2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
11 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
13 * Derived from book3s_rmhandlers.S and other files, which are:
15 * Copyright SUSE Linux Products GmbH 2009
17 * Authors: Alexander Graf <agraf@suse.de>
20 #include <asm/ppc_asm.h>
21 #include <asm/code-patching-asm.h>
22 #include <asm/kvm_asm.h>
26 #include <asm/ptrace.h>
27 #include <asm/hvcall.h>
28 #include <asm/asm-offsets.h>
29 #include <asm/exception-64s.h>
30 #include <asm/kvm_book3s_asm.h>
31 #include <asm/book3s/64/mmu-hash.h>
34 #include <asm/xive-regs.h>
36 /* Sign-extend HDEC if not on POWER9 */
37 #define EXTEND_HDEC(reg) \
40 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
42 #define VCPU_GPRS_TM(reg) (((reg) * ULONG_SIZE) + VCPU_GPR_TM)
44 /* Values in HSTATE_NAPPING(r13) */
45 #define NAPPING_CEDE 1
46 #define NAPPING_NOVCPU 2
48 /* Stack frame offsets for kvmppc_hv_entry */
50 #define STACK_SLOT_TRAP (SFS-4)
51 #define STACK_SLOT_TID (SFS-16)
52 #define STACK_SLOT_PSSCR (SFS-24)
53 #define STACK_SLOT_PID (SFS-32)
54 #define STACK_SLOT_IAMR (SFS-40)
55 #define STACK_SLOT_CIABR (SFS-48)
56 #define STACK_SLOT_DAWR (SFS-56)
57 #define STACK_SLOT_DAWRX (SFS-64)
58 #define STACK_SLOT_HFSCR (SFS-72)
61 * Call kvmppc_hv_entry in real mode.
62 * Must be called with interrupts hard-disabled.
66 * LR = return address to continue at after eventually re-enabling MMU
68 _GLOBAL_TOC(kvmppc_hv_entry_trampoline)
70 std r0, PPC_LR_STKOFF(r1)
73 std r10, HSTATE_HOST_MSR(r13)
74 LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
79 mtmsrd r0,1 /* clear RI in MSR */
85 ld r4, HSTATE_KVM_VCPU(r13)
88 /* Back from guest - restore host state and return to caller */
91 /* Restore host DABR and DABRX */
92 ld r5,HSTATE_DABR(r13)
96 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
99 ld r3,PACA_SPRG_VDSO(r13)
100 mtspr SPRN_SPRG_VDSO_WRITE,r3
102 /* Reload the host's PMU registers */
103 ld r3, PACALPPACAPTR(r13) /* is the host using the PMU? */
104 lbz r4, LPPACA_PMCINUSE(r3)
106 beq 23f /* skip if not */
108 ld r3, HSTATE_MMCR0(r13)
109 andi. r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
112 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
113 lwz r3, HSTATE_PMC1(r13)
114 lwz r4, HSTATE_PMC2(r13)
115 lwz r5, HSTATE_PMC3(r13)
116 lwz r6, HSTATE_PMC4(r13)
117 lwz r8, HSTATE_PMC5(r13)
118 lwz r9, HSTATE_PMC6(r13)
125 ld r3, HSTATE_MMCR0(r13)
126 ld r4, HSTATE_MMCR1(r13)
127 ld r5, HSTATE_MMCRA(r13)
128 ld r6, HSTATE_SIAR(r13)
129 ld r7, HSTATE_SDAR(r13)
135 ld r8, HSTATE_MMCR2(r13)
136 ld r9, HSTATE_SIER(r13)
139 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
145 * Reload DEC. HDEC interrupts were disabled when
146 * we reloaded the host's LPCR value.
148 ld r3, HSTATE_DECEXP(r13)
154 /* hwthread_req may have got set by cede or no vcpu, so clear it */
156 stb r0, HSTATE_HWTHREAD_REQ(r13)
157 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
160 * For external interrupts we need to call the Linux
161 * handler to process the interrupt. We do that by jumping
162 * to absolute address 0x500 for external interrupts.
163 * The [h]rfid at the end of the handler will return to
164 * the book3s_hv_interrupts.S code. For other interrupts
165 * we do the rfid to get back to the book3s_hv_interrupts.S
168 ld r8, 112+PPC_LR_STKOFF(r1)
170 ld r7, HSTATE_HOST_MSR(r13)
172 /* Return the trap number on this thread as the return value */
176 * If we came back from the guest via a relocation-on interrupt,
177 * we will be in virtual mode at this point, which makes it a
178 * little easier to get back to the caller.
181 andi. r0, r0, MSR_IR /* in real mode? */
184 /* RFI into the highmem handler */
188 mtmsrd r6, 1 /* Clear RI in MSR */
193 /* Virtual-mode return */
198 kvmppc_primary_no_guest:
199 /* We handle this much like a ceded vcpu */
200 /* put the HDEC into the DEC, since HDEC interrupts don't wake us */
201 /* HDEC may be larger than DEC for arch >= v3.00, but since the */
202 /* HDEC value came from DEC in the first place, it will fit */
206 * Make sure the primary has finished the MMU switch.
207 * We should never get here on a secondary thread, but
208 * check it for robustness' sake.
210 ld r5, HSTATE_KVM_VCORE(r13)
211 65: lbz r0, VCORE_IN_GUEST(r5)
218 /* set our bit in napping_threads */
219 ld r5, HSTATE_KVM_VCORE(r13)
220 lbz r7, HSTATE_PTID(r13)
223 addi r6, r5, VCORE_NAPPING_THREADS
228 /* order napping_threads update vs testing entry_exit_map */
231 lwz r7, VCORE_ENTRY_EXIT(r5)
233 bge kvm_novcpu_exit /* another thread already exiting */
234 li r3, NAPPING_NOVCPU
235 stb r3, HSTATE_NAPPING(r13)
237 li r3, 0 /* Don't wake on privileged (OS) doorbell */
242 * Entered from kvm_start_guest if kvm_hstate.napping is set
248 ld r1, HSTATE_HOST_R1(r13)
249 ld r5, HSTATE_KVM_VCORE(r13)
251 stb r0, HSTATE_NAPPING(r13)
253 /* check the wake reason */
254 bl kvmppc_check_wake_reason
257 * Restore volatile registers since we could have called
258 * a C routine in kvmppc_check_wake_reason.
261 ld r5, HSTATE_KVM_VCORE(r13)
263 /* see if any other thread is already exiting */
264 lwz r0, VCORE_ENTRY_EXIT(r5)
268 /* clear our bit in napping_threads */
269 lbz r7, HSTATE_PTID(r13)
272 addi r6, r5, VCORE_NAPPING_THREADS
278 /* See if the wake reason means we need to exit */
282 /* See if our timeslice has expired (HDEC is negative) */
285 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
289 /* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
290 ld r4, HSTATE_KVM_VCPU(r13)
292 beq kvmppc_primary_no_guest
294 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
295 addi r3, r4, VCPU_TB_RMENTRY
296 bl kvmhv_start_timing
301 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
302 ld r4, HSTATE_KVM_VCPU(r13)
305 addi r3, r4, VCPU_TB_RMEXIT
306 bl kvmhv_accumulate_time
309 stw r12, STACK_SLOT_TRAP(r1)
310 bl kvmhv_commence_exit
312 b kvmhv_switch_to_host
315 * We come in here when wakened from nap mode.
316 * Relocation is off and most register values are lost.
317 * r13 points to the PACA.
318 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
319 * This is not used by ISAv3.0B processors.
321 .globl kvm_start_guest
323 /* Set runlatch bit the minute you wake up from nap */
329 * Could avoid this and pass it through in r3. For now,
330 * code expects it to be in SRR1.
336 li r0,KVM_HWTHREAD_IN_KVM
337 stb r0,HSTATE_HWTHREAD_STATE(r13)
339 /* NV GPR values from power7_idle() will no longer be valid */
341 stb r0,PACA_NAPSTATELOST(r13)
343 /* were we napping due to cede? */
344 lbz r0,HSTATE_NAPPING(r13)
345 cmpwi r0,NAPPING_CEDE
347 cmpwi r0,NAPPING_NOVCPU
348 beq kvm_novcpu_wakeup
350 ld r1,PACAEMERGSP(r13)
351 subi r1,r1,STACK_FRAME_OVERHEAD
354 * We weren't napping due to cede, so this must be a secondary
355 * thread being woken up to run a guest, or being woken up due
356 * to a stray IPI. (Or due to some machine check or hypervisor
357 * maintenance interrupt while the core is in KVM.)
360 /* Check the wake reason in SRR1 to see why we got here */
361 bl kvmppc_check_wake_reason
363 * kvmppc_check_wake_reason could invoke a C routine, but we
364 * have no volatile registers to restore when we return.
370 /* get vcore pointer, NULL if we have nothing to run */
371 ld r5,HSTATE_KVM_VCORE(r13)
373 /* if we have no vcore to run, go back to sleep */
376 kvm_secondary_got_guest:
378 /* Set HSTATE_DSCR(r13) to something sensible */
379 ld r6, PACA_DSCR_DEFAULT(r13)
380 std r6, HSTATE_DSCR(r13)
382 /* On thread 0 of a subcore, set HDEC to max */
383 lbz r4, HSTATE_PTID(r13)
386 LOAD_REG_ADDR(r6, decrementer_max)
389 /* and set per-LPAR registers, if doing dynamic micro-threading */
390 ld r6, HSTATE_SPLIT_MODE(r13)
393 ld r0, KVM_SPLIT_RPR(r6)
395 ld r0, KVM_SPLIT_PMMAR(r6)
397 ld r0, KVM_SPLIT_LDBAR(r6)
401 /* Order load of vcpu after load of vcore */
403 ld r4, HSTATE_KVM_VCPU(r13)
406 /* Back from the guest, go back to nap */
407 /* Clear our vcpu and vcore pointers so we don't come back in early */
409 std r0, HSTATE_KVM_VCPU(r13)
411 * Once we clear HSTATE_KVM_VCORE(r13), the code in
412 * kvmppc_run_core() is going to assume that all our vcpu
413 * state is visible in memory. This lwsync makes sure
417 std r0, HSTATE_KVM_VCORE(r13)
420 * All secondaries exiting guest will fall through this path.
421 * Before proceeding, just check for HMI interrupt and
422 * invoke opal hmi handler. By now we are sure that the
423 * primary thread on this core/subcore has already made partition
424 * switch/TB resync and we are good to call opal hmi handler.
426 cmpwi r12, BOOK3S_INTERRUPT_HMI
429 li r3,0 /* NULL argument */
430 bl hmi_exception_realmode
432 * At this point we have finished executing in the guest.
433 * We need to wait for hwthread_req to become zero, since
434 * we may not turn on the MMU while hwthread_req is non-zero.
435 * While waiting we also need to check if we get given a vcpu to run.
440 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
441 lbz r3, HSTATE_HWTHREAD_REQ(r13)
445 li r0, KVM_HWTHREAD_IN_KERNEL
446 stb r0, HSTATE_HWTHREAD_STATE(r13)
447 /* need to recheck hwthread_req after a barrier, to avoid race */
449 lbz r3, HSTATE_HWTHREAD_REQ(r13)
453 * We jump to pnv_wakeup_loss, which will return to the caller
454 * of power7_nap in the powernv cpu offline loop. The value we
455 * put in r3 becomes the return value for power7_nap. pnv_wakeup_loss
456 * requires SRR1 in r12.
460 rlwimi r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
467 ld r5, HSTATE_KVM_VCORE(r13)
470 ld r3, HSTATE_SPLIT_MODE(r13)
473 lbz r0, KVM_SPLIT_DO_NAP(r3)
479 b kvm_secondary_got_guest
481 54: li r0, KVM_HWTHREAD_IN_KVM
482 stb r0, HSTATE_HWTHREAD_STATE(r13)
486 * Here the primary thread is trying to return the core to
487 * whole-core mode, so we need to nap.
491 * When secondaries are napping in kvm_unsplit_nap() with
492 * hwthread_req = 1, HMI goes ignored even though subcores are
493 * already exited the guest. Hence HMI keeps waking up secondaries
494 * from nap in a loop and secondaries always go back to nap since
495 * no vcore is assigned to them. This makes impossible for primary
496 * thread to get hold of secondary threads resulting into a soft
497 * lockup in KVM path.
499 * Let us check if HMI is pending and handle it before we go to nap.
501 cmpwi r12, BOOK3S_INTERRUPT_HMI
503 li r3, 0 /* NULL argument */
504 bl hmi_exception_realmode
507 * Ensure that secondary doesn't nap when it has
508 * its vcore pointer set.
510 sync /* matches smp_mb() before setting split_info.do_nap */
511 ld r0, HSTATE_KVM_VCORE(r13)
514 /* clear any pending message */
516 lis r6, (PPC_DBELL_SERVER << (63-36))@h
518 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
519 /* Set kvm_split_mode.napped[tid] = 1 */
520 ld r3, HSTATE_SPLIT_MODE(r13)
522 lhz r4, PACAPACAINDEX(r13)
523 clrldi r4, r4, 61 /* micro-threading => P8 => 8 threads/core */
524 addi r4, r4, KVM_SPLIT_NAPPED
526 /* Check the do_nap flag again after setting napped[] */
528 lbz r0, KVM_SPLIT_DO_NAP(r3)
531 li r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
533 rlwimi r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
540 /******************************************************************************
544 *****************************************************************************/
546 .global kvmppc_hv_entry
551 * R4 = vcpu pointer (or NULL)
556 * all other volatile GPRS = free
557 * Does not preserve non-volatile GPRs or CR fields
560 std r0, PPC_LR_STKOFF(r1)
563 /* Save R1 in the PACA */
564 std r1, HSTATE_HOST_R1(r13)
566 li r6, KVM_GUEST_MODE_HOST_HV
567 stb r6, HSTATE_IN_GUEST(r13)
569 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
570 /* Store initial timestamp */
573 addi r3, r4, VCPU_TB_RMENTRY
574 bl kvmhv_start_timing
578 /* Use cr7 as an indication of radix mode */
579 ld r5, HSTATE_KVM_VCORE(r13)
580 ld r9, VCORE_KVM(r5) /* pointer to struct kvm */
581 lbz r0, KVM_RADIX(r9)
584 /* Clear out SLB if hash */
592 * POWER7/POWER8 host -> guest partition switch code.
593 * We don't have to lock against concurrent tlbies,
594 * but we do have to coordinate across hardware threads.
596 /* Set bit in entry map iff exit map is zero. */
598 lbz r6, HSTATE_PTID(r13)
600 addi r8, r5, VCORE_ENTRY_EXIT
602 cmpwi r3, 0x100 /* any threads starting to exit? */
603 bge secondary_too_late /* if so we're too late to the party */
608 /* Primary thread switches to guest partition. */
614 li r0,LPID_RSVD /* switch to reserved LPID */
617 mtspr SPRN_SDR1,r6 /* switch to partition page table */
618 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
622 /* See if we need to flush the TLB */
623 lhz r6,PACAPACAINDEX(r13) /* test_bit(cpu, need_tlb_flush) */
626 * On POWER9, individual threads can come in here, but the
627 * TLB is shared between the 4 threads in a core, hence
628 * invalidating on one thread invalidates for all.
629 * Thus we make all 4 threads use the same bit here.
632 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
633 clrldi r7,r6,64-6 /* extract bit number (6 bits) */
634 srdi r6,r6,6 /* doubleword number */
635 sldi r6,r6,3 /* address offset */
637 addi r6,r6,KVM_NEED_FLUSH /* dword in kvm->arch.need_tlb_flush */
643 /* Flush the TLB of any entries for this LPID */
644 lwz r0,KVM_TLB_SETS(r9)
646 li r7,0x800 /* IS field = 0b10 */
648 li r0,0 /* RS for P9 version of tlbiel */
650 28: tlbiel r7 /* On P9, rs=0, RIC=0, PRS=0, R=0 */
654 29: PPC_TLBIEL(7,0,2,1,1) /* for radix, RIC=2, PRS=1, R=1 */
658 23: ldarx r7,0,r6 /* clear the bit after TLB flushed */
663 /* Add timebase offset onto timebase */
664 22: ld r8,VCORE_TB_OFFSET(r5)
667 mftb r6 /* current host timebase */
669 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
670 mftb r7 /* check if lower 24 bits overflowed */
675 addis r8,r8,0x100 /* if so, increment upper 40 bits */
678 /* Load guest PCR value to select appropriate compat mode */
679 37: ld r7, VCORE_PCR(r5)
686 /* DPDES and VTB are shared between threads */
687 ld r8, VCORE_DPDES(r5)
691 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
693 /* Mark the subcore state as inside guest */
694 bl kvmppc_subcore_enter_guest
696 ld r5, HSTATE_KVM_VCORE(r13)
697 ld r4, HSTATE_KVM_VCPU(r13)
699 stb r0,VCORE_IN_GUEST(r5) /* signal secondaries to continue */
701 /* Do we have a guest vcpu to run? */
703 beq kvmppc_primary_no_guest
706 /* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
707 lwz r5,VCPU_SLB_MAX(r4)
712 1: ld r8,VCPU_SLB_E(r6)
715 addi r6,r6,VCPU_SLB_SIZE
718 /* Increment yield count if they have a VPA */
722 li r6, LPPACA_YIELDCOUNT
727 stb r6, VCPU_VPA_DIRTY(r4)
730 /* Save purr/spurr */
733 std r5,HSTATE_PURR(r13)
734 std r6,HSTATE_SPURR(r13)
740 /* Save host values of some registers */
746 std r5, STACK_SLOT_TID(r1)
747 std r6, STACK_SLOT_PSSCR(r1)
748 std r7, STACK_SLOT_PID(r1)
749 std r8, STACK_SLOT_IAMR(r1)
751 std r5, STACK_SLOT_HFSCR(r1)
752 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
757 std r5, STACK_SLOT_CIABR(r1)
758 std r6, STACK_SLOT_DAWR(r1)
759 std r7, STACK_SLOT_DAWRX(r1)
760 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
763 /* Set partition DABR */
764 /* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
765 lwz r5,VCPU_DABRX(r4)
770 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
772 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
775 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
778 END_FTR_SECTION_IFSET(CPU_FTR_TM)
781 /* Load guest PMU registers */
782 /* R4 is live here (vcpu pointer) */
784 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
785 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
789 andi. r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
792 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
793 lwz r3, VCPU_PMC(r4) /* always load up guest PMU registers */
794 lwz r5, VCPU_PMC + 4(r4) /* to prevent information leak */
795 lwz r6, VCPU_PMC + 8(r4)
796 lwz r7, VCPU_PMC + 12(r4)
797 lwz r8, VCPU_PMC + 16(r4)
798 lwz r9, VCPU_PMC + 20(r4)
806 ld r5, VCPU_MMCR + 8(r4)
807 ld r6, VCPU_MMCR + 16(r4)
815 ld r5, VCPU_MMCR + 24(r4)
819 BEGIN_FTR_SECTION_NESTED(96)
820 lwz r7, VCPU_PMC + 24(r4)
821 lwz r8, VCPU_PMC + 28(r4)
822 ld r9, VCPU_MMCR + 32(r4)
826 END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
827 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
831 /* Load up FP, VMX and VSX registers */
834 ld r14, VCPU_GPR(R14)(r4)
835 ld r15, VCPU_GPR(R15)(r4)
836 ld r16, VCPU_GPR(R16)(r4)
837 ld r17, VCPU_GPR(R17)(r4)
838 ld r18, VCPU_GPR(R18)(r4)
839 ld r19, VCPU_GPR(R19)(r4)
840 ld r20, VCPU_GPR(R20)(r4)
841 ld r21, VCPU_GPR(R21)(r4)
842 ld r22, VCPU_GPR(R22)(r4)
843 ld r23, VCPU_GPR(R23)(r4)
844 ld r24, VCPU_GPR(R24)(r4)
845 ld r25, VCPU_GPR(R25)(r4)
846 ld r26, VCPU_GPR(R26)(r4)
847 ld r27, VCPU_GPR(R27)(r4)
848 ld r28, VCPU_GPR(R28)(r4)
849 ld r29, VCPU_GPR(R29)(r4)
850 ld r30, VCPU_GPR(R30)(r4)
851 ld r31, VCPU_GPR(R31)(r4)
853 /* Switch DSCR to guest value */
858 /* Skip next section on POWER7 */
860 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
861 /* Load up POWER8-specific registers */
863 lwz r6, VCPU_PSPB(r4)
869 ld r6, VCPU_DAWRX(r4)
870 ld r7, VCPU_CIABR(r4)
877 ld r8, VCPU_EBBHR(r4)
880 ld r5, VCPU_EBBRR(r4)
881 ld r6, VCPU_BESCR(r4)
882 lwz r7, VCPU_GUEST_PID(r4)
890 END_FTR_SECTION_IFSET(CPU_FTR_POWER9_DD1)
892 /* POWER8-only registers */
893 ld r5, VCPU_TCSCR(r4)
895 ld r7, VCPU_CSIGR(r4)
902 /* POWER9-only registers */
904 ld r6, VCPU_PSSCR(r4)
905 oris r6, r6, PSSCR_EC@h /* This makes stop trap to HV */
906 ld r7, VCPU_HFSCR(r4)
910 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
914 * Set the decrementer to the guest decrementer.
916 ld r8,VCPU_DEC_EXPIRES(r4)
917 /* r8 is a host timebase value here, convert to guest TB */
918 ld r5,HSTATE_KVM_VCORE(r13)
919 ld r6,VCORE_TB_OFFSET(r5)
926 ld r5, VCPU_SPRG0(r4)
927 ld r6, VCPU_SPRG1(r4)
928 ld r7, VCPU_SPRG2(r4)
929 ld r8, VCPU_SPRG3(r4)
935 /* Load up DAR and DSISR */
937 lwz r6, VCPU_DSISR(r4)
941 /* Restore AMR and UAMOR, set AMOR to all 1s */
949 /* Restore state of CTRL run bit; assume 1 on entry */
957 /* Secondary threads wait for primary to have done partition switch */
958 ld r5, HSTATE_KVM_VCORE(r13)
959 lbz r6, HSTATE_PTID(r13)
962 lbz r0, VCORE_IN_GUEST(r5)
966 20: lwz r3, VCORE_ENTRY_EXIT(r5)
969 lbz r0, VCORE_IN_GUEST(r5)
979 /* Check if HDEC expires soon */
982 cmpdi r3, 512 /* 1 microsecond */
985 #ifdef CONFIG_KVM_XICS
986 /* We are entering the guest on that thread, push VCPU to XIVE */
987 ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
990 ld r11, VCPU_XIVE_SAVED_STATE(r4)
994 lwz r11, VCPU_XIVE_CAM_WORD(r4)
995 li r9, TM_QW1_OS + TM_WORD2
998 stw r9, VCPU_XIVE_PUSHED(r4)
1001 #endif /* CONFIG_KVM_XICS */
1003 deliver_guest_interrupt:
1010 kvmppc_cede_reentry: /* r4 = vcpu, r13 = paca */
1012 ld r11, VCPU_MSR(r4)
1013 ld r6, VCPU_SRR0(r4)
1014 ld r7, VCPU_SRR1(r4)
1018 /* r11 = vcpu->arch.msr & ~MSR_HV */
1019 rldicl r11, r11, 63 - MSR_HV_LG, 1
1020 rotldi r11, r11, 1 + MSR_HV_LG
1021 ori r11, r11, MSR_ME
1023 /* Check if we can deliver an external or decrementer interrupt now */
1024 ld r0, VCPU_PENDING_EXC(r4)
1025 rldicl r0, r0, 64 - BOOK3S_IRQPRIO_EXTERNAL_LEVEL, 63
1027 andi. r8, r11, MSR_EE
1029 /* Insert EXTERNAL_LEVEL bit into LPCR at the MER bit position */
1030 rldimi r8, r0, LPCR_MER_SH, 63 - LPCR_MER_SH
1034 li r0, BOOK3S_INTERRUPT_EXTERNAL
1038 /* On POWER9 check whether the guest has large decrementer enabled */
1039 andis. r8, r8, LPCR_LD@h
1041 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1044 li r0, BOOK3S_INTERRUPT_DECREMENTER
1047 12: mtspr SPRN_SRR0, r10
1049 mtspr SPRN_SRR1, r11
1051 bl kvmppc_msr_interrupt
1055 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1056 /* On POWER9, check for pending doorbell requests */
1057 lbz r0, VCPU_DBELL_REQ(r4)
1059 beq fast_guest_return
1060 ld r5, HSTATE_KVM_VCORE(r13)
1061 /* Set DPDES register so the CPU will take a doorbell interrupt */
1063 mtspr SPRN_DPDES, r0
1064 std r0, VCORE_DPDES(r5)
1065 /* Make sure other cpus see vcore->dpdes set before dbell req clear */
1067 /* Clear the pending doorbell request */
1069 stb r0, VCPU_DBELL_REQ(r4)
1074 * R10: value for HSRR0
1075 * R11: value for HSRR1
1080 stb r0,VCPU_CEDED(r4) /* cancel cede */
1081 mtspr SPRN_HSRR0,r10
1082 mtspr SPRN_HSRR1,r11
1084 /* Activate guest mode, so faults get handled by KVM */
1085 li r9, KVM_GUEST_MODE_GUEST_HV
1086 stb r9, HSTATE_IN_GUEST(r13)
1088 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1089 /* Accumulate timing */
1090 addi r3, r4, VCPU_TB_GUEST
1091 bl kvmhv_accumulate_time
1097 ld r5, VCPU_CFAR(r4)
1099 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1102 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1109 ld r1, VCPU_GPR(R1)(r4)
1110 ld r2, VCPU_GPR(R2)(r4)
1111 ld r3, VCPU_GPR(R3)(r4)
1112 ld r5, VCPU_GPR(R5)(r4)
1113 ld r6, VCPU_GPR(R6)(r4)
1114 ld r7, VCPU_GPR(R7)(r4)
1115 ld r8, VCPU_GPR(R8)(r4)
1116 ld r9, VCPU_GPR(R9)(r4)
1117 ld r10, VCPU_GPR(R10)(r4)
1118 ld r11, VCPU_GPR(R11)(r4)
1119 ld r12, VCPU_GPR(R12)(r4)
1120 ld r13, VCPU_GPR(R13)(r4)
1124 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1126 /* Move canary into DSISR to check for later */
1129 mtspr SPRN_HDSISR, r0
1130 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1132 ld r0, VCPU_GPR(R0)(r4)
1133 ld r4, VCPU_GPR(R4)(r4)
1139 stw r12, STACK_SLOT_TRAP(r1)
1142 stw r12, VCPU_TRAP(r4)
1143 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1144 addi r3, r4, VCPU_TB_RMEXIT
1145 bl kvmhv_accumulate_time
1147 11: b kvmhv_switch_to_host
1154 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
1155 12: stw r12, VCPU_TRAP(r4)
1157 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1158 addi r3, r4, VCPU_TB_RMEXIT
1159 bl kvmhv_accumulate_time
1163 /******************************************************************************
1167 *****************************************************************************/
1170 * We come here from the first-level interrupt handlers.
1172 .globl kvmppc_interrupt_hv
1173 kvmppc_interrupt_hv:
1175 * Register contents:
1176 * R12 = (guest CR << 32) | interrupt vector
1178 * guest R12 saved in shadow VCPU SCRATCH0
1179 * guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
1180 * guest R13 saved in SPRN_SCRATCH0
1182 std r9, HSTATE_SCRATCH2(r13)
1183 lbz r9, HSTATE_IN_GUEST(r13)
1184 cmpwi r9, KVM_GUEST_MODE_HOST_HV
1185 beq kvmppc_bad_host_intr
1186 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1187 cmpwi r9, KVM_GUEST_MODE_GUEST
1188 ld r9, HSTATE_SCRATCH2(r13)
1189 beq kvmppc_interrupt_pr
1191 /* We're now back in the host but in guest MMU context */
1192 li r9, KVM_GUEST_MODE_HOST_HV
1193 stb r9, HSTATE_IN_GUEST(r13)
1195 ld r9, HSTATE_KVM_VCPU(r13)
1197 /* Save registers */
1199 std r0, VCPU_GPR(R0)(r9)
1200 std r1, VCPU_GPR(R1)(r9)
1201 std r2, VCPU_GPR(R2)(r9)
1202 std r3, VCPU_GPR(R3)(r9)
1203 std r4, VCPU_GPR(R4)(r9)
1204 std r5, VCPU_GPR(R5)(r9)
1205 std r6, VCPU_GPR(R6)(r9)
1206 std r7, VCPU_GPR(R7)(r9)
1207 std r8, VCPU_GPR(R8)(r9)
1208 ld r0, HSTATE_SCRATCH2(r13)
1209 std r0, VCPU_GPR(R9)(r9)
1210 std r10, VCPU_GPR(R10)(r9)
1211 std r11, VCPU_GPR(R11)(r9)
1212 ld r3, HSTATE_SCRATCH0(r13)
1213 std r3, VCPU_GPR(R12)(r9)
1214 /* CR is in the high half of r12 */
1218 ld r3, HSTATE_CFAR(r13)
1219 std r3, VCPU_CFAR(r9)
1220 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1222 ld r4, HSTATE_PPR(r13)
1223 std r4, VCPU_PPR(r9)
1224 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1226 /* Restore R1/R2 so we can handle faults */
1227 ld r1, HSTATE_HOST_R1(r13)
1230 mfspr r10, SPRN_SRR0
1231 mfspr r11, SPRN_SRR1
1232 std r10, VCPU_SRR0(r9)
1233 std r11, VCPU_SRR1(r9)
1234 /* trap is in the low half of r12, clear CR from the high half */
1236 andi. r0, r12, 2 /* need to read HSRR0/1? */
1238 mfspr r10, SPRN_HSRR0
1239 mfspr r11, SPRN_HSRR1
1241 1: std r10, VCPU_PC(r9)
1242 std r11, VCPU_MSR(r9)
1246 std r3, VCPU_GPR(R13)(r9)
1249 stw r12,VCPU_TRAP(r9)
1252 * Now that we have saved away SRR0/1 and HSRR0/1,
1253 * interrupts are recoverable in principle, so set MSR_RI.
1254 * This becomes important for relocation-on interrupts from
1255 * the guest, which we can get in radix mode on POWER9.
1260 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1261 addi r3, r9, VCPU_TB_RMINTR
1263 bl kvmhv_accumulate_time
1264 ld r5, VCPU_GPR(R5)(r9)
1265 ld r6, VCPU_GPR(R6)(r9)
1266 ld r7, VCPU_GPR(R7)(r9)
1267 ld r8, VCPU_GPR(R8)(r9)
1270 /* Save HEIR (HV emulation assist reg) in emul_inst
1271 if this is an HEI (HV emulation interrupt, e40) */
1272 li r3,KVM_INST_FETCH_FAILED
1273 stw r3,VCPU_LAST_INST(r9)
1274 cmpwi r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1277 11: stw r3,VCPU_HEIR(r9)
1279 /* these are volatile across C function calls */
1280 #ifdef CONFIG_RELOCATABLE
1281 ld r3, HSTATE_SCRATCH1(r13)
1287 std r3, VCPU_CTR(r9)
1288 std r4, VCPU_XER(r9)
1290 /* If this is a page table miss then see if it's theirs or ours */
1291 cmpwi r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1293 cmpwi r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1296 /* See if this is a leftover HDEC interrupt */
1297 cmpwi r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1303 bge fast_guest_return
1305 /* See if this is an hcall we can handle in real mode */
1306 cmpwi r12,BOOK3S_INTERRUPT_SYSCALL
1307 beq hcall_try_real_mode
1309 /* Hypervisor doorbell - exit only if host IPI flag set */
1310 cmpwi r12, BOOK3S_INTERRUPT_H_DOORBELL
1315 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1316 lbz r0, HSTATE_HOST_IPI(r13)
1321 /* If it's a hypervisor facility unavailable interrupt, save HFSCR */
1322 cmpwi r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
1324 mfspr r3, SPRN_HFSCR
1325 std r3, VCPU_HFSCR(r9)
1328 /* External interrupt ? */
1329 cmpwi r12, BOOK3S_INTERRUPT_EXTERNAL
1330 bne+ guest_exit_cont
1332 /* External interrupt, first check for host_ipi. If this is
1333 * set, we know the host wants us out so let's do it now
1338 * Restore the active volatile registers after returning from
1341 ld r9, HSTATE_KVM_VCPU(r13)
1342 li r12, BOOK3S_INTERRUPT_EXTERNAL
1345 * kvmppc_read_intr return codes:
1347 * Exit to host (r3 > 0)
1348 * 1 An interrupt is pending that needs to be handled by the host
1349 * Exit guest and return to host by branching to guest_exit_cont
1351 * 2 Passthrough that needs completion in the host
1352 * Exit guest and return to host by branching to guest_exit_cont
1353 * However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
1354 * to indicate to the host to complete handling the interrupt
1356 * Before returning to guest, we check if any CPU is heading out
1357 * to the host and if so, we head out also. If no CPUs are heading
1358 * check return values <= 0.
1360 * Return to guest (r3 <= 0)
1361 * 0 No external interrupt is pending
1362 * -1 A guest wakeup IPI (which has now been cleared)
1363 * In either case, we return to guest to deliver any pending
1366 * -2 A PCI passthrough external interrupt was handled
1367 * (interrupt was delivered directly to guest)
1368 * Return to guest to deliver any pending guest interrupts.
1374 /* Return code = 2 */
1375 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
1376 stw r12, VCPU_TRAP(r9)
1379 1: /* Return code <= 1 */
1383 /* Return code <= 0 */
1384 4: ld r5, HSTATE_KVM_VCORE(r13)
1385 lwz r0, VCORE_ENTRY_EXIT(r5)
1388 blt deliver_guest_interrupt
1390 guest_exit_cont: /* r9 = vcpu, r12 = trap, r13 = paca */
1391 /* Save more register state */
1394 std r6, VCPU_DAR(r9)
1395 stw r7, VCPU_DSISR(r9)
1396 /* don't overwrite fault_dar/fault_dsisr if HDSI */
1397 cmpwi r12,BOOK3S_INTERRUPT_H_DATA_STORAGE
1399 std r6, VCPU_FAULT_DAR(r9)
1400 stw r7, VCPU_FAULT_DSISR(r9)
1402 /* See if it is a machine check */
1403 cmpwi r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1404 beq machine_check_realmode
1406 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1407 addi r3, r9, VCPU_TB_RMEXIT
1409 bl kvmhv_accumulate_time
1411 #ifdef CONFIG_KVM_XICS
1412 /* We are exiting, pull the VP from the XIVE */
1413 lwz r0, VCPU_XIVE_PUSHED(r9)
1416 li r7, TM_SPC_PULL_OS_CTX
1419 andi. r0, r0, MSR_IR /* in real mode? */
1421 ld r10, HSTATE_XIVE_TIMA_VIRT(r13)
1424 /* First load to pull the context, we ignore the value */
1427 /* Second load to recover the context state (Words 0 and 1) */
1430 2: ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
1433 /* First load to pull the context, we ignore the value */
1436 /* Second load to recover the context state (Words 0 and 1) */
1438 3: std r11, VCPU_XIVE_SAVED_STATE(r9)
1439 /* Fixup some of the state for the next load */
1442 stw r10, VCPU_XIVE_PUSHED(r9)
1443 stb r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
1444 stb r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
1447 #endif /* CONFIG_KVM_XICS */
1449 /* Possibly flush the link stack here. */
1451 patch_site 1b patch__call_kvm_flush_link_stack
1453 stw r12, STACK_SLOT_TRAP(r1)
1455 /* Increment exit count, poke other threads to exit */
1456 bl kvmhv_commence_exit
1458 ld r9, HSTATE_KVM_VCPU(r13)
1460 /* Stop others sending VCPU interrupts to this physical CPU */
1462 stw r0, VCPU_CPU(r9)
1463 stw r0, VCPU_THREAD_CPU(r9)
1465 /* Save guest CTRL register, set runlatch to 1 */
1467 stw r6,VCPU_CTRL(r9)
1473 /* Check if we are running hash or radix and store it in cr2 */
1475 lbz r0, KVM_RADIX(r5)
1478 /* Read the guest SLB and save it away */
1480 bne cr2, 3f /* for radix, save 0 entries */
1481 lwz r0,VCPU_SLB_NR(r9) /* number of entries in SLB */
1486 andis. r0,r8,SLB_ESID_V@h
1488 add r8,r8,r6 /* put index in */
1490 std r8,VCPU_SLB_E(r7)
1491 std r3,VCPU_SLB_V(r7)
1492 addi r7,r7,VCPU_SLB_SIZE
1496 3: stw r5,VCPU_SLB_MAX(r9)
1499 * Save the guest PURR/SPURR
1504 ld r8,VCPU_SPURR(r9)
1505 std r5,VCPU_PURR(r9)
1506 std r6,VCPU_SPURR(r9)
1511 * Restore host PURR/SPURR and add guest times
1512 * so that the time in the guest gets accounted.
1514 ld r3,HSTATE_PURR(r13)
1515 ld r4,HSTATE_SPURR(r13)
1522 ld r3, HSTATE_KVM_VCORE(r13)
1525 /* On P9, if the guest has large decr enabled, don't sign extend */
1527 ld r4, VCORE_LPCR(r3)
1528 andis. r4, r4, LPCR_LD@h
1530 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1533 /* r5 is a guest timebase value here, convert to host TB */
1534 ld r4,VCORE_TB_OFFSET(r3)
1536 std r5,VCPU_DEC_EXPIRES(r9)
1540 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1541 /* Save POWER8-specific registers */
1545 std r5, VCPU_IAMR(r9)
1546 stw r6, VCPU_PSPB(r9)
1547 std r7, VCPU_FSCR(r9)
1551 std r7, VCPU_TAR(r9)
1552 mfspr r8, SPRN_EBBHR
1553 std r8, VCPU_EBBHR(r9)
1554 mfspr r5, SPRN_EBBRR
1555 mfspr r6, SPRN_BESCR
1558 std r5, VCPU_EBBRR(r9)
1559 std r6, VCPU_BESCR(r9)
1560 stw r7, VCPU_GUEST_PID(r9)
1561 std r8, VCPU_WORT(r9)
1563 mfspr r5, SPRN_TCSCR
1565 mfspr r7, SPRN_CSIGR
1567 std r5, VCPU_TCSCR(r9)
1568 std r6, VCPU_ACOP(r9)
1569 std r7, VCPU_CSIGR(r9)
1570 std r8, VCPU_TACR(r9)
1573 mfspr r6, SPRN_PSSCR
1574 std r5, VCPU_TID(r9)
1575 rldicl r6, r6, 4, 50 /* r6 &= PSSCR_GUEST_VIS */
1577 std r6, VCPU_PSSCR(r9)
1578 /* Restore host HFSCR value */
1579 ld r7, STACK_SLOT_HFSCR(r1)
1580 mtspr SPRN_HFSCR, r7
1581 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
1583 * Restore various registers to 0, where non-zero values
1584 * set by the guest could disrupt the host.
1591 mtspr SPRN_TCSCR, r0
1592 /* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1595 mtspr SPRN_MMCRS, r0
1596 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1599 /* Save and reset AMR and UAMOR before turning on the MMU */
1603 std r6,VCPU_UAMOR(r9)
1606 mtspr SPRN_UAMOR, r6
1608 /* Switch DSCR back to host value */
1610 ld r7, HSTATE_DSCR(r13)
1611 std r8, VCPU_DSCR(r9)
1614 /* Save non-volatile GPRs */
1615 std r14, VCPU_GPR(R14)(r9)
1616 std r15, VCPU_GPR(R15)(r9)
1617 std r16, VCPU_GPR(R16)(r9)
1618 std r17, VCPU_GPR(R17)(r9)
1619 std r18, VCPU_GPR(R18)(r9)
1620 std r19, VCPU_GPR(R19)(r9)
1621 std r20, VCPU_GPR(R20)(r9)
1622 std r21, VCPU_GPR(R21)(r9)
1623 std r22, VCPU_GPR(R22)(r9)
1624 std r23, VCPU_GPR(R23)(r9)
1625 std r24, VCPU_GPR(R24)(r9)
1626 std r25, VCPU_GPR(R25)(r9)
1627 std r26, VCPU_GPR(R26)(r9)
1628 std r27, VCPU_GPR(R27)(r9)
1629 std r28, VCPU_GPR(R28)(r9)
1630 std r29, VCPU_GPR(R29)(r9)
1631 std r30, VCPU_GPR(R30)(r9)
1632 std r31, VCPU_GPR(R31)(r9)
1635 mfspr r3, SPRN_SPRG0
1636 mfspr r4, SPRN_SPRG1
1637 mfspr r5, SPRN_SPRG2
1638 mfspr r6, SPRN_SPRG3
1639 std r3, VCPU_SPRG0(r9)
1640 std r4, VCPU_SPRG1(r9)
1641 std r5, VCPU_SPRG2(r9)
1642 std r6, VCPU_SPRG3(r9)
1648 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1651 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
1654 END_FTR_SECTION_IFSET(CPU_FTR_TM)
1657 /* Increment yield count if they have a VPA */
1658 ld r8, VCPU_VPA(r9) /* do they have a VPA? */
1661 li r4, LPPACA_YIELDCOUNT
1666 stb r3, VCPU_VPA_DIRTY(r9)
1668 /* Save PMU registers if requested */
1669 /* r8 and cr0.eq are live here */
1672 * POWER8 seems to have a hardware bug where setting
1673 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
1674 * when some counters are already negative doesn't seem
1675 * to cause a performance monitor alert (and hence interrupt).
1676 * The effect of this is that when saving the PMU state,
1677 * if there is no PMU alert pending when we read MMCR0
1678 * before freezing the counters, but one becomes pending
1679 * before we read the counters, we lose it.
1680 * To work around this, we need a way to freeze the counters
1681 * before reading MMCR0. Normally, freezing the counters
1682 * is done by writing MMCR0 (to set MMCR0[FC]) which
1683 * unavoidably writes MMCR0[PMA0] as well. On POWER8,
1684 * we can also freeze the counters using MMCR2, by writing
1685 * 1s to all the counter freeze condition bits (there are
1686 * 9 bits each for 6 counters).
1688 li r3, -1 /* set all freeze bits */
1690 mfspr r10, SPRN_MMCR2
1691 mtspr SPRN_MMCR2, r3
1693 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1695 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
1696 mfspr r4, SPRN_MMCR0 /* save MMCR0 */
1697 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
1698 mfspr r6, SPRN_MMCRA
1699 /* Clear MMCRA in order to disable SDAR updates */
1701 mtspr SPRN_MMCRA, r7
1703 beq 21f /* if no VPA, save PMU stuff anyway */
1704 lbz r7, LPPACA_PMCINUSE(r8)
1705 cmpwi r7, 0 /* did they ask for PMU stuff to be saved? */
1707 std r3, VCPU_MMCR(r9) /* if not, set saved MMCR0 to FC */
1709 21: mfspr r5, SPRN_MMCR1
1712 std r4, VCPU_MMCR(r9)
1713 std r5, VCPU_MMCR + 8(r9)
1714 std r6, VCPU_MMCR + 16(r9)
1716 std r10, VCPU_MMCR + 24(r9)
1717 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1718 std r7, VCPU_SIAR(r9)
1719 std r8, VCPU_SDAR(r9)
1726 stw r3, VCPU_PMC(r9)
1727 stw r4, VCPU_PMC + 4(r9)
1728 stw r5, VCPU_PMC + 8(r9)
1729 stw r6, VCPU_PMC + 12(r9)
1730 stw r7, VCPU_PMC + 16(r9)
1731 stw r8, VCPU_PMC + 20(r9)
1734 std r5, VCPU_SIER(r9)
1735 BEGIN_FTR_SECTION_NESTED(96)
1736 mfspr r6, SPRN_SPMC1
1737 mfspr r7, SPRN_SPMC2
1738 mfspr r8, SPRN_MMCRS
1739 stw r6, VCPU_PMC + 24(r9)
1740 stw r7, VCPU_PMC + 28(r9)
1741 std r8, VCPU_MMCR + 32(r9)
1743 mtspr SPRN_MMCRS, r4
1744 END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
1745 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1748 /* Restore host values of some registers */
1750 ld r5, STACK_SLOT_CIABR(r1)
1751 ld r6, STACK_SLOT_DAWR(r1)
1752 ld r7, STACK_SLOT_DAWRX(r1)
1753 mtspr SPRN_CIABR, r5
1755 mtspr SPRN_DAWRX, r7
1756 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1758 ld r5, STACK_SLOT_TID(r1)
1759 ld r6, STACK_SLOT_PSSCR(r1)
1760 ld r7, STACK_SLOT_PID(r1)
1761 ld r8, STACK_SLOT_IAMR(r1)
1763 mtspr SPRN_PSSCR, r6
1766 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1768 #ifdef CONFIG_PPC_RADIX_MMU
1770 * Are we running hash or radix ?
1773 lbz r0, KVM_RADIX(r5)
1777 /* Radix: Handle the case where the guest used an illegal PID */
1778 LOAD_REG_ADDR(r4, mmu_base_pid)
1779 lwz r3, VCPU_GUEST_PID(r9)
1785 * Illegal PID, the HW might have prefetched and cached in the TLB
1786 * some translations for the LPID 0 / guest PID combination which
1787 * Linux doesn't know about, so we need to flush that PID out of
1788 * the TLB. First we need to set LPIDR to 0 so tlbiel applies to
1789 * the right context.
1795 /* Then do a congruence class local flush */
1797 lwz r0,KVM_TLB_SETS(r6)
1799 li r7,0x400 /* IS field = 0b01 */
1801 sldi r0,r3,32 /* RS has PID */
1802 1: PPC_TLBIEL(7,0,2,1,1) /* RIC=2, PRS=1, R=1 */
1807 2: /* Flush the ERAT on radix P9 DD1 guest exit */
1810 END_FTR_SECTION_IFSET(CPU_FTR_POWER9_DD1)
1812 #endif /* CONFIG_PPC_RADIX_MMU */
1814 /* Hash: clear out SLB */
1821 * POWER7/POWER8 guest -> host partition switch code.
1822 * We don't have to lock against tlbies but we do
1823 * have to coordinate the hardware threads.
1824 * Here STACK_SLOT_TRAP(r1) contains the trap number.
1826 kvmhv_switch_to_host:
1827 /* Secondary threads wait for primary to do partition switch */
1828 ld r5,HSTATE_KVM_VCORE(r13)
1829 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1830 lbz r3,HSTATE_PTID(r13)
1834 13: lbz r3,VCORE_IN_GUEST(r5)
1840 /* Primary thread waits for all the secondaries to exit guest */
1841 15: lwz r3,VCORE_ENTRY_EXIT(r5)
1842 rlwinm r0,r3,32-8,0xff
1848 /* Did we actually switch to the guest at all? */
1849 lbz r6, VCORE_IN_GUEST(r5)
1853 /* Primary thread switches back to host partition */
1854 lwz r7,KVM_HOST_LPID(r4)
1856 ld r6,KVM_HOST_SDR1(r4)
1857 li r8,LPID_RSVD /* switch to reserved LPID */
1860 mtspr SPRN_SDR1,r6 /* switch to host page table */
1861 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1866 /* DPDES and VTB are shared between threads */
1867 mfspr r7, SPRN_DPDES
1869 std r7, VCORE_DPDES(r5)
1870 std r8, VCORE_VTB(r5)
1871 /* clear DPDES so we don't get guest doorbells in the host */
1873 mtspr SPRN_DPDES, r8
1874 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1876 /* If HMI, call kvmppc_realmode_hmi_handler() */
1877 lwz r12, STACK_SLOT_TRAP(r1)
1878 cmpwi r12, BOOK3S_INTERRUPT_HMI
1880 bl kvmppc_realmode_hmi_handler
1883 * At this point kvmppc_realmode_hmi_handler would have resync-ed
1884 * the TB. Hence it is not required to subtract guest timebase
1885 * offset from timebase. So, skip it.
1887 * Also, do not call kvmppc_subcore_exit_guest() because it has
1888 * been invoked as part of kvmppc_realmode_hmi_handler().
1893 /* Subtract timebase offset from timebase */
1894 ld r8,VCORE_TB_OFFSET(r5)
1897 mftb r6 /* current guest timebase */
1899 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
1900 mftb r7 /* check if lower 24 bits overflowed */
1905 addis r8,r8,0x100 /* if so, increment upper 40 bits */
1908 17: bl kvmppc_subcore_exit_guest
1910 30: ld r5,HSTATE_KVM_VCORE(r13)
1911 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1914 ld r0, VCORE_PCR(r5)
1920 /* Signal secondary CPUs to continue */
1921 stb r0,VCORE_IN_GUEST(r5)
1922 19: lis r8,0x7fff /* MAX_INT@h */
1925 16: ld r8,KVM_HOST_LPCR(r4)
1929 /* load host SLB entries */
1930 BEGIN_MMU_FTR_SECTION
1932 END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
1933 ld r8,PACA_SLBSHADOWPTR(r13)
1935 .rept SLB_NUM_BOLTED
1936 li r3, SLBSHADOW_SAVEAREA
1940 andis. r7,r5,SLB_ESID_V@h
1946 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1947 /* Finish timing, if we have a vcpu */
1948 ld r4, HSTATE_KVM_VCPU(r13)
1952 bl kvmhv_accumulate_time
1955 /* Unset guest mode */
1956 li r0, KVM_GUEST_MODE_NONE
1957 stb r0, HSTATE_IN_GUEST(r13)
1959 lwz r12, STACK_SLOT_TRAP(r1) /* return trap # in r12 */
1960 ld r0, SFS+PPC_LR_STKOFF(r1)
1966 .global kvm_flush_link_stack
1967 kvm_flush_link_stack:
1968 /* Save LR into r0 */
1971 /* Flush the link stack. On Power8 it's up to 32 entries in size. */
1976 /* And on Power9 it's up to 64. */
1981 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1988 * Check whether an HDSI is an HPTE not found fault or something else.
1989 * If it is an HPTE not found fault that is due to the guest accessing
1990 * a page that they have mapped but which we have paged out, then
1991 * we continue on with the guest exit path. In all other cases,
1992 * reflect the HDSI to the guest as a DSI.
1996 lbz r0, KVM_RADIX(r3)
1998 mfspr r6, SPRN_HDSISR
2000 /* Look for DSISR canary. If we find it, retry instruction */
2003 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2005 bne .Lradix_hdsi /* on radix, just save DAR/DSISR/ASDR */
2006 /* HPTE not found fault or protection fault? */
2007 andis. r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
2008 beq 1f /* if not, send it to the guest */
2009 andi. r0, r11, MSR_DR /* data relocation enabled? */
2012 mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
2014 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2016 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
2017 li r0, BOOK3S_INTERRUPT_DATA_SEGMENT
2018 bne 7f /* if no SLB entry found */
2019 4: std r4, VCPU_FAULT_DAR(r9)
2020 stw r6, VCPU_FAULT_DSISR(r9)
2022 /* Search the hash table. */
2023 mr r3, r9 /* vcpu pointer */
2024 li r7, 1 /* data fault */
2025 bl kvmppc_hpte_hv_fault
2026 ld r9, HSTATE_KVM_VCPU(r13)
2028 ld r11, VCPU_MSR(r9)
2029 li r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
2030 cmpdi r3, 0 /* retry the instruction */
2032 cmpdi r3, -1 /* handle in kernel mode */
2034 cmpdi r3, -2 /* MMIO emulation; need instr word */
2037 /* Synthesize a DSI (or DSegI) for the guest */
2038 ld r4, VCPU_FAULT_DAR(r9)
2040 1: li r0, BOOK3S_INTERRUPT_DATA_STORAGE
2041 mtspr SPRN_DSISR, r6
2042 7: mtspr SPRN_DAR, r4
2043 mtspr SPRN_SRR0, r10
2044 mtspr SPRN_SRR1, r11
2046 bl kvmppc_msr_interrupt
2047 fast_interrupt_c_return:
2048 6: ld r7, VCPU_CTR(r9)
2055 3: ld r5, VCPU_KVM(r9) /* not relocated, use VRMA */
2056 ld r5, KVM_VRMA_SLB_V(r5)
2059 /* If this is for emulated MMIO, load the instruction word */
2060 2: li r8, KVM_INST_FETCH_FAILED /* In case lwz faults */
2062 /* Set guest mode to 'jump over instruction' so if lwz faults
2063 * we'll just continue at the next IP. */
2064 li r0, KVM_GUEST_MODE_SKIP
2065 stb r0, HSTATE_IN_GUEST(r13)
2067 /* Do the access with MSR:DR enabled */
2069 ori r4, r3, MSR_DR /* Enable paging for data */
2074 /* Store the result */
2075 stw r8, VCPU_LAST_INST(r9)
2077 /* Unset guest mode. */
2078 li r0, KVM_GUEST_MODE_HOST_HV
2079 stb r0, HSTATE_IN_GUEST(r13)
2083 std r4, VCPU_FAULT_DAR(r9)
2084 stw r6, VCPU_FAULT_DSISR(r9)
2087 std r5, VCPU_FAULT_GPA(r9)
2091 * Similarly for an HISI, reflect it to the guest as an ISI unless
2092 * it is an HPTE not found fault for a page that we have paged out.
2096 lbz r0, KVM_RADIX(r3)
2098 bne .Lradix_hisi /* for radix, just save ASDR */
2099 andis. r0, r11, SRR1_ISI_NOPT@h
2101 andi. r0, r11, MSR_IR /* instruction relocation enabled? */
2104 mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
2106 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2108 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
2109 li r0, BOOK3S_INTERRUPT_INST_SEGMENT
2110 bne 7f /* if no SLB entry found */
2112 /* Search the hash table. */
2113 mr r3, r9 /* vcpu pointer */
2116 li r7, 0 /* instruction fault */
2117 bl kvmppc_hpte_hv_fault
2118 ld r9, HSTATE_KVM_VCPU(r13)
2120 ld r11, VCPU_MSR(r9)
2121 li r12, BOOK3S_INTERRUPT_H_INST_STORAGE
2122 cmpdi r3, 0 /* retry the instruction */
2123 beq fast_interrupt_c_return
2124 cmpdi r3, -1 /* handle in kernel mode */
2127 /* Synthesize an ISI (or ISegI) for the guest */
2129 1: li r0, BOOK3S_INTERRUPT_INST_STORAGE
2130 7: mtspr SPRN_SRR0, r10
2131 mtspr SPRN_SRR1, r11
2133 bl kvmppc_msr_interrupt
2134 b fast_interrupt_c_return
2136 3: ld r6, VCPU_KVM(r9) /* not relocated, use VRMA */
2137 ld r5, KVM_VRMA_SLB_V(r6)
2141 * Try to handle an hcall in real mode.
2142 * Returns to the guest if we handle it, or continues on up to
2143 * the kernel if we can't (i.e. if we don't have a handler for
2144 * it, or if the handler returns H_TOO_HARD).
2146 * r5 - r8 contain hcall args,
2147 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
2149 hcall_try_real_mode:
2150 ld r3,VCPU_GPR(R3)(r9)
2152 /* sc 1 from userspace - reflect to guest syscall */
2153 bne sc_1_fast_return
2155 cmpldi r3,hcall_real_table_end - hcall_real_table
2157 /* See if this hcall is enabled for in-kernel handling */
2159 srdi r0, r3, 8 /* r0 = (r3 / 4) >> 6 */
2160 sldi r0, r0, 3 /* index into kvm->arch.enabled_hcalls[] */
2162 ld r0, KVM_ENABLED_HCALLS(r4)
2163 rlwinm r4, r3, 32-2, 0x3f /* r4 = (r3 / 4) & 0x3f */
2167 /* Get pointer to handler, if any, and call it */
2168 LOAD_REG_ADDR(r4, hcall_real_table)
2174 mr r3,r9 /* get vcpu pointer */
2175 ld r4,VCPU_GPR(R4)(r9)
2178 beq hcall_real_fallback
2179 ld r4,HSTATE_KVM_VCPU(r13)
2180 std r3,VCPU_GPR(R3)(r4)
2188 li r10, BOOK3S_INTERRUPT_SYSCALL
2189 bl kvmppc_msr_interrupt
2193 /* We've attempted a real mode hcall, but it's punted it back
2194 * to userspace. We need to restore some clobbered volatiles
2195 * before resuming the pass-it-to-qemu path */
2196 hcall_real_fallback:
2197 li r12,BOOK3S_INTERRUPT_SYSCALL
2198 ld r9, HSTATE_KVM_VCPU(r13)
2202 .globl hcall_real_table
2204 .long 0 /* 0 - unused */
2205 .long DOTSYM(kvmppc_h_remove) - hcall_real_table
2206 .long DOTSYM(kvmppc_h_enter) - hcall_real_table
2207 .long DOTSYM(kvmppc_h_read) - hcall_real_table
2208 .long DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
2209 .long DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
2210 .long DOTSYM(kvmppc_h_protect) - hcall_real_table
2211 .long DOTSYM(kvmppc_h_get_tce) - hcall_real_table
2212 .long DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
2213 .long 0 /* 0x24 - H_SET_SPRG0 */
2214 .long DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
2229 #ifdef CONFIG_KVM_XICS
2230 .long DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
2231 .long DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
2232 .long DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
2233 .long DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
2234 .long DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
2236 .long 0 /* 0x64 - H_EOI */
2237 .long 0 /* 0x68 - H_CPPR */
2238 .long 0 /* 0x6c - H_IPI */
2239 .long 0 /* 0x70 - H_IPOLL */
2240 .long 0 /* 0x74 - H_XIRR */
2268 .long DOTSYM(kvmppc_h_cede) - hcall_real_table
2269 .long DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
2285 .long DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
2289 .long DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
2290 .long DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
2291 .long DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
2403 #ifdef CONFIG_KVM_XICS
2404 .long DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
2406 .long 0 /* 0x2fc - H_XIRR_X*/
2408 .long DOTSYM(kvmppc_h_random) - hcall_real_table
2409 .globl hcall_real_table_end
2410 hcall_real_table_end:
2412 _GLOBAL(kvmppc_h_set_xdabr)
2413 andi. r0, r5, DABRX_USER | DABRX_KERNEL
2415 li r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2418 6: li r3, H_PARAMETER
2421 _GLOBAL(kvmppc_h_set_dabr)
2422 li r5, DABRX_USER | DABRX_KERNEL
2426 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2427 std r4,VCPU_DABR(r3)
2428 stw r5, VCPU_DABRX(r3)
2429 mtspr SPRN_DABRX, r5
2430 /* Work around P7 bug where DABR can get corrupted on mtspr */
2431 1: mtspr SPRN_DABR,r4
2439 /* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
2440 2: rlwimi r5, r4, 5, DAWRX_DR | DAWRX_DW
2441 rlwimi r5, r4, 2, DAWRX_WT
2443 std r4, VCPU_DAWR(r3)
2444 std r5, VCPU_DAWRX(r3)
2446 mtspr SPRN_DAWRX, r5
2450 _GLOBAL(kvmppc_h_cede) /* r3 = vcpu pointer, r11 = msr, r13 = paca */
2452 std r11,VCPU_MSR(r3)
2454 stb r0,VCPU_CEDED(r3)
2455 sync /* order setting ceded vs. testing prodded */
2456 lbz r5,VCPU_PRODDED(r3)
2458 bne kvm_cede_prodded
2459 li r12,0 /* set trap to 0 to say hcall is handled */
2460 stw r12,VCPU_TRAP(r3)
2462 std r0,VCPU_GPR(R3)(r3)
2465 * Set our bit in the bitmask of napping threads unless all the
2466 * other threads are already napping, in which case we send this
2469 ld r5,HSTATE_KVM_VCORE(r13)
2470 lbz r6,HSTATE_PTID(r13)
2471 lwz r8,VCORE_ENTRY_EXIT(r5)
2475 addi r6,r5,VCORE_NAPPING_THREADS
2482 /* order napping_threads update vs testing entry_exit_map */
2485 stb r0,HSTATE_NAPPING(r13)
2486 lwz r7,VCORE_ENTRY_EXIT(r5)
2488 bge 33f /* another thread already exiting */
2491 * Although not specifically required by the architecture, POWER7
2492 * preserves the following registers in nap mode, even if an SMT mode
2493 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2494 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2496 /* Save non-volatile GPRs */
2497 std r14, VCPU_GPR(R14)(r3)
2498 std r15, VCPU_GPR(R15)(r3)
2499 std r16, VCPU_GPR(R16)(r3)
2500 std r17, VCPU_GPR(R17)(r3)
2501 std r18, VCPU_GPR(R18)(r3)
2502 std r19, VCPU_GPR(R19)(r3)
2503 std r20, VCPU_GPR(R20)(r3)
2504 std r21, VCPU_GPR(R21)(r3)
2505 std r22, VCPU_GPR(R22)(r3)
2506 std r23, VCPU_GPR(R23)(r3)
2507 std r24, VCPU_GPR(R24)(r3)
2508 std r25, VCPU_GPR(R25)(r3)
2509 std r26, VCPU_GPR(R26)(r3)
2510 std r27, VCPU_GPR(R27)(r3)
2511 std r28, VCPU_GPR(R28)(r3)
2512 std r29, VCPU_GPR(R29)(r3)
2513 std r30, VCPU_GPR(R30)(r3)
2514 std r31, VCPU_GPR(R31)(r3)
2519 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2522 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
2524 ld r9, HSTATE_KVM_VCPU(r13)
2526 END_FTR_SECTION_IFSET(CPU_FTR_TM)
2530 * Set DEC to the smaller of DEC and HDEC, so that we wake
2531 * no later than the end of our timeslice (HDEC interrupts
2532 * don't wake us from nap).
2538 /* On P9 check whether the guest has large decrementer mode enabled */
2539 ld r6, HSTATE_KVM_VCORE(r13)
2540 ld r6, VCORE_LPCR(r6)
2541 andis. r6, r6, LPCR_LD@h
2543 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2550 /* save expiry time of guest decrementer */
2552 ld r4, HSTATE_KVM_VCPU(r13)
2553 ld r5, HSTATE_KVM_VCORE(r13)
2554 ld r6, VCORE_TB_OFFSET(r5)
2555 subf r3, r6, r3 /* convert to host TB value */
2556 std r3, VCPU_DEC_EXPIRES(r4)
2558 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2559 ld r4, HSTATE_KVM_VCPU(r13)
2560 addi r3, r4, VCPU_TB_CEDE
2561 bl kvmhv_accumulate_time
2564 lis r3, LPCR_PECEDP@h /* Do wake on privileged doorbell */
2567 * Take a nap until a decrementer or external or doobell interrupt
2568 * occurs, with PECE1 and PECE0 set in LPCR.
2569 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2570 * Also clear the runlatch bit before napping.
2573 mfspr r0, SPRN_CTRLF
2575 mtspr SPRN_CTRLT, r0
2579 stb r0,HSTATE_HWTHREAD_REQ(r13)
2580 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
2582 ori r5,r5,LPCR_PECE0 | LPCR_PECE1
2584 ori r5, r5, LPCR_PECEDH
2585 rlwimi r5, r3, 0, LPCR_PECEDP
2586 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2588 kvm_nap_sequence: /* desired LPCR value in r5 */
2591 * PSSCR bits: exit criterion = 1 (wakeup based on LPCR at sreset)
2592 * enable state loss = 1 (allow SMT mode switch)
2593 * requested level = 0 (just stop dispatching)
2595 lis r3, (PSSCR_EC | PSSCR_ESL)@h
2596 mtspr SPRN_PSSCR, r3
2597 /* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
2598 li r4, LPCR_PECE_HVEE@higher
2601 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2605 std r0, HSTATE_SCRATCH0(r13)
2607 ld r0, HSTATE_SCRATCH0(r13)
2614 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
2623 /* get vcpu pointer */
2624 ld r4, HSTATE_KVM_VCPU(r13)
2626 /* Woken by external or decrementer interrupt */
2627 ld r1, HSTATE_HOST_R1(r13)
2629 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2630 addi r3, r4, VCPU_TB_RMINTR
2631 bl kvmhv_accumulate_time
2634 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2637 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
2639 bl kvmppc_restore_tm
2640 END_FTR_SECTION_IFSET(CPU_FTR_TM)
2643 /* load up FP state */
2646 /* Restore guest decrementer */
2647 ld r3, VCPU_DEC_EXPIRES(r4)
2648 ld r5, HSTATE_KVM_VCORE(r13)
2649 ld r6, VCORE_TB_OFFSET(r5)
2650 add r3, r3, r6 /* convert host TB to guest TB value */
2656 ld r14, VCPU_GPR(R14)(r4)
2657 ld r15, VCPU_GPR(R15)(r4)
2658 ld r16, VCPU_GPR(R16)(r4)
2659 ld r17, VCPU_GPR(R17)(r4)
2660 ld r18, VCPU_GPR(R18)(r4)
2661 ld r19, VCPU_GPR(R19)(r4)
2662 ld r20, VCPU_GPR(R20)(r4)
2663 ld r21, VCPU_GPR(R21)(r4)
2664 ld r22, VCPU_GPR(R22)(r4)
2665 ld r23, VCPU_GPR(R23)(r4)
2666 ld r24, VCPU_GPR(R24)(r4)
2667 ld r25, VCPU_GPR(R25)(r4)
2668 ld r26, VCPU_GPR(R26)(r4)
2669 ld r27, VCPU_GPR(R27)(r4)
2670 ld r28, VCPU_GPR(R28)(r4)
2671 ld r29, VCPU_GPR(R29)(r4)
2672 ld r30, VCPU_GPR(R30)(r4)
2673 ld r31, VCPU_GPR(R31)(r4)
2675 /* Check the wake reason in SRR1 to see why we got here */
2676 bl kvmppc_check_wake_reason
2679 * Restore volatile registers since we could have called a
2680 * C routine in kvmppc_check_wake_reason
2682 * r3 tells us whether we need to return to host or not
2683 * WARNING: it gets checked further down:
2684 * should not modify r3 until this check is done.
2686 ld r4, HSTATE_KVM_VCPU(r13)
2688 /* clear our bit in vcore->napping_threads */
2689 34: ld r5,HSTATE_KVM_VCORE(r13)
2690 lbz r7,HSTATE_PTID(r13)
2693 addi r6,r5,VCORE_NAPPING_THREADS
2699 stb r0,HSTATE_NAPPING(r13)
2701 /* See if the wake reason saved in r3 means we need to exit */
2702 stw r12, VCPU_TRAP(r4)
2707 /* see if any other thread is already exiting */
2708 lwz r0,VCORE_ENTRY_EXIT(r5)
2712 b kvmppc_cede_reentry /* if not go back to guest */
2714 /* cede when already previously prodded case */
2717 stb r0,VCPU_PRODDED(r3)
2718 sync /* order testing prodded vs. clearing ceded */
2719 stb r0,VCPU_CEDED(r3)
2723 /* we've ceded but we want to give control to the host */
2725 ld r9, HSTATE_KVM_VCPU(r13)
2728 /* Try to handle a machine check in real mode */
2729 machine_check_realmode:
2730 mr r3, r9 /* get vcpu pointer */
2731 bl kvmppc_realmode_machine_check
2733 ld r9, HSTATE_KVM_VCPU(r13)
2734 li r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2736 * For the guest that is FWNMI capable, deliver all the MCE errors
2737 * (handled/unhandled) by exiting the guest with KVM_EXIT_NMI exit
2738 * reason. This new approach injects machine check errors in guest
2739 * address space to guest with additional information in the form
2740 * of RTAS event, thus enabling guest kernel to suitably handle
2743 * For the guest that is not FWNMI capable (old QEMU) fallback
2744 * to old behaviour for backward compatibility:
2745 * Deliver unhandled/fatal (e.g. UE) MCE errors to guest either
2746 * through machine check interrupt (set HSRR0 to 0x200).
2747 * For handled errors (no-fatal), just go back to guest execution
2748 * with current HSRR0.
2749 * if we receive machine check with MSR(RI=0) then deliver it to
2750 * guest as machine check causing guest to crash.
2752 ld r11, VCPU_MSR(r9)
2753 rldicl. r0, r11, 64-MSR_HV_LG, 63 /* check if it happened in HV mode */
2754 bne mc_cont /* if so, exit to host */
2755 /* Check if guest is capable of handling NMI exit */
2756 ld r10, VCPU_KVM(r9)
2757 lbz r10, KVM_FWNMI(r10)
2758 cmpdi r10, 1 /* FWNMI capable? */
2759 beq mc_cont /* if so, exit with KVM_EXIT_NMI. */
2761 /* if not, fall through for backward compatibility. */
2762 andi. r10, r11, MSR_RI /* check for unrecoverable exception */
2763 beq 1f /* Deliver a machine check to guest */
2765 cmpdi r3, 0 /* Did we handle MCE ? */
2766 bne 2f /* Continue guest execution. */
2767 /* If not, deliver a machine check. SRR0/1 are already set */
2768 1: li r10, BOOK3S_INTERRUPT_MACHINE_CHECK
2769 bl kvmppc_msr_interrupt
2770 2: b fast_interrupt_c_return
2773 * Check the reason we woke from nap, and take appropriate action.
2775 * 0 if nothing needs to be done
2776 * 1 if something happened that needs to be handled by the host
2777 * -1 if there was a guest wakeup (IPI or msgsnd)
2778 * -2 if we handled a PCI passthrough interrupt (returned by
2779 * kvmppc_read_intr only)
2781 * Also sets r12 to the interrupt vector for any interrupt that needs
2782 * to be handled now by the host (0x500 for external interrupt), or zero.
2783 * Modifies all volatile registers (since it may call a C function).
2784 * This routine calls kvmppc_read_intr, a C function, if an external
2785 * interrupt is pending.
2787 kvmppc_check_wake_reason:
2790 rlwinm r6, r6, 45-31, 0xf /* extract wake reason field (P8) */
2792 rlwinm r6, r6, 45-31, 0xe /* P7 wake reason field is 3 bits */
2793 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2794 cmpwi r6, 8 /* was it an external interrupt? */
2795 beq 7f /* if so, see what it was */
2798 cmpwi r6, 6 /* was it the decrementer? */
2801 cmpwi r6, 5 /* privileged doorbell? */
2803 cmpwi r6, 3 /* hypervisor doorbell? */
2805 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2806 cmpwi r6, 0xa /* Hypervisor maintenance ? */
2808 li r3, 1 /* anything else, return 1 */
2811 /* hypervisor doorbell */
2812 3: li r12, BOOK3S_INTERRUPT_H_DOORBELL
2815 * Clear the doorbell as we will invoke the handler
2816 * explicitly in the guest exit path.
2818 lis r6, (PPC_DBELL_SERVER << (63-36))@h
2820 /* see if it's a host IPI */
2825 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2826 lbz r0, HSTATE_HOST_IPI(r13)
2829 /* if not, return -1 */
2833 /* Woken up due to Hypervisor maintenance interrupt */
2834 4: li r12, BOOK3S_INTERRUPT_HMI
2838 /* external interrupt - create a stack frame so we can call C */
2840 std r0, PPC_LR_STKOFF(r1)
2841 stdu r1, -PPC_MIN_STKFRM(r1)
2844 li r12, BOOK3S_INTERRUPT_EXTERNAL
2849 * Return code of 2 means PCI passthrough interrupt, but
2850 * we need to return back to host to complete handling the
2851 * interrupt. Trap reason is expected in r12 by guest
2854 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
2856 ld r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
2857 addi r1, r1, PPC_MIN_STKFRM
2862 * Save away FP, VMX and VSX registers.
2864 * N.B. r30 and r31 are volatile across this function,
2865 * thus it is not callable from C.
2872 #ifdef CONFIG_ALTIVEC
2874 oris r8,r8,MSR_VEC@h
2875 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2879 oris r8,r8,MSR_VSX@h
2880 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2883 addi r3,r3,VCPU_FPRS
2885 #ifdef CONFIG_ALTIVEC
2887 addi r3,r31,VCPU_VRS
2889 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2891 mfspr r6,SPRN_VRSAVE
2892 stw r6,VCPU_VRSAVE(r31)
2897 * Load up FP, VMX and VSX registers
2899 * N.B. r30 and r31 are volatile across this function,
2900 * thus it is not callable from C.
2907 #ifdef CONFIG_ALTIVEC
2909 oris r8,r8,MSR_VEC@h
2910 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2914 oris r8,r8,MSR_VSX@h
2915 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2918 addi r3,r4,VCPU_FPRS
2920 #ifdef CONFIG_ALTIVEC
2922 addi r3,r31,VCPU_VRS
2924 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2926 lwz r7,VCPU_VRSAVE(r31)
2927 mtspr SPRN_VRSAVE,r7
2932 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2934 * Save transactional state and TM-related registers.
2935 * Called with r9 pointing to the vcpu struct.
2936 * This can modify all checkpointed registers, but
2937 * restores r1, r2 and r9 (vcpu pointer) before exit.
2941 std r0, PPC_LR_STKOFF(r1)
2946 rldimi r8, r0, MSR_TM_LG, 63-MSR_TM_LG
2950 rldicl. r5, r5, 64 - MSR_TS_S_LG, 62
2951 beq 1f /* TM not active in guest. */
2953 std r1, HSTATE_HOST_R1(r13)
2954 li r3, TM_CAUSE_KVM_RESCHED
2956 /* Clear the MSR RI since r1, r13 are all going to be foobar. */
2960 /* All GPRs are volatile at this point. */
2963 /* Temporarily store r13 and r9 so we have some regs to play with */
2966 std r9, PACATMSCRATCH(r13)
2967 ld r9, HSTATE_KVM_VCPU(r13)
2969 /* Get a few more GPRs free. */
2970 std r29, VCPU_GPRS_TM(29)(r9)
2971 std r30, VCPU_GPRS_TM(30)(r9)
2972 std r31, VCPU_GPRS_TM(31)(r9)
2974 /* Save away PPR and DSCR soon so don't run with user values. */
2977 mfspr r30, SPRN_DSCR
2978 ld r29, HSTATE_DSCR(r13)
2979 mtspr SPRN_DSCR, r29
2981 /* Save all but r9, r13 & r29-r31 */
2984 .if (reg != 9) && (reg != 13)
2985 std reg, VCPU_GPRS_TM(reg)(r9)
2989 /* ... now save r13 */
2991 std r4, VCPU_GPRS_TM(13)(r9)
2992 /* ... and save r9 */
2993 ld r4, PACATMSCRATCH(r13)
2994 std r4, VCPU_GPRS_TM(9)(r9)
2996 /* Reload stack pointer and TOC. */
2997 ld r1, HSTATE_HOST_R1(r13)
3000 /* Set MSR RI now we have r1 and r13 back. */
3004 /* Save away checkpinted SPRs. */
3005 std r31, VCPU_PPR_TM(r9)
3006 std r30, VCPU_DSCR_TM(r9)
3013 std r5, VCPU_LR_TM(r9)
3014 stw r6, VCPU_CR_TM(r9)
3015 std r7, VCPU_CTR_TM(r9)
3016 std r8, VCPU_AMR_TM(r9)
3017 std r10, VCPU_TAR_TM(r9)
3018 std r11, VCPU_XER_TM(r9)
3020 /* Restore r12 as trap number. */
3021 lwz r12, VCPU_TRAP(r9)
3024 addi r3, r9, VCPU_FPRS_TM
3026 addi r3, r9, VCPU_VRS_TM
3028 mfspr r6, SPRN_VRSAVE
3029 stw r6, VCPU_VRSAVE_TM(r9)
3032 * We need to save these SPRs after the treclaim so that the software
3033 * error code is recorded correctly in the TEXASR. Also the user may
3034 * change these outside of a transaction, so they must always be
3037 mfspr r5, SPRN_TFHAR
3038 mfspr r6, SPRN_TFIAR
3039 mfspr r7, SPRN_TEXASR
3040 std r5, VCPU_TFHAR(r9)
3041 std r6, VCPU_TFIAR(r9)
3042 std r7, VCPU_TEXASR(r9)
3044 ld r0, PPC_LR_STKOFF(r1)
3049 * Restore transactional state and TM-related registers.
3050 * Called with r4 pointing to the vcpu struct.
3051 * This potentially modifies all checkpointed registers.
3052 * It restores r1, r2, r4 from the PACA.
3056 std r0, PPC_LR_STKOFF(r1)
3058 /* Turn on TM/FP/VSX/VMX so we can restore them. */
3064 oris r5, r5, (MSR_VEC | MSR_VSX)@h
3068 * The user may change these outside of a transaction, so they must
3069 * always be context switched.
3071 ld r5, VCPU_TFHAR(r4)
3072 ld r6, VCPU_TFIAR(r4)
3073 ld r7, VCPU_TEXASR(r4)
3074 mtspr SPRN_TFHAR, r5
3075 mtspr SPRN_TFIAR, r6
3076 mtspr SPRN_TEXASR, r7
3079 rldicl. r5, r5, 64 - MSR_TS_S_LG, 62
3080 beqlr /* TM not active in guest */
3081 std r1, HSTATE_HOST_R1(r13)
3083 /* Make sure the failure summary is set, otherwise we'll program check
3084 * when we trechkpt. It's possible that this might have been not set
3085 * on a kvmppc_set_one_reg() call but we shouldn't let this crash the
3088 oris r7, r7, (TEXASR_FS)@h
3089 mtspr SPRN_TEXASR, r7
3092 * We need to load up the checkpointed state for the guest.
3093 * We need to do this early as it will blow away any GPRs, VSRs and
3098 addi r3, r31, VCPU_FPRS_TM
3100 addi r3, r31, VCPU_VRS_TM
3103 lwz r7, VCPU_VRSAVE_TM(r4)
3104 mtspr SPRN_VRSAVE, r7
3106 ld r5, VCPU_LR_TM(r4)
3107 lwz r6, VCPU_CR_TM(r4)
3108 ld r7, VCPU_CTR_TM(r4)
3109 ld r8, VCPU_AMR_TM(r4)
3110 ld r9, VCPU_TAR_TM(r4)
3111 ld r10, VCPU_XER_TM(r4)
3120 * Load up PPR and DSCR values but don't put them in the actual SPRs
3121 * till the last moment to avoid running with userspace PPR and DSCR for
3124 ld r29, VCPU_DSCR_TM(r4)
3125 ld r30, VCPU_PPR_TM(r4)
3127 std r2, PACATMSCRATCH(r13) /* Save TOC */
3129 /* Clear the MSR RI since r1, r13 are all going to be foobar. */
3133 /* Load GPRs r0-r28 */
3136 ld reg, VCPU_GPRS_TM(reg)(r31)
3140 mtspr SPRN_DSCR, r29
3143 /* Load final GPRs */
3144 ld 29, VCPU_GPRS_TM(29)(r31)
3145 ld 30, VCPU_GPRS_TM(30)(r31)
3146 ld 31, VCPU_GPRS_TM(31)(r31)
3148 /* TM checkpointed state is now setup. All GPRs are now volatile. */
3151 /* Now let's get back the state we need. */
3154 ld r29, HSTATE_DSCR(r13)
3155 mtspr SPRN_DSCR, r29
3156 ld r4, HSTATE_KVM_VCPU(r13)
3157 ld r1, HSTATE_HOST_R1(r13)
3158 ld r2, PACATMSCRATCH(r13)
3160 /* Set the MSR RI since we have our registers back. */
3164 ld r0, PPC_LR_STKOFF(r1)
3170 * We come here if we get any exception or interrupt while we are
3171 * executing host real mode code while in guest MMU context.
3172 * For now just spin, but we should do something better.
3174 kvmppc_bad_host_intr:
3178 * This mimics the MSR transition on IRQ delivery. The new guest MSR is taken
3179 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
3180 * r11 has the guest MSR value (in/out)
3181 * r9 has a vcpu pointer (in)
3182 * r0 is used as a scratch register
3184 kvmppc_msr_interrupt:
3185 rldicl r0, r11, 64 - MSR_TS_S_LG, 62
3186 cmpwi r0, 2 /* Check if we are in transactional state.. */
3187 ld r11, VCPU_INTR_MSR(r9)
3189 /* ... if transactional, change to suspended */
3191 1: rldimi r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
3195 * This works around a hardware bug on POWER8E processors, where
3196 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
3197 * performance monitor interrupt. Instead, when we need to have
3198 * an interrupt pending, we have to arrange for a counter to overflow.
3202 mtspr SPRN_MMCR2, r3
3203 lis r3, (MMCR0_PMXE | MMCR0_FCECE)@h
3204 ori r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
3205 mtspr SPRN_MMCR0, r3
3212 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
3214 * Start timing an activity
3215 * r3 = pointer to time accumulation struct, r4 = vcpu
3218 ld r5, HSTATE_KVM_VCORE(r13)
3219 lbz r6, VCORE_IN_GUEST(r5)
3221 beq 5f /* if in guest, need to */
3222 ld r6, VCORE_TB_OFFSET(r5) /* subtract timebase offset */
3225 std r3, VCPU_CUR_ACTIVITY(r4)
3226 std r5, VCPU_ACTIVITY_START(r4)
3230 * Accumulate time to one activity and start another.
3231 * r3 = pointer to new time accumulation struct, r4 = vcpu
3233 kvmhv_accumulate_time:
3234 ld r5, HSTATE_KVM_VCORE(r13)
3235 lbz r8, VCORE_IN_GUEST(r5)
3237 beq 4f /* if in guest, need to */
3238 ld r8, VCORE_TB_OFFSET(r5) /* subtract timebase offset */
3239 4: ld r5, VCPU_CUR_ACTIVITY(r4)
3240 ld r6, VCPU_ACTIVITY_START(r4)
3241 std r3, VCPU_CUR_ACTIVITY(r4)
3244 std r7, VCPU_ACTIVITY_START(r4)
3248 ld r8, TAS_SEQCOUNT(r5)
3251 std r8, TAS_SEQCOUNT(r5)
3253 ld r7, TAS_TOTAL(r5)
3255 std r7, TAS_TOTAL(r5)
3261 3: std r3, TAS_MIN(r5)
3267 std r8, TAS_SEQCOUNT(r5)