028         vcpu->stat.exit_wait_state++;
1029
1030         /* fast path */
1031         if (kvm_arch_vcpu_runnable(vcpu))
1032                 return 0;
1033
1034         if (psw_interrupts_disabled(vcpu)) {
1035                 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1036                 return -EOPNOTSUPP; /* disabled wait */
1037         }
1038
1039         if (!ckc_interrupts_enabled(vcpu) &&
1040             !cpu_timer_interrupts_enabled(vcpu)) {
1041                 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1042                 __set_cpu_idle(vcpu);
1043                 goto no_timer;
1044         }
1045
1046         sltime = __calculate_sltime(vcpu);
1047         if (!sltime)
1048                 return 0;
1049
1050         __set_cpu_idle(vcpu);
1051         hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1052         VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1053 no_timer:
1054         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1055         kvm_vcpu_block(vcpu);
1056         __unset_cpu_idle(vcpu);
1057         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1058
1059         hrtimer_cancel(&vcpu->arch.ckc_timer);
1060         return 0;
1061 }
1062
1063 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1064 {
1065         /*
1066          * We cannot move this into the if, as the CPU might be already
1067          * in kvm_vcpu_block without having the waitqueue set (polling)
1068          */
1069         vcpu->valid_wakeup = true;
1070         if (swait_active(&vcpu->wq)) {
1071                 /*
1072                  * The vcpu gave up the cpu voluntarily, mark it as a good
1073                  * yield-candidate.
1074                  */
1075                 vcpu->preempted = true;
1076                 swake_up(&vcpu->wq);
1077                 vcpu->stat.halt_wakeup++;
1078         }
1079         /*
1080          * The VCPU might not be sleeping but is executing the VSIE. Let's
1081          * kick it, so it leaves the SIE to process the request.
1082          */
1083         kvm_s390_vsie_kick(vcpu);
1084 }
1085
1086 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1087 {
1088         struct kvm_vcpu *vcpu;
1089         u64 sltime;
1090
1091         vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1092         sltime = __calculate_sltime(vcpu);
1093
1094         /*
1095          * If the monotonic clock runs faster than the tod clock we might be
1096          * woken up too early and have to go back to sleep to avoid deadlocks.
1097          */
1098         if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1099                 return HRTIMER_RESTART;
1100         kvm_s390_vcpu_wakeup(vcpu);
1101         return HRTIMER_NORESTART;
1102 }
1103
1104 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1105 {
1106         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1107
1108         spin_lock(&li->lock);
1109         li->pending_irqs = 0;
1110         bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1111         memset(&li->irq, 0, sizeof(li->irq));
1112         spin_unlock(&li->lock);
1113
1114         sca_clear_ext_call(vcpu);
1115 }
1116
1117 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1118 {
1119         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1120         deliver_irq_t func;
1121         int rc = 0;
1122         unsigned long irq_type;
1123         unsigned long irqs;
1124
1125         __reset_intercept_indicators(vcpu);
1126
1127         /* pending ckc conditions might have been invalidated */
1128         clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1129         if (ckc_irq_pending(vcpu))
1130                 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1131
1132         /* pending cpu timer conditions might have been invalidated */
1133         clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1134         if (cpu_timer_irq_pending(vcpu))
1135                 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1136
1137         while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1138                 /* bits are in the order of interrupt priority */
1139                 irq_type = find_first_bit(&irqs, IRQ_PEND_COUNT);
1140                 if (is_ioirq(irq_type)) {
1141                         rc = __deliver_io(vcpu, irq_type);
1142                 } else {
1143                         func = deliver_irq_funcs[irq_type];
1144                         if (!func) {
1145                                 WARN_ON_ONCE(func == NULL);
1146                                 clear_bit(irq_type, &li->pending_irqs);
1147                                 continue;
1148                         }
1149                         rc = func(vcpu);
1150                 }
1151         }
1152
1153         set_intercept_indicators(vcpu);
1154
1155         return rc;
1156 }
1157
1158 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1159 {
1160         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1161
1162         VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1163         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1164                                    irq->u.pgm.code, 0);
1165
1166         if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1167                 /* auto detection if no valid ILC was given */
1168                 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1169                 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1170                 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1171         }
1172
1173         if (irq->u.pgm.code == PGM_PER) {
1174                 li->irq.pgm.code |= PGM_PER;
1175                 li->irq.pgm.flags = irq->u.pgm.flags;
1176                 /* only modify PER related information */
1177                 li->irq.pgm.per_address = irq->u.pgm.per_address;
1178                 li->irq.pgm.per_code = irq->u.pgm.per_code;
1179                 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1180                 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1181         } else if (!(irq->u.pgm.code & PGM_PER)) {
1182                 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1183                                    irq->u.pgm.code;
1184                 li->irq.pgm.flags = irq->u.pgm.flags;
1185                 /* only modify non-PER information */
1186                 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1187                 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1188                 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1189                 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1190                 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1191                 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1192         } else {
1193                 li->irq.pgm = irq->u.pgm;
1194         }
1195         set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1196         return 0;
1197 }
1198
1199 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1200 {
1201         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1202
1203         VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1204                    irq->u.ext.ext_params2);
1205         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1206                                    irq->u.ext.ext_params,
1207                                    irq->u.ext.ext_params2);
1208
1209         li->irq.ext = irq->u.ext;
1210         set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1211         atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1212         return 0;
1213 }
1214
1215 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1216 {
1217         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1218         struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1219         uint16_t src_id = irq->u.extcall.code;
1220
1221         VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1222                    src_id);
1223         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1224                                    src_id, 0);
1225
1226         /* sending vcpu invalid */
1227         if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1228                 return -EINVAL;
1229
1230         if (sclp.has_sigpif)
1231                 return sca_inject_ext_call(vcpu, src_id);
1232
1233         if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1234                 return -EBUSY;
1235         *extcall = irq->u.extcall;
1236         atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1237         return 0;
1238 }
1239
1240 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1241 {
1242         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1243         struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1244
1245         VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1246                    irq->u.prefix.address);
1247         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1248                                    irq->u.prefix.address, 0);
1249
1250         if (!is_vcpu_stopped(vcpu))
1251                 return -EBUSY;
1252
1253         *prefix = irq->u.prefix;
1254         set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1255         return 0;
1256 }
1257
1258 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1259 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1260 {
1261         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1262         struct kvm_s390_stop_info *stop = &li->irq.stop;
1263         int rc = 0;
1264
1265         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1266
1267         if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1268                 return -EINVAL;
1269
1270         if (is_vcpu_stopped(vcpu)) {
1271                 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1272                         rc = kvm_s390_store_status_unloaded(vcpu,
1273                                                 KVM_S390_STORE_STATUS_NOADDR);
1274                 return rc;
1275         }
1276
1277         if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1278                 return -EBUSY;
1279         stop->flags = irq->u.stop.flags;
1280         __set_cpuflag(vcpu, CPUSTAT_STOP_INT);
1281         return 0;
1282 }
1283
1284 static int __inject_sigp_restart(struct kvm_vcpu *vcpu,
1285                                  struct kvm_s390_irq *irq)
1286 {
1287         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1288
1289         VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1290         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1291
1292         set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1293         return 0;
1294 }
1295
1296 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1297                                    struct kvm_s390_irq *irq)
1298 {
1299         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1300
1301         VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1302                    irq->u.emerg.code);
1303         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1304                                    irq->u.emerg.code, 0);
1305
1306         /* sending vcpu invalid */
1307         if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1308                 return -EINVAL;
1309
1310         set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1311         set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1312         atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1313         return 0;
1314 }
1315
1316 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1317 {
1318         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1319         struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1320
1321         VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1322                    irq->u.mchk.mcic);
1323         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1324                                    irq->u.mchk.mcic);
1325
1326         /*
1327          * Because repressible machine checks can be indicated along with
1328          * exigent machine checks (PoP, Chapter 11, Interruption action)
1329          * we need to combine cr14, mcic and external damage code.
1330          * Failing storage address and the logout area should not be or'ed
1331          * together, we just indicate the last occurrence of the corresponding
1332          * machine check
1333          */
1334         mchk->cr14 |= irq->u.mchk.cr14;
1335         mchk->mcic |= irq->u.mchk.mcic;
1336         mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1337         mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1338         memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1339                sizeof(mchk->fixed_logout));
1340         if (mchk->mcic & MCHK_EX_MASK)
1341                 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1342         else if (mchk->mcic & MCHK_REP_MASK)
1343                 set_bit(IRQ_PEND_MCHK_REP,  &li->pending_irqs);
1344         return 0;
1345 }
1346
1347 static int __inject_ckc(struct kvm_vcpu *vcpu)
1348 {
1349         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1350
1351         VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1352         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1353                                    0, 0);
1354
1355         set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1356         atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1357         return 0;
1358 }
1359
1360 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1361 {
1362         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1363
1364         VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1365         trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1366                                    0, 0);
1367
1368         set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1369         atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1370         return 0;
1371 }
1372
1373 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1374                                                   int isc, u32 schid)
1375 {
1376         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1377         struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1378         struct kvm_s390_interrupt_info *iter;
1379         u16 id = (schid & 0xffff0000U) >> 16;
1380         u16 nr = schid & 0x0000ffffU;
1381
1382         spin_lock(&fi->lock);
1383         list_for_each_entry(iter, isc_list, list) {
1384                 if (schid && (id != iter->io.subchannel_id ||
1385                               nr != iter->io.subchannel_nr))
1386                         continue;
1387                 /* found an appropriate entry */
1388                 list_del_init(&iter->list);
1389                 fi->counters[FIRQ_CNTR_IO] -= 1;
1390                 if (list_empty(isc_list))
1391                         clear_bit(IRQ_PEND_IO_ISC_0 + isc, &fi->pending_irqs);
1392                 spin_unlock(&fi->lock);
1393                 return iter;
1394         }
1395         spin_unlock(&fi->lock);
1396         return NULL;
1397 }
1398
1399 /*
1400  * Dequeue and return an I/O interrupt matching any of the interruption
1401  * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1402  */
1403 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1404                                                     u64 isc_mask, u32 schid)
1405 {
1406         struct kvm_s390_interrupt_info *inti = NULL;
1407         int isc;
1408
1409         for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1410                 if (isc_mask & isc_to_isc_bits(isc))
1411                         inti = get_io_int(kvm, isc, schid);
1412         }
1413         return inti;
1414 }
1415
1416 #define SCCB_MASK 0xFFFFFFF8
1417 #define SCCB_EVENT_PENDING 0x3
1418
1419 static int __inject_service(struct kvm *kvm,
1420                              struct kvm_s390_interrupt_info *inti)
1421 {
1422         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1423
1424         spin_lock(&fi->lock);
1425         fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1426         /*
1427          * Early versions of the QEMU s390 bios will inject several
1428          * service interrupts after another without handling a
1429          * condition code indicating busy.
1430          * We will silently ignore those superfluous sccb values.
1431          * A future version of QEMU will take care of serialization
1432          * of servc requests
1433          */
1434         if (fi->srv_signal.ext_params & SCCB_MASK)
1435                 goto out;
1436         fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1437         set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1438 out:
1439         spin_unlock(&fi->lock);
1440         kfree(inti);
1441         return 0;
1442 }
1443
1444 static int __inject_virtio(struct kvm *kvm,
1445                             struct kvm_s390_interrupt_info *inti)
1446 {
1447         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1448
1449         spin_lock(&fi->lock);
1450         if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1451                 spin_unlock(&fi->lock);
1452                 return -EBUSY;
1453         }
1454         fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1455         list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1456         set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1457         spin_unlock(&fi->lock);
1458         return 0;
1459 }
1460
1461 static int __inject_pfault_done(struct kvm *kvm,
1462                                  struct kvm_s390_interrupt_info *inti)
1463 {
1464         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1465
1466         spin_lock(&fi->lock);
1467         if (fi->counters[FIRQ_CNTR_PFAULT] >=
1468                 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1469                 spin_unlock(&fi->lock);
1470                 return -EBUSY;
1471         }
1472         fi->counters[FIRQ_CNTR_PFAULT] += 1;
1473         list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1474         set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1475         spin_unlock(&fi->lock);
1476         return 0;
1477 }
1478
1479 #define CR_PENDING_SUBCLASS 28
1480 static int __inject_float_mchk(struct kvm *kvm,
1481                                 struct kvm_s390_interrupt_info *inti)
1482 {
1483         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1484
1485         spin_lock(&fi->lock);
1486         fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1487         fi->mchk.mcic |= inti->mchk.mcic;
1488         set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1489         spin_unlock(&fi->lock);
1490         kfree(inti);
1491         return 0;
1492 }
1493
1494 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1495 {
1496         struct kvm_s390_float_interrupt *fi;
1497         struct list_head *list;
1498         int isc;
1499
1500         fi = &kvm->arch.float_int;
1501         spin_lock(&fi->lock);
1502         if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1503                 spin_unlock(&fi->lock);
1504                 return -EBUSY;
1505         }
1506         fi->counters[FIRQ_CNTR_IO] += 1;
1507
1508         if (inti->type & KVM_S390_INT_IO_AI_MASK)
1509                 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1510         else
1511                 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1512                         inti->io.subchannel_id >> 8,
1513                         inti->io.subchannel_id >> 1 & 0x3,
1514                         inti->io.subchannel_nr);
1515         isc = int_word_to_isc(inti->io.io_int_word);
1516         list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1517         list_add_tail(&inti->list, list);
1518         set_bit(IRQ_PEND_IO_ISC_0 + isc, &fi->pending_irqs);
1519         spin_unlock(&fi->lock);
1520         return 0;
1521 }
1522
1523 /*
1524  * Find a destination VCPU for a floating irq and kick it.
1525  */
1526 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1527 {
1528         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1529         struct kvm_s390_local_interrupt *li;
1530         struct kvm_vcpu *dst_vcpu;
1531         int sigcpu, online_vcpus, nr_tries = 0;
1532
1533         online_vcpus = atomic_read(&kvm->online_vcpus);
1534         if (!online_vcpus)
1535                 return;
1536
1537         /* find idle VCPUs first, then round robin */
1538         sigcpu = find_first_bit(fi->idle_mask, online_vcpus);
1539         if (sigcpu == online_vcpus) {
1540                 do {
1541                         sigcpu = fi->next_rr_cpu;
1542                         fi->next_rr_cpu = (fi->next_rr_cpu + 1) % online_vcpus;
1543                         /* avoid endless loops if all vcpus are stopped */
1544                         if (nr_tries++ >= online_vcpus)
1545                                 return;
1546                 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1547         }
1548         dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1549
1550         /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1551         li = &dst_vcpu->arch.local_int;
1552         spin_lock(&li->lock);
1553         switch (type) {
1554         case KVM_S390_MCHK:
1555                 atomic_or(CPUSTAT_STOP_INT, li->cpuflags);
1556                 break;
1557         case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1558                 atomic_or(CPUSTAT_IO_INT, li->cpuflags);
1559                 break;
1560         default:
1561                 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1562                 break;
1563         }
1564         spin_unlock(&li->lock);
1565         kvm_s390_vcpu_wakeup(dst_vcpu);
1566 }
1567
1568 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1569 {
1570         u64 type = READ_ONCE(inti->type);
1571         int rc;
1572
1573         switch (type) {
1574         case KVM_S390_MCHK:
1575                 rc = __inject_float_mchk(kvm, inti);
1576                 break;
1577         case KVM_S390_INT_VIRTIO:
1578                 rc = __inject_virtio(kvm, inti);
1579                 break;
1580         case KVM_S390_INT_SERVICE:
1581                 rc = __inject_service(kvm, inti);
1582                 break;
1583         case KVM_S390_INT_PFAULT_DONE:
1584                 rc = __inject_pfault_done(kvm, inti);
1585                 break;
1586         case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1587                 rc = __inject_io(kvm, inti);
1588                 break;
1589         default:
1590                 rc = -EINVAL;
1591         }
1592         if (rc)
1593                 return rc;
1594
1595         __floating_irq_kick(kvm, type);
1596         return 0;
1597 }
1598
1599 int kvm_s390_inject_vm(struct kvm *kvm,
1600                        struct kvm_s390_interrupt *s390int)
1601 {
1602         struct kvm_s390_interrupt_info *inti;
1603         int rc;
1604
1605         inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1606         if (!inti)
1607                 return -ENOMEM;
1608
1609         inti->type = s390int->type;
1610         switch (inti->type) {
1611         case KVM_S390_INT_VIRTIO:
1612                 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
1613                          s390int->parm, s390int->parm64);
1614                 inti->ext.ext_params = s390int->parm;
1615                 inti->ext.ext_params2 = s390int->parm64;
1616                 break;
1617         case KVM_S390_INT_SERVICE:
1618                 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
1619                 inti->ext.ext_params = s390int->parm;
1620                 break;
1621         case KVM_S390_INT_PFAULT_DONE:
1622                 inti->ext.ext_params2 = s390int->parm64;
1623                 break;
1624         case KVM_S390_MCHK:
1625                 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
1626                          s390int->parm64);
1627                 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
1628                 inti->mchk.mcic = s390int->parm64;
1629                 break;
1630         case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1631                 inti->io.subchannel_id = s390int->parm >> 16;
1632                 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
1633                 inti->io.io_int_parm = s390int->parm64 >> 32;
1634                 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
1635                 break;
1636         default:
1637                 kfree(inti);
1638                 return -EINVAL;
1639         }
1640         trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
1641                                  2);
1642
1643         rc = __inject_vm(kvm, inti);
1644         if (rc)
1645                 kfree(inti);
1646         return rc;
1647 }
1648
1649 int kvm_s390_reinject_io_int(struct kvm *kvm,
1650                               struct kvm_s390_interrupt_info *inti)
1651 {
1652         return __inject_vm(kvm, inti);
1653 }
1654
1655 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
1656                        struct kvm_s390_irq *irq)
1657 {
1658         irq->type = s390int->type;
1659         switch (irq->type) {
1660         case KVM_S390_PROGRAM_INT:
1661                 if (s390int->parm & 0xffff0000)
1662                         return -EINVAL;
1663                 irq->u.pgm.code = s390int->parm;
1664                 break;
1665         case KVM_S390_SIGP_SET_PREFIX:
1666                 irq->u.prefix.address = s390int->parm;
1667                 break;
1668         case KVM_S390_SIGP_STOP:
1669                 irq->u.stop.flags = s390int->parm;
1670                 break;
1671         case KVM_S390_INT_EXTERNAL_CALL:
1672                 if (s390int->parm & 0xffff0000)
1673                         return -EINVAL;
1674                 irq->u.extcall.code = s390int->parm;
1675                 break;
1676         case KVM_S390_INT_EMERGENCY:
1677                 if (s390int->parm & 0xffff0000)
1678                         return -EINVAL;
1679                 irq->u.emerg.code = s390int->parm;
1680                 break;
1681         case KVM_S390_MCHK:
1682                 irq->u.mchk.mcic = s390int->parm64;
1683                 break;
1684         }
1685         return 0;
1686 }
1687
1688 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
1689 {
1690         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1691
1692         return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1693 }
1694
1695 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
1696 {
1697         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1698
1699         spin_lock(&li->lock);
1700         li->irq.stop.flags = 0;
1701         clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1702         spin_unlock(&li->lock);
1703 }
1704
1705 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1706 {
1707         int rc;
1708
1709         switch (irq->type) {
1710         case KVM_S390_PROGRAM_INT:
1711                 rc = __inject_prog(vcpu, irq);
1712                 break;
1713         case KVM_S390_SIGP_SET_PREFIX:
1714                 rc = __inject_set_prefix(vcpu, irq);
1715                 break;
1716         case KVM_S390_SIGP_STOP:
1717                 rc = __inject_sigp_stop(vcpu, irq);
1718                 break;
1719         case KVM_S390_RESTART:
1720                 rc = __inject_sigp_restart(vcpu, irq);
1721                 break;
1722         case KVM_S390_INT_CLOCK_COMP:
1723                 rc = __inject_ckc(vcpu);
1724                 break;
1725         case KVM_S390_INT_CPU_TIMER:
1726                 rc = __inject_cpu_timer(vcpu);
1727                 break;
1728         case KVM_S390_INT_EXTERNAL_CALL:
1729                 rc = __inject_extcall(vcpu, irq);
1730                 break;
1731         case KVM_S390_INT_EMERGENCY:
1732                 rc = __inject_sigp_emergency(vcpu, irq);
1733                 break;
1734         case KVM_S390_MCHK:
1735                 rc = __inject_mchk(vcpu, irq);
1736                 break;
1737         case KVM_S390_INT_PFAULT_INIT:
1738                 rc = __inject_pfault_init(vcpu, irq);
1739                 break;
1740         case KVM_S390_INT_VIRTIO:
1741         case KVM_S390_INT_SERVICE:
1742         case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1743         default:
1744                 rc = -EINVAL;
1745         }
1746
1747         return rc;
1748 }
1749
1750 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1751 {
1752         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1753         int rc;
1754
1755         spin_lock(&li->lock);
1756         rc = do_inject_vcpu(vcpu, irq);
1757         spin_unlock(&li->lock);
1758         if (!rc)
1759                 kvm_s390_vcpu_wakeup(vcpu);
1760         return rc;
1761 }
1762
1763 static inline void clear_irq_list(struct list_head *_list)
1764 {
1765         struct kvm_s390_interrupt_info *inti, *n;
1766
1767         list_for_each_entry_safe(inti, n, _list, list) {
1768                 list_del(&inti->list);
1769                 kfree(inti);
1770         }
1771 }
1772
1773 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
1774                        struct kvm_s390_irq *irq)
1775 {
1776         irq->type = inti->type;
1777         switch (inti->type) {
1778         case KVM_S390_INT_PFAULT_INIT:
1779         case KVM_S390_INT_PFAULT_DONE:
1780         case KVM_S390_INT_VIRTIO:
1781                 irq->u.ext = inti->ext;
1782                 break;
1783         case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1784                 irq->u.io = inti->io;
1785                 break;
1786         }
1787 }
1788
1789 void kvm_s390_clear_float_irqs(struct kvm *kvm)
1790 {
1791         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1792         int i;
1793
1794         spin_lock(&fi->lock);
1795         fi->pending_irqs = 0;
1796         memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
1797         memset(&fi->mchk, 0, sizeof(fi->mchk));
1798         for (i = 0; i < FIRQ_LIST_COUNT; i++)
1799                 clear_irq_list(&fi->lists[i]);
1800         for (i = 0; i < FIRQ_MAX_COUNT; i++)
1801                 fi->counters[i] = 0;
1802         spin_unlock(&fi->lock);
1803 };
1804
1805 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
1806 {
1807         struct kvm_s390_interrupt_info *inti;
1808         struct kvm_s390_float_interrupt *fi;
1809         struct kvm_s390_irq *buf;
1810         struct kvm_s390_irq *irq;
1811         int max_irqs;
1812         int ret = 0;
1813         int n = 0;
1814         int i;
1815
1816         if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
1817                 return -EINVAL;
1818
1819         /*
1820          * We are already using -ENOMEM to signal
1821          * userspace it may retry with a bigger buffer,
1822          * so we need to use something else for this case
1823          */
1824         buf = vzalloc(len);
1825         if (!buf)
1826                 return -ENOBUFS;
1827
1828         max_irqs = len / sizeof(struct kvm_s390_irq);
1829
1830         fi = &kvm->arch.float_int;
1831         spin_lock(&fi->lock);
1832         for (i = 0; i < FIRQ_LIST_COUNT; i++) {
1833                 list_for_each_entry(inti, &fi->lists[i], list) {
1834                         if (n == max_irqs) {
1835                                 /* signal userspace to try again */
1836                                 ret = -ENOMEM;
1837                                 goto out;
1838                         }
1839                         inti_to_irq(inti, &buf[n]);
1840                         n++;
1841                 }
1842         }
1843         if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs)) {
1844                 if (n == max_irqs) {
1845                         /* signal userspace to try again */
1846                         ret = -ENOMEM;
1847                         goto out;
1848                 }
1849                 irq = (struct kvm_s390_irq *) &buf[n];
1850                 irq->type = KVM_S390_INT_SERVICE;
1851                 irq->u.ext = fi->srv_signal;
1852                 n++;
1853         }
1854         if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
1855                 if (n == max_irqs) {
1856                                 /* signal userspace to try again */
1857                                 ret = -ENOMEM;
1858                                 goto out;
1859                 }
1860                 irq = (struct kvm_s390_irq *) &buf[n];
1861                 irq->type = KVM_S390_MCHK;
1862                 irq->u.mchk = fi->mchk;
1863                 n++;
1864 }
1865
1866 out:
1867         spin_unlock(&fi->lock);
1868         if (!ret && n > 0) {
1869                 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
1870                         ret = -EFAULT;
1871         }
1872         vfree(buf);
1873
1874         return ret < 0 ? ret : n;
1875 }
1876
1877 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1878 {
1879         int r;
1880
1881         switch (attr->group) {
1882         case KVM_DEV_FLIC_GET_ALL_IRQS:
1883                 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
1884                                           attr->attr);
1885                 break;
1886         default:
1887                 r = -EINVAL;
1888         }
1889
1890         return r;
1891 }
1892
1893 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
1894                                      u64 addr)
1895 {
1896         struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
1897         void *target = NULL;
1898         void __user *source;
1899         u64 size;
1900
1901         if (get_user(inti->type, (u64 __user *)addr))
1902                 return -EFAULT;
1903
1904         switch (inti->type) {
1905         case KVM_S390_INT_PFAULT_INIT:
1906         case KVM_S390_INT_PFAULT_DONE:
1907         case KVM_S390_INT_VIRTIO:
1908         case KVM_S390_INT_SERVICE:
1909                 target = (void *) &inti->ext;
1910                 source = &uptr->u.ext;
1911                 size = sizeof(inti->ext);
1912                 break;
1913         case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1914                 target = (void *) &inti->io;
1915                 source = &uptr->u.io;
1916                 size = sizeof(inti->io);
1917                 break;
1918         case KVM_S390_MCHK:
1919                 target = (void *) &inti->mchk;
1920                 source = &uptr->u.mchk;
1921                 size = sizeof(inti->mchk);
1922                 break;
1923         default:
1924                 return -EINVAL;
1925         }
1926
1927         if (copy_from_user(target, source, size))
1928                 return -EFAULT;
1929
1930         return 0;
1931 }
1932
1933 static int enqueue_floating_irq(struct kvm_device *dev,
1934                                 struct kvm_device_attr *attr)
1935 {
1936         struct kvm_s390_interrupt_info *inti = NULL;
1937         int r = 0;
1938         int len = attr->attr;
1939
1940         if (len % sizeof(struct kvm_s390_irq) != 0)
1941                 return -EINVAL;
1942         else if (len > KVM_S390_FLIC_MAX_BUFFER)
1943                 return -EINVAL;
1944
1945         while (len >= sizeof(struct kvm_s390_irq)) {
1946                 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1947                 if (!inti)
1948                         return -ENOMEM;
1949
1950                 r = copy_irq_from_user(inti, attr->addr);
1951                 if (r) {
1952                         kfree(inti);
1953                         return r;
1954                 }
1955                 r = __inject_vm(dev->kvm, inti);
1956                 if (r) {
1957                         kfree(inti);
1958                         return r;
1959                 }
1960                 len -= sizeof(struct kvm_s390_irq);
1961                 attr->addr += sizeof(struct kvm_s390_irq);
1962         }
1963
1964         return r;
1965 }
1966
1967 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
1968 {
1969         if (id >= MAX_S390_IO_ADAPTERS)
1970                 return NULL;
1971         return kvm->arch.adapters[id];
1972 }
1973
1974 static int register_io_adapter(struct kvm_device *dev,
1975                                struct kvm_device_attr *attr)
1976 {
1977         struct s390_io_adapter *adapter;
1978         struct kvm_s390_io_adapter adapter_info;
1979
1980         if (copy_from_user(&adapter_info,
1981                            (void __user *)attr->addr, sizeof(adapter_info)))
1982                 return -EFAULT;
1983
1984         if ((adapter_info.id >= MAX_S390_IO_ADAPTERS) ||
1985             (dev->kvm->arch.adapters[adapter_info.id] != NULL))
1986                 return -EINVAL;
1987
1988         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
1989         if (!adapter)
1990                 return -ENOMEM;
1991
1992         INIT_LIST_HEAD(&adapter->maps);
1993         init_rwsem(&adapter->maps_lock);
1994         atomic_set(&adapter->nr_maps, 0);
1995         adapter->id = adapter_info.id;
1996         adapter->isc = adapter_info.isc;
1997         adapter->maskable = adapter_info.maskable;
1998         adapter->masked = false;
1999         adapter->swap = adapter_info.swap;
2000         adapter->suppressible = (adapter_info.flags) &
2001                                 KVM_S390_ADAPTER_SUPPRESSIBLE;
2002         dev->kvm->arch.adapters[adapter->id] = adapter;
2003
2004         return 0;
2005 }
2006
2007 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2008 {
2009         int ret;
2010         struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2011
2012         if (!adapter || !adapter->maskable)
2013                 return -EINVAL;
2014         ret = adapter->masked;
2015         adapter->masked = masked;
2016         return ret;
2017 }
2018
2019 static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
2020 {
2021         struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2022         struct s390_map_info *map;
2023         int ret;
2024
2025         if (!adapter || !addr)
2026                 return -EINVAL;
2027
2028         map = kzalloc(sizeof(*map), GFP_KERNEL);
2029         if (!map) {
2030                 ret = -ENOMEM;
2031                 goto out;
2032         }
2033         INIT_LIST_HEAD(&map->list);
2034         map->guest_addr = addr;
2035         map->addr = gmap_translate(kvm->arch.gmap, addr);
2036         if (map->addr == -EFAULT) {
2037                 ret = -EFAULT;
2038                 goto out;
2039         }
2040         ret = get_user_pages_fast(map->addr, 1, 1, &map->page);
2041         if (ret < 0)
2042                 goto out;
2043         BUG_ON(ret != 1);
2044         down_write(&adapter->maps_lock);
2045         if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
2046                 list_add_tail(&map->list, &adapter->maps);
2047                 ret = 0;
2048         } else {
2049                 put_page(map->page);
2050                 ret = -EINVAL;
2051         }
2052         up_write(&adapter->maps_lock);
2053 out:
2054         if (ret)
2055                 kfree(map);
2056         return ret;
2057 }
2058
2059 static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
2060 {
2061         struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2062         struct s390_map_info *map, *tmp;
2063         int found = 0;
2064
2065         if (!adapter || !addr)
2066                 return -EINVAL;
2067
2068         down_write(&adapter->maps_lock);
2069         list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
2070                 if (map->guest_addr == addr) {
2071                         found = 1;
2072                         atomic_dec(&adapter->nr_maps);
2073                         list_del(&map->list);
2074                         put_page(map->page);
2075                         kfree(map);
2076                         break;
2077                 }
2078         }
2079         up_write(&adapter->maps_lock);
2080
2081         return found ? 0 : -EINVAL;
2082 }
2083
2084 void kvm_s390_destroy_adapters(struct kvm *kvm)
2085 {
2086         int i;
2087         struct s390_map_info *map, *tmp;
2088
2089         for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
2090                 if (!kvm->arch.adapters[i])
2091                         continue;
2092                 list_for_each_entry_safe(map, tmp,
2093                                          &kvm->arch.adapters[i]->maps, list) {
2094                         list_del(&map->list);
2095                         put_page(map->page);
2096                         kfree(map);
2097                 }
2098                 kfree(kvm->arch.adapters[i]);
2099         }
2100 }
2101
2102 static int modify_io_adapter(struct kvm_device *dev,
2103                              struct kvm_device_attr *attr)
2104 {
2105         struct kvm_s390_io_adapter_req req;
2106         struct s390_io_adapter *adapter;
2107         int ret;
2108
2109         if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2110                 return -EFAULT;
2111
2112         adapter = get_io_adapter(dev->kvm, req.id);
2113         if (!adapter)
2114                 return -EINVAL;
2115         switch (req.type) {
2116         case KVM_S390_IO_ADAPTER_MASK:
2117                 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2118                 if (ret > 0)
2119                         ret = 0;
2120                 break;
2121         case KVM_S390_IO_ADAPTER_MAP:
2122                 ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
2123                 break;
2124         case KVM_S390_IO_ADAPTER_UNMAP:
2125                 ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
2126                 break;
2127         default:
2128                 ret = -EINVAL;
2129         }
2130
2131         return ret;
2132 }
2133
2134 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2135
2136 {
2137         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2138         u32 schid;
2139
2140         if (attr->flags)
2141                 return -EINVAL;
2142         if (attr->attr != sizeof(schid))
2143                 return -EINVAL;
2144         if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2145                 return -EFAULT;
2146         kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2147         /*
2148          * If userspace is conforming to the architecture, we can have at most
2149          * one pending I/O interrupt per subchannel, so this is effectively a
2150          * clear all.
2151          */
2152         return 0;
2153 }
2154
2155 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2156 {
2157         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2158         struct kvm_s390_ais_req req;
2159         int ret = 0;
2160
2161         if (!fi->ais_enabled)
2162                 return -ENOTSUPP;
2163
2164         if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2165                 return -EFAULT;
2166
2167         if (req.isc > MAX_ISC)
2168                 return -EINVAL;
2169
2170         trace_kvm_s390_modify_ais_mode(req.isc,
2171                                        (fi->simm & AIS_MODE_MASK(req.isc)) ?
2172                                        (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2173                                        2 : KVM_S390_AIS_MODE_SINGLE :
2174                                        KVM_S390_AIS_MODE_ALL, req.mode);
2175
2176         mutex_lock(&fi->ais_lock);
2177         switch (req.mode) {
2178         case KVM_S390_AIS_MODE_ALL:
2179                 fi->simm &= ~AIS_MODE_MASK(req.isc);
2180                 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2181                 break;
2182         case KVM_S390_AIS_MODE_SINGLE:
2183                 fi->simm |= AIS_MODE_MASK(req.isc);
2184                 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2185                 break;
2186         default:
2187                 ret = -EINVAL;
2188         }
2189         mutex_unlock(&fi->ais_lock);
2190
2191         return ret;
2192 }
2193
2194 static int kvm_s390_inject_airq(struct kvm *kvm,
2195                                 struct s390_io_adapter *adapter)
2196 {
2197         struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2198         struct kvm_s390_interrupt s390int = {
2199                 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2200                 .parm = 0,
2201                 .parm64 = (adapter->isc << 27) | 0x80000000,
2202         };
2203         int ret = 0;
2204
2205         if (!fi->ais_enabled || !adapter->suppressible)
2206                 return kvm_s390_inject_vm(kvm, &s390int);
2207
2208         mutex_lock(&fi->ais_lock);
2209         if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2210                 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2211                 goto out;
2212         }
2213
2214         ret = kvm_s390_inject_vm(kvm, &s390int);
2215         if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2216                 fi->nimm |= AIS_MODE_MASK(adapter->isc);
2217                 trace_kvm_s390_modify_ais_mode(adapter->isc,
2218                                                KVM_S390_AIS_MODE_SINGLE, 2);
2219         }
2220 out:
2221         mutex_unlock(&fi->ais_lock);
2222         return ret;
2223 }
2224
2225 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2226 {
2227         unsigned int id = attr->attr;
2228         struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2229
2230         if (!adapter)
2231                 return -EINVAL;
2232
2233         return kvm_s390_inject_airq(kvm, adapter);
2234 }
2235
2236 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2237 {
2238         int r = 0;
2239         unsigned int i;
2240         struct kvm_vcpu *vcpu;
2241
2242         switch (attr->group) {
2243         case KVM_DEV_FLIC_ENQUEUE:
2244                 r = enqueue_floating_irq(dev, attr);
2245                 break;
2246         case KVM_DEV_FLIC_CLEAR_IRQS:
2247                 kvm_s390_clear_float_irqs(dev->kvm);
2248                 break;
2249         case KVM_DEV_FLIC_APF_ENABLE:
2250                 dev->kvm->arch.gmap->pfault_enabled = 1;
2251                 break;
2252         case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2253                 dev->kvm->arch.gmap->pfault_enabled = 0;
2254                 /*
2255                  * Make sure no async faults are in transition when
2256                  * clearing the queues. So we don't need to worry
2257                  * about late coming workers.
2258                  */
2259                 synchronize_srcu(&dev->kvm->srcu);
2260                 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2261                         kvm_clear_async_pf_completion_queue(vcpu);
2262                 break;
2263         case KVM_DEV_FLIC_ADAPTER_REGISTER:
2264                 r = register_io_adapter(dev, attr);
2265                 break;
2266         case KVM_DEV_FLIC_ADAPTER_MODIFY:
2267                 r = modify_io_adapter(dev, attr);
2268                 break;
2269         case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2270                 r = clear_io_irq(dev->kvm, attr);
2271                 break;
2272         case KVM_DEV_FLIC_AISM:
2273                 r = modify_ais_mode(dev->kvm, attr);
2274                 break;
2275         case KVM_DEV_FLIC_AIRQ_INJECT:
2276                 r = flic_inject_airq(dev->kvm, attr);
2277                 break;
2278         default:
2279                 r = -EINVAL;
2280         }
2281
2282         return r;
2283 }
2284
2285 static int flic_has_attr(struct kvm_device *dev,
2286                              struct kvm_device_attr *attr)
2287 {
2288         switch (attr->group) {
2289         case KVM_DEV_FLIC_GET_ALL_IRQS:
2290         case KVM_DEV_FLIC_ENQUEUE:
2291         case KVM_DEV_FLIC_CLEAR_IRQS:
2292         case KVM_DEV_FLIC_APF_ENABLE:
2293         case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2294         case KVM_DEV_FLIC_ADAPTER_REGISTER:
2295         case KVM_DEV_FLIC_ADAPTER_MODIFY:
2296         case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2297         case KVM_DEV_FLIC_AISM:
2298         case KVM_DEV_FLIC_AIRQ_INJECT:
2299                 return 0;
2300         }
2301         return -ENXIO;
2302 }
2303
2304 static int flic_create(struct kvm_device *dev, u32 type)
2305 {
2306         if (!dev)
2307                 return -EINVAL;
2308         if (dev->kvm->arch.flic)
2309                 return -EINVAL;
2310         dev->kvm->arch.flic = dev;
2311         return 0;
2312 }
2313
2314 static void flic_destroy(struct kvm_device *dev)
2315 {
2316         dev->kvm->arch.flic = NULL;
2317         kfree(dev);
2318 }
2319
2320 /* s390 floating irq controller (flic) */
2321 struct kvm_device_ops kvm_flic_ops = {
2322         .name = "kvm-flic",
2323         .get_attr = flic_get_attr,
2324         .set_attr = flic_set_attr,
2325         .has_attr = flic_has_attr,
2326         .create = flic_create,
2327         .destroy = flic_destroy,
2328 };
2329
2330 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2331 {
2332         unsigned long bit;
2333
2334         bit = bit_nr + (addr % PAGE_SIZE) * 8;
2335
2336         return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2337 }
2338
2339 static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
2340                                           u64 addr)
2341 {
2342         struct s390_map_info *map;
2343
2344         if (!adapter)
2345                 return NULL;
2346
2347         list_for_each_entry(map, &adapter->maps, list) {
2348                 if (map->guest_addr == addr)
2349                         return map;
2350         }
2351         return NULL;
2352 }
2353
2354 static int adapter_indicators_set(struct kvm *kvm,
2355                                   struct s390_io_adapter *adapter,
2356                                   struct kvm_s390_adapter_int *adapter_int)
2357 {
2358         unsigned long bit;
2359         int summary_set, idx;
2360         struct s390_map_info *info;
2361         void *map;
2362
2363         info = get_map_info(adapter, adapter_int->ind_addr);
2364         if (!info)
2365                 return -1;
2366         map = page_address(info->page);
2367         bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
2368         set_bit(bit, map);
2369         idx = srcu_read_lock(&kvm->srcu);
2370         mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2371         set_page_dirty_lock(info->page);
2372         info = get_map_info(adapter, adapter_int->summary_addr);
2373         if (!info) {
2374                 srcu_read_unlock(&kvm->srcu, idx);
2375                 return -1;
2376         }
2377         map = page_address(info->page);
2378         bit = get_ind_bit(info->addr, adapter_int->summary_offset,
2379                           adapter->swap);
2380         summary_set = test_and_set_bit(bit, map);
2381         mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2382         set_page_dirty_lock(info->page);
2383         srcu_read_unlock(&kvm->srcu, idx);
2384         return summary_set ? 0 : 1;
2385 }
2386
2387 /*
2388  * < 0 - not injected due to error
2389  * = 0 - coalesced, summary indicator already active
2390  * > 0 - injected interrupt
2391  */
2392 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2393                            struct kvm *kvm, int irq_source_id, int level,
2394                            bool line_status)
2395 {
2396         int ret;
2397         struct s390_io_adapter *adapter;
2398
2399         /* We're only interested in the 0->1 transition. */
2400         if (!level)
2401                 return 0;
2402         adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2403         if (!adapter)
2404                 return -1;
2405         down_read(&adapter->maps_lock);
2406         ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2407         up_read(&adapter->maps_lock);
2408         if ((ret > 0) && !adapter->masked) {
2409                 ret = kvm_s390_inject_airq(kvm, adapter);
2410                 if (ret == 0)
2411                         ret = 1;
2412         }
2413         return ret;
2414 }
2415
2416 int kvm_set_routing_entry(struct kvm *kvm,
2417                           struct kvm_kernel_irq_routing_entry *e,
2418                           const struct kvm_irq_routing_entry *ue)
2419 {
2420         int ret;
2421
2422         switch (ue->type) {
2423         case KVM_IRQ_ROUTING_S390_ADAPTER:
2424                 e->set = set_adapter_int;
2425                 e->adapter.summary_addr = ue->u.adapter.summary_addr;
2426                 e->adapter.ind_addr = ue->u.adapter.ind_addr;
2427                 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2428                 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2429                 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2430                 ret = 0;
2431                 break;
2432         default:
2433                 ret = -EINVAL;
2434         }
2435
2436         return ret;
2437 }
2438
2439 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2440                 int irq_source_id, int level, bool line_status)
2441 {
2442         return -EINVAL;
2443 }
2444
2445 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2446 {
2447         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2448         struct kvm_s390_irq *buf;
2449         int r = 0;
2450         int n;
2451
2452         buf = vmalloc(len);
2453         if (!buf)
2454                 return -ENOMEM;
2455
2456         if (copy_from_user((void *) buf, irqstate, len)) {
2457                 r = -EFAULT;
2458                 goto out_free;
2459         }
2460
2461         /*
2462          * Don't allow setting the interrupt state
2463          * when there are already interrupts pending
2464          */
2465         spin_lock(&li->lock);
2466         if (li->pending_irqs) {
2467                 r = -EBUSY;
2468                 goto out_unlock;
2469         }
2470
2471         for (n = 0; n < len / sizeof(*buf); n++) {
2472                 r = do_inject_vcpu(vcpu, &buf[n]);
2473                 if (r)
2474                         break;
2475         }
2476
2477 out_unlock:
2478         spin_unlock(&li->lock);
2479 out_free:
2480         vfree(buf);
2481
2482         return r;
2483 }
2484
2485 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2486                             struct kvm_s390_irq *irq,
2487                             unsigned long irq_type)
2488 {
2489         switch (irq_type) {
2490         case IRQ_PEND_MCHK_EX:
2491         case IRQ_PEND_MCHK_REP:
2492                 irq->type = KVM_S390_MCHK;
2493                 irq->u.mchk = li->irq.mchk;
2494                 break;
2495         case IRQ_PEND_PROG:
2496                 irq->type = KVM_S390_PROGRAM_INT;
2497                 irq->u.pgm = li->irq.pgm;
2498                 break;
2499         case IRQ_PEND_PFAULT_INIT:
2500                 irq->type = KVM_S390_INT_PFAULT_INIT;
2501                 irq->u.ext = li->irq.ext;
2502                 break;
2503         case IRQ_PEND_EXT_EXTERNAL:
2504                 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2505                 irq->u.extcall = li->irq.extcall;
2506                 break;
2507         case IRQ_PEND_EXT_CLOCK_COMP:
2508                 irq->type = KVM_S390_INT_CLOCK_COMP;
2509                 break;
2510         case IRQ_PEND_EXT_CPU_TIMER:
2511                 irq->type = KVM_S390_INT_CPU_TIMER;
2512                 break;
2513         case IRQ_PEND_SIGP_STOP:
2514                 irq->type = KVM_S390_SIGP_STOP;
2515                 irq->u.stop = li->irq.stop;
2516                 break;
2517         case IRQ_PEND_RESTART:
2518                 irq->type = KVM_S390_RESTART;
2519                 break;
2520         case IRQ_PEND_SET_PREFIX:
2521                 irq->type = KVM_S390_SIGP_SET_PREFIX;
2522                 irq->u.prefix = li->irq.prefix;
2523                 break;
2524         }
2525 }
2526
2527 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2528 {
2529         int scn;
2530         unsigned long sigp_emerg_pending[BITS_TO_LONGS(KVM_MAX_VCPUS)];
2531         struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2532         unsigned long pending_irqs;
2533         struct kvm_s390_irq irq;
2534         unsigned long irq_type;
2535         int cpuaddr;
2536         int n = 0;
2537
2538         spin_lock(&li->lock);
2539         pending_irqs = li->pending_irqs;
2540         memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
2541                sizeof(sigp_emerg_pending));
2542         spin_unlock(&li->lock);
2543
2544         for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
2545                 memset(&irq, 0, sizeof(irq));
2546                 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
2547                         continue;
2548                 if (n + sizeof(irq) > len)
2549                         return -ENOBUFS;
2550                 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
2551                 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2552                         return -EFAULT;
2553                 n += sizeof(irq);
2554         }
2555
2556         if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
2557                 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
2558                         memset(&irq, 0, sizeof(irq));
2559                         if (n + sizeof(irq) > len)
2560                                 return -ENOBUFS;
2561                         irq.type = KVM_S390_INT_EMERGENCY;
2562                         irq.u.emerg.code = cpuaddr;
2563                         if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2564                                 return -EFAULT;
2565                         n += sizeof(irq);
2566                 }
2567         }
2568
2569         if (sca_ext_call_pending(vcpu, &scn)) {
2570                 if (n + sizeof(irq) > len)
2571                         return -ENOBUFS;
2572                 memset(&irq, 0, sizeof(irq));
2573                 irq.type = KVM_S390_INT_EXTERNAL_CALL;
2574                 irq.u.extcall.code = scn;
2575                 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2576                         return -EFAULT;
2577                 n += sizeof(irq);
2578         }
2579
2580         return n;
2581 }