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/kvm_asm.h>
25 #include <asm/ptrace.h>
26 #include <asm/hvcall.h>
27 #include <asm/asm-offsets.h>
28 #include <asm/exception-64s.h>
29 #include <asm/kvm_book3s_asm.h>
30 #include <asm/book3s/64/mmu-hash.h>
33 #include <asm/xive-regs.h>
35 /* Sign-extend HDEC if not on POWER9 */
36 #define EXTEND_HDEC(reg) \
39 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
41 #define VCPU_GPRS_TM(reg) (((reg) * ULONG_SIZE) + VCPU_GPR_TM)
43 /* Values in HSTATE_NAPPING(r13) */
44 #define NAPPING_CEDE 1
45 #define NAPPING_NOVCPU 2
47 /* Stack frame offsets for kvmppc_hv_entry */
49 #define STACK_SLOT_TRAP (SFS-4)
50 #define STACK_SLOT_TID (SFS-16)
51 #define STACK_SLOT_PSSCR (SFS-24)
52 #define STACK_SLOT_PID (SFS-32)
53 #define STACK_SLOT_IAMR (SFS-40)
54 #define STACK_SLOT_CIABR (SFS-48)
55 #define STACK_SLOT_DAWR (SFS-56)
56 #define STACK_SLOT_DAWRX (SFS-64)
57 #define STACK_SLOT_HFSCR (SFS-72)
60 * Call kvmppc_hv_entry in real mode.
61 * Must be called with interrupts hard-disabled.
65 * LR = return address to continue at after eventually re-enabling MMU
67 _GLOBAL_TOC(kvmppc_hv_entry_trampoline)
69 std r0, PPC_LR_STKOFF(r1)
72 std r10, HSTATE_HOST_MSR(r13)
73 LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
78 mtmsrd r0,1 /* clear RI in MSR */
84 ld r4, HSTATE_KVM_VCPU(r13)
87 /* Back from guest - restore host state and return to caller */
90 /* Restore host DABR and DABRX */
91 ld r5,HSTATE_DABR(r13)
95 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
98 ld r3,PACA_SPRG_VDSO(r13)
99 mtspr SPRN_SPRG_VDSO_WRITE,r3
101 /* Reload the host's PMU registers */
102 ld r3, PACALPPACAPTR(r13) /* is the host using the PMU? */
103 lbz r4, LPPACA_PMCINUSE(r3)
105 beq 23f /* skip if not */
107 ld r3, HSTATE_MMCR0(r13)
108 andi. r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
111 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
112 lwz r3, HSTATE_PMC1(r13)
113 lwz r4, HSTATE_PMC2(r13)
114 lwz r5, HSTATE_PMC3(r13)
115 lwz r6, HSTATE_PMC4(r13)
116 lwz r8, HSTATE_PMC5(r13)
117 lwz r9, HSTATE_PMC6(r13)
124 ld r3, HSTATE_MMCR0(r13)
125 ld r4, HSTATE_MMCR1(r13)
126 ld r5, HSTATE_MMCRA(r13)
127 ld r6, HSTATE_SIAR(r13)
128 ld r7, HSTATE_SDAR(r13)
134 ld r8, HSTATE_MMCR2(r13)
135 ld r9, HSTATE_SIER(r13)
138 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
144 * Reload DEC. HDEC interrupts were disabled when
145 * we reloaded the host's LPCR value.
147 ld r3, HSTATE_DECEXP(r13)
152 /* hwthread_req may have got set by cede or no vcpu, so clear it */
154 stb r0, HSTATE_HWTHREAD_REQ(r13)
157 * For external interrupts we need to call the Linux
158 * handler to process the interrupt. We do that by jumping
159 * to absolute address 0x500 for external interrupts.
160 * The [h]rfid at the end of the handler will return to
161 * the book3s_hv_interrupts.S code. For other interrupts
162 * we do the rfid to get back to the book3s_hv_interrupts.S
165 ld r8, 112+PPC_LR_STKOFF(r1)
167 ld r7, HSTATE_HOST_MSR(r13)
169 /* Return the trap number on this thread as the return value */
173 * If we came back from the guest via a relocation-on interrupt,
174 * we will be in virtual mode at this point, which makes it a
175 * little easier to get back to the caller.
178 andi. r0, r0, MSR_IR /* in real mode? */
181 /* RFI into the highmem handler */
185 mtmsrd r6, 1 /* Clear RI in MSR */
190 /* Virtual-mode return */
195 kvmppc_primary_no_guest:
196 /* We handle this much like a ceded vcpu */
197 /* put the HDEC into the DEC, since HDEC interrupts don't wake us */
198 /* HDEC may be larger than DEC for arch >= v3.00, but since the */
199 /* HDEC value came from DEC in the first place, it will fit */
203 * Make sure the primary has finished the MMU switch.
204 * We should never get here on a secondary thread, but
205 * check it for robustness' sake.
207 ld r5, HSTATE_KVM_VCORE(r13)
208 65: lbz r0, VCORE_IN_GUEST(r5)
215 /* set our bit in napping_threads */
216 ld r5, HSTATE_KVM_VCORE(r13)
217 lbz r7, HSTATE_PTID(r13)
220 addi r6, r5, VCORE_NAPPING_THREADS
225 /* order napping_threads update vs testing entry_exit_map */
228 lwz r7, VCORE_ENTRY_EXIT(r5)
230 bge kvm_novcpu_exit /* another thread already exiting */
231 li r3, NAPPING_NOVCPU
232 stb r3, HSTATE_NAPPING(r13)
234 li r3, 0 /* Don't wake on privileged (OS) doorbell */
239 * Entered from kvm_start_guest if kvm_hstate.napping is set
245 ld r1, HSTATE_HOST_R1(r13)
246 ld r5, HSTATE_KVM_VCORE(r13)
248 stb r0, HSTATE_NAPPING(r13)
250 /* check the wake reason */
251 bl kvmppc_check_wake_reason
254 * Restore volatile registers since we could have called
255 * a C routine in kvmppc_check_wake_reason.
258 ld r5, HSTATE_KVM_VCORE(r13)
260 /* see if any other thread is already exiting */
261 lwz r0, VCORE_ENTRY_EXIT(r5)
265 /* clear our bit in napping_threads */
266 lbz r7, HSTATE_PTID(r13)
269 addi r6, r5, VCORE_NAPPING_THREADS
275 /* See if the wake reason means we need to exit */
279 /* See if our timeslice has expired (HDEC is negative) */
282 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
286 /* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
287 ld r4, HSTATE_KVM_VCPU(r13)
289 beq kvmppc_primary_no_guest
291 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
292 addi r3, r4, VCPU_TB_RMENTRY
293 bl kvmhv_start_timing
298 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
299 ld r4, HSTATE_KVM_VCPU(r13)
302 addi r3, r4, VCPU_TB_RMEXIT
303 bl kvmhv_accumulate_time
306 stw r12, STACK_SLOT_TRAP(r1)
307 bl kvmhv_commence_exit
309 lwz r12, STACK_SLOT_TRAP(r1)
310 b kvmhv_switch_to_host
313 * We come in here when wakened from nap mode.
314 * Relocation is off and most register values are lost.
315 * r13 points to the PACA.
316 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
318 .globl kvm_start_guest
320 /* Set runlatch bit the minute you wake up from nap */
326 * Could avoid this and pass it through in r3. For now,
327 * code expects it to be in SRR1.
333 li r0,KVM_HWTHREAD_IN_KVM
334 stb r0,HSTATE_HWTHREAD_STATE(r13)
336 /* NV GPR values from power7_idle() will no longer be valid */
338 stb r0,PACA_NAPSTATELOST(r13)
340 /* were we napping due to cede? */
341 lbz r0,HSTATE_NAPPING(r13)
342 cmpwi r0,NAPPING_CEDE
344 cmpwi r0,NAPPING_NOVCPU
345 beq kvm_novcpu_wakeup
347 ld r1,PACAEMERGSP(r13)
348 subi r1,r1,STACK_FRAME_OVERHEAD
351 * We weren't napping due to cede, so this must be a secondary
352 * thread being woken up to run a guest, or being woken up due
353 * to a stray IPI. (Or due to some machine check or hypervisor
354 * maintenance interrupt while the core is in KVM.)
357 /* Check the wake reason in SRR1 to see why we got here */
358 bl kvmppc_check_wake_reason
360 * kvmppc_check_wake_reason could invoke a C routine, but we
361 * have no volatile registers to restore when we return.
367 /* get vcore pointer, NULL if we have nothing to run */
368 ld r5,HSTATE_KVM_VCORE(r13)
370 /* if we have no vcore to run, go back to sleep */
373 kvm_secondary_got_guest:
375 /* Set HSTATE_DSCR(r13) to something sensible */
376 ld r6, PACA_DSCR_DEFAULT(r13)
377 std r6, HSTATE_DSCR(r13)
379 /* On thread 0 of a subcore, set HDEC to max */
380 lbz r4, HSTATE_PTID(r13)
383 LOAD_REG_ADDR(r6, decrementer_max)
386 /* and set per-LPAR registers, if doing dynamic micro-threading */
387 ld r6, HSTATE_SPLIT_MODE(r13)
390 ld r0, KVM_SPLIT_RPR(r6)
392 ld r0, KVM_SPLIT_PMMAR(r6)
394 ld r0, KVM_SPLIT_LDBAR(r6)
398 /* Order load of vcpu after load of vcore */
400 ld r4, HSTATE_KVM_VCPU(r13)
403 /* Back from the guest, go back to nap */
404 /* Clear our vcpu and vcore pointers so we don't come back in early */
406 std r0, HSTATE_KVM_VCPU(r13)
408 * Once we clear HSTATE_KVM_VCORE(r13), the code in
409 * kvmppc_run_core() is going to assume that all our vcpu
410 * state is visible in memory. This lwsync makes sure
414 std r0, HSTATE_KVM_VCORE(r13)
417 * All secondaries exiting guest will fall through this path.
418 * Before proceeding, just check for HMI interrupt and
419 * invoke opal hmi handler. By now we are sure that the
420 * primary thread on this core/subcore has already made partition
421 * switch/TB resync and we are good to call opal hmi handler.
423 cmpwi r12, BOOK3S_INTERRUPT_HMI
426 li r3,0 /* NULL argument */
427 bl hmi_exception_realmode
429 * At this point we have finished executing in the guest.
430 * We need to wait for hwthread_req to become zero, since
431 * we may not turn on the MMU while hwthread_req is non-zero.
432 * While waiting we also need to check if we get given a vcpu to run.
435 lbz r3, HSTATE_HWTHREAD_REQ(r13)
439 li r0, KVM_HWTHREAD_IN_KERNEL
440 stb r0, HSTATE_HWTHREAD_STATE(r13)
441 /* need to recheck hwthread_req after a barrier, to avoid race */
443 lbz r3, HSTATE_HWTHREAD_REQ(r13)
447 * We jump to pnv_wakeup_loss, which will return to the caller
448 * of power7_nap in the powernv cpu offline loop. The value we
449 * put in r3 becomes the return value for power7_nap. pnv_wakeup_loss
450 * requires SRR1 in r12.
454 rlwimi r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
461 ld r5, HSTATE_KVM_VCORE(r13)
464 ld r3, HSTATE_SPLIT_MODE(r13)
467 lbz r0, KVM_SPLIT_DO_NAP(r3)
473 b kvm_secondary_got_guest
475 54: li r0, KVM_HWTHREAD_IN_KVM
476 stb r0, HSTATE_HWTHREAD_STATE(r13)
480 * Here the primary thread is trying to return the core to
481 * whole-core mode, so we need to nap.
485 * When secondaries are napping in kvm_unsplit_nap() with
486 * hwthread_req = 1, HMI goes ignored even though subcores are
487 * already exited the guest. Hence HMI keeps waking up secondaries
488 * from nap in a loop and secondaries always go back to nap since
489 * no vcore is assigned to them. This makes impossible for primary
490 * thread to get hold of secondary threads resulting into a soft
491 * lockup in KVM path.
493 * Let us check if HMI is pending and handle it before we go to nap.
495 cmpwi r12, BOOK3S_INTERRUPT_HMI
497 li r3, 0 /* NULL argument */
498 bl hmi_exception_realmode
501 * Ensure that secondary doesn't nap when it has
502 * its vcore pointer set.
504 sync /* matches smp_mb() before setting split_info.do_nap */
505 ld r0, HSTATE_KVM_VCORE(r13)
508 /* clear any pending message */
510 lis r6, (PPC_DBELL_SERVER << (63-36))@h
512 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
513 /* Set kvm_split_mode.napped[tid] = 1 */
514 ld r3, HSTATE_SPLIT_MODE(r13)
516 lhz r4, PACAPACAINDEX(r13)
517 clrldi r4, r4, 61 /* micro-threading => P8 => 8 threads/core */
518 addi r4, r4, KVM_SPLIT_NAPPED
520 /* Check the do_nap flag again after setting napped[] */
522 lbz r0, KVM_SPLIT_DO_NAP(r3)
525 li r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
527 rlwimi r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
534 /******************************************************************************
538 *****************************************************************************/
540 .global kvmppc_hv_entry
545 * R4 = vcpu pointer (or NULL)
550 * all other volatile GPRS = free
551 * Does not preserve non-volatile GPRs or CR fields
554 std r0, PPC_LR_STKOFF(r1)
557 /* Save R1 in the PACA */
558 std r1, HSTATE_HOST_R1(r13)
560 li r6, KVM_GUEST_MODE_HOST_HV
561 stb r6, HSTATE_IN_GUEST(r13)
563 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
564 /* Store initial timestamp */
567 addi r3, r4, VCPU_TB_RMENTRY
568 bl kvmhv_start_timing
572 /* Use cr7 as an indication of radix mode */
573 ld r5, HSTATE_KVM_VCORE(r13)
574 ld r9, VCORE_KVM(r5) /* pointer to struct kvm */
575 lbz r0, KVM_RADIX(r9)
578 /* Clear out SLB if hash */
586 * POWER7/POWER8 host -> guest partition switch code.
587 * We don't have to lock against concurrent tlbies,
588 * but we do have to coordinate across hardware threads.
590 /* Set bit in entry map iff exit map is zero. */
592 lbz r6, HSTATE_PTID(r13)
594 addi r8, r5, VCORE_ENTRY_EXIT
596 cmpwi r3, 0x100 /* any threads starting to exit? */
597 bge secondary_too_late /* if so we're too late to the party */
602 /* Primary thread switches to guest partition. */
608 li r0,LPID_RSVD /* switch to reserved LPID */
611 mtspr SPRN_SDR1,r6 /* switch to partition page table */
612 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
616 /* See if we need to flush the TLB */
617 lhz r6,PACAPACAINDEX(r13) /* test_bit(cpu, need_tlb_flush) */
620 * On POWER9, individual threads can come in here, but the
621 * TLB is shared between the 4 threads in a core, hence
622 * invalidating on one thread invalidates for all.
623 * Thus we make all 4 threads use the same bit here.
626 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
627 clrldi r7,r6,64-6 /* extract bit number (6 bits) */
628 srdi r6,r6,6 /* doubleword number */
629 sldi r6,r6,3 /* address offset */
631 addi r6,r6,KVM_NEED_FLUSH /* dword in kvm->arch.need_tlb_flush */
637 /* Flush the TLB of any entries for this LPID */
638 lwz r0,KVM_TLB_SETS(r9)
640 li r7,0x800 /* IS field = 0b10 */
642 li r0,0 /* RS for P9 version of tlbiel */
644 28: tlbiel r7 /* On P9, rs=0, RIC=0, PRS=0, R=0 */
648 29: PPC_TLBIEL(7,0,2,1,1) /* for radix, RIC=2, PRS=1, R=1 */
652 23: ldarx r7,0,r6 /* clear the bit after TLB flushed */
657 /* Add timebase offset onto timebase */
658 22: ld r8,VCORE_TB_OFFSET(r5)
661 mftb r6 /* current host timebase */
663 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
664 mftb r7 /* check if lower 24 bits overflowed */
669 addis r8,r8,0x100 /* if so, increment upper 40 bits */
672 /* Load guest PCR value to select appropriate compat mode */
673 37: ld r7, VCORE_PCR(r5)
680 /* DPDES and VTB are shared between threads */
681 ld r8, VCORE_DPDES(r5)
685 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
687 /* Mark the subcore state as inside guest */
688 bl kvmppc_subcore_enter_guest
690 ld r5, HSTATE_KVM_VCORE(r13)
691 ld r4, HSTATE_KVM_VCPU(r13)
693 stb r0,VCORE_IN_GUEST(r5) /* signal secondaries to continue */
695 /* Do we have a guest vcpu to run? */
697 beq kvmppc_primary_no_guest
700 /* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
701 lwz r5,VCPU_SLB_MAX(r4)
706 1: ld r8,VCPU_SLB_E(r6)
709 addi r6,r6,VCPU_SLB_SIZE
712 /* Increment yield count if they have a VPA */
716 li r6, LPPACA_YIELDCOUNT
721 stb r6, VCPU_VPA_DIRTY(r4)
724 /* Save purr/spurr */
727 std r5,HSTATE_PURR(r13)
728 std r6,HSTATE_SPURR(r13)
734 /* Save host values of some registers */
740 std r5, STACK_SLOT_TID(r1)
741 std r6, STACK_SLOT_PSSCR(r1)
742 std r7, STACK_SLOT_PID(r1)
743 std r8, STACK_SLOT_IAMR(r1)
745 std r5, STACK_SLOT_HFSCR(r1)
746 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
751 std r5, STACK_SLOT_CIABR(r1)
752 std r6, STACK_SLOT_DAWR(r1)
753 std r7, STACK_SLOT_DAWRX(r1)
754 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
757 /* Set partition DABR */
758 /* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
759 lwz r5,VCPU_DABRX(r4)
764 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
766 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
769 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
772 END_FTR_SECTION_IFSET(CPU_FTR_TM)
775 /* Load guest PMU registers */
776 /* R4 is live here (vcpu pointer) */
778 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
779 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
783 andi. r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
786 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
787 lwz r3, VCPU_PMC(r4) /* always load up guest PMU registers */
788 lwz r5, VCPU_PMC + 4(r4) /* to prevent information leak */
789 lwz r6, VCPU_PMC + 8(r4)
790 lwz r7, VCPU_PMC + 12(r4)
791 lwz r8, VCPU_PMC + 16(r4)
792 lwz r9, VCPU_PMC + 20(r4)
800 ld r5, VCPU_MMCR + 8(r4)
801 ld r6, VCPU_MMCR + 16(r4)
809 ld r5, VCPU_MMCR + 24(r4)
813 BEGIN_FTR_SECTION_NESTED(96)
814 lwz r7, VCPU_PMC + 24(r4)
815 lwz r8, VCPU_PMC + 28(r4)
816 ld r9, VCPU_MMCR + 32(r4)
820 END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
821 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
825 /* Load up FP, VMX and VSX registers */
828 ld r14, VCPU_GPR(R14)(r4)
829 ld r15, VCPU_GPR(R15)(r4)
830 ld r16, VCPU_GPR(R16)(r4)
831 ld r17, VCPU_GPR(R17)(r4)
832 ld r18, VCPU_GPR(R18)(r4)
833 ld r19, VCPU_GPR(R19)(r4)
834 ld r20, VCPU_GPR(R20)(r4)
835 ld r21, VCPU_GPR(R21)(r4)
836 ld r22, VCPU_GPR(R22)(r4)
837 ld r23, VCPU_GPR(R23)(r4)
838 ld r24, VCPU_GPR(R24)(r4)
839 ld r25, VCPU_GPR(R25)(r4)
840 ld r26, VCPU_GPR(R26)(r4)
841 ld r27, VCPU_GPR(R27)(r4)
842 ld r28, VCPU_GPR(R28)(r4)
843 ld r29, VCPU_GPR(R29)(r4)
844 ld r30, VCPU_GPR(R30)(r4)
845 ld r31, VCPU_GPR(R31)(r4)
847 /* Switch DSCR to guest value */
852 /* Skip next section on POWER7 */
854 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
855 /* Load up POWER8-specific registers */
857 lwz r6, VCPU_PSPB(r4)
863 ld r6, VCPU_DAWRX(r4)
864 ld r7, VCPU_CIABR(r4)
871 ld r8, VCPU_EBBHR(r4)
874 ld r5, VCPU_EBBRR(r4)
875 ld r6, VCPU_BESCR(r4)
876 lwz r7, VCPU_GUEST_PID(r4)
884 END_FTR_SECTION_IFSET(CPU_FTR_POWER9_DD1)
886 /* POWER8-only registers */
887 ld r5, VCPU_TCSCR(r4)
889 ld r7, VCPU_CSIGR(r4)
896 /* POWER9-only registers */
898 ld r6, VCPU_PSSCR(r4)
899 oris r6, r6, PSSCR_EC@h /* This makes stop trap to HV */
900 ld r7, VCPU_HFSCR(r4)
904 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
908 * Set the decrementer to the guest decrementer.
910 ld r8,VCPU_DEC_EXPIRES(r4)
911 /* r8 is a host timebase value here, convert to guest TB */
912 ld r5,HSTATE_KVM_VCORE(r13)
913 ld r6,VCORE_TB_OFFSET(r5)
920 ld r5, VCPU_SPRG0(r4)
921 ld r6, VCPU_SPRG1(r4)
922 ld r7, VCPU_SPRG2(r4)
923 ld r8, VCPU_SPRG3(r4)
929 /* Load up DAR and DSISR */
931 lwz r6, VCPU_DSISR(r4)
935 /* Restore AMR and UAMOR, set AMOR to all 1s */
943 /* Restore state of CTRL run bit; assume 1 on entry */
951 /* Secondary threads wait for primary to have done partition switch */
952 ld r5, HSTATE_KVM_VCORE(r13)
953 lbz r6, HSTATE_PTID(r13)
956 lbz r0, VCORE_IN_GUEST(r5)
960 20: lwz r3, VCORE_ENTRY_EXIT(r5)
963 lbz r0, VCORE_IN_GUEST(r5)
973 /* Check if HDEC expires soon */
976 cmpdi r3, 512 /* 1 microsecond */
979 #ifdef CONFIG_KVM_XICS
980 /* We are entering the guest on that thread, push VCPU to XIVE */
981 ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
984 ld r11, VCPU_XIVE_SAVED_STATE(r4)
988 lwz r11, VCPU_XIVE_CAM_WORD(r4)
989 li r9, TM_QW1_OS + TM_WORD2
992 stw r9, VCPU_XIVE_PUSHED(r4)
995 #endif /* CONFIG_KVM_XICS */
997 deliver_guest_interrupt:
1004 kvmppc_cede_reentry: /* r4 = vcpu, r13 = paca */
1006 ld r11, VCPU_MSR(r4)
1007 ld r6, VCPU_SRR0(r4)
1008 ld r7, VCPU_SRR1(r4)
1012 /* r11 = vcpu->arch.msr & ~MSR_HV */
1013 rldicl r11, r11, 63 - MSR_HV_LG, 1
1014 rotldi r11, r11, 1 + MSR_HV_LG
1015 ori r11, r11, MSR_ME
1017 /* Check if we can deliver an external or decrementer interrupt now */
1018 ld r0, VCPU_PENDING_EXC(r4)
1019 rldicl r0, r0, 64 - BOOK3S_IRQPRIO_EXTERNAL_LEVEL, 63
1021 andi. r8, r11, MSR_EE
1023 /* Insert EXTERNAL_LEVEL bit into LPCR at the MER bit position */
1024 rldimi r8, r0, LPCR_MER_SH, 63 - LPCR_MER_SH
1028 li r0, BOOK3S_INTERRUPT_EXTERNAL
1032 /* On POWER9 check whether the guest has large decrementer enabled */
1033 andis. r8, r8, LPCR_LD@h
1035 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1038 li r0, BOOK3S_INTERRUPT_DECREMENTER
1041 12: mtspr SPRN_SRR0, r10
1043 mtspr SPRN_SRR1, r11
1045 bl kvmppc_msr_interrupt
1049 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1050 /* On POWER9, check for pending doorbell requests */
1051 lbz r0, VCPU_DBELL_REQ(r4)
1053 beq fast_guest_return
1054 ld r5, HSTATE_KVM_VCORE(r13)
1055 /* Set DPDES register so the CPU will take a doorbell interrupt */
1057 mtspr SPRN_DPDES, r0
1058 std r0, VCORE_DPDES(r5)
1059 /* Make sure other cpus see vcore->dpdes set before dbell req clear */
1061 /* Clear the pending doorbell request */
1063 stb r0, VCPU_DBELL_REQ(r4)
1068 * R10: value for HSRR0
1069 * R11: value for HSRR1
1074 stb r0,VCPU_CEDED(r4) /* cancel cede */
1075 mtspr SPRN_HSRR0,r10
1076 mtspr SPRN_HSRR1,r11
1078 /* Activate guest mode, so faults get handled by KVM */
1079 li r9, KVM_GUEST_MODE_GUEST_HV
1080 stb r9, HSTATE_IN_GUEST(r13)
1082 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1083 /* Accumulate timing */
1084 addi r3, r4, VCPU_TB_GUEST
1085 bl kvmhv_accumulate_time
1091 ld r5, VCPU_CFAR(r4)
1093 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1096 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1103 ld r1, VCPU_GPR(R1)(r4)
1104 ld r2, VCPU_GPR(R2)(r4)
1105 ld r3, VCPU_GPR(R3)(r4)
1106 ld r5, VCPU_GPR(R5)(r4)
1107 ld r6, VCPU_GPR(R6)(r4)
1108 ld r7, VCPU_GPR(R7)(r4)
1109 ld r8, VCPU_GPR(R8)(r4)
1110 ld r9, VCPU_GPR(R9)(r4)
1111 ld r10, VCPU_GPR(R10)(r4)
1112 ld r11, VCPU_GPR(R11)(r4)
1113 ld r12, VCPU_GPR(R12)(r4)
1114 ld r13, VCPU_GPR(R13)(r4)
1118 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1120 /* Move canary into DSISR to check for later */
1123 mtspr SPRN_HDSISR, r0
1124 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1126 ld r0, VCPU_GPR(R0)(r4)
1127 ld r4, VCPU_GPR(R4)(r4)
1136 stw r12, VCPU_TRAP(r4)
1137 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1138 addi r3, r4, VCPU_TB_RMEXIT
1139 bl kvmhv_accumulate_time
1141 11: b kvmhv_switch_to_host
1148 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
1149 12: stw r12, VCPU_TRAP(r4)
1151 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1152 addi r3, r4, VCPU_TB_RMEXIT
1153 bl kvmhv_accumulate_time
1157 /******************************************************************************
1161 *****************************************************************************/
1164 * We come here from the first-level interrupt handlers.
1166 .globl kvmppc_interrupt_hv
1167 kvmppc_interrupt_hv:
1169 * Register contents:
1170 * R12 = (guest CR << 32) | interrupt vector
1172 * guest R12 saved in shadow VCPU SCRATCH0
1173 * guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
1174 * guest R13 saved in SPRN_SCRATCH0
1176 std r9, HSTATE_SCRATCH2(r13)
1177 lbz r9, HSTATE_IN_GUEST(r13)
1178 cmpwi r9, KVM_GUEST_MODE_HOST_HV
1179 beq kvmppc_bad_host_intr
1180 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1181 cmpwi r9, KVM_GUEST_MODE_GUEST
1182 ld r9, HSTATE_SCRATCH2(r13)
1183 beq kvmppc_interrupt_pr
1185 /* We're now back in the host but in guest MMU context */
1186 li r9, KVM_GUEST_MODE_HOST_HV
1187 stb r9, HSTATE_IN_GUEST(r13)
1189 ld r9, HSTATE_KVM_VCPU(r13)
1191 /* Save registers */
1193 std r0, VCPU_GPR(R0)(r9)
1194 std r1, VCPU_GPR(R1)(r9)
1195 std r2, VCPU_GPR(R2)(r9)
1196 std r3, VCPU_GPR(R3)(r9)
1197 std r4, VCPU_GPR(R4)(r9)
1198 std r5, VCPU_GPR(R5)(r9)
1199 std r6, VCPU_GPR(R6)(r9)
1200 std r7, VCPU_GPR(R7)(r9)
1201 std r8, VCPU_GPR(R8)(r9)
1202 ld r0, HSTATE_SCRATCH2(r13)
1203 std r0, VCPU_GPR(R9)(r9)
1204 std r10, VCPU_GPR(R10)(r9)
1205 std r11, VCPU_GPR(R11)(r9)
1206 ld r3, HSTATE_SCRATCH0(r13)
1207 std r3, VCPU_GPR(R12)(r9)
1208 /* CR is in the high half of r12 */
1212 ld r3, HSTATE_CFAR(r13)
1213 std r3, VCPU_CFAR(r9)
1214 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1216 ld r4, HSTATE_PPR(r13)
1217 std r4, VCPU_PPR(r9)
1218 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1220 /* Restore R1/R2 so we can handle faults */
1221 ld r1, HSTATE_HOST_R1(r13)
1224 mfspr r10, SPRN_SRR0
1225 mfspr r11, SPRN_SRR1
1226 std r10, VCPU_SRR0(r9)
1227 std r11, VCPU_SRR1(r9)
1228 /* trap is in the low half of r12, clear CR from the high half */
1230 andi. r0, r12, 2 /* need to read HSRR0/1? */
1232 mfspr r10, SPRN_HSRR0
1233 mfspr r11, SPRN_HSRR1
1235 1: std r10, VCPU_PC(r9)
1236 std r11, VCPU_MSR(r9)
1240 std r3, VCPU_GPR(R13)(r9)
1243 stw r12,VCPU_TRAP(r9)
1246 * Now that we have saved away SRR0/1 and HSRR0/1,
1247 * interrupts are recoverable in principle, so set MSR_RI.
1248 * This becomes important for relocation-on interrupts from
1249 * the guest, which we can get in radix mode on POWER9.
1254 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1255 addi r3, r9, VCPU_TB_RMINTR
1257 bl kvmhv_accumulate_time
1258 ld r5, VCPU_GPR(R5)(r9)
1259 ld r6, VCPU_GPR(R6)(r9)
1260 ld r7, VCPU_GPR(R7)(r9)
1261 ld r8, VCPU_GPR(R8)(r9)
1264 /* Save HEIR (HV emulation assist reg) in emul_inst
1265 if this is an HEI (HV emulation interrupt, e40) */
1266 li r3,KVM_INST_FETCH_FAILED
1267 stw r3,VCPU_LAST_INST(r9)
1268 cmpwi r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1271 11: stw r3,VCPU_HEIR(r9)
1273 /* these are volatile across C function calls */
1274 #ifdef CONFIG_RELOCATABLE
1275 ld r3, HSTATE_SCRATCH1(r13)
1281 std r3, VCPU_CTR(r9)
1282 std r4, VCPU_XER(r9)
1284 /* If this is a page table miss then see if it's theirs or ours */
1285 cmpwi r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1287 cmpwi r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1290 /* See if this is a leftover HDEC interrupt */
1291 cmpwi r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1297 bge fast_guest_return
1299 /* See if this is an hcall we can handle in real mode */
1300 cmpwi r12,BOOK3S_INTERRUPT_SYSCALL
1301 beq hcall_try_real_mode
1303 /* Hypervisor doorbell - exit only if host IPI flag set */
1304 cmpwi r12, BOOK3S_INTERRUPT_H_DOORBELL
1309 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1310 lbz r0, HSTATE_HOST_IPI(r13)
1315 /* If it's a hypervisor facility unavailable interrupt, save HFSCR */
1316 cmpwi r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
1318 mfspr r3, SPRN_HFSCR
1319 std r3, VCPU_HFSCR(r9)
1322 /* External interrupt ? */
1323 cmpwi r12, BOOK3S_INTERRUPT_EXTERNAL
1324 bne+ guest_exit_cont
1326 /* External interrupt, first check for host_ipi. If this is
1327 * set, we know the host wants us out so let's do it now
1332 * Restore the active volatile registers after returning from
1335 ld r9, HSTATE_KVM_VCPU(r13)
1336 li r12, BOOK3S_INTERRUPT_EXTERNAL
1339 * kvmppc_read_intr return codes:
1341 * Exit to host (r3 > 0)
1342 * 1 An interrupt is pending that needs to be handled by the host
1343 * Exit guest and return to host by branching to guest_exit_cont
1345 * 2 Passthrough that needs completion in the host
1346 * Exit guest and return to host by branching to guest_exit_cont
1347 * However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
1348 * to indicate to the host to complete handling the interrupt
1350 * Before returning to guest, we check if any CPU is heading out
1351 * to the host and if so, we head out also. If no CPUs are heading
1352 * check return values <= 0.
1354 * Return to guest (r3 <= 0)
1355 * 0 No external interrupt is pending
1356 * -1 A guest wakeup IPI (which has now been cleared)
1357 * In either case, we return to guest to deliver any pending
1360 * -2 A PCI passthrough external interrupt was handled
1361 * (interrupt was delivered directly to guest)
1362 * Return to guest to deliver any pending guest interrupts.
1368 /* Return code = 2 */
1369 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
1370 stw r12, VCPU_TRAP(r9)
1373 1: /* Return code <= 1 */
1377 /* Return code <= 0 */
1378 4: ld r5, HSTATE_KVM_VCORE(r13)
1379 lwz r0, VCORE_ENTRY_EXIT(r5)
1382 blt deliver_guest_interrupt
1384 guest_exit_cont: /* r9 = vcpu, r12 = trap, r13 = paca */
1385 #ifdef CONFIG_KVM_XICS
1386 /* We are exiting, pull the VP from the XIVE */
1387 lwz r0, VCPU_XIVE_PUSHED(r9)
1390 li r7, TM_SPC_PULL_OS_CTX
1393 andi. r0, r0, MSR_IR /* in real mode? */
1395 ld r10, HSTATE_XIVE_TIMA_VIRT(r13)
1398 /* First load to pull the context, we ignore the value */
1401 /* Second load to recover the context state (Words 0 and 1) */
1404 2: ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
1407 /* First load to pull the context, we ignore the value */
1410 /* Second load to recover the context state (Words 0 and 1) */
1412 3: std r11, VCPU_XIVE_SAVED_STATE(r9)
1413 /* Fixup some of the state for the next load */
1416 stw r10, VCPU_XIVE_PUSHED(r9)
1417 stb r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
1418 stb r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
1421 #endif /* CONFIG_KVM_XICS */
1422 /* Save more register state */
1425 std r6, VCPU_DAR(r9)
1426 stw r7, VCPU_DSISR(r9)
1427 /* don't overwrite fault_dar/fault_dsisr if HDSI */
1428 cmpwi r12,BOOK3S_INTERRUPT_H_DATA_STORAGE
1430 std r6, VCPU_FAULT_DAR(r9)
1431 stw r7, VCPU_FAULT_DSISR(r9)
1433 /* See if it is a machine check */
1434 cmpwi r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1435 beq machine_check_realmode
1437 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1438 addi r3, r9, VCPU_TB_RMEXIT
1440 bl kvmhv_accumulate_time
1444 /* Increment exit count, poke other threads to exit */
1445 bl kvmhv_commence_exit
1447 ld r9, HSTATE_KVM_VCPU(r13)
1448 lwz r12, VCPU_TRAP(r9)
1450 /* Stop others sending VCPU interrupts to this physical CPU */
1452 stw r0, VCPU_CPU(r9)
1453 stw r0, VCPU_THREAD_CPU(r9)
1455 /* Save guest CTRL register, set runlatch to 1 */
1457 stw r6,VCPU_CTRL(r9)
1463 /* Check if we are running hash or radix and store it in cr2 */
1465 lbz r0, KVM_RADIX(r5)
1468 /* Read the guest SLB and save it away */
1470 bne cr2, 3f /* for radix, save 0 entries */
1471 lwz r0,VCPU_SLB_NR(r9) /* number of entries in SLB */
1476 andis. r0,r8,SLB_ESID_V@h
1478 add r8,r8,r6 /* put index in */
1480 std r8,VCPU_SLB_E(r7)
1481 std r3,VCPU_SLB_V(r7)
1482 addi r7,r7,VCPU_SLB_SIZE
1486 3: stw r5,VCPU_SLB_MAX(r9)
1489 * Save the guest PURR/SPURR
1494 ld r8,VCPU_SPURR(r9)
1495 std r5,VCPU_PURR(r9)
1496 std r6,VCPU_SPURR(r9)
1501 * Restore host PURR/SPURR and add guest times
1502 * so that the time in the guest gets accounted.
1504 ld r3,HSTATE_PURR(r13)
1505 ld r4,HSTATE_SPURR(r13)
1512 ld r3, HSTATE_KVM_VCORE(r13)
1515 /* On P9, if the guest has large decr enabled, don't sign extend */
1517 ld r4, VCORE_LPCR(r3)
1518 andis. r4, r4, LPCR_LD@h
1520 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1523 /* r5 is a guest timebase value here, convert to host TB */
1524 ld r4,VCORE_TB_OFFSET(r3)
1526 std r5,VCPU_DEC_EXPIRES(r9)
1530 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1531 /* Save POWER8-specific registers */
1535 std r5, VCPU_IAMR(r9)
1536 stw r6, VCPU_PSPB(r9)
1537 std r7, VCPU_FSCR(r9)
1541 std r7, VCPU_TAR(r9)
1542 mfspr r8, SPRN_EBBHR
1543 std r8, VCPU_EBBHR(r9)
1544 mfspr r5, SPRN_EBBRR
1545 mfspr r6, SPRN_BESCR
1548 std r5, VCPU_EBBRR(r9)
1549 std r6, VCPU_BESCR(r9)
1550 stw r7, VCPU_GUEST_PID(r9)
1551 std r8, VCPU_WORT(r9)
1553 mfspr r5, SPRN_TCSCR
1555 mfspr r7, SPRN_CSIGR
1557 std r5, VCPU_TCSCR(r9)
1558 std r6, VCPU_ACOP(r9)
1559 std r7, VCPU_CSIGR(r9)
1560 std r8, VCPU_TACR(r9)
1563 mfspr r6, SPRN_PSSCR
1564 std r5, VCPU_TID(r9)
1565 rldicl r6, r6, 4, 50 /* r6 &= PSSCR_GUEST_VIS */
1567 std r6, VCPU_PSSCR(r9)
1568 /* Restore host HFSCR value */
1569 ld r7, STACK_SLOT_HFSCR(r1)
1570 mtspr SPRN_HFSCR, r7
1571 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
1573 * Restore various registers to 0, where non-zero values
1574 * set by the guest could disrupt the host.
1581 mtspr SPRN_TCSCR, r0
1582 /* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1585 mtspr SPRN_MMCRS, r0
1586 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1589 /* Save and reset AMR and UAMOR before turning on the MMU */
1593 std r6,VCPU_UAMOR(r9)
1596 mtspr SPRN_UAMOR, r6
1598 /* Switch DSCR back to host value */
1600 ld r7, HSTATE_DSCR(r13)
1601 std r8, VCPU_DSCR(r9)
1604 /* Save non-volatile GPRs */
1605 std r14, VCPU_GPR(R14)(r9)
1606 std r15, VCPU_GPR(R15)(r9)
1607 std r16, VCPU_GPR(R16)(r9)
1608 std r17, VCPU_GPR(R17)(r9)
1609 std r18, VCPU_GPR(R18)(r9)
1610 std r19, VCPU_GPR(R19)(r9)
1611 std r20, VCPU_GPR(R20)(r9)
1612 std r21, VCPU_GPR(R21)(r9)
1613 std r22, VCPU_GPR(R22)(r9)
1614 std r23, VCPU_GPR(R23)(r9)
1615 std r24, VCPU_GPR(R24)(r9)
1616 std r25, VCPU_GPR(R25)(r9)
1617 std r26, VCPU_GPR(R26)(r9)
1618 std r27, VCPU_GPR(R27)(r9)
1619 std r28, VCPU_GPR(R28)(r9)
1620 std r29, VCPU_GPR(R29)(r9)
1621 std r30, VCPU_GPR(R30)(r9)
1622 std r31, VCPU_GPR(R31)(r9)
1625 mfspr r3, SPRN_SPRG0
1626 mfspr r4, SPRN_SPRG1
1627 mfspr r5, SPRN_SPRG2
1628 mfspr r6, SPRN_SPRG3
1629 std r3, VCPU_SPRG0(r9)
1630 std r4, VCPU_SPRG1(r9)
1631 std r5, VCPU_SPRG2(r9)
1632 std r6, VCPU_SPRG3(r9)
1638 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1641 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
1644 END_FTR_SECTION_IFSET(CPU_FTR_TM)
1647 /* Increment yield count if they have a VPA */
1648 ld r8, VCPU_VPA(r9) /* do they have a VPA? */
1651 li r4, LPPACA_YIELDCOUNT
1656 stb r3, VCPU_VPA_DIRTY(r9)
1658 /* Save PMU registers if requested */
1659 /* r8 and cr0.eq are live here */
1662 * POWER8 seems to have a hardware bug where setting
1663 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
1664 * when some counters are already negative doesn't seem
1665 * to cause a performance monitor alert (and hence interrupt).
1666 * The effect of this is that when saving the PMU state,
1667 * if there is no PMU alert pending when we read MMCR0
1668 * before freezing the counters, but one becomes pending
1669 * before we read the counters, we lose it.
1670 * To work around this, we need a way to freeze the counters
1671 * before reading MMCR0. Normally, freezing the counters
1672 * is done by writing MMCR0 (to set MMCR0[FC]) which
1673 * unavoidably writes MMCR0[PMA0] as well. On POWER8,
1674 * we can also freeze the counters using MMCR2, by writing
1675 * 1s to all the counter freeze condition bits (there are
1676 * 9 bits each for 6 counters).
1678 li r3, -1 /* set all freeze bits */
1680 mfspr r10, SPRN_MMCR2
1681 mtspr SPRN_MMCR2, r3
1683 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1685 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
1686 mfspr r4, SPRN_MMCR0 /* save MMCR0 */
1687 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
1688 mfspr r6, SPRN_MMCRA
1689 /* Clear MMCRA in order to disable SDAR updates */
1691 mtspr SPRN_MMCRA, r7
1693 beq 21f /* if no VPA, save PMU stuff anyway */
1694 lbz r7, LPPACA_PMCINUSE(r8)
1695 cmpwi r7, 0 /* did they ask for PMU stuff to be saved? */
1697 std r3, VCPU_MMCR(r9) /* if not, set saved MMCR0 to FC */
1699 21: mfspr r5, SPRN_MMCR1
1702 std r4, VCPU_MMCR(r9)
1703 std r5, VCPU_MMCR + 8(r9)
1704 std r6, VCPU_MMCR + 16(r9)
1706 std r10, VCPU_MMCR + 24(r9)
1707 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1708 std r7, VCPU_SIAR(r9)
1709 std r8, VCPU_SDAR(r9)
1716 stw r3, VCPU_PMC(r9)
1717 stw r4, VCPU_PMC + 4(r9)
1718 stw r5, VCPU_PMC + 8(r9)
1719 stw r6, VCPU_PMC + 12(r9)
1720 stw r7, VCPU_PMC + 16(r9)
1721 stw r8, VCPU_PMC + 20(r9)
1724 std r5, VCPU_SIER(r9)
1725 BEGIN_FTR_SECTION_NESTED(96)
1726 mfspr r6, SPRN_SPMC1
1727 mfspr r7, SPRN_SPMC2
1728 mfspr r8, SPRN_MMCRS
1729 stw r6, VCPU_PMC + 24(r9)
1730 stw r7, VCPU_PMC + 28(r9)
1731 std r8, VCPU_MMCR + 32(r9)
1733 mtspr SPRN_MMCRS, r4
1734 END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
1735 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1738 /* Restore host values of some registers */
1740 ld r5, STACK_SLOT_CIABR(r1)
1741 ld r6, STACK_SLOT_DAWR(r1)
1742 ld r7, STACK_SLOT_DAWRX(r1)
1743 mtspr SPRN_CIABR, r5
1745 mtspr SPRN_DAWRX, r7
1746 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1748 ld r5, STACK_SLOT_TID(r1)
1749 ld r6, STACK_SLOT_PSSCR(r1)
1750 ld r7, STACK_SLOT_PID(r1)
1751 ld r8, STACK_SLOT_IAMR(r1)
1753 mtspr SPRN_PSSCR, r6
1756 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1758 #ifdef CONFIG_PPC_RADIX_MMU
1760 * Are we running hash or radix ?
1763 lbz r0, KVM_RADIX(r5)
1767 /* Radix: Handle the case where the guest used an illegal PID */
1768 LOAD_REG_ADDR(r4, mmu_base_pid)
1769 lwz r3, VCPU_GUEST_PID(r9)
1775 * Illegal PID, the HW might have prefetched and cached in the TLB
1776 * some translations for the LPID 0 / guest PID combination which
1777 * Linux doesn't know about, so we need to flush that PID out of
1778 * the TLB. First we need to set LPIDR to 0 so tlbiel applies to
1779 * the right context.
1785 /* Then do a congruence class local flush */
1787 lwz r0,KVM_TLB_SETS(r6)
1789 li r7,0x400 /* IS field = 0b01 */
1791 sldi r0,r3,32 /* RS has PID */
1792 1: PPC_TLBIEL(7,0,2,1,1) /* RIC=2, PRS=1, R=1 */
1797 2: /* Flush the ERAT on radix P9 DD1 guest exit */
1800 END_FTR_SECTION_IFSET(CPU_FTR_POWER9_DD1)
1802 #endif /* CONFIG_PPC_RADIX_MMU */
1804 /* Hash: clear out SLB */
1811 * POWER7/POWER8 guest -> host partition switch code.
1812 * We don't have to lock against tlbies but we do
1813 * have to coordinate the hardware threads.
1815 kvmhv_switch_to_host:
1816 /* Secondary threads wait for primary to do partition switch */
1817 ld r5,HSTATE_KVM_VCORE(r13)
1818 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1819 lbz r3,HSTATE_PTID(r13)
1823 13: lbz r3,VCORE_IN_GUEST(r5)
1829 /* Primary thread waits for all the secondaries to exit guest */
1830 15: lwz r3,VCORE_ENTRY_EXIT(r5)
1831 rlwinm r0,r3,32-8,0xff
1837 /* Did we actually switch to the guest at all? */
1838 lbz r6, VCORE_IN_GUEST(r5)
1842 /* Primary thread switches back to host partition */
1843 lwz r7,KVM_HOST_LPID(r4)
1845 ld r6,KVM_HOST_SDR1(r4)
1846 li r8,LPID_RSVD /* switch to reserved LPID */
1849 mtspr SPRN_SDR1,r6 /* switch to host page table */
1850 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1855 /* DPDES and VTB are shared between threads */
1856 mfspr r7, SPRN_DPDES
1858 std r7, VCORE_DPDES(r5)
1859 std r8, VCORE_VTB(r5)
1860 /* clear DPDES so we don't get guest doorbells in the host */
1862 mtspr SPRN_DPDES, r8
1863 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1865 /* If HMI, call kvmppc_realmode_hmi_handler() */
1866 cmpwi r12, BOOK3S_INTERRUPT_HMI
1868 bl kvmppc_realmode_hmi_handler
1870 li r12, BOOK3S_INTERRUPT_HMI
1872 * At this point kvmppc_realmode_hmi_handler would have resync-ed
1873 * the TB. Hence it is not required to subtract guest timebase
1874 * offset from timebase. So, skip it.
1876 * Also, do not call kvmppc_subcore_exit_guest() because it has
1877 * been invoked as part of kvmppc_realmode_hmi_handler().
1882 /* Subtract timebase offset from timebase */
1883 ld r8,VCORE_TB_OFFSET(r5)
1886 mftb r6 /* current guest timebase */
1888 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
1889 mftb r7 /* check if lower 24 bits overflowed */
1894 addis r8,r8,0x100 /* if so, increment upper 40 bits */
1897 17: bl kvmppc_subcore_exit_guest
1899 30: ld r5,HSTATE_KVM_VCORE(r13)
1900 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1903 ld r0, VCORE_PCR(r5)
1909 /* Signal secondary CPUs to continue */
1910 stb r0,VCORE_IN_GUEST(r5)
1911 19: lis r8,0x7fff /* MAX_INT@h */
1914 16: ld r8,KVM_HOST_LPCR(r4)
1918 /* load host SLB entries */
1919 BEGIN_MMU_FTR_SECTION
1921 END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
1922 ld r8,PACA_SLBSHADOWPTR(r13)
1924 .rept SLB_NUM_BOLTED
1925 li r3, SLBSHADOW_SAVEAREA
1929 andis. r7,r5,SLB_ESID_V@h
1935 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1936 /* Finish timing, if we have a vcpu */
1937 ld r4, HSTATE_KVM_VCPU(r13)
1941 bl kvmhv_accumulate_time
1944 /* Unset guest mode */
1945 li r0, KVM_GUEST_MODE_NONE
1946 stb r0, HSTATE_IN_GUEST(r13)
1948 ld r0, SFS+PPC_LR_STKOFF(r1)
1954 * Check whether an HDSI is an HPTE not found fault or something else.
1955 * If it is an HPTE not found fault that is due to the guest accessing
1956 * a page that they have mapped but which we have paged out, then
1957 * we continue on with the guest exit path. In all other cases,
1958 * reflect the HDSI to the guest as a DSI.
1962 lbz r0, KVM_RADIX(r3)
1964 mfspr r6, SPRN_HDSISR
1966 /* Look for DSISR canary. If we find it, retry instruction */
1969 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1971 bne .Lradix_hdsi /* on radix, just save DAR/DSISR/ASDR */
1972 /* HPTE not found fault or protection fault? */
1973 andis. r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
1974 beq 1f /* if not, send it to the guest */
1975 andi. r0, r11, MSR_DR /* data relocation enabled? */
1978 mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
1980 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1982 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
1983 li r0, BOOK3S_INTERRUPT_DATA_SEGMENT
1984 bne 7f /* if no SLB entry found */
1985 4: std r4, VCPU_FAULT_DAR(r9)
1986 stw r6, VCPU_FAULT_DSISR(r9)
1988 /* Search the hash table. */
1989 mr r3, r9 /* vcpu pointer */
1990 li r7, 1 /* data fault */
1991 bl kvmppc_hpte_hv_fault
1992 ld r9, HSTATE_KVM_VCPU(r13)
1994 ld r11, VCPU_MSR(r9)
1995 li r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1996 cmpdi r3, 0 /* retry the instruction */
1998 cmpdi r3, -1 /* handle in kernel mode */
2000 cmpdi r3, -2 /* MMIO emulation; need instr word */
2003 /* Synthesize a DSI (or DSegI) for the guest */
2004 ld r4, VCPU_FAULT_DAR(r9)
2006 1: li r0, BOOK3S_INTERRUPT_DATA_STORAGE
2007 mtspr SPRN_DSISR, r6
2008 7: mtspr SPRN_DAR, r4
2009 mtspr SPRN_SRR0, r10
2010 mtspr SPRN_SRR1, r11
2012 bl kvmppc_msr_interrupt
2013 fast_interrupt_c_return:
2014 6: ld r7, VCPU_CTR(r9)
2021 3: ld r5, VCPU_KVM(r9) /* not relocated, use VRMA */
2022 ld r5, KVM_VRMA_SLB_V(r5)
2025 /* If this is for emulated MMIO, load the instruction word */
2026 2: li r8, KVM_INST_FETCH_FAILED /* In case lwz faults */
2028 /* Set guest mode to 'jump over instruction' so if lwz faults
2029 * we'll just continue at the next IP. */
2030 li r0, KVM_GUEST_MODE_SKIP
2031 stb r0, HSTATE_IN_GUEST(r13)
2033 /* Do the access with MSR:DR enabled */
2035 ori r4, r3, MSR_DR /* Enable paging for data */
2040 /* Store the result */
2041 stw r8, VCPU_LAST_INST(r9)
2043 /* Unset guest mode. */
2044 li r0, KVM_GUEST_MODE_HOST_HV
2045 stb r0, HSTATE_IN_GUEST(r13)
2049 std r4, VCPU_FAULT_DAR(r9)
2050 stw r6, VCPU_FAULT_DSISR(r9)
2053 std r5, VCPU_FAULT_GPA(r9)
2057 * Similarly for an HISI, reflect it to the guest as an ISI unless
2058 * it is an HPTE not found fault for a page that we have paged out.
2062 lbz r0, KVM_RADIX(r3)
2064 bne .Lradix_hisi /* for radix, just save ASDR */
2065 andis. r0, r11, SRR1_ISI_NOPT@h
2067 andi. r0, r11, MSR_IR /* instruction relocation enabled? */
2070 mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
2072 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2074 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
2075 li r0, BOOK3S_INTERRUPT_INST_SEGMENT
2076 bne 7f /* if no SLB entry found */
2078 /* Search the hash table. */
2079 mr r3, r9 /* vcpu pointer */
2082 li r7, 0 /* instruction fault */
2083 bl kvmppc_hpte_hv_fault
2084 ld r9, HSTATE_KVM_VCPU(r13)
2086 ld r11, VCPU_MSR(r9)
2087 li r12, BOOK3S_INTERRUPT_H_INST_STORAGE
2088 cmpdi r3, 0 /* retry the instruction */
2089 beq fast_interrupt_c_return
2090 cmpdi r3, -1 /* handle in kernel mode */
2093 /* Synthesize an ISI (or ISegI) for the guest */
2095 1: li r0, BOOK3S_INTERRUPT_INST_STORAGE
2096 7: mtspr SPRN_SRR0, r10
2097 mtspr SPRN_SRR1, r11
2099 bl kvmppc_msr_interrupt
2100 b fast_interrupt_c_return
2102 3: ld r6, VCPU_KVM(r9) /* not relocated, use VRMA */
2103 ld r5, KVM_VRMA_SLB_V(r6)
2107 * Try to handle an hcall in real mode.
2108 * Returns to the guest if we handle it, or continues on up to
2109 * the kernel if we can't (i.e. if we don't have a handler for
2110 * it, or if the handler returns H_TOO_HARD).
2112 * r5 - r8 contain hcall args,
2113 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
2115 hcall_try_real_mode:
2116 ld r3,VCPU_GPR(R3)(r9)
2118 /* sc 1 from userspace - reflect to guest syscall */
2119 bne sc_1_fast_return
2121 cmpldi r3,hcall_real_table_end - hcall_real_table
2123 /* See if this hcall is enabled for in-kernel handling */
2125 srdi r0, r3, 8 /* r0 = (r3 / 4) >> 6 */
2126 sldi r0, r0, 3 /* index into kvm->arch.enabled_hcalls[] */
2128 ld r0, KVM_ENABLED_HCALLS(r4)
2129 rlwinm r4, r3, 32-2, 0x3f /* r4 = (r3 / 4) & 0x3f */
2133 /* Get pointer to handler, if any, and call it */
2134 LOAD_REG_ADDR(r4, hcall_real_table)
2140 mr r3,r9 /* get vcpu pointer */
2141 ld r4,VCPU_GPR(R4)(r9)
2144 beq hcall_real_fallback
2145 ld r4,HSTATE_KVM_VCPU(r13)
2146 std r3,VCPU_GPR(R3)(r4)
2154 li r10, BOOK3S_INTERRUPT_SYSCALL
2155 bl kvmppc_msr_interrupt
2159 /* We've attempted a real mode hcall, but it's punted it back
2160 * to userspace. We need to restore some clobbered volatiles
2161 * before resuming the pass-it-to-qemu path */
2162 hcall_real_fallback:
2163 li r12,BOOK3S_INTERRUPT_SYSCALL
2164 ld r9, HSTATE_KVM_VCPU(r13)
2168 .globl hcall_real_table
2170 .long 0 /* 0 - unused */
2171 .long DOTSYM(kvmppc_h_remove) - hcall_real_table
2172 .long DOTSYM(kvmppc_h_enter) - hcall_real_table
2173 .long DOTSYM(kvmppc_h_read) - hcall_real_table
2174 .long DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
2175 .long DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
2176 .long DOTSYM(kvmppc_h_protect) - hcall_real_table
2177 .long DOTSYM(kvmppc_h_get_tce) - hcall_real_table
2178 .long DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
2179 .long 0 /* 0x24 - H_SET_SPRG0 */
2180 .long DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
2195 #ifdef CONFIG_KVM_XICS
2196 .long DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
2197 .long DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
2198 .long DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
2199 .long DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
2200 .long DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
2202 .long 0 /* 0x64 - H_EOI */
2203 .long 0 /* 0x68 - H_CPPR */
2204 .long 0 /* 0x6c - H_IPI */
2205 .long 0 /* 0x70 - H_IPOLL */
2206 .long 0 /* 0x74 - H_XIRR */
2234 .long DOTSYM(kvmppc_h_cede) - hcall_real_table
2235 .long DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
2251 .long DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
2255 .long DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
2256 .long DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
2257 .long DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
2369 #ifdef CONFIG_KVM_XICS
2370 .long DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
2372 .long 0 /* 0x2fc - H_XIRR_X*/
2374 .long DOTSYM(kvmppc_h_random) - hcall_real_table
2375 .globl hcall_real_table_end
2376 hcall_real_table_end:
2378 _GLOBAL(kvmppc_h_set_xdabr)
2379 andi. r0, r5, DABRX_USER | DABRX_KERNEL
2381 li r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2384 6: li r3, H_PARAMETER
2387 _GLOBAL(kvmppc_h_set_dabr)
2388 li r5, DABRX_USER | DABRX_KERNEL
2392 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2393 std r4,VCPU_DABR(r3)
2394 stw r5, VCPU_DABRX(r3)
2395 mtspr SPRN_DABRX, r5
2396 /* Work around P7 bug where DABR can get corrupted on mtspr */
2397 1: mtspr SPRN_DABR,r4
2405 /* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
2406 2: rlwimi r5, r4, 5, DAWRX_DR | DAWRX_DW
2407 rlwimi r5, r4, 2, DAWRX_WT
2409 std r4, VCPU_DAWR(r3)
2410 std r5, VCPU_DAWRX(r3)
2412 mtspr SPRN_DAWRX, r5
2416 _GLOBAL(kvmppc_h_cede) /* r3 = vcpu pointer, r11 = msr, r13 = paca */
2418 std r11,VCPU_MSR(r3)
2420 stb r0,VCPU_CEDED(r3)
2421 sync /* order setting ceded vs. testing prodded */
2422 lbz r5,VCPU_PRODDED(r3)
2424 bne kvm_cede_prodded
2425 li r12,0 /* set trap to 0 to say hcall is handled */
2426 stw r12,VCPU_TRAP(r3)
2428 std r0,VCPU_GPR(R3)(r3)
2431 * Set our bit in the bitmask of napping threads unless all the
2432 * other threads are already napping, in which case we send this
2435 ld r5,HSTATE_KVM_VCORE(r13)
2436 lbz r6,HSTATE_PTID(r13)
2437 lwz r8,VCORE_ENTRY_EXIT(r5)
2441 addi r6,r5,VCORE_NAPPING_THREADS
2448 /* order napping_threads update vs testing entry_exit_map */
2451 stb r0,HSTATE_NAPPING(r13)
2452 lwz r7,VCORE_ENTRY_EXIT(r5)
2454 bge 33f /* another thread already exiting */
2457 * Although not specifically required by the architecture, POWER7
2458 * preserves the following registers in nap mode, even if an SMT mode
2459 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2460 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2462 /* Save non-volatile GPRs */
2463 std r14, VCPU_GPR(R14)(r3)
2464 std r15, VCPU_GPR(R15)(r3)
2465 std r16, VCPU_GPR(R16)(r3)
2466 std r17, VCPU_GPR(R17)(r3)
2467 std r18, VCPU_GPR(R18)(r3)
2468 std r19, VCPU_GPR(R19)(r3)
2469 std r20, VCPU_GPR(R20)(r3)
2470 std r21, VCPU_GPR(R21)(r3)
2471 std r22, VCPU_GPR(R22)(r3)
2472 std r23, VCPU_GPR(R23)(r3)
2473 std r24, VCPU_GPR(R24)(r3)
2474 std r25, VCPU_GPR(R25)(r3)
2475 std r26, VCPU_GPR(R26)(r3)
2476 std r27, VCPU_GPR(R27)(r3)
2477 std r28, VCPU_GPR(R28)(r3)
2478 std r29, VCPU_GPR(R29)(r3)
2479 std r30, VCPU_GPR(R30)(r3)
2480 std r31, VCPU_GPR(R31)(r3)
2485 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2488 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
2490 ld r9, HSTATE_KVM_VCPU(r13)
2492 END_FTR_SECTION_IFSET(CPU_FTR_TM)
2496 * Set DEC to the smaller of DEC and HDEC, so that we wake
2497 * no later than the end of our timeslice (HDEC interrupts
2498 * don't wake us from nap).
2504 /* On P9 check whether the guest has large decrementer mode enabled */
2505 ld r6, HSTATE_KVM_VCORE(r13)
2506 ld r6, VCORE_LPCR(r6)
2507 andis. r6, r6, LPCR_LD@h
2509 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2516 /* save expiry time of guest decrementer */
2518 ld r4, HSTATE_KVM_VCPU(r13)
2519 ld r5, HSTATE_KVM_VCORE(r13)
2520 ld r6, VCORE_TB_OFFSET(r5)
2521 subf r3, r6, r3 /* convert to host TB value */
2522 std r3, VCPU_DEC_EXPIRES(r4)
2524 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2525 ld r4, HSTATE_KVM_VCPU(r13)
2526 addi r3, r4, VCPU_TB_CEDE
2527 bl kvmhv_accumulate_time
2530 lis r3, LPCR_PECEDP@h /* Do wake on privileged doorbell */
2533 * Take a nap until a decrementer or external or doobell interrupt
2534 * occurs, with PECE1 and PECE0 set in LPCR.
2535 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2536 * Also clear the runlatch bit before napping.
2539 mfspr r0, SPRN_CTRLF
2541 mtspr SPRN_CTRLT, r0
2544 stb r0,HSTATE_HWTHREAD_REQ(r13)
2546 ori r5,r5,LPCR_PECE0 | LPCR_PECE1
2548 ori r5, r5, LPCR_PECEDH
2549 rlwimi r5, r3, 0, LPCR_PECEDP
2550 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2552 kvm_nap_sequence: /* desired LPCR value in r5 */
2555 * PSSCR bits: exit criterion = 1 (wakeup based on LPCR at sreset)
2556 * enable state loss = 1 (allow SMT mode switch)
2557 * requested level = 0 (just stop dispatching)
2559 lis r3, (PSSCR_EC | PSSCR_ESL)@h
2560 mtspr SPRN_PSSCR, r3
2561 /* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
2562 li r4, LPCR_PECE_HVEE@higher
2565 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2569 std r0, HSTATE_SCRATCH0(r13)
2571 ld r0, HSTATE_SCRATCH0(r13)
2578 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
2587 /* get vcpu pointer */
2588 ld r4, HSTATE_KVM_VCPU(r13)
2590 /* Woken by external or decrementer interrupt */
2591 ld r1, HSTATE_HOST_R1(r13)
2593 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2594 addi r3, r4, VCPU_TB_RMINTR
2595 bl kvmhv_accumulate_time
2598 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2601 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
2603 bl kvmppc_restore_tm
2604 END_FTR_SECTION_IFSET(CPU_FTR_TM)
2607 /* load up FP state */
2610 /* Restore guest decrementer */
2611 ld r3, VCPU_DEC_EXPIRES(r4)
2612 ld r5, HSTATE_KVM_VCORE(r13)
2613 ld r6, VCORE_TB_OFFSET(r5)
2614 add r3, r3, r6 /* convert host TB to guest TB value */
2620 ld r14, VCPU_GPR(R14)(r4)
2621 ld r15, VCPU_GPR(R15)(r4)
2622 ld r16, VCPU_GPR(R16)(r4)
2623 ld r17, VCPU_GPR(R17)(r4)
2624 ld r18, VCPU_GPR(R18)(r4)
2625 ld r19, VCPU_GPR(R19)(r4)
2626 ld r20, VCPU_GPR(R20)(r4)
2627 ld r21, VCPU_GPR(R21)(r4)
2628 ld r22, VCPU_GPR(R22)(r4)
2629 ld r23, VCPU_GPR(R23)(r4)
2630 ld r24, VCPU_GPR(R24)(r4)
2631 ld r25, VCPU_GPR(R25)(r4)
2632 ld r26, VCPU_GPR(R26)(r4)
2633 ld r27, VCPU_GPR(R27)(r4)
2634 ld r28, VCPU_GPR(R28)(r4)
2635 ld r29, VCPU_GPR(R29)(r4)
2636 ld r30, VCPU_GPR(R30)(r4)
2637 ld r31, VCPU_GPR(R31)(r4)
2639 /* Check the wake reason in SRR1 to see why we got here */
2640 bl kvmppc_check_wake_reason
2643 * Restore volatile registers since we could have called a
2644 * C routine in kvmppc_check_wake_reason
2646 * r3 tells us whether we need to return to host or not
2647 * WARNING: it gets checked further down:
2648 * should not modify r3 until this check is done.
2650 ld r4, HSTATE_KVM_VCPU(r13)
2652 /* clear our bit in vcore->napping_threads */
2653 34: ld r5,HSTATE_KVM_VCORE(r13)
2654 lbz r7,HSTATE_PTID(r13)
2657 addi r6,r5,VCORE_NAPPING_THREADS
2663 stb r0,HSTATE_NAPPING(r13)
2665 /* See if the wake reason saved in r3 means we need to exit */
2666 stw r12, VCPU_TRAP(r4)
2671 /* see if any other thread is already exiting */
2672 lwz r0,VCORE_ENTRY_EXIT(r5)
2676 b kvmppc_cede_reentry /* if not go back to guest */
2678 /* cede when already previously prodded case */
2681 stb r0,VCPU_PRODDED(r3)
2682 sync /* order testing prodded vs. clearing ceded */
2683 stb r0,VCPU_CEDED(r3)
2687 /* we've ceded but we want to give control to the host */
2689 ld r9, HSTATE_KVM_VCPU(r13)
2692 /* Try to handle a machine check in real mode */
2693 machine_check_realmode:
2694 mr r3, r9 /* get vcpu pointer */
2695 bl kvmppc_realmode_machine_check
2697 ld r9, HSTATE_KVM_VCPU(r13)
2698 li r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2700 * For the guest that is FWNMI capable, deliver all the MCE errors
2701 * (handled/unhandled) by exiting the guest with KVM_EXIT_NMI exit
2702 * reason. This new approach injects machine check errors in guest
2703 * address space to guest with additional information in the form
2704 * of RTAS event, thus enabling guest kernel to suitably handle
2707 * For the guest that is not FWNMI capable (old QEMU) fallback
2708 * to old behaviour for backward compatibility:
2709 * Deliver unhandled/fatal (e.g. UE) MCE errors to guest either
2710 * through machine check interrupt (set HSRR0 to 0x200).
2711 * For handled errors (no-fatal), just go back to guest execution
2712 * with current HSRR0.
2713 * if we receive machine check with MSR(RI=0) then deliver it to
2714 * guest as machine check causing guest to crash.
2716 ld r11, VCPU_MSR(r9)
2717 rldicl. r0, r11, 64-MSR_HV_LG, 63 /* check if it happened in HV mode */
2718 bne mc_cont /* if so, exit to host */
2719 /* Check if guest is capable of handling NMI exit */
2720 ld r10, VCPU_KVM(r9)
2721 lbz r10, KVM_FWNMI(r10)
2722 cmpdi r10, 1 /* FWNMI capable? */
2723 beq mc_cont /* if so, exit with KVM_EXIT_NMI. */
2725 /* if not, fall through for backward compatibility. */
2726 andi. r10, r11, MSR_RI /* check for unrecoverable exception */
2727 beq 1f /* Deliver a machine check to guest */
2729 cmpdi r3, 0 /* Did we handle MCE ? */
2730 bne 2f /* Continue guest execution. */
2731 /* If not, deliver a machine check. SRR0/1 are already set */
2732 1: li r10, BOOK3S_INTERRUPT_MACHINE_CHECK
2733 bl kvmppc_msr_interrupt
2734 2: b fast_interrupt_c_return
2737 * Check the reason we woke from nap, and take appropriate action.
2739 * 0 if nothing needs to be done
2740 * 1 if something happened that needs to be handled by the host
2741 * -1 if there was a guest wakeup (IPI or msgsnd)
2742 * -2 if we handled a PCI passthrough interrupt (returned by
2743 * kvmppc_read_intr only)
2745 * Also sets r12 to the interrupt vector for any interrupt that needs
2746 * to be handled now by the host (0x500 for external interrupt), or zero.
2747 * Modifies all volatile registers (since it may call a C function).
2748 * This routine calls kvmppc_read_intr, a C function, if an external
2749 * interrupt is pending.
2751 kvmppc_check_wake_reason:
2754 rlwinm r6, r6, 45-31, 0xf /* extract wake reason field (P8) */
2756 rlwinm r6, r6, 45-31, 0xe /* P7 wake reason field is 3 bits */
2757 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2758 cmpwi r6, 8 /* was it an external interrupt? */
2759 beq 7f /* if so, see what it was */
2762 cmpwi r6, 6 /* was it the decrementer? */
2765 cmpwi r6, 5 /* privileged doorbell? */
2767 cmpwi r6, 3 /* hypervisor doorbell? */
2769 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2770 cmpwi r6, 0xa /* Hypervisor maintenance ? */
2772 li r3, 1 /* anything else, return 1 */
2775 /* hypervisor doorbell */
2776 3: li r12, BOOK3S_INTERRUPT_H_DOORBELL
2779 * Clear the doorbell as we will invoke the handler
2780 * explicitly in the guest exit path.
2782 lis r6, (PPC_DBELL_SERVER << (63-36))@h
2784 /* see if it's a host IPI */
2789 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2790 lbz r0, HSTATE_HOST_IPI(r13)
2793 /* if not, return -1 */
2797 /* Woken up due to Hypervisor maintenance interrupt */
2798 4: li r12, BOOK3S_INTERRUPT_HMI
2802 /* external interrupt - create a stack frame so we can call C */
2804 std r0, PPC_LR_STKOFF(r1)
2805 stdu r1, -PPC_MIN_STKFRM(r1)
2808 li r12, BOOK3S_INTERRUPT_EXTERNAL
2813 * Return code of 2 means PCI passthrough interrupt, but
2814 * we need to return back to host to complete handling the
2815 * interrupt. Trap reason is expected in r12 by guest
2818 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
2820 ld r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
2821 addi r1, r1, PPC_MIN_STKFRM
2826 * Save away FP, VMX and VSX registers.
2828 * N.B. r30 and r31 are volatile across this function,
2829 * thus it is not callable from C.
2836 #ifdef CONFIG_ALTIVEC
2838 oris r8,r8,MSR_VEC@h
2839 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2843 oris r8,r8,MSR_VSX@h
2844 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2847 addi r3,r3,VCPU_FPRS
2849 #ifdef CONFIG_ALTIVEC
2851 addi r3,r31,VCPU_VRS
2853 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2855 mfspr r6,SPRN_VRSAVE
2856 stw r6,VCPU_VRSAVE(r31)
2861 * Load up FP, VMX and VSX registers
2863 * N.B. r30 and r31 are volatile across this function,
2864 * thus it is not callable from C.
2871 #ifdef CONFIG_ALTIVEC
2873 oris r8,r8,MSR_VEC@h
2874 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2878 oris r8,r8,MSR_VSX@h
2879 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2882 addi r3,r4,VCPU_FPRS
2884 #ifdef CONFIG_ALTIVEC
2886 addi r3,r31,VCPU_VRS
2888 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2890 lwz r7,VCPU_VRSAVE(r31)
2891 mtspr SPRN_VRSAVE,r7
2896 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2898 * Save transactional state and TM-related registers.
2899 * Called with r9 pointing to the vcpu struct.
2900 * This can modify all checkpointed registers, but
2901 * restores r1, r2 and r9 (vcpu pointer) before exit.
2905 std r0, PPC_LR_STKOFF(r1)
2910 rldimi r8, r0, MSR_TM_LG, 63-MSR_TM_LG
2914 rldicl. r5, r5, 64 - MSR_TS_S_LG, 62
2915 beq 1f /* TM not active in guest. */
2917 std r1, HSTATE_HOST_R1(r13)
2918 li r3, TM_CAUSE_KVM_RESCHED
2920 /* Clear the MSR RI since r1, r13 are all going to be foobar. */
2924 /* All GPRs are volatile at this point. */
2927 /* Temporarily store r13 and r9 so we have some regs to play with */
2930 std r9, PACATMSCRATCH(r13)
2931 ld r9, HSTATE_KVM_VCPU(r13)
2933 /* Get a few more GPRs free. */
2934 std r29, VCPU_GPRS_TM(29)(r9)
2935 std r30, VCPU_GPRS_TM(30)(r9)
2936 std r31, VCPU_GPRS_TM(31)(r9)
2938 /* Save away PPR and DSCR soon so don't run with user values. */
2941 mfspr r30, SPRN_DSCR
2942 ld r29, HSTATE_DSCR(r13)
2943 mtspr SPRN_DSCR, r29
2945 /* Save all but r9, r13 & r29-r31 */
2948 .if (reg != 9) && (reg != 13)
2949 std reg, VCPU_GPRS_TM(reg)(r9)
2953 /* ... now save r13 */
2955 std r4, VCPU_GPRS_TM(13)(r9)
2956 /* ... and save r9 */
2957 ld r4, PACATMSCRATCH(r13)
2958 std r4, VCPU_GPRS_TM(9)(r9)
2960 /* Reload stack pointer and TOC. */
2961 ld r1, HSTATE_HOST_R1(r13)
2964 /* Set MSR RI now we have r1 and r13 back. */
2968 /* Save away checkpinted SPRs. */
2969 std r31, VCPU_PPR_TM(r9)
2970 std r30, VCPU_DSCR_TM(r9)
2977 std r5, VCPU_LR_TM(r9)
2978 stw r6, VCPU_CR_TM(r9)
2979 std r7, VCPU_CTR_TM(r9)
2980 std r8, VCPU_AMR_TM(r9)
2981 std r10, VCPU_TAR_TM(r9)
2982 std r11, VCPU_XER_TM(r9)
2984 /* Restore r12 as trap number. */
2985 lwz r12, VCPU_TRAP(r9)
2988 addi r3, r9, VCPU_FPRS_TM
2990 addi r3, r9, VCPU_VRS_TM
2992 mfspr r6, SPRN_VRSAVE
2993 stw r6, VCPU_VRSAVE_TM(r9)
2996 * We need to save these SPRs after the treclaim so that the software
2997 * error code is recorded correctly in the TEXASR. Also the user may
2998 * change these outside of a transaction, so they must always be
3001 mfspr r5, SPRN_TFHAR
3002 mfspr r6, SPRN_TFIAR
3003 mfspr r7, SPRN_TEXASR
3004 std r5, VCPU_TFHAR(r9)
3005 std r6, VCPU_TFIAR(r9)
3006 std r7, VCPU_TEXASR(r9)
3008 ld r0, PPC_LR_STKOFF(r1)
3013 * Restore transactional state and TM-related registers.
3014 * Called with r4 pointing to the vcpu struct.
3015 * This potentially modifies all checkpointed registers.
3016 * It restores r1, r2, r4 from the PACA.
3020 std r0, PPC_LR_STKOFF(r1)
3022 /* Turn on TM/FP/VSX/VMX so we can restore them. */
3028 oris r5, r5, (MSR_VEC | MSR_VSX)@h
3032 * The user may change these outside of a transaction, so they must
3033 * always be context switched.
3035 ld r5, VCPU_TFHAR(r4)
3036 ld r6, VCPU_TFIAR(r4)
3037 ld r7, VCPU_TEXASR(r4)
3038 mtspr SPRN_TFHAR, r5
3039 mtspr SPRN_TFIAR, r6
3040 mtspr SPRN_TEXASR, r7
3043 rldicl. r5, r5, 64 - MSR_TS_S_LG, 62
3044 beqlr /* TM not active in guest */
3045 std r1, HSTATE_HOST_R1(r13)
3047 /* Make sure the failure summary is set, otherwise we'll program check
3048 * when we trechkpt. It's possible that this might have been not set
3049 * on a kvmppc_set_one_reg() call but we shouldn't let this crash the
3052 oris r7, r7, (TEXASR_FS)@h
3053 mtspr SPRN_TEXASR, r7
3056 * We need to load up the checkpointed state for the guest.
3057 * We need to do this early as it will blow away any GPRs, VSRs and
3062 addi r3, r31, VCPU_FPRS_TM
3064 addi r3, r31, VCPU_VRS_TM
3067 lwz r7, VCPU_VRSAVE_TM(r4)
3068 mtspr SPRN_VRSAVE, r7
3070 ld r5, VCPU_LR_TM(r4)
3071 lwz r6, VCPU_CR_TM(r4)
3072 ld r7, VCPU_CTR_TM(r4)
3073 ld r8, VCPU_AMR_TM(r4)
3074 ld r9, VCPU_TAR_TM(r4)
3075 ld r10, VCPU_XER_TM(r4)
3084 * Load up PPR and DSCR values but don't put them in the actual SPRs
3085 * till the last moment to avoid running with userspace PPR and DSCR for
3088 ld r29, VCPU_DSCR_TM(r4)
3089 ld r30, VCPU_PPR_TM(r4)
3091 std r2, PACATMSCRATCH(r13) /* Save TOC */
3093 /* Clear the MSR RI since r1, r13 are all going to be foobar. */
3097 /* Load GPRs r0-r28 */
3100 ld reg, VCPU_GPRS_TM(reg)(r31)
3104 mtspr SPRN_DSCR, r29
3107 /* Load final GPRs */
3108 ld 29, VCPU_GPRS_TM(29)(r31)
3109 ld 30, VCPU_GPRS_TM(30)(r31)
3110 ld 31, VCPU_GPRS_TM(31)(r31)
3112 /* TM checkpointed state is now setup. All GPRs are now volatile. */
3115 /* Now let's get back the state we need. */
3118 ld r29, HSTATE_DSCR(r13)
3119 mtspr SPRN_DSCR, r29
3120 ld r4, HSTATE_KVM_VCPU(r13)
3121 ld r1, HSTATE_HOST_R1(r13)
3122 ld r2, PACATMSCRATCH(r13)
3124 /* Set the MSR RI since we have our registers back. */
3128 ld r0, PPC_LR_STKOFF(r1)
3134 * We come here if we get any exception or interrupt while we are
3135 * executing host real mode code while in guest MMU context.
3136 * For now just spin, but we should do something better.
3138 kvmppc_bad_host_intr:
3142 * This mimics the MSR transition on IRQ delivery. The new guest MSR is taken
3143 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
3144 * r11 has the guest MSR value (in/out)
3145 * r9 has a vcpu pointer (in)
3146 * r0 is used as a scratch register
3148 kvmppc_msr_interrupt:
3149 rldicl r0, r11, 64 - MSR_TS_S_LG, 62
3150 cmpwi r0, 2 /* Check if we are in transactional state.. */
3151 ld r11, VCPU_INTR_MSR(r9)
3153 /* ... if transactional, change to suspended */
3155 1: rldimi r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
3159 * This works around a hardware bug on POWER8E processors, where
3160 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
3161 * performance monitor interrupt. Instead, when we need to have
3162 * an interrupt pending, we have to arrange for a counter to overflow.
3166 mtspr SPRN_MMCR2, r3
3167 lis r3, (MMCR0_PMXE | MMCR0_FCECE)@h
3168 ori r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
3169 mtspr SPRN_MMCR0, r3
3176 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
3178 * Start timing an activity
3179 * r3 = pointer to time accumulation struct, r4 = vcpu
3182 ld r5, HSTATE_KVM_VCORE(r13)
3183 lbz r6, VCORE_IN_GUEST(r5)
3185 beq 5f /* if in guest, need to */
3186 ld r6, VCORE_TB_OFFSET(r5) /* subtract timebase offset */
3189 std r3, VCPU_CUR_ACTIVITY(r4)
3190 std r5, VCPU_ACTIVITY_START(r4)
3194 * Accumulate time to one activity and start another.
3195 * r3 = pointer to new time accumulation struct, r4 = vcpu
3197 kvmhv_accumulate_time:
3198 ld r5, HSTATE_KVM_VCORE(r13)
3199 lbz r8, VCORE_IN_GUEST(r5)
3201 beq 4f /* if in guest, need to */
3202 ld r8, VCORE_TB_OFFSET(r5) /* subtract timebase offset */
3203 4: ld r5, VCPU_CUR_ACTIVITY(r4)
3204 ld r6, VCPU_ACTIVITY_START(r4)
3205 std r3, VCPU_CUR_ACTIVITY(r4)
3208 std r7, VCPU_ACTIVITY_START(r4)
3212 ld r8, TAS_SEQCOUNT(r5)
3215 std r8, TAS_SEQCOUNT(r5)
3217 ld r7, TAS_TOTAL(r5)
3219 std r7, TAS_TOTAL(r5)
3225 3: std r3, TAS_MIN(r5)
3231 std r8, TAS_SEQCOUNT(r5)